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
  17 #include "f2fs.h"
  18 #include "node.h"
  19 #include "segment.h"
  20 #include "gc.h"
  21 #include <trace/events/f2fs.h>
  22
  23 static int gc_thread_func(void *data)
  24 {
  25         struct f2fs_sb_info *sbi = data;
  26         struct f2fs_gc_kthread *gc_th = sbi->gc_thread;
  27         wait_queue_head_t *wq = &sbi->gc_thread->gc_wait_queue_head;
  28         unsigned int wait_ms;
  29
  30         wait_ms = gc_th->min_sleep_time;
  31
  32         set_freezable();
  33         do {
  34                 wait_event_interruptible_timeout(*wq,
  35                                 kthread_should_stop() || freezing(current) ||
  36                                 gc_th->gc_wake,
  37                                 msecs_to_jiffies(wait_ms));
  38
  39                 /* give it a try one time */
  40                 if (gc_th->gc_wake)
  41                         gc_th->gc_wake = 0;
  42
  43                 if (try_to_freeze()) {
  44                         stat_other_skip_bggc_count(sbi);
  45                         continue;
  46                 }
  47                 if (kthread_should_stop())
  48                         break;
  49
  50                 if (sbi->sb->s_writers.frozen >= SB_FREEZE_WRITE) {
  51                         increase_sleep_time(gc_th, &wait_ms);
  52                         stat_other_skip_bggc_count(sbi);
  53                         continue;
  54                 }
  55
  56                 if (time_to_inject(sbi, FAULT_CHECKPOINT)) {
  57                         f2fs_show_injection_info(FAULT_CHECKPOINT);
  58                         f2fs_stop_checkpoint(sbi, false);
  59                 }
  60
  61                 if (!sb_start_write_trylock(sbi->sb)) {
  62                         stat_other_skip_bggc_count(sbi);
  63                         continue;
  64                 }
  65
  66                 /*
  67                  * [GC triggering condition]
  68                  * 0. GC is not conducted currently.
  69                  * 1. There are enough dirty segments.
  70                  * 2. IO subsystem is idle by checking the # of writeback pages.
  71                  * 3. IO subsystem is idle by checking the # of requests in
  72                  *    bdev's request list.
  73                  *
  74                  * Note) We have to avoid triggering GCs frequently.
  75                  * Because it is possible that some segments can be
  76                  * invalidated soon after by user update or deletion.
  77                  * So, I'd like to wait some time to collect dirty segments.
  78                  */
  79                 if (sbi->gc_mode == GC_URGENT) {
  80                         wait_ms = gc_th->urgent_sleep_time;
  81                         mutex_lock(&sbi->gc_mutex);
  82                         goto do_gc;
  83                 }
  84
  85                 if (!mutex_trylock(&sbi->gc_mutex)) {
  86                         stat_other_skip_bggc_count(sbi);
  87                         goto next;
  88                 }
  89
  90                 if (!is_idle(sbi, GC_TIME)) {
  91                         increase_sleep_time(gc_th, &wait_ms);
  92                         mutex_unlock(&sbi->gc_mutex);
  93                         stat_io_skip_bggc_count(sbi);
  94                         goto next;
  95                 }
  96
  97                 if (has_enough_invalid_blocks(sbi))
  98                         decrease_sleep_time(gc_th, &wait_ms);
  99                 else
 100                         increase_sleep_time(gc_th, &wait_ms);
 101 do_gc:
 102                 stat_inc_bggc_count(sbi);
 103
 104                 /* if return value is not zero, no victim was selected */
 105                 if (f2fs_gc(sbi, test_opt(sbi, FORCE_FG_GC), true, NULL_SEGNO))
 106                         wait_ms = gc_th->no_gc_sleep_time;
 107
 108                 trace_f2fs_background_gc(sbi->sb, wait_ms,
 109                                 prefree_segments(sbi), free_segments(sbi));
 110
 111                 /* balancing f2fs's metadata periodically */
 112                 f2fs_balance_fs_bg(sbi);
 113 next:
 114                 sb_end_write(sbi->sb);
 115
 116         } while (!kthread_should_stop());
 117         return 0;
 118 }
 119
 120 int f2fs_start_gc_thread(struct f2fs_sb_info *sbi)
 121 {
 122         struct f2fs_gc_kthread *gc_th;
 123         dev_t dev = sbi->sb->s_bdev->bd_dev;
 124         int err = 0;
 125
 126         gc_th = f2fs_kmalloc(sbi, sizeof(struct f2fs_gc_kthread), GFP_KERNEL);
 127         if (!gc_th) {
 128                 err = -ENOMEM;
 129                 goto out;
 130         }
 131
 132         gc_th->urgent_sleep_time = DEF_GC_THREAD_URGENT_SLEEP_TIME;
 133         gc_th->min_sleep_time = DEF_GC_THREAD_MIN_SLEEP_TIME;
 134         gc_th->max_sleep_time = DEF_GC_THREAD_MAX_SLEEP_TIME;
 135         gc_th->no_gc_sleep_time = DEF_GC_THREAD_NOGC_SLEEP_TIME;
 136
 137         gc_th->gc_wake= 0;
 138
 139         sbi->gc_thread = gc_th;
 140         init_waitqueue_head(&sbi->gc_thread->gc_wait_queue_head);
 141         sbi->gc_thread->f2fs_gc_task = kthread_run(gc_thread_func, sbi,
 142                         "f2fs_gc-%u:%u", MAJOR(dev), MINOR(dev));
 143         if (IS_ERR(gc_th->f2fs_gc_task)) {
 144                 err = PTR_ERR(gc_th->f2fs_gc_task);
 145                 kvfree(gc_th);
 146                 sbi->gc_thread = NULL;
 147         }
 148 out:
 149         return err;
 150 }
 151
 152 void f2fs_stop_gc_thread(struct f2fs_sb_info *sbi)
 153 {
 154         struct f2fs_gc_kthread *gc_th = sbi->gc_thread;
 155         if (!gc_th)
 156                 return;
 157         kthread_stop(gc_th->f2fs_gc_task);
 158         kvfree(gc_th);
 159         sbi->gc_thread = NULL;
