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 int gc_thread_func(void *data)
  28 {
  29         struct f2fs_sb_info *sbi = data;
  30         struct f2fs_gc_kthread *gc_th = sbi->gc_thread;
  31         wait_queue_head_t *wq = &sbi->gc_thread->gc_wait_queue_head;
  32         long wait_ms;
  33
  34         wait_ms = gc_th->min_sleep_time;
  35
  36         do {
  37                 if (try_to_freeze())
  38                         continue;
  39                 else
  40                         wait_event_interruptible_timeout(*wq,
  41                                                 kthread_should_stop(),
  42                                                 msecs_to_jiffies(wait_ms));
  43                 if (kthread_should_stop())
  44                         break;
  45
  46                 if (sbi->sb->s_writers.frozen >= SB_FREEZE_WRITE) {
  47                         increase_sleep_time(gc_th, &wait_ms);
  48                         continue;
  49                 }
  50
  51                 /*
  52                  * [GC triggering condition]
  53                  * 0. GC is not conducted currently.
  54                  * 1. There are enough dirty segments.
  55                  * 2. IO subsystem is idle by checking the # of writeback pages.
  56                  * 3. IO subsystem is idle by checking the # of requests in
  57                  *    bdev's request list.
  58                  *
  59                  * Note) We have to avoid triggering GCs frequently.
  60                  * Because it is possible that some segments can be
  61                  * invalidated soon after by user update or deletion.
  62                  * So, I'd like to wait some time to collect dirty segments.
  63                  */
  64                 if (!mutex_trylock(&sbi->gc_mutex))
  65                         continue;
  66
  67                 if (!is_idle(sbi)) {
  68                         increase_sleep_time(gc_th, &wait_ms);
  69                         mutex_unlock(&sbi->gc_mutex);
  70                         continue;
  71                 }
  72
  73                 if (has_enough_invalid_blocks(sbi))
  74                         decrease_sleep_time(gc_th, &wait_ms);
  75                 else
  76                         increase_sleep_time(gc_th, &wait_ms);
  77
  78                 stat_inc_bggc_count(sbi);
  79
  80                 /* if return value is not zero, no victim was selected */
  81                 if (f2fs_gc(sbi))
  82                         wait_ms = gc_th->no_gc_sleep_time;
  83
  84                 /* balancing f2fs's metadata periodically */
  85                 f2fs_balance_fs_bg(sbi);
  86
  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         gc_th = kmalloc(sizeof(struct f2fs_gc_kthread), GFP_KERNEL);
  98         if (!gc_th) {
  99                 err = -ENOMEM;
 100                 goto out;
 101         }
 102
 103         gc_th->min_sleep_time = DEF_GC_THREAD_MIN_SLEEP_TIME;
 104         gc_th->max_sleep_time = DEF_GC_THREAD_MAX_SLEEP_TIME;
 105         gc_th->no_gc_sleep_time = DEF_GC_THREAD_NOGC_SLEEP_TIME;
 106
 107         gc_th->gc_idle = 0;
 108
 109         sbi->gc_thread = gc_th;
 110         init_waitqueue_head(&sbi->gc_thread->gc_wait_queue_head);
 111         sbi->gc_thread->f2fs_gc_task = kthread_run(gc_thread_func, sbi,
 112                         "f2fs_gc-%u:%u", MAJOR(dev), MINOR(dev));
 113         if (IS_ERR(gc_th->f2fs_gc_task)) {
 114                 err = PTR_ERR(gc_th->f2fs_gc_task);
 115                 kfree(gc_th);
 116                 sbi->gc_thread = NULL;
 117         }
 118 out:
 119         return err;
 120 }
 121
 122 void stop_gc_thread(struct f2fs_sb_info *sbi)
 123 {
 124         struct f2fs_gc_kthread *gc_th = sbi->gc_thread;
 125         if (!gc_th)
 126                 return;
 127         kthread_stop(gc_th->f2fs_gc_task);
 128         kfree(gc_th);
 129         sbi->gc_thread = NULL;
 130 }
 131
 132 static int select_gc_type(struct f2fs_gc_kthread *gc_th, int gc_type)
 133 {
 134         int gc_mode = (gc_type == BG_GC) ? GC_CB : GC_GREEDY;
 135
 136         if (gc_th && gc_th->gc_idle) {
 137                 if (gc_th->gc_idle == 1)
 138                         gc_mode = GC_CB;
 139                 else if (gc_th->gc_idle == 2)
