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 f2fs's metadata periodically */
  87                 f2fs_balance_fs_bg(sbi);
  88
  89         } while (!kthread_should_stop());
  90         return 0;
  91 }
  92
  93 int start_gc_thread(struct f2fs_sb_info *sbi)
  94 {
  95         struct f2fs_gc_kthread *gc_th;
  96         dev_t dev = sbi->sb->s_bdev->bd_dev;
  97         int err = 0;
  98
  99         if (!test_opt(sbi, BG_GC))
 100                 goto out;
 101         gc_th = kmalloc(sizeof(struct f2fs_gc_kthread), GFP_KERNEL);
 102         if (!gc_th) {
 103                 err = -ENOMEM;
 104                 goto out;
 105         }
 106
 107         gc_th->min_sleep_time = DEF_GC_THREAD_MIN_SLEEP_TIME;
 108         gc_th->max_sleep_time = DEF_GC_THREAD_MAX_SLEEP_TIME;
 109         gc_th->no_gc_sleep_time = DEF_GC_THREAD_NOGC_SLEEP_TIME;
 110
 111         gc_th->gc_idle = 0;
 112
 113         sbi->gc_thread = gc_th;
 114         init_waitqueue_head(&sbi->gc_thread->gc_wait_queue_head);
 115         sbi->gc_thread->f2fs_gc_task = kthread_run(gc_thread_func, sbi,
 116                         "f2fs_gc-%u:%u", MAJOR(dev), MINOR(dev));
 117         if (IS_ERR(gc_th->f2fs_gc_task)) {
 118                 err = PTR_ERR(gc_th->f2fs_gc_task);
 119                 kfree(gc_th);
 120                 sbi->gc_thread = NULL;
 121         }
 122 out:
 123         return err;
 124 }
 125
 126 void stop_gc_thread(struct f2fs_sb_info *sbi)
 127 {
 128         struct f2fs_gc_kthread *gc_th = sbi->gc_thread;
 129         if (!gc_th)
 130                 return;
 131         kthread_stop(gc_th->f2fs_gc_task);
 132         kfree(gc_th);
 133         sbi->gc_thread = NULL;
 134 }
 135
 136 static int select_gc_type(struct f2fs_gc_kthread *gc_th, int gc_type)
 137 {
 138         int gc_mode = (gc_type == BG_GC) ? GC_CB : GC_GREEDY;
 139
 140         if (gc_th && gc_th->gc_idle) {
 141                 if (gc_th->gc_idle == 1)
 142                         gc_mode = GC_CB;
 143                 else if (gc_th->gc_idle == 2)
 144                         gc_mode = GC_GREEDY;
 145         }
 146         return gc_mode;
 147 }
 148
 149 static void select_policy(struct f2fs_sb_info *sbi, int gc_type,
 150                         int type, struct victim_sel_policy *p)
 151 {
 152         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
 153
 154         if (p->alloc_mode == SSR) {
 155                 p->gc_mode = GC_GREEDY;
 156                 p->dirty_segmap = dirty_i->dirty_segmap[type];
 157                 p->max_search = dirty_i->nr_dirty[type];
 158                 p->ofs_unit = 1;
 159         } else {
 160                 p->gc_mode = select_gc_type(sbi->gc_thread, gc_type);
 161                 p->dirty_segmap = dirty_i->dirty_segmap[DIRTY];
 162                 p->max_search = dirty_i->nr_dirty[DIRTY];
 163                 p->ofs_unit = sbi->segs_per_sec;
 164         }
 165
 166         if (p->max_search > sbi->max_victim_search)
 167                 p->max_search = sbi->max_victim_search;
 168
 169         p->offset = sbi->last_victim[p->gc_mode];
 170 }
 171
 172 static unsigned int get_max_cost(struct f2fs_sb_info *sbi,
 173                                 struct victim_sel_policy *p)
 174 {
 175         /* SSR allocates in a segment unit */
 176         if (p->alloc_mode == SSR)
 177                 return 1 << sbi->log_blocks_per_seg;
 178         if (p->gc_mode == GC_GREEDY)
 179                 return (1 << sbi->log_blocks_per_seg) * p->ofs_unit;
 180         else if (p->gc_mode == GC_CB)
 181                 return UINT_MAX;
 182         else /* No other gc_mode */
 183                 return 0;
 184 }
 185
 186 static unsigned int check_bg_victims(struct f2fs_sb_info *sbi)
 187 {
 188         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
 189         unsigned int hint = 0;
 190         unsigned int secno;
 191
 192         /*
 193          * If the gc_type is FG_GC, we can select victim segments
 194          * selected by background GC before.
