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
  20 #include "f2fs.h"
  21 #include "node.h"
  22 #include "segment.h"
  23 #include "gc.h"
  24 #include <trace/events/f2fs.h>
  25
  26 static int gc_thread_func(void *data)
  27 {
  28         struct f2fs_sb_info *sbi = data;
  29         struct f2fs_gc_kthread *gc_th = sbi->gc_thread;
  30         wait_queue_head_t *wq = &sbi->gc_thread->gc_wait_queue_head;
  31         unsigned int wait_ms;
  32
  33         wait_ms = gc_th->min_sleep_time;
  34
  35         set_freezable();
  36         do {
  37                 wait_event_interruptible_timeout(*wq,
  38                                 kthread_should_stop() || freezing(current) ||
  39                                 gc_th->gc_wake,
  40                                 msecs_to_jiffies(wait_ms));
  41
  42                 /* give it a try one time */
  43                 if (gc_th->gc_wake)
  44                         gc_th->gc_wake = 0;
  45
  46                 if (try_to_freeze())
  47                         continue;
  48                 if (kthread_should_stop())
  49                         break;
  50
  51                 if (sbi->sb->s_writers.frozen >= SB_FREEZE_WRITE) {
  52                         increase_sleep_time(gc_th, &wait_ms);
  53                         continue;
  54                 }
  55
  56 #ifdef CONFIG_F2FS_FAULT_INJECTION
  57                 if (time_to_inject(sbi, FAULT_CHECKPOINT)) {
  58                         f2fs_show_injection_info(FAULT_CHECKPOINT);
  59                         f2fs_stop_checkpoint(sbi, false);
  60                 }
  61 #endif
  62
  63                 if (!sb_start_write_trylock(sbi->sb))
  64                         continue;
  65
  66                 /*
  67                  * [GC triggering condition]
  68                  * 0. GC is not conducted currently.
  69                  * 1. There are enough dirty segments.
  70                  * 2. IO subsystem is idle by checking the # of writeback pages.
  71                  * 3. IO subsystem is idle by checking the # of requests in
  72                  *    bdev's request list.
  73                  *
  74                  * Note) We have to avoid triggering GCs frequently.
  75                  * Because it is possible that some segments can be
  76                  * invalidated soon after by user update or deletion.
  77                  * So, I'd like to wait some time to collect dirty segments.
  78                  */
  79                 if (sbi->gc_mode == GC_URGENT) {
  80                         wait_ms = gc_th->urgent_sleep_time;
  81                         mutex_lock(&sbi->gc_mutex);
  82                         goto do_gc;
  83                 }
  84
  85                 if (!mutex_trylock(&sbi->gc_mutex))
  86                         goto next;
  87
  88                 if (!is_idle(sbi)) {
  89                         increase_sleep_time(gc_th, &wait_ms);
  90                         mutex_unlock(&sbi->gc_mutex);
  91                         goto next;
  92                 }
  93
  94                 if (has_enough_invalid_blocks(sbi))
  95                         decrease_sleep_time(gc_th, &wait_ms);
  96                 else
  97                         increase_sleep_time(gc_th, &wait_ms);
  98 do_gc:
  99                 stat_inc_bggc_count(sbi);
 100
 101                 /* if return value is not zero, no victim was selected */
 102                 if (f2fs_gc(sbi, test_opt(sbi, FORCE_FG_GC), true, NULL_SEGNO))
 103                         wait_ms = gc_th->no_gc_sleep_time;
 104
 105                 trace_f2fs_background_gc(sbi->sb, wait_ms,
 106                                 prefree_segments(sbi), free_segments(sbi));
 107
 108                 /* balancing f2fs's metadata periodically */
 109                 f2fs_balance_fs_bg(sbi);
 110 next:
 111                 sb_end_write(sbi->sb);
 112
 113         } while (!kthread_should_stop());
 114         return 0;
 115 }
 116
 117 int f2fs_start_gc_thread(struct f2fs_sb_info *sbi)
 118 {
 119         struct f2fs_gc_kthread *gc_th;
 120         dev_t dev = sbi->sb->s_bdev->bd_dev;
 121         int err = 0;
 122
 123         gc_th = f2fs_kmalloc(sbi, sizeof(struct f2fs_gc_kthread), GFP_KERNEL);
 124         if (!gc_th) {
 125                 err = -ENOMEM;
 126                 goto out;
 127         }
 128
 129         gc_th->urgent_sleep_time = DEF_GC_THREAD_URGENT_SLEEP_TIME;
 130         gc_th->min_sleep_time = DEF_GC_THREAD_MIN_SLEEP_TIME;
 131         gc_th->max_sleep_time = DEF_GC_THREAD_MAX_SLEEP_TIME;
 132         gc_th->no_gc_sleep_time = DEF_GC_THREAD_NOGC_SLEEP_TIME;
 133
 134         gc_th->gc_wake= 0;
 135
 136         sbi->gc_thread = gc_th;
 137         init_waitqueue_head(&sbi->gc_thread->gc_wait_queue_head);
 138         sbi->gc_thread->f2fs_gc_task = kthread_run(gc_thread_func, sbi,
 139                         "f2fs_gc-%u:%u", MAJOR(dev), MINOR(dev));
 140         if (IS_ERR(gc_th->f2fs_gc_task)) {
 141                 err = PTR_ERR(gc_th->f2fs_gc_task);
 142                 kfree(gc_th);
 143                 sbi->gc_thread = NULL;
 144         }
 145 out:
 146         return err;
 147 }
 148
 149 void f2fs_stop_gc_thread(struct f2fs_sb_info *sbi)
 150 {
 151         struct f2fs_gc_kthread *gc_th = sbi->gc_thread;
 152         if (!gc_th)
 153                 return;
 154         kthread_stop(gc_th->f2fs_gc_task);
 155         kfree(gc_th);
 156         sbi->gc_thread = NULL;