 140                         gc_mode = GC_GREEDY;
 141         }
 142         return gc_mode;
 143 }
 144
 145 static void select_policy(struct f2fs_sb_info *sbi, int gc_type,
 146                         int type, struct victim_sel_policy *p)
 147 {
 148         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
 149
 150         if (p->alloc_mode == SSR) {
 151                 p->gc_mode = GC_GREEDY;
 152                 p->dirty_segmap = dirty_i->dirty_segmap[type];
 153                 p->max_search = dirty_i->nr_dirty[type];
 154                 p->ofs_unit = 1;
 155         } else {
 156                 p->gc_mode = select_gc_type(sbi->gc_thread, gc_type);
 157                 p->dirty_segmap = dirty_i->dirty_segmap[DIRTY];
 158                 p->max_search = dirty_i->nr_dirty[DIRTY];
 159                 p->ofs_unit = sbi->segs_per_sec;
 160         }
 161
 162         if (p->max_search > sbi->max_victim_search)
 163                 p->max_search = sbi->max_victim_search;
 164
 165         p->offset = sbi->last_victim[p->gc_mode];
 166 }
 167
 168 static unsigned int get_max_cost(struct f2fs_sb_info *sbi,
 169                                 struct victim_sel_policy *p)
 170 {
 171         /* SSR allocates in a segment unit */
 172         if (p->alloc_mode == SSR)
 173                 return 1 << sbi->log_blocks_per_seg;
 174         if (p->gc_mode == GC_GREEDY)
 175                 return (1 << sbi->log_blocks_per_seg) * p->ofs_unit;
 176         else if (p->gc_mode == GC_CB)
 177                 return UINT_MAX;
 178         else /* No other gc_mode */
 179                 return 0;
 180 }
 181
 182 static unsigned int check_bg_victims(struct f2fs_sb_info *sbi)
 183 {
 184         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
 185         unsigned int secno;
 186
 187         /*
 188          * If the gc_type is FG_GC, we can select victim segments
 189          * selected by background GC before.
 190          * Those segments guarantee they have small valid blocks.
 191          */
 192         for_each_set_bit(secno, dirty_i->victim_secmap, MAIN_SECS(sbi)) {
 193                 if (sec_usage_check(sbi, secno))
 194                         continue;
 195                 clear_bit(secno, dirty_i->victim_secmap);
 196                 return secno * sbi->segs_per_sec;
 197         }
 198         return NULL_SEGNO;
 199 }
 200
 201 static unsigned int get_cb_cost(struct f2fs_sb_info *sbi, unsigned int segno)
 202 {
 203         struct sit_info *sit_i = SIT_I(sbi);
 204         unsigned int secno = GET_SECNO(sbi, segno);
 205         unsigned int start = secno * sbi->segs_per_sec;
 206         unsigned long long mtime = 0;
 207         unsigned int vblocks;
 208         unsigned char age = 0;
 209         unsigned char u;
 210         unsigned int i;
 211
 212         for (i = 0; i < sbi->segs_per_sec; i++)
 213                 mtime += get_seg_entry(sbi, start + i)->mtime;
 214         vblocks = get_valid_blocks(sbi, segno, sbi->segs_per_sec);
 215
 216         mtime = div_u64(mtime, sbi->segs_per_sec);
 217         vblocks = div_u64(vblocks, sbi->segs_per_sec);
 218
 219         u = (vblocks * 100) >> sbi->log_blocks_per_seg;
 220
 221         /* Handle if the system time has changed by the user */
 222         if (mtime < sit_i->min_mtime)
 223                 sit_i->min_mtime = mtime;
 224         if (mtime > sit_i->max_mtime)
 225                 sit_i->max_mtime = mtime;
 226         if (sit_i->max_mtime != sit_i->min_mtime)
 227                 age = 100 - div64_u64(100 * (mtime - sit_i->min_mtime),
 228                                 sit_i->max_mtime - sit_i->min_mtime);
 229
 230         return UINT_MAX - ((100 * (100 - u) * age) / (100 + u));
 231 }
 232
 233 static inline unsigned int get_gc_cost(struct f2fs_sb_info *sbi,
 234                         unsigned int segno, struct victim_sel_policy *p)
 235 {
 236         if (p->alloc_mode == SSR)
 237                 return get_seg_entry(sbi, segno)->ckpt_valid_blocks;
 238
 239         /* alloc_mode == LFS */
 240         if (p->gc_mode == GC_GREEDY)
 241                 return get_valid_blocks(sbi, segno, sbi->segs_per_sec);
 242         else
 243                 return get_cb_cost(sbi, segno);
 244 }
 245
 246 /*
 247  * This function is called from two paths.
