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/mount.h>
  11 #include <linux/backing-dev.h>
  12 #include <linux/init.h>
  13 #include <linux/f2fs_fs.h>
  14 #include <linux/kthread.h>
  15 #include <linux/delay.h>
  16 #include <linux/freezer.h>
  17 #include <linux/sched/signal.h>
  18
  19 #include "f2fs.h"
  20 #include "node.h"
  21 #include "segment.h"
  22 #include "gc.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         unsigned int wait_ms;
  36
  37         wait_ms = gc_th->min_sleep_time;
  38
  39         set_freezable();
  40         do {
  41                 bool sync_mode;
  42
  43                 wait_event_interruptible_timeout(*wq,
  44                                 kthread_should_stop() || freezing(current) ||
  45                                 gc_th->gc_wake,
  46                                 msecs_to_jiffies(wait_ms));
  47
  48                 /* give it a try one time */
  49                 if (gc_th->gc_wake)
  50                         gc_th->gc_wake = 0;
  51
  52                 if (try_to_freeze()) {
  53                         stat_other_skip_bggc_count(sbi);
  54                         continue;
  55                 }
  56                 if (kthread_should_stop())
  57                         break;
  58
  59                 if (sbi->sb->s_writers.frozen >= SB_FREEZE_WRITE) {
  60                         increase_sleep_time(gc_th, &wait_ms);
  61                         stat_other_skip_bggc_count(sbi);
  62                         continue;
  63                 }
  64
  65                 if (time_to_inject(sbi, FAULT_CHECKPOINT)) {
  66                         f2fs_show_injection_info(sbi, FAULT_CHECKPOINT);
  67                         f2fs_stop_checkpoint(sbi, false);
  68                 }
  69
  70                 if (!sb_start_write_trylock(sbi->sb)) {
  71                         stat_other_skip_bggc_count(sbi);
  72                         continue;
  73                 }
  74
  75                 /*
  76                  * [GC triggering condition]
  77                  * 0. GC is not conducted currently.
  78                  * 1. There are enough dirty segments.
  79                  * 2. IO subsystem is idle by checking the # of writeback pages.
  80                  * 3. IO subsystem is idle by checking the # of requests in
  81                  *    bdev's request list.
  82                  *
  83                  * Note) We have to avoid triggering GCs frequently.
  84                  * Because it is possible that some segments can be
  85                  * invalidated soon after by user update or deletion.
  86                  * So, I'd like to wait some time to collect dirty segments.
  87                  */
  88                 if (sbi->gc_mode == GC_URGENT_HIGH) {
  89                         wait_ms = gc_th->urgent_sleep_time;
  90                         down_write(&sbi->gc_lock);
  91                         goto do_gc;
  92                 }
  93
  94                 if (!down_write_trylock(&sbi->gc_lock)) {
  95                         stat_other_skip_bggc_count(sbi);
  96                         goto next;
  97                 }
  98
  99                 if (!is_idle(sbi, GC_TIME)) {
 100                         increase_sleep_time(gc_th, &wait_ms);
 101                         up_write(&sbi->gc_lock);
 102                         stat_io_skip_bggc_count(sbi);
 103                         goto next;
 104                 }
 105
 106                 if (has_enough_invalid_blocks(sbi))
 107                         decrease_sleep_time(gc_th, &wait_ms);
 108                 else
 109                         increase_sleep_time(gc_th, &wait_ms);
 110 do_gc:
 111                 stat_inc_bggc_count(sbi->stat_info);
 112
 113                 sync_mode = F2FS_OPTION(sbi).bggc_mode == BGGC_MODE_SYNC;
 114
 115                 /* if return value is not zero, no victim was selected */
 116                 if (f2fs_gc(sbi, sync_mode, true, false, NULL_SEGNO))
 117                         wait_ms = gc_th->no_gc_sleep_time;
 118
 119                 trace_f2fs_background_gc(sbi->sb, wait_ms,
 120                                 prefree_segments(sbi), free_segments(sbi));
 121
 122                 /* balancing f2fs's metadata periodically */
 123                 f2fs_balance_fs_bg(sbi, true);
 124 next:
 125                 sb_end_write(sbi->sb);
 126
 127         } while (!kthread_should_stop());
 128         return 0;
 129 }
 130
 131 int f2fs_start_gc_thread(struct f2fs_sb_info *sbi)
 132 {
 133         struct f2fs_gc_kthread *gc_th;
 134         dev_t dev = sbi->sb->s_bdev->bd_dev;
 135         int err = 0;
 136
 137         gc_th = f2fs_kmalloc(sbi, sizeof(struct f2fs_gc_kthread), GFP_KERNEL);
 138         if (!gc_th) {
 139                 err = -ENOMEM;
 140                 goto out;
 141         }
 142
 143         gc_th->urgent_sleep_time = DEF_GC_THREAD_URGENT_SLEEP_TIME;
 144         gc_th->min_sleep_time = DEF_GC_THREAD_MIN_SLEEP_TIME;
 145         gc_th->max_sleep_time = DEF_GC_THREAD_MAX_SLEEP_TIME;
 146         gc_th->no_gc_sleep_time = DEF_GC_THREAD_NOGC_SLEEP_TIME;
 147
 148         gc_th->gc_wake= 0;
 149
 150         sbi->gc_thread = gc_th;
 151         init_waitqueue_head(&sbi->gc_thread->gc_wait_queue_head);
 152         sbi->gc_thread->f2fs_gc_task = kthread_run(gc_thread_func, sbi,
 153                         "f2fs_gc-%u:%u", MAJOR(dev), MINOR(dev));
 154         if (IS_ERR(gc_th->f2fs_gc_task)) {
 155                 err = PTR_ERR(gc_th->f2fs_gc_task);
 156                 kfree(gc_th);
 157                 sbi->gc_thread = NULL;
 158         }
 159 out:
 160         return err;
 161 }
 162
 163 void f2fs_stop_gc_thread(struct f2fs_sb_info *sbi)
 164 {
 165         struct f2fs_gc_kthread *gc_th = sbi->gc_thread;
 166         if (!gc_th)
 167                 return;
 168         kthread_stop(gc_th->f2fs_gc_task);
 169         kfree(gc_th);
 170         sbi->gc_thread = NULL;
 171 }
 172
 173 static int select_gc_type(struct f2fs_sb_info *sbi, int gc_type)
 174 {
 175         int gc_mode;
 176
 177         if (gc_type == BG_GC) {
 178                 if (sbi->am.atgc_enabled)
 179                         gc_mode = GC_AT;
 180                 else
 181                         gc_mode = GC_CB;
 182         } else {
 183                 gc_mode = GC_GREEDY;
 184         }
 185
 186         switch (sbi->gc_mode) {
 187         case GC_IDLE_CB:
 188                 gc_mode = GC_CB;
 189                 break;
 190         case GC_IDLE_GREEDY:
 191         case GC_URGENT_HIGH:
 192                 gc_mode = GC_GREEDY;
 193                 break;
 194         case GC_IDLE_AT:
 195                 gc_mode = GC_AT;
 196                 break;
 197         }
 198
 199         return gc_mode;
 200 }
 201
 202 static void select_policy(struct f2fs_sb_info *sbi, int gc_type,
 203                         int type, struct victim_sel_policy *p)
 204 {
 205         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
 206
 207         if (p->alloc_mode == SSR) {
 208                 p->gc_mode = GC_GREEDY;
 209                 p->dirty_bitmap = dirty_i->dirty_segmap[type];
 210                 p->max_search = dirty_i->nr_dirty[type];
 211                 p->ofs_unit = 1;
 212         } else if (p->alloc_mode == AT_SSR) {
 213                 p->gc_mode = GC_GREEDY;
 214                 p->dirty_bitmap = dirty_i->dirty_segmap[type];
 215                 p->max_search = dirty_i->nr_dirty[type];
 216                 p->ofs_unit = 1;
 217         } else {
 218                 p->gc_mode = select_gc_type(sbi, gc_type);
 219                 p->ofs_unit = sbi->segs_per_sec;
 220                 if (__is_large_section(sbi)) {
 221                         p->dirty_bitmap = dirty_i->dirty_secmap;
 222                         p->max_search = count_bits(p->dirty_bitmap,
 223                                                 0, MAIN_SECS(sbi));
 224                 } else {
 225                         p->dirty_bitmap = dirty_i->dirty_segmap[DIRTY];
 226                         p->max_search = dirty_i->nr_dirty[DIRTY];
 227                 }
 228         }
 229
 230         /*
 231          * adjust candidates range, should select all dirty segments for
 232          * foreground GC and urgent GC cases.