 157 }
 158
 159 static int select_gc_type(struct f2fs_sb_info *sbi, int gc_type)
 160 {
 161         int gc_mode = (gc_type == BG_GC) ? GC_CB : GC_GREEDY;
 162
 163         switch (sbi->gc_mode) {
 164         case GC_IDLE_CB:
 165                 gc_mode = GC_CB;
 166                 break;
 167         case GC_IDLE_GREEDY:
 168         case GC_URGENT:
 169                 gc_mode = GC_GREEDY;
 170                 break;
 171         }
 172         return gc_mode;
 173 }
 174
 175 static void select_policy(struct f2fs_sb_info *sbi, int gc_type,
 176                         int type, struct victim_sel_policy *p)
 177 {
 178         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
 179
 180         if (p->alloc_mode == SSR) {
 181                 p->gc_mode = GC_GREEDY;
 182                 p->dirty_segmap = dirty_i->dirty_segmap[type];
 183                 p->max_search = dirty_i->nr_dirty[type];
 184                 p->ofs_unit = 1;
 185         } else {
 186                 p->gc_mode = select_gc_type(sbi, gc_type);
 187                 p->dirty_segmap = dirty_i->dirty_segmap[DIRTY];
 188                 p->max_search = dirty_i->nr_dirty[DIRTY];
 189                 p->ofs_unit = sbi->segs_per_sec;
 190         }
 191
 192         /* we need to check every dirty segments in the FG_GC case */
 193         if (gc_type != FG_GC &&
 194                         (sbi->gc_mode != GC_URGENT) &&
 195                         p->max_search > sbi->max_victim_search)
 196                 p->max_search = sbi->max_victim_search;
 197
 198         /* let's select beginning hot/small space first in no_heap mode*/
 199         if (test_opt(sbi, NOHEAP) &&
 200                 (type == CURSEG_HOT_DATA || IS_NODESEG(type)))
 201                 p->offset = 0;
 202         else
 203                 p->offset = SIT_I(sbi)->last_victim[p->gc_mode];
 204 }
 205
 206 static unsigned int get_max_cost(struct f2fs_sb_info *sbi,
 207                                 struct victim_sel_policy *p)
 208 {
 209         /* SSR allocates in a segment unit */
 210         if (p->alloc_mode == SSR)
 211                 return sbi->blocks_per_seg;
 212         if (p->gc_mode == GC_GREEDY)
 213                 return 2 * sbi->blocks_per_seg * p->ofs_unit;
 214         else if (p->gc_mode == GC_CB)
 215                 return UINT_MAX;
 216         else /* No other gc_mode */
 217                 return 0;
 218 }
 219
 220 static unsigned int check_bg_victims(struct f2fs_sb_info *sbi)
 221 {
 222         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
 223         unsigned int secno;
 224
 225         /*
 226          * If the gc_type is FG_GC, we can select victim segments
 227          * selected by background GC before.
 228          * Those segments guarantee they have small valid blocks.
 229          */
 230         for_each_set_bit(secno, dirty_i->victim_secmap, MAIN_SECS(sbi)) {
 231                 if (sec_usage_check(sbi, secno))
 232                         continue;
 233                 clear_bit(secno, dirty_i->victim_secmap);
 234                 return GET_SEG_FROM_SEC(sbi, secno);
 235         }
 236         return NULL_SEGNO;
 237 }
 238
 239 static unsigned int get_cb_cost(struct f2fs_sb_info *sbi, unsigned int segno)
 240 {
 241         struct sit_info *sit_i = SIT_I(sbi);
 242         unsigned int secno = GET_SEC_FROM_SEG(sbi, segno);
 243         unsigned int start = GET_SEG_FROM_SEC(sbi, secno);
 244         unsigned long long mtime = 0;
 245         unsigned int vblocks;
 246         unsigned char age = 0;
 247         unsigned char u;
 248         unsigned int i;
 249
 250         for (i = 0; i < sbi->segs_per_sec; i++)
 251                 mtime += get_seg_entry(sbi, start + i)->mtime;
 252         vblocks = get_valid_blocks(sbi, segno, true);
 253
 254         mtime = div_u64(mtime, sbi->segs_per_sec);
 255         vblocks = div_u64(vblocks, sbi->segs_per_sec);
 256
 257         u = (vblocks * 100) >> sbi->log_blocks_per_seg;
 258
 259         /* Handle if the system time has changed by the user */
 260         if (mtime < sit_i->min_mtime)
 261                 sit_i->min_mtime = mtime;
 262         if (mtime > sit_i->max_mtime)
 263                 sit_i->max_mtime = mtime;
 264         if (sit_i->max_mtime != sit_i->min_mtime)
 265                 age = 100 - div64_u64(100 * (mtime - sit_i->min_mtime),
 266                                 sit_i->max_mtime - sit_i->min_mtime);
 267
 268         return UINT_MAX - ((100 * (100 - u) * age) / (100 + u));
 269 }
 270
 271 static inline unsigned int get_gc_cost(struct f2fs_sb_info *sbi,
 272                         unsigned int segno, struct victim_sel_policy *p)
 273 {
 274         if (p->alloc_mode == SSR)
 275                 return get_seg_entry(sbi, segno)->ckpt_valid_blocks;
 276
 277         /* alloc_mode == LFS */
 278         if (p->gc_mode == GC_GREEDY)
 279                 return get_valid_blocks(sbi, segno, true);
 280         else
 281                 return get_cb_cost(sbi, segno);
 282 }
 283
 284 static unsigned int count_bits(const unsigned long *addr,
 285                                 unsigned int offset, unsigned int len)
 286 {
 287         unsigned int end = offset + len, sum = 0;
 288
 289         while (offset < end) {
 290                 if (test_bit(offset++, addr))
 291                         ++sum;
 292         }
 293         return sum;
 294 }
 295
 296 /*
 297  * This function is called from two paths.
