fs/f2fs/gc.c

   1 /*
   2  * fs/f2fs/gc.c
   3  *
   4  * Copyright (c) 2012 Samsung Electronics Co., Ltd.
   5  *             http://www.samsung.com/
   6  *
   7  * This program is free software; you can redistribute it and/or modify
   8  * it under the terms of the GNU General Public License version 2 as
   9  * published by the Free Software Foundation.
  10  */
  11 #include <linux/fs.h>
  12 #include <linux/module.h>
  13 #include <linux/backing-dev.h>
  14 #include <linux/init.h>
  15 #include <linux/f2fs_fs.h>
  16 #include <linux/kthread.h>
  17 #include <linux/delay.h>
  18 #include <linux/freezer.h>
  19 #include <linux/blkdev.h>
  20
  21 #include "f2fs.h"
  22 #include "node.h"
  23 #include "segment.h"
  24 #include "gc.h"
  25 #include <trace/events/f2fs.h>
  26
  27 static struct kmem_cache *winode_slab;
  28
  29 static int gc_thread_func(void *data)
  30 {
  31         struct f2fs_sb_info *sbi = data;
  32         struct f2fs_gc_kthread *gc_th = sbi->gc_thread;
  33         wait_queue_head_t *wq = &sbi->gc_thread->gc_wait_queue_head;
  34         long wait_ms;
  35
  36         wait_ms = gc_th->min_sleep_time;
  37
  38         do {
  39                 if (try_to_freeze())
  40                         continue;
  41                 else
  42                         wait_event_interruptible_timeout(*wq,
  43                                                 kthread_should_stop(),
  44                                                 msecs_to_jiffies(wait_ms));
  45                 if (kthread_should_stop())
  46                         break;
  47
  48                 if (sbi->sb->s_writers.frozen >= SB_FREEZE_WRITE) {
  49                         wait_ms = increase_sleep_time(gc_th, wait_ms);
  50                         continue;
  51                 }
  52
  53                 /*
  54                  * [GC triggering condition]
  55                  * 0. GC is not conducted currently.
  56                  * 1. There are enough dirty segments.
  57                  * 2. IO subsystem is idle by checking the # of writeback pages.
  58                  * 3. IO subsystem is idle by checking the # of requests in
  59                  *    bdev's request list.
  60                  *
  61                  * Note) We have to avoid triggering GCs too much frequently.
  62                  * Because it is possible that some segments can be
  63                  * invalidated soon after by user update or deletion.
  64                  * So, I'd like to wait some time to collect dirty segments.
  65                  */
  66                 if (!mutex_trylock(&sbi->gc_mutex))
  67                         continue;
  68
  69                 if (!is_idle(sbi)) {
  70                         wait_ms = increase_sleep_time(gc_th, wait_ms);
  71                         mutex_unlock(&sbi->gc_mutex);
  72                         continue;
  73                 }
  74
  75                 if (has_enough_invalid_blocks(sbi))
  76                         wait_ms = decrease_sleep_time(gc_th, wait_ms);
  77                 else
  78                         wait_ms = increase_sleep_time(gc_th, wait_ms);
  79
  80 #ifdef CONFIG_F2FS_STAT_FS
  81                 sbi->bg_gc++;
  82 #endif
  83
  84                 /* if return value is not zero, no victim was selected */
  85                 if (f2fs_gc(sbi))
  86                         wait_ms = gc_th->no_gc_sleep_time;
  87         } while (!kthread_should_stop());
  88         return 0;
  89 }
  90
  91 int start_gc_thread(struct f2fs_sb_info *sbi)
  92 {
  93         struct f2fs_gc_kthread *gc_th;
  94         dev_t dev = sbi->sb->s_bdev->bd_dev;
  95         int err = 0;
  96