 160 }
 161
 162 static int select_gc_type(struct f2fs_sb_info *sbi, int gc_type)
 163 {
 164         int gc_mode = (gc_type == BG_GC) ? GC_CB : GC_GREEDY;
 165
 166         switch (sbi->gc_mode) {
 167         case GC_IDLE_CB:
 168                 gc_mode = GC_CB;
 169                 break;
 170         case GC_IDLE_GREEDY:
 171         case GC_URGENT:
 172                 gc_mode = GC_GREEDY;
 173                 break;
 174         }
 175         return gc_mode;
 176 }
 177
 178 static void select_policy(struct f2fs_sb_info *sbi, int gc_type,
 179                         int type, struct victim_sel_policy *p)
 180 {
 181         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
 182
 183         if (p->alloc_mode == SSR) {
 184                 p->gc_mode = GC_GREEDY;
 185                 p->dirty_segmap = dirty_i->dirty_segmap[type];
 186                 p->max_search = dirty_i->nr_dirty[type];
 187                 p->ofs_unit = 1;
 188         } else {
 189                 p->gc_mode = select_gc_type(sbi, gc_type);
 190                 p->dirty_segmap = dirty_i->dirty_segmap[DIRTY];
 191                 p->max_search = dirty_i->nr_dirty[DIRTY];
 192                 p->ofs_unit = sbi->segs_per_sec;
 193         }
 194
 195         /* we need to check every dirty segments in the FG_GC case */
 196         if (gc_type != FG_GC &&
 197                         (sbi->gc_mode != GC_URGENT) &&
 198                         p->max_search > sbi->max_victim_search)
 199                 p->max_search = sbi->max_victim_search;
 200
 201         /* let's select beginning hot/small space first in no_heap mode*/
 202         if (test_opt(sbi, NOHEAP) &&
 203                 (type == CURSEG_HOT_DATA || IS_NODESEG(type)))
 204                 p->offset = 0;
 205         else
 206                 p->offset = SIT_I(sbi)->last_victim[p->gc_mode];
 207 }
 208
 209 static unsigned int get_max_cost(struct f2fs_sb_info *sbi,
 210                                 struct victim_sel_policy *p)
 211 {
 212         /* SSR allocates in a segment unit */
 213         if (p->alloc_mode == SSR)
 214                 return sbi->blocks_per_seg;
 215         if (p->gc_mode == GC_GREEDY)
 216                 return 2 * sbi->blocks_per_seg * p->ofs_unit;
 217         else if (p->gc_mode == GC_CB)
 218                 return UINT_MAX;
 219         else /* No other gc_mode */
 220                 return 0;
 221 }
 222
 223 static unsigned int check_bg_victims(struct f2fs_sb_info *sbi)
 224 {
 225         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
 226         unsigned int secno;
 227
 228         /*
 229          * If the gc_type is FG_GC, we can select victim segments
 230          * selected by background GC before.
 231          * Those segments guarantee they have small valid blocks.
 232          */
 233         for_each_set_bit(secno, dirty_i->victim_secmap, MAIN_SECS(sbi)) {
 234                 if (sec_usage_check(sbi, secno))
 235                         continue;
 236                 clear_bit(secno, dirty_i->victim_secmap);
 237                 return GET_SEG_FROM_SEC(sbi, secno);
 238         }
 239         return NULL_SEGNO;
 240 }
 241
 242 static unsigned int get_cb_cost(struct f2fs_sb_info *sbi, unsigned int segno)
 243 {
 244         struct sit_info *sit_i = SIT_I(sbi);
 245         unsigned int secno = GET_SEC_FROM_SEG(sbi, segno);
 246         unsigned int start = GET_SEG_FROM_SEC(sbi, secno);
 247         unsigned long long mtime = 0;
 248         unsigned int vblocks;
 249         unsigned char age = 0;
 250         unsigned char u;
 251         unsigned int i;
 252
 253         for (i = 0; i < sbi->segs_per_sec; i++)
 254                 mtime += get_seg_entry(sbi, start + i)->mtime;
 255         vblocks = get_valid_blocks(sbi, segno, true);
 256
 257         mtime = div_u64(mtime, sbi->segs_per_sec);
 258         vblocks = div_u64(vblocks, sbi->segs_per_sec);
 259
 260         u = (vblocks * 100) >> sbi->log_blocks_per_seg;
 261
 262         /* Handle if the system time has changed by the user */
 263         if (mtime < sit_i->min_mtime)
 264                 sit_i->min_mtime = mtime;
 265         if (mtime > sit_i->max_mtime)
 266                 sit_i->max_mtime = mtime;
 267         if (sit_i->max_mtime != sit_i->min_mtime)
 268                 age = 100 - div64_u64(100 * (mtime - sit_i->min_mtime),
 269                                 sit_i->max_mtime - sit_i->min_mtime);
 270
 271         return UINT_MAX - ((100 * (100 - u) * age) / (100 + u));
 272 }
 273
 274 static inline unsigned int get_gc_cost(struct f2fs_sb_info *sbi,
 275                         unsigned int segno, struct victim_sel_policy *p)
 276 {
 277         if (p->alloc_mode == SSR)
 278                 return get_seg_entry(sbi, segno)->ckpt_valid_blocks;
 279
 280         /* alloc_mode == LFS */
 281         if (p->gc_mode == GC_GREEDY)
 282                 return get_valid_blocks(sbi, segno, true);
 283         else
 284                 return get_cb_cost(sbi, segno);
 285 }
 286
 287 static unsigned int count_bits(const unsigned long *addr,
 288                                 unsigned int offset, unsigned int len)
 289 {
 290         unsigned int end = offset + len, sum = 0;
 291
 292         while (offset < end) {
 293                 if (test_bit(offset++, addr))
 294                         ++sum;
 295         }
 296         return sum;
 297 }
 298
 299 /*
 300  * This function is called from two paths.
 301  * One is garbage collection and the other is SSR segment selection.
 302  * When it is called during GC, it just gets a victim segment
 303  * and it does not remove it from dirty seglist.
 304  * When it is called from SSR segment selection, it finds a segment
 305  * which has minimum valid blocks and removes it from dirty seglist.
