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/backing-dev.h>
  11 #include <linux/init.h>
  12 #include <linux/f2fs_fs.h>
  13 #include <linux/kthread.h>
  14 #include <linux/delay.h>
  15 #include <linux/freezer.h>
  16 #include <linux/sched/signal.h>
  17
  18 #include "f2fs.h"
  19 #include "node.h"
  20 #include "segment.h"
  21 #include "gc.h"
  22 #include "iostat.h"
  23 #include <trace/events/f2fs.h>
  24
  25 static struct kmem_cache *victim_entry_slab;
  26
  27 static unsigned int count_bits(const unsigned long *addr,
  28                                 unsigned int offset, unsigned int len);
  29
  30 static int gc_thread_func(void *data)
  31 {
  32         struct f2fs_sb_info *sbi = data;
  33         struct f2fs_gc_kthread *gc_th = sbi->gc_thread;
  34         wait_queue_head_t *wq = &sbi->gc_thread->gc_wait_queue_head;
  35         wait_queue_head_t *fggc_wq = &sbi->gc_thread->fggc_wq;
  36         unsigned int wait_ms;
  37
  38         wait_ms = gc_th->min_sleep_time;
  39
  40         set_freezable();
  41         do {
  42                 bool sync_mode, foreground = false;
  43
  44                 wait_event_interruptible_timeout(*wq,
  45                                 kthread_should_stop() || freezing(current) ||
  46                                 waitqueue_active(fggc_wq) ||
  47                                 gc_th->gc_wake,
  48                                 msecs_to_jiffies(wait_ms));
  49
  50                 if (test_opt(sbi, GC_MERGE) && waitqueue_active(fggc_wq))
  51                         foreground = true;
  52
  53                 /* give it a try one time */
  54                 if (gc_th->gc_wake)
  55                         gc_th->gc_wake = 0;
  56
  57                 if (try_to_freeze()) {
  58                         stat_other_skip_bggc_count(sbi);
  59                         continue;
  60                 }
  61                 if (kthread_should_stop())
  62                         break;
  63
  64                 if (sbi->sb->s_writers.frozen >= SB_FREEZE_WRITE) {
  65                         increase_sleep_time(gc_th, &wait_ms);
  66                         stat_other_skip_bggc_count(sbi);
  67                         continue;
  68                 }
  69
  70                 if (time_to_inject(sbi, FAULT_CHECKPOINT)) {
  71                         f2fs_show_injection_info(sbi, FAULT_CHECKPOINT);
  72                         f2fs_stop_checkpoint(sbi, false);
  73                 }
  74
  75                 if (!sb_start_write_trylock(sbi->sb)) {
  76                         stat_other_skip_bggc_count(sbi);
  77                         continue;
  78                 }
  79
  80                 /*
  81                  * [GC triggering condition]
  82                  * 0. GC is not conducted currently.
  83                  * 1. There are enough dirty segments.
  84                  * 2. IO subsystem is idle by checking the # of writeback pages.
  85                  * 3. IO subsystem is idle by checking the # of requests in
  86                  *    bdev's request list.
  87                  *
  88                  * Note) We have to avoid triggering GCs frequently.
  89                  * Because it is possible that some segments can be
  90                  * invalidated soon after by user update or deletion.
  91                  * So, I'd like to wait some time to collect dirty segments.
  92                  */
  93                 if (sbi->gc_mode == GC_URGENT_HIGH) {
  94                         wait_ms = gc_th->urgent_sleep_time;
  95                         down_write(&sbi->gc_lock);
  96                         goto do_gc;
  97                 }
  98
  99                 if (foreground) {
 100                         down_write(&sbi->gc_lock);
 101                         goto do_gc;
 102                 } else if (!down_write_trylock(&sbi->gc_lock)) {
 103                         stat_other_skip_bggc_count(sbi);
 104                         goto next;
 105                 }
 106
 107                 if (!is_idle(sbi, GC_TIME)) {
 108                         increase_sleep_time(gc_th, &wait_ms);
 109                         up_write(&sbi->gc_lock);
 110                         stat_io_skip_bggc_count(sbi);
 111                         goto next;
 112                 }
 113
 114                 if (has_enough_invalid_blocks(sbi))
 115                         decrease_sleep_time(gc_th, &wait_ms);
 116                 else
 117                         increase_sleep_time(gc_th, &wait_ms);
 118 do_gc:
 119                 if (!foreground)
 120                         stat_inc_bggc_count(sbi->stat_info);
 121
 122                 sync_mode = F2FS_OPTION(sbi).bggc_mode == BGGC_MODE_SYNC;
 123
 124                 /* foreground GC was been triggered via f2fs_balance_fs() */
 125                 if (foreground)
 126                         sync_mode = false;
 127
 128                 /* if return value is not zero, no victim was selected */
 129                 if (f2fs_gc(sbi, sync_mode, !foreground, false, NULL_SEGNO))
 130                         wait_ms = gc_th->no_gc_sleep_time;
 131
 132                 if (foreground)
 133                         wake_up_all(&gc_th->fggc_wq);
 134
 135                 trace_f2fs_background_gc(sbi->sb, wait_ms,
 136                                 prefree_segments(sbi), free_segments(sbi));
 137
 138                 /* balancing f2fs's metadata periodically */
 139                 f2fs_balance_fs_bg(sbi, true);
 140 next:
 141                 sb_end_write(sbi->sb);
 142
 143         } while (!kthread_should_stop());
 144         return 0;
 145 }
 146
 147 int f2fs_start_gc_thread(struct f2fs_sb_info *sbi)
 148 {
 149         struct f2fs_gc_kthread *gc_th;
 150         dev_t dev = sbi->sb->s_bdev->bd_dev;
 151         int err = 0;
 152
 153         gc_th = f2fs_kmalloc(sbi, sizeof(struct f2fs_gc_kthread), GFP_KERNEL);
 154         if (!gc_th) {
 155                 err = -ENOMEM;
 156                 goto out;
 157         }
 158
 159         gc_th->urgent_sleep_time = DEF_GC_THREAD_URGENT_SLEEP_TIME;
 160         gc_th->min_sleep_time = DEF_GC_THREAD_MIN_SLEEP_TIME;
 161         gc_th->max_sleep_time = DEF_GC_THREAD_MAX_SLEEP_TIME;
 162         gc_th->no_gc_sleep_time = DEF_GC_THREAD_NOGC_SLEEP_TIME;
 163
 164         gc_th->gc_wake = 0;
 165
 166         sbi->gc_thread = gc_th;
 167         init_waitqueue_head(&sbi->gc_thread->gc_wait_queue_head);
 168         init_waitqueue_head(&sbi->gc_thread->fggc_wq);
 169         sbi->gc_thread->f2fs_gc_task = kthread_run(gc_thread_func, sbi,
 170                         "f2fs_gc-%u:%u", MAJOR(dev), MINOR(dev));
 171         if (IS_ERR(gc_th->f2fs_gc_task)) {
 172                 err = PTR_ERR(gc_th->f2fs_gc_task);
 173                 kfree(gc_th);
 174                 sbi->gc_thread = NULL;
 175         }
 176 out:
 177         return err;
 178 }
 179
 180 void f2fs_stop_gc_thread(struct f2fs_sb_info *sbi)
 181 {
 182         struct f2fs_gc_kthread *gc_th = sbi->gc_thread;
 183
 184         if (!gc_th)
 185                 return;
 186         kthread_stop(gc_th->f2fs_gc_task);
 187         wake_up_all(&gc_th->fggc_wq);
 188         kfree(gc_th);
 189         sbi->gc_thread = NULL;
 190 }
 191
 192 static int select_gc_type(struct f2fs_sb_info *sbi, int gc_type)
 193 {
 194         int gc_mode;
 195
 196         if (gc_type == BG_GC) {
 197                 if (sbi->am.atgc_enabled)
 198                         gc_mode = GC_AT;
 199                 else
 200                         gc_mode = GC_CB;
 201         } else {
 202                 gc_mode = GC_GREEDY;
 203         }
 204
 205         switch (sbi->gc_mode) {
 206         case GC_IDLE_CB:
 207                 gc_mode = GC_CB;
 208                 break;
 209         case GC_IDLE_GREEDY:
 210         case GC_URGENT_HIGH:
 211                 gc_mode = GC_GREEDY;
 212                 break;
 213         case GC_IDLE_AT:
 214                 gc_mode = GC_AT;
 215                 break;
 216         }
 217
 218         return gc_mode;
 219 }
 220
 221 static void select_policy(struct f2fs_sb_info *sbi, int gc_type,
 222                         int type, struct victim_sel_policy *p)
 223 {
 224         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
 225
 226         if (p->alloc_mode == SSR) {
 227                 p->gc_mode = GC_GREEDY;
 228                 p->dirty_bitmap = dirty_i->dirty_segmap[type];
 229                 p->max_search = dirty_i->nr_dirty[type];
 230                 p->ofs_unit = 1;
 231         } else if (p->alloc_mode == AT_SSR) {
 232                 p->gc_mode = GC_GREEDY;
 233                 p->dirty_bitmap = dirty_i->dirty_segmap[type];
 234                 p->max_search = dirty_i->nr_dirty[type];
 235                 p->ofs_unit = 1;
 236         } else {
 237                 p->gc_mode = select_gc_type(sbi, gc_type);
 238                 p->ofs_unit = sbi->segs_per_sec;
 239                 if (__is_large_section(sbi)) {
 240                         p->dirty_bitmap = dirty_i->dirty_secmap;
 241                         p->max_search = count_bits(p->dirty_bitmap,
 242                                                 0, MAIN_SECS(sbi));
 243                 } else {
 244                         p->dirty_bitmap = dirty_i->dirty_segmap[DIRTY];
 245                         p->max_search = dirty_i->nr_dirty[DIRTY];
 246                 }
 247         }
 248
 249         /*
 250          * adjust candidates range, should select all dirty segments for
 251          * foreground GC and urgent GC cases.