 298  * One is garbage collection and the other is SSR segment selection.
 299  * When it is called during GC, it just gets a victim segment
 300  * and it does not remove it from dirty seglist.
 301  * When it is called from SSR segment selection, it finds a segment
 302  * which has minimum valid blocks and removes it from dirty seglist.
 303  */
 304 static int get_victim_by_default(struct f2fs_sb_info *sbi,
 305                 unsigned int *result, int gc_type, int type, char alloc_mode)
 306 {
 307         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
 308         struct sit_info *sm = SIT_I(sbi);
 309         struct victim_sel_policy p;
 310         unsigned int secno, last_victim;
 311         unsigned int last_segment = MAIN_SEGS(sbi);
 312         unsigned int nsearched = 0;
 313
 314         mutex_lock(&dirty_i->seglist_lock);
 315
 316         p.alloc_mode = alloc_mode;
 317         select_policy(sbi, gc_type, type, &p);
 318
 319         p.min_segno = NULL_SEGNO;
 320         p.min_cost = get_max_cost(sbi, &p);
 321
 322         if (*result != NULL_SEGNO) {
 323                 if (IS_DATASEG(get_seg_entry(sbi, *result)->type) &&
 324                         get_valid_blocks(sbi, *result, false) &&
 325                         !sec_usage_check(sbi, GET_SEC_FROM_SEG(sbi, *result)))
 326                         p.min_segno = *result;
 327                 goto out;
 328         }
 329
 330         if (p.max_search == 0)
 331                 goto out;
 332
 333         last_victim = sm->last_victim[p.gc_mode];
 334         if (p.alloc_mode == LFS && gc_type == FG_GC) {
 335                 p.min_segno = check_bg_victims(sbi);
 336                 if (p.min_segno != NULL_SEGNO)
 337                         goto got_it;
 338         }
 339
 340         while (1) {
 341                 unsigned long cost;
 342                 unsigned int segno;
 343
 344                 segno = find_next_bit(p.dirty_segmap, last_segment, p.offset);
 345                 if (segno >= last_segment) {
 346                         if (sm->last_victim[p.gc_mode]) {
 347                                 last_segment =
 348                                         sm->last_victim[p.gc_mode];
 349                                 sm->last_victim[p.gc_mode] = 0;
 350                                 p.offset = 0;
 351                                 continue;
 352                         }
 353                         break;
 354                 }
 355
 356                 p.offset = segno + p.ofs_unit;
 357                 if (p.ofs_unit > 1) {
 358                         p.offset -= segno % p.ofs_unit;
 359                         nsearched += count_bits(p.dirty_segmap,
 360                                                 p.offset - p.ofs_unit,
 361                                                 p.ofs_unit);
 362                 } else {
 363                         nsearched++;
 364                 }
 365
 366                 secno = GET_SEC_FROM_SEG(sbi, segno);
 367
 368                 if (sec_usage_check(sbi, secno))
 369                         goto next;
 370                 if (gc_type == BG_GC && test_bit(secno, dirty_i->victim_secmap))
 371                         goto next;
 372
 373                 cost = get_gc_cost(sbi, segno, &p);
 374
 375                 if (p.min_cost > cost) {
 376                         p.min_segno = segno;
 377                         p.min_cost = cost;
 378                 }
 379 next:
 380                 if (nsearched >= p.max_search) {
 381                         if (!sm->last_victim[p.gc_mode] && segno <= last_victim)
 382                                 sm->last_victim[p.gc_mode] = last_victim + 1;
 383                         else
 384                                 sm->last_victim[p.gc_mode] = segno + 1;
 385                         sm->last_victim[p.gc_mode] %= MAIN_SEGS(sbi);
 386                         break;
 387                 }
 388         }
 389         if (p.min_segno != NULL_SEGNO) {
 390 got_it:
 391                 if (p.alloc_mode == LFS) {
 392                         secno = GET_SEC_FROM_SEG(sbi, p.min_segno);
 393                         if (gc_type == FG_GC)
 394                                 sbi->cur_victim_sec = secno;
 395                         else
 396                                 set_bit(secno, dirty_i->victim_secmap);
 397                 }
 398                 *result = (p.min_segno / p.ofs_unit) * p.ofs_unit;
 399
 400                 trace_f2fs_get_victim(sbi->sb, type, gc_type, &p,
 401                                 sbi->cur_victim_sec,
 402                                 prefree_segments(sbi), free_segments(sbi));
 403         }
 404 out:
 405         mutex_unlock(&dirty_i->seglist_lock);
 406
 407         return (p.min_segno == NULL_SEGNO) ? 0 : 1;
 408 }
 409
 410 static const struct victim_selection default_v_ops = {
 411         .get_victim = get_victim_by_default,
 412 };
 413
 414 static struct inode *find_gc_inode(struct gc_inode_list *gc_list, nid_t ino)
 415 {
 416         struct inode_entry *ie;
 417
 418         ie = radix_tree_lookup(&gc_list->iroot, ino);
 419         if (ie)
 420                 return ie->inode;
 421         return NULL;
 422 }
 423
 424 static void add_gc_inode(struct gc_inode_list *gc_list, struct inode *inode)
 425 {
 426         struct inode_entry *new_ie;
 427
 428         if (inode == find_gc_inode(gc_list, inode->i_ino)) {
 429                 iput(inode);
 430                 return;
 431         }
 432         new_ie = f2fs_kmem_cache_alloc(f2fs_inode_entry_slab, GFP_NOFS);
 433         new_ie->inode = inode;
 434
 435         f2fs_radix_tree_insert(&gc_list->iroot, inode->i_ino, new_ie);
 436         list_add_tail(&new_ie->list, &gc_list->ilist);
 437 }
 438
 439 static void put_gc_inode(struct gc_inode_list *gc_list)
 440 {
 441         struct inode_entry *ie, *next_ie;
 442         list_for_each_entry_safe(ie, next_ie, &gc_list->ilist, list) {
 443                 radix_tree_delete(&gc_list->iroot, ie->inode->i_ino);
 444                 iput(ie->inode);
 445                 list_del(&ie->list);
 446                 kmem_cache_free(f2fs_inode_entry_slab, ie);
 447         }
 448 }
 449
 450 static int check_valid_map(struct f2fs_sb_info *sbi,
 451                                 unsigned int segno, int offset)
 452 {
 453         struct sit_info *sit_i = SIT_I(sbi);
 454         struct seg_entry *sentry;
 455         int ret;
 456
 457         down_read(&sit_i->sentry_lock);
 458         sentry = get_seg_entry(sbi, segno);
 459         ret = f2fs_test_bit(offset, sentry->cur_valid_map);
 460         up_read(&sit_i->sentry_lock);
 461         return ret;
 462 }
 463
 464 /*
 465  * This function compares node address got in summary with that in NAT.