 233          */
 234         if (gc_type != FG_GC &&
 235                         (sbi->gc_mode != GC_URGENT_HIGH) &&
 236                         (p->gc_mode != GC_AT && p->alloc_mode != AT_SSR) &&
 237                         p->max_search > sbi->max_victim_search)
 238                 p->max_search = sbi->max_victim_search;
 239
 240         /* let's select beginning hot/small space first in no_heap mode*/
 241         if (test_opt(sbi, NOHEAP) &&
 242                 (type == CURSEG_HOT_DATA || IS_NODESEG(type)))
 243                 p->offset = 0;
 244         else
 245                 p->offset = SIT_I(sbi)->last_victim[p->gc_mode];
 246 }
 247
 248 static unsigned int get_max_cost(struct f2fs_sb_info *sbi,
 249                                 struct victim_sel_policy *p)
 250 {
 251         /* SSR allocates in a segment unit */
 252         if (p->alloc_mode == SSR)
 253                 return sbi->blocks_per_seg;
 254         else if (p->alloc_mode == AT_SSR)
 255                 return UINT_MAX;
 256
 257         /* LFS */
 258         if (p->gc_mode == GC_GREEDY)
 259                 return 2 * sbi->blocks_per_seg * p->ofs_unit;
 260         else if (p->gc_mode == GC_CB)
 261                 return UINT_MAX;
 262         else if (p->gc_mode == GC_AT)
 263                 return UINT_MAX;
 264         else /* No other gc_mode */
 265                 return 0;
 266 }
 267
 268 static unsigned int check_bg_victims(struct f2fs_sb_info *sbi)
 269 {
 270         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
 271         unsigned int secno;
 272
 273         /*
 274          * If the gc_type is FG_GC, we can select victim segments
 275          * selected by background GC before.
 276          * Those segments guarantee they have small valid blocks.
 277          */
 278         for_each_set_bit(secno, dirty_i->victim_secmap, MAIN_SECS(sbi)) {
 279                 if (sec_usage_check(sbi, secno))
 280                         continue;
 281                 clear_bit(secno, dirty_i->victim_secmap);
 282                 return GET_SEG_FROM_SEC(sbi, secno);
 283         }
 284         return NULL_SEGNO;
 285 }
 286
 287 static unsigned int get_cb_cost(struct f2fs_sb_info *sbi, unsigned int segno)
 288 {
 289         struct sit_info *sit_i = SIT_I(sbi);
 290         unsigned int secno = GET_SEC_FROM_SEG(sbi, segno);
 291         unsigned int start = GET_SEG_FROM_SEC(sbi, secno);
 292         unsigned long long mtime = 0;
 293         unsigned int vblocks;
 294         unsigned char age = 0;
 295         unsigned char u;
 296         unsigned int i;
 297         unsigned int usable_segs_per_sec = f2fs_usable_segs_in_sec(sbi, segno);
 298
 299         for (i = 0; i < usable_segs_per_sec; i++)
 300                 mtime += get_seg_entry(sbi, start + i)->mtime;
 301         vblocks = get_valid_blocks(sbi, segno, true);
 302
 303         mtime = div_u64(mtime, usable_segs_per_sec);
 304         vblocks = div_u64(vblocks, usable_segs_per_sec);
 305
 306         u = (vblocks * 100) >> sbi->log_blocks_per_seg;
 307
 308         /* Handle if the system time has changed by the user */
 309         if (mtime < sit_i->min_mtime)
 310                 sit_i->min_mtime = mtime;
 311         if (mtime > sit_i->max_mtime)
 312                 sit_i->max_mtime = mtime;
 313         if (sit_i->max_mtime != sit_i->min_mtime)
 314                 age = 100 - div64_u64(100 * (mtime - sit_i->min_mtime),
 315                                 sit_i->max_mtime - sit_i->min_mtime);
 316
 317         return UINT_MAX - ((100 * (100 - u) * age) / (100 + u));
 318 }
 319
 320 static inline unsigned int get_gc_cost(struct f2fs_sb_info *sbi,
 321                         unsigned int segno, struct victim_sel_policy *p)
 322 {
 323         if (p->alloc_mode == SSR)
 324                 return get_seg_entry(sbi, segno)->ckpt_valid_blocks;
 325
 326         /* alloc_mode == LFS */
 327         if (p->gc_mode == GC_GREEDY)
 328                 return get_valid_blocks(sbi, segno, true);
 329         else if (p->gc_mode == GC_CB)
 330                 return get_cb_cost(sbi, segno);
 331
 332         f2fs_bug_on(sbi, 1);
 333         return 0;
 334 }
 335
 336 static unsigned int count_bits(const unsigned long *addr,
 337                                 unsigned int offset, unsigned int len)
 338 {
 339         unsigned int end = offset + len, sum = 0;
 340
 341         while (offset < end) {
 342                 if (test_bit(offset++, addr))
 343                         ++sum;
 344         }
 345         return sum;
 346 }
 347
 348 static struct victim_entry *attach_victim_entry(struct f2fs_sb_info *sbi,
 349                                 unsigned long long mtime, unsigned int segno,
 350                                 struct rb_node *parent, struct rb_node **p,
 351                                 bool left_most)
 352 {
 353         struct atgc_management *am = &sbi->am;
 354         struct victim_entry *ve;
 355
 356         ve =  f2fs_kmem_cache_alloc(victim_entry_slab, GFP_NOFS);
 357
 358         ve->mtime = mtime;
 359         ve->segno = segno;
 360
 361         rb_link_node(&ve->rb_node, parent, p);
 362         rb_insert_color_cached(&ve->rb_node, &am->root, left_most);
 363
 364         list_add_tail(&ve->list, &am->victim_list);
 365
 366         am->victim_count++;
 367
 368         return ve;
 369 }
 370
 371 static void insert_victim_entry(struct f2fs_sb_info *sbi,
 372                                 unsigned long long mtime, unsigned int segno)
 373 {
 374         struct atgc_management *am = &sbi->am;
 375         struct rb_node **p;
 376         struct rb_node *parent = NULL;
 377         bool left_most = true;
 378
 379         p = f2fs_lookup_rb_tree_ext(sbi, &am->root, &parent, mtime, &left_most);
 380         attach_victim_entry(sbi, mtime, segno, parent, p, left_most);
 381 }
 382
 383 static void add_victim_entry(struct f2fs_sb_info *sbi,
 384                                 struct victim_sel_policy *p, unsigned int segno)
 385 {
 386         struct sit_info *sit_i = SIT_I(sbi);
 387         unsigned int secno = GET_SEC_FROM_SEG(sbi, segno);
 388         unsigned int start = GET_SEG_FROM_SEC(sbi, secno);
 389         unsigned long long mtime = 0;
 390         unsigned int i;
 391
 392         if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED))) {
 393                 if (p->gc_mode == GC_AT &&
 394                         get_valid_blocks(sbi, segno, true) == 0)
 395                         return;
 396         }
 397
 398         for (i = 0; i < sbi->segs_per_sec; i++)
 399                 mtime += get_seg_entry(sbi, start + i)->mtime;
 400         mtime = div_u64(mtime, sbi->segs_per_sec);
 401
 402         /* Handle if the system time has changed by the user */
 403         if (mtime < sit_i->min_mtime)
 404                 sit_i->min_mtime = mtime;
 405         if (mtime > sit_i->max_mtime)
 406                 sit_i->max_mtime = mtime;
 407         if (mtime < sit_i->dirty_min_mtime)
 408                 sit_i->dirty_min_mtime = mtime;
 409         if (mtime > sit_i->dirty_max_mtime)
 410                 sit_i->dirty_max_mtime = mtime;
 411
 412         /* don't choose young section as candidate */
 413         if (sit_i->dirty_max_mtime - mtime < p->age_threshold)
 414                 return;
 415
 416         insert_victim_entry(sbi, mtime, segno);
 417 }
 418
 419 static struct rb_node *lookup_central_victim(struct f2fs_sb_info *sbi,
 420                                                 struct victim_sel_policy *p)
 421 {
 422         struct atgc_management *am = &sbi->am;
 423         struct rb_node *parent = NULL;
 424         bool left_most;
 425
 426         f2fs_lookup_rb_tree_ext(sbi, &am->root, &parent, p->age, &left_most);
 427
 428         return parent;
 429 }
 430
 431 static void atgc_lookup_victim(struct f2fs_sb_info *sbi,
 432                                                 struct victim_sel_policy *p)
 433 {
 434         struct sit_info *sit_i = SIT_I(sbi);
 435         struct atgc_management *am = &sbi->am;
 436         struct rb_root_cached *root = &am->root;
 437         struct rb_node *node;
 438         struct rb_entry *re;
 439         struct victim_entry *ve;
 440         unsigned long long total_time;
 441         unsigned long long age, u, accu;
 442         unsigned long long max_mtime = sit_i->dirty_max_mtime;
 443         unsigned long long min_mtime = sit_i->dirty_min_mtime;
 444         unsigned int sec_blocks = BLKS_PER_SEC(sbi);
 445         unsigned int vblocks;
 446         unsigned int dirty_threshold = max(am->max_candidate_count,
 447                                         am->candidate_ratio *
 448                                         am->victim_count / 100);
 449         unsigned int age_weight = am->age_weight;
 450         unsigned int cost;
 451         unsigned int iter = 0;
 452
 453         if (max_mtime < min_mtime)
 454                 return;
 455
 456         max_mtime += 1;
 457         total_time = max_mtime - min_mtime;
 458
 459         accu = div64_u64(ULLONG_MAX, total_time);
 460         accu = min_t(unsigned long long, div_u64(accu, 100),
 461                                         DEFAULT_ACCURACY_CLASS);
 462
 463         node = rb_first_cached(root);
 464 next:
 465         re = rb_entry_safe(node, struct rb_entry, rb_node);
 466         if (!re)
 467                 return;
 468
 469         ve = (struct victim_entry *)re;
 470
 471         if (ve->mtime >= max_mtime || ve->mtime < min_mtime)