 252          */
 253         if (gc_type != FG_GC &&
 254                         (sbi->gc_mode != GC_URGENT_HIGH) &&
 255                         (p->gc_mode != GC_AT && p->alloc_mode != AT_SSR) &&
 256                         p->max_search > sbi->max_victim_search)
 257                 p->max_search = sbi->max_victim_search;
 258
 259         /* let's select beginning hot/small space first in no_heap mode*/
 260         if (test_opt(sbi, NOHEAP) &&
 261                 (type == CURSEG_HOT_DATA || IS_NODESEG(type)))
 262                 p->offset = 0;
 263         else
 264                 p->offset = SIT_I(sbi)->last_victim[p->gc_mode];
 265 }
 266
 267 static unsigned int get_max_cost(struct f2fs_sb_info *sbi,
 268                                 struct victim_sel_policy *p)
 269 {
 270         /* SSR allocates in a segment unit */
 271         if (p->alloc_mode == SSR)
 272                 return sbi->blocks_per_seg;
 273         else if (p->alloc_mode == AT_SSR)
 274                 return UINT_MAX;
 275
 276         /* LFS */
 277         if (p->gc_mode == GC_GREEDY)
 278                 return 2 * sbi->blocks_per_seg * p->ofs_unit;
 279         else if (p->gc_mode == GC_CB)
 280                 return UINT_MAX;
 281         else if (p->gc_mode == GC_AT)
 282                 return UINT_MAX;
 283         else /* No other gc_mode */
 284                 return 0;
 285 }
 286
 287 static unsigned int check_bg_victims(struct f2fs_sb_info *sbi)
 288 {
 289         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
 290         unsigned int secno;
 291
 292         /*
 293          * If the gc_type is FG_GC, we can select victim segments
 294          * selected by background GC before.
 295          * Those segments guarantee they have small valid blocks.
 296          */
 297         for_each_set_bit(secno, dirty_i->victim_secmap, MAIN_SECS(sbi)) {
 298                 if (sec_usage_check(sbi, secno))
 299                         continue;
 300                 clear_bit(secno, dirty_i->victim_secmap);
 301                 return GET_SEG_FROM_SEC(sbi, secno);
 302         }
 303         return NULL_SEGNO;
 304 }
 305
 306 static unsigned int get_cb_cost(struct f2fs_sb_info *sbi, unsigned int segno)
 307 {
 308         struct sit_info *sit_i = SIT_I(sbi);
 309         unsigned int secno = GET_SEC_FROM_SEG(sbi, segno);
 310         unsigned int start = GET_SEG_FROM_SEC(sbi, secno);
 311         unsigned long long mtime = 0;
 312         unsigned int vblocks;
 313         unsigned char age = 0;
 314         unsigned char u;
 315         unsigned int i;
 316         unsigned int usable_segs_per_sec = f2fs_usable_segs_in_sec(sbi, segno);
 317
 318         for (i = 0; i < usable_segs_per_sec; i++)
 319                 mtime += get_seg_entry(sbi, start + i)->mtime;
 320         vblocks = get_valid_blocks(sbi, segno, true);
 321
 322         mtime = div_u64(mtime, usable_segs_per_sec);
 323         vblocks = div_u64(vblocks, usable_segs_per_sec);
 324
 325         u = (vblocks * 100) >> sbi->log_blocks_per_seg;
 326
 327         /* Handle if the system time has changed by the user */
 328         if (mtime < sit_i->min_mtime)
 329                 sit_i->min_mtime = mtime;
 330         if (mtime > sit_i->max_mtime)
 331                 sit_i->max_mtime = mtime;
 332         if (sit_i->max_mtime != sit_i->min_mtime)
 333                 age = 100 - div64_u64(100 * (mtime - sit_i->min_mtime),
 334                                 sit_i->max_mtime - sit_i->min_mtime);
 335
 336         return UINT_MAX - ((100 * (100 - u) * age) / (100 + u));
 337 }
 338
 339 static inline unsigned int get_gc_cost(struct f2fs_sb_info *sbi,
 340                         unsigned int segno, struct victim_sel_policy *p)
 341 {
 342         if (p->alloc_mode == SSR)
 343                 return get_seg_entry(sbi, segno)->ckpt_valid_blocks;
 344
 345         /* alloc_mode == LFS */
 346         if (p->gc_mode == GC_GREEDY)
 347                 return get_valid_blocks(sbi, segno, true);
 348         else if (p->gc_mode == GC_CB)
 349                 return get_cb_cost(sbi, segno);
 350
 351         f2fs_bug_on(sbi, 1);
 352         return 0;
 353 }
 354
 355 static unsigned int count_bits(const unsigned long *addr,
 356                                 unsigned int offset, unsigned int len)
 357 {
 358         unsigned int end = offset + len, sum = 0;
 359
 360         while (offset < end) {
 361                 if (test_bit(offset++, addr))
 362                         ++sum;
 363         }
 364         return sum;
 365 }
 366
 367 static struct victim_entry *attach_victim_entry(struct f2fs_sb_info *sbi,
 368                                 unsigned long long mtime, unsigned int segno,
 369                                 struct rb_node *parent, struct rb_node **p,
 370                                 bool left_most)
 371 {
 372         struct atgc_management *am = &sbi->am;
 373         struct victim_entry *ve;
 374
 375         ve =  f2fs_kmem_cache_alloc(victim_entry_slab,
 376                                 GFP_NOFS, true, NULL);
 377
 378         ve->mtime = mtime;
 379         ve->segno = segno;
 380
 381         rb_link_node(&ve->rb_node, parent, p);
 382         rb_insert_color_cached(&ve->rb_node, &am->root, left_most);
 383
 384         list_add_tail(&ve->list, &am->victim_list);
 385
 386         am->victim_count++;
 387
 388         return ve;
 389 }
 390
 391 static void insert_victim_entry(struct f2fs_sb_info *sbi,
 392                                 unsigned long long mtime, unsigned int segno)
 393 {
 394         struct atgc_management *am = &sbi->am;
 395         struct rb_node **p;
 396         struct rb_node *parent = NULL;
 397         bool left_most = true;
 398
 399         p = f2fs_lookup_rb_tree_ext(sbi, &am->root, &parent, mtime, &left_most);
 400         attach_victim_entry(sbi, mtime, segno, parent, p, left_most);
 401 }
 402
 403 static void add_victim_entry(struct f2fs_sb_info *sbi,
 404                                 struct victim_sel_policy *p, unsigned int segno)
 405 {
 406         struct sit_info *sit_i = SIT_I(sbi);
 407         unsigned int secno = GET_SEC_FROM_SEG(sbi, segno);
 408         unsigned int start = GET_SEG_FROM_SEC(sbi, secno);
 409         unsigned long long mtime = 0;
 410         unsigned int i;
 411
 412         if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED))) {
 413                 if (p->gc_mode == GC_AT &&
 414                         get_valid_blocks(sbi, segno, true) == 0)
 415                         return;
 416         }
 417
 418         for (i = 0; i < sbi->segs_per_sec; i++)
 419                 mtime += get_seg_entry(sbi, start + i)->mtime;
 420         mtime = div_u64(mtime, sbi->segs_per_sec);
 421
 422         /* Handle if the system time has changed by the user */
 423         if (mtime < sit_i->min_mtime)
 424                 sit_i->min_mtime = mtime;
 425         if (mtime > sit_i->max_mtime)
 426                 sit_i->max_mtime = mtime;
 427         if (mtime < sit_i->dirty_min_mtime)
 428                 sit_i->dirty_min_mtime = mtime;
 429         if (mtime > sit_i->dirty_max_mtime)
 430                 sit_i->dirty_max_mtime = mtime;
 431
 432         /* don't choose young section as candidate */
 433         if (sit_i->dirty_max_mtime - mtime < p->age_threshold)
 434                 return;
 435
 436         insert_victim_entry(sbi, mtime, segno);
 437 }
 438
 439 static struct rb_node *lookup_central_victim(struct f2fs_sb_info *sbi,
 440                                                 struct victim_sel_policy *p)
 441 {
 442         struct atgc_management *am = &sbi->am;
 443         struct rb_node *parent = NULL;
 444         bool left_most;
 445
 446         f2fs_lookup_rb_tree_ext(sbi, &am->root, &parent, p->age, &left_most);
 447
 448         return parent;
 449 }
 450
 451 static void atgc_lookup_victim(struct f2fs_sb_info *sbi,
 452                                                 struct victim_sel_policy *p)
 453 {
 454         struct sit_info *sit_i = SIT_I(sbi);
 455         struct atgc_management *am = &sbi->am;
 456         struct rb_root_cached *root = &am->root;
 457         struct rb_node *node;
 458         struct rb_entry *re;
 459         struct victim_entry *ve;
 460         unsigned long long total_time;
 461         unsigned long long age, u, accu;
 462         unsigned long long max_mtime = sit_i->dirty_max_mtime;
 463         unsigned long long min_mtime = sit_i->dirty_min_mtime;
 464         unsigned int sec_blocks = BLKS_PER_SEC(sbi);
 465         unsigned int vblocks;
 466         unsigned int dirty_threshold = max(am->max_candidate_count,
 467                                         am->candidate_ratio *
 468                                         am->victim_count / 100);
 469         unsigned int age_weight = am->age_weight;
 470         unsigned int cost;
 471         unsigned int iter = 0;
 472
 473         if (max_mtime < min_mtime)
 474                 return;
 475
 476         max_mtime += 1;
 477         total_time = max_mtime - min_mtime;
 478
 479         accu = div64_u64(ULLONG_MAX, total_time);
 480         accu = min_t(unsigned long long, div_u64(accu, 100),
 481                                         DEFAULT_ACCURACY_CLASS);
 482
 483         node = rb_first_cached(root);
 484 next:
 485         re = rb_entry_safe(node, struct rb_entry, rb_node);
 486         if (!re)
 487                 return;
 488
 489         ve = (struct victim_entry *)re;
 490
 491         if (ve->mtime >= max_mtime || ve->mtime < min_mtime)
 492                 goto skip;
 493
 494         /* age = 10000 * x% * 60 */
 495         age = div64_u64(accu * (max_mtime - ve->mtime), total_time) *
 496                                                                 age_weight;
 497
 498         vblocks = get_valid_blocks(sbi, ve->segno, true);
 499         f2fs_bug_on(sbi, !vblocks || vblocks == sec_blocks);
 500
 501         /* u = 10000 * x% * 40 */
 502         u = div64_u64(accu * (sec_blocks - vblocks), sec_blocks) *
 503                                                         (100 - age_weight);
 504