  97         if (!test_opt(sbi, BG_GC))
  98                 goto out;
  99         gc_th = kmalloc(sizeof(struct f2fs_gc_kthread), GFP_KERNEL);
 100         if (!gc_th) {
 101                 err = -ENOMEM;
 102                 goto out;
 103         }
 104
 105         gc_th->min_sleep_time = DEF_GC_THREAD_MIN_SLEEP_TIME;
 106         gc_th->max_sleep_time = DEF_GC_THREAD_MAX_SLEEP_TIME;
 107         gc_th->no_gc_sleep_time = DEF_GC_THREAD_NOGC_SLEEP_TIME;
 108
 109         gc_th->gc_idle = 0;
 110
 111         sbi->gc_thread = gc_th;
 112         init_waitqueue_head(&sbi->gc_thread->gc_wait_queue_head);
 113         sbi->gc_thread->f2fs_gc_task = kthread_run(gc_thread_func, sbi,
 114                         "f2fs_gc-%u:%u", MAJOR(dev), MINOR(dev));
 115         if (IS_ERR(gc_th->f2fs_gc_task)) {
 116                 err = PTR_ERR(gc_th->f2fs_gc_task);
 117                 kfree(gc_th);
 118                 sbi->gc_thread = NULL;
 119         }
 120
 121 out:
 122         return err;
 123 }
 124
 125 void stop_gc_thread(struct f2fs_sb_info *sbi)
 126 {
 127         struct f2fs_gc_kthread *gc_th = sbi->gc_thread;
 128         if (!gc_th)
 129                 return;
 130         kthread_stop(gc_th->f2fs_gc_task);
 131         kfree(gc_th);
 132         sbi->gc_thread = NULL;
 133 }
 134
 135 static int select_gc_type(struct f2fs_gc_kthread *gc_th, int gc_type)
 136 {
 137         int gc_mode = (gc_type == BG_GC) ? GC_CB : GC_GREEDY;
 138
 139         if (gc_th && gc_th->gc_idle) {
 140                 if (gc_th->gc_idle == 1)
 141                         gc_mode = GC_CB;
 142                 else if (gc_th->gc_idle == 2)
 143                         gc_mode = GC_GREEDY;
 144         }
 145         return gc_mode;
 146 }
 147
 148 static void select_policy(struct f2fs_sb_info *sbi, int gc_type,
 149                         int type, struct victim_sel_policy *p)
 150 {
 151         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
 152
 153         if (p->alloc_mode == SSR) {
 154                 p->gc_mode = GC_GREEDY;
 155                 p->dirty_segmap = dirty_i->dirty_segmap[type];
 156                 p->max_search = dirty_i->nr_dirty[type];
 157                 p->ofs_unit = 1;
 158         } else {
 159                 p->gc_mode = select_gc_type(sbi->gc_thread, gc_type);
 160                 p->dirty_segmap = dirty_i->dirty_segmap[DIRTY];
 161                 p->max_search = dirty_i->nr_dirty[DIRTY];
 162                 p->ofs_unit = sbi->segs_per_sec;
 163         }
 164
 165         if (p->max_search > MAX_VICTIM_SEARCH)
 166                 p->max_search = MAX_VICTIM_SEARCH;
 167
 168         p->offset = sbi->last_victim[p->gc_mode];
 169 }
 170
 171 static unsigned int get_max_cost(struct f2fs_sb_info *sbi,
 172                                 struct victim_sel_policy *p)
 173 {
 174         /* SSR allocates in a segment unit */
 175         if (p->alloc_mode == SSR)
 176                 return 1 << sbi->log_blocks_per_seg;
 177         if (p->gc_mode == GC_GREEDY)
 178                 return (1 << sbi->log_blocks_per_seg) * p->ofs_unit;
 179         else if (p->gc_mode == GC_CB)
 180                 return UINT_MAX;
 181         else /* No other gc_mode */
 182                 return 0;
 183 }
 184
 185 static unsigned int check_bg_victims(struct f2fs_sb_info *sbi)
 186 {
 187         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
 188         unsigned int hint = 0;
 189         unsigned int secno;
 190
 191         /*
 192          * If the gc_type is FG_GC, we can select victim segments
 193          * selected by background GC before.
 194          * Those segments guarantee they have small valid blocks.
