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         long wait_ms;
  32
  33         wait_ms = gc_th->min_sleep_time;
  34
  35         do {
  36                 if (try_to_freeze())
  37                         continue;
  38                 else
  39                         wait_event_interruptible_timeout(*wq,
  40                                                 kthread_should_stop(),
  41                                                 msecs_to_jiffies(wait_ms));
  42                 if (kthread_should_stop())
  43                         break;
  44
  45                 if (sbi->sb->s_writers.frozen >= SB_FREEZE_WRITE) {
  46                         increase_sleep_time(gc_th, &wait_ms);
  47                         continue;
  48                 }
  49
  50 #ifdef CONFIG_F2FS_FAULT_INJECTION
  51                 if (time_to_inject(sbi, FAULT_CHECKPOINT))
  52                         f2fs_stop_checkpoint(sbi, false);
  53 #endif
  54
  55                 /*
  56                  * [GC triggering condition]
  57                  * 0. GC is not conducted currently.
  58                  * 1. There are enough dirty segments.
  59                  * 2. IO subsystem is idle by checking the # of writeback pages.
  60                  * 3. IO subsystem is idle by checking the # of requests in
  61                  *    bdev's request list.
  62                  *
  63                  * Note) We have to avoid triggering GCs frequently.
  64                  * Because it is possible that some segments can be
  65                  * invalidated soon after by user update or deletion.
  66                  * So, I'd like to wait some time to collect dirty segments.
  67                  */
  68                 if (!mutex_trylock(&sbi->gc_mutex))
  69                         continue;
  70
  71                 if (!is_idle(sbi)) {
  72                         increase_sleep_time(gc_th, &wait_ms);
  73                         mutex_unlock(&sbi->gc_mutex);
  74                         continue;
  75                 }
  76
  77                 if (has_enough_invalid_blocks(sbi))
  78                         decrease_sleep_time(gc_th, &wait_ms);
  79                 else
  80                         increase_sleep_time(gc_th, &wait_ms);
  81
  82                 stat_inc_bggc_count(sbi);
  83
  84                 /* if return value is not zero, no victim was selected */
  85                 if (f2fs_gc(sbi, test_opt(sbi, FORCE_FG_GC), true))
  86                         wait_ms = gc_th->no_gc_sleep_time;
  87
  88                 trace_f2fs_background_gc(sbi->sb, wait_ms,
  89                                 prefree_segments(sbi), free_segments(sbi));
  90
  91                 /* balancing f2fs's metadata periodically */
  92                 f2fs_balance_fs_bg(sbi);
  93
  94         } while (!kthread_should_stop());
  95         return 0;
  96 }
  97
  98 int start_gc_thread(struct f2fs_sb_info *sbi)
  99 {
 100         struct f2fs_gc_kthread *gc_th;
 101         dev_t dev = sbi->sb->s_bdev->bd_dev;
 102         int err = 0;
 103
 104         gc_th = f2fs_kmalloc(sbi, sizeof(struct f2fs_gc_kthread), GFP_KERNEL);
 105         if (!gc_th) {
 106                 err = -ENOMEM;
 107                 goto out;
 108         }
 109
 110         gc_th->min_sleep_time = DEF_GC_THREAD_MIN_SLEEP_TIME;
 111         gc_th->max_sleep_time = DEF_GC_THREAD_MAX_SLEEP_TIME;
 112         gc_th->no_gc_sleep_time = DEF_GC_THREAD_NOGC_SLEEP_TIME;
 113
 114         gc_th->gc_idle = 0;
 115
 116         sbi->gc_thread = gc_th;
 117         init_waitqueue_head(&sbi->gc_thread->gc_wait_queue_head);
 118         sbi->gc_thread->f2fs_gc_task = kthread_run(gc_thread_func, sbi,
 119                         "f2fs_gc-%u:%u", MAJOR(dev), MINOR(dev));
 120         if (IS_ERR(gc_th->f2fs_gc_task)) {
 121                 err = PTR_ERR(gc_th->f2fs_gc_task);
 122                 kfree(gc_th);