 505         f2fs_bug_on(sbi, age + u >= UINT_MAX);
 506
 507         cost = UINT_MAX - (age + u);
 508         iter++;
 509
 510         if (cost < p->min_cost ||
 511                         (cost == p->min_cost && age > p->oldest_age)) {
 512                 p->min_cost = cost;
 513                 p->oldest_age = age;
 514                 p->min_segno = ve->segno;
 515         }
 516 skip:
 517         if (iter < dirty_threshold) {
 518                 node = rb_next(node);
 519                 goto next;
 520         }
 521 }
 522
 523 /*
 524  * select candidates around source section in range of
 525  * [target - dirty_threshold, target + dirty_threshold]
 526  */
 527 static void atssr_lookup_victim(struct f2fs_sb_info *sbi,
 528                                                 struct victim_sel_policy *p)
 529 {
 530         struct sit_info *sit_i = SIT_I(sbi);
 531         struct atgc_management *am = &sbi->am;
 532         struct rb_node *node;
 533         struct rb_entry *re;
 534         struct victim_entry *ve;
 535         unsigned long long age;
 536         unsigned long long max_mtime = sit_i->dirty_max_mtime;
 537         unsigned long long min_mtime = sit_i->dirty_min_mtime;
 538         unsigned int seg_blocks = sbi->blocks_per_seg;
 539         unsigned int vblocks;
 540         unsigned int dirty_threshold = max(am->max_candidate_count,
 541                                         am->candidate_ratio *
 542                                         am->victim_count / 100);
 543         unsigned int cost;
 544         unsigned int iter = 0;
 545         int stage = 0;
 546
 547         if (max_mtime < min_mtime)
 548                 return;
 549         max_mtime += 1;
 550 next_stage:
 551         node = lookup_central_victim(sbi, p);
 552 next_node:
 553         re = rb_entry_safe(node, struct rb_entry, rb_node);
 554         if (!re) {
 555                 if (stage == 0)
 556                         goto skip_stage;
 557                 return;
 558         }
 559
 560         ve = (struct victim_entry *)re;
 561
 562         if (ve->mtime >= max_mtime || ve->mtime < min_mtime)
 563                 goto skip_node;
 564
 565         age = max_mtime - ve->mtime;
 566
 567         vblocks = get_seg_entry(sbi, ve->segno)->ckpt_valid_blocks;
 568         f2fs_bug_on(sbi, !vblocks);
 569
 570         /* rare case */
 571         if (vblocks == seg_blocks)
 572                 goto skip_node;
 573
 574         iter++;
 575
 576         age = max_mtime - abs(p->age - age);
 577         cost = UINT_MAX - vblocks;
 578
 579         if (cost < p->min_cost ||
 580                         (cost == p->min_cost && age > p->oldest_age)) {
 581                 p->min_cost = cost;
 582                 p->oldest_age = age;
 583                 p->min_segno = ve->segno;
 584         }
 585 skip_node:
 586         if (iter < dirty_threshold) {
 587                 if (stage == 0)
 588                         node = rb_prev(node);
 589                 else if (stage == 1)
 590                         node = rb_next(node);
 591                 goto next_node;
 592         }
 593 skip_stage:
 594         if (stage < 1) {
 595                 stage++;
 596                 iter = 0;
 597                 goto next_stage;
 598         }
 599 }
 600 static void lookup_victim_by_age(struct f2fs_sb_info *sbi,
 601                                                 struct victim_sel_policy *p)
 602 {
 603         f2fs_bug_on(sbi, !f2fs_check_rb_tree_consistence(sbi,
 604                                                 &sbi->am.root, true));
 605
 606         if (p->gc_mode == GC_AT)
 607                 atgc_lookup_victim(sbi, p);
 608         else if (p->alloc_mode == AT_SSR)
 609                 atssr_lookup_victim(sbi, p);
 610         else
 611                 f2fs_bug_on(sbi, 1);
 612 }
 613
 614 static void release_victim_entry(struct f2fs_sb_info *sbi)
 615 {
 616         struct atgc_management *am = &sbi->am;
 617         struct victim_entry *ve, *tmp;
 618
 619         list_for_each_entry_safe(ve, tmp, &am->victim_list, list) {
 620                 list_del(&ve->list);
 621                 kmem_cache_free(victim_entry_slab, ve);
 622                 am->victim_count--;
 623         }
 624
 625         am->root = RB_ROOT_CACHED;
 626
 627         f2fs_bug_on(sbi, am->victim_count);
 628         f2fs_bug_on(sbi, !list_empty(&am->victim_list));
 629 }
 630
 631 /*
 632  * This function is called from two paths.
 633  * One is garbage collection and the other is SSR segment selection.
 634  * When it is called during GC, it just gets a victim segment
 635  * and it does not remove it from dirty seglist.
 636  * When it is called from SSR segment selection, it finds a segment
 637  * which has minimum valid blocks and removes it from dirty seglist.
 638  */
 639 static int get_victim_by_default(struct f2fs_sb_info *sbi,
 640                         unsigned int *result, int gc_type, int type,
 641                         char alloc_mode, unsigned long long age)
 642 {
 643         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
 644         struct sit_info *sm = SIT_I(sbi);
 645         struct victim_sel_policy p;
 646         unsigned int secno, last_victim;
 647         unsigned int last_segment;
 648         unsigned int nsearched;
 649         bool is_atgc;
 650         int ret = 0;
 651
 652         mutex_lock(&dirty_i->seglist_lock);
 653         last_segment = MAIN_SECS(sbi) * sbi->segs_per_sec;
 654
 655         p.alloc_mode = alloc_mode;
 656         p.age = age;
 657         p.age_threshold = sbi->am.age_threshold;
 658
 659 retry:
 660         select_policy(sbi, gc_type, type, &p);
 661         p.min_segno = NULL_SEGNO;
 662         p.oldest_age = 0;
 663         p.min_cost = get_max_cost(sbi, &p);
 664
 665         is_atgc = (p.gc_mode == GC_AT || p.alloc_mode == AT_SSR);
 666         nsearched = 0;
 667
 668         if (is_atgc)
 669                 SIT_I(sbi)->dirty_min_mtime = ULLONG_MAX;
 670
 671         if (*result != NULL_SEGNO) {
 672                 if (!get_valid_blocks(sbi, *result, false)) {
 673                         ret = -ENODATA;
 674                         goto out;
 675                 }
 676
 677                 if (sec_usage_check(sbi, GET_SEC_FROM_SEG(sbi, *result)))
 678                         ret = -EBUSY;
 679                 else
 680                         p.min_segno = *result;
 681                 goto out;
 682         }
 683
 684         ret = -ENODATA;
 685         if (p.max_search == 0)
 686                 goto out;
 687
 688         if (__is_large_section(sbi) && p.alloc_mode == LFS) {
 689                 if (sbi->next_victim_seg[BG_GC] != NULL_SEGNO) {
 690                         p.min_segno = sbi->next_victim_seg[BG_GC];
 691                         *result = p.min_segno;
 692                         sbi->next_victim_seg[BG_GC] = NULL_SEGNO;
 693                         goto got_result;
 694                 }
 695                 if (gc_type == FG_GC &&
 696                                 sbi->next_victim_seg[FG_GC] != NULL_SEGNO) {
 697                         p.min_segno = sbi->next_victim_seg[FG_GC];
 698                         *result = p.min_segno;
 699                         sbi->next_victim_seg[FG_GC] = NULL_SEGNO;
 700                         goto got_result;
 701                 }
 702         }
 703
 704         last_victim = sm->last_victim[p.gc_mode];
 705         if (p.alloc_mode == LFS && gc_type == FG_GC) {
 706                 p.min_segno = check_bg_victims(sbi);
 707                 if (p.min_segno != NULL_SEGNO)
 708                         goto got_it;
 709         }
 710
 711         while (1) {
 712                 unsigned long cost, *dirty_bitmap;
 713                 unsigned int unit_no, segno;
 714
 715                 dirty_bitmap = p.dirty_bitmap;
 716                 unit_no = find_next_bit(dirty_bitmap,
 717                                 last_segment / p.ofs_unit,
 718                                 p.offset / p.ofs_unit);
 719                 segno = unit_no * p.ofs_unit;
 720                 if (segno >= last_segment) {
 721                         if (sm->last_victim[p.gc_mode]) {
 722                                 last_segment =
 723                                         sm->last_victim[p.gc_mode];
 724                                 sm->last_victim[p.gc_mode] = 0;
 725                                 p.offset = 0;
 726                                 continue;
 727                         }
 728                         break;
 729                 }
 730
 731                 p.offset = segno + p.ofs_unit;
 732                 nsearched++;
 733
 734 #ifdef CONFIG_F2FS_CHECK_FS
 735                 /*
 736                  * skip selecting the invalid segno (that is failed due to block
 737                  * validity check failure during GC) to avoid endless GC loop in
 738                  * such cases.
 739                  */
 740                 if (test_bit(segno, sm->invalid_segmap))
 741                         goto next;
 742 #endif
 743
 744                 secno = GET_SEC_FROM_SEG(sbi, segno);
 745
 746                 if (sec_usage_check(sbi, secno))
 747                         goto next;
 748
 749                 /* Don't touch checkpointed data */
 750                 if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED))) {
 751                         if (p.alloc_mode == LFS) {
 752                                 /*
 753                                  * LFS is set to find source section during GC.
 754                                  * The victim should have no checkpointed data.
 755                                  */
 756                                 if (get_ckpt_valid_blocks(sbi, segno, true))
 757                                         goto next;
 758                         } else {
 759                                 /*
 760                                  * SSR | AT_SSR are set to find target segment
 761                                  * for writes which can be full by checkpointed
 762                                  * and newly written blocks.