 195          */
 196 next:
 197         secno = find_next_bit(dirty_i->victim_secmap, TOTAL_SECS(sbi), hint++);
 198         if (secno < TOTAL_SECS(sbi)) {
 199                 if (sec_usage_check(sbi, secno))
 200                         goto next;
 201                 clear_bit(secno, dirty_i->victim_secmap);
 202                 return secno * sbi->segs_per_sec;
 203         }
 204         return NULL_SEGNO;
 205 }
 206
 207 static unsigned int get_cb_cost(struct f2fs_sb_info *sbi, unsigned int segno)
 208 {
 209         struct sit_info *sit_i = SIT_I(sbi);
 210         unsigned int secno = GET_SECNO(sbi, segno);
 211         unsigned int start = secno * sbi->segs_per_sec;
 212         unsigned long long mtime = 0;
 213         unsigned int vblocks;
 214         unsigned char age = 0;
 215         unsigned char u;
 216         unsigned int i;
 217
 218         for (i = 0; i < sbi->segs_per_sec; i++)
 219                 mtime += get_seg_entry(sbi, start + i)->mtime;
 220         vblocks = get_valid_blocks(sbi, segno, sbi->segs_per_sec);
 221
 222         mtime = div_u64(mtime, sbi->segs_per_sec);
 223         vblocks = div_u64(vblocks, sbi->segs_per_sec);
 224
 225         u = (vblocks * 100) >> sbi->log_blocks_per_seg;
 226
 227         /* Handle if the system time is changed by user */
 228         if (mtime < sit_i->min_mtime)
 229                 sit_i->min_mtime = mtime;
 230         if (mtime > sit_i->max_mtime)
 231                 sit_i->max_mtime = mtime;
 232         if (sit_i->max_mtime != sit_i->min_mtime)
 233                 age = 100 - div64_u64(100 * (mtime - sit_i->min_mtime),
 234                                 sit_i->max_mtime - sit_i->min_mtime);
 235
 236         return UINT_MAX - ((100 * (100 - u) * age) / (100 + u));
 237 }
 238
 239 static inline unsigned int get_gc_cost(struct f2fs_sb_info *sbi,
 240                         unsigned int segno, struct victim_sel_policy *p)
 241 {
 242         if (p->alloc_mode == SSR)
 243                 return get_seg_entry(sbi, segno)->ckpt_valid_blocks;
 244
 245         /* alloc_mode == LFS */
 246         if (p->gc_mode == GC_GREEDY)
 247                 return get_valid_blocks(sbi, segno, sbi->segs_per_sec);
 248         else
 249                 return get_cb_cost(sbi, segno);
 250 }
 251
 252 /*
 253  * This function is called from two paths.
 254  * One is garbage collection and the other is SSR segment selection.
 255  * When it is called during GC, it just gets a victim segment
 256  * and it does not remove it from dirty seglist.
 257  * When it is called from SSR segment selection, it finds a segment
 258  * which has minimum valid blocks and removes it from dirty seglist.
 259  */
 260 static int get_victim_by_default(struct f2fs_sb_info *sbi,
 261                 unsigned int *result, int gc_type, int type, char alloc_mode)
 262 {
 263         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
 264         struct victim_sel_policy p;
 265         unsigned int secno, max_cost;
 266         int nsearched = 0;
 267
 268         p.alloc_mode = alloc_mode;
 269         select_policy(sbi, gc_type, type, &p);
 270
 271         p.min_segno = NULL_SEGNO;
 272         p.min_cost = max_cost = get_max_cost(sbi, &p);
 273
 274         mutex_lock(&dirty_i->seglist_lock);
 275
 276         if (p.alloc_mode == LFS && gc_type == FG_GC) {
 277                 p.min_segno = check_bg_victims(sbi);
 278                 if (p.min_segno != NULL_SEGNO)
 279                         goto got_it;
 280         }
 281
 282         while (1) {
 283                 unsigned long cost;
 284                 unsigned int segno;
 285
 286                 segno = find_next_bit(p.dirty_segmap,
 287                                                 TOTAL_SEGS(sbi), p.offset);
 288                 if (segno >= TOTAL_SEGS(sbi)) {
 289                         if (sbi->last_victim[p.gc_mode]) {
 290                                 sbi->last_victim[p.gc_mode] = 0;
