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