 763                                  */
 764                                 if (!f2fs_segment_has_free_slot(sbi, segno))
 765                                         goto next;
 766                         }
 767                 }
 768
 769                 if (gc_type == BG_GC && test_bit(secno, dirty_i->victim_secmap))
 770                         goto next;
 771
 772                 if (is_atgc) {
 773                         add_victim_entry(sbi, &p, segno);
 774                         goto next;
 775                 }
 776
 777                 cost = get_gc_cost(sbi, segno, &p);
 778
 779                 if (p.min_cost > cost) {
 780                         p.min_segno = segno;
 781                         p.min_cost = cost;
 782                 }
 783 next:
 784                 if (nsearched >= p.max_search) {
 785                         if (!sm->last_victim[p.gc_mode] && segno <= last_victim)
 786                                 sm->last_victim[p.gc_mode] =
 787                                         last_victim + p.ofs_unit;
 788                         else
 789                                 sm->last_victim[p.gc_mode] = segno + p.ofs_unit;
 790                         sm->last_victim[p.gc_mode] %=
 791                                 (MAIN_SECS(sbi) * sbi->segs_per_sec);
 792                         break;
 793                 }
 794         }
 795
 796         /* get victim for GC_AT/AT_SSR */
 797         if (is_atgc) {
 798                 lookup_victim_by_age(sbi, &p);
 799                 release_victim_entry(sbi);
 800         }
 801
 802         if (is_atgc && p.min_segno == NULL_SEGNO &&
 803                         sm->elapsed_time < p.age_threshold) {
 804                 p.age_threshold = 0;
 805                 goto retry;
 806         }
 807
 808         if (p.min_segno != NULL_SEGNO) {
 809 got_it:
 810                 *result = (p.min_segno / p.ofs_unit) * p.ofs_unit;
 811 got_result:
 812                 if (p.alloc_mode == LFS) {
 813                         secno = GET_SEC_FROM_SEG(sbi, p.min_segno);
 814                         if (gc_type == FG_GC)
 815                                 sbi->cur_victim_sec = secno;
 816                         else
 817                                 set_bit(secno, dirty_i->victim_secmap);
 818                 }
 819                 ret = 0;
 820
 821         }
 822 out:
 823         if (p.min_segno != NULL_SEGNO)
 824                 trace_f2fs_get_victim(sbi->sb, type, gc_type, &p,
 825                                 sbi->cur_victim_sec,
 826                                 prefree_segments(sbi), free_segments(sbi));
 827         mutex_unlock(&dirty_i->seglist_lock);
 828
 829         return ret;
 830 }
 831
 832 static const struct victim_selection default_v_ops = {
 833         .get_victim = get_victim_by_default,
 834 };
 835
 836 static struct inode *find_gc_inode(struct gc_inode_list *gc_list, nid_t ino)
 837 {
 838         struct inode_entry *ie;
 839
 840         ie = radix_tree_lookup(&gc_list->iroot, ino);
 841         if (ie)
 842                 return ie->inode;
 843         return NULL;
 844 }
 845
 846 static void add_gc_inode(struct gc_inode_list *gc_list, struct inode *inode)
 847 {
 848         struct inode_entry *new_ie;
 849
 850         if (inode == find_gc_inode(gc_list, inode->i_ino)) {
 851                 iput(inode);
 852                 return;
 853         }
 854         new_ie = f2fs_kmem_cache_alloc(f2fs_inode_entry_slab,
 855                                         GFP_NOFS, true, NULL);
 856         new_ie->inode = inode;
 857
 858         f2fs_radix_tree_insert(&gc_list->iroot, inode->i_ino, new_ie);
 859         list_add_tail(&new_ie->list, &gc_list->ilist);
 860 }
 861
 862 static void put_gc_inode(struct gc_inode_list *gc_list)
 863 {
 864         struct inode_entry *ie, *next_ie;
 865
 866         list_for_each_entry_safe(ie, next_ie, &gc_list->ilist, list) {
 867                 radix_tree_delete(&gc_list->iroot, ie->inode->i_ino);
 868                 iput(ie->inode);
 869                 list_del(&ie->list);
 870                 kmem_cache_free(f2fs_inode_entry_slab, ie);
 871         }
 872 }
 873
 874 static int check_valid_map(struct f2fs_sb_info *sbi,
 875                                 unsigned int segno, int offset)
 876 {
 877         struct sit_info *sit_i = SIT_I(sbi);
 878         struct seg_entry *sentry;
 879         int ret;
 880
 881         down_read(&sit_i->sentry_lock);
 882         sentry = get_seg_entry(sbi, segno);
 883         ret = f2fs_test_bit(offset, sentry->cur_valid_map);
 884         up_read(&sit_i->sentry_lock);
 885         return ret;
 886 }
 887
 888 /*
 889  * This function compares node address got in summary with that in NAT.
 890  * On validity, copy that node with cold status, otherwise (invalid node)
 891  * ignore that.
 892  */
 893 static int gc_node_segment(struct f2fs_sb_info *sbi,
 894                 struct f2fs_summary *sum, unsigned int segno, int gc_type)
 895 {
 896         struct f2fs_summary *entry;
 897         block_t start_addr;
 898         int off;
 899         int phase = 0;
 900         bool fggc = (gc_type == FG_GC);
 901         int submitted = 0;
 902         unsigned int usable_blks_in_seg = f2fs_usable_blks_in_seg(sbi, segno);
 903
 904         start_addr = START_BLOCK(sbi, segno);
 905
 906 next_step:
 907         entry = sum;
 908
 909         if (fggc && phase == 2)
 910                 atomic_inc(&sbi->wb_sync_req[NODE]);
 911
 912         for (off = 0; off < usable_blks_in_seg; off++, entry++) {
 913                 nid_t nid = le32_to_cpu(entry->nid);
 914                 struct page *node_page;
 915                 struct node_info ni;
 916                 int err;
 917
 918                 /* stop BG_GC if there is not enough free sections. */
 919                 if (gc_type == BG_GC && has_not_enough_free_secs(sbi, 0, 0))
 920                         return submitted;
 921
 922                 if (check_valid_map(sbi, segno, off) == 0)
 923                         continue;
 924
 925                 if (phase == 0) {
 926                         f2fs_ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nid), 1,
 927                                                         META_NAT, true);
 928                         continue;
 929                 }
 930
 931                 if (phase == 1) {
 932                         f2fs_ra_node_page(sbi, nid);
 933                         continue;
 934                 }
 935
 936                 /* phase == 2 */
 937                 node_page = f2fs_get_node_page(sbi, nid);
 938                 if (IS_ERR(node_page))
 939                         continue;
 940
 941                 /* block may become invalid during f2fs_get_node_page */
 942                 if (check_valid_map(sbi, segno, off) == 0) {
 943                         f2fs_put_page(node_page, 1);
 944                         continue;
 945                 }
 946
 947                 if (f2fs_get_node_info(sbi, nid, &ni)) {
 948                         f2fs_put_page(node_page, 1);
 949                         continue;
 950                 }
 951
 952                 if (ni.blk_addr != start_addr + off) {
 953                         f2fs_put_page(node_page, 1);
 954                         continue;
 955                 }
 956
 957                 err = f2fs_move_node_page(node_page, gc_type);
 958                 if (!err && gc_type == FG_GC)
 959                         submitted++;
 960                 stat_inc_node_blk_count(sbi, 1, gc_type);
 961         }
 962
 963         if (++phase < 3)
 964                 goto next_step;
 965
 966         if (fggc)
 967                 atomic_dec(&sbi->wb_sync_req[NODE]);
 968         return submitted;
 969 }
 970
 971 /*
 972  * Calculate start block index indicating the given node offset.
 973  * Be careful, caller should give this node offset only indicating direct node
 974  * blocks. If any node offsets, which point the other types of node blocks such
 975  * as indirect or double indirect node blocks, are given, it must be a caller's
 976  * bug.
 977  */
 978 block_t f2fs_start_bidx_of_node(unsigned int node_ofs, struct inode *inode)
 979 {
 980         unsigned int indirect_blks = 2 * NIDS_PER_BLOCK + 4;
 981         unsigned int bidx;
 982
 983         if (node_ofs == 0)
 984                 return 0;
 985
 986         if (node_ofs <= 2) {
 987                 bidx = node_ofs - 1;
 988         } else if (node_ofs <= indirect_blks) {
 989                 int dec = (node_ofs - 4) / (NIDS_PER_BLOCK + 1);
 990
 991                 bidx = node_ofs - 2 - dec;
 992         } else {
 993                 int dec = (node_ofs - indirect_blks - 3) / (NIDS_PER_BLOCK + 1);
 994
 995                 bidx = node_ofs - 5 - dec;
 996         }
 997         return bidx * ADDRS_PER_BLOCK(inode) + ADDRS_PER_INODE(inode);
 998 }
 999
1000 static bool is_alive(struct f2fs_sb_info *sbi, struct f2fs_summary *sum,
1001                 struct node_info *dni, block_t blkaddr, unsigned int *nofs)
1002 {
1003         struct page *node_page;
1004         nid_t nid;
1005         unsigned int ofs_in_node;
1006         block_t source_blkaddr;
1007
1008         nid = le32_to_cpu(sum->nid);
1009         ofs_in_node = le16_to_cpu(sum->ofs_in_node);
1010
1011         node_page = f2fs_get_node_page(sbi, nid);
1012         if (IS_ERR(node_page))
1013                 return false;
1014
1015         if (f2fs_get_node_info(sbi, nid, dni)) {
1016                 f2fs_put_page(node_page, 1);
1017                 return false;
1018         }
1019
1020         if (sum->version != dni->version) {
1021                 f2fs_warn(sbi, "%s: valid data with mismatched node version.",
1022                           __func__);
1023                 set_sbi_flag(sbi, SBI_NEED_FSCK);
1024         }
1025
1026         *nofs = ofs_of_node(node_page);
1027         source_blkaddr = data_blkaddr(NULL, node_page, ofs_in_node);
1028         f2fs_put_page(node_page, 1);
1029
1030         if (source_blkaddr != blkaddr) {
1031 #ifdef CONFIG_F2FS_CHECK_FS
1032                 unsigned int segno = GET_SEGNO(sbi, blkaddr);
1033                 unsigned long offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
1034
1035                 if (unlikely(check_valid_map(sbi, segno, offset))) {
1036                         if (!test_and_set_bit(segno, SIT_I(sbi)->invalid_segmap)) {
1037                                 f2fs_err(sbi, "mismatched blkaddr %u (source_blkaddr %u) in seg %u",
1038                                          blkaddr, source_blkaddr, segno);
1039                                 f2fs_bug_on(sbi, 1);
1040                         }
1041                 }
1042 #endif
1043                 return false;
1044         }
1045         return true;
1046 }
1047
1048 static int ra_data_block(struct inode *inode, pgoff_t index)
1049 {
1050         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1051         struct address_space *mapping = inode->i_mapping;
1052         struct dnode_of_data dn;
1053         struct page *page;
1054         struct extent_info ei = {0, 0, 0};
1055         struct f2fs_io_info fio = {
1056                 .sbi = sbi,
1057                 .ino = inode->i_ino,
1058                 .type = DATA,
1059                 .temp = COLD,
1060                 .op = REQ_OP_READ,
1061                 .op_flags = 0,
1062                 .encrypted_page = NULL,
1063                 .in_list = false,
1064                 .retry = false,
1065         };
1066         int err;
1067
1068         page = f2fs_grab_cache_page(mapping, index, true);
1069         if (!page)
1070                 return -ENOMEM;
1071
1072         if (f2fs_lookup_extent_cache(inode, index, &ei)) {
1073                 dn.data_blkaddr = ei.blk + index - ei.fofs;
1074                 if (unlikely(!f2fs_is_valid_blkaddr(sbi, dn.data_blkaddr,
1075                                                 DATA_GENERIC_ENHANCE_READ))) {
1076                         err = -EFSCORRUPTED;
1077                         goto put_page;
1078                 }
1079                 goto got_it;
1080         }
1081
1082         set_new_dnode(&dn, inode, NULL, NULL, 0);
1083         err = f2fs_get_dnode_of_data(&dn, index, LOOKUP_NODE);
1084         if (err)
1085                 goto put_page;
1086         f2fs_put_dnode(&dn);
1087
1088         if (!__is_valid_data_blkaddr(dn.data_blkaddr)) {
1089                 err = -ENOENT;
1090                 goto put_page;
1091         }
1092         if (unlikely(!f2fs_is_valid_blkaddr(sbi, dn.data_blkaddr,
1093                                                 DATA_GENERIC_ENHANCE))) {
1094                 err = -EFSCORRUPTED;
1095                 goto put_page;
1096         }
1097 got_it:
1098         /* read page */
1099         fio.page = page;
1100         fio.new_blkaddr = fio.old_blkaddr = dn.data_blkaddr;
1101
1102         /*
1103          * don't cache encrypted data into meta inode until previous dirty
1104          * data were writebacked to avoid racing between GC and flush.
