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 (gc_th->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 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_idle = 0;
 135         gc_th->gc_urgent = 0;
 136         gc_th->gc_wake= 0;
 137
 138         sbi->gc_thread = gc_th;
 139         init_waitqueue_head(&sbi->gc_thread->gc_wait_queue_head);
 140         sbi->gc_thread->f2fs_gc_task = kthread_run(gc_thread_func, sbi,
 141                         "f2fs_gc-%u:%u", MAJOR(dev), MINOR(dev));
 142         if (IS_ERR(gc_th->f2fs_gc_task)) {
 143                 err = PTR_ERR(gc_th->f2fs_gc_task);
 144                 kfree(gc_th);
 145                 sbi->gc_thread = NULL;
 146         }
 147 out:
 148         return err;
 149 }
 150
 151 void stop_gc_thread(struct f2fs_sb_info *sbi)
 152 {
 153         struct f2fs_gc_kthread *gc_th = sbi->gc_thread;
 154         if (!gc_th)
 155                 return;
 156         kthread_stop(gc_th->f2fs_gc_task);
 157         kfree(gc_th);
 158         sbi->gc_thread = NULL;
 159 }
 160
 161 static int select_gc_type(struct f2fs_gc_kthread *gc_th, int gc_type)
 162 {
 163         int gc_mode = (gc_type == BG_GC) ? GC_CB : GC_GREEDY;
 164
 165         if (!gc_th)
 166                 return gc_mode;
 167
 168         if (gc_th->gc_idle) {
 169                 if (gc_th->gc_idle == 1)
 170                         gc_mode = GC_CB;
 171                 else if (gc_th->gc_idle == 2)
 172                         gc_mode = GC_GREEDY;
 173         }
 174         if (gc_th->gc_urgent)
 175                 gc_mode = GC_GREEDY;
 176         return gc_mode;
 177 }
 178
 179 static void select_policy(struct f2fs_sb_info *sbi, int gc_type,
 180                         int type, struct victim_sel_policy *p)
 181 {
 182         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
 183
 184         if (p->alloc_mode == SSR) {
 185                 p->gc_mode = GC_GREEDY;
 186                 p->dirty_segmap = dirty_i->dirty_segmap[type];
 187                 p->max_search = dirty_i->nr_dirty[type];
 188                 p->ofs_unit = 1;
 189         } else {
 190                 p->gc_mode = select_gc_type(sbi->gc_thread, gc_type);
 191                 p->dirty_segmap = dirty_i->dirty_segmap[DIRTY];
 192                 p->max_search = dirty_i->nr_dirty[DIRTY];
 193                 p->ofs_unit = sbi->segs_per_sec;
 194         }
 195
 196         /* we need to check every dirty segments in the FG_GC case */
 197         if (gc_type != FG_GC &&
 198                         (sbi->gc_thread && !sbi->gc_thread->gc_urgent) &&
 199                         p->max_search > sbi->max_victim_search)
 200                 p->max_search = sbi->max_victim_search;
 201
 202         /* let's select beginning hot/small space first in no_heap mode*/
 203         if (test_opt(sbi, NOHEAP) &&
 204                 (type == CURSEG_HOT_DATA || IS_NODESEG(type)))
 205                 p->offset = 0;
 206         else
 207                 p->offset = SIT_I(sbi)->last_victim[p->gc_mode];
 208 }
 209
 210 static unsigned int get_max_cost(struct f2fs_sb_info *sbi,
 211                                 struct victim_sel_policy *p)
 212 {
 213         /* SSR allocates in a segment unit */
 214         if (p->alloc_mode == SSR)
 215                 return sbi->blocks_per_seg;
 216         if (p->gc_mode == GC_GREEDY)
 217                 return 2 * sbi->blocks_per_seg * p->ofs_unit;
 218         else if (p->gc_mode == GC_CB)
 219                 return UINT_MAX;
 220         else /* No other gc_mode */
 221                 return 0;
 222 }
 223
 224 static unsigned int check_bg_victims(struct f2fs_sb_info *sbi)
 225 {
 226         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
 227         unsigned int secno;
 228
 229         /*
 230          * If the gc_type is FG_GC, we can select victim segments
 231          * selected by background GC before.
 232          * Those segments guarantee they have small valid blocks.