 291                                 p.offset = 0;
 292                                 continue;
 293                         }
 294                         break;
 295                 }
 296
 297                 p.offset = segno + p.ofs_unit;
 298                 if (p.ofs_unit > 1)
 299                         p.offset -= segno % p.ofs_unit;
 300
 301                 secno = GET_SECNO(sbi, segno);
 302
 303                 if (sec_usage_check(sbi, secno))
 304                         continue;
 305                 if (gc_type == BG_GC && test_bit(secno, dirty_i->victim_secmap))
 306                         continue;
 307
 308                 cost = get_gc_cost(sbi, segno, &p);
 309
 310                 if (p.min_cost > cost) {
 311                         p.min_segno = segno;
 312                         p.min_cost = cost;
 313                 } else if (unlikely(cost == max_cost)) {
 314                         continue;
 315                 }
 316
 317                 if (nsearched++ >= p.max_search) {
 318                         sbi->last_victim[p.gc_mode] = segno;
 319                         break;
 320                 }
 321         }
 322         if (p.min_segno != NULL_SEGNO) {
 323 got_it:
 324                 if (p.alloc_mode == LFS) {
 325                         secno = GET_SECNO(sbi, p.min_segno);
 326                         if (gc_type == FG_GC)
 327                                 sbi->cur_victim_sec = secno;
 328                         else
 329                                 set_bit(secno, dirty_i->victim_secmap);
 330                 }
 331                 *result = (p.min_segno / p.ofs_unit) * p.ofs_unit;
 332
 333                 trace_f2fs_get_victim(sbi->sb, type, gc_type, &p,
 334                                 sbi->cur_victim_sec,
 335                                 prefree_segments(sbi), free_segments(sbi));
 336         }
 337         mutex_unlock(&dirty_i->seglist_lock);
 338
 339         return (p.min_segno == NULL_SEGNO) ? 0 : 1;
 340 }
 341
 342 static const struct victim_selection default_v_ops = {
 343         .get_victim = get_victim_by_default,
 344 };
 345
 346 static struct inode *find_gc_inode(nid_t ino, struct list_head *ilist)
 347 {
 348         struct inode_entry *ie;
 349
 350         list_for_each_entry(ie, ilist, list)
 351                 if (ie->inode->i_ino == ino)
 352                         return ie->inode;
 353         return NULL;
 354 }
 355
 356 static void add_gc_inode(struct inode *inode, struct list_head *ilist)
 357 {
 358         struct inode_entry *new_ie;
 359
 360         if (inode == find_gc_inode(inode->i_ino, ilist)) {
 361                 iput(inode);
 362                 return;
 363         }
 364 repeat:
 365         new_ie = kmem_cache_alloc(winode_slab, GFP_NOFS);
 366         if (!new_ie) {
 367                 cond_resched();
 368                 goto repeat;
 369         }
 370         new_ie->inode = inode;
 371         list_add_tail(&new_ie->list, ilist);
 372 }
 373
 374 static void put_gc_inode(struct list_head *ilist)
 375 {
 376         struct inode_entry *ie, *next_ie;
 377         list_for_each_entry_safe(ie, next_ie, ilist, list) {
 378                 iput(ie->inode);
 379                 list_del(&ie->list);
 380                 kmem_cache_free(winode_slab, ie);
 381         }
 382 }
 383
 384 static int check_valid_map(struct f2fs_sb_info *sbi,
 385                                 unsigned int segno, int offset)
 386 {
 387         struct sit_info *sit_i = SIT_I(sbi);
 388         struct seg_entry *sentry;
 389         int ret;
 390
 391         mutex_lock(&sit_i->sentry_lock);
 392         sentry = get_seg_entry(sbi, segno);
 393         ret = f2fs_test_bit(offset, sentry->cur_valid_map);
 394         mutex_unlock(&sit_i->sentry_lock);
 395         return ret;
 396 }
 397
 398 /*
 399  * This function compares node address got in summary with that in NAT.
 400  * On validity, copy that node with cold status, otherwise (invalid node)
 401  * ignore that.
