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                 stat_inc_bggc_count(sbi);
  81
  82                 /* if return value is not zero, no victim was selected */
  83                 if (f2fs_gc(sbi))
  84                         wait_ms = gc_th->no_gc_sleep_time;
  85
  86                 /* balancing prefree segments */
  87                 if (excess_prefree_segs(sbi))
  88                         f2fs_sync_fs(sbi->sb, true);
  89
  90         } while (!kthread_should_stop());
  91         return 0;
  92 }
  93
  94 int start_gc_thread(struct f2fs_sb_info *sbi)
  95 {
  96         struct f2fs_gc_kthread *gc_th;
  97         dev_t dev = sbi->sb->s_bdev->bd_dev;
  98         int err = 0;
  99
 100         if (!test_opt(sbi, BG_GC))
 101                 goto out;
 102         gc_th = kmalloc(sizeof(struct f2fs_gc_kthread), GFP_KERNEL);
 103         if (!gc_th) {
 104                 err = -ENOMEM;
 105                 goto out;
 106         }
 107
 108         gc_th->min_sleep_time = DEF_GC_THREAD_MIN_SLEEP_TIME;
 109         gc_th->max_sleep_time = DEF_GC_THREAD_MAX_SLEEP_TIME;
 110         gc_th->no_gc_sleep_time = DEF_GC_THREAD_NOGC_SLEEP_TIME;
 111
 112         gc_th->gc_idle = 0;
 113
 114         sbi->gc_thread = gc_th;
 115         init_waitqueue_head(&sbi->gc_thread->gc_wait_queue_head);
 116         sbi->gc_thread->f2fs_gc_task = kthread_run(gc_thread_func, sbi,
 117                         "f2fs_gc-%u:%u", MAJOR(dev), MINOR(dev));
 118         if (IS_ERR(gc_th->f2fs_gc_task)) {
 119                 err = PTR_ERR(gc_th->f2fs_gc_task);
 120                 kfree(gc_th);
 121                 sbi->gc_thread = NULL;
 122         }
 123
 124 out:
 125         return err;
 126 }
 127
 128 void stop_gc_thread(struct f2fs_sb_info *sbi)
 129 {
 130         struct f2fs_gc_kthread *gc_th = sbi->gc_thread;
 131         if (!gc_th)
 132                 return;
 133         kthread_stop(gc_th->f2fs_gc_task);
 134         kfree(gc_th);
 135         sbi->gc_thread = NULL;
 136 }
 137
 138 static int select_gc_type(struct f2fs_gc_kthread *gc_th, int gc_type)
 139 {
 140         int gc_mode = (gc_type == BG_GC) ? GC_CB : GC_GREEDY;
 141
 142         if (gc_th && gc_th->gc_idle) {
 143                 if (gc_th->gc_idle == 1)
 144                         gc_mode = GC_CB;
 145                 else if (gc_th->gc_idle == 2)
 146                         gc_mode = GC_GREEDY;
 147         }
 148         return gc_mode;
 149 }
 150
 151 static void select_policy(struct f2fs_sb_info *sbi, int gc_type,
 152                         int type, struct victim_sel_policy *p)
 153 {
 154         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
 155
 156         if (p->alloc_mode == SSR) {
 157                 p->gc_mode = GC_GREEDY;
 158                 p->dirty_segmap = dirty_i->dirty_segmap[type];
 159                 p->max_search = dirty_i->nr_dirty[type];
 160                 p->ofs_unit = 1;
 161         } else {
 162                 p->gc_mode = select_gc_type(sbi->gc_thread, gc_type);
 163                 p->dirty_segmap = dirty_i->dirty_segmap[DIRTY];
 164                 p->max_search = dirty_i->nr_dirty[DIRTY];
 165                 p->ofs_unit = sbi->segs_per_sec;
 166         }
 167
 168         if (p->max_search > MAX_VICTIM_SEARCH)
 169                 p->max_search = MAX_VICTIM_SEARCH;
 170
 171         p->offset = sbi->last_victim[p->gc_mode];
 172 }
 173
 174 static unsigned int get_max_cost(struct f2fs_sb_info *sbi,
 175                                 struct victim_sel_policy *p)
 176 {
 177         /* SSR allocates in a segment unit */
 178         if (p->alloc_mode == SSR)
 179                 return 1 << sbi->log_blocks_per_seg;
 180         if (p->gc_mode == GC_GREEDY)
 181                 return (1 << sbi->log_blocks_per_seg) * p->ofs_unit;
 182         else if (p->gc_mode == GC_CB)
 183                 return UINT_MAX;
 184         else /* No other gc_mode */
 185                 return 0;
 186 }
 187
 188 static unsigned int check_bg_victims(struct f2fs_sb_info *sbi)
 189 {
 190         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
 191         unsigned int hint = 0;
 192         unsigned int secno;
 193
 194         /*
 195          * If the gc_type is FG_GC, we can select victim segments
 196          * selected by background GC before.