 248  * One is garbage collection and the other is SSR segment selection.
 249  * When it is called during GC, it just gets a victim segment
 250  * and it does not remove it from dirty seglist.
 251  * When it is called from SSR segment selection, it finds a segment
 252  * which has minimum valid blocks and removes it from dirty seglist.
 253  */
 254 static int get_victim_by_default(struct f2fs_sb_info *sbi,
 255                 unsigned int *result, int gc_type, int type, char alloc_mode)
 256 {
 257         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
 258         struct victim_sel_policy p;
 259         unsigned int secno, max_cost;
 260         int nsearched = 0;
 261
 262         mutex_lock(&dirty_i->seglist_lock);
 263
 264         p.alloc_mode = alloc_mode;
 265         select_policy(sbi, gc_type, type, &p);
 266
 267         p.min_segno = NULL_SEGNO;
 268         p.min_cost = max_cost = get_max_cost(sbi, &p);
 269
 270         if (p.alloc_mode == LFS && gc_type == FG_GC) {
 271                 p.min_segno = check_bg_victims(sbi);
 272                 if (p.min_segno != NULL_SEGNO)
 273                         goto got_it;
 274         }
 275
 276         while (1) {
 277                 unsigned long cost;
 278                 unsigned int segno;
 279
 280                 segno = find_next_bit(p.dirty_segmap, MAIN_SEGS(sbi), p.offset);
 281                 if (segno >= MAIN_SEGS(sbi)) {
 282                         if (sbi->last_victim[p.gc_mode]) {
 283                                 sbi->last_victim[p.gc_mode] = 0;
 284                                 p.offset = 0;
 285                                 continue;
 286                         }
 287                         break;
 288                 }
 289
 290                 p.offset = segno + p.ofs_unit;
 291                 if (p.ofs_unit > 1)
 292                         p.offset -= segno % p.ofs_unit;
 293
 294                 secno = GET_SECNO(sbi, segno);
 295
 296                 if (sec_usage_check(sbi, secno))
 297                         continue;
 298                 if (gc_type == BG_GC && test_bit(secno, dirty_i->victim_secmap))
 299                         continue;
 300
 301                 cost = get_gc_cost(sbi, segno, &p);
 302
 303                 if (p.min_cost > cost) {
 304                         p.min_segno = segno;
 305                         p.min_cost = cost;
 306                 } else if (unlikely(cost == max_cost)) {
 307                         continue;
 308                 }
 309
 310                 if (nsearched++ >= p.max_search) {
 311                         sbi->last_victim[p.gc_mode] = segno;
 312                         break;
 313                 }
 314         }
 315         if (p.min_segno != NULL_SEGNO) {
 316 got_it:
 317                 if (p.alloc_mode == LFS) {
 318                         secno = GET_SECNO(sbi, p.min_segno);
 319                         if (gc_type == FG_GC)
 320                                 sbi->cur_victim_sec = secno;
 321                         else
 322                                 set_bit(secno, dirty_i->victim_secmap);
 323                 }
 324                 *result = (p.min_segno / p.ofs_unit) * p.ofs_unit;
 325
 326                 trace_f2fs_get_victim(sbi->sb, type, gc_type, &p,
 327                                 sbi->cur_victim_sec,
 328                                 prefree_segments(sbi), free_segments(sbi));
 329         }
 330         mutex_unlock(&dirty_i->seglist_lock);
 331
 332         return (p.min_segno == NULL_SEGNO) ? 0 : 1;
 333 }
 334
 335 static const struct victim_selection default_v_ops = {
 336         .get_victim = get_victim_by_default,
 337 };
 338
 339 static struct inode *find_gc_inode(struct gc_inode_list *gc_list, nid_t ino)