 195          * Those segments guarantee they have small valid blocks.
 196          */
 197 next:
 198         secno = find_next_bit(dirty_i->victim_secmap, TOTAL_SECS(sbi), hint++);
 199         if (secno < TOTAL_SECS(sbi)) {
 200                 if (sec_usage_check(sbi, secno))
 201                         goto next;
 202                 clear_bit(secno, dirty_i->victim_secmap);
 203                 return secno * sbi->segs_per_sec;
 204         }
 205         return NULL_SEGNO;
 206 }
 207
 208 static unsigned int get_cb_cost(struct f2fs_sb_info *sbi, unsigned int segno)
 209 {
 210         struct sit_info *sit_i = SIT_I(sbi);
 211         unsigned int secno = GET_SECNO(sbi, segno);
 212         unsigned int start = secno * sbi->segs_per_sec;
 213         unsigned long long mtime = 0;
 214         unsigned int vblocks;
 215         unsigned char age = 0;
 216         unsigned char u;
 217         unsigned int i;
 218
 219         for (i = 0; i < sbi->segs_per_sec; i++)
 220                 mtime += get_seg_entry(sbi, start + i)->mtime;
 221         vblocks = get_valid_blocks(sbi, segno, sbi->segs_per_sec);
 222
 223         mtime = div_u64(mtime, sbi->segs_per_sec);
 224         vblocks = div_u64(vblocks, sbi->segs_per_sec);
 225
 226         u = (vblocks * 100) >> sbi->log_blocks_per_seg;
 227
 228         /* Handle if the system time is changed by user */
 229         if (mtime < sit_i->min_mtime)
 230                 sit_i->min_mtime = mtime;
 231         if (mtime > sit_i->max_mtime)
 232                 sit_i->max_mtime = mtime;
 233         if (sit_i->max_mtime != sit_i->min_mtime)
 234                 age = 100 - div64_u64(100 * (mtime - sit_i->min_mtime),
 235                                 sit_i->max_mtime - sit_i->min_mtime);
 236
 237         return UINT_MAX - ((100 * (100 - u) * age) / (100 + u));
 238 }
 239
 240 static inline unsigned int get_gc_cost(struct f2fs_sb_info *sbi,
 241                         unsigned int segno, struct victim_sel_policy *p)
 242 {
 243         if (p->alloc_mode == SSR)
 244                 return get_seg_entry(sbi, segno)->ckpt_valid_blocks;
 245
 246         /* alloc_mode == LFS */
 247         if (p->gc_mode == GC_GREEDY)
 248                 return get_valid_blocks(sbi, segno, sbi->segs_per_sec);
 249         else
 250                 return get_cb_cost(sbi, segno);
 251 }
 252
 253 /*
 254  * This function is called from two paths.
 255  * One is garbage collection and the other is SSR segment selection.
 256  * When it is called during GC, it just gets a victim segment
 257  * and it does not remove it from dirty seglist.
 258  * When it is called from SSR segment selection, it finds a segment
 259  * which has minimum valid blocks and removes it from dirty seglist.