 472                 goto skip;
 473
 474         /* age = 10000 * x% * 60 */
 475         age = div64_u64(accu * (max_mtime - ve->mtime), total_time) *
 476                                                                 age_weight;
 477
 478         vblocks = get_valid_blocks(sbi, ve->segno, true);
 479         f2fs_bug_on(sbi, !vblocks || vblocks == sec_blocks);
 480
 481         /* u = 10000 * x% * 40 */
 482         u = div64_u64(accu * (sec_blocks - vblocks), sec_blocks) *
 483                                                         (100 - age_weight);
 484
 485         f2fs_bug_on(sbi, age + u >= UINT_MAX);
 486
 487         cost = UINT_MAX - (age + u);
 488         iter++;
 489
 490         if (cost < p->min_cost ||
 491                         (cost == p->min_cost && age > p->oldest_age)) {
 492                 p->min_cost = cost;
 493                 p->oldest_age = age;
 494                 p->min_segno = ve->segno;
 495         }
 496 skip:
 497         if (iter < dirty_threshold) {
 498                 node = rb_next(node);
 499                 goto next;
 500         }
 501 }
 502
 503 /*
 504  * select candidates around source section in range of
 505  * [target - dirty_threshold, target + dirty_threshold]
 506  */
 507 static void atssr_lookup_victim(struct f2fs_sb_info *sbi,
 508                                                 struct victim_sel_policy *p)
 509 {
 510         struct sit_info *sit_i = SIT_I(sbi);
 511         struct atgc_management *am = &sbi->am;
 512         struct rb_node *node;
 513         struct rb_entry *re;
 514         struct victim_entry *ve;
 515         unsigned long long age;
 516         unsigned long long max_mtime = sit_i->dirty_max_mtime;
 517         unsigned long long min_mtime = sit_i->dirty_min_mtime;
 518         unsigned int seg_blocks = sbi->blocks_per_seg;
 519         unsigned int vblocks;
 520         unsigned int dirty_threshold = max(am->max_candidate_count,
 521                                         am->candidate_ratio *
 522                                         am->victim_count / 100);
 523         unsigned int cost;
 524         unsigned int iter = 0;
 525         int stage = 0;
 526
 527         if (max_mtime < min_mtime)
 528                 return;
 529         max_mtime += 1;
 530 next_stage:
 531         node = lookup_central_victim(sbi, p);
 532 next_node:
 533         re = rb_entry_safe(node, struct rb_entry, rb_node);
 534         if (!re) {
 535                 if (stage == 0)
 536                         goto skip_stage;
 537                 return;
 538         }
 539
 540         ve = (struct victim_entry *)re;
 541
 542         if (ve->mtime >= max_mtime || ve->mtime < min_mtime)
 543                 goto skip_node;
 544
 545         age = max_mtime - ve->mtime;
 546
 547         vblocks = get_seg_entry(sbi, ve->segno)->ckpt_valid_blocks;
 548         f2fs_bug_on(sbi, !vblocks);
 549
 550         /* rare case */
 551         if (vblocks == seg_blocks)
 552                 goto skip_node;
 553
 554         iter++;
 555
 556         age = max_mtime - abs(p->age - age);
 557         cost = UINT_MAX - vblocks;
 558
 559         if (cost < p->min_cost ||
 560                         (cost == p->min_cost && age > p->oldest_age)) {
 561                 p->min_cost = cost;
 562                 p->oldest_age = age;
 563                 p->min_segno = ve->segno;
 564         }
 565 skip_node:
 566         if (iter < dirty_threshold) {
 567                 if (stage == 0)
 568                         node = rb_prev(node);
 569                 else if (stage == 1)
 570                         node = rb_next(node);
 571                 goto next_node;
 572         }
 573 skip_stage:
 574         if (stage < 1) {
 575                 stage++;
 576                 iter = 0;
 577                 goto next_stage;
 578         }
 579 }
 580 static void lookup_victim_by_age(struct f2fs_sb_info *sbi,
 581                                                 struct victim_sel_policy *p)
 582 {
 583         f2fs_bug_on(sbi, !f2fs_check_rb_tree_consistence(sbi,
 584                                                 &sbi->am.root, true));
 585
 586         if (p->gc_mode == GC_AT)
 587                 atgc_lookup_victim(sbi, p);
 588         else if (p->alloc_mode == AT_SSR)
 589                 atssr_lookup_victim(sbi, p);
 590         else
 591                 f2fs_bug_on(sbi, 1);
 592 }
 593
 594 static void release_victim_entry(struct f2fs_sb_info *sbi)
 595 {
 596         struct atgc_management *am = &sbi->am;
 597         struct victim_entry *ve, *tmp;
 598
 599         list_for_each_entry_safe(ve, tmp, &am->victim_list, list) {
 600                 list_del(&ve->list);
 601                 kmem_cache_free(victim_entry_slab, ve);
 602                 am->victim_count--;
 603         }
 604
 605         am->root = RB_ROOT_CACHED;
 606
 607         f2fs_bug_on(sbi, am->victim_count);
 608         f2fs_bug_on(sbi, !list_empty(&am->victim_list));
 609 }
 610
 611 /*
 612  * This function is called from two paths.
 613  * One is garbage collection and the other is SSR segment selection.
 614  * When it is called during GC, it just gets a victim segment
 615  * and it does not remove it from dirty seglist.
 616  * When it is called from SSR segment selection, it finds a segment
 617  * which has minimum valid blocks and removes it from dirty seglist.
 618  */
 619 static int get_victim_by_default(struct f2fs_sb_info *sbi,
 620                         unsigned int *result, int gc_type, int type,
 621                         char alloc_mode, unsigned long long age)
 622 {
 623         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
 624         struct sit_info *sm = SIT_I(sbi);
 625         struct victim_sel_policy p;
 626         unsigned int secno, last_victim;
 627         unsigned int last_segment;
 628         unsigned int nsearched;
 629         bool is_atgc;
 630         int ret = 0;
 631
 632         mutex_lock(&dirty_i->seglist_lock);
 633         last_segment = MAIN_SECS(sbi) * sbi->segs_per_sec;
 634
 635         p.alloc_mode = alloc_mode;
 636         p.age = age;
 637         p.age_threshold = sbi->am.age_threshold;
 638
 639 retry:
 640         select_policy(sbi, gc_type, type, &p);
 641         p.min_segno = NULL_SEGNO;
 642         p.oldest_age = 0;
 643         p.min_cost = get_max_cost(sbi, &p);
 644
 645         is_atgc = (p.gc_mode == GC_AT || p.alloc_mode == AT_SSR);
 646         nsearched = 0;
 647
 648         if (is_atgc)
 649                 SIT_I(sbi)->dirty_min_mtime = ULLONG_MAX;
 650
 651         if (*result != NULL_SEGNO) {
 652                 if (!get_valid_blocks(sbi, *result, false)) {
 653                         ret = -ENODATA;
 654                         goto out;
 655                 }
 656
 657                 if (sec_usage_check(sbi, GET_SEC_FROM_SEG(sbi, *result)))
 658                         ret = -EBUSY;
 659                 else
 660                         p.min_segno = *result;
 661                 goto out;
 662         }
 663
 664         ret = -ENODATA;
 665         if (p.max_search == 0)
 666                 goto out;
 667
 668         if (__is_large_section(sbi) && p.alloc_mode == LFS) {
 669                 if (sbi->next_victim_seg[BG_GC] != NULL_SEGNO) {
 670                         p.min_segno = sbi->next_victim_seg[BG_GC];
 671                         *result = p.min_segno;
 672                         sbi->next_victim_seg[BG_GC] = NULL_SEGNO;
 673                         goto got_result;
 674                 }
 675                 if (gc_type == FG_GC &&
 676                                 sbi->next_victim_seg[FG_GC] != NULL_SEGNO) {
 677                         p.min_segno = sbi->next_victim_seg[FG_GC];
 678                         *result = p.min_segno;
 679                         sbi->next_victim_seg[FG_GC] = NULL_SEGNO;
 680                         goto got_result;
 681                 }
 682         }
 683
 684         last_victim = sm->last_victim[p.gc_mode];
 685         if (p.alloc_mode == LFS && gc_type == FG_GC) {
 686                 p.min_segno = check_bg_victims(sbi);
 687                 if (p.min_segno != NULL_SEGNO)
 688                         goto got_it;
 689         }
 690
 691         while (1) {
 692                 unsigned long cost, *dirty_bitmap;
 693                 unsigned int unit_no, segno;
 694
 695                 dirty_bitmap = p.dirty_bitmap;
 696                 unit_no = find_next_bit(dirty_bitmap,
 697                                 last_segment / p.ofs_unit,
 698                                 p.offset / p.ofs_unit);
 699                 segno = unit_no * p.ofs_unit;
 700                 if (segno >= last_segment) {
 701                         if (sm->last_victim[p.gc_mode]) {
 702                                 last_segment =
 703                                         sm->last_victim[p.gc_mode];
 704                                 sm->last_victim[p.gc_mode] = 0;
 705                                 p.offset = 0;
 706                                 continue;
 707                         }
 708                         break;
 709                 }
 710
 711                 p.offset = segno + p.ofs_unit;
 712                 nsearched++;
 713
 714 #ifdef CONFIG_F2FS_CHECK_FS
 715                 /*
 716                  * skip selecting the invalid segno (that is failed due to block
 717                  * validity check failure during GC) to avoid endless GC loop in
 718                  * such cases.