 197          * Those segments guarantee they have small valid blocks.
 198          */
 199 next:
 200         secno = find_next_bit(dirty_i->victim_secmap, TOTAL_SECS(sbi), hint++);
 201         if (secno < TOTAL_SECS(sbi)) {
 202                 if (sec_usage_check(sbi, secno))
 203                         goto next;
 204                 clear_bit(secno, dirty_i->victim_secmap);
 205                 return secno * sbi->segs_per_sec;
 206         }
 207         return NULL_SEGNO;
 208 }
 209
 210 static unsigned int get_cb_cost(struct f2fs_sb_info *sbi, unsigned int segno)
 211 {
 212         struct sit_info *sit_i = SIT_I(sbi);
 213         unsigned int secno = GET_SECNO(sbi, segno);
 214         unsigned int start = secno * sbi->segs_per_sec;
 215         unsigned long long mtime = 0;
 216         unsigned int vblocks;
 217         unsigned char age = 0;
 218         unsigned char u;
 219         unsigned int i;
 220
 221         for (i = 0; i < sbi->segs_per_sec; i++)
 222                 mtime += get_seg_entry(sbi, start + i)->mtime;
 223         vblocks = get_valid_blocks(sbi, segno, sbi->segs_per_sec);
 224
 225         mtime = div_u64(mtime, sbi->segs_per_sec);
 226         vblocks = div_u64(vblocks, sbi->segs_per_sec);
 227
 228         u = (vblocks * 100) >> sbi->log_blocks_per_seg;
 229
 230         /* Handle if the system time is changed by user */
 231         if (mtime < sit_i->min_mtime)
 232                 sit_i->min_mtime = mtime;
 233         if (mtime > sit_i->max_mtime)
 234                 sit_i->max_mtime = mtime;
 235         if (sit_i->max_mtime != sit_i->min_mtime)
 236                 age = 100 - div64_u64(100 * (mtime - sit_i->min_mtime),
 237                                 sit_i->max_mtime - sit_i->min_mtime);
 238
 239         return UINT_MAX - ((100 * (100 - u) * age) / (100 + u));
 240 }
 241
 242 static inline unsigned int get_gc_cost(struct f2fs_sb_info *sbi,
 243                         unsigned int segno, struct victim_sel_policy *p)
 244 {
 245         if (p->alloc_mode == SSR)
 246                 return get_seg_entry(sbi, segno)->ckpt_valid_blocks;
 247
 248         /* alloc_mode == LFS */
 249         if (p->gc_mode == GC_GREEDY)
 250                 return get_valid_blocks(sbi, segno, sbi->segs_per_sec);
 251         else
 252                 return get_cb_cost(sbi, segno);
 253 }
 254
 255 /*
 256  * This function is called from two paths.
 257  * One is garbage collection and the other is SSR segment selection.
 258  * When it is called during GC, it just gets a victim segment
 259  * and it does not remove it from dirty seglist.
 260  * When it is called from SSR segment selection, it finds a segment
 261  * which has minimum valid blocks and removes it from dirty seglist.