 123                 sbi->gc_thread = NULL;
 124         }
 125 out:
 126         return err;
 127 }
 128
 129 void stop_gc_thread(struct f2fs_sb_info *sbi)
 130 {
 131         struct f2fs_gc_kthread *gc_th = sbi->gc_thread;
 132         if (!gc_th)
 133                 return;
 134         kthread_stop(gc_th->f2fs_gc_task);
 135         kfree(gc_th);
 136         sbi->gc_thread = NULL;
 137 }
 138
 139 static int select_gc_type(struct f2fs_gc_kthread *gc_th, int gc_type)
 140 {
 141         int gc_mode = (gc_type == BG_GC) ? GC_CB : GC_GREEDY;
 142
 143         if (gc_th && gc_th->gc_idle) {
 144                 if (gc_th->gc_idle == 1)
 145                         gc_mode = GC_CB;
 146                 else if (gc_th->gc_idle == 2)
 147                         gc_mode = GC_GREEDY;
 148         }
 149         return gc_mode;
 150 }
 151
 152 static void select_policy(struct f2fs_sb_info *sbi, int gc_type,
 153                         int type, struct victim_sel_policy *p)
 154 {
 155         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
 156
 157         if (p->alloc_mode == SSR) {
 158                 p->gc_mode = GC_GREEDY;
 159                 p->dirty_segmap = dirty_i->dirty_segmap[type];
 160                 p->max_search = dirty_i->nr_dirty[type];
 161                 p->ofs_unit = 1;
 162         } else {
 163                 p->gc_mode = select_gc_type(sbi->gc_thread, gc_type);
 164                 p->dirty_segmap = dirty_i->dirty_segmap[DIRTY];
 165                 p->max_search = dirty_i->nr_dirty[DIRTY];
 166                 p->ofs_unit = sbi->segs_per_sec;
 167         }
 168
 169         if (p->max_search > sbi->max_victim_search)
 170                 p->max_search = sbi->max_victim_search;
 171
 172         p->offset = sbi->last_victim[p->gc_mode];
 173 }
 174
 175 static unsigned int get_max_cost(struct f2fs_sb_info *sbi,
 176                                 struct victim_sel_policy *p)
 177 {
 178         /* SSR allocates in a segment unit */
 179         if (p->alloc_mode == SSR)
 180                 return sbi->blocks_per_seg;
 181         if (p->gc_mode == GC_GREEDY)
 182                 return sbi->blocks_per_seg * p->ofs_unit;
 183         else if (p->gc_mode == GC_CB)
 184                 return UINT_MAX;
 185         else /* No other gc_mode */
 186                 return 0;
 187 }
 188
 189 static unsigned int check_bg_victims(struct f2fs_sb_info *sbi)
 190 {
 191         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
 192         unsigned int secno;
 193
 194         /*
 195          * If the gc_type is FG_GC, we can select victim segments
 196          * selected by background GC before.
 197          * Those segments guarantee they have small valid blocks.
 198          */
 199         for_each_set_bit(secno, dirty_i->victim_secmap, MAIN_SECS(sbi)) {
 200                 if (sec_usage_check(sbi, secno))
 201                         continue;
 202                 clear_bit(secno, dirty_i->victim_secmap);
 203                 return secno * sbi->segs_per_sec;
 204         }
 205         return NULL_SEGNO;
 206 }
 207
 208 static unsigned int get_cb_cost(struct f2fs_sb_info *sbi, unsigned int segno)
 209 {
 210         struct sit_info *sit_i = SIT_I(sbi);
 211         unsigned int secno = GET_SECNO(sbi, segno);
 212         unsigned int start = secno * sbi->segs_per_sec;
 213         unsigned long long mtime = 0;
 214         unsigned int vblocks;
 215         unsigned char age = 0;
 216         unsigned char u;
 217         unsigned int i;
 218
 219         for (i = 0; i < sbi->segs_per_sec; i++)
 220                 mtime += get_seg_entry(sbi, start + i)->mtime;
 221         vblocks = get_valid_blocks(sbi, segno, sbi->segs_per_sec);
 222
 223         mtime = div_u64(mtime, sbi->segs_per_sec);
 224         vblocks = div_u64(vblocks, sbi->segs_per_sec);
 225
 226         u = (vblocks * 100) >> sbi->log_blocks_per_seg;