 306  */
 307 static int get_victim_by_default(struct f2fs_sb_info *sbi,
 308                 unsigned int *result, int gc_type, int type, char alloc_mode)
 309 {
 310         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
 311         struct sit_info *sm = SIT_I(sbi);
 312         struct victim_sel_policy p;
 313         unsigned int secno, last_victim;
 314         unsigned int last_segment = MAIN_SEGS(sbi);
 315         unsigned int nsearched = 0;
 316
 317         mutex_lock(&dirty_i->seglist_lock);
 318
 319         p.alloc_mode = alloc_mode;
 320         select_policy(sbi, gc_type, type, &p);
 321
 322         p.min_segno = NULL_SEGNO;
 323         p.min_cost = get_max_cost(sbi, &p);
 324
 325         if (*result != NULL_SEGNO) {
 326                 if (get_valid_blocks(sbi, *result, false) &&
 327                         !sec_usage_check(sbi, GET_SEC_FROM_SEG(sbi, *result)))
 328                         p.min_segno = *result;
 329                 goto out;
 330         }
 331
 332         if (p.max_search == 0)
 333                 goto out;
 334
 335         if (__is_large_section(sbi) && p.alloc_mode == LFS) {
 336                 if (sbi->next_victim_seg[BG_GC] != NULL_SEGNO) {
 337                         p.min_segno = sbi->next_victim_seg[BG_GC];
 338                         *result = p.min_segno;
 339                         sbi->next_victim_seg[BG_GC] = NULL_SEGNO;
 340                         goto got_result;
 341                 }
 342                 if (gc_type == FG_GC &&
 343                                 sbi->next_victim_seg[FG_GC] != NULL_SEGNO) {
 344                         p.min_segno = sbi->next_victim_seg[FG_GC];
 345                         *result = p.min_segno;
 346                         sbi->next_victim_seg[FG_GC] = NULL_SEGNO;
 347                         goto got_result;
 348                 }
 349         }
 350
 351         last_victim = sm->last_victim[p.gc_mode];
 352         if (p.alloc_mode == LFS && gc_type == FG_GC) {
 353                 p.min_segno = check_bg_victims(sbi);
 354                 if (p.min_segno != NULL_SEGNO)
 355                         goto got_it;
 356         }
 357
 358         while (1) {
 359                 unsigned long cost;
 360                 unsigned int segno;
 361
 362                 segno = find_next_bit(p.dirty_segmap, last_segment, p.offset);
 363                 if (segno >= last_segment) {
 364                         if (sm->last_victim[p.gc_mode]) {
 365                                 last_segment =
 366                                         sm->last_victim[p.gc_mode];
 367                                 sm->last_victim[p.gc_mode] = 0;
 368                                 p.offset = 0;
 369                                 continue;
 370                         }
 371                         break;
 372                 }
 373
 374                 p.offset = segno + p.ofs_unit;
 375                 if (p.ofs_unit > 1) {
 376                         p.offset -= segno % p.ofs_unit;
 377                         nsearched += count_bits(p.dirty_segmap,
 378                                                 p.offset - p.ofs_unit,
 379                                                 p.ofs_unit);
 380                 } else {
 381                         nsearched++;
 382                 }
 383
 384                 secno = GET_SEC_FROM_SEG(sbi, segno);
 385
 386                 if (sec_usage_check(sbi, secno))
 387                         goto next;
 388                 /* Don't touch checkpointed data */
 389                 if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED) &&
 390                                         get_ckpt_valid_blocks(sbi, segno)))
 391                         goto next;
 392                 if (gc_type == BG_GC && test_bit(secno, dirty_i->victim_secmap))
 393                         goto next;
 394
 395                 cost = get_gc_cost(sbi, segno, &p);
 396
 397                 if (p.min_cost > cost) {
 398                         p.min_segno = segno;
 399                         p.min_cost = cost;
 400                 }
 401 next:
 402                 if (nsearched >= p.max_search) {
 403                         if (!sm->last_victim[p.gc_mode] && segno <= last_victim)
 404                                 sm->last_victim[p.gc_mode] = last_victim + 1;
 405                         else
 406                                 sm->last_victim[p.gc_mode] = segno + 1;
 407                         sm->last_victim[p.gc_mode] %= MAIN_SEGS(sbi);
 408                         break;
 409                 }
 410         }
 411         if (p.min_segno != NULL_SEGNO) {
 412 got_it:
 413                 *result = (p.min_segno / p.ofs_unit) * p.ofs_unit;
 414 got_result:
 415                 if (p.alloc_mode == LFS) {
 416                         secno = GET_SEC_FROM_SEG(sbi, p.min_segno);
 417                         if (gc_type == FG_GC)
 418                                 sbi->cur_victim_sec = secno;
 419                         else
 420                                 set_bit(secno, dirty_i->victim_secmap);
 421                 }
 422
 423         }
 424 out:
 425         if (p.min_segno != NULL_SEGNO)
 426                 trace_f2fs_get_victim(sbi->sb, type, gc_type, &p,
 427                                 sbi->cur_victim_sec,
 428                                 prefree_segments(sbi), free_segments(sbi));
 429         mutex_unlock(&dirty_i->seglist_lock);
 430
 431         return (p.min_segno == NULL_SEGNO) ? 0 : 1;
 432 }
 433
 434 static const struct victim_selection default_v_ops = {
 435         .get_victim = get_victim_by_default,
 436 };
 437
 438 static struct inode *find_gc_inode(struct gc_inode_list *gc_list, nid_t ino)
 439 {
 440         struct inode_entry *ie;
 441
 442         ie = radix_tree_lookup(&gc_list->iroot, ino);
 443         if (ie)
 444                 return ie->inode;
 445         return NULL;
 446 }
 447
 448 static void add_gc_inode(struct gc_inode_list *gc_list, struct inode *inode)
 449 {
 450         struct inode_entry *new_ie;
 451
 452         if (inode == find_gc_inode(gc_list, inode->i_ino)) {
 453                 iput(inode);
 454                 return;
 455         }
 456         new_ie = f2fs_kmem_cache_alloc(f2fs_inode_entry_slab, GFP_NOFS);
 457         new_ie->inode = inode;
 458
 459         f2fs_radix_tree_insert(&gc_list->iroot, inode->i_ino, new_ie);
 460         list_add_tail(&new_ie->list, &gc_list->ilist);
 461 }
 462
 463 static void put_gc_inode(struct gc_inode_list *gc_list)
 464 {
 465         struct inode_entry *ie, *next_ie;
 466         list_for_each_entry_safe(ie, next_ie, &gc_list->ilist, list) {
 467                 radix_tree_delete(&gc_list->iroot, ie->inode->i_ino);
 468                 iput(ie->inode);
 469                 list_del(&ie->list);
 470                 kmem_cache_free(f2fs_inode_entry_slab, ie);
 471         }
 472 }
 473
 474 static int check_valid_map(struct f2fs_sb_info *sbi,
 475                                 unsigned int segno, int offset)
 476 {
 477         struct sit_info *sit_i = SIT_I(sbi);
 478         struct seg_entry *sentry;
 479         int ret;
 480
 481         down_read(&sit_i->sentry_lock);
 482         sentry = get_seg_entry(sbi, segno);
 483         ret = f2fs_test_bit(offset, sentry->cur_valid_map);
 484         up_read(&sit_i->sentry_lock);
 485         return ret;
 486 }
 487
 488 /*
 489  * This function compares node address got in summary with that in NAT.
 490  * On validity, copy that node with cold status, otherwise (invalid node)
 491  * ignore that.