 260  */
 261 static int get_victim_by_default(struct f2fs_sb_info *sbi,
 262                 unsigned int *result, int gc_type, int type, char alloc_mode)
 263 {
 264         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
 265         struct victim_sel_policy p;
 266         unsigned int secno, max_cost;
 267         int nsearched = 0;
 268
 269         p.alloc_mode = alloc_mode;
 270         select_policy(sbi, gc_type, type, &p);
 271
 272         p.min_segno = NULL_SEGNO;
 273         p.min_cost = max_cost = get_max_cost(sbi, &p);
 274
 275         mutex_lock(&dirty_i->seglist_lock);
 276
 277         if (p.alloc_mode == LFS && gc_type == FG_GC) {
 278                 p.min_segno = check_bg_victims(sbi);
 279                 if (p.min_segno != NULL_SEGNO)
 280                         goto got_it;
 281         }
 282
 283         while (1) {
 284                 unsigned long cost;
 285                 unsigned int segno;
 286
 287                 segno = find_next_bit(p.dirty_segmap,
 288                                                 TOTAL_SEGS(sbi), p.offset);
 289                 if (segno >= TOTAL_SEGS(sbi)) {
 290                         if (sbi->last_victim[p.gc_mode]) {
 291                                 sbi->last_victim[p.gc_mode] = 0;
 292                                 p.offset = 0;
 293                                 continue;
 294                         }
 295                         break;
 296                 }
 297
 298                 p.offset = segno + p.ofs_unit;
 299                 if (p.ofs_unit > 1)
 300                         p.offset -= segno % p.ofs_unit;
 301
 302                 secno = GET_SECNO(sbi, segno);
 303
 304                 if (sec_usage_check(sbi, secno))
 305                         continue;
 306                 if (gc_type == BG_GC && test_bit(secno, dirty_i->victim_secmap))
 307                         continue;
 308
 309                 cost = get_gc_cost(sbi, segno, &p);
 310
 311                 if (p.min_cost > cost) {
 312                         p.min_segno = segno;
 313                         p.min_cost = cost;
 314                 } else if (unlikely(cost == max_cost)) {
 315                         continue;
 316                 }
 317
 318                 if (nsearched++ >= p.max_search) {
 319                         sbi->last_victim[p.gc_mode] = segno;
 320                         break;
 321                 }
 322         }
 323         if (p.min_segno != NULL_SEGNO) {
 324 got_it:
 325                 if (p.alloc_mode == LFS) {
 326                         secno = GET_SECNO(sbi, p.min_segno);
 327                         if (gc_type == FG_GC)
 328                                 sbi->cur_victim_sec = secno;
 329                         else
 330                                 set_bit(secno, dirty_i->victim_secmap);
 331                 }
 332                 *result = (p.min_segno / p.ofs_unit) * p.ofs_unit;
 333
 334                 trace_f2fs_get_victim(sbi->sb, type, gc_type, &p,
 335                                 sbi->cur_victim_sec,
 336                                 prefree_segments(sbi), free_segments(sbi));
 337         }
 338         mutex_unlock(&dirty_i->seglist_lock);
 339
 340         return (p.min_segno == NULL_SEGNO) ? 0 : 1;
 341 }
 342
 343 static const struct victim_selection default_v_ops = {
 344         .get_victim = get_victim_by_default,
 345 };
 346
 347 static struct inode *find_gc_inode(nid_t ino, struct list_head *ilist)
 348 {
 349         struct inode_entry *ie;
 350
 351         list_for_each_entry(ie, ilist, list)
 352                 if (ie->inode->i_ino == ino)
 353                         return ie->inode;
 354         return NULL;
 355 }
 356
 357 static void add_gc_inode(struct inode *inode, struct list_head *ilist)
 358 {
 359         struct inode_entry *new_ie;
 360
 361         if (inode == find_gc_inode(inode->i_ino, ilist)) {
 362                 iput(inode);
 363                 return;
 364         }
 365
 366         new_ie = f2fs_kmem_cache_alloc(winode_slab, GFP_NOFS);
 367         new_ie->inode = inode;
 368         list_add_tail(&new_ie->list, ilist);
 369 }
 370
 371 static void put_gc_inode(struct list_head *ilist)
 372 {
 373         struct inode_entry *ie, *next_ie;
 374         list_for_each_entry_safe(ie, next_ie, ilist, list) {
 375                 iput(ie->inode);
 376                 list_del(&ie->list);
 377                 kmem_cache_free(winode_slab, ie);
 378         }
 379 }
 380
 381 static int check_valid_map(struct f2fs_sb_info *sbi,
 382                                 unsigned int segno, int offset)
 383 {
 384         struct sit_info *sit_i = SIT_I(sbi);
 385         struct seg_entry *sentry;
 386         int ret;
 387
 388         mutex_lock(&sit_i->sentry_lock);
 389         sentry = get_seg_entry(sbi, segno);
 390         ret = f2fs_test_bit(offset, sentry->cur_valid_map);
 391         mutex_unlock(&sit_i->sentry_lock);
 392         return ret;
 393 }
 394
 395 /*
 396  * This function compares node address got in summary with that in NAT.
 397  * On validity, copy that node with cold status, otherwise (invalid node)
 398  * ignore that.