 227
 228         /* Handle if the system time has changed by the user */
 229         if (mtime < sit_i->min_mtime)
 230                 sit_i->min_mtime = mtime;
 231         if (mtime > sit_i->max_mtime)
 232                 sit_i->max_mtime = mtime;
 233         if (sit_i->max_mtime != sit_i->min_mtime)
 234                 age = 100 - div64_u64(100 * (mtime - sit_i->min_mtime),
 235                                 sit_i->max_mtime - sit_i->min_mtime);
 236
 237         return UINT_MAX - ((100 * (100 - u) * age) / (100 + u));
 238 }
 239
 240 static inline unsigned int get_gc_cost(struct f2fs_sb_info *sbi,
 241                         unsigned int segno, struct victim_sel_policy *p)
 242 {
 243         if (p->alloc_mode == SSR)
 244                 return get_seg_entry(sbi, segno)->ckpt_valid_blocks;
 245
 246         /* alloc_mode == LFS */
 247         if (p->gc_mode == GC_GREEDY)
 248                 return get_valid_blocks(sbi, segno, sbi->segs_per_sec);
 249         else
 250                 return get_cb_cost(sbi, segno);
 251 }
 252
 253 static unsigned int count_bits(const unsigned long *addr,
 254                                 unsigned int offset, unsigned int len)
 255 {
 256         unsigned int end = offset + len, sum = 0;
 257
 258         while (offset < end) {
 259                 if (test_bit(offset++, addr))
 260                         ++sum;
 261         }
 262         return sum;
 263 }
 264
 265 /*
 266  * This function is called from two paths.
 267  * One is garbage collection and the other is SSR segment selection.
 268  * When it is called during GC, it just gets a victim segment
 269  * and it does not remove it from dirty seglist.
 270  * When it is called from SSR segment selection, it finds a segment
 271  * which has minimum valid blocks and removes it from dirty seglist.
 272  */
 273 static int get_victim_by_default(struct f2fs_sb_info *sbi,
 274                 unsigned int *result, int gc_type, int type, char alloc_mode)
 275 {
 276         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
 277         struct victim_sel_policy p;
 278         unsigned int secno, last_victim;
 279         unsigned int last_segment = MAIN_SEGS(sbi);
 280         unsigned int nsearched = 0;
 281
 282         mutex_lock(&dirty_i->seglist_lock);
 283
 284         p.alloc_mode = alloc_mode;
 285         select_policy(sbi, gc_type, type, &p);
 286
 287         p.min_segno = NULL_SEGNO;
 288         p.min_cost = get_max_cost(sbi, &p);
 289
 290         if (p.max_search == 0)
 291                 goto out;
 292
 293         last_victim = sbi->last_victim[p.gc_mode];
 294         if (p.alloc_mode == LFS && gc_type == FG_GC) {
 295                 p.min_segno = check_bg_victims(sbi);
 296                 if (p.min_segno != NULL_SEGNO)
 297                         goto got_it;
 298         }
 299
 300         while (1) {
 301                 unsigned long cost;
 302                 unsigned int segno;
 303
 304                 segno = find_next_bit(p.dirty_segmap, last_segment, p.offset);
 305                 if (segno >= last_segment) {
 306                         if (sbi->last_victim[p.gc_mode]) {
 307                                 last_segment = sbi->last_victim[p.gc_mode];
 308                                 sbi->last_victim[p.gc_mode] = 0;
 309                                 p.offset = 0;
 310                                 continue;
 311                         }
 312                         break;
 313                 }
 314
 315                 p.offset = segno + p.ofs_unit;
 316                 if (p.ofs_unit > 1) {
 317                         p.offset -= segno % p.ofs_unit;
 318                         nsearched += count_bits(p.dirty_segmap,
 319                                                 p.offset - p.ofs_unit,
 320                                                 p.ofs_unit);
 321                 } else {