 492  */
 493 static int gc_node_segment(struct f2fs_sb_info *sbi,
 494                 struct f2fs_summary *sum, unsigned int segno, int gc_type)
 495 {
 496         struct f2fs_summary *entry;
 497         block_t start_addr;
 498         int off;
 499         int phase = 0;
 500         bool fggc = (gc_type == FG_GC);
 501         int submitted = 0;
 502
 503         start_addr = START_BLOCK(sbi, segno);
 504
 505 next_step:
 506         entry = sum;
 507
 508         if (fggc && phase == 2)
 509                 atomic_inc(&sbi->wb_sync_req[NODE]);
 510
 511         for (off = 0; off < sbi->blocks_per_seg; off++, entry++) {
 512                 nid_t nid = le32_to_cpu(entry->nid);
 513                 struct page *node_page;
 514                 struct node_info ni;
 515                 int err;
 516
 517                 /* stop BG_GC if there is not enough free sections. */
 518                 if (gc_type == BG_GC && has_not_enough_free_secs(sbi, 0, 0))
 519                         return submitted;
 520
 521                 if (check_valid_map(sbi, segno, off) == 0)
 522                         continue;
 523
 524                 if (phase == 0) {
 525                         f2fs_ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nid), 1,
 526                                                         META_NAT, true);
 527                         continue;
 528                 }
 529
 530                 if (phase == 1) {
 531                         f2fs_ra_node_page(sbi, nid);
 532                         continue;
 533                 }
 534
 535                 /* phase == 2 */
 536                 node_page = f2fs_get_node_page(sbi, nid);
 537                 if (IS_ERR(node_page))
 538                         continue;
 539
 540                 /* block may become invalid during f2fs_get_node_page */
 541                 if (check_valid_map(sbi, segno, off) == 0) {
 542                         f2fs_put_page(node_page, 1);
 543                         continue;
 544                 }
 545
 546                 if (f2fs_get_node_info(sbi, nid, &ni)) {
 547                         f2fs_put_page(node_page, 1);
 548                         continue;
 549                 }
 550
 551                 if (ni.blk_addr != start_addr + off) {
 552                         f2fs_put_page(node_page, 1);
 553                         continue;
 554                 }
 555
 556                 err = f2fs_move_node_page(node_page, gc_type);
 557                 if (!err && gc_type == FG_GC)
 558                         submitted++;
 559                 stat_inc_node_blk_count(sbi, 1, gc_type);
 560         }
 561
 562         if (++phase < 3)
 563                 goto next_step;
 564
 565         if (fggc)
 566                 atomic_dec(&sbi->wb_sync_req[NODE]);
 567         return submitted;
 568 }
 569
 570 /*
 571  * Calculate start block index indicating the given node offset.
 572  * Be careful, caller should give this node offset only indicating direct node
 573  * blocks. If any node offsets, which point the other types of node blocks such
 574  * as indirect or double indirect node blocks, are given, it must be a caller's
 575  * bug.
 576  */
 577 block_t f2fs_start_bidx_of_node(unsigned int node_ofs, struct inode *inode)
 578 {
 579         unsigned int indirect_blks = 2 * NIDS_PER_BLOCK + 4;
 580         unsigned int bidx;
 581
 582         if (node_ofs == 0)
 583                 return 0;
 584
 585         if (node_ofs <= 2) {
 586                 bidx = node_ofs - 1;
 587         } else if (node_ofs <= indirect_blks) {
 588                 int dec = (node_ofs - 4) / (NIDS_PER_BLOCK + 1);
 589                 bidx = node_ofs - 2 - dec;
 590         } else {
 591                 int dec = (node_ofs - indirect_blks - 3) / (NIDS_PER_BLOCK + 1);
 592                 bidx = node_ofs - 5 - dec;
 593         }
 594         return bidx * ADDRS_PER_BLOCK(inode) + ADDRS_PER_INODE(inode);
 595 }
 596
 597 static bool is_alive(struct f2fs_sb_info *sbi, struct f2fs_summary *sum,
 598                 struct node_info *dni, block_t blkaddr, unsigned int *nofs)
 599 {
 600         struct page *node_page;
 601         nid_t nid;
 602         unsigned int ofs_in_node;
 603         block_t source_blkaddr;
 604
 605         nid = le32_to_cpu(sum->nid);
 606         ofs_in_node = le16_to_cpu(sum->ofs_in_node);
 607
 608         node_page = f2fs_get_node_page(sbi, nid);
 609         if (IS_ERR(node_page))
 610                 return false;
 611
 612         if (f2fs_get_node_info(sbi, nid, dni)) {
 613                 f2fs_put_page(node_page, 1);
 614                 return false;
 615         }
 616
 617         if (sum->version != dni->version) {
 618                 f2fs_msg(sbi->sb, KERN_WARNING,
 619                                 "%s: valid data with mismatched node version.",
 620                                 __func__);
 621                 set_sbi_flag(sbi, SBI_NEED_FSCK);
 622         }
 623
 624         *nofs = ofs_of_node(node_page);
 625         source_blkaddr = datablock_addr(NULL, node_page, ofs_in_node);
 626         f2fs_put_page(node_page, 1);
 627
 628         if (source_blkaddr != blkaddr)
 629                 return false;
 630         return true;
 631 }
 632
 633 static int ra_data_block(struct inode *inode, pgoff_t index)
 634 {
 635         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 636         struct address_space *mapping = inode->i_mapping;
 637         struct dnode_of_data dn;
 638         struct page *page;
 639         struct extent_info ei = {0, 0, 0};
 640         struct f2fs_io_info fio = {
 641                 .sbi = sbi,
 642                 .ino = inode->i_ino,
 643                 .type = DATA,
 644                 .temp = COLD,
 645                 .op = REQ_OP_READ,
 646                 .op_flags = 0,
 647                 .encrypted_page = NULL,
 648                 .in_list = false,
 649                 .retry = false,
 650         };
 651         int err;
 652
 653         page = f2fs_grab_cache_page(mapping, index, true);
 654         if (!page)
 655                 return -ENOMEM;
 656
 657         if (f2fs_lookup_extent_cache(inode, index, &ei)) {
 658                 dn.data_blkaddr = ei.blk + index - ei.fofs;
 659                 if (unlikely(!f2fs_is_valid_blkaddr(sbi, dn.data_blkaddr,
 660                                                 DATA_GENERIC_ENHANCE_READ))) {
 661                         err = -EFAULT;
 662                         goto put_page;
 663                 }
 664                 goto got_it;
 665         }
 666
 667         set_new_dnode(&dn, inode, NULL, NULL, 0);
 668         err = f2fs_get_dnode_of_data(&dn, index, LOOKUP_NODE);
 669         if (err)
 670                 goto put_page;
 671         f2fs_put_dnode(&dn);
 672
 673         if (!__is_valid_data_blkaddr(dn.data_blkaddr)) {
 674                 err = -ENOENT;
 675                 goto put_page;
 676         }
 677         if (unlikely(!f2fs_is_valid_blkaddr(sbi, dn.data_blkaddr,
 678                                                 DATA_GENERIC_ENHANCE))) {
 679                 err = -EFAULT;
 680                 goto put_page;
 681         }
 682 got_it:
 683         /* read page */
 684         fio.page = page;
 685         fio.new_blkaddr = fio.old_blkaddr = dn.data_blkaddr;
 686
 687         /*
 688          * don't cache encrypted data into meta inode until previous dirty
 689          * data were writebacked to avoid racing between GC and flush.