 233          */
 234         for_each_set_bit(secno, dirty_i->victim_secmap, MAIN_SECS(sbi)) {
 235                 if (sec_usage_check(sbi, secno))
 236                         continue;
 237
 238                 if (no_fggc_candidate(sbi, secno))
 239                         continue;
 240
 241                 clear_bit(secno, dirty_i->victim_secmap);
 242                 return GET_SEG_FROM_SEC(sbi, secno);
 243         }
 244         return NULL_SEGNO;
 245 }
 246
 247 static unsigned int get_cb_cost(struct f2fs_sb_info *sbi, unsigned int segno)
 248 {
 249         struct sit_info *sit_i = SIT_I(sbi);
 250         unsigned int secno = GET_SEC_FROM_SEG(sbi, segno);
 251         unsigned int start = GET_SEG_FROM_SEC(sbi, secno);
 252         unsigned long long mtime = 0;
 253         unsigned int vblocks;
 254         unsigned char age = 0;
 255         unsigned char u;
 256         unsigned int i;
 257
 258         for (i = 0; i < sbi->segs_per_sec; i++)
 259                 mtime += get_seg_entry(sbi, start + i)->mtime;
 260         vblocks = get_valid_blocks(sbi, segno, true);
 261
 262         mtime = div_u64(mtime, sbi->segs_per_sec);
 263         vblocks = div_u64(vblocks, sbi->segs_per_sec);
 264
 265         u = (vblocks * 100) >> sbi->log_blocks_per_seg;
 266
 267         /* Handle if the system time has changed by the user */
 268         if (mtime < sit_i->min_mtime)
 269                 sit_i->min_mtime = mtime;
 270         if (mtime > sit_i->max_mtime)
 271                 sit_i->max_mtime = mtime;
 272         if (sit_i->max_mtime != sit_i->min_mtime)
 273                 age = 100 - div64_u64(100 * (mtime - sit_i->min_mtime),
 274                                 sit_i->max_mtime - sit_i->min_mtime);
 275
 276         return UINT_MAX - ((100 * (100 - u) * age) / (100 + u));
 277 }
 278
 279 static inline unsigned int get_gc_cost(struct f2fs_sb_info *sbi,
 280                         unsigned int segno, struct victim_sel_policy *p)
 281 {
 282         if (p->alloc_mode == SSR)
 283                 return get_seg_entry(sbi, segno)->ckpt_valid_blocks;
 284
 285         /* alloc_mode == LFS */
 286         if (p->gc_mode == GC_GREEDY)
 287                 return get_valid_blocks(sbi, segno, true);
 288         else
 289                 return get_cb_cost(sbi, segno);
 290 }
 291
 292 static unsigned int count_bits(const unsigned long *addr,
 293                                 unsigned int offset, unsigned int len)
 294 {
 295         unsigned int end = offset + len, sum = 0;
 296
 297         while (offset < end) {
 298                 if (test_bit(offset++, addr))
 299                         ++sum;
 300         }
 301         return sum;
 302 }
 303
 304 /*
 305  * This function is called from two paths.
 306  * One is garbage collection and the other is SSR segment selection.
 307  * When it is called during GC, it just gets a victim segment
 308  * and it does not remove it from dirty seglist.
 309  * When it is called from SSR segment selection, it finds a segment
 310  * which has minimum valid blocks and removes it from dirty seglist.
 311  */
 312 static int get_victim_by_default(struct f2fs_sb_info *sbi,
 313                 unsigned int *result, int gc_type, int type, char alloc_mode)
 314 {
 315         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
 316         struct sit_info *sm = SIT_I(sbi);
 317         struct victim_sel_policy p;
 318         unsigned int secno, last_victim;
 319         unsigned int last_segment = MAIN_SEGS(sbi);
 320         unsigned int nsearched = 0;
 321
 322         mutex_lock(&dirty_i->seglist_lock);
 323
 324         p.alloc_mode = alloc_mode;
 325         select_policy(sbi, gc_type, type, &p);
 326
 327         p.min_segno = NULL_SEGNO;
 328         p.min_cost = get_max_cost(sbi, &p);
 329
 330         if (*result != NULL_SEGNO) {
 331                 if (IS_DATASEG(get_seg_entry(sbi, *result)->type) &&
 332                         get_valid_blocks(sbi, *result, false) &&
 333                         !sec_usage_check(sbi, GET_SEC_FROM_SEG(sbi, *result)))
 334                         p.min_segno = *result;
 335                 goto out;
 336         }
 337
 338         if (p.max_search == 0)
 339                 goto out;
 340
 341         last_victim = sm->last_victim[p.gc_mode];
 342         if (p.alloc_mode == LFS && gc_type == FG_GC) {
 343                 p.min_segno = check_bg_victims(sbi);
 344                 if (p.min_segno != NULL_SEGNO)
 345                         goto got_it;
 346         }
 347
 348         while (1) {
 349                 unsigned long cost;
 350                 unsigned int segno;
 351
 352                 segno = find_next_bit(p.dirty_segmap, last_segment, p.offset);
 353                 if (segno >= last_segment) {
 354                         if (sm->last_victim[p.gc_mode]) {
 355                                 last_segment =