 466  * On validity, copy that node with cold status, otherwise (invalid node)
 467  * ignore that.
 468  */
 469 static void gc_node_segment(struct f2fs_sb_info *sbi,
 470                 struct f2fs_summary *sum, unsigned int segno, int gc_type)
 471 {
 472         struct f2fs_summary *entry;
 473         block_t start_addr;
 474         int off;
 475         int phase = 0;
 476         bool fggc = (gc_type == FG_GC);
 477
 478         start_addr = START_BLOCK(sbi, segno);
 479
 480 next_step:
 481         entry = sum;
 482
 483         if (fggc && phase == 2)
 484                 atomic_inc(&sbi->wb_sync_req[NODE]);
 485
 486         for (off = 0; off < sbi->blocks_per_seg; off++, entry++) {
 487                 nid_t nid = le32_to_cpu(entry->nid);
 488                 struct page *node_page;
 489                 struct node_info ni;
 490
 491                 /* stop BG_GC if there is not enough free sections. */
 492                 if (gc_type == BG_GC && has_not_enough_free_secs(sbi, 0, 0))
 493                         return;
 494
 495                 if (check_valid_map(sbi, segno, off) == 0)
 496                         continue;
 497
 498                 if (phase == 0) {
 499                         f2fs_ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nid), 1,
 500                                                         META_NAT, true);
 501                         continue;
 502                 }
 503
 504                 if (phase == 1) {
 505                         f2fs_ra_node_page(sbi, nid);
 506                         continue;
 507                 }
 508
 509                 /* phase == 2 */
 510                 node_page = f2fs_get_node_page(sbi, nid);
 511                 if (IS_ERR(node_page))
 512                         continue;
 513
 514                 /* block may become invalid during f2fs_get_node_page */
 515                 if (check_valid_map(sbi, segno, off) == 0) {
 516                         f2fs_put_page(node_page, 1);
 517                         continue;
 518                 }
 519
 520                 f2fs_get_node_info(sbi, nid, &ni);
 521                 if (ni.blk_addr != start_addr + off) {
 522                         f2fs_put_page(node_page, 1);
 523                         continue;
 524                 }
 525
 526                 f2fs_move_node_page(node_page, gc_type);
 527                 stat_inc_node_blk_count(sbi, 1, gc_type);
 528         }
 529
 530         if (++phase < 3)
 531                 goto next_step;
 532
 533         if (fggc)
 534                 atomic_dec(&sbi->wb_sync_req[NODE]);
 535 }
 536
 537 /*
 538  * Calculate start block index indicating the given node offset.
 539  * Be careful, caller should give this node offset only indicating direct node
 540  * blocks. If any node offsets, which point the other types of node blocks such
 541  * as indirect or double indirect node blocks, are given, it must be a caller's
 542  * bug.
 543  */
 544 block_t f2fs_start_bidx_of_node(unsigned int node_ofs, struct inode *inode)
 545 {
 546         unsigned int indirect_blks = 2 * NIDS_PER_BLOCK + 4;
 547         unsigned int bidx;
 548
 549         if (node_ofs == 0)
 550                 return 0;
 551
 552         if (node_ofs <= 2) {
 553                 bidx = node_ofs - 1;
 554         } else if (node_ofs <= indirect_blks) {
 555                 int dec = (node_ofs - 4) / (NIDS_PER_BLOCK + 1);
 556                 bidx = node_ofs - 2 - dec;
 557         } else {
 558                 int dec = (node_ofs - indirect_blks - 3) / (NIDS_PER_BLOCK + 1);
 559                 bidx = node_ofs - 5 - dec;
 560         }
 561         return bidx * ADDRS_PER_BLOCK + ADDRS_PER_INODE(inode);
 562 }
 563
 564 static bool is_alive(struct f2fs_sb_info *sbi, struct f2fs_summary *sum,
 565                 struct node_info *dni, block_t blkaddr, unsigned int *nofs)
 566 {
 567         struct page *node_page;
 568         nid_t nid;
 569         unsigned int ofs_in_node;
 570         block_t source_blkaddr;
 571
 572         nid = le32_to_cpu(sum->nid);
 573         ofs_in_node = le16_to_cpu(sum->ofs_in_node);
 574
 575         node_page = f2fs_get_node_page(sbi, nid);
 576         if (IS_ERR(node_page))
 577                 return false;
 578
 579         f2fs_get_node_info(sbi, nid, dni);
 580
 581         if (sum->version != dni->version) {
 582                 f2fs_msg(sbi->sb, KERN_WARNING,
 583                                 "%s: valid data with mismatched node version.",
 584                                 __func__);
 585                 set_sbi_flag(sbi, SBI_NEED_FSCK);
 586         }
 587
 588         *nofs = ofs_of_node(node_page);
 589         source_blkaddr = datablock_addr(NULL, node_page, ofs_in_node);
 590         f2fs_put_page(node_page, 1);
 591
 592         if (source_blkaddr != blkaddr)
 593                 return false;
 594         return true;
 595 }
 596
 597 /*
 598  * Move data block via META_MAPPING while keeping locked data page.