 690          */
 691         f2fs_wait_on_page_writeback(page, DATA, true, true);
 692
 693         f2fs_wait_on_block_writeback(inode, dn.data_blkaddr);
 694
 695         fio.encrypted_page = f2fs_pagecache_get_page(META_MAPPING(sbi),
 696                                         dn.data_blkaddr,
 697                                         FGP_LOCK | FGP_CREAT, GFP_NOFS);
 698         if (!fio.encrypted_page) {
 699                 err = -ENOMEM;
 700                 goto put_page;
 701         }
 702
 703         err = f2fs_submit_page_bio(&fio);
 704         if (err)
 705                 goto put_encrypted_page;
 706         f2fs_put_page(fio.encrypted_page, 0);
 707         f2fs_put_page(page, 1);
 708         return 0;
 709 put_encrypted_page:
 710         f2fs_put_page(fio.encrypted_page, 1);
 711 put_page:
 712         f2fs_put_page(page, 1);
 713         return err;
 714 }
 715
 716 /*
 717  * Move data block via META_MAPPING while keeping locked data page.
 718  * This can be used to move blocks, aka LBAs, directly on disk.
 719  */
 720 static int move_data_block(struct inode *inode, block_t bidx,
 721                                 int gc_type, unsigned int segno, int off)
 722 {
 723         struct f2fs_io_info fio = {
 724                 .sbi = F2FS_I_SB(inode),
 725                 .ino = inode->i_ino,
 726                 .type = DATA,
 727                 .temp = COLD,
 728                 .op = REQ_OP_READ,
 729                 .op_flags = 0,
 730                 .encrypted_page = NULL,
 731                 .in_list = false,
 732                 .retry = false,
 733         };
 734         struct dnode_of_data dn;
 735         struct f2fs_summary sum;
 736         struct node_info ni;
 737         struct page *page, *mpage;
 738         block_t newaddr;
 739         int err = 0;
 740         bool lfs_mode = test_opt(fio.sbi, LFS);
 741
 742         /* do not read out */
 743         page = f2fs_grab_cache_page(inode->i_mapping, bidx, false);
 744         if (!page)
 745                 return -ENOMEM;
 746
 747         if (!check_valid_map(F2FS_I_SB(inode), segno, off)) {
 748                 err = -ENOENT;
 749                 goto out;
 750         }
 751
 752         if (f2fs_is_atomic_file(inode)) {
 753                 F2FS_I(inode)->i_gc_failures[GC_FAILURE_ATOMIC]++;
 754                 F2FS_I_SB(inode)->skipped_atomic_files[gc_type]++;
 755                 err = -EAGAIN;
 756                 goto out;
 757         }
 758
 759         if (f2fs_is_pinned_file(inode)) {
 760                 f2fs_pin_file_control(inode, true);
 761                 err = -EAGAIN;
 762                 goto out;
 763         }
 764
 765         set_new_dnode(&dn, inode, NULL, NULL, 0);
 766         err = f2fs_get_dnode_of_data(&dn, bidx, LOOKUP_NODE);
 767         if (err)
 768                 goto out;
 769
 770         if (unlikely(dn.data_blkaddr == NULL_ADDR)) {
 771                 ClearPageUptodate(page);
 772                 err = -ENOENT;
 773                 goto put_out;
 774         }
 775
 776         /*
 777          * don't cache encrypted data into meta inode until previous dirty
 778          * data were writebacked to avoid racing between GC and flush.
 779          */
 780         f2fs_wait_on_page_writeback(page, DATA, true, true);
 781
 782         f2fs_wait_on_block_writeback(inode, dn.data_blkaddr);
 783
 784         err = f2fs_get_node_info(fio.sbi, dn.nid, &ni);
 785         if (err)
 786                 goto put_out;
 787
 788         set_summary(&sum, dn.nid, dn.ofs_in_node, ni.version);
 789
 790         /* read page */
 791         fio.page = page;
 792         fio.new_blkaddr = fio.old_blkaddr = dn.data_blkaddr;
 793
 794         if (lfs_mode)
 795                 down_write(&fio.sbi->io_order_lock);
 796
 797         f2fs_allocate_data_block(fio.sbi, NULL, fio.old_blkaddr, &newaddr,
 798                                         &sum, CURSEG_COLD_DATA, NULL, false);
 799
 800         fio.encrypted_page = f2fs_pagecache_get_page(META_MAPPING(fio.sbi),
 801                                 newaddr, FGP_LOCK | FGP_CREAT, GFP_NOFS);
 802         if (!fio.encrypted_page) {
 803                 err = -ENOMEM;
 804                 goto recover_block;
 805         }
 806
 807         mpage = f2fs_pagecache_get_page(META_MAPPING(fio.sbi),
 808                                         fio.old_blkaddr, FGP_LOCK, GFP_NOFS);
 809         if (mpage) {
 810                 bool updated = false;
 811
 812                 if (PageUptodate(mpage)) {
 813                         memcpy(page_address(fio.encrypted_page),
 814                                         page_address(mpage), PAGE_SIZE);
 815                         updated = true;
 816                 }
 817                 f2fs_put_page(mpage, 1);
 818                 invalidate_mapping_pages(META_MAPPING(fio.sbi),
 819                                         fio.old_blkaddr, fio.old_blkaddr);
 820                 if (updated)
 821                         goto write_page;
 822         }
 823
 824         err = f2fs_submit_page_bio(&fio);
 825         if (err)
 826                 goto put_page_out;
 827
 828         /* write page */
 829         lock_page(fio.encrypted_page);
 830
 831         if (unlikely(fio.encrypted_page->mapping != META_MAPPING(fio.sbi))) {
 832                 err = -EIO;
 833                 goto put_page_out;
 834         }
 835         if (unlikely(!PageUptodate(fio.encrypted_page))) {
 836                 err = -EIO;
 837                 goto put_page_out;
 838         }
 839
 840 write_page:
 841         f2fs_wait_on_page_writeback(fio.encrypted_page, DATA, true, true);
 842         set_page_dirty(fio.encrypted_page);
 843         if (clear_page_dirty_for_io(fio.encrypted_page))
 844                 dec_page_count(fio.sbi, F2FS_DIRTY_META);
 845
 846         set_page_writeback(fio.encrypted_page);
 847         ClearPageError(page);
 848
 849         /* allocate block address */
 850         f2fs_wait_on_page_writeback(dn.node_page, NODE, true, true);
 851
 852         fio.op = REQ_OP_WRITE;
 853         fio.op_flags = REQ_SYNC;
 854         fio.new_blkaddr = newaddr;
 855         f2fs_submit_page_write(&fio);
 856         if (fio.retry) {
 857                 err = -EAGAIN;
 858                 if (PageWriteback(fio.encrypted_page))
 859                         end_page_writeback(fio.encrypted_page);
 860                 goto put_page_out;
 861         }
 862
 863         f2fs_update_iostat(fio.sbi, FS_GC_DATA_IO, F2FS_BLKSIZE);
 864
 865         f2fs_update_data_blkaddr(&dn, newaddr);
 866         set_inode_flag(inode, FI_APPEND_WRITE);
 867         if (page->index == 0)
 868                 set_inode_flag(inode, FI_FIRST_BLOCK_WRITTEN);
 869 put_page_out:
 870         f2fs_put_page(fio.encrypted_page, 1);
 871 recover_block:
 872         if (lfs_mode)
 873                 up_write(&fio.sbi->io_order_lock);
 874         if (err)
 875                 f2fs_do_replace_block(fio.sbi, &sum, newaddr, fio.old_blkaddr,