 399  */
 400 static void gc_node_segment(struct f2fs_sb_info *sbi,
 401                 struct f2fs_summary *sum, unsigned int segno, int gc_type)
 402 {
 403         bool initial = true;
 404         struct f2fs_summary *entry;
 405         int off;
 406
 407 next_step:
 408         entry = sum;
 409
 410         for (off = 0; off < sbi->blocks_per_seg; off++, entry++) {
 411                 nid_t nid = le32_to_cpu(entry->nid);
 412                 struct page *node_page;
 413
 414                 /* stop BG_GC if there is not enough free sections. */
 415                 if (gc_type == BG_GC && has_not_enough_free_secs(sbi, 0))
 416                         return;
 417
 418                 if (check_valid_map(sbi, segno, off) == 0)
 419                         continue;
 420
 421                 if (initial) {
 422                         ra_node_page(sbi, nid);
 423                         continue;
 424                 }
 425                 node_page = get_node_page(sbi, nid);
 426                 if (IS_ERR(node_page))
 427                         continue;
 428
 429                 /* set page dirty and write it */
 430                 if (gc_type == FG_GC) {
 431                         f2fs_wait_on_page_writeback(node_page, NODE);
 432                         set_page_dirty(node_page);
 433                 } else {
 434                         if (!PageWriteback(node_page))
 435                                 set_page_dirty(node_page);
 436                 }
 437                 f2fs_put_page(node_page, 1);
 438                 stat_inc_node_blk_count(sbi, 1);
 439         }
 440
 441         if (initial) {
 442                 initial = false;
 443                 goto next_step;
 444         }
 445
 446         if (gc_type == FG_GC) {
 447                 struct writeback_control wbc = {
 448                         .sync_mode = WB_SYNC_ALL,
 449                         .nr_to_write = LONG_MAX,
 450                         .for_reclaim = 0,
 451                 };
 452                 sync_node_pages(sbi, 0, &wbc);
 453
 454                 /*
 455                  * In the case of FG_GC, it'd be better to reclaim this victim
 456                  * completely.
 457                  */
 458                 if (get_valid_blocks(sbi, segno, 1) != 0)
 459                         goto next_step;
 460         }
 461 }
 462
 463 /*
 464  * Calculate start block index indicating the given node offset.
 465  * Be careful, caller should give this node offset only indicating direct node
 466  * blocks. If any node offsets, which point the other types of node blocks such
 467  * as indirect or double indirect node blocks, are given, it must be a caller's
 468  * bug.
 469  */
 470 block_t start_bidx_of_node(unsigned int node_ofs, struct f2fs_inode_info *fi)
 471 {
 472         unsigned int indirect_blks = 2 * NIDS_PER_BLOCK + 4;
 473         unsigned int bidx;
 474
 475         if (node_ofs == 0)
 476                 return 0;
 477
 478         if (node_ofs <= 2) {
 479                 bidx = node_ofs - 1;
 480         } else if (node_ofs <= indirect_blks) {
 481                 int dec = (node_ofs - 4) / (NIDS_PER_BLOCK + 1);
 482                 bidx = node_ofs - 2 - dec;
 483         } else {
 484                 int dec = (node_ofs - indirect_blks - 3) / (NIDS_PER_BLOCK + 1);
 485                 bidx = node_ofs - 5 - dec;
 486         }
 487         return bidx * ADDRS_PER_BLOCK + ADDRS_PER_INODE(fi);
 488 }
 489
 490 static int check_dnode(struct f2fs_sb_info *sbi, struct f2fs_summary *sum,
 491                 struct node_info *dni, block_t blkaddr, unsigned int *nofs)
 492 {
 493         struct page *node_page;
 494         nid_t nid;
 495         unsigned int ofs_in_node;
 496         block_t source_blkaddr;
 497
 498         nid = le32_to_cpu(sum->nid);
 499         ofs_in_node = le16_to_cpu(sum->ofs_in_node);
 500
 501         node_page = get_node_page(sbi, nid);
 502         if (IS_ERR(node_page))
 503                 return 0;
 504
 505         get_node_info(sbi, nid, dni);
 506
 507         if (sum->version != dni->version) {
 508                 f2fs_put_page(node_page, 1);
 509                 return 0;
 510         }
 511
 512         *nofs = ofs_of_node(node_page);
 513         source_blkaddr = datablock_addr(node_page, ofs_in_node);
 514         f2fs_put_page(node_page, 1);
 515
 516         if (source_blkaddr != blkaddr)
 517                 return 0;
 518         return 1;
 519 }
 520
 521 static void move_data_page(struct inode *inode, struct page *page, int gc_type)
 522 {
 523         struct f2fs_io_info fio = {
 524                 .type = DATA,
 525                 .rw = WRITE_SYNC,
 526         };
 527
 528         if (gc_type == BG_GC) {
 529                 if (PageWriteback(page))
 530                         goto out;
 531                 set_page_dirty(page);
 532                 set_cold_data(page);
 533         } else {
 534                 f2fs_wait_on_page_writeback(page, DATA);
 535
 536                 if (clear_page_dirty_for_io(page))
 537                         inode_dec_dirty_dents(inode);
 538                 set_cold_data(page);
 539                 do_write_data_page(page, &fio);
 540                 clear_cold_data(page);
 541         }
 542 out:
 543         f2fs_put_page(page, 1);
 544 }
 545
 546 /*
 547  * This function tries to get parent node of victim data block, and identifies
 548  * data block validity. If the block is valid, copy that with cold status and
 549  * modify parent node.