 356                                         sm->last_victim[p.gc_mode];
 357                                 sm->last_victim[p.gc_mode] = 0;
 358                                 p.offset = 0;
 359                                 continue;
 360                         }
 361                         break;
 362                 }
 363
 364                 p.offset = segno + p.ofs_unit;
 365                 if (p.ofs_unit > 1) {
 366                         p.offset -= segno % p.ofs_unit;
 367                         nsearched += count_bits(p.dirty_segmap,
 368                                                 p.offset - p.ofs_unit,
 369                                                 p.ofs_unit);
 370                 } else {
 371                         nsearched++;
 372                 }
 373
 374                 secno = GET_SEC_FROM_SEG(sbi, segno);
 375
 376                 if (sec_usage_check(sbi, secno))
 377                         goto next;
 378                 if (gc_type == BG_GC && test_bit(secno, dirty_i->victim_secmap))
 379                         goto next;
 380                 if (gc_type == FG_GC && p.alloc_mode == LFS &&
 381                                         no_fggc_candidate(sbi, secno))
 382                         goto next;
 383
 384                 cost = get_gc_cost(sbi, segno, &p);
 385
 386                 if (p.min_cost > cost) {
 387                         p.min_segno = segno;
 388                         p.min_cost = cost;
 389                 }
 390 next:
 391                 if (nsearched >= p.max_search) {
 392                         if (!sm->last_victim[p.gc_mode] && segno <= last_victim)
 393                                 sm->last_victim[p.gc_mode] = last_victim + 1;
 394                         else
 395                                 sm->last_victim[p.gc_mode] = segno + 1;
 396                         sm->last_victim[p.gc_mode] %= MAIN_SEGS(sbi);
 397                         break;
 398                 }
 399         }
 400         if (p.min_segno != NULL_SEGNO) {
 401 got_it:
 402                 if (p.alloc_mode == LFS) {
 403                         secno = GET_SEC_FROM_SEG(sbi, p.min_segno);
 404                         if (gc_type == FG_GC)
 405                                 sbi->cur_victim_sec = secno;
 406                         else
 407                                 set_bit(secno, dirty_i->victim_secmap);
 408                 }
 409                 *result = (p.min_segno / p.ofs_unit) * p.ofs_unit;
 410
 411                 trace_f2fs_get_victim(sbi->sb, type, gc_type, &p,
 412                                 sbi->cur_victim_sec,
 413                                 prefree_segments(sbi), free_segments(sbi));
 414         }
 415 out:
 416         mutex_unlock(&dirty_i->seglist_lock);
 417
 418         return (p.min_segno == NULL_SEGNO) ? 0 : 1;
 419 }
 420
 421 static const struct victim_selection default_v_ops = {
 422         .get_victim = get_victim_by_default,
 423 };
 424
 425 static struct inode *find_gc_inode(struct gc_inode_list *gc_list, nid_t ino)
 426 {
 427         struct inode_entry *ie;
 428
 429         ie = radix_tree_lookup(&gc_list->iroot, ino);
 430         if (ie)
 431                 return ie->inode;
 432         return NULL;
 433 }
 434
 435 static void add_gc_inode(struct gc_inode_list *gc_list, struct inode *inode)
 436 {
 437         struct inode_entry *new_ie;
 438
 439         if (inode == find_gc_inode(gc_list, inode->i_ino)) {
 440                 iput(inode);
 441                 return;
 442         }
 443         new_ie = f2fs_kmem_cache_alloc(inode_entry_slab, GFP_NOFS);
 444         new_ie->inode = inode;
 445
 446         f2fs_radix_tree_insert(&gc_list->iroot, inode->i_ino, new_ie);
 447         list_add_tail(&new_ie->list, &gc_list->ilist);
 448 }
 449
 450 static void put_gc_inode(struct gc_inode_list *gc_list)
 451 {
 452         struct inode_entry *ie, *next_ie;
 453         list_for_each_entry_safe(ie, next_ie, &gc_list->ilist, list) {
 454                 radix_tree_delete(&gc_list->iroot, ie->inode->i_ino);
 455                 iput(ie->inode);
 456                 list_del(&ie->list);
 457                 kmem_cache_free(inode_entry_slab, ie);
 458         }
 459 }
 460
 461 static int check_valid_map(struct f2fs_sb_info *sbi,
 462                                 unsigned int segno, int offset)
 463 {
 464         struct sit_info *sit_i = SIT_I(sbi);
 465         struct seg_entry *sentry;
 466         int ret;
 467
 468         down_read(&sit_i->sentry_lock);
 469         sentry = get_seg_entry(sbi, segno);
 470         ret = f2fs_test_bit(offset, sentry->cur_valid_map);
 471         up_read(&sit_i->sentry_lock);
 472         return ret;
 473 }
 474
 475 /*
 476  * This function compares node address got in summary with that in NAT.
 477  * On validity, copy that node with cold status, otherwise (invalid node)
 478  * ignore that.