 322                         nsearched++;
 323                 }
 324
 325
 326                 secno = GET_SECNO(sbi, segno);
 327
 328                 if (sec_usage_check(sbi, secno))
 329                         goto next;
 330                 if (gc_type == BG_GC && test_bit(secno, dirty_i->victim_secmap))
 331                         goto next;
 332
 333                 cost = get_gc_cost(sbi, segno, &p);
 334
 335                 if (p.min_cost > cost) {
 336                         p.min_segno = segno;
 337                         p.min_cost = cost;
 338                 }
 339 next:
 340                 if (nsearched >= p.max_search) {
 341                         if (!sbi->last_victim[p.gc_mode] && segno <= last_victim)
 342                                 sbi->last_victim[p.gc_mode] = last_victim + 1;
 343                         else
 344                                 sbi->last_victim[p.gc_mode] = segno + 1;
 345                         break;
 346                 }
 347         }
 348         if (p.min_segno != NULL_SEGNO) {
 349 got_it:
 350                 if (p.alloc_mode == LFS) {
 351                         secno = GET_SECNO(sbi, p.min_segno);
 352                         if (gc_type == FG_GC)
 353                                 sbi->cur_victim_sec = secno;
 354                         else
 355                                 set_bit(secno, dirty_i->victim_secmap);
 356                 }
 357                 *result = (p.min_segno / p.ofs_unit) * p.ofs_unit;
 358
 359                 trace_f2fs_get_victim(sbi->sb, type, gc_type, &p,
 360                                 sbi->cur_victim_sec,
 361                                 prefree_segments(sbi), free_segments(sbi));
 362         }
 363 out:
 364         mutex_unlock(&dirty_i->seglist_lock);
 365
 366         return (p.min_segno == NULL_SEGNO) ? 0 : 1;
 367 }
 368
 369 static const struct victim_selection default_v_ops = {
 370         .get_victim = get_victim_by_default,
 371 };
 372
 373 static struct inode *find_gc_inode(struct gc_inode_list *gc_list, nid_t ino)
 374 {
 375         struct inode_entry *ie;
 376
 377         ie = radix_tree_lookup(&gc_list->iroot, ino);
 378         if (ie)
 379                 return ie->inode;
 380         return NULL;
 381 }
 382
 383 static void add_gc_inode(struct gc_inode_list *gc_list, struct inode *inode)
 384 {
 385         struct inode_entry *new_ie;
 386
 387         if (inode == find_gc_inode(gc_list, inode->i_ino)) {
 388                 iput(inode);
 389                 return;
 390         }
 391         new_ie = f2fs_kmem_cache_alloc(inode_entry_slab, GFP_NOFS);
 392         new_ie->inode = inode;
 393
 394         f2fs_radix_tree_insert(&gc_list->iroot, inode->i_ino, new_ie);
 395         list_add_tail(&new_ie->list, &gc_list->ilist);
 396 }
 397
 398 static void put_gc_inode(struct gc_inode_list *gc_list)
 399 {
 400         struct inode_entry *ie, *next_ie;
 401         list_for_each_entry_safe(ie, next_ie, &gc_list->ilist, list) {
 402                 radix_tree_delete(&gc_list->iroot, ie->inode->i_ino);
 403                 iput(ie->inode);
 404                 list_del(&ie->list);
 405                 kmem_cache_free(inode_entry_slab, ie);
 406         }
 407 }
 408
 409 static int check_valid_map(struct f2fs_sb_info *sbi,
 410                                 unsigned int segno, int offset)
 411 {
 412         struct sit_info *sit_i = SIT_I(sbi);
 413         struct seg_entry *sentry;
 414         int ret;
 415
 416         mutex_lock(&sit_i->sentry_lock);
 417         sentry = get_seg_entry(sbi, segno);
 418         ret = f2fs_test_bit(offset, sentry->cur_valid_map);
 419         mutex_unlock(&sit_i->sentry_lock);
 420         return ret;
 421 }
 422
 423 /*
 424  * This function compares node address got in summary with that in NAT.
 425  * On validity, copy that node with cold status, otherwise (invalid node)
 426  * ignore that.