 550  * If the parent node is not valid or the data block address is different,
 551  * the victim data block is ignored.
 552  */
 553 static void gc_data_segment(struct f2fs_sb_info *sbi, struct f2fs_summary *sum,
 554                 struct list_head *ilist, unsigned int segno, int gc_type)
 555 {
 556         struct super_block *sb = sbi->sb;
 557         struct f2fs_summary *entry;
 558         block_t start_addr;
 559         int off;
 560         int phase = 0;
 561
 562         start_addr = START_BLOCK(sbi, segno);
 563
 564 next_step:
 565         entry = sum;
 566
 567         for (off = 0; off < sbi->blocks_per_seg; off++, entry++) {
 568                 struct page *data_page;
 569                 struct inode *inode;
 570                 struct node_info dni; /* dnode info for the data */
 571                 unsigned int ofs_in_node, nofs;
 572                 block_t start_bidx;
 573
 574                 /* stop BG_GC if there is not enough free sections. */
 575                 if (gc_type == BG_GC && has_not_enough_free_secs(sbi, 0))
 576                         return;
 577
 578                 if (check_valid_map(sbi, segno, off) == 0)
 579                         continue;
 580
 581                 if (phase == 0) {
 582                         ra_node_page(sbi, le32_to_cpu(entry->nid));
 583                         continue;
 584                 }
 585
 586                 /* Get an inode by ino with checking validity */
 587                 if (check_dnode(sbi, entry, &dni, start_addr + off, &nofs) == 0)
 588                         continue;
 589
 590                 if (phase == 1) {
 591                         ra_node_page(sbi, dni.ino);
 592                         continue;
 593                 }
 594
 595                 ofs_in_node = le16_to_cpu(entry->ofs_in_node);
 596
 597                 if (phase == 2) {
 598                         inode = f2fs_iget(sb, dni.ino);
 599                         if (IS_ERR(inode))
 600                                 continue;
 601
 602                         start_bidx = start_bidx_of_node(nofs, F2FS_I(inode));
 603
 604                         data_page = find_data_page(inode,
 605                                         start_bidx + ofs_in_node, false);
 606                         if (IS_ERR(data_page))
 607                                 goto next_iput;
 608
 609                         f2fs_put_page(data_page, 0);
 610                         add_gc_inode(inode, ilist);
 611                 } else {
 612                         inode = find_gc_inode(dni.ino, ilist);
 613                         if (inode) {
 614                                 start_bidx = start_bidx_of_node(nofs,
 615                                                                 F2FS_I(inode));
 616                                 data_page = get_lock_data_page(inode,
 617                                                 start_bidx + ofs_in_node);
 618                                 if (IS_ERR(data_page))
 619                                         continue;
 620                                 move_data_page(inode, data_page, gc_type);
 621                                 stat_inc_data_blk_count(sbi, 1);
 622                         }
 623                 }
 624                 continue;
 625 next_iput:
 626                 iput(inode);
 627         }
 628
 629         if (++phase < 4)
 630                 goto next_step;
 631
 632         if (gc_type == FG_GC) {
 633                 f2fs_submit_merged_bio(sbi, DATA, WRITE);
 634
 635                 /*
 636                  * In the case of FG_GC, it'd be better to reclaim this victim
 637                  * completely.