 479  */
 480 static void gc_node_segment(struct f2fs_sb_info *sbi,
 481                 struct f2fs_summary *sum, unsigned int segno, int gc_type)
 482 {
 483         struct f2fs_summary *entry;
 484         block_t start_addr;
 485         int off;
 486         int phase = 0;
 487
 488         start_addr = START_BLOCK(sbi, segno);
 489
 490 next_step:
 491         entry = sum;
 492
 493         for (off = 0; off < sbi->blocks_per_seg; off++, entry++) {
 494                 nid_t nid = le32_to_cpu(entry->nid);
 495                 struct page *node_page;
 496                 struct node_info ni;
 497
 498                 /* stop BG_GC if there is not enough free sections. */
 499                 if (gc_type == BG_GC && has_not_enough_free_secs(sbi, 0, 0))
 500                         return;
 501
 502                 if (check_valid_map(sbi, segno, off) == 0)
 503                         continue;
 504
 505                 if (phase == 0) {
 506                         ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nid), 1,
 507                                                         META_NAT, true);
 508                         continue;
 509                 }
 510
 511                 if (phase == 1) {
 512                         ra_node_page(sbi, nid);
 513                         continue;
 514                 }
 515
 516                 /* phase == 2 */
 517                 node_page = get_node_page(sbi, nid);
 518                 if (IS_ERR(node_page))
 519                         continue;
 520
 521                 /* block may become invalid during get_node_page */
 522                 if (check_valid_map(sbi, segno, off) == 0) {
 523                         f2fs_put_page(node_page, 1);
 524                         continue;
 525                 }
 526
 527                 get_node_info(sbi, nid, &ni);
 528                 if (ni.blk_addr != start_addr + off) {
 529                         f2fs_put_page(node_page, 1);
 530                         continue;
 531                 }
 532
 533                 move_node_page(node_page, gc_type);
 534                 stat_inc_node_blk_count(sbi, 1, gc_type);
 535         }
 536
 537         if (++phase < 3)
 538                 goto next_step;
 539 }
 540
 541 /*
 542  * Calculate start block index indicating the given node offset.
 543  * Be careful, caller should give this node offset only indicating direct node
 544  * blocks. If any node offsets, which point the other types of node blocks such
 545  * as indirect or double indirect node blocks, are given, it must be a caller's
 546  * bug.
 547  */
 548 block_t start_bidx_of_node(unsigned int node_ofs, struct inode *inode)
 549 {
 550         unsigned int indirect_blks = 2 * NIDS_PER_BLOCK + 4;
 551         unsigned int bidx;
 552
 553         if (node_ofs == 0)
 554                 return 0;
 555
 556         if (node_ofs <= 2) {
 557                 bidx = node_ofs - 1;
 558         } else if (node_ofs <= indirect_blks) {
 559                 int dec = (node_ofs - 4) / (NIDS_PER_BLOCK + 1);
 560                 bidx = node_ofs - 2 - dec;
 561         } else {
 562                 int dec = (node_ofs - indirect_blks - 3) / (NIDS_PER_BLOCK + 1);
 563                 bidx = node_ofs - 5 - dec;
 564         }
 565         return bidx * ADDRS_PER_BLOCK + ADDRS_PER_INODE(inode);
 566 }
 567
 568 static bool is_alive(struct f2fs_sb_info *sbi, struct f2fs_summary *sum,
 569                 struct node_info *dni, block_t blkaddr, unsigned int *nofs)
 570 {
 571         struct page *node_page;
 572         nid_t nid;
 573         unsigned int ofs_in_node;
 574         block_t source_blkaddr;
 575
 576         nid = le32_to_cpu(sum->nid);
 577         ofs_in_node = le16_to_cpu(sum->ofs_in_node);
 578
 579         node_page = get_node_page(sbi, nid);
 580         if (IS_ERR(node_page))
 581                 return false;
 582
 583         get_node_info(sbi, nid, dni);
 584
 585         if (sum->version != dni->version) {
 586                 f2fs_msg(sbi->sb, KERN_WARNING,
 587                                 "%s: valid data with mismatched node version.",
 588                                 __func__);
 589                 set_sbi_flag(sbi, SBI_NEED_FSCK);
 590         }
 591
 592         *nofs = ofs_of_node(node_page);
 593         source_blkaddr = datablock_addr(NULL, node_page, ofs_in_node);
 594         f2fs_put_page(node_page, 1);
 595
 596         if (source_blkaddr != blkaddr)
 597                 return false;
 598         return true;
 599 }
 600
 601 /*
 602  * Move data block via META_MAPPING while keeping locked data page.
 603  * This can be used to move blocks, aka LBAs, directly on disk.
 604  */
 605 static void move_data_block(struct inode *inode, block_t bidx,
 606                                         unsigned int segno, int off)
 607 {
 608         struct f2fs_io_info fio = {
 609                 .sbi = F2FS_I_SB(inode),
 610                 .ino = inode->i_ino,
 611                 .type = DATA,
 612                 .temp = COLD,
 613                 .op = REQ_OP_READ,
 614                 .op_flags = 0,
 615                 .encrypted_page = NULL,
 616                 .in_list = false,
 617         };
 618         struct dnode_of_data dn;
 619         struct f2fs_summary sum;
 620         struct node_info ni;
 621         struct page *page;
 622         block_t newaddr;
 623         int err;
 624
 625         /* do not read out */
 626         page = f2fs_grab_cache_page(inode->i_mapping, bidx, false);
 627         if (!page)
 628                 return;
 629
 630         if (!check_valid_map(F2FS_I_SB(inode), segno, off))
 631                 goto out;
 632
 633         if (f2fs_is_atomic_file(inode))
 634                 goto out;
 635
 636         if (f2fs_is_pinned_file(inode)) {
 637                 f2fs_pin_file_control(inode, true);
 638                 goto out;
 639         }
 640
 641         set_new_dnode(&dn, inode, NULL, NULL, 0);
 642         err = get_dnode_of_data(&dn, bidx, LOOKUP_NODE);
 643         if (err)
 644                 goto out;
 645
 646         if (unlikely(dn.data_blkaddr == NULL_ADDR)) {
 647                 ClearPageUptodate(page);
 648                 goto put_out;
 649         }
 650
 651         /*
 652          * don't cache encrypted data into meta inode until previous dirty
 653          * data were writebacked to avoid racing between GC and flush.