 262  */
 263 static int get_victim_by_default(struct f2fs_sb_info *sbi,
 264                 unsigned int *result, int gc_type, int type, char alloc_mode)
 265 {
 266         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
 267         struct victim_sel_policy p;
 268         unsigned int secno, max_cost;
 269         int nsearched = 0;
 270
 271         p.alloc_mode = alloc_mode;
 272         select_policy(sbi, gc_type, type, &p);
 273
 274         p.min_segno = NULL_SEGNO;
 275         p.min_cost = max_cost = get_max_cost(sbi, &p);
 276
 277         mutex_lock(&dirty_i->seglist_lock);
 278
 279         if (p.alloc_mode == LFS && gc_type == FG_GC) {
 280                 p.min_segno = check_bg_victims(sbi);
 281                 if (p.min_segno != NULL_SEGNO)
 282                         goto got_it;
 283         }
 284
 285         while (1) {
 286                 unsigned long cost;
 287                 unsigned int segno;
 288
 289                 segno = find_next_bit(p.dirty_segmap,
 290                                                 TOTAL_SEGS(sbi), p.offset);
 291                 if (segno >= TOTAL_SEGS(sbi)) {
 292                         if (sbi->last_victim[p.gc_mode]) {
 293                                 sbi->last_victim[p.gc_mode] = 0;
 294                                 p.offset = 0;
 295                                 continue;
 296                         }
 297                         break;
 298                 }
 299
 300                 p.offset = segno + p.ofs_unit;
 301                 if (p.ofs_unit > 1)
 302                         p.offset -= segno % p.ofs_unit;
 303
 304                 secno = GET_SECNO(sbi, segno);
 305
 306                 if (sec_usage_check(sbi, secno))
 307                         continue;
 308                 if (gc_type == BG_GC && test_bit(secno, dirty_i->victim_secmap))
 309                         continue;
 310
 311                 cost = get_gc_cost(sbi, segno, &p);
 312
 313                 if (p.min_cost > cost) {
 314                         p.min_segno = segno;
 315                         p.min_cost = cost;
 316                 } else if (unlikely(cost == max_cost)) {
 317                         continue;
 318                 }
 319
 320                 if (nsearched++ >= p.max_search) {
 321                         sbi->last_victim[p.gc_mode] = segno;
 322                         break;
 323                 }
 324         }
 325         if (p.min_segno != NULL_SEGNO) {
 326 got_it:
 327                 if (p.alloc_mode == LFS) {
 328                         secno = GET_SECNO(sbi, p.min_segno);
 329                         if (gc_type == FG_GC)
 330                                 sbi->cur_victim_sec = secno;
 331                         else
 332                                 set_bit(secno, dirty_i->victim_secmap);
 333                 }
 334                 *result = (p.min_segno / p.ofs_unit) * p.ofs_unit;
 335
 336                 trace_f2fs_get_victim(sbi->sb, type, gc_type, &p,
 337                                 sbi->cur_victim_sec,
 338                                 prefree_segments(sbi), free_segments(sbi));
 339         }
 340         mutex_unlock(&dirty_i->seglist_lock);
 341
 342         return (p.min_segno == NULL_SEGNO) ? 0 : 1;
 343 }
 344
 345 static const struct victim_selection default_v_ops = {
 346         .get_victim = get_victim_by_default,
 347 };
 348
 349 static struct inode *find_gc_inode(nid_t ino, struct list_head *ilist)
 350 {
 351         struct inode_entry *ie;
 352
 353         list_for_each_entry(ie, ilist, list)
 354                 if (ie->inode->i_ino == ino)
 355                         return ie->inode;
 356         return NULL;
 357 }
 358
 359 static void add_gc_inode(struct inode *inode, struct list_head *ilist)
 360 {
 361         struct inode_entry *new_ie;
 362
 363         if (inode == find_gc_inode(inode->i_ino, ilist)) {
 364                 iput(inode);
 365                 return;
 366         }
 367
 368         new_ie = f2fs_kmem_cache_alloc(winode_slab, GFP_NOFS);
 369         new_ie->inode = inode;
 370         list_add_tail(&new_ie->list, ilist);
 371 }
 372
 373 static void put_gc_inode(struct list_head *ilist)
 374 {
 375         struct inode_entry *ie, *next_ie;
 376         list_for_each_entry_safe(ie, next_ie, ilist, list) {
 377                 iput(ie->inode);
 378                 list_del(&ie->list);
 379                 kmem_cache_free(winode_slab, ie);
 380         }
 381 }
 382
 383 static int check_valid_map(struct f2fs_sb_info *sbi,
 384                                 unsigned int segno, int offset)
 385 {
 386         struct sit_info *sit_i = SIT_I(sbi);
 387         struct seg_entry *sentry;
 388         int ret;
 389
 390         mutex_lock(&sit_i->sentry_lock);
 391         sentry = get_seg_entry(sbi, segno);
 392         ret = f2fs_test_bit(offset, sentry->cur_valid_map);
 393         mutex_unlock(&sit_i->sentry_lock);
 394         return ret;
 395 }
 396
 397 /*
 398  * This function compares node address got in summary with that in NAT.