 719                  */
 720                 if (test_bit(segno, sm->invalid_segmap))
 721                         goto next;
 722 #endif
 723
 724                 secno = GET_SEC_FROM_SEG(sbi, segno);
 725
 726                 if (sec_usage_check(sbi, secno))
 727                         goto next;
 728
 729                 /* Don't touch checkpointed data */
 730                 if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED))) {
 731                         if (p.alloc_mode == LFS) {
 732                                 /*
 733                                  * LFS is set to find source section during GC.
 734                                  * The victim should have no checkpointed data.
 735                                  */
 736                                 if (get_ckpt_valid_blocks(sbi, segno, true))
 737                                         goto next;
 738                         } else {
 739                                 /*
 740                                  * SSR | AT_SSR are set to find target segment
 741                                  * for writes which can be full by checkpointed
 742                                  * and newly written blocks.
 743                                  */
 744                                 if (!f2fs_segment_has_free_slot(sbi, segno))
 745                                         goto next;
 746                         }
 747                 }
 748
 749                 if (gc_type == BG_GC && test_bit(secno, dirty_i->victim_secmap))
 750                         goto next;
 751
 752                 if (is_atgc) {
 753                         add_victim_entry(sbi, &p, segno);
 754                         goto next;
 755                 }
 756
 757                 cost = get_gc_cost(sbi, segno, &p);
 758
 759                 if (p.min_cost > cost) {
 760                         p.min_segno = segno;
 761                         p.min_cost = cost;
 762                 }
 763 next:
 764                 if (nsearched >= p.max_search) {
 765                         if (!sm->last_victim[p.gc_mode] && segno <= last_victim)
 766                                 sm->last_victim[p.gc_mode] =
 767                                         last_victim + p.ofs_unit;
 768                         else
 769                                 sm->last_victim[p.gc_mode] = segno + p.ofs_unit;
 770                         sm->last_victim[p.gc_mode] %=
 771                                 (MAIN_SECS(sbi) * sbi->segs_per_sec);
 772                         break;
 773                 }
 774         }
 775
 776         /* get victim for GC_AT/AT_SSR */
 777         if (is_atgc) {
 778                 lookup_victim_by_age(sbi, &p);
 779                 release_victim_entry(sbi);
 780         }
 781
 782         if (is_atgc && p.min_segno == NULL_SEGNO &&
 783                         sm->elapsed_time < p.age_threshold) {
 784                 p.age_threshold = 0;
 785                 goto retry;
 786         }
 787
 788         if (p.min_segno != NULL_SEGNO) {
 789 got_it:
 790                 *result = (p.min_segno / p.ofs_unit) * p.ofs_unit;
 791 got_result:
 792                 if (p.alloc_mode == LFS) {
 793                         secno = GET_SEC_FROM_SEG(sbi, p.min_segno);
 794                         if (gc_type == FG_GC)
 795                                 sbi->cur_victim_sec = secno;
 796                         else
 797                                 set_bit(secno, dirty_i->victim_secmap);
 798                 }
 799                 ret = 0;
 800
 801         }
 802 out:
 803         if (p.min_segno != NULL_SEGNO)
 804                 trace_f2fs_get_victim(sbi->sb, type, gc_type, &p,
 805                                 sbi->cur_victim_sec,
 806                                 prefree_segments(sbi), free_segments(sbi));
 807         mutex_unlock(&dirty_i->seglist_lock);
 808
 809         return ret;
 810 }
 811
 812 static const struct victim_selection default_v_ops = {
 813         .get_victim = get_victim_by_default,
 814 };
 815
 816 static struct inode *find_gc_inode(struct gc_inode_list *gc_list, nid_t ino)
 817 {
 818         struct inode_entry *ie;
 819
 820         ie = radix_tree_lookup(&gc_list->iroot, ino);
 821         if (ie)
 822                 return ie->inode;
 823         return NULL;
 824 }
 825
 826 static void add_gc_inode(struct gc_inode_list *gc_list, struct inode *inode)
 827 {
 828         struct inode_entry *new_ie;
 829
 830         if (inode == find_gc_inode(gc_list, inode->i_ino)) {
 831                 iput(inode);
 832                 return;
 833         }
 834         new_ie = f2fs_kmem_cache_alloc(f2fs_inode_entry_slab, GFP_NOFS);
 835         new_ie->inode = inode;
 836
 837         f2fs_radix_tree_insert(&gc_list->iroot, inode->i_ino, new_ie);
 838         list_add_tail(&new_ie->list, &gc_list->ilist);
 839 }
 840
 841 static void put_gc_inode(struct gc_inode_list *gc_list)
 842 {
 843         struct inode_entry *ie, *next_ie;
 844         list_for_each_entry_safe(ie, next_ie, &gc_list->ilist, list) {
 845                 radix_tree_delete(&gc_list->iroot, ie->inode->i_ino);
 846                 iput(ie->inode);
 847                 list_del(&ie->list);
 848                 kmem_cache_free(f2fs_inode_entry_slab, ie);
 849         }
 850 }
 851
 852 static int check_valid_map(struct f2fs_sb_info *sbi,
 853                                 unsigned int segno, int offset)
 854 {
 855         struct sit_info *sit_i = SIT_I(sbi);
 856         struct seg_entry *sentry;
 857         int ret;
 858
 859         down_read(&sit_i->sentry_lock);
 860         sentry = get_seg_entry(sbi, segno);
 861         ret = f2fs_test_bit(offset, sentry->cur_valid_map);
 862         up_read(&sit_i->sentry_lock);
 863         return ret;
 864 }
 865
 866 /*
 867  * This function compares node address got in summary with that in NAT.
 868  * On validity, copy that node with cold status, otherwise (invalid node)
 869  * ignore that.
 870  */
 871 static int gc_node_segment(struct f2fs_sb_info *sbi,
 872                 struct f2fs_summary *sum, unsigned int segno, int gc_type)
 873 {
 874         struct f2fs_summary *entry;
 875         block_t start_addr;
 876         int off;
 877         int phase = 0;
 878         bool fggc = (gc_type == FG_GC);
 879         int submitted = 0;
 880         unsigned int usable_blks_in_seg = f2fs_usable_blks_in_seg(sbi, segno);
 881
 882         start_addr = START_BLOCK(sbi, segno);
 883
 884 next_step:
 885         entry = sum;
 886
 887         if (fggc && phase == 2)
 888                 atomic_inc(&sbi->wb_sync_req[NODE]);
 889
 890         for (off = 0; off < usable_blks_in_seg; off++, entry++) {
 891                 nid_t nid = le32_to_cpu(entry->nid);
 892                 struct page *node_page;
 893                 struct node_info ni;
 894                 int err;
 895
 896                 /* stop BG_GC if there is not enough free sections. */
 897                 if (gc_type == BG_GC && has_not_enough_free_secs(sbi, 0, 0))
 898                         return submitted;
 899
 900                 if (check_valid_map(sbi, segno, off) == 0)
 901                         continue;
 902
 903                 if (phase == 0) {
 904                         f2fs_ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nid), 1,
 905                                                         META_NAT, true);
 906                         continue;
 907                 }
 908
 909                 if (phase == 1) {
 910                         f2fs_ra_node_page(sbi, nid);
 911                         continue;
 912                 }
 913
 914                 /* phase == 2 */
 915                 node_page = f2fs_get_node_page(sbi, nid);
 916                 if (IS_ERR(node_page))
 917                         continue;
 918
 919                 /* block may become invalid during f2fs_get_node_page */
 920                 if (check_valid_map(sbi, segno, off) == 0) {
 921                         f2fs_put_page(node_page, 1);
 922                         continue;
 923                 }
 924
 925                 if (f2fs_get_node_info(sbi, nid, &ni)) {
 926                         f2fs_put_page(node_page, 1);
 927                         continue;
 928                 }
 929
 930                 if (ni.blk_addr != start_addr + off) {
 931                         f2fs_put_page(node_page, 1);
 932                         continue;
 933                 }
 934
 935                 err = f2fs_move_node_page(node_page, gc_type);
 936                 if (!err && gc_type == FG_GC)
 937                         submitted++;
 938                 stat_inc_node_blk_count(sbi, 1, gc_type);
 939         }
 940
 941         if (++phase < 3)
 942                 goto next_step;
 943
 944         if (fggc)
 945                 atomic_dec(&sbi->wb_sync_req[NODE]);
 946         return submitted;
 947 }
 948
 949 /*
 950  * Calculate start block index indicating the given node offset.