1105          */
1106         f2fs_wait_on_page_writeback(page, DATA, true, true);
1107
1108         f2fs_wait_on_block_writeback(inode, dn.data_blkaddr);
1109
1110         fio.encrypted_page = f2fs_pagecache_get_page(META_MAPPING(sbi),
1111                                         dn.data_blkaddr,
1112                                         FGP_LOCK | FGP_CREAT, GFP_NOFS);
1113         if (!fio.encrypted_page) {
1114                 err = -ENOMEM;
1115                 goto put_page;
1116         }
1117
1118         err = f2fs_submit_page_bio(&fio);
1119         if (err)
1120                 goto put_encrypted_page;
1121         f2fs_put_page(fio.encrypted_page, 0);
1122         f2fs_put_page(page, 1);
1123
1124         f2fs_update_iostat(sbi, FS_DATA_READ_IO, F2FS_BLKSIZE);
1125         f2fs_update_iostat(sbi, FS_GDATA_READ_IO, F2FS_BLKSIZE);
1126
1127         return 0;
1128 put_encrypted_page:
1129         f2fs_put_page(fio.encrypted_page, 1);
1130 put_page:
1131         f2fs_put_page(page, 1);
1132         return err;
1133 }
1134
1135 /*
1136  * Move data block via META_MAPPING while keeping locked data page.
1137  * This can be used to move blocks, aka LBAs, directly on disk.
1138  */
1139 static int move_data_block(struct inode *inode, block_t bidx,
1140                                 int gc_type, unsigned int segno, int off)
1141 {
1142         struct f2fs_io_info fio = {
1143                 .sbi = F2FS_I_SB(inode),
1144                 .ino = inode->i_ino,
1145                 .type = DATA,
1146                 .temp = COLD,
1147                 .op = REQ_OP_READ,
1148                 .op_flags = 0,
1149                 .encrypted_page = NULL,
1150                 .in_list = false,
1151                 .retry = false,
1152         };
1153         struct dnode_of_data dn;
1154         struct f2fs_summary sum;
1155         struct node_info ni;
1156         struct page *page, *mpage;
1157         block_t newaddr;
1158         int err = 0;
1159         bool lfs_mode = f2fs_lfs_mode(fio.sbi);
1160         int type = fio.sbi->am.atgc_enabled && (gc_type == BG_GC) &&
1161                                 (fio.sbi->gc_mode != GC_URGENT_HIGH) ?
1162                                 CURSEG_ALL_DATA_ATGC : CURSEG_COLD_DATA;
1163
1164         /* do not read out */
1165         page = f2fs_grab_cache_page(inode->i_mapping, bidx, false);
1166         if (!page)
1167                 return -ENOMEM;
1168
1169         if (!check_valid_map(F2FS_I_SB(inode), segno, off)) {
1170                 err = -ENOENT;
1171                 goto out;
1172         }
1173
1174         if (f2fs_is_atomic_file(inode)) {
1175                 F2FS_I(inode)->i_gc_failures[GC_FAILURE_ATOMIC]++;
1176                 F2FS_I_SB(inode)->skipped_atomic_files[gc_type]++;
1177                 err = -EAGAIN;
1178                 goto out;
1179         }
1180
1181         if (f2fs_is_pinned_file(inode)) {
1182                 f2fs_pin_file_control(inode, true);
1183                 err = -EAGAIN;
1184                 goto out;
1185         }
1186
1187         set_new_dnode(&dn, inode, NULL, NULL, 0);
1188         err = f2fs_get_dnode_of_data(&dn, bidx, LOOKUP_NODE);
1189         if (err)
1190                 goto out;
1191
1192         if (unlikely(dn.data_blkaddr == NULL_ADDR)) {
1193                 ClearPageUptodate(page);
1194                 err = -ENOENT;
1195                 goto put_out;
1196         }
1197
1198         /*
1199          * don't cache encrypted data into meta inode until previous dirty
1200          * data were writebacked to avoid racing between GC and flush.
1201          */
1202         f2fs_wait_on_page_writeback(page, DATA, true, true);
1203
1204         f2fs_wait_on_block_writeback(inode, dn.data_blkaddr);
1205
1206         err = f2fs_get_node_info(fio.sbi, dn.nid, &ni);
1207         if (err)
1208                 goto put_out;
1209
1210         /* read page */
1211         fio.page = page;
1212         fio.new_blkaddr = fio.old_blkaddr = dn.data_blkaddr;
1213
1214         if (lfs_mode)
1215                 down_write(&fio.sbi->io_order_lock);
1216
1217         mpage = f2fs_grab_cache_page(META_MAPPING(fio.sbi),
1218                                         fio.old_blkaddr, false);
1219         if (!mpage) {
1220                 err = -ENOMEM;
1221                 goto up_out;
1222         }
1223
1224         fio.encrypted_page = mpage;
1225
1226         /* read source block in mpage */
1227         if (!PageUptodate(mpage)) {
1228                 err = f2fs_submit_page_bio(&fio);
1229                 if (err) {
1230                         f2fs_put_page(mpage, 1);
1231                         goto up_out;
1232                 }
1233
1234                 f2fs_update_iostat(fio.sbi, FS_DATA_READ_IO, F2FS_BLKSIZE);
1235                 f2fs_update_iostat(fio.sbi, FS_GDATA_READ_IO, F2FS_BLKSIZE);
1236
1237                 lock_page(mpage);
1238                 if (unlikely(mpage->mapping != META_MAPPING(fio.sbi) ||
1239                                                 !PageUptodate(mpage))) {
1240                         err = -EIO;
1241                         f2fs_put_page(mpage, 1);
1242                         goto up_out;
1243                 }
1244         }
1245
1246         set_summary(&sum, dn.nid, dn.ofs_in_node, ni.version);
1247
1248         /* allocate block address */
1249         f2fs_allocate_data_block(fio.sbi, NULL, fio.old_blkaddr, &newaddr,
1250                                 &sum, type, NULL);
1251
1252         fio.encrypted_page = f2fs_pagecache_get_page(META_MAPPING(fio.sbi),
1253                                 newaddr, FGP_LOCK | FGP_CREAT, GFP_NOFS);
1254         if (!fio.encrypted_page) {
1255                 err = -ENOMEM;
1256                 f2fs_put_page(mpage, 1);
1257                 goto recover_block;
1258         }
1259
1260         /* write target block */
1261         f2fs_wait_on_page_writeback(fio.encrypted_page, DATA, true, true);
1262         memcpy(page_address(fio.encrypted_page),
1263                                 page_address(mpage), PAGE_SIZE);
1264         f2fs_put_page(mpage, 1);
1265         invalidate_mapping_pages(META_MAPPING(fio.sbi),
1266                                 fio.old_blkaddr, fio.old_blkaddr);
1267         f2fs_invalidate_compress_page(fio.sbi, fio.old_blkaddr);
1268
1269         set_page_dirty(fio.encrypted_page);
1270         if (clear_page_dirty_for_io(fio.encrypted_page))
1271                 dec_page_count(fio.sbi, F2FS_DIRTY_META);
1272
1273         set_page_writeback(fio.encrypted_page);
1274         ClearPageError(page);
1275
1276         fio.op = REQ_OP_WRITE;
1277         fio.op_flags = REQ_SYNC;
1278         fio.new_blkaddr = newaddr;
1279         f2fs_submit_page_write(&fio);
1280         if (fio.retry) {
1281                 err = -EAGAIN;
1282                 if (PageWriteback(fio.encrypted_page))
1283                         end_page_writeback(fio.encrypted_page);
1284                 goto put_page_out;
1285         }
1286
1287         f2fs_update_iostat(fio.sbi, FS_GC_DATA_IO, F2FS_BLKSIZE);
1288
1289         f2fs_update_data_blkaddr(&dn, newaddr);
1290         set_inode_flag(inode, FI_APPEND_WRITE);
1291         if (page->index == 0)
1292                 set_inode_flag(inode, FI_FIRST_BLOCK_WRITTEN);
1293 put_page_out:
1294         f2fs_put_page(fio.encrypted_page, 1);
1295 recover_block:
1296         if (err)
1297                 f2fs_do_replace_block(fio.sbi, &sum, newaddr, fio.old_blkaddr,
1298                                                         true, true, true);
1299 up_out:
1300         if (lfs_mode)
1301                 up_write(&fio.sbi->io_order_lock);
1302 put_out:
1303         f2fs_put_dnode(&dn);
1304 out:
1305         f2fs_put_page(page, 1);
1306         return err;
1307 }
1308
1309 static int move_data_page(struct inode *inode, block_t bidx, int gc_type,
1310                                                         unsigned int segno, int off)
1311 {
1312         struct page *page;
1313         int err = 0;
1314
1315         page = f2fs_get_lock_data_page(inode, bidx, true);
1316         if (IS_ERR(page))
1317                 return PTR_ERR(page);
1318
1319         if (!check_valid_map(F2FS_I_SB(inode), segno, off)) {
1320                 err = -ENOENT;
1321                 goto out;
1322         }
1323
1324         if (f2fs_is_atomic_file(inode)) {
1325                 F2FS_I(inode)->i_gc_failures[GC_FAILURE_ATOMIC]++;
1326                 F2FS_I_SB(inode)->skipped_atomic_files[gc_type]++;
1327                 err = -EAGAIN;
1328                 goto out;
1329         }
1330         if (f2fs_is_pinned_file(inode)) {
1331                 if (gc_type == FG_GC)
1332                         f2fs_pin_file_control(inode, true);
1333                 err = -EAGAIN;
1334                 goto out;
1335         }
1336
1337         if (gc_type == BG_GC) {
1338                 if (PageWriteback(page)) {
1339                         err = -EAGAIN;
1340                         goto out;
1341                 }
1342                 set_page_dirty(page);
1343                 set_page_private_gcing(page);
1344         } else {
1345                 struct f2fs_io_info fio = {
1346                         .sbi = F2FS_I_SB(inode),
1347                         .ino = inode->i_ino,
1348                         .type = DATA,
1349                         .temp = COLD,
1350                         .op = REQ_OP_WRITE,
1351                         .op_flags = REQ_SYNC,
1352                         .old_blkaddr = NULL_ADDR,
1353                         .page = page,
1354                         .encrypted_page = NULL,
1355                         .need_lock = LOCK_REQ,
1356                         .io_type = FS_GC_DATA_IO,
1357                 };
1358                 bool is_dirty = PageDirty(page);
1359
1360 retry:
1361                 f2fs_wait_on_page_writeback(page, DATA, true, true);
1362
1363                 set_page_dirty(page);
1364                 if (clear_page_dirty_for_io(page)) {
1365                         inode_dec_dirty_pages(inode);
1366                         f2fs_remove_dirty_inode(inode);
1367                 }
1368
1369                 set_page_private_gcing(page);
1370
1371                 err = f2fs_do_write_data_page(&fio);
1372                 if (err) {
1373                         clear_page_private_gcing(page);
1374                         if (err == -ENOMEM) {
1375                                 congestion_wait(BLK_RW_ASYNC,
1376                                                 DEFAULT_IO_TIMEOUT);
1377                                 goto retry;
1378                         }
1379                         if (is_dirty)
1380                                 set_page_dirty(page);
1381                 }
1382         }
1383 out:
1384         f2fs_put_page(page, 1);
1385         return err;
1386 }
1387
1388 /*
1389  * This function tries to get parent node of victim data block, and identifies
1390  * data block validity. If the block is valid, copy that with cold status and
1391  * modify parent node.