 599  * This can be used to move blocks, aka LBAs, directly on disk.
 600  */
 601 static void move_data_block(struct inode *inode, block_t bidx,
 602                                 int gc_type, unsigned int segno, int off)
 603 {
 604         struct f2fs_io_info fio = {
 605                 .sbi = F2FS_I_SB(inode),
 606                 .ino = inode->i_ino,
 607                 .type = DATA,
 608                 .temp = COLD,
 609                 .op = REQ_OP_READ,
 610                 .op_flags = 0,
 611                 .encrypted_page = NULL,
 612                 .in_list = false,
 613                 .retry = false,
 614         };
 615         struct dnode_of_data dn;
 616         struct f2fs_summary sum;
 617         struct node_info ni;
 618         struct page *page;
 619         block_t newaddr;
 620         int err;
 621         bool lfs_mode = test_opt(fio.sbi, LFS);
 622
 623         /* do not read out */
 624         page = f2fs_grab_cache_page(inode->i_mapping, bidx, false);
 625         if (!page)
 626                 return;
 627
 628         if (!check_valid_map(F2FS_I_SB(inode), segno, off))
 629                 goto out;
 630
 631         if (f2fs_is_atomic_file(inode)) {
 632                 F2FS_I(inode)->i_gc_failures[GC_FAILURE_ATOMIC]++;
 633                 F2FS_I_SB(inode)->skipped_atomic_files[gc_type]++;
 634                 goto out;
 635         }
 636
 637         if (f2fs_is_pinned_file(inode)) {
 638                 f2fs_pin_file_control(inode, true);
 639                 goto out;
 640         }
 641
 642         set_new_dnode(&dn, inode, NULL, NULL, 0);
 643         err = f2fs_get_dnode_of_data(&dn, bidx, LOOKUP_NODE);
 644         if (err)
 645                 goto out;
 646
 647         if (unlikely(dn.data_blkaddr == NULL_ADDR)) {
 648                 ClearPageUptodate(page);
 649                 goto put_out;
 650         }
 651
 652         /*
 653          * don't cache encrypted data into meta inode until previous dirty
 654          * data were writebacked to avoid racing between GC and flush.
 655          */
 656         f2fs_wait_on_page_writeback(page, DATA, true);
 657
 658         f2fs_get_node_info(fio.sbi, dn.nid, &ni);
 659         set_summary(&sum, dn.nid, dn.ofs_in_node, ni.version);
 660
 661         /* read page */
 662         fio.page = page;
 663         fio.new_blkaddr = fio.old_blkaddr = dn.data_blkaddr;
 664
 665         if (lfs_mode)
 666                 down_write(&fio.sbi->io_order_lock);
 667
 668         f2fs_allocate_data_block(fio.sbi, NULL, fio.old_blkaddr, &newaddr,
 669                                         &sum, CURSEG_COLD_DATA, NULL, false);
 670
 671         fio.encrypted_page = f2fs_pagecache_get_page(META_MAPPING(fio.sbi),
 672                                 newaddr, FGP_LOCK | FGP_CREAT, GFP_NOFS);
 673         if (!fio.encrypted_page) {
 674                 err = -ENOMEM;
 675                 goto recover_block;
 676         }
 677
 678         err = f2fs_submit_page_bio(&fio);
 679         if (err)
 680                 goto put_page_out;
 681
 682         /* write page */
 683         lock_page(fio.encrypted_page);
 684
 685         if (unlikely(fio.encrypted_page->mapping != META_MAPPING(fio.sbi))) {
 686                 err = -EIO;
 687                 goto put_page_out;
 688         }
 689         if (unlikely(!PageUptodate(fio.encrypted_page))) {
 690                 err = -EIO;
 691                 goto put_page_out;
 692         }
 693
 694         set_page_dirty(fio.encrypted_page);
 695         f2fs_wait_on_page_writeback(fio.encrypted_page, DATA, true);
 696         if (clear_page_dirty_for_io(fio.encrypted_page))
 697                 dec_page_count(fio.sbi, F2FS_DIRTY_META);
 698
 699         set_page_writeback(fio.encrypted_page);
 700         ClearPageError(page);
 701
 702         /* allocate block address */
 703         f2fs_wait_on_page_writeback(dn.node_page, NODE, true);
 704
 705         fio.op = REQ_OP_WRITE;
 706         fio.op_flags = REQ_SYNC;
 707         fio.new_blkaddr = newaddr;
 708         f2fs_submit_page_write(&fio);
 709         if (fio.retry) {
 710                 if (PageWriteback(fio.encrypted_page))
 711                         end_page_writeback(fio.encrypted_page);
 712                 goto put_page_out;
 713         }
 714
 715         f2fs_update_iostat(fio.sbi, FS_GC_DATA_IO, F2FS_BLKSIZE);
 716
 717         f2fs_update_data_blkaddr(&dn, newaddr);
 718         set_inode_flag(inode, FI_APPEND_WRITE);
 719         if (page->index == 0)
 720                 set_inode_flag(inode, FI_FIRST_BLOCK_WRITTEN);
 721 put_page_out:
 722         f2fs_put_page(fio.encrypted_page, 1);
 723 recover_block:
 724         if (lfs_mode)
 725                 up_write(&fio.sbi->io_order_lock);
 726         if (err)
 727                 f2fs_do_replace_block(fio.sbi, &sum, newaddr, fio.old_blkaddr,
 728                                                                 true, true);
 729 put_out:
 730         f2fs_put_dnode(&dn);
 731 out:
 732         f2fs_put_page(page, 1);
 733 }
 734
 735 static void move_data_page(struct inode *inode, block_t bidx, int gc_type,
 736                                                         unsigned int segno, int off)
 737 {
 738         struct page *page;
 739
 740         page = f2fs_get_lock_data_page(inode, bidx, true);
 741         if (IS_ERR(page))
 742                 return;
 743
 744         if (!check_valid_map(F2FS_I_SB(inode), segno, off))
 745                 goto out;
 746
 747         if (f2fs_is_atomic_file(inode)) {
 748                 F2FS_I(inode)->i_gc_failures[GC_FAILURE_ATOMIC]++;