 951  * Be careful, caller should give this node offset only indicating direct node
 952  * blocks. If any node offsets, which point the other types of node blocks such
 953  * as indirect or double indirect node blocks, are given, it must be a caller's
 954  * bug.
 955  */
 956 block_t f2fs_start_bidx_of_node(unsigned int node_ofs, struct inode *inode)
 957 {
 958         unsigned int indirect_blks = 2 * NIDS_PER_BLOCK + 4;
 959         unsigned int bidx;
 960
 961         if (node_ofs == 0)
 962                 return 0;
 963
 964         if (node_ofs <= 2) {
 965                 bidx = node_ofs - 1;
 966         } else if (node_ofs <= indirect_blks) {
 967                 int dec = (node_ofs - 4) / (NIDS_PER_BLOCK + 1);
 968                 bidx = node_ofs - 2 - dec;
 969         } else {
 970                 int dec = (node_ofs - indirect_blks - 3) / (NIDS_PER_BLOCK + 1);
 971                 bidx = node_ofs - 5 - dec;
 972         }
 973         return bidx * ADDRS_PER_BLOCK(inode) + ADDRS_PER_INODE(inode);
 974 }
 975
 976 static bool is_alive(struct f2fs_sb_info *sbi, struct f2fs_summary *sum,
 977                 struct node_info *dni, block_t blkaddr, unsigned int *nofs)
 978 {
 979         struct page *node_page;
 980         nid_t nid;
 981         unsigned int ofs_in_node, max_addrs, base;
 982         block_t source_blkaddr;
 983
 984         nid = le32_to_cpu(sum->nid);
 985         ofs_in_node = le16_to_cpu(sum->ofs_in_node);
 986
 987         node_page = f2fs_get_node_page(sbi, nid);
 988         if (IS_ERR(node_page))
 989                 return false;
 990
 991         if (f2fs_get_node_info(sbi, nid, dni)) {
 992                 f2fs_put_page(node_page, 1);
 993                 return false;
 994         }
 995
 996         if (sum->version != dni->version) {
 997                 f2fs_warn(sbi, "%s: valid data with mismatched node version.",
 998                           __func__);
 999                 set_sbi_flag(sbi, SBI_NEED_FSCK);
1000         }
1001
1002         if (f2fs_check_nid_range(sbi, dni->ino)) {
1003                 f2fs_put_page(node_page, 1);
1004                 return false;
1005         }
1006
1007         if (IS_INODE(node_page)) {
1008                 base = offset_in_addr(F2FS_INODE(node_page));
1009                 max_addrs = DEF_ADDRS_PER_INODE;
1010         } else {
1011                 base = 0;
1012                 max_addrs = DEF_ADDRS_PER_BLOCK;
1013         }
1014
1015         if (base + ofs_in_node >= max_addrs) {
1016                 f2fs_err(sbi, "Inconsistent blkaddr offset: base:%u, ofs_in_node:%u, max:%u, ino:%u, nid:%u",
1017                         base, ofs_in_node, max_addrs, dni->ino, dni->nid);
1018                 f2fs_put_page(node_page, 1);
1019                 return false;
1020         }
1021
1022         *nofs = ofs_of_node(node_page);
1023         source_blkaddr = data_blkaddr(NULL, node_page, ofs_in_node);
1024         f2fs_put_page(node_page, 1);
1025
1026         if (source_blkaddr != blkaddr) {
1027 #ifdef CONFIG_F2FS_CHECK_FS
1028                 unsigned int segno = GET_SEGNO(sbi, blkaddr);
1029                 unsigned long offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
1030
1031                 if (unlikely(check_valid_map(sbi, segno, offset))) {
1032                         if (!test_and_set_bit(segno, SIT_I(sbi)->invalid_segmap)) {
1033                                 f2fs_err(sbi, "mismatched blkaddr %u (source_blkaddr %u) in seg %u\n",
1034                                                 blkaddr, source_blkaddr, segno);
1035                                 f2fs_bug_on(sbi, 1);
1036                         }
1037                 }
1038 #endif
1039                 return false;
1040         }
1041         return true;
1042 }
1043
1044 static int ra_data_block(struct inode *inode, pgoff_t index)
1045 {
1046         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1047         struct address_space *mapping = inode->i_mapping;
1048         struct dnode_of_data dn;
1049         struct page *page;
1050         struct extent_info ei = {0, 0, 0};
1051         struct f2fs_io_info fio = {
1052                 .sbi = sbi,
1053                 .ino = inode->i_ino,
1054                 .type = DATA,
1055                 .temp = COLD,
1056                 .op = REQ_OP_READ,
1057                 .op_flags = 0,
1058                 .encrypted_page = NULL,
1059                 .in_list = false,
1060                 .retry = false,
1061         };
1062         int err;
1063
1064         page = f2fs_grab_cache_page(mapping, index, true);
1065         if (!page)
1066                 return -ENOMEM;
1067
1068         if (f2fs_lookup_extent_cache(inode, index, &ei)) {
1069                 dn.data_blkaddr = ei.blk + index - ei.fofs;
1070                 if (unlikely(!f2fs_is_valid_blkaddr(sbi, dn.data_blkaddr,
1071                                                 DATA_GENERIC_ENHANCE_READ))) {
1072                         err = -EFSCORRUPTED;
1073                         goto put_page;
1074                 }
1075                 goto got_it;
1076         }
1077
1078         set_new_dnode(&dn, inode, NULL, NULL, 0);
1079         err = f2fs_get_dnode_of_data(&dn, index, LOOKUP_NODE);
1080         if (err)
1081                 goto put_page;
1082         f2fs_put_dnode(&dn);
1083
1084         if (!__is_valid_data_blkaddr(dn.data_blkaddr)) {
1085                 err = -ENOENT;
1086                 goto put_page;
1087         }
1088         if (unlikely(!f2fs_is_valid_blkaddr(sbi, dn.data_blkaddr,
1089                                                 DATA_GENERIC_ENHANCE))) {
1090                 err = -EFSCORRUPTED;
1091                 goto put_page;
1092         }
1093 got_it:
1094         /* read page */
1095         fio.page = page;
1096         fio.new_blkaddr = fio.old_blkaddr = dn.data_blkaddr;
1097
1098         /*
1099          * don't cache encrypted data into meta inode until previous dirty
1100          * data were writebacked to avoid racing between GC and flush.
1101          */
1102         f2fs_wait_on_page_writeback(page, DATA, true, true);
1103
1104         f2fs_wait_on_block_writeback(inode, dn.data_blkaddr);
1105
1106         fio.encrypted_page = f2fs_pagecache_get_page(META_MAPPING(sbi),
1107                                         dn.data_blkaddr,
1108                                         FGP_LOCK | FGP_CREAT, GFP_NOFS);
1109         if (!fio.encrypted_page) {
1110                 err = -ENOMEM;
1111                 goto put_page;
1112         }
1113
1114         err = f2fs_submit_page_bio(&fio);
1115         if (err)
1116                 goto put_encrypted_page;
1117         f2fs_put_page(fio.encrypted_page, 0);
1118         f2fs_put_page(page, 1);
1119
1120         f2fs_update_iostat(sbi, FS_DATA_READ_IO, F2FS_BLKSIZE);
1121         f2fs_update_iostat(sbi, FS_GDATA_READ_IO, F2FS_BLKSIZE);
1122
1123         return 0;
1124 put_encrypted_page:
1125         f2fs_put_page(fio.encrypted_page, 1);
1126 put_page:
1127         f2fs_put_page(page, 1);
1128         return err;
1129 }
1130
1131 /*
1132  * Move data block via META_MAPPING while keeping locked data page.
1133  * This can be used to move blocks, aka LBAs, directly on disk.
1134  */
1135 static int move_data_block(struct inode *inode, block_t bidx,
1136                                 int gc_type, unsigned int segno, int off)
1137 {
1138         struct f2fs_io_info fio = {
1139                 .sbi = F2FS_I_SB(inode),
1140                 .ino = inode->i_ino,
1141                 .type = DATA,
1142                 .temp = COLD,
1143                 .op = REQ_OP_READ,
1144                 .op_flags = 0,
1145                 .encrypted_page = NULL,
1146                 .in_list = false,
1147                 .retry = false,
1148         };
1149         struct dnode_of_data dn;
1150         struct f2fs_summary sum;
1151         struct node_info ni;
1152         struct page *page, *mpage;
1153         block_t newaddr;
1154         int err = 0;
1155         bool lfs_mode = f2fs_lfs_mode(fio.sbi);
1156         int type = fio.sbi->am.atgc_enabled ?