 340 {
 341         struct inode_entry *ie;
 342
 343         ie = radix_tree_lookup(&gc_list->iroot, ino);
 344         if (ie)
 345                 return ie->inode;
 346         return NULL;
 347 }
 348
 349 static void add_gc_inode(struct gc_inode_list *gc_list, struct inode *inode)
 350 {
 351         struct inode_entry *new_ie;
 352
 353         if (inode == find_gc_inode(gc_list, inode->i_ino)) {
 354                 iput(inode);
 355                 return;
 356         }
 357         new_ie = f2fs_kmem_cache_alloc(inode_entry_slab, GFP_NOFS);
 358         new_ie->inode = inode;
 359
 360         f2fs_radix_tree_insert(&gc_list->iroot, inode->i_ino, new_ie);
 361         list_add_tail(&new_ie->list, &gc_list->ilist);
 362 }
 363
 364 static void put_gc_inode(struct gc_inode_list *gc_list)
 365 {
 366         struct inode_entry *ie, *next_ie;
 367         list_for_each_entry_safe(ie, next_ie, &gc_list->ilist, list) {
 368                 radix_tree_delete(&gc_list->iroot, ie->inode->i_ino);
 369                 iput(ie->inode);
 370                 list_del(&ie->list);
 371                 kmem_cache_free(inode_entry_slab, ie);
 372         }
 373 }
 374
 375 static int check_valid_map(struct f2fs_sb_info *sbi,
 376                                 unsigned int segno, int offset)
 377 {
 378         struct sit_info *sit_i = SIT_I(sbi);
 379         struct seg_entry *sentry;
 380         int ret;
 381
 382         mutex_lock(&sit_i->sentry_lock);
 383         sentry = get_seg_entry(sbi, segno);
 384         ret = f2fs_test_bit(offset, sentry->cur_valid_map);
 385         mutex_unlock(&sit_i->sentry_lock);
 386         return ret;
 387 }
 388
 389 /*
 390  * This function compares node address got in summary with that in NAT.
 391  * On validity, copy that node with cold status, otherwise (invalid node)
 392  * ignore that.
 393  */
 394 static void gc_node_segment(struct f2fs_sb_info *sbi,
 395                 struct f2fs_summary *sum, unsigned int segno, int gc_type)
 396 {
 397         bool initial = true;
 398         struct f2fs_summary *entry;
 399         int off;
 400
 401 next_step:
 402         entry = sum;
 403
 404         for (off = 0; off < sbi->blocks_per_seg; off++, entry++) {
 405                 nid_t nid = le32_to_cpu(entry->nid);
 406                 struct page *node_page;
 407
 408                 /* stop BG_GC if there is not enough free sections. */
 409                 if (gc_type == BG_GC && has_not_enough_free_secs(sbi, 0))
 410                         return;
 411
 412                 if (check_valid_map(sbi, segno, off) == 0)
 413                         continue;
 414
 415                 if (initial) {
 416                         ra_node_page(sbi, nid);
 417                         continue;
 418                 }
 419                 node_page = get_node_page(sbi, nid);
 420                 if (IS_ERR(node_page))
 421                         continue;
 422
 423                 /* block may become invalid during get_node_page */
 424                 if (check_valid_map(sbi, segno, off) == 0) {
 425                         f2fs_put_page(node_page, 1);
 426                         continue;
 427                 }
 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, gc_type);
 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_encrypted_block(struct inode *inode, block_t bidx)
 522 {
 523         struct f2fs_io_info fio = {
 524                 .sbi = F2FS_I_SB(inode),
 525                 .type = DATA,
 526                 .rw = READ_SYNC,
 527                 .encrypted_page = NULL,
 528         };
 529         struct dnode_of_data dn;
 530         struct f2fs_summary sum;
 531         struct node_info ni;
 532         struct page *page;
 533         int err;