 654          */
 655         f2fs_wait_on_page_writeback(page, DATA, true);
 656
 657         get_node_info(fio.sbi, dn.nid, &ni);
 658         set_summary(&sum, dn.nid, dn.ofs_in_node, ni.version);
 659
 660         /* read page */
 661         fio.page = page;
 662         fio.new_blkaddr = fio.old_blkaddr = dn.data_blkaddr;
 663
 664         allocate_data_block(fio.sbi, NULL, fio.old_blkaddr, &newaddr,
 665                                         &sum, CURSEG_COLD_DATA, NULL, false);
 666
 667         fio.encrypted_page = f2fs_pagecache_get_page(META_MAPPING(fio.sbi),
 668                                 newaddr, FGP_LOCK | FGP_CREAT, GFP_NOFS);
 669         if (!fio.encrypted_page) {
 670                 err = -ENOMEM;
 671                 goto recover_block;
 672         }
 673
 674         err = f2fs_submit_page_bio(&fio);
 675         if (err)
 676                 goto put_page_out;
 677
 678         /* write page */
 679         lock_page(fio.encrypted_page);
 680
 681         if (unlikely(fio.encrypted_page->mapping != META_MAPPING(fio.sbi))) {
 682                 err = -EIO;
 683                 goto put_page_out;
 684         }
 685         if (unlikely(!PageUptodate(fio.encrypted_page))) {
 686                 err = -EIO;
 687                 goto put_page_out;
 688         }
 689
 690         set_page_dirty(fio.encrypted_page);
 691         f2fs_wait_on_page_writeback(fio.encrypted_page, DATA, true);
 692         if (clear_page_dirty_for_io(fio.encrypted_page))
 693                 dec_page_count(fio.sbi, F2FS_DIRTY_META);
 694
 695         set_page_writeback(fio.encrypted_page);
 696
 697         /* allocate block address */
 698         f2fs_wait_on_page_writeback(dn.node_page, NODE, true);
 699
 700         fio.op = REQ_OP_WRITE;
 701         fio.op_flags = REQ_SYNC;
 702         fio.new_blkaddr = newaddr;
 703         err = f2fs_submit_page_write(&fio);
 704         if (err) {
 705                 if (PageWriteback(fio.encrypted_page))
 706                         end_page_writeback(fio.encrypted_page);
 707                 goto put_page_out;
 708         }
 709
 710         f2fs_update_iostat(fio.sbi, FS_GC_DATA_IO, F2FS_BLKSIZE);
 711
 712         f2fs_update_data_blkaddr(&dn, newaddr);
 713         set_inode_flag(inode, FI_APPEND_WRITE);
 714         if (page->index == 0)
 715                 set_inode_flag(inode, FI_FIRST_BLOCK_WRITTEN);
 716 put_page_out:
 717         f2fs_put_page(fio.encrypted_page, 1);
 718 recover_block:
 719         if (err)
 720                 __f2fs_replace_block(fio.sbi, &sum, newaddr, fio.old_blkaddr,
 721                                                                 true, true);
 722 put_out:
 723         f2fs_put_dnode(&dn);
 724 out:
 725         f2fs_put_page(page, 1);
 726 }
 727
 728 static void move_data_page(struct inode *inode, block_t bidx, int gc_type,
 729                                                         unsigned int segno, int off)
 730 {
 731         struct page *page;
 732
 733         page = get_lock_data_page(inode, bidx, true);
 734         if (IS_ERR(page))
 735                 return;
 736
 737         if (!check_valid_map(F2FS_I_SB(inode), segno, off))
 738                 goto out;
 739
 740         if (f2fs_is_atomic_file(inode))
 741                 goto out;
 742         if (f2fs_is_pinned_file(inode)) {
 743                 if (gc_type == FG_GC)
 744                         f2fs_pin_file_control(inode, true);
 745                 goto out;
 746         }
 747
 748         if (gc_type == BG_GC) {
 749                 if (PageWriteback(page))
 750                         goto out;
 751                 set_page_dirty(page);
 752                 set_cold_data(page);
 753         } else {
 754                 struct f2fs_io_info fio = {
 755                         .sbi = F2FS_I_SB(inode),
 756                         .ino = inode->i_ino,
 757                         .type = DATA,
 758                         .temp = COLD,
 759                         .op = REQ_OP_WRITE,
 760                         .op_flags = REQ_SYNC,
 761                         .old_blkaddr = NULL_ADDR,
 762                         .page = page,
 763                         .encrypted_page = NULL,
 764                         .need_lock = LOCK_REQ,
 765                         .io_type = FS_GC_DATA_IO,
 766                 };
 767                 bool is_dirty = PageDirty(page);
 768                 int err;
 769
 770 retry:
 771                 set_page_dirty(page);
 772                 f2fs_wait_on_page_writeback(page, DATA, true);
 773                 if (clear_page_dirty_for_io(page)) {
 774                         inode_dec_dirty_pages(inode);
 775                         remove_dirty_inode(inode);
 776                 }
 777
 778                 set_cold_data(page);
 779
 780                 err = do_write_data_page(&fio);
 781                 if (err == -ENOMEM && is_dirty) {
 782                         congestion_wait(BLK_RW_ASYNC, HZ/50);
 783                         goto retry;
 784                 }
 785         }
 786 out:
 787         f2fs_put_page(page, 1);
 788 }
 789
 790 /*
 791  * This function tries to get parent node of victim data block, and identifies
 792  * data block validity. If the block is valid, copy that with cold status and
 793  * modify parent node.