1157                                 CURSEG_ALL_DATA_ATGC : CURSEG_COLD_DATA;
1158
1159         /* do not read out */
1160         page = f2fs_grab_cache_page(inode->i_mapping, bidx, false);
1161         if (!page)
1162                 return -ENOMEM;
1163
1164         if (!check_valid_map(F2FS_I_SB(inode), segno, off)) {
1165                 err = -ENOENT;
1166                 goto out;
1167         }
1168
1169         if (f2fs_is_atomic_file(inode)) {
1170                 F2FS_I(inode)->i_gc_failures[GC_FAILURE_ATOMIC]++;
1171                 F2FS_I_SB(inode)->skipped_atomic_files[gc_type]++;
1172                 err = -EAGAIN;
1173                 goto out;
1174         }
1175
1176         if (f2fs_is_pinned_file(inode)) {
1177                 if (gc_type == FG_GC)
1178                         f2fs_pin_file_control(inode, true);
1179                 err = -EAGAIN;
1180                 goto out;
1181         }
1182
1183         set_new_dnode(&dn, inode, NULL, NULL, 0);
1184         err = f2fs_get_dnode_of_data(&dn, bidx, LOOKUP_NODE);
1185         if (err)
1186                 goto out;
1187
1188         if (unlikely(dn.data_blkaddr == NULL_ADDR)) {
1189                 ClearPageUptodate(page);
1190                 err = -ENOENT;
1191                 goto put_out;
1192         }
1193
1194         /*
1195          * don't cache encrypted data into meta inode until previous dirty
1196          * data were writebacked to avoid racing between GC and flush.
1197          */
1198         f2fs_wait_on_page_writeback(page, DATA, true, true);
1199
1200         f2fs_wait_on_block_writeback(inode, dn.data_blkaddr);
1201
1202         err = f2fs_get_node_info(fio.sbi, dn.nid, &ni);
1203         if (err)
1204                 goto put_out;
1205
1206         set_summary(&sum, dn.nid, dn.ofs_in_node, ni.version);
1207
1208         /* read page */
1209         fio.page = page;
1210         fio.new_blkaddr = fio.old_blkaddr = dn.data_blkaddr;
1211
1212         if (lfs_mode)
1213                 down_write(&fio.sbi->io_order_lock);
1214
1215         mpage = f2fs_grab_cache_page(META_MAPPING(fio.sbi),
1216                                         fio.old_blkaddr, false);
1217         if (!mpage) {
1218                 err = -ENOMEM;
1219                 goto up_out;
1220         }
1221
1222         fio.encrypted_page = mpage;
1223
1224         /* read source block in mpage */
1225         if (!PageUptodate(mpage)) {
1226                 err = f2fs_submit_page_bio(&fio);
1227                 if (err) {
1228                         f2fs_put_page(mpage, 1);
1229                         goto up_out;
1230                 }
1231
1232                 f2fs_update_iostat(fio.sbi, FS_DATA_READ_IO, F2FS_BLKSIZE);
1233                 f2fs_update_iostat(fio.sbi, FS_GDATA_READ_IO, F2FS_BLKSIZE);
1234
1235                 lock_page(mpage);
1236                 if (unlikely(mpage->mapping != META_MAPPING(fio.sbi) ||
1237                                                 !PageUptodate(mpage))) {
1238                         err = -EIO;
1239                         f2fs_put_page(mpage, 1);
1240                         goto up_out;
1241                 }
1242         }
1243
1244         f2fs_allocate_data_block(fio.sbi, NULL, fio.old_blkaddr, &newaddr,
1245                                 &sum, type, NULL);
1246
1247         fio.encrypted_page = f2fs_pagecache_get_page(META_MAPPING(fio.sbi),
1248                                 newaddr, FGP_LOCK | FGP_CREAT, GFP_NOFS);
1249         if (!fio.encrypted_page) {
1250                 err = -ENOMEM;
1251                 f2fs_put_page(mpage, 1);
1252                 goto recover_block;
1253         }
1254
1255         /* write target block */
1256         f2fs_wait_on_page_writeback(fio.encrypted_page, DATA, true, true);
1257         memcpy(page_address(fio.encrypted_page),
1258                                 page_address(mpage), PAGE_SIZE);
1259         f2fs_put_page(mpage, 1);
1260         invalidate_mapping_pages(META_MAPPING(fio.sbi),
1261                                 fio.old_blkaddr, fio.old_blkaddr);
1262
1263         set_page_dirty(fio.encrypted_page);
1264         if (clear_page_dirty_for_io(fio.encrypted_page))
1265                 dec_page_count(fio.sbi, F2FS_DIRTY_META);
1266
1267         set_page_writeback(fio.encrypted_page);
1268         ClearPageError(page);
1269
1270         /* allocate block address */
1271         f2fs_wait_on_page_writeback(dn.node_page, NODE, true, true);
1272
1273         fio.op = REQ_OP_WRITE;
1274         fio.op_flags = REQ_SYNC;
1275         fio.new_blkaddr = newaddr;
1276         f2fs_submit_page_write(&fio);
1277         if (fio.retry) {
1278                 err = -EAGAIN;
1279                 if (PageWriteback(fio.encrypted_page))
1280                         end_page_writeback(fio.encrypted_page);
1281                 goto put_page_out;
1282         }
1283
1284         f2fs_update_iostat(fio.sbi, FS_GC_DATA_IO, F2FS_BLKSIZE);
1285
1286         f2fs_update_data_blkaddr(&dn, newaddr);
1287         set_inode_flag(inode, FI_APPEND_WRITE);
1288         if (page->index == 0)
1289                 set_inode_flag(inode, FI_FIRST_BLOCK_WRITTEN);
1290 put_page_out:
1291         f2fs_put_page(fio.encrypted_page, 1);
1292 recover_block:
1293         if (err)
1294                 f2fs_do_replace_block(fio.sbi, &sum, newaddr, fio.old_blkaddr,
1295                                                         true, true, true);
1296 up_out:
1297         if (lfs_mode)
1298                 up_write(&fio.sbi->io_order_lock);
1299 put_out:
1300         f2fs_put_dnode(&dn);
1301 out:
1302         f2fs_put_page(page, 1);
1303         return err;
1304 }
1305
1306 static int move_data_page(struct inode *inode, block_t bidx, int gc_type,
1307                                                         unsigned int segno, int off)
1308 {
1309         struct page *page;
1310         int err = 0;
1311
1312         page = f2fs_get_lock_data_page(inode, bidx, true);
1313         if (IS_ERR(page))
1314                 return PTR_ERR(page);
1315
1316         if (!check_valid_map(F2FS_I_SB(inode), segno, off)) {
1317                 err = -ENOENT;
1318                 goto out;
1319         }
1320
1321         if (f2fs_is_atomic_file(inode)) {
1322                 F2FS_I(inode)->i_gc_failures[GC_FAILURE_ATOMIC]++;
1323                 F2FS_I_SB(inode)->skipped_atomic_files[gc_type]++;
1324                 err = -EAGAIN;
1325                 goto out;
1326         }
1327         if (f2fs_is_pinned_file(inode)) {
1328                 if (gc_type == FG_GC)
1329                         f2fs_pin_file_control(inode, true);
1330                 err = -EAGAIN;
1331                 goto out;
1332         }
1333
1334         if (gc_type == BG_GC) {
1335                 if (PageWriteback(page)) {
1336                         err = -EAGAIN;
1337                         goto out;
1338                 }
1339                 set_page_dirty(page);
1340                 set_cold_data(page);
1341         } else {
1342                 struct f2fs_io_info fio = {
1343                         .sbi = F2FS_I_SB(inode),
1344                         .ino = inode->i_ino,
1345                         .type = DATA,
1346                         .temp = COLD,
1347                         .op = REQ_OP_WRITE,
1348                         .op_flags = REQ_SYNC,
1349                         .old_blkaddr = NULL_ADDR,
1350                         .page = page,
1351                         .encrypted_page = NULL,
1352                         .need_lock = LOCK_REQ,
1353                         .io_type = FS_GC_DATA_IO,
1354                 };
1355                 bool is_dirty = PageDirty(page);
1356
1357 retry:
1358                 f2fs_wait_on_page_writeback(page, DATA, true, true);
1359
1360                 set_page_dirty(page);
1361                 if (clear_page_dirty_for_io(page)) {
1362                         inode_dec_dirty_pages(inode);
1363                         f2fs_remove_dirty_inode(inode);
1364                 }
1365
1366                 set_cold_data(page);
1367
1368                 err = f2fs_do_write_data_page(&fio);
1369                 if (err) {
1370                         clear_cold_data(page);
1371                         if (err == -ENOMEM) {
1372                                 congestion_wait(BLK_RW_ASYNC,
1373                                                 DEFAULT_IO_TIMEOUT);
1374                                 goto retry;
1375                         }
1376                         if (is_dirty)
1377                                 set_page_dirty(page);
1378                 }
1379         }
1380 out:
1381         f2fs_put_page(page, 1);
1382         return err;
1383 }
1384
1385 /*
1386  * This function tries to get parent node of victim data block, and identifies
1387  * data block validity. If the block is valid, copy that with cold status and
1388  * modify parent node.
1389  * If the parent node is not valid or the data block address is different,
1390  * the victim data block is ignored.