 534
 535         /* do not read out */
 536         page = grab_cache_page(inode->i_mapping, bidx);
 537         if (!page)
 538                 return;
 539
 540         set_new_dnode(&dn, inode, NULL, NULL, 0);
 541         err = get_dnode_of_data(&dn, bidx, LOOKUP_NODE);
 542         if (err)
 543                 goto out;
 544
 545         if (unlikely(dn.data_blkaddr == NULL_ADDR))
 546                 goto put_out;
 547
 548         get_node_info(fio.sbi, dn.nid, &ni);
 549         set_summary(&sum, dn.nid, dn.ofs_in_node, ni.version);
 550
 551         /* read page */
 552         fio.page = page;
 553         fio.blk_addr = dn.data_blkaddr;
 554
 555         fio.encrypted_page = grab_cache_page(META_MAPPING(fio.sbi), fio.blk_addr);
 556         if (!fio.encrypted_page)
 557                 goto put_out;
 558
 559         err = f2fs_submit_page_bio(&fio);
 560         if (err)
 561                 goto put_page_out;
 562
 563         /* write page */
 564         lock_page(fio.encrypted_page);
 565
 566         if (unlikely(!PageUptodate(fio.encrypted_page)))
 567                 goto put_page_out;
 568         if (unlikely(fio.encrypted_page->mapping != META_MAPPING(fio.sbi)))
 569                 goto put_page_out;
 570
 571         set_page_dirty(fio.encrypted_page);
 572         f2fs_wait_on_page_writeback(fio.encrypted_page, META);
 573         if (clear_page_dirty_for_io(fio.encrypted_page))
 574                 dec_page_count(fio.sbi, F2FS_DIRTY_META);
 575
 576         set_page_writeback(fio.encrypted_page);
 577
 578         /* allocate block address */
 579         f2fs_wait_on_page_writeback(dn.node_page, NODE);
 580         allocate_data_block(fio.sbi, NULL, fio.blk_addr,
 581                                         &fio.blk_addr, &sum, CURSEG_COLD_DATA);
 582         fio.rw = WRITE_SYNC;
 583         f2fs_submit_page_mbio(&fio);
 584
 585         dn.data_blkaddr = fio.blk_addr;
 586         set_data_blkaddr(&dn);
 587         f2fs_update_extent_cache(&dn);
 588         set_inode_flag(F2FS_I(inode), FI_APPEND_WRITE);
 589         if (page->index == 0)
 590                 set_inode_flag(F2FS_I(inode), FI_FIRST_BLOCK_WRITTEN);
 591 put_page_out:
 592         f2fs_put_page(fio.encrypted_page, 1);
 593 put_out:
 594         f2fs_put_dnode(&dn);
 595 out:
 596         f2fs_put_page(page, 1);
 597 }
 598
 599 static void move_data_page(struct inode *inode, block_t bidx, int gc_type)
 600 {
 601         struct page *page;
 602
 603         page = get_lock_data_page(inode, bidx);
 604         if (IS_ERR(page))
 605                 return;
 606
 607         if (gc_type == BG_GC) {
 608                 if (PageWriteback(page))
 609                         goto out;
 610                 set_page_dirty(page);
 611                 set_cold_data(page);
 612         } else {
 613                 struct f2fs_io_info fio = {
 614                         .sbi = F2FS_I_SB(inode),
 615                         .type = DATA,
 616                         .rw = WRITE_SYNC,
 617                         .page = page,
 618                         .encrypted_page = NULL,
 619                 };
 620                 set_page_dirty(page);
 621                 f2fs_wait_on_page_writeback(page, DATA);
 622                 if (clear_page_dirty_for_io(page))
 623                         inode_dec_dirty_pages(inode);
 624                 set_cold_data(page);
 625                 do_write_data_page(&fio);
 626                 clear_cold_data(page);
 627         }
 628 out:
 629         f2fs_put_page(page, 1);
 630 }
 631
 632 /*
 633  * This function tries to get parent node of victim data block, and identifies
 634  * data block validity. If the block is valid, copy that with cold status and
 635  * modify parent node.