 876                                                                 true, true);
 877 put_out:
 878         f2fs_put_dnode(&dn);
 879 out:
 880         f2fs_put_page(page, 1);
 881         return err;
 882 }
 883
 884 static int move_data_page(struct inode *inode, block_t bidx, int gc_type,
 885                                                         unsigned int segno, int off)
 886 {
 887         struct page *page;
 888         int err = 0;
 889
 890         page = f2fs_get_lock_data_page(inode, bidx, true);
 891         if (IS_ERR(page))
 892                 return PTR_ERR(page);
 893
 894         if (!check_valid_map(F2FS_I_SB(inode), segno, off)) {
 895                 err = -ENOENT;
 896                 goto out;
 897         }
 898
 899         if (f2fs_is_atomic_file(inode)) {
 900                 F2FS_I(inode)->i_gc_failures[GC_FAILURE_ATOMIC]++;
 901                 F2FS_I_SB(inode)->skipped_atomic_files[gc_type]++;
 902                 err = -EAGAIN;
 903                 goto out;
 904         }
 905         if (f2fs_is_pinned_file(inode)) {
 906                 if (gc_type == FG_GC)
 907                         f2fs_pin_file_control(inode, true);
 908                 err = -EAGAIN;
 909                 goto out;
 910         }
 911
 912         if (gc_type == BG_GC) {
 913                 if (PageWriteback(page)) {
 914                         err = -EAGAIN;
 915                         goto out;
 916                 }
 917                 set_page_dirty(page);
 918                 set_cold_data(page);
 919         } else {
 920                 struct f2fs_io_info fio = {
 921                         .sbi = F2FS_I_SB(inode),
 922                         .ino = inode->i_ino,
 923                         .type = DATA,
 924                         .temp = COLD,
 925                         .op = REQ_OP_WRITE,
 926                         .op_flags = REQ_SYNC,
 927                         .old_blkaddr = NULL_ADDR,
 928                         .page = page,
 929                         .encrypted_page = NULL,
 930                         .need_lock = LOCK_REQ,
 931                         .io_type = FS_GC_DATA_IO,
 932                 };
 933                 bool is_dirty = PageDirty(page);
 934
 935 retry:
 936                 f2fs_wait_on_page_writeback(page, DATA, true, true);
 937
 938                 set_page_dirty(page);
 939                 if (clear_page_dirty_for_io(page)) {
 940                         inode_dec_dirty_pages(inode);
 941                         f2fs_remove_dirty_inode(inode);
 942                 }
 943
 944                 set_cold_data(page);
 945
 946                 err = f2fs_do_write_data_page(&fio);
 947                 if (err) {
 948                         clear_cold_data(page);
 949                         if (err == -ENOMEM) {
 950                                 congestion_wait(BLK_RW_ASYNC, HZ/50);
 951                                 goto retry;
 952                         }
 953                         if (is_dirty)
 954                                 set_page_dirty(page);
 955                 }
 956         }
 957 out:
 958         f2fs_put_page(page, 1);
 959         return err;
 960 }
 961
 962 /*
 963  * This function tries to get parent node of victim data block, and identifies
 964  * data block validity. If the block is valid, copy that with cold status and
 965  * modify parent node.
 966  * If the parent node is not valid or the data block address is different,
 967  * the victim data block is ignored.
 968  */
 969 static int gc_data_segment(struct f2fs_sb_info *sbi, struct f2fs_summary *sum,
 970                 struct gc_inode_list *gc_list, unsigned int segno, int gc_type)
 971 {
 972         struct super_block *sb = sbi->sb;
 973         struct f2fs_summary *entry;
 974         block_t start_addr;
 975         int off;
 976         int phase = 0;
 977         int submitted = 0;
 978
 979         start_addr = START_BLOCK(sbi, segno);
 980
 981 next_step:
 982         entry = sum;
 983
 984         for (off = 0; off < sbi->blocks_per_seg; off++, entry++) {
 985                 struct page *data_page;
 986                 struct inode *inode;
 987                 struct node_info dni; /* dnode info for the data */
 988                 unsigned int ofs_in_node, nofs;
 989                 block_t start_bidx;
 990                 nid_t nid = le32_to_cpu(entry->nid);
 991
 992                 /* stop BG_GC if there is not enough free sections. */
 993                 if (gc_type == BG_GC && has_not_enough_free_secs(sbi, 0, 0))
 994                         return submitted;
 995
 996                 if (check_valid_map(sbi, segno, off) == 0)
 997                         continue;
 998
 999                 if (phase == 0) {
1000                         f2fs_ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nid), 1,
1001                                                         META_NAT, true);
1002                         continue;
1003                 }
1004
1005                 if (phase == 1) {
1006                         f2fs_ra_node_page(sbi, nid);
1007                         continue;
1008                 }
1009
1010                 /* Get an inode by ino with checking validity */
1011                 if (!is_alive(sbi, entry, &dni, start_addr + off, &nofs))
1012                         continue;
1013
1014                 if (phase == 2) {
1015                         f2fs_ra_node_page(sbi, dni.ino);
1016                         continue;
1017                 }
1018
1019                 ofs_in_node = le16_to_cpu(entry->ofs_in_node);
1020
1021                 if (phase == 3) {
1022                         inode = f2fs_iget(sb, dni.ino);
1023                         if (IS_ERR(inode) || is_bad_inode(inode))
1024                                 continue;
1025
1026                         if (!down_write_trylock(
1027                                 &F2FS_I(inode)->i_gc_rwsem[WRITE])) {
1028                                 iput(inode);
1029                                 sbi->skipped_gc_rwsem++;
1030                                 continue;
1031                         }
1032
1033                         start_bidx = f2fs_start_bidx_of_node(nofs, inode) +
1034                                                                 ofs_in_node;
1035
1036                         if (f2fs_post_read_required(inode)) {
1037                                 int err = ra_data_block(inode, start_bidx);
1038
1039                                 up_write(&F2FS_I(inode)->i_gc_rwsem[WRITE]);
1040                                 if (err) {
1041                                         iput(inode);
1042                                         continue;
1043                                 }
1044                                 add_gc_inode(gc_list, inode);
1045                                 continue;
1046                         }
1047
1048                         data_page = f2fs_get_read_data_page(inode,