1392  * If the parent node is not valid or the data block address is different,
1393  * the victim data block is ignored.
1394  */
1395 static int gc_data_segment(struct f2fs_sb_info *sbi, struct f2fs_summary *sum,
1396                 struct gc_inode_list *gc_list, unsigned int segno, int gc_type,
1397                 bool force_migrate)
1398 {
1399         struct super_block *sb = sbi->sb;
1400         struct f2fs_summary *entry;
1401         block_t start_addr;
1402         int off;
1403         int phase = 0;
1404         int submitted = 0;
1405         unsigned int usable_blks_in_seg = f2fs_usable_blks_in_seg(sbi, segno);
1406
1407         start_addr = START_BLOCK(sbi, segno);
1408
1409 next_step:
1410         entry = sum;
1411
1412         for (off = 0; off < usable_blks_in_seg; off++, entry++) {
1413                 struct page *data_page;
1414                 struct inode *inode;
1415                 struct node_info dni; /* dnode info for the data */
1416                 unsigned int ofs_in_node, nofs;
1417                 block_t start_bidx;
1418                 nid_t nid = le32_to_cpu(entry->nid);
1419
1420                 /*
1421                  * stop BG_GC if there is not enough free sections.
1422                  * Or, stop GC if the segment becomes fully valid caused by
1423                  * race condition along with SSR block allocation.
1424                  */
1425                 if ((gc_type == BG_GC && has_not_enough_free_secs(sbi, 0, 0)) ||
1426                         (!force_migrate && get_valid_blocks(sbi, segno, true) ==
1427                                                         BLKS_PER_SEC(sbi)))
1428                         return submitted;
1429
1430                 if (check_valid_map(sbi, segno, off) == 0)
1431                         continue;
1432
1433                 if (phase == 0) {
1434                         f2fs_ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nid), 1,
1435                                                         META_NAT, true);
1436                         continue;
1437                 }
1438
1439                 if (phase == 1) {
1440                         f2fs_ra_node_page(sbi, nid);
1441                         continue;
1442                 }
1443
1444                 /* Get an inode by ino with checking validity */
1445                 if (!is_alive(sbi, entry, &dni, start_addr + off, &nofs))
1446                         continue;
1447
1448                 if (phase == 2) {
1449                         f2fs_ra_node_page(sbi, dni.ino);
1450                         continue;
1451                 }
1452
1453                 ofs_in_node = le16_to_cpu(entry->ofs_in_node);
1454
1455                 if (phase == 3) {
1456                         inode = f2fs_iget(sb, dni.ino);
1457                         if (IS_ERR(inode) || is_bad_inode(inode))
1458                                 continue;
1459
1460                         if (!down_write_trylock(
1461                                 &F2FS_I(inode)->i_gc_rwsem[WRITE])) {
1462                                 iput(inode);
1463                                 sbi->skipped_gc_rwsem++;
1464                                 continue;
1465                         }
1466
1467                         start_bidx = f2fs_start_bidx_of_node(nofs, inode) +
1468                                                                 ofs_in_node;
1469
1470                         if (f2fs_post_read_required(inode)) {
1471                                 int err = ra_data_block(inode, start_bidx);
1472
1473                                 up_write(&F2FS_I(inode)->i_gc_rwsem[WRITE]);
1474                                 if (err) {
1475                                         iput(inode);
1476                                         continue;
1477                                 }
1478                                 add_gc_inode(gc_list, inode);
1479                                 continue;
1480                         }
1481
1482                         data_page = f2fs_get_read_data_page(inode,
1483                                                 start_bidx, REQ_RAHEAD, true);
1484                         up_write(&F2FS_I(inode)->i_gc_rwsem[WRITE]);
1485                         if (IS_ERR(data_page)) {
1486                                 iput(inode);
1487                                 continue;
1488                         }
1489
1490                         f2fs_put_page(data_page, 0);
1491                         add_gc_inode(gc_list, inode);
1492                         continue;
1493                 }
1494
1495                 /* phase 4 */
1496                 inode = find_gc_inode(gc_list, dni.ino);
1497                 if (inode) {
1498                         struct f2fs_inode_info *fi = F2FS_I(inode);
1499                         bool locked = false;
1500                         int err;
1501
1502                         if (S_ISREG(inode->i_mode)) {
1503                                 if (!down_write_trylock(&fi->i_gc_rwsem[READ])) {
1504                                         sbi->skipped_gc_rwsem++;
1505                                         continue;
1506                                 }
1507                                 if (!down_write_trylock(
1508                                                 &fi->i_gc_rwsem[WRITE])) {
1509                                         sbi->skipped_gc_rwsem++;
1510                                         up_write(&fi->i_gc_rwsem[READ]);
1511                                         continue;
1512                                 }
1513                                 locked = true;
1514
1515                                 /* wait for all inflight aio data */
1516                                 inode_dio_wait(inode);
1517                         }
1518
1519                         start_bidx = f2fs_start_bidx_of_node(nofs, inode)
1520                                                                 + ofs_in_node;
1521                         if (f2fs_post_read_required(inode))
1522                                 err = move_data_block(inode, start_bidx,
1523                                                         gc_type, segno, off);
1524                         else
1525                                 err = move_data_page(inode, start_bidx, gc_type,
1526                                                                 segno, off);
1527
1528                         if (!err && (gc_type == FG_GC ||
1529                                         f2fs_post_read_required(inode)))
1530                                 submitted++;
1531
1532                         if (locked) {
1533                                 up_write(&fi->i_gc_rwsem[WRITE]);
1534                                 up_write(&fi->i_gc_rwsem[READ]);
1535                         }
1536
1537                         stat_inc_data_blk_count(sbi, 1, gc_type);
1538                 }
1539         }
1540
1541         if (++phase < 5)
1542                 goto next_step;
1543
1544         return submitted;
1545 }
1546
1547 static int __get_victim(struct f2fs_sb_info *sbi, unsigned int *victim,
1548                         int gc_type)
1549 {
1550         struct sit_info *sit_i = SIT_I(sbi);
1551         int ret;
1552
1553         down_write(&sit_i->sentry_lock);
1554         ret = DIRTY_I(sbi)->v_ops->get_victim(sbi, victim, gc_type,
1555                                               NO_CHECK_TYPE, LFS, 0);
1556         up_write(&sit_i->sentry_lock);
1557         return ret;
1558 }
1559
1560 static int do_garbage_collect(struct f2fs_sb_info *sbi,
1561                                 unsigned int start_segno,
1562                                 struct gc_inode_list *gc_list, int gc_type,
1563                                 bool force_migrate)
1564 {
1565         struct page *sum_page;
1566         struct f2fs_summary_block *sum;
1567         struct blk_plug plug;
1568         unsigned int segno = start_segno;
1569         unsigned int end_segno = start_segno + sbi->segs_per_sec;
1570         int seg_freed = 0, migrated = 0;
1571         unsigned char type = IS_DATASEG(get_seg_entry(sbi, segno)->type) ?