 399  * On validity, copy that node with cold status, otherwise (invalid node)
 400  * ignore that.
 401  */
 402 static void gc_node_segment(struct f2fs_sb_info *sbi,
 403                 struct f2fs_summary *sum, unsigned int segno, int gc_type)
 404 {
 405         bool initial = true;
 406         struct f2fs_summary *entry;
 407         int off;
 408
 409 next_step:
 410         entry = sum;
 411
 412         for (off = 0; off < sbi->blocks_per_seg; off++, entry++) {
 413                 nid_t nid = le32_to_cpu(entry->nid);
 414                 struct page *node_page;
 415
 416                 /* stop BG_GC if there is not enough free sections. */
 417                 if (gc_type == BG_GC && has_not_enough_free_secs(sbi, 0))
 418                         return;
 419
 420                 if (check_valid_map(sbi, segno, off) == 0)
 421                         continue;
 422
 423                 if (initial) {
 424                         ra_node_page(sbi, nid);
 425                         continue;
 426                 }
 427                 node_page = get_node_page(sbi, nid);
 428                 if (IS_ERR(node_page))
 429                         continue;
 430
 431                 /* set page dirty and write it */
 432                 if (gc_type == FG_GC) {
 433                         f2fs_wait_on_page_writeback(node_page, NODE, true);
 434                         set_page_dirty(node_page);
 435                 } else {
 436                         if (!PageWriteback(node_page))
 437                                 set_page_dirty(node_page);
 438                 }
 439                 f2fs_put_page(node_page, 1);
 440                 stat_inc_node_blk_count(sbi, 1);
 441         }
 442
 443         if (initial) {
 444                 initial = false;
 445                 goto next_step;
 446         }
 447
 448         if (gc_type == FG_GC) {
 449                 struct writeback_control wbc = {
 450                         .sync_mode = WB_SYNC_ALL,
 451                         .nr_to_write = LONG_MAX,
 452                         .for_reclaim = 0,
 453                 };
 454                 sync_node_pages(sbi, 0, &wbc);
 455
 456                 /*
 457                  * In the case of FG_GC, it'd be better to reclaim this victim
 458                  * completely.
 459                  */
 460                 if (get_valid_blocks(sbi, segno, 1) != 0)
 461                         goto next_step;
 462         }
 463 }
 464
 465 /*
 466  * Calculate start block index indicating the given node offset.
 467  * Be careful, caller should give this node offset only indicating direct node
 468  * blocks. If any node offsets, which point the other types of node blocks such
 469  * as indirect or double indirect node blocks, are given, it must be a caller's
 470  * bug.
 471  */
 472 block_t start_bidx_of_node(unsigned int node_ofs, struct f2fs_inode_info *fi)
 473 {
 474         unsigned int indirect_blks = 2 * NIDS_PER_BLOCK + 4;
 475         unsigned int bidx;
 476
 477         if (node_ofs == 0)
 478                 return 0;
 479
 480         if (node_ofs <= 2) {
 481                 bidx = node_ofs - 1;
 482         } else if (node_ofs <= indirect_blks) {
 483                 int dec = (node_ofs - 4) / (NIDS_PER_BLOCK + 1);
 484                 bidx = node_ofs - 2 - dec;
 485         } else {
 486                 int dec = (node_ofs - indirect_blks - 3) / (NIDS_PER_BLOCK + 1);
 487                 bidx = node_ofs - 5 - dec;
 488         }
 489         return bidx * ADDRS_PER_BLOCK + ADDRS_PER_INODE(fi);
 490 }
 491
 492 static int check_dnode(struct f2fs_sb_info *sbi, struct f2fs_summary *sum,
 493                 struct node_info *dni, block_t blkaddr, unsigned int *nofs)
 494 {
 495         struct page *node_page;
 496         nid_t nid;
 497         unsigned int ofs_in_node;
 498         block_t source_blkaddr;
 499
 500         nid = le32_to_cpu(sum->nid);
 501         ofs_in_node = le16_to_cpu(sum->ofs_in_node);
 502
 503         node_page = get_node_page(sbi, nid);
 504         if (IS_ERR(node_page))
 505                 return 0;
 506
 507         get_node_info(sbi, nid, dni);
 508
 509         if (sum->version != dni->version) {
 510                 f2fs_put_page(node_page, 1);
 511                 return 0;
 512         }
 513
 514         *nofs = ofs_of_node(node_page);
 515         source_blkaddr = datablock_addr(node_page, ofs_in_node);
 516         f2fs_put_page(node_page, 1);
 517
 518         if (source_blkaddr != blkaddr)
 519                 return 0;
 520         return 1;
 521 }
 522
 523 static void move_data_page(struct inode *inode, struct page *page, int gc_type)
 524 {
 525         if (gc_type == BG_GC) {