1049                                                 start_bidx, REQ_RAHEAD, true);
1050                         up_write(&F2FS_I(inode)->i_gc_rwsem[WRITE]);
1051                         if (IS_ERR(data_page)) {
1052                                 iput(inode);
1053                                 continue;
1054                         }
1055
1056                         f2fs_put_page(data_page, 0);
1057                         add_gc_inode(gc_list, inode);
1058                         continue;
1059                 }
1060
1061                 /* phase 4 */
1062                 inode = find_gc_inode(gc_list, dni.ino);
1063                 if (inode) {
1064                         struct f2fs_inode_info *fi = F2FS_I(inode);
1065                         bool locked = false;
1066                         int err;
1067
1068                         if (S_ISREG(inode->i_mode)) {
1069                                 if (!down_write_trylock(&fi->i_gc_rwsem[READ]))
1070                                         continue;
1071                                 if (!down_write_trylock(
1072                                                 &fi->i_gc_rwsem[WRITE])) {
1073                                         sbi->skipped_gc_rwsem++;
1074                                         up_write(&fi->i_gc_rwsem[READ]);
1075                                         continue;
1076                                 }
1077                                 locked = true;
1078
1079                                 /* wait for all inflight aio data */
1080                                 inode_dio_wait(inode);
1081                         }
1082
1083                         start_bidx = f2fs_start_bidx_of_node(nofs, inode)
1084                                                                 + ofs_in_node;
1085                         if (f2fs_post_read_required(inode))
1086                                 err = move_data_block(inode, start_bidx,
1087                                                         gc_type, segno, off);
1088                         else
1089                                 err = move_data_page(inode, start_bidx, gc_type,
1090                                                                 segno, off);
1091
1092                         if (!err && (gc_type == FG_GC ||
1093                                         f2fs_post_read_required(inode)))
1094                                 submitted++;
1095
1096                         if (locked) {
1097                                 up_write(&fi->i_gc_rwsem[WRITE]);
1098                                 up_write(&fi->i_gc_rwsem[READ]);
1099                         }
1100
1101                         stat_inc_data_blk_count(sbi, 1, gc_type);
1102                 }
1103         }
1104
1105         if (++phase < 5)
1106                 goto next_step;
1107
1108         return submitted;
1109 }
1110
1111 static int __get_victim(struct f2fs_sb_info *sbi, unsigned int *victim,
1112                         int gc_type)
1113 {
1114         struct sit_info *sit_i = SIT_I(sbi);
1115         int ret;
1116
1117         down_write(&sit_i->sentry_lock);
1118         ret = DIRTY_I(sbi)->v_ops->get_victim(sbi, victim, gc_type,
1119                                               NO_CHECK_TYPE, LFS);
1120         up_write(&sit_i->sentry_lock);
1121         return ret;
1122 }
1123
1124 static int do_garbage_collect(struct f2fs_sb_info *sbi,
1125                                 unsigned int start_segno,
1126                                 struct gc_inode_list *gc_list, int gc_type)
1127 {
1128         struct page *sum_page;
1129         struct f2fs_summary_block *sum;
1130         struct blk_plug plug;
1131         unsigned int segno = start_segno;
1132         unsigned int end_segno = start_segno + sbi->segs_per_sec;
1133         int seg_freed = 0, migrated = 0;
1134         unsigned char type = IS_DATASEG(get_seg_entry(sbi, segno)->type) ?
1135                                                 SUM_TYPE_DATA : SUM_TYPE_NODE;
1136         int submitted = 0;
1137
1138         if (__is_large_section(sbi))
1139                 end_segno = rounddown(end_segno, sbi->segs_per_sec);
1140
1141         /* readahead multi ssa blocks those have contiguous address */
1142         if (__is_large_section(sbi))
1143                 f2fs_ra_meta_pages(sbi, GET_SUM_BLOCK(sbi, segno),
1144                                         end_segno - segno, META_SSA, true);
1145
1146         /* reference all summary page */
1147         while (segno < end_segno) {
1148                 sum_page = f2fs_get_sum_page(sbi, segno++);
1149                 if (IS_ERR(sum_page)) {
1150                         int err = PTR_ERR(sum_page);
1151
1152                         end_segno = segno - 1;
1153                         for (segno = start_segno; segno < end_segno; segno++) {
1154                                 sum_page = find_get_page(META_MAPPING(sbi),
1155                                                 GET_SUM_BLOCK(sbi, segno));
1156                                 f2fs_put_page(sum_page, 0);
1157                                 f2fs_put_page(sum_page, 0);
1158                         }
1159                         return err;
1160                 }
1161                 unlock_page(sum_page);
1162         }
1163
1164         blk_start_plug(&plug);
1165
1166         for (segno = start_segno; segno < end_segno; segno++) {
1167
1168                 /* find segment summary of victim */
1169                 sum_page = find_get_page(META_MAPPING(sbi),
1170                                         GET_SUM_BLOCK(sbi, segno));
1171                 f2fs_put_page(sum_page, 0);
1172
1173                 if (get_valid_blocks(sbi, segno, false) == 0)
1174                         goto freed;
1175                 if (__is_large_section(sbi) &&
1176                                 migrated >= sbi->migration_granularity)
1177                         goto skip;
1178                 if (!PageUptodate(sum_page) || unlikely(f2fs_cp_error(sbi)))
1179                         goto skip;
1180
1181                 sum = page_address(sum_page);
1182                 if (type != GET_SUM_TYPE((&sum->footer))) {
1183                         f2fs_msg(sbi->sb, KERN_ERR, "Inconsistent segment (%u) "
1184                                 "type [%d, %d] in SSA and SIT",
1185                                 segno, type, GET_SUM_TYPE((&sum->footer)));
1186                         set_sbi_flag(sbi, SBI_NEED_FSCK);
1187                         f2fs_stop_checkpoint(sbi, false);
1188                         goto skip;
1189                 }
1190
1191                 /*
1192                  * this is to avoid deadlock:
1193                  * - lock_page(sum_page)         - f2fs_replace_block
1194                  *  - check_valid_map()            - down_write(sentry_lock)
1195                  *   - down_read(sentry_lock)     - change_curseg()
1196                  *                                  - lock_page(sum_page)