 636  * If the parent node is not valid or the data block address is different,
 637  * the victim data block is ignored.
 638  */
 639 static void gc_data_segment(struct f2fs_sb_info *sbi, struct f2fs_summary *sum,
 640                 struct gc_inode_list *gc_list, unsigned int segno, int gc_type)
 641 {
 642         struct super_block *sb = sbi->sb;
 643         struct f2fs_summary *entry;
 644         block_t start_addr;
 645         int off;
 646         int phase = 0;
 647
 648         start_addr = START_BLOCK(sbi, segno);
 649
 650 next_step:
 651         entry = sum;
 652
 653         for (off = 0; off < sbi->blocks_per_seg; off++, entry++) {
 654                 struct page *data_page;
 655                 struct inode *inode;
 656                 struct node_info dni; /* dnode info for the data */
 657                 unsigned int ofs_in_node, nofs;
 658                 block_t start_bidx;
 659
 660                 /* stop BG_GC if there is not enough free sections. */
 661                 if (gc_type == BG_GC && has_not_enough_free_secs(sbi, 0))
 662                         return;
 663
 664                 if (check_valid_map(sbi, segno, off) == 0)
 665                         continue;
 666
 667                 if (phase == 0) {
 668                         ra_node_page(sbi, le32_to_cpu(entry->nid));
 669                         continue;
 670                 }
 671
 672                 /* Get an inode by ino with checking validity */
 673                 if (check_dnode(sbi, entry, &dni, start_addr + off, &nofs) == 0)
 674                         continue;
 675
 676                 if (phase == 1) {
 677                         ra_node_page(sbi, dni.ino);
 678                         continue;
 679                 }
 680
 681                 ofs_in_node = le16_to_cpu(entry->ofs_in_node);
 682
 683                 if (phase == 2) {
 684                         inode = f2fs_iget(sb, dni.ino);
 685                         if (IS_ERR(inode) || is_bad_inode(inode))
 686                                 continue;
 687
 688                         /* if encrypted inode, let's go phase 3 */
 689                         if (f2fs_encrypted_inode(inode) &&
 690                                                 S_ISREG(inode->i_mode)) {
 691                                 add_gc_inode(gc_list, inode);
 692                                 continue;
 693                         }
 694
 695                         start_bidx = start_bidx_of_node(nofs, F2FS_I(inode));
 696                         data_page = get_read_data_page(inode,
 697                                         start_bidx + ofs_in_node, READA);
 698                         if (IS_ERR(data_page)) {
 699                                 iput(inode);
 700                                 continue;
 701                         }
 702
 703                         f2fs_put_page(data_page, 0);
 704                         add_gc_inode(gc_list, inode);
 705                         continue;
 706                 }
 707
 708                 /* phase 3 */
 709                 inode = find_gc_inode(gc_list, dni.ino);
 710                 if (inode) {
 711                         start_bidx = start_bidx_of_node(nofs, F2FS_I(inode))
 712                                                                 + ofs_in_node;
 713                         if (f2fs_encrypted_inode(inode) && S_ISREG(inode->i_mode))
 714                                 move_encrypted_block(inode, start_bidx);
 715                         else
 716                                 move_data_page(inode, start_bidx, gc_type);
 717                         stat_inc_data_blk_count(sbi, 1, gc_type);
 718                 }
 719         }
 720
 721         if (++phase < 4)
 722                 goto next_step;
 723
 724         if (gc_type == FG_GC) {
 725                 f2fs_submit_merged_bio(sbi, DATA, WRITE);
 726
 727                 /*
 728                  * In the case of FG_GC, it'd be better to reclaim this victim
 729                  * completely.