 526                 if (PageWriteback(page))
 527                         goto out;
 528                 set_page_dirty(page);
 529                 set_cold_data(page);
 530         } else {
 531                 struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
 532
 533                 f2fs_wait_on_page_writeback(page, DATA, true);
 534
 535                 if (clear_page_dirty_for_io(page) &&
 536                         S_ISDIR(inode->i_mode)) {
 537                         dec_page_count(sbi, F2FS_DIRTY_DENTS);
 538                         inode_dec_dirty_dents(inode);
 539                 }
 540                 set_cold_data(page);
 541                 do_write_data_page(page);
 542                 clear_cold_data(page);
 543         }
 544 out:
 545         f2fs_put_page(page, 1);
 546 }
 547
 548 /*
 549  * This function tries to get parent node of victim data block, and identifies
 550  * data block validity. If the block is valid, copy that with cold status and
 551  * modify parent node.
 552  * If the parent node is not valid or the data block address is different,
 553  * the victim data block is ignored.
 554  */
 555 static void gc_data_segment(struct f2fs_sb_info *sbi, struct f2fs_summary *sum,
 556                 struct list_head *ilist, unsigned int segno, int gc_type)
 557 {
 558         struct super_block *sb = sbi->sb;
 559         struct f2fs_summary *entry;
 560         block_t start_addr;
 561         int off;
 562         int phase = 0;
 563
 564         start_addr = START_BLOCK(sbi, segno);
 565
 566 next_step:
 567         entry = sum;
 568
 569         for (off = 0; off < sbi->blocks_per_seg; off++, entry++) {
 570                 struct page *data_page;
 571                 struct inode *inode;
 572                 struct node_info dni; /* dnode info for the data */
 573                 unsigned int ofs_in_node, nofs;
 574                 block_t start_bidx;
 575
 576                 /* stop BG_GC if there is not enough free sections. */
 577                 if (gc_type == BG_GC && has_not_enough_free_secs(sbi, 0))
 578                         return;
 579
 580                 if (check_valid_map(sbi, segno, off) == 0)
 581                         continue;
 582
 583                 if (phase == 0) {
 584                         ra_node_page(sbi, le32_to_cpu(entry->nid));
 585                         continue;
 586                 }
 587
 588                 /* Get an inode by ino with checking validity */
 589                 if (check_dnode(sbi, entry, &dni, start_addr + off, &nofs) == 0)
 590                         continue;
 591
 592                 if (phase == 1) {
 593                         ra_node_page(sbi, dni.ino);
 594                         continue;
 595                 }
 596
 597                 ofs_in_node = le16_to_cpu(entry->ofs_in_node);
 598
 599                 if (phase == 2) {
 600                         inode = f2fs_iget(sb, dni.ino);
 601                         if (IS_ERR(inode))
 602                                 continue;
 603
 604                         start_bidx = start_bidx_of_node(nofs, F2FS_I(inode));
 605
 606                         data_page = find_data_page(inode,
 607                                         start_bidx + ofs_in_node, false);
 608                         if (IS_ERR(data_page))
 609                                 goto next_iput;
 610
 611                         f2fs_put_page(data_page, 0);
 612                         add_gc_inode(inode, ilist);
 613                 } else {
 614                         inode = find_gc_inode(dni.ino, ilist);
 615                         if (inode) {
 616                                 start_bidx = start_bidx_of_node(nofs,
 617                                                                 F2FS_I(inode));
 618                                 data_page = get_lock_data_page(inode,
 619                                                 start_bidx + ofs_in_node);
 620                                 if (IS_ERR(data_page))
 621                                         continue;
 622                                 move_data_page(inode, data_page, gc_type);
 623                                 stat_inc_data_blk_count(sbi, 1);
 624                         }
 625                 }
 626                 continue;
 627 next_iput:
 628                 iput(inode);
 629         }
 630
 631         if (++phase < 4)
 632                 goto next_step;
 633
 634         if (gc_type == FG_GC) {
 635                 f2fs_submit_bio(sbi, DATA, true);
 636
 637                 /*
 638                  * In the case of FG_GC, it'd be better to reclaim this victim
 639                  * completely.