1391  */
1392 static int gc_data_segment(struct f2fs_sb_info *sbi, struct f2fs_summary *sum,
1393                 struct gc_inode_list *gc_list, unsigned int segno, int gc_type,
1394                 bool force_migrate)
1395 {
1396         struct super_block *sb = sbi->sb;
1397         struct f2fs_summary *entry;
1398         block_t start_addr;
1399         int off;
1400         int phase = 0;
1401         int submitted = 0;
1402         unsigned int usable_blks_in_seg = f2fs_usable_blks_in_seg(sbi, segno);
1403
1404         start_addr = START_BLOCK(sbi, segno);
1405
1406 next_step:
1407         entry = sum;
1408
1409         for (off = 0; off < usable_blks_in_seg; off++, entry++) {
1410                 struct page *data_page;
1411                 struct inode *inode;
1412                 struct node_info dni; /* dnode info for the data */
1413                 unsigned int ofs_in_node, nofs;
1414                 block_t start_bidx;
1415                 nid_t nid = le32_to_cpu(entry->nid);
1416
1417                 /*
1418                  * stop BG_GC if there is not enough free sections.
1419                  * Or, stop GC if the segment becomes fully valid caused by
1420                  * race condition along with SSR block allocation.
1421                  */
1422                 if ((gc_type == BG_GC && has_not_enough_free_secs(sbi, 0, 0)) ||
1423                         (!force_migrate && get_valid_blocks(sbi, segno, true) ==
1424                                                         BLKS_PER_SEC(sbi)))
1425                         return submitted;
1426
1427                 if (check_valid_map(sbi, segno, off) == 0)
1428                         continue;
1429
1430                 if (phase == 0) {
1431                         f2fs_ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nid), 1,
1432                                                         META_NAT, true);
1433                         continue;
1434                 }
1435
1436                 if (phase == 1) {
1437                         f2fs_ra_node_page(sbi, nid);
1438                         continue;
1439                 }
1440
1441                 /* Get an inode by ino with checking validity */
1442                 if (!is_alive(sbi, entry, &dni, start_addr + off, &nofs))
1443                         continue;
1444
1445                 if (phase == 2) {
1446                         f2fs_ra_node_page(sbi, dni.ino);
1447                         continue;
1448                 }
1449
1450                 ofs_in_node = le16_to_cpu(entry->ofs_in_node);
1451
1452                 if (phase == 3) {
1453                         inode = f2fs_iget(sb, dni.ino);
1454                         if (IS_ERR(inode) || is_bad_inode(inode) ||
1455                                         special_file(inode->i_mode)) {
1456                                 set_sbi_flag(sbi, SBI_NEED_FSCK);
1457                                 continue;
1458                         }
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                 migrated++;
1655
1656 freed:
1657                 if (gc_type == FG_GC &&
1658                                 get_valid_blocks(sbi, segno, false) == 0)
1659                         seg_freed++;
1660
1661                 if (__is_large_section(sbi))
1662                         sbi->next_victim_seg[gc_type] =
1663                                 (segno + 1 < end_segno) ? segno + 1 : NULL_SEGNO;
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 && seg_freed)
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                 goto stop;
1764
1765         if (skipped_round <= MAX_SKIP_GC_COUNT || skipped_round * 2 < round) {
1766
1767                 /* Write checkpoint to reclaim prefree segments */
1768                 if (free_sections(sbi) < NR_CURSEG_PERSIST_TYPE &&
1769                                 prefree_segments(sbi) &&
1770                                 !is_sbi_flag_set(sbi, SBI_CP_DISABLED)) {
1771                         ret = f2fs_write_checkpoint(sbi, &cpc);
1772                         if (ret)
1773                                 goto stop;
1774                 }
1775                 segno = NULL_SEGNO;
1776                 goto gc_more;
1777         }
1778         if (first_skipped < last_skipped &&
1779                         (last_skipped - first_skipped) >
1780                                         sbi->skipped_gc_rwsem) {
1781                 f2fs_drop_inmem_pages_all(sbi, true);
1782                 segno = NULL_SEGNO;
1783                 goto gc_more;
1784         }
1785         if (gc_type == FG_GC && !is_sbi_flag_set(sbi, SBI_CP_DISABLED))
1786                 ret = f2fs_write_checkpoint(sbi, &cpc);
1787 stop:
1788         SIT_I(sbi)->last_victim[ALLOC_NEXT] = 0;
1789         SIT_I(sbi)->last_victim[FLUSH_DEVICE] = init_segno;
1790
1791         trace_f2fs_gc_end(sbi->sb, ret, total_freed, sec_freed,
1792                                 get_pages(sbi, F2FS_DIRTY_NODES),
1793                                 get_pages(sbi, F2FS_DIRTY_DENTS),
1794                                 get_pages(sbi, F2FS_DIRTY_IMETA),
1795                                 free_sections(sbi),
1796                                 free_segments(sbi),
1797                                 reserved_segments(sbi),
1798                                 prefree_segments(sbi));
1799
1800         up_write(&sbi->gc_lock);
1801
1802         put_gc_inode(&gc_list);
1803
1804         if (sync && !ret)
1805                 ret = sec_freed ? 0 : -EAGAIN;
1806         return ret;
1807 }
1808
1809 int __init f2fs_create_garbage_collection_cache(void)
1810 {
1811         victim_entry_slab = f2fs_kmem_cache_create("f2fs_victim_entry",
1812                                         sizeof(struct victim_entry));
1813         if (!victim_entry_slab)
1814                 return -ENOMEM;
1815         return 0;
1816 }
1817
1818 void f2fs_destroy_garbage_collection_cache(void)
1819 {
1820         kmem_cache_destroy(victim_entry_slab);
1821 }
1822
1823 static void init_atgc_management(struct f2fs_sb_info *sbi)
1824 {
1825         struct atgc_management *am = &sbi->am;
1826
1827         if (test_opt(sbi, ATGC) &&
1828                 SIT_I(sbi)->elapsed_time >= DEF_GC_THREAD_AGE_THRESHOLD)
1829                 am->atgc_enabled = true;
1830
1831         am->root = RB_ROOT_CACHED;
1832         INIT_LIST_HEAD(&am->victim_list);
1833         am->victim_count = 0;
1834
1835         am->candidate_ratio = DEF_GC_THREAD_CANDIDATE_RATIO;
1836         am->max_candidate_count = DEF_GC_THREAD_MAX_CANDIDATE_COUNT;
1837         am->age_weight = DEF_GC_THREAD_AGE_WEIGHT;
1838         am->age_threshold = DEF_GC_THREAD_AGE_THRESHOLD;
1839 }
1840
1841 void f2fs_build_gc_manager(struct f2fs_sb_info *sbi)
1842 {
1843         DIRTY_I(sbi)->v_ops = &default_v_ops;
1844
1845         sbi->gc_pin_file_threshold = DEF_GC_FAILED_PINNED_FILES;
1846
1847         /* give warm/cold data area from slower device */
1848         if (f2fs_is_multi_device(sbi) && !__is_large_section(sbi))
1849                 SIT_I(sbi)->last_victim[ALLOC_NEXT] =
1850                                 GET_SEGNO(sbi, FDEV(0).end_blk) + 1;
1851
1852         init_atgc_management(sbi);
1853 }
1854
1855 static int free_segment_range(struct f2fs_sb_info *sbi,
1856                                 unsigned int secs, bool gc_only)
1857 {
1858         unsigned int segno, next_inuse, start, end;
1859         struct cp_control cpc = { CP_RESIZE, 0, 0, 0 };
1860         int gc_mode, gc_type;
1861         int err = 0;
1862         int type;
1863
1864         /* Force block allocation for GC */
1865         MAIN_SECS(sbi) -= secs;
1866         start = MAIN_SECS(sbi) * sbi->segs_per_sec;
1867         end = MAIN_SEGS(sbi) - 1;
1868
1869         mutex_lock(&DIRTY_I(sbi)->seglist_lock);
1870         for (gc_mode = 0; gc_mode < MAX_GC_POLICY; gc_mode++)
1871                 if (SIT_I(sbi)->last_victim[gc_mode] >= start)
1872                         SIT_I(sbi)->last_victim[gc_mode] = 0;
1873
1874         for (gc_type = BG_GC; gc_type <= FG_GC; gc_type++)
1875                 if (sbi->next_victim_seg[gc_type] >= start)
1876                         sbi->next_victim_seg[gc_type] = NULL_SEGNO;
1877         mutex_unlock(&DIRTY_I(sbi)->seglist_lock);
1878
1879         /* Move out cursegs from the target range */
1880         for (type = CURSEG_HOT_DATA; type < NR_CURSEG_PERSIST_TYPE; type++)
1881                 f2fs_allocate_segment_for_resize(sbi, type, start, end);
1882
1883         /* do GC to move out valid blocks in the range */
1884         for (segno = start; segno <= end; segno += sbi->segs_per_sec) {
1885                 struct gc_inode_list gc_list = {
1886                         .ilist = LIST_HEAD_INIT(gc_list.ilist),