1572                                                 SUM_TYPE_DATA : SUM_TYPE_NODE;
1573         int submitted = 0;
1574
1575         if (__is_large_section(sbi))
1576                 end_segno = rounddown(end_segno, sbi->segs_per_sec);
1577
1578         /*
1579          * zone-capacity can be less than zone-size in zoned devices,
1580          * resulting in less than expected usable segments in the zone,
1581          * calculate the end segno in the zone which can be garbage collected
1582          */
1583         if (f2fs_sb_has_blkzoned(sbi))
1584                 end_segno -= sbi->segs_per_sec -
1585                                         f2fs_usable_segs_in_sec(sbi, segno);
1586
1587         sanity_check_seg_type(sbi, get_seg_entry(sbi, segno)->type);
1588
1589         /* readahead multi ssa blocks those have contiguous address */
1590         if (__is_large_section(sbi))
1591                 f2fs_ra_meta_pages(sbi, GET_SUM_BLOCK(sbi, segno),
1592                                         end_segno - segno, META_SSA, true);
1593
1594         /* reference all summary page */
1595         while (segno < end_segno) {
1596                 sum_page = f2fs_get_sum_page(sbi, segno++);
1597                 if (IS_ERR(sum_page)) {
1598                         int err = PTR_ERR(sum_page);
1599
1600                         end_segno = segno - 1;
1601                         for (segno = start_segno; segno < end_segno; segno++) {
1602                                 sum_page = find_get_page(META_MAPPING(sbi),
1603                                                 GET_SUM_BLOCK(sbi, segno));
1604                                 f2fs_put_page(sum_page, 0);
1605                                 f2fs_put_page(sum_page, 0);
1606                         }
1607                         return err;
1608                 }
1609                 unlock_page(sum_page);
1610         }
1611
1612         blk_start_plug(&plug);
1613
1614         for (segno = start_segno; segno < end_segno; segno++) {
1615
1616                 /* find segment summary of victim */
1617                 sum_page = find_get_page(META_MAPPING(sbi),
1618                                         GET_SUM_BLOCK(sbi, segno));
1619                 f2fs_put_page(sum_page, 0);
1620
1621                 if (get_valid_blocks(sbi, segno, false) == 0)
1622                         goto freed;
1623                 if (gc_type == BG_GC && __is_large_section(sbi) &&
1624                                 migrated >= sbi->migration_granularity)
1625                         goto skip;
1626                 if (!PageUptodate(sum_page) || unlikely(f2fs_cp_error(sbi)))
1627                         goto skip;
1628
1629                 sum = page_address(sum_page);
1630                 if (type != GET_SUM_TYPE((&sum->footer))) {
1631                         f2fs_err(sbi, "Inconsistent segment (%u) type [%d, %d] in SSA and SIT",
1632                                  segno, type, GET_SUM_TYPE((&sum->footer)));
1633                         set_sbi_flag(sbi, SBI_NEED_FSCK);
1634                         f2fs_stop_checkpoint(sbi, false);
1635                         goto skip;
1636                 }
1637
1638                 /*
1639                  * this is to avoid deadlock:
1640                  * - lock_page(sum_page)         - f2fs_replace_block
1641                  *  - check_valid_map()            - down_write(sentry_lock)
1642                  *   - down_read(sentry_lock)     - change_curseg()
1643                  *                                  - lock_page(sum_page)
1644                  */
1645                 if (type == SUM_TYPE_NODE)
1646                         submitted += gc_node_segment(sbi, sum->entries, segno,
1647                                                                 gc_type);
1648                 else
1649                         submitted += gc_data_segment(sbi, sum->entries, gc_list,
1650                                                         segno, gc_type,
1651                                                         force_migrate);
1652
1653                 stat_inc_seg_count(sbi, type, gc_type);
1654                 sbi->gc_reclaimed_segs[sbi->gc_mode]++;
1655                 migrated++;
1656
1657 freed:
1658                 if (gc_type == FG_GC &&
1659                                 get_valid_blocks(sbi, segno, false) == 0)
1660                         seg_freed++;
1661
1662                 if (__is_large_section(sbi) && segno + 1 < end_segno)
1663                         sbi->next_victim_seg[gc_type] = segno + 1;
1664 skip:
1665                 f2fs_put_page(sum_page, 0);
1666         }
1667
1668         if (submitted)
1669                 f2fs_submit_merged_write(sbi,
1670                                 (type == SUM_TYPE_NODE) ? NODE : DATA);
1671
1672         blk_finish_plug(&plug);
1673
1674         stat_inc_call_count(sbi->stat_info);
1675
1676         return seg_freed;
1677 }
1678
1679 int f2fs_gc(struct f2fs_sb_info *sbi, bool sync,
1680                         bool background, bool force, unsigned int segno)
1681 {
1682         int gc_type = sync ? FG_GC : BG_GC;
1683         int sec_freed = 0, seg_freed = 0, total_freed = 0;
1684         int ret = 0;
1685         struct cp_control cpc;
1686         unsigned int init_segno = segno;
1687         struct gc_inode_list gc_list = {
1688                 .ilist = LIST_HEAD_INIT(gc_list.ilist),
1689                 .iroot = RADIX_TREE_INIT(gc_list.iroot, GFP_NOFS),
1690         };
1691         unsigned long long last_skipped = sbi->skipped_atomic_files[FG_GC];
1692         unsigned long long first_skipped;
1693         unsigned int skipped_round = 0, round = 0;
1694
1695         trace_f2fs_gc_begin(sbi->sb, sync, background,
1696                                 get_pages(sbi, F2FS_DIRTY_NODES),
1697                                 get_pages(sbi, F2FS_DIRTY_DENTS),
1698                                 get_pages(sbi, F2FS_DIRTY_IMETA),
1699                                 free_sections(sbi),
1700                                 free_segments(sbi),
1701                                 reserved_segments(sbi),
1702                                 prefree_segments(sbi));
1703
1704         cpc.reason = __get_cp_reason(sbi);
1705         sbi->skipped_gc_rwsem = 0;
1706         first_skipped = last_skipped;
1707 gc_more:
1708         if (unlikely(!(sbi->sb->s_flags & SB_ACTIVE))) {
1709                 ret = -EINVAL;
1710                 goto stop;
1711         }
1712         if (unlikely(f2fs_cp_error(sbi))) {
1713                 ret = -EIO;
1714                 goto stop;
1715         }
1716
1717         if (gc_type == BG_GC && has_not_enough_free_secs(sbi, 0, 0)) {
1718                 /*
1719                  * For example, if there are many prefree_segments below given
1720                  * threshold, we can make them free by checkpoint. Then, we
1721                  * secure free segments which doesn't need fggc any more.
1722                  */
1723                 if (prefree_segments(sbi) &&
1724                                 !is_sbi_flag_set(sbi, SBI_CP_DISABLED)) {
1725                         ret = f2fs_write_checkpoint(sbi, &cpc);
1726                         if (ret)
1727                                 goto stop;
1728                 }
1729                 if (has_not_enough_free_secs(sbi, 0, 0))
1730                         gc_type = FG_GC;
1731         }
1732
1733         /* f2fs_balance_fs doesn't need to do BG_GC in critical path. */
1734         if (gc_type == BG_GC && !background) {
1735                 ret = -EINVAL;
1736                 goto stop;
1737         }
1738         ret = __get_victim(sbi, &segno, gc_type);
1739         if (ret)
1740                 goto stop;
1741
1742         seg_freed = do_garbage_collect(sbi, segno, &gc_list, gc_type, force);
1743         if (gc_type == FG_GC &&
1744                 seg_freed == f2fs_usable_segs_in_sec(sbi, segno))
1745                 sec_freed++;
1746         total_freed += seg_freed;
1747
1748         if (gc_type == FG_GC) {
1749                 if (sbi->skipped_atomic_files[FG_GC] > last_skipped ||
1750                                                 sbi->skipped_gc_rwsem)
1751                         skipped_round++;
1752                 last_skipped = sbi->skipped_atomic_files[FG_GC];
1753                 round++;
1754         }
1755
1756         if (gc_type == FG_GC)
1757                 sbi->cur_victim_sec = NULL_SEGNO;
1758
1759         if (sync)
1760                 goto stop;
1761
1762         if (has_not_enough_free_secs(sbi, sec_freed, 0)) {
1763                 if (skipped_round <= MAX_SKIP_GC_COUNT ||
1764                                         skipped_round * 2 < round) {
1765                         segno = NULL_SEGNO;
1766                         goto gc_more;
1767                 }
1768
1769                 if (first_skipped < last_skipped &&
1770                                 (last_skipped - first_skipped) >
1771                                                 sbi->skipped_gc_rwsem) {
1772                         f2fs_drop_inmem_pages_all(sbi, true);
1773                         segno = NULL_SEGNO;
1774                         goto gc_more;
1775                 }
1776                 if (gc_type == FG_GC && !is_sbi_flag_set(sbi, SBI_CP_DISABLED))
1777                         ret = f2fs_write_checkpoint(sbi, &cpc);
1778         }
1779 stop:
1780         SIT_I(sbi)->last_victim[ALLOC_NEXT] = 0;
1781         SIT_I(sbi)->last_victim[FLUSH_DEVICE] = init_segno;
1782
1783         trace_f2fs_gc_end(sbi->sb, ret, total_freed, sec_freed,
1784                                 get_pages(sbi, F2FS_DIRTY_NODES),
1785                                 get_pages(sbi, F2FS_DIRTY_DENTS),
1786                                 get_pages(sbi, F2FS_DIRTY_IMETA),
1787                                 free_sections(sbi),
1788                                 free_segments(sbi),
1789                                 reserved_segments(sbi),
1790                                 prefree_segments(sbi));
1791
1792         up_write(&sbi->gc_lock);
1793
1794         put_gc_inode(&gc_list);
1795
1796         if (sync && !ret)
1797                 ret = sec_freed ? 0 : -EAGAIN;
1798         return ret;
1799 }
1800
1801 int __init f2fs_create_garbage_collection_cache(void)
1802 {
1803         victim_entry_slab = f2fs_kmem_cache_create("f2fs_victim_entry",
1804                                         sizeof(struct victim_entry));
1805         if (!victim_entry_slab)
1806                 return -ENOMEM;
1807         return 0;
1808 }
1809
1810 void f2fs_destroy_garbage_collection_cache(void)
1811 {
1812         kmem_cache_destroy(victim_entry_slab);
1813 }
1814
1815 static void init_atgc_management(struct f2fs_sb_info *sbi)
1816 {
1817         struct atgc_management *am = &sbi->am;
1818
1819         if (test_opt(sbi, ATGC) &&
1820                 SIT_I(sbi)->elapsed_time >= DEF_GC_THREAD_AGE_THRESHOLD)