 640                  */
 641                 if (get_valid_blocks(sbi, segno, 1) != 0) {
 642                         phase = 2;
 643                         goto next_step;
 644                 }
 645         }
 646 }
 647
 648 static int __get_victim(struct f2fs_sb_info *sbi, unsigned int *victim,
 649                                                 int gc_type, int type)
 650 {
 651         struct sit_info *sit_i = SIT_I(sbi);
 652         int ret;
 653         mutex_lock(&sit_i->sentry_lock);
 654         ret = DIRTY_I(sbi)->v_ops->get_victim(sbi, victim, gc_type, type, LFS);
 655         mutex_unlock(&sit_i->sentry_lock);
 656         return ret;
 657 }
 658
 659 static void do_garbage_collect(struct f2fs_sb_info *sbi, unsigned int segno,
 660                                 struct list_head *ilist, int gc_type)
 661 {
 662         struct page *sum_page;
 663         struct f2fs_summary_block *sum;
 664         struct blk_plug plug;
 665
 666         /* read segment summary of victim */
 667         sum_page = get_sum_page(sbi, segno);
 668         if (IS_ERR(sum_page))
 669                 return;
 670
 671         blk_start_plug(&plug);
 672
 673         sum = page_address(sum_page);
 674
 675         switch (GET_SUM_TYPE((&sum->footer))) {
 676         case SUM_TYPE_NODE:
 677                 gc_node_segment(sbi, sum->entries, segno, gc_type);
 678                 break;
 679         case SUM_TYPE_DATA:
 680                 gc_data_segment(sbi, sum->entries, ilist, segno, gc_type);
 681                 break;
 682         }
 683         blk_finish_plug(&plug);
 684
 685         stat_inc_seg_count(sbi, GET_SUM_TYPE((&sum->footer)));
 686         stat_inc_call_count(sbi->stat_info);
 687
 688         f2fs_put_page(sum_page, 1);
 689 }
 690
 691 int f2fs_gc(struct f2fs_sb_info *sbi)
 692 {
 693         struct list_head ilist;
 694         unsigned int segno, i;
 695         int gc_type = BG_GC;
 696         int nfree = 0;
 697         int ret = -1;
 698
 699         INIT_LIST_HEAD(&ilist);
 700 gc_more:
 701         if (!(sbi->sb->s_flags & MS_ACTIVE))
 702                 goto stop;
 703
 704         if (gc_type == BG_GC && has_not_enough_free_secs(sbi, nfree)) {
 705                 gc_type = FG_GC;
 706                 write_checkpoint(sbi, false);
 707         }
 708
 709         if (!__get_victim(sbi, &segno, gc_type, NO_CHECK_TYPE))
 710                 goto stop;
 711         ret = 0;
 712
 713         for (i = 0; i < sbi->segs_per_sec; i++)
 714                 do_garbage_collect(sbi, segno + i, &ilist, gc_type);
 715
 716         if (gc_type == FG_GC) {
 717                 sbi->cur_victim_sec = NULL_SEGNO;
 718                 nfree++;
 719                 WARN_ON(get_valid_blocks(sbi, segno, sbi->segs_per_sec));
 720         }
 721
 722         if (has_not_enough_free_secs(sbi, nfree))
 723                 goto gc_more;
 724
 725         if (gc_type == FG_GC)
 726                 write_checkpoint(sbi, false);
 727 stop:
 728         mutex_unlock(&sbi->gc_mutex);
 729
 730         put_gc_inode(&ilist);
 731         return ret;
 732 }
 733
 734 void build_gc_manager(struct f2fs_sb_info *sbi)
 735 {
 736         DIRTY_I(sbi)->v_ops = &default_v_ops;
 737 }
 738
 739 int __init create_gc_caches(void)
 740 {
 741         winode_slab = f2fs_kmem_cache_create("f2fs_gc_inodes",
 742                         sizeof(struct inode_entry), NULL);
 743         if (!winode_slab)
 744                 return -ENOMEM;
 745         return 0;
 746 }
 747
 748 void destroy_gc_caches(void)
 749 {
 750         kmem_cache_destroy(winode_slab);
 751 }