 749                 F2FS_I_SB(inode)->skipped_atomic_files[gc_type]++;
 750                 goto out;
 751         }
 752         if (f2fs_is_pinned_file(inode)) {
 753                 if (gc_type == FG_GC)
 754                         f2fs_pin_file_control(inode, true);
 755                 goto out;
 756         }
 757
 758         if (gc_type == BG_GC) {
 759                 if (PageWriteback(page))
 760                         goto out;
 761                 set_page_dirty(page);
 762                 set_cold_data(page);
 763         } else {
 764                 struct f2fs_io_info fio = {
 765                         .sbi = F2FS_I_SB(inode),
 766                         .ino = inode->i_ino,
 767                         .type = DATA,
 768                         .temp = COLD,
 769                         .op = REQ_OP_WRITE,
 770                         .op_flags = REQ_SYNC,
 771                         .old_blkaddr = NULL_ADDR,
 772                         .page = page,
 773                         .encrypted_page = NULL,
 774                         .need_lock = LOCK_REQ,
 775                         .io_type = FS_GC_DATA_IO,
 776                 };
 777                 bool is_dirty = PageDirty(page);
 778                 int err;
 779
 780 retry:
 781                 set_page_dirty(page);
 782                 f2fs_wait_on_page_writeback(page, DATA, true);
 783                 if (clear_page_dirty_for_io(page)) {
 784                         inode_dec_dirty_pages(inode);
 785                         f2fs_remove_dirty_inode(inode);
 786                 }
 787
 788                 set_cold_data(page);
 789
 790                 err = f2fs_do_write_data_page(&fio);
 791                 if (err) {
 792                         clear_cold_data(page);
 793                         if (err == -ENOMEM) {
 794                                 congestion_wait(BLK_RW_ASYNC, HZ/50);
 795                                 goto retry;
 796                         }
 797                         if (is_dirty)
 798                                 set_page_dirty(page);
 799                 }
 800         }
 801 out:
 802         f2fs_put_page(page, 1);
 803 }
 804
 805 /*
 806  * This function tries to get parent node of victim data block, and identifies
 807  * data block validity. If the block is valid, copy that with cold status and
 808  * modify parent node.
 809  * If the parent node is not valid or the data block address is different,
 810  * the victim data block is ignored.
 811  */
 812 static void gc_data_segment(struct f2fs_sb_info *sbi, struct f2fs_summary *sum,
 813                 struct gc_inode_list *gc_list, unsigned int segno, int gc_type)
 814 {
 815         struct super_block *sb = sbi->sb;
 816         struct f2fs_summary *entry;
 817         block_t start_addr;
 818         int off;
 819         int phase = 0;
 820
 821         start_addr = START_BLOCK(sbi, segno);
 822
 823 next_step:
 824         entry = sum;
 825
 826         for (off = 0; off < sbi->blocks_per_seg; off++, entry++) {
 827                 struct page *data_page;
 828                 struct inode *inode;
 829                 struct node_info dni; /* dnode info for the data */
 830                 unsigned int ofs_in_node, nofs;
 831                 block_t start_bidx;
 832                 nid_t nid = le32_to_cpu(entry->nid);
 833
 834                 /* stop BG_GC if there is not enough free sections. */
 835                 if (gc_type == BG_GC && has_not_enough_free_secs(sbi, 0, 0))
 836                         return;
 837
 838                 if (check_valid_map(sbi, segno, off) == 0)
 839                         continue;
 840
 841                 if (phase == 0) {
 842                         f2fs_ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nid), 1,
 843                                                         META_NAT, true);
 844                         continue;
 845                 }
 846
 847                 if (phase == 1) {
 848                         f2fs_ra_node_page(sbi, nid);
 849                         continue;
 850                 }
 851
 852                 /* Get an inode by ino with checking validity */
 853                 if (!is_alive(sbi, entry, &dni, start_addr + off, &nofs))
 854                         continue;
 855
 856                 if (phase == 2) {
 857                         f2fs_ra_node_page(sbi, dni.ino);
 858                         continue;
 859                 }
 860
 861                 ofs_in_node = le16_to_cpu(entry->ofs_in_node);
 862
 863                 if (phase == 3) {
 864                         inode = f2fs_iget(sb, dni.ino);
 865                         if (IS_ERR(inode) || is_bad_inode(inode))
 866                                 continue;
 867
 868                         /* if inode uses special I/O path, let's go phase 3 */
 869                         if (f2fs_post_read_required(inode)) {
 870                                 add_gc_inode(gc_list, inode);
 871                                 continue;
 872                         }
 873
 874                         if (!down_write_trylock(
 875                                 &F2FS_I(inode)->i_gc_rwsem[WRITE])) {
 876                                 iput(inode);
 877                                 continue;
 878                         }
 879
 880                         start_bidx = f2fs_start_bidx_of_node(nofs, inode);
 881                         data_page = f2fs_get_read_data_page(inode,
 882                                         start_bidx + ofs_in_node, REQ_RAHEAD,
 883                                         true);