 794  * If the parent node is not valid or the data block address is different,
 795  * the victim data block is ignored.
 796  */
 797 static void gc_data_segment(struct f2fs_sb_info *sbi, struct f2fs_summary *sum,
 798                 struct gc_inode_list *gc_list, unsigned int segno, int gc_type)
 799 {
 800         struct super_block *sb = sbi->sb;
 801         struct f2fs_summary *entry;
 802         block_t start_addr;
 803         int off;
 804         int phase = 0;
 805
 806         start_addr = START_BLOCK(sbi, segno);
 807
 808 next_step:
 809         entry = sum;
 810
 811         for (off = 0; off < sbi->blocks_per_seg; off++, entry++) {
 812                 struct page *data_page;
 813                 struct inode *inode;
 814                 struct node_info dni; /* dnode info for the data */
 815                 unsigned int ofs_in_node, nofs;
 816                 block_t start_bidx;
 817                 nid_t nid = le32_to_cpu(entry->nid);
 818
 819                 /* stop BG_GC if there is not enough free sections. */
 820                 if (gc_type == BG_GC && has_not_enough_free_secs(sbi, 0, 0))
 821                         return;
 822
 823                 if (check_valid_map(sbi, segno, off) == 0)
 824                         continue;
 825
 826                 if (phase == 0) {
 827                         ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nid), 1,
 828                                                         META_NAT, true);
 829                         continue;
 830                 }
 831
 832                 if (phase == 1) {
 833                         ra_node_page(sbi, nid);
 834                         continue;
 835                 }
 836
 837                 /* Get an inode by ino with checking validity */
 838                 if (!is_alive(sbi, entry, &dni, start_addr + off, &nofs))
 839                         continue;
 840
 841                 if (phase == 2) {
 842                         ra_node_page(sbi, dni.ino);
 843                         continue;
 844                 }
 845
 846                 ofs_in_node = le16_to_cpu(entry->ofs_in_node);
 847
 848                 if (phase == 3) {
 849                         inode = f2fs_iget(sb, dni.ino);
 850                         if (IS_ERR(inode) || is_bad_inode(inode))
 851                                 continue;
 852
 853                         /* if inode uses special I/O path, let's go phase 3 */
 854                         if (f2fs_post_read_required(inode)) {
 855                                 add_gc_inode(gc_list, inode);
 856                                 continue;
 857                         }
 858
 859                         if (!down_write_trylock(
 860                                 &F2FS_I(inode)->dio_rwsem[WRITE])) {
 861                                 iput(inode);
 862                                 continue;
 863                         }
 864
 865                         start_bidx = start_bidx_of_node(nofs, inode);
 866                         data_page = get_read_data_page(inode,
 867                                         start_bidx + ofs_in_node, REQ_RAHEAD,
 868                                         true);
 869                         up_write(&F2FS_I(inode)->dio_rwsem[WRITE]);
 870                         if (IS_ERR(data_page)) {
 871                                 iput(inode);
 872                                 continue;
 873                         }
 874
 875                         f2fs_put_page(data_page, 0);
 876                         add_gc_inode(gc_list, inode);
 877                         continue;
 878                 }
 879
 880                 /* phase 4 */
 881                 inode = find_gc_inode(gc_list, dni.ino);
 882                 if (inode) {
 883                         struct f2fs_inode_info *fi = F2FS_I(inode);
 884                         bool locked = false;
 885
 886                         if (S_ISREG(inode->i_mode)) {
 887                                 if (!down_write_trylock(&fi->dio_rwsem[READ]))
 888                                         continue;
 889                                 if (!down_write_trylock(
 890                                                 &fi->dio_rwsem[WRITE])) {
 891                                         up_write(&fi->dio_rwsem[READ]);
 892                                         continue;
 893                                 }
 894                                 locked = true;
 895
 896                                 /* wait for all inflight aio data */
 897                                 inode_dio_wait(inode);
 898                         }
 899
 900                         start_bidx = start_bidx_of_node(nofs, inode)
 901                                                                 + ofs_in_node;
 902                         if (f2fs_post_read_required(inode))
 903                                 move_data_block(inode, start_bidx, segno, off);
 904                         else
 905                                 move_data_page(inode, start_bidx, gc_type,
 906                                                                 segno, off);
 907
 908                         if (locked) {
 909                                 up_write(&fi->dio_rwsem[WRITE]);
 910                                 up_write(&fi->dio_rwsem[READ]);
 911                         }
 912
 913                         stat_inc_data_blk_count(sbi, 1, gc_type);
 914                 }
 915         }
 916