 402  */
 403 static void gc_node_segment(struct f2fs_sb_info *sbi,
 404                 struct f2fs_summary *sum, unsigned int segno, int gc_type)
 405 {
 406         bool initial = true;
 407         struct f2fs_summary *entry;
 408         int off;
 409
 410 next_step:
 411         entry = sum;
 412
 413         for (off = 0; off < sbi->blocks_per_seg; off++, entry++) {
 414                 nid_t nid = le32_to_cpu(entry->nid);
 415                 struct page *node_page;
 416
 417                 /* stop BG_GC if there is not enough free sections. */
 418                 if (gc_type == BG_GC && has_not_enough_free_secs(sbi, 0))
 419                         return;
 420
 421                 if (check_valid_map(sbi, segno, off) == 0)
 422                         continue;
 423
 424                 if (initial) {
 425                         ra_node_page(sbi, nid);
 426                         continue;
 427                 }
 428                 node_page = get_node_page(sbi, nid);
 429                 if (IS_ERR(node_page))
 430                         continue;
 431
 432                 /* set page dirty and write it */
 433                 if (gc_type == FG_GC) {
 434                         f2fs_wait_on_page_writeback(node_page, NODE, true);
 435                         set_page_dirty(node_page);
 436                 } else {
 437                         if (!PageWriteback(node_page))
 438                                 set_page_dirty(node_page);
 439                 }
 440                 f2fs_put_page(node_page, 1);
 441                 stat_inc_node_blk_count(sbi, 1);
 442         }
 443
 444         if (initial) {
 445                 initial = false;
 446                 goto next_step;
 447         }
 448
 449         if (gc_type == FG_GC) {
 450                 struct writeback_control wbc = {
 451                         .sync_mode = WB_SYNC_ALL,
 452                         .nr_to_write = LONG_MAX,
 453                         .for_reclaim = 0,
 454                 };
 455                 sync_node_pages(sbi, 0, &wbc);
 456
 457                 /*
 458                  * In the case of FG_GC, it'd be better to reclaim this victim
 459                  * completely.
 460                  */
 461                 if (get_valid_blocks(sbi, segno, 1) != 0)
 462                         goto next_step;
 463         }
 464 }
 465
 466 /*
 467  * Calculate start block index indicating the given node offset.
 468  * Be careful, caller should give this node offset only indicating direct node
 469  * blocks. If any node offsets, which point the other types of node blocks such
 470  * as indirect or double indirect node blocks, are given, it must be a caller's
 471  * bug.
 472  */
 473 block_t start_bidx_of_node(unsigned int node_ofs, struct f2fs_inode_info *fi)
 474 {
 475         unsigned int indirect_blks = 2 * NIDS_PER_BLOCK + 4;
 476         unsigned int bidx;
 477
 478         if (node_ofs == 0)
 479                 return 0;
 480
 481         if (node_ofs <= 2) {
 482                 bidx = node_ofs - 1;
 483         } else if (node_ofs <= indirect_blks) {
 484                 int dec = (node_ofs - 4) / (NIDS_PER_BLOCK + 1);
 485                 bidx = node_ofs - 2 - dec;
 486         } else {
 487                 int dec = (node_ofs - indirect_blks - 3) / (NIDS_PER_BLOCK + 1);
 488                 bidx = node_ofs - 5 - dec;
 489         }
 490         return bidx * ADDRS_PER_BLOCK + ADDRS_PER_INODE(fi);
 491 }
 492
 493 static int check_dnode(struct f2fs_sb_info *sbi, struct f2fs_summary *sum,
 494                 struct node_info *dni, block_t blkaddr, unsigned int *nofs)
 495 {
 496         struct page *node_page;
 497         nid_t nid;
 498         unsigned int ofs_in_node;
 499         block_t source_blkaddr;
 500
 501         nid = le32_to_cpu(sum->nid);
 502         ofs_in_node = le16_to_cpu(sum->ofs_in_node);
 503
 504         node_page = get_node_page(sbi, nid);
 505         if (IS_ERR(node_page))
 506                 return 0;
 507
 508         get_node_info(sbi, nid, dni);
 509
 510         if (sum->version != dni->version) {
 511                 f2fs_put_page(node_page, 1);
 512                 return 0;
 513         }
 514
 515         *nofs = ofs_of_node(node_page);
 516         source_blkaddr = datablock_addr(node_page, ofs_in_node);
 517         f2fs_put_page(node_page, 1);
 518
 519         if (source_blkaddr != blkaddr)
 520                 return 0;
 521         return 1;
 522 }
 523
 524 static void move_data_page(struct inode *inode, struct page *page, int gc_type)
 525 {
 526         if (gc_type == BG_GC) {
 527                 if (PageWriteback(page))
 528                         goto out;
 529                 set_page_dirty(page);
 530                 set_cold_data(page);
 531         } else {
 532                 struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
 533
 534                 f2fs_wait_on_page_writeback(page, DATA, true);
 535
 536                 if (clear_page_dirty_for_io(page) &&
 537                         S_ISDIR(inode->i_mode)) {
 538                         dec_page_count(sbi, F2FS_DIRTY_DENTS);
 539                         inode_dec_dirty_dents(inode);
 540                 }
 541                 set_cold_data(page);
 542                 do_write_data_page(page);
 543                 clear_cold_data(page);
 544         }
 545 out:
 546         f2fs_put_page(page, 1);
 547 }
 548
 549 /*
 550  * This function tries to get parent node of victim data block, and identifies
 551  * data block validity. If the block is valid, copy that with cold status and
 552  * modify parent node.