1887                         .iroot = RADIX_TREE_INIT(gc_list.iroot, GFP_NOFS),
1888                 };
1889
1890                 do_garbage_collect(sbi, segno, &gc_list, FG_GC, true);
1891                 put_gc_inode(&gc_list);
1892
1893                 if (!gc_only && get_valid_blocks(sbi, segno, true)) {
1894                         err = -EAGAIN;
1895                         goto out;
1896                 }
1897                 if (fatal_signal_pending(current)) {
1898                         err = -ERESTARTSYS;
1899                         goto out;
1900                 }
1901         }
1902         if (gc_only)
1903                 goto out;
1904
1905         err = f2fs_write_checkpoint(sbi, &cpc);
1906         if (err)
1907                 goto out;
1908
1909         next_inuse = find_next_inuse(FREE_I(sbi), end + 1, start);
1910         if (next_inuse <= end) {
1911                 f2fs_err(sbi, "segno %u should be free but still inuse!",
1912                          next_inuse);
1913                 f2fs_bug_on(sbi, 1);
1914         }
1915 out:
1916         MAIN_SECS(sbi) += secs;
1917         return err;
1918 }
1919
1920 static void update_sb_metadata(struct f2fs_sb_info *sbi, int secs)
1921 {
1922         struct f2fs_super_block *raw_sb = F2FS_RAW_SUPER(sbi);
1923         int section_count;
1924         int segment_count;
1925         int segment_count_main;
1926         long long block_count;
1927         int segs = secs * sbi->segs_per_sec;
1928
1929         down_write(&sbi->sb_lock);
1930
1931         section_count = le32_to_cpu(raw_sb->section_count);
1932         segment_count = le32_to_cpu(raw_sb->segment_count);
1933         segment_count_main = le32_to_cpu(raw_sb->segment_count_main);
1934         block_count = le64_to_cpu(raw_sb->block_count);
1935
1936         raw_sb->section_count = cpu_to_le32(section_count + secs);
1937         raw_sb->segment_count = cpu_to_le32(segment_count + segs);
1938         raw_sb->segment_count_main = cpu_to_le32(segment_count_main + segs);
1939         raw_sb->block_count = cpu_to_le64(block_count +
1940                                         (long long)segs * sbi->blocks_per_seg);
1941         if (f2fs_is_multi_device(sbi)) {
1942                 int last_dev = sbi->s_ndevs - 1;
1943                 int dev_segs =
1944                         le32_to_cpu(raw_sb->devs[last_dev].total_segments);
1945
1946                 raw_sb->devs[last_dev].total_segments =
1947                                                 cpu_to_le32(dev_segs + segs);
1948         }
1949
1950         up_write(&sbi->sb_lock);
1951 }
1952
1953 static void update_fs_metadata(struct f2fs_sb_info *sbi, int secs)
1954 {
1955         int segs = secs * sbi->segs_per_sec;
1956         long long blks = (long long)segs * sbi->blocks_per_seg;
1957         long long user_block_count =
1958                                 le64_to_cpu(F2FS_CKPT(sbi)->user_block_count);
1959
1960         SM_I(sbi)->segment_count = (int)SM_I(sbi)->segment_count + segs;
1961         MAIN_SEGS(sbi) = (int)MAIN_SEGS(sbi) + segs;
1962         MAIN_SECS(sbi) += secs;
1963         FREE_I(sbi)->free_sections = (int)FREE_I(sbi)->free_sections + secs;
1964         FREE_I(sbi)->free_segments = (int)FREE_I(sbi)->free_segments + segs;
1965         F2FS_CKPT(sbi)->user_block_count = cpu_to_le64(user_block_count + blks);
1966
1967         if (f2fs_is_multi_device(sbi)) {
1968                 int last_dev = sbi->s_ndevs - 1;
1969
1970                 FDEV(last_dev).total_segments =
1971                                 (int)FDEV(last_dev).total_segments + segs;
1972                 FDEV(last_dev).end_blk =
1973                                 (long long)FDEV(last_dev).end_blk + blks;
1974 #ifdef CONFIG_BLK_DEV_ZONED
1975                 FDEV(last_dev).nr_blkz = (int)FDEV(last_dev).nr_blkz +
1976                                         (int)(blks >> sbi->log_blocks_per_blkz);
1977 #endif
1978         }
1979 }
1980
1981 int f2fs_resize_fs(struct file *filp, __u64 block_count)
1982 {
1983         struct f2fs_sb_info *sbi = F2FS_I_SB(file_inode(filp));
1984         __u64 old_block_count, shrunk_blocks;
1985         struct cp_control cpc = { CP_RESIZE, 0, 0, 0 };
1986         unsigned int secs;
1987         int err = 0;
1988         __u32 rem;
1989
1990         old_block_count = le64_to_cpu(F2FS_RAW_SUPER(sbi)->block_count);
1991         if (block_count > old_block_count)
1992                 return -EINVAL;
1993
1994         if (f2fs_is_multi_device(sbi)) {
1995                 int last_dev = sbi->s_ndevs - 1;
1996                 __u64 last_segs = FDEV(last_dev).total_segments;
1997
1998                 if (block_count + last_segs * sbi->blocks_per_seg <=
1999                                                                 old_block_count)
2000                         return -EINVAL;
2001         }
2002
2003         /* new fs size should align to section size */
2004         div_u64_rem(block_count, BLKS_PER_SEC(sbi), &rem);
2005         if (rem)
2006                 return -EINVAL;
2007
2008         if (block_count == old_block_count)
2009                 return 0;
2010
2011         if (is_sbi_flag_set(sbi, SBI_NEED_FSCK)) {
2012                 f2fs_err(sbi, "Should run fsck to repair first.");
2013                 return -EFSCORRUPTED;
2014         }
2015
2016         if (test_opt(sbi, DISABLE_CHECKPOINT)) {
2017                 f2fs_err(sbi, "Checkpoint should be enabled.");
2018                 return -EINVAL;
2019         }
2020
2021         err = mnt_want_write_file(filp);
2022         if (err)
2023                 return err;
2024
2025         shrunk_blocks = old_block_count - block_count;
2026         secs = div_u64(shrunk_blocks, BLKS_PER_SEC(sbi));
2027
2028         /* stop other GC */
2029         if (!down_write_trylock(&sbi->gc_lock)) {
2030                 err = -EAGAIN;
2031                 goto out_drop_write;
2032         }
2033
2034         /* stop CP to protect MAIN_SEC in free_segment_range */
2035         f2fs_lock_op(sbi);
2036
2037         spin_lock(&sbi->stat_lock);
2038         if (shrunk_blocks + valid_user_blocks(sbi) +
2039                 sbi->current_reserved_blocks + sbi->unusable_block_count +
2040                 F2FS_OPTION(sbi).root_reserved_blocks > sbi->user_block_count)
2041                 err = -ENOSPC;
2042         spin_unlock(&sbi->stat_lock);
2043
2044         if (err)
2045                 goto out_unlock;
2046
2047         err = free_segment_range(sbi, secs, true);
2048
2049 out_unlock:
2050         f2fs_unlock_op(sbi);
2051         up_write(&sbi->gc_lock);
2052 out_drop_write:
2053         mnt_drop_write_file(filp);
2054         if (err)
2055                 return err;
2056
2057         freeze_super(sbi->sb);
2058
2059         if (f2fs_readonly(sbi->sb)) {
2060                 thaw_super(sbi->sb);
2061                 return -EROFS;
2062         }
2063
2064         down_write(&sbi->gc_lock);
2065         mutex_lock(&sbi->cp_mutex);
2066
2067         spin_lock(&sbi->stat_lock);
2068         if (shrunk_blocks + valid_user_blocks(sbi) +
2069                 sbi->current_reserved_blocks + sbi->unusable_block_count +
2070                 F2FS_OPTION(sbi).root_reserved_blocks > sbi->user_block_count)
2071                 err = -ENOSPC;
2072         else
2073                 sbi->user_block_count -= shrunk_blocks;
2074         spin_unlock(&sbi->stat_lock);
2075         if (err)
2076                 goto out_err;
2077
2078         set_sbi_flag(sbi, SBI_IS_RESIZEFS);
2079         err = free_segment_range(sbi, secs, false);
2080         if (err)
2081                 goto recover_out;
2082
2083         update_sb_metadata(sbi, -secs);
2084
2085         err = f2fs_commit_super(sbi, false);
2086         if (err) {
2087                 update_sb_metadata(sbi, secs);
2088                 goto recover_out;
2089         }
2090
2091         update_fs_metadata(sbi, -secs);
2092         clear_sbi_flag(sbi, SBI_IS_RESIZEFS);
2093         set_sbi_flag(sbi, SBI_IS_DIRTY);
2094
2095         err = f2fs_write_checkpoint(sbi, &cpc);
2096         if (err) {
2097                 update_fs_metadata(sbi, secs);
2098                 update_sb_metadata(sbi, secs);
2099                 f2fs_commit_super(sbi, false);
2100         }
2101 recover_out:
2102         clear_sbi_flag(sbi, SBI_IS_RESIZEFS);
2103         if (err) {
2104                 set_sbi_flag(sbi, SBI_NEED_FSCK);
2105                 f2fs_err(sbi, "resize_fs failed, should run fsck to repair!");
2106
2107                 spin_lock(&sbi->stat_lock);
2108                 sbi->user_block_count += shrunk_blocks;
2109                 spin_unlock(&sbi->stat_lock);
2110         }
2111 out_err:
2112         mutex_unlock(&sbi->cp_mutex);
2113         up_write(&sbi->gc_lock);
2114         thaw_super(sbi->sb);
2115         return err;
2116 }