 917         if (++phase < 5)
 918                 goto next_step;
 919 }
 920
 921 static int __get_victim(struct f2fs_sb_info *sbi, unsigned int *victim,
 922                         int gc_type)
 923 {
 924         struct sit_info *sit_i = SIT_I(sbi);
 925         int ret;
 926
 927         down_write(&sit_i->sentry_lock);
 928         ret = DIRTY_I(sbi)->v_ops->get_victim(sbi, victim, gc_type,
 929                                               NO_CHECK_TYPE, LFS);
 930         up_write(&sit_i->sentry_lock);
 931         return ret;
 932 }
 933
 934 static int do_garbage_collect(struct f2fs_sb_info *sbi,
 935                                 unsigned int start_segno,
 936                                 struct gc_inode_list *gc_list, int gc_type)
 937 {
 938         struct page *sum_page;
 939         struct f2fs_summary_block *sum;
 940         struct blk_plug plug;
 941         unsigned int segno = start_segno;
 942         unsigned int end_segno = start_segno + sbi->segs_per_sec;
 943         int seg_freed = 0;
 944         unsigned char type = IS_DATASEG(get_seg_entry(sbi, segno)->type) ?
 945                                                 SUM_TYPE_DATA : SUM_TYPE_NODE;
 946
 947         /* readahead multi ssa blocks those have contiguous address */
 948         if (sbi->segs_per_sec > 1)
 949                 ra_meta_pages(sbi, GET_SUM_BLOCK(sbi, segno),
 950                                         sbi->segs_per_sec, META_SSA, true);
 951
 952         /* reference all summary page */
 953         while (segno < end_segno) {
 954                 sum_page = get_sum_page(sbi, segno++);
 955                 unlock_page(sum_page);
 956         }
 957
 958         blk_start_plug(&plug);
 959
 960         for (segno = start_segno; segno < end_segno; segno++) {
 961
 962                 /* find segment summary of victim */
 963                 sum_page = find_get_page(META_MAPPING(sbi),
 964                                         GET_SUM_BLOCK(sbi, segno));
 965                 f2fs_put_page(sum_page, 0);
 966
 967                 if (get_valid_blocks(sbi, segno, false) == 0 ||
 968                                 !PageUptodate(sum_page) ||
 969                                 unlikely(f2fs_cp_error(sbi)))
 970                         goto next;
 971
 972                 sum = page_address(sum_page);
 973                 f2fs_bug_on(sbi, type != GET_SUM_TYPE((&sum->footer)));
 974
 975                 /*
 976                  * this is to avoid deadlock:
 977                  * - lock_page(sum_page)         - f2fs_replace_block
 978                  *  - check_valid_map()            - down_write(sentry_lock)
 979                  *   - down_read(sentry_lock)     - change_curseg()
 980                  *                                  - lock_page(sum_page)
 981                  */
 982                 if (type == SUM_TYPE_NODE)
 983                         gc_node_segment(sbi, sum->entries, segno, gc_type);
 984                 else
 985                         gc_data_segment(sbi, sum->entries, gc_list, segno,
 986                                                                 gc_type);
 987
 988                 stat_inc_seg_count(sbi, type, gc_type);
 989
 990                 if (gc_type == FG_GC &&
 991                                 get_valid_blocks(sbi, segno, false) == 0)
 992                         seg_freed++;
 993 next:
 994                 f2fs_put_page(sum_page, 0);
 995         }
 996
 997         if (gc_type == FG_GC)
 998                 f2fs_submit_merged_write(sbi,
 999                                 (type == SUM_TYPE_NODE) ? NODE : DATA);
1000
1001         blk_finish_plug(&plug);
1002
1003         stat_inc_call_count(sbi->stat_info);
1004
1005         return seg_freed;
1006 }
1007
1008 int f2fs_gc(struct f2fs_sb_info *sbi, bool sync,
1009                         bool background, unsigned int segno)
1010 {
1011         int gc_type = sync ? FG_GC : BG_GC;
1012         int sec_freed = 0, seg_freed = 0, total_freed = 0;
1013         int ret = 0;
1014         struct cp_control cpc;
1015         unsigned int init_segno = segno;
1016         struct gc_inode_list gc_list = {
1017                 .ilist = LIST_HEAD_INIT(gc_list.ilist),
1018                 .iroot = RADIX_TREE_INIT(GFP_NOFS),
1019         };
1020
1021         trace_f2fs_gc_begin(sbi->sb, sync, background,
1022                                 get_pages(sbi, F2FS_DIRTY_NODES),
1023                                 get_pages(sbi, F2FS_DIRTY_DENTS),
1024                                 get_pages(sbi, F2FS_DIRTY_IMETA),
1025                                 free_sections(sbi),
1026                                 free_segments(sbi),
1027                                 reserved_segments(sbi),
1028                                 prefree_segments(sbi));
1029
1030         cpc.reason = __get_cp_reason(sbi);
1031 gc_more:
1032         if (unlikely(!(sbi->sb->s_flags & MS_ACTIVE))) {
1033                 ret = -EINVAL;
1034                 goto stop;
1035         }
1036         if (unlikely(f2fs_cp_error(sbi))) {
1037                 ret = -EIO;
1038                 goto stop;
1039         }
1040
1041         if (gc_type == BG_GC && has_not_enough_free_secs(sbi, 0, 0)) {
1042                 /*
1043                  * For example, if there are many prefree_segments below given
1044                  * threshold, we can make them free by checkpoint. Then, we
1045                  * secure free segments which doesn't need fggc any more.