 884                         up_write(&F2FS_I(inode)->i_gc_rwsem[WRITE]);
 885                         if (IS_ERR(data_page)) {
 886                                 iput(inode);
 887                                 continue;
 888                         }
 889
 890                         f2fs_put_page(data_page, 0);
 891                         add_gc_inode(gc_list, inode);
 892                         continue;
 893                 }
 894
 895                 /* phase 4 */
 896                 inode = find_gc_inode(gc_list, dni.ino);
 897                 if (inode) {
 898                         struct f2fs_inode_info *fi = F2FS_I(inode);
 899                         bool locked = false;
 900
 901                         if (S_ISREG(inode->i_mode)) {
 902                                 if (!down_write_trylock(&fi->i_gc_rwsem[READ]))
 903                                         continue;
 904                                 if (!down_write_trylock(
 905                                                 &fi->i_gc_rwsem[WRITE])) {
 906                                         up_write(&fi->i_gc_rwsem[READ]);
 907                                         continue;
 908                                 }
 909                                 locked = true;
 910
 911                                 /* wait for all inflight aio data */
 912                                 inode_dio_wait(inode);
 913                         }
 914
 915                         start_bidx = f2fs_start_bidx_of_node(nofs, inode)
 916                                                                 + ofs_in_node;
 917                         if (f2fs_post_read_required(inode))
 918                                 move_data_block(inode, start_bidx, gc_type,
 919                                                                 segno, off);
 920                         else
 921                                 move_data_page(inode, start_bidx, gc_type,
 922                                                                 segno, off);
 923
 924                         if (locked) {
 925                                 up_write(&fi->i_gc_rwsem[WRITE]);
 926                                 up_write(&fi->i_gc_rwsem[READ]);
 927                         }
 928
 929                         stat_inc_data_blk_count(sbi, 1, gc_type);
 930                 }
 931         }
 932
 933         if (++phase < 5)
 934                 goto next_step;
 935 }
 936
 937 static int __get_victim(struct f2fs_sb_info *sbi, unsigned int *victim,
 938                         int gc_type)
 939 {
 940         struct sit_info *sit_i = SIT_I(sbi);
 941         int ret;
 942
 943         down_write(&sit_i->sentry_lock);
 944         ret = DIRTY_I(sbi)->v_ops->get_victim(sbi, victim, gc_type,
 945                                               NO_CHECK_TYPE, LFS);
 946         up_write(&sit_i->sentry_lock);
 947         return ret;
 948 }
 949
 950 static int do_garbage_collect(struct f2fs_sb_info *sbi,
 951                                 unsigned int start_segno,
 952                                 struct gc_inode_list *gc_list, int gc_type)
 953 {
 954         struct page *sum_page;
 955         struct f2fs_summary_block *sum;
 956         struct blk_plug plug;
 957         unsigned int segno = start_segno;
 958         unsigned int end_segno = start_segno + sbi->segs_per_sec;
 959         int seg_freed = 0;
 960         unsigned char type = IS_DATASEG(get_seg_entry(sbi, segno)->type) ?
 961                                                 SUM_TYPE_DATA : SUM_TYPE_NODE;
 962
 963         /* readahead multi ssa blocks those have contiguous address */
 964         if (sbi->segs_per_sec > 1)
 965                 f2fs_ra_meta_pages(sbi, GET_SUM_BLOCK(sbi, segno),
 966                                         sbi->segs_per_sec, META_SSA, true);
 967
 968         /* reference all summary page */
 969         while (segno < end_segno) {
 970                 sum_page = f2fs_get_sum_page(sbi, segno++);
 971                 unlock_page(sum_page);
 972         }
 973
 974         blk_start_plug(&plug);
 975
 976         for (segno = start_segno; segno < end_segno; segno++) {
 977
 978                 /* find segment summary of victim */
 979                 sum_page = find_get_page(META_MAPPING(sbi),
 980                                         GET_SUM_BLOCK(sbi, segno));
 981                 f2fs_put_page(sum_page, 0);
 982
 983                 if (get_valid_blocks(sbi, segno, false) == 0 ||
 984                                 !PageUptodate(sum_page) ||
 985                                 unlikely(f2fs_cp_error(sbi)))
 986                         goto next;
 987
 988                 sum = page_address(sum_page);
 989                 f2fs_bug_on(sbi, type != GET_SUM_TYPE((&sum->footer)));
 990
 991                 /*
 992                  * this is to avoid deadlock:
 993                  * - lock_page(sum_page)         - f2fs_replace_block
 994                  *  - check_valid_map()            - down_write(sentry_lock)
 995                  *   - down_read(sentry_lock)     - change_curseg()
 996                  *                                  - lock_page(sum_page)
 997                  */
 998                 if (type == SUM_TYPE_NODE)
 999                         gc_node_segment(sbi, sum->entries, segno, gc_type);
1000                 else
1001                         gc_data_segment(sbi, sum->entries, gc_list, segno,
1002                                                                 gc_type);
1003
1004                 stat_inc_seg_count(sbi, type, gc_type);
1005