 427  */
 428 static void gc_node_segment(struct f2fs_sb_info *sbi,
 429                 struct f2fs_summary *sum, unsigned int segno, int gc_type)
 430 {
 431         struct f2fs_summary *entry;
 432         block_t start_addr;
 433         int off;
 434         int phase = 0;
 435
 436         start_addr = START_BLOCK(sbi, segno);
 437
 438 next_step:
 439         entry = sum;
 440
 441         for (off = 0; off < sbi->blocks_per_seg; off++, entry++) {
 442                 nid_t nid = le32_to_cpu(entry->nid);
 443                 struct page *node_page;
 444                 struct node_info ni;
 445
 446                 /* stop BG_GC if there is not enough free sections. */
 447                 if (gc_type == BG_GC && has_not_enough_free_secs(sbi, 0, 0))
 448                         return;
 449
 450                 if (check_valid_map(sbi, segno, off) == 0)
 451                         continue;
 452
 453                 if (phase == 0) {
 454                         ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nid), 1,
 455                                                         META_NAT, true);
 456                         continue;
 457                 }
 458
 459                 if (phase == 1) {
 460                         ra_node_page(sbi, nid);
 461                         continue;
 462                 }
 463
 464                 /* phase == 2 */
 465                 node_page = get_node_page(sbi, nid);
 466                 if (IS_ERR(node_page))
 467                         continue;
 468
 469                 /* block may become invalid during get_node_page */
 470                 if (check_valid_map(sbi, segno, off) == 0) {
 471                         f2fs_put_page(node_page, 1);
 472                         continue;
 473                 }
 474
 475                 get_node_info(sbi, nid, &ni);
 476                 if (ni.blk_addr != start_addr + off) {
 477                         f2fs_put_page(node_page, 1);
 478                         continue;
 479                 }
 480
 481                 move_node_page(node_page, gc_type);
 482                 stat_inc_node_blk_count(sbi, 1, gc_type);
 483         }
 484
 485         if (++phase < 3)
 486                 goto next_step;
 487 }
 488
 489 /*
 490  * Calculate start block index indicating the given node offset.
 491  * Be careful, caller should give this node offset only indicating direct node
 492  * blocks. If any node offsets, which point the other types of node blocks such
 493  * as indirect or double indirect node blocks, are given, it must be a caller's
 494  * bug.
 495  */
 496 block_t start_bidx_of_node(unsigned int node_ofs, struct inode *inode)
 497 {
 498         unsigned int indirect_blks = 2 * NIDS_PER_BLOCK + 4;
 499         unsigned int bidx;
 500
 501         if (node_ofs == 0)
 502                 return 0;
 503
 504         if (node_ofs <= 2) {
 505                 bidx = node_ofs - 1;
 506         } else if (node_ofs <= indirect_blks) {
 507                 int dec = (node_ofs - 4) / (NIDS_PER_BLOCK + 1);
 508                 bidx = node_ofs - 2 - dec;
 509         } else {
 510                 int dec = (node_ofs - indirect_blks - 3) / (NIDS_PER_BLOCK + 1);
 511                 bidx = node_ofs - 5 - dec;
 512         }
 513         return bidx * ADDRS_PER_BLOCK + ADDRS_PER_INODE(inode);
 514 }
 515
 516 static bool is_alive(struct f2fs_sb_info *sbi, struct f2fs_summary *sum,
 517                 struct node_info *dni, block_t blkaddr, unsigned int *nofs)
 518 {
 519         struct page *node_page;
 520         nid_t nid;
 521         unsigned int ofs_in_node;
 522         block_t source_blkaddr;
 523
 524         nid = le32_to_cpu(sum->nid);
 525         ofs_in_node = le16_to_cpu(sum->ofs_in_node);
 526
 527         node_page = get_node_page(sbi, nid);
 528         if (IS_ERR(node_page))
 529                 return false;
 530
 531         get_node_info(sbi, nid, dni);
 532
 533         if (sum->version != dni->version) {
 534                 f2fs_put_page(node_page, 1);
 535                 return false;
 536         }
 537
 538         *nofs = ofs_of_node(node_page);
 539         source_blkaddr = datablock_addr(node_page, ofs_in_node);
 540         f2fs_put_page(node_page, 1);
 541
 542         if (source_blkaddr != blkaddr)
 543                 return false;
 544         return true;
 545 }
 546
 547 static void move_encrypted_block(struct inode *inode, block_t bidx,
 548                                                         unsigned int segno, int off)
 549 {
 550         struct f2fs_io_info fio = {
 551                 .sbi = F2FS_I_SB(inode),
 552                 .type = DATA,
 553                 .op = REQ_OP_READ,
 554                 .op_flags = 0,
 555                 .encrypted_page = NULL,
 556         };
 557         struct dnode_of_data dn;
 558         struct f2fs_summary sum;
 559         struct node_info ni;
 560         struct page *page;
 561         block_t newaddr;
 562         int err;
 563
 564         /* do not read out */
 565         page = f2fs_grab_cache_page(inode->i_mapping, bidx, false);
 566         if (!page)
 567                 return;
 568
 569         if (!check_valid_map(F2FS_I_SB(inode), segno, off))
 570                 goto out;
 571
 572         set_new_dnode(&dn, inode, NULL, NULL, 0);
 573         err = get_dnode_of_data(&dn, bidx, LOOKUP_NODE);
 574         if (err)
 575                 goto out;
 576
 577         if (unlikely(dn.data_blkaddr == NULL_ADDR)) {
 578                 ClearPageUptodate(page);
 579                 goto put_out;
 580         }
 581
 582         /*
 583          * don't cache encrypted data into meta inode until previous dirty
 584          * data were writebacked to avoid racing between GC and flush.