 730                  */
 731                 if (get_valid_blocks(sbi, segno, 1) != 0) {
 732                         phase = 2;
 733                         goto next_step;
 734                 }
 735         }
 736 }
 737
 738 static int __get_victim(struct f2fs_sb_info *sbi, unsigned int *victim,
 739                         int gc_type)
 740 {
 741         struct sit_info *sit_i = SIT_I(sbi);
 742         int ret;
 743
 744         mutex_lock(&sit_i->sentry_lock);
 745         ret = DIRTY_I(sbi)->v_ops->get_victim(sbi, victim, gc_type,
 746                                               NO_CHECK_TYPE, LFS);
 747         mutex_unlock(&sit_i->sentry_lock);
 748         return ret;
 749 }
 750
 751 static void do_garbage_collect(struct f2fs_sb_info *sbi, unsigned int segno,
 752                                 struct gc_inode_list *gc_list, int gc_type)
 753 {
 754         struct page *sum_page;
 755         struct f2fs_summary_block *sum;
 756         struct blk_plug plug;
 757
 758         /* read segment summary of victim */
 759         sum_page = get_sum_page(sbi, segno);
 760
 761         blk_start_plug(&plug);
 762
 763         sum = page_address(sum_page);
 764
 765         /*
 766          * this is to avoid deadlock:
 767          * - lock_page(sum_page)         - f2fs_replace_block
 768          *  - check_valid_map()            - mutex_lock(sentry_lock)
 769          *   - mutex_lock(sentry_lock)     - change_curseg()
 770          *                                  - lock_page(sum_page)
 771          */
 772         unlock_page(sum_page);
 773
 774         switch (GET_SUM_TYPE((&sum->footer))) {
 775         case SUM_TYPE_NODE:
 776                 gc_node_segment(sbi, sum->entries, segno, gc_type);
 777                 break;
 778         case SUM_TYPE_DATA:
 779                 gc_data_segment(sbi, sum->entries, gc_list, segno, gc_type);
 780                 break;
 781         }
 782         blk_finish_plug(&plug);
 783
 784         stat_inc_seg_count(sbi, GET_SUM_TYPE((&sum->footer)), gc_type);
 785         stat_inc_call_count(sbi->stat_info);
 786
 787         f2fs_put_page(sum_page, 0);
 788 }
 789
 790 int f2fs_gc(struct f2fs_sb_info *sbi)
 791 {
 792         unsigned int segno, i;
 793         int gc_type = BG_GC;
 794         int nfree = 0;
 795         int ret = -1;
 796         struct cp_control cpc;
 797         struct gc_inode_list gc_list = {
 798                 .ilist = LIST_HEAD_INIT(gc_list.ilist),
 799                 .iroot = RADIX_TREE_INIT(GFP_NOFS),
 800         };
 801
 802         cpc.reason = __get_cp_reason(sbi);
 803 gc_more:
 804         if (unlikely(!(sbi->sb->s_flags & MS_ACTIVE)))
 805                 goto stop;
 806         if (unlikely(f2fs_cp_error(sbi)))
 807                 goto stop;
 808
 809         if (gc_type == BG_GC && has_not_enough_free_secs(sbi, nfree)) {
 810                 gc_type = FG_GC;
 811                 write_checkpoint(sbi, &cpc);
 812         }
 813
 814         if (!__get_victim(sbi, &segno, gc_type))
 815                 goto stop;
 816         ret = 0;
 817
 818         /* readahead multi ssa blocks those have contiguous address */
 819         if (sbi->segs_per_sec > 1)
 820                 ra_meta_pages(sbi, GET_SUM_BLOCK(sbi, segno), sbi->segs_per_sec,
 821                                                                 META_SSA);
 822
 823         for (i = 0; i < sbi->segs_per_sec; i++)
 824                 do_garbage_collect(sbi, segno + i, &gc_list, gc_type);
 825
 826         if (gc_type == FG_GC) {
 827                 sbi->cur_victim_sec = NULL_SEGNO;
 828                 nfree++;
 829                 WARN_ON(get_valid_blocks(sbi, segno, sbi->segs_per_sec));
 830         }
 831
 832         if (has_not_enough_free_secs(sbi, nfree))
 833                 goto gc_more;
 834
 835         if (gc_type == FG_GC)
 836                 write_checkpoint(sbi, &cpc);
 837 stop:
 838         mutex_unlock(&sbi->gc_mutex);
 839
 840         put_gc_inode(&gc_list);
 841         return ret;
 842 }
 843
 844 void build_gc_manager(struct f2fs_sb_info *sbi)
 845 {
 846         DIRTY_I(sbi)->v_ops = &default_v_ops;
 847 }