1006                 if (gc_type == FG_GC &&
1007                                 get_valid_blocks(sbi, segno, false) == 0)
1008                         seg_freed++;
1009 next:
1010                 f2fs_put_page(sum_page, 0);
1011         }
1012
1013         if (gc_type == FG_GC)
1014                 f2fs_submit_merged_write(sbi,
1015                                 (type == SUM_TYPE_NODE) ? NODE : DATA);
1016
1017         blk_finish_plug(&plug);
1018
1019         stat_inc_call_count(sbi->stat_info);
1020
1021         return seg_freed;
1022 }
1023
1024 int f2fs_gc(struct f2fs_sb_info *sbi, bool sync,
1025                         bool background, unsigned int segno)
1026 {
1027         int gc_type = sync ? FG_GC : BG_GC;
1028         int sec_freed = 0, seg_freed = 0, total_freed = 0;
1029         int ret = 0;
1030         struct cp_control cpc;
1031         unsigned int init_segno = segno;
1032         struct gc_inode_list gc_list = {
1033                 .ilist = LIST_HEAD_INIT(gc_list.ilist),
1034                 .iroot = RADIX_TREE_INIT(gc_list.iroot, GFP_NOFS),
1035         };
1036         unsigned long long last_skipped = sbi->skipped_atomic_files[FG_GC];
1037         unsigned int skipped_round = 0, round = 0;
1038
1039         trace_f2fs_gc_begin(sbi->sb, sync, background,
1040                                 get_pages(sbi, F2FS_DIRTY_NODES),
1041                                 get_pages(sbi, F2FS_DIRTY_DENTS),
1042                                 get_pages(sbi, F2FS_DIRTY_IMETA),
1043                                 free_sections(sbi),
1044                                 free_segments(sbi),
1045                                 reserved_segments(sbi),
1046                                 prefree_segments(sbi));
1047
1048         cpc.reason = __get_cp_reason(sbi);
1049 gc_more:
1050         if (unlikely(!(sbi->sb->s_flags & SB_ACTIVE))) {
1051                 ret = -EINVAL;
1052                 goto stop;
1053         }
1054         if (unlikely(f2fs_cp_error(sbi))) {
1055                 ret = -EIO;
1056                 goto stop;
1057         }
1058
1059         if (gc_type == BG_GC && has_not_enough_free_secs(sbi, 0, 0)) {
1060                 /*
1061                  * For example, if there are many prefree_segments below given
1062                  * threshold, we can make them free by checkpoint. Then, we
1063                  * secure free segments which doesn't need fggc any more.
1064                  */
1065                 if (prefree_segments(sbi)) {
1066                         ret = f2fs_write_checkpoint(sbi, &cpc);
1067                         if (ret)
1068                                 goto stop;
1069                 }
1070                 if (has_not_enough_free_secs(sbi, 0, 0))
1071                         gc_type = FG_GC;
1072         }
1073
1074         /* f2fs_balance_fs doesn't need to do BG_GC in critical path. */
1075         if (gc_type == BG_GC && !background) {
1076                 ret = -EINVAL;
1077                 goto stop;
1078         }
1079         if (!__get_victim(sbi, &segno, gc_type)) {
1080                 ret = -ENODATA;
1081                 goto stop;
1082         }
1083
1084         seg_freed = do_garbage_collect(sbi, segno, &gc_list, gc_type);
1085         if (gc_type == FG_GC && seg_freed == sbi->segs_per_sec)
1086                 sec_freed++;
1087         total_freed += seg_freed;
1088
1089         if (gc_type == FG_GC) {
1090                 if (sbi->skipped_atomic_files[FG_GC] > last_skipped)
1091                         skipped_round++;
1092                 last_skipped = sbi->skipped_atomic_files[FG_GC];
1093                 round++;
1094         }
1095
1096         if (gc_type == FG_GC)
1097                 sbi->cur_victim_sec = NULL_SEGNO;
1098
1099         if (!sync) {
1100                 if (has_not_enough_free_secs(sbi, sec_freed, 0)) {
1101                         if (skipped_round > MAX_SKIP_ATOMIC_COUNT &&
1102                                 skipped_round * 2 >= round)
1103                                 f2fs_drop_inmem_pages_all(sbi, true);
1104                         segno = NULL_SEGNO;
1105                         goto gc_more;
1106                 }
1107
1108                 if (gc_type == FG_GC)
1109                         ret = f2fs_write_checkpoint(sbi, &cpc);
1110         }
1111 stop:
1112         SIT_I(sbi)->last_victim[ALLOC_NEXT] = 0;
1113         SIT_I(sbi)->last_victim[FLUSH_DEVICE] = init_segno;
1114
1115         trace_f2fs_gc_end(sbi->sb, ret, total_freed, sec_freed,
1116                                 get_pages(sbi, F2FS_DIRTY_NODES),
1117                                 get_pages(sbi, F2FS_DIRTY_DENTS),
1118                                 get_pages(sbi, F2FS_DIRTY_IMETA),
1119                                 free_sections(sbi),
1120                                 free_segments(sbi),
1121                                 reserved_segments(sbi),
1122                                 prefree_segments(sbi));
1123
1124         mutex_unlock(&sbi->gc_mutex);
1125
1126         put_gc_inode(&gc_list);
1127
1128         if (sync)
1129                 ret = sec_freed ? 0 : -EAGAIN;
1130         return ret;
1131 }
1132
1133 void f2fs_build_gc_manager(struct f2fs_sb_info *sbi)
1134 {
1135         DIRTY_I(sbi)->v_ops = &default_v_ops;
1136
1137         sbi->gc_pin_file_threshold = DEF_GC_FAILED_PINNED_FILES;
1138
1139         /* give warm/cold data area from slower device */
1140         if (sbi->s_ndevs && sbi->segs_per_sec == 1)
1141                 SIT_I(sbi)->last_victim[ALLOC_NEXT] =
1142                                 GET_SEGNO(sbi, FDEV(0).end_blk) + 1;
1143 }