 585          */
 586         f2fs_wait_on_page_writeback(page, DATA, true);
 587
 588         get_node_info(fio.sbi, dn.nid, &ni);
 589         set_summary(&sum, dn.nid, dn.ofs_in_node, ni.version);
 590
 591         /* read page */
 592         fio.page = page;
 593         fio.new_blkaddr = fio.old_blkaddr = dn.data_blkaddr;
 594
 595         allocate_data_block(fio.sbi, NULL, fio.old_blkaddr, &newaddr,
 596                                                         &sum, CURSEG_COLD_DATA);
 597
 598         fio.encrypted_page = pagecache_get_page(META_MAPPING(fio.sbi), newaddr,
 599                                         FGP_LOCK | FGP_CREAT, GFP_NOFS);
 600         if (!fio.encrypted_page) {
 601                 err = -ENOMEM;
 602                 goto recover_block;
 603         }
 604
 605         err = f2fs_submit_page_bio(&fio);
 606         if (err)
 607                 goto put_page_out;
 608
 609         /* write page */
 610         lock_page(fio.encrypted_page);
 611
 612         if (unlikely(fio.encrypted_page->mapping != META_MAPPING(fio.sbi))) {
 613                 err = -EIO;
 614                 goto put_page_out;
 615         }
 616         if (unlikely(!PageUptodate(fio.encrypted_page))) {
 617                 err = -EIO;
 618                 goto put_page_out;
 619         }
 620
 621         set_page_dirty(fio.encrypted_page);
 622         f2fs_wait_on_page_writeback(fio.encrypted_page, DATA, true);
 623         if (clear_page_dirty_for_io(fio.encrypted_page))
 624                 dec_page_count(fio.sbi, F2FS_DIRTY_META);
 625
 626         set_page_writeback(fio.encrypted_page);
 627
 628         /* allocate block address */
 629         f2fs_wait_on_page_writeback(dn.node_page, NODE, true);
 630
 631         fio.op = REQ_OP_WRITE;
 632         fio.op_flags = REQ_SYNC;
 633         fio.new_blkaddr = newaddr;
 634         f2fs_submit_page_mbio(&fio);
 635
 636         f2fs_update_data_blkaddr(&dn, newaddr);
 637         set_inode_flag(inode, FI_APPEND_WRITE);
 638         if (page->index == 0)
 639                 set_inode_flag(inode, FI_FIRST_BLOCK_WRITTEN);
 640 put_page_out:
 641         f2fs_put_page(fio.encrypted_page, 1);
 642 recover_block:
 643         if (err)
 644                 __f2fs_replace_block(fio.sbi, &sum, newaddr, fio.old_blkaddr,
 645                                                                 true, true);
 646 put_out:
 647         f2fs_put_dnode(&dn);
 648 out:
 649         f2fs_put_page(page, 1);
 650 }
 651
 652 static void move_data_page(struct inode *inode, block_t bidx, int gc_type,
 653                                                         unsigned int segno, int off)
 654 {
 655         struct page *page;
 656
 657         page = get_lock_data_page(inode, bidx, true);
 658         if (IS_ERR(page))
 659                 return;
 660
 661         if (!check_valid_map(F2FS_I_SB(inode), segno, off))
 662                 goto out;
 663
 664         if (gc_type == BG_GC) {
 665                 if (PageWriteback(page))
 666                         goto out;
 667                 set_page_dirty(page);
 668                 set_cold_data(page);
 669         } else {
 670                 struct f2fs_io_info fio = {
 671                         .sbi = F2FS_I_SB(inode),
 672                         .type = DATA,
 673                         .op = REQ_OP_WRITE,
 674                         .op_flags = REQ_SYNC,
 675                         .page = page,
 676                         .encrypted_page = NULL,
 677                 };
 678                 bool is_dirty = PageDirty(page);
 679                 int err;
 680
 681 retry:
 682                 set_page_dirty(page);