1197                  */
1198                 if (type == SUM_TYPE_NODE)
1199                         submitted += gc_node_segment(sbi, sum->entries, segno,
1200                                                                 gc_type);
1201                 else
1202                         submitted += gc_data_segment(sbi, sum->entries, gc_list,
1203                                                         segno, gc_type);
1204
1205                 stat_inc_seg_count(sbi, type, gc_type);
1206
1207 freed:
1208                 if (gc_type == FG_GC &&
1209                                 get_valid_blocks(sbi, segno, false) == 0)
1210                         seg_freed++;
1211                 migrated++;
1212
1213                 if (__is_large_section(sbi) && segno + 1 < end_segno)
1214                         sbi->next_victim_seg[gc_type] = segno + 1;
1215 skip:
1216                 f2fs_put_page(sum_page, 0);
1217         }
1218
1219         if (submitted)
1220                 f2fs_submit_merged_write(sbi,
1221                                 (type == SUM_TYPE_NODE) ? NODE : DATA);
1222
1223         blk_finish_plug(&plug);
1224
1225         stat_inc_call_count(sbi->stat_info);
1226
1227         return seg_freed;
1228 }
1229
1230 int f2fs_gc(struct f2fs_sb_info *sbi, bool sync,
1231                         bool background, unsigned int segno)
1232 {
1233         int gc_type = sync ? FG_GC : BG_GC;
1234         int sec_freed = 0, seg_freed = 0, total_freed = 0;
1235         int ret = 0;
1236         struct cp_control cpc;
1237         unsigned int init_segno = segno;
1238         struct gc_inode_list gc_list = {
1239                 .ilist = LIST_HEAD_INIT(gc_list.ilist),
1240                 .iroot = RADIX_TREE_INIT(gc_list.iroot, GFP_NOFS),
1241         };
1242         unsigned long long last_skipped = sbi->skipped_atomic_files[FG_GC];
1243         unsigned long long first_skipped;
1244         unsigned int skipped_round = 0, round = 0;
1245
1246         trace_f2fs_gc_begin(sbi->sb, sync, background,
1247                                 get_pages(sbi, F2FS_DIRTY_NODES),
1248                                 get_pages(sbi, F2FS_DIRTY_DENTS),
1249                                 get_pages(sbi, F2FS_DIRTY_IMETA),
1250                                 free_sections(sbi),
1251                                 free_segments(sbi),
1252                                 reserved_segments(sbi),
1253                                 prefree_segments(sbi));
1254
1255         cpc.reason = __get_cp_reason(sbi);
1256         sbi->skipped_gc_rwsem = 0;
1257         first_skipped = last_skipped;
1258 gc_more:
1259         if (unlikely(!(sbi->sb->s_flags & SB_ACTIVE))) {
1260                 ret = -EINVAL;
1261                 goto stop;
1262         }
1263         if (unlikely(f2fs_cp_error(sbi))) {
1264                 ret = -EIO;
1265                 goto stop;
1266         }
1267
1268         if (gc_type == BG_GC && has_not_enough_free_secs(sbi, 0, 0)) {
1269                 /*
1270                  * For example, if there are many prefree_segments below given
1271                  * threshold, we can make them free by checkpoint. Then, we
1272                  * secure free segments which doesn't need fggc any more.
1273                  */
1274                 if (prefree_segments(sbi) &&
1275                                 !is_sbi_flag_set(sbi, SBI_CP_DISABLED)) {
1276                         ret = f2fs_write_checkpoint(sbi, &cpc);
1277                         if (ret)
1278                                 goto stop;
1279                 }
1280                 if (has_not_enough_free_secs(sbi, 0, 0))
1281                         gc_type = FG_GC;
1282         }
1283
1284         /* f2fs_balance_fs doesn't need to do BG_GC in critical path. */
1285         if (gc_type == BG_GC && !background) {
1286                 ret = -EINVAL;
1287                 goto stop;
1288         }
1289         if (!__get_victim(sbi, &segno, gc_type)) {
1290                 ret = -ENODATA;
1291                 goto stop;
1292         }
1293
1294         seg_freed = do_garbage_collect(sbi, segno, &gc_list, gc_type);
1295         if (gc_type == FG_GC && seg_freed == sbi->segs_per_sec)
1296                 sec_freed++;
1297         total_freed += seg_freed;
1298
1299         if (gc_type == FG_GC) {
1300                 if (sbi->skipped_atomic_files[FG_GC] > last_skipped ||
1301                                                 sbi->skipped_gc_rwsem)
1302                         skipped_round++;
1303                 last_skipped = sbi->skipped_atomic_files[FG_GC];
1304                 round++;
1305         }
1306
1307         if (gc_type == FG_GC)
1308                 sbi->cur_victim_sec = NULL_SEGNO;
1309
1310         if (sync)
1311                 goto stop;
1312
1313         if (has_not_enough_free_secs(sbi, sec_freed, 0)) {
1314                 if (skipped_round <= MAX_SKIP_GC_COUNT ||
1315                                         skipped_round * 2 < round) {
1316                         segno = NULL_SEGNO;
1317                         goto gc_more;
1318                 }
1319
1320                 if (first_skipped < last_skipped &&
1321                                 (last_skipped - first_skipped) >
1322                                                 sbi->skipped_gc_rwsem) {
1323                         f2fs_drop_inmem_pages_all(sbi, true);
1324                         segno = NULL_SEGNO;
1325                         goto gc_more;
1326                 }
1327                 if (gc_type == FG_GC && !is_sbi_flag_set(sbi, SBI_CP_DISABLED))
1328                         ret = f2fs_write_checkpoint(sbi, &cpc);
1329         }
1330 stop:
1331         SIT_I(sbi)->last_victim[ALLOC_NEXT] = 0;
1332         SIT_I(sbi)->last_victim[FLUSH_DEVICE] = init_segno;
1333
1334         trace_f2fs_gc_end(sbi->sb, ret, total_freed, sec_freed,
1335                                 get_pages(sbi, F2FS_DIRTY_NODES),
1336                                 get_pages(sbi, F2FS_DIRTY_DENTS),
1337                                 get_pages(sbi, F2FS_DIRTY_IMETA),
1338                                 free_sections(sbi),
1339                                 free_segments(sbi),
1340                                 reserved_segments(sbi),
1341                                 prefree_segments(sbi));
1342
1343         mutex_unlock(&sbi->gc_mutex);
1344
1345         put_gc_inode(&gc_list);
1346
1347         if (sync && !ret)
1348                 ret = sec_freed ? 0 : -EAGAIN;
1349         return ret;
1350 }
1351
1352 void f2fs_build_gc_manager(struct f2fs_sb_info *sbi)
1353 {
1354         DIRTY_I(sbi)->v_ops = &default_v_ops;
1355
1356         sbi->gc_pin_file_threshold = DEF_GC_FAILED_PINNED_FILES;
1357
1358         /* give warm/cold data area from slower device */
1359         if (f2fs_is_multi_device(sbi) && !__is_large_section(sbi))
1360                 SIT_I(sbi)->last_victim[ALLOC_NEXT] =
1361                                 GET_SEGNO(sbi, FDEV(0).end_blk) + 1;
1362 }