 553  * If the parent node is not valid or the data block address is different,
 554  * the victim data block is ignored.
 555  */
 556 static void gc_data_segment(struct f2fs_sb_info *sbi, struct f2fs_summary *sum,
 557                 struct list_head *ilist, unsigned int segno, int gc_type)
 558 {
 559         struct super_block *sb = sbi->sb;
 560         struct f2fs_summary *entry;
 561         block_t start_addr;
 562         int off;
 563         int phase = 0;
 564
 565         start_addr = START_BLOCK(sbi, segno);
 566
 567 next_step:
 568         entry = sum;
 569
 570         for (off = 0; off < sbi->blocks_per_seg; off++, entry++) {
 571                 struct page *data_page;
 572                 struct inode *inode;
 573                 struct node_info dni; /* dnode info for the data */
 574                 unsigned int ofs_in_node, nofs;
 575                 block_t start_bidx;
 576
 577                 /* stop BG_GC if there is not enough free sections. */
 578                 if (gc_type == BG_GC && has_not_enough_free_secs(sbi, 0))
 579                         return;
 580
 581                 if (check_valid_map(sbi, segno, off) == 0)
 582                         continue;
 583
 584                 if (phase == 0) {
 585                         ra_node_page(sbi, le32_to_cpu(entry->nid));
 586                         continue;
 587                 }
 588
 589                 /* Get an inode by ino with checking validity */
 590                 if (check_dnode(sbi, entry, &dni, start_addr + off, &nofs) == 0)
 591                         continue;
 592
 593                 if (phase == 1) {
 594                         ra_node_page(sbi, dni.ino);
 595                         continue;
 596                 }
 597
 598                 ofs_in_node = le16_to_cpu(entry->ofs_in_node);
 599
 600                 if (phase == 2) {
 601                         inode = f2fs_iget(sb, dni.ino);
 602                         if (IS_ERR(inode))
 603                                 continue;
 604
 605                         start_bidx = start_bidx_of_node(nofs, F2FS_I(inode));
 606
 607                         data_page = find_data_page(inode,
 608                                         start_bidx + ofs_in_node, false);
 609                         if (IS_ERR(data_page))
 610                                 goto next_iput;
 611
 612                         f2fs_put_page(data_page, 0);
 613                         add_gc_inode(inode, ilist);
 614                 } else {
 615                         inode = find_gc_inode(dni.ino, ilist);
 616                         if (inode) {
 617                                 start_bidx = start_bidx_of_node(nofs,
 618                                                                 F2FS_I(inode));
 619                                 data_page = get_lock_data_page(inode,
 620                                                 start_bidx + ofs_in_node);
 621                                 if (IS_ERR(data_page))
 622                                         continue;
 623                                 move_data_page(inode, data_page, gc_type);
 624                                 stat_inc_data_blk_count(sbi, 1);
 625                         }
 626                 }
 627                 continue;
 628 next_iput:
 629                 iput(inode);
 630         }
 631
 632         if (++phase < 4)
 633                 goto next_step;
 634
 635         if (gc_type == FG_GC) {
 636                 f2fs_submit_bio(sbi, DATA, true);
 637
 638                 /*
 639                  * In the case of FG_GC, it'd be better to reclaim this victim
 640                  * completely.
