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