1821                 am->atgc_enabled = true;
1822
1823         am->root = RB_ROOT_CACHED;
1824         INIT_LIST_HEAD(&am->victim_list);
1825         am->victim_count = 0;
1826
1827         am->candidate_ratio = DEF_GC_THREAD_CANDIDATE_RATIO;
1828         am->max_candidate_count = DEF_GC_THREAD_MAX_CANDIDATE_COUNT;
1829         am->age_weight = DEF_GC_THREAD_AGE_WEIGHT;
1830         am->age_threshold = DEF_GC_THREAD_AGE_THRESHOLD;
1831 }
1832
1833 void f2fs_build_gc_manager(struct f2fs_sb_info *sbi)
1834 {
1835         DIRTY_I(sbi)->v_ops = &default_v_ops;
1836
1837         sbi->gc_pin_file_threshold = DEF_GC_FAILED_PINNED_FILES;
1838
1839         /* give warm/cold data area from slower device */
1840         if (f2fs_is_multi_device(sbi) && !__is_large_section(sbi))
1841                 SIT_I(sbi)->last_victim[ALLOC_NEXT] =
1842                                 GET_SEGNO(sbi, FDEV(0).end_blk) + 1;
1843
1844         init_atgc_management(sbi);
1845 }
1846
1847 static int free_segment_range(struct f2fs_sb_info *sbi,
1848                                 unsigned int secs, bool gc_only)
1849 {
1850         unsigned int segno, next_inuse, start, end;
1851         struct cp_control cpc = { CP_RESIZE, 0, 0, 0 };
1852         int gc_mode, gc_type;
1853         int err = 0;
1854         int type;
1855
1856         /* Force block allocation for GC */
1857         MAIN_SECS(sbi) -= secs;
1858         start = MAIN_SECS(sbi) * sbi->segs_per_sec;
1859         end = MAIN_SEGS(sbi) - 1;
1860
1861         mutex_lock(&DIRTY_I(sbi)->seglist_lock);
1862         for (gc_mode = 0; gc_mode < MAX_GC_POLICY; gc_mode++)
1863                 if (SIT_I(sbi)->last_victim[gc_mode] >= start)
1864                         SIT_I(sbi)->last_victim[gc_mode] = 0;
1865
1866         for (gc_type = BG_GC; gc_type <= FG_GC; gc_type++)
1867                 if (sbi->next_victim_seg[gc_type] >= start)
1868                         sbi->next_victim_seg[gc_type] = NULL_SEGNO;
1869         mutex_unlock(&DIRTY_I(sbi)->seglist_lock);
1870
1871         /* Move out cursegs from the target range */
1872         for (type = CURSEG_HOT_DATA; type < NR_CURSEG_PERSIST_TYPE; type++)
1873                 f2fs_allocate_segment_for_resize(sbi, type, start, end);
1874
1875         /* do GC to move out valid blocks in the range */
1876         for (segno = start; segno <= end; segno += sbi->segs_per_sec) {
1877                 struct gc_inode_list gc_list = {
1878                         .ilist = LIST_HEAD_INIT(gc_list.ilist),
1879                         .iroot = RADIX_TREE_INIT(gc_list.iroot, GFP_NOFS),
1880                 };
1881
1882                 do_garbage_collect(sbi, segno, &gc_list, FG_GC, true);
1883                 put_gc_inode(&gc_list);
1884
1885                 if (!gc_only && get_valid_blocks(sbi, segno, true)) {
1886                         err = -EAGAIN;
1887                         goto out;
1888                 }
1889                 if (fatal_signal_pending(current)) {
1890                         err = -ERESTARTSYS;
1891                         goto out;
1892                 }
1893         }
1894         if (gc_only)
1895                 goto out;
1896
1897         err = f2fs_write_checkpoint(sbi, &cpc);
1898         if (err)
1899                 goto out;
1900
1901         next_inuse = find_next_inuse(FREE_I(sbi), end + 1, start);
1902         if (next_inuse <= end) {
1903                 f2fs_err(sbi, "segno %u should be free but still inuse!",
1904                          next_inuse);
1905                 f2fs_bug_on(sbi, 1);
1906         }
1907 out:
1908         MAIN_SECS(sbi) += secs;
1909         return err;
1910 }
1911
1912 static void update_sb_metadata(struct f2fs_sb_info *sbi, int secs)
1913 {
1914         struct f2fs_super_block *raw_sb = F2FS_RAW_SUPER(sbi);
1915         int section_count;
1916         int segment_count;
1917         int segment_count_main;
1918         long long block_count;
1919         int segs = secs * sbi->segs_per_sec;
1920
1921         down_write(&sbi->sb_lock);
1922
1923         section_count = le32_to_cpu(raw_sb->section_count);
1924         segment_count = le32_to_cpu(raw_sb->segment_count);
1925         segment_count_main = le32_to_cpu(raw_sb->segment_count_main);
1926         block_count = le64_to_cpu(raw_sb->block_count);
1927
1928         raw_sb->section_count = cpu_to_le32(section_count + secs);
1929         raw_sb->segment_count = cpu_to_le32(segment_count + segs);
1930         raw_sb->segment_count_main = cpu_to_le32(segment_count_main + segs);
1931         raw_sb->block_count = cpu_to_le64(block_count +
1932                                         (long long)segs * sbi->blocks_per_seg);
1933         if (f2fs_is_multi_device(sbi)) {
1934                 int last_dev = sbi->s_ndevs - 1;
1935                 int dev_segs =
1936                         le32_to_cpu(raw_sb->devs[last_dev].total_segments);
1937
1938                 raw_sb->devs[last_dev].total_segments =
1939                                                 cpu_to_le32(dev_segs + segs);
1940         }
1941
1942         up_write(&sbi->sb_lock);
1943 }
1944
1945 static void update_fs_metadata(struct f2fs_sb_info *sbi, int secs)
1946 {
1947         int segs = secs * sbi->segs_per_sec;
1948         long long blks = (long long)segs * sbi->blocks_per_seg;
1949         long long user_block_count =
1950                                 le64_to_cpu(F2FS_CKPT(sbi)->user_block_count);
1951
1952         SM_I(sbi)->segment_count = (int)SM_I(sbi)->segment_count + segs;
1953         MAIN_SEGS(sbi) = (int)MAIN_SEGS(sbi) + segs;
1954         MAIN_SECS(sbi) += secs;
1955         FREE_I(sbi)->free_sections = (int)FREE_I(sbi)->free_sections + secs;
1956         FREE_I(sbi)->free_segments = (int)FREE_I(sbi)->free_segments + segs;
1957         F2FS_CKPT(sbi)->user_block_count = cpu_to_le64(user_block_count + blks);
1958
1959         if (f2fs_is_multi_device(sbi)) {
1960                 int last_dev = sbi->s_ndevs - 1;
1961
1962                 FDEV(last_dev).total_segments =
1963                                 (int)FDEV(last_dev).total_segments + segs;
1964                 FDEV(last_dev).end_blk =
1965                                 (long long)FDEV(last_dev).end_blk + blks;
1966 #ifdef CONFIG_BLK_DEV_ZONED
1967                 FDEV(last_dev).nr_blkz = (int)FDEV(last_dev).nr_blkz +
1968                                         (int)(blks >> sbi->log_blocks_per_blkz);
1969 #endif
1970         }
1971 }
1972
1973 int f2fs_resize_fs(struct f2fs_sb_info *sbi, __u64 block_count)
1974 {
1975         __u64 old_block_count, shrunk_blocks;
1976         struct cp_control cpc = { CP_RESIZE, 0, 0, 0 };
1977         unsigned int secs;
1978         int err = 0;
1979         __u32 rem;
1980
1981         old_block_count = le64_to_cpu(F2FS_RAW_SUPER(sbi)->block_count);
1982         if (block_count > old_block_count)
1983                 return -EINVAL;
1984
1985         if (f2fs_is_multi_device(sbi)) {
1986                 int last_dev = sbi->s_ndevs - 1;
1987                 __u64 last_segs = FDEV(last_dev).total_segments;
1988
1989                 if (block_count + last_segs * sbi->blocks_per_seg <=
1990                                                                 old_block_count)
1991                         return -EINVAL;
1992         }
1993
1994         /* new fs size should align to section size */
1995         div_u64_rem(block_count, BLKS_PER_SEC(sbi), &rem);
1996         if (rem)
1997                 return -EINVAL;
1998
1999         if (block_count == old_block_count)
2000                 return 0;
2001
2002         if (is_sbi_flag_set(sbi, SBI_NEED_FSCK)) {
2003                 f2fs_err(sbi, "Should run fsck to repair first.");
2004                 return -EFSCORRUPTED;
2005         }
2006
2007         if (test_opt(sbi, DISABLE_CHECKPOINT)) {
2008                 f2fs_err(sbi, "Checkpoint should be enabled.");
2009                 return -EINVAL;
2010         }
2011
2012         shrunk_blocks = old_block_count - block_count;
2013         secs = div_u64(shrunk_blocks, BLKS_PER_SEC(sbi));
2014
2015         /* stop other GC */
2016         if (!down_write_trylock(&sbi->gc_lock))
2017                 return -EAGAIN;
2018
2019         /* stop CP to protect MAIN_SEC in free_segment_range */
2020         f2fs_lock_op(sbi);
2021
2022         spin_lock(&sbi->stat_lock);
2023         if (shrunk_blocks + valid_user_blocks(sbi) +
2024                 sbi->current_reserved_blocks + sbi->unusable_block_count +
2025                 F2FS_OPTION(sbi).root_reserved_blocks > sbi->user_block_count)
2026                 err = -ENOSPC;
2027         spin_unlock(&sbi->stat_lock);
2028
2029         if (err)
2030                 goto out_unlock;
2031
2032         err = free_segment_range(sbi, secs, true);
2033
2034 out_unlock:
2035         f2fs_unlock_op(sbi);
2036         up_write(&sbi->gc_lock);
2037         if (err)
2038                 return err;
2039
2040         set_sbi_flag(sbi, SBI_IS_RESIZEFS);
2041
2042         freeze_super(sbi->sb);
2043         down_write(&sbi->gc_lock);
2044         down_write(&sbi->cp_global_sem);
2045
2046         spin_lock(&sbi->stat_lock);
2047         if (shrunk_blocks + valid_user_blocks(sbi) +
2048                 sbi->current_reserved_blocks + sbi->unusable_block_count +
2049                 F2FS_OPTION(sbi).root_reserved_blocks > sbi->user_block_count)
2050                 err = -ENOSPC;
2051         else
2052                 sbi->user_block_count -= shrunk_blocks;
2053         spin_unlock(&sbi->stat_lock);
2054         if (err)
2055                 goto out_err;
2056
2057         err = free_segment_range(sbi, secs, false);
2058         if (err)
2059                 goto recover_out;
2060
2061         update_sb_metadata(sbi, -secs);
2062
2063         err = f2fs_commit_super(sbi, false);
2064         if (err) {
2065                 update_sb_metadata(sbi, secs);
2066                 goto recover_out;
2067         }
2068
2069         update_fs_metadata(sbi, -secs);
2070         clear_sbi_flag(sbi, SBI_IS_RESIZEFS);
2071         set_sbi_flag(sbi, SBI_IS_DIRTY);
2072
2073         err = f2fs_write_checkpoint(sbi, &cpc);
2074         if (err) {
2075                 update_fs_metadata(sbi, secs);
2076                 update_sb_metadata(sbi, secs);
2077                 f2fs_commit_super(sbi, false);
2078         }
2079 recover_out:
2080         if (err) {
2081                 set_sbi_flag(sbi, SBI_NEED_FSCK);
2082                 f2fs_err(sbi, "resize_fs failed, should run fsck to repair!");
2083
2084                 spin_lock(&sbi->stat_lock);
2085                 sbi->user_block_count += shrunk_blocks;
2086                 spin_unlock(&sbi->stat_lock);
2087         }
2088 out_err:
2089         up_write(&sbi->cp_global_sem);
2090         up_write(&sbi->gc_lock);
2091         thaw_super(sbi->sb);
2092         clear_sbi_flag(sbi, SBI_IS_RESIZEFS);
2093         return err;
2094 }