 641                  */
 642                 if (get_valid_blocks(sbi, segno, 1) != 0) {
 643                         phase = 2;
 644                         goto next_step;
 645                 }
 646         }
 647 }
 648
 649 static int __get_victim(struct f2fs_sb_info *sbi, unsigned int *victim,
 650                                                 int gc_type, int type)
 651 {
 652         struct sit_info *sit_i = SIT_I(sbi);
 653         int ret;
 654         mutex_lock(&sit_i->sentry_lock);
 655         ret = DIRTY_I(sbi)->v_ops->get_victim(sbi, victim, gc_type, type, LFS);
 656         mutex_unlock(&sit_i->sentry_lock);
 657         return ret;
 658 }
 659
 660 static void do_garbage_collect(struct f2fs_sb_info *sbi, unsigned int segno,
 661                                 struct list_head *ilist, int gc_type)
 662 {
 663         struct page *sum_page;
 664         struct f2fs_summary_block *sum;
 665         struct blk_plug plug;
 666
 667         /* read segment summary of victim */
 668         sum_page = get_sum_page(sbi, segno);
 669         if (IS_ERR(sum_page))
 670                 return;
 671
 672         blk_start_plug(&plug);
 673
 674         sum = page_address(sum_page);
 675
 676         switch (GET_SUM_TYPE((&sum->footer))) {
 677         case SUM_TYPE_NODE:
 678                 gc_node_segment(sbi, sum->entries, segno, gc_type);
 679                 break;
 680         case SUM_TYPE_DATA:
 681                 gc_data_segment(sbi, sum->entries, ilist, segno, gc_type);
 682                 break;
 683         }
 684         blk_finish_plug(&plug);
 685
 686         stat_inc_seg_count(sbi, GET_SUM_TYPE((&sum->footer)));
 687         stat_inc_call_count(sbi->stat_info);
 688
 689         f2fs_put_page(sum_page, 1);
 690 }
 691
 692 int f2fs_gc(struct f2fs_sb_info *sbi)
 693 {
 694         struct list_head ilist;
 695         unsigned int segno, i;
 696         int gc_type = BG_GC;
 697         int nfree = 0;
 698         int ret = -1;
 699
 700         INIT_LIST_HEAD(&ilist);
 701 gc_more:
 702         if (!(sbi->sb->s_flags & MS_ACTIVE))
 703                 goto stop;
 704
 705         if (gc_type == BG_GC && has_not_enough_free_secs(sbi, nfree)) {
 706                 gc_type = FG_GC;
 707                 write_checkpoint(sbi, false);
 708         }
 709
 710         if (!__get_victim(sbi, &segno, gc_type, NO_CHECK_TYPE))
 711                 goto stop;
 712         ret = 0;
 713
 714         for (i = 0; i < sbi->segs_per_sec; i++)
 715                 do_garbage_collect(sbi, segno + i, &ilist, gc_type);
 716
 717         if (gc_type == FG_GC) {
 718                 sbi->cur_victim_sec = NULL_SEGNO;
 719                 nfree++;
 720                 WARN_ON(get_valid_blocks(sbi, segno, sbi->segs_per_sec));
 721         }
 722
 723         if (has_not_enough_free_secs(sbi, nfree))
 724                 goto gc_more;
 725
 726         if (gc_type == FG_GC)
 727                 write_checkpoint(sbi, false);
 728 stop:
 729         mutex_unlock(&sbi->gc_mutex);
 730
 731         put_gc_inode(&ilist);
 732         return ret;
 733 }
 734
 735 void build_gc_manager(struct f2fs_sb_info *sbi)
 736 {
 737         DIRTY_I(sbi)->v_ops = &default_v_ops;
 738 }
 739
 740 int __init create_gc_caches(void)
 741 {
 742         winode_slab = f2fs_kmem_cache_create("f2fs_gc_inodes",
 743                         sizeof(struct inode_entry), NULL);
 744         if (!winode_slab)
 745                 return -ENOMEM;
 746         return 0;
 747 }
 748
 749 void destroy_gc_caches(void)
 750 {
 751         kmem_cache_destroy(winode_slab);
 752 }