1046                  */
1047                 if (prefree_segments(sbi)) {
1048                         ret = write_checkpoint(sbi, &cpc);
1049                         if (ret)
1050                                 goto stop;
1051                 }
1052                 if (has_not_enough_free_secs(sbi, 0, 0))
1053                         gc_type = FG_GC;
1054         }
1055
1056         /* f2fs_balance_fs doesn't need to do BG_GC in critical path. */
1057         if (gc_type == BG_GC && !background) {
1058                 ret = -EINVAL;
1059                 goto stop;
1060         }
1061         if (!__get_victim(sbi, &segno, gc_type)) {
1062                 ret = -ENODATA;
1063                 goto stop;
1064         }
1065
1066         seg_freed = do_garbage_collect(sbi, segno, &gc_list, gc_type);
1067         if (gc_type == FG_GC && seg_freed == sbi->segs_per_sec)
1068                 sec_freed++;
1069         total_freed += seg_freed;
1070
1071         if (gc_type == FG_GC)
1072                 sbi->cur_victim_sec = NULL_SEGNO;
1073
1074         if (!sync) {
1075                 if (has_not_enough_free_secs(sbi, sec_freed, 0)) {
1076                         segno = NULL_SEGNO;
1077                         goto gc_more;
1078                 }
1079
1080                 if (gc_type == FG_GC)
1081                         ret = write_checkpoint(sbi, &cpc);
1082         }
1083 stop:
1084         SIT_I(sbi)->last_victim[ALLOC_NEXT] = 0;
1085         SIT_I(sbi)->last_victim[FLUSH_DEVICE] = init_segno;
1086
1087         trace_f2fs_gc_end(sbi->sb, ret, total_freed, sec_freed,
1088                                 get_pages(sbi, F2FS_DIRTY_NODES),
1089                                 get_pages(sbi, F2FS_DIRTY_DENTS),
1090                                 get_pages(sbi, F2FS_DIRTY_IMETA),
1091                                 free_sections(sbi),
1092                                 free_segments(sbi),
1093                                 reserved_segments(sbi),
1094                                 prefree_segments(sbi));
1095
1096         mutex_unlock(&sbi->gc_mutex);
1097
1098         put_gc_inode(&gc_list);
1099
1100         if (sync)
1101                 ret = sec_freed ? 0 : -EAGAIN;
1102         return ret;
1103 }
1104
1105 void build_gc_manager(struct f2fs_sb_info *sbi)
1106 {
1107         u64 main_count, resv_count, ovp_count;
1108
1109         DIRTY_I(sbi)->v_ops = &default_v_ops;
1110
1111         /* threshold of # of valid blocks in a section for victims of FG_GC */
1112         main_count = SM_I(sbi)->main_segments << sbi->log_blocks_per_seg;
1113         resv_count = SM_I(sbi)->reserved_segments << sbi->log_blocks_per_seg;
1114         ovp_count = SM_I(sbi)->ovp_segments << sbi->log_blocks_per_seg;
1115
1116         sbi->fggc_threshold = div64_u64((main_count - ovp_count) *
1117                                 BLKS_PER_SEC(sbi), (main_count - resv_count));
1118         sbi->gc_pin_file_threshold = DEF_GC_FAILED_PINNED_FILES;
1119
1120         /* give warm/cold data area from slower device */
1121         if (sbi->s_ndevs && sbi->segs_per_sec == 1)
1122                 SIT_I(sbi)->last_victim[ALLOC_NEXT] =
1123                                 GET_SEGNO(sbi, FDEV(0).end_blk) + 1;
1124 }