 638                  */
 639                 if (get_valid_blocks(sbi, segno, 1) != 0) {
 640                         phase = 2;
 641                         goto next_step;
 642                 }
 643         }
 644 }
 645
 646 static int __get_victim(struct f2fs_sb_info *sbi, unsigned int *victim,
 647                                                 int gc_type, int type)
 648 {
 649         struct sit_info *sit_i = SIT_I(sbi);
 650         int ret;
 651         mutex_lock(&sit_i->sentry_lock);
 652         ret = DIRTY_I(sbi)->v_ops->get_victim(sbi, victim, gc_type, type, LFS);
 653         mutex_unlock(&sit_i->sentry_lock);
 654         return ret;
 655 }
 656
 657 static void do_garbage_collect(struct f2fs_sb_info *sbi, unsigned int segno,
 658                                 struct list_head *ilist, int gc_type)
 659 {
 660         struct page *sum_page;
 661         struct f2fs_summary_block *sum;
 662         struct blk_plug plug;
 663
 664         /* read segment summary of victim */
 665         sum_page = get_sum_page(sbi, segno);
 666
 667         blk_start_plug(&plug);
 668
 669         sum = page_address(sum_page);
 670
 671         switch (GET_SUM_TYPE((&sum->footer))) {
 672         case SUM_TYPE_NODE:
 673                 gc_node_segment(sbi, sum->entries, segno, gc_type);
 674                 break;
 675         case SUM_TYPE_DATA:
 676                 gc_data_segment(sbi, sum->entries, ilist, segno, gc_type);
 677                 break;
 678         }
 679         blk_finish_plug(&plug);
 680
 681         stat_inc_seg_count(sbi, GET_SUM_TYPE((&sum->footer)));
 682         stat_inc_call_count(sbi->stat_info);
 683
 684         f2fs_put_page(sum_page, 1);
 685 }
 686
 687 int f2fs_gc(struct f2fs_sb_info *sbi)
 688 {
 689         struct list_head ilist;
 690         unsigned int segno, i;
 691         int gc_type = BG_GC;
 692         int nfree = 0;
 693         int ret = -1;
 694
 695         INIT_LIST_HEAD(&ilist);
 696 gc_more:
 697         if (unlikely(!(sbi->sb->s_flags & MS_ACTIVE)))
 698                 goto stop;
 699         if (unlikely(is_set_ckpt_flags(F2FS_CKPT(sbi), CP_ERROR_FLAG)))
 700                 goto stop;
 701
 702         if (gc_type == BG_GC && has_not_enough_free_secs(sbi, nfree)) {
 703                 gc_type = FG_GC;
 704                 write_checkpoint(sbi, false);
 705         }
 706
 707         if (!__get_victim(sbi, &segno, gc_type, NO_CHECK_TYPE))
 708                 goto stop;
 709         ret = 0;
 710
 711         /* readahead multi ssa blocks those have contiguous address */
 712         if (sbi->segs_per_sec > 1)
 713                 ra_meta_pages(sbi, GET_SUM_BLOCK(sbi, segno), sbi->segs_per_sec,
 714                                                                 META_SSA);
 715
 716         for (i = 0; i < sbi->segs_per_sec; i++)
 717                 do_garbage_collect(sbi, segno + i, &ilist, gc_type);
 718
 719         if (gc_type == FG_GC) {
 720                 sbi->cur_victim_sec = NULL_SEGNO;
 721                 nfree++;
 722                 WARN_ON(get_valid_blocks(sbi, segno, sbi->segs_per_sec));
 723         }
 724
 725         if (has_not_enough_free_secs(sbi, nfree))
 726                 goto gc_more;
 727
 728         if (gc_type == FG_GC)
 729                 write_checkpoint(sbi, false);
 730 stop:
 731         mutex_unlock(&sbi->gc_mutex);
 732
 733         put_gc_inode(&ilist);
 734         return ret;
 735 }
 736
 737 void build_gc_manager(struct f2fs_sb_info *sbi)
 738 {
 739         DIRTY_I(sbi)->v_ops = &default_v_ops;
 740 }
 741
 742 int __init create_gc_caches(void)
 743 {
 744         winode_slab = f2fs_kmem_cache_create("f2fs_gc_inodes",
 745                         sizeof(struct inode_entry));
 746         if (!winode_slab)
 747                 return -ENOMEM;
 748         return 0;
 749 }
 750
 751 void destroy_gc_caches(void)
 752 {
 753         kmem_cache_destroy(winode_slab);
 754 }