 683                 f2fs_wait_on_page_writeback(page, DATA, true);
 684                 if (clear_page_dirty_for_io(page)) {
 685                         inode_dec_dirty_pages(inode);
 686                         remove_dirty_inode(inode);
 687                 }
 688
 689                 set_cold_data(page);
 690
 691                 err = do_write_data_page(&fio);
 692                 if (err == -ENOMEM && is_dirty) {
 693                         congestion_wait(BLK_RW_ASYNC, HZ/50);
 694                         goto retry;
 695                 }
 696         }
 697 out:
 698         f2fs_put_page(page, 1);
 699 }
 700
 701 /*
 702  * This function tries to get parent node of victim data block, and identifies
 703  * data block validity. If the block is valid, copy that with cold status and
 704  * modify parent node.
 705  * If the parent node is not valid or the data block address is different,
 706  * the victim data block is ignored.
 707  */
 708 static void gc_data_segment(struct f2fs_sb_info *sbi, struct f2fs_summary *sum,
 709                 struct gc_inode_list *gc_list, unsigned int segno, int gc_type)
 710 {
 711         struct super_block *sb = sbi->sb;
 712         struct f2fs_summary *entry;
 713         block_t start_addr;
 714         int off;
 715         int phase = 0;
 716
 717         start_addr = START_BLOCK(sbi, segno);
 718
 719 next_step:
 720         entry = sum;
 721
 722         for (off = 0; off < sbi->blocks_per_seg; off++, entry++) {
 723                 struct page *data_page;
 724                 struct inode *inode;
 725                 struct node_info dni; /* dnode info for the data */
 726                 unsigned int ofs_in_node, nofs;
 727                 block_t start_bidx;
 728                 nid_t nid = le32_to_cpu(entry->nid);
 729
 730                 /* stop BG_GC if there is not enough free sections. */
 731                 if (gc_type == BG_GC && has_not_enough_free_secs(sbi, 0, 0))
 732                         return;
 733
 734                 if (check_valid_map(sbi, segno, off) == 0)
 735                         continue;
 736
 737                 if (phase == 0) {
 738                         ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nid), 1,
 739                                                         META_NAT, true);
 740                         continue;
 741                 }
 742
 743                 if (phase == 1) {
 744                         ra_node_page(sbi, nid);
 745                         continue;
 746                 }
 747
 748                 /* Get an inode by ino with checking validity */
 749                 if (!is_alive(sbi, entry, &dni, start_addr + off, &nofs))
 750                         continue;
 751
 752                 if (phase == 2) {
 753                         ra_node_page(sbi, dni.ino);
 754                         continue;
 755                 }
 756
 757                 ofs_in_node = le16_to_cpu(entry->ofs_in_node);
 758
 759                 if (phase == 3) {
 760                         inode = f2fs_iget(sb, dni.ino);
 761                         if (IS_ERR(inode) || is_bad_inode(inode))
 762                                 continue;
 763
 764                         /* if encrypted inode, let's go phase 3 */
 765                         if (f2fs_encrypted_inode(inode) &&
 766                                                 S_ISREG(inode->i_mode)) {
 767                                 add_gc_inode(gc_list, inode);
 768                                 continue;
 769                         }
 770
 771                         start_bidx = start_bidx_of_node(nofs, inode);
 772                         data_page = get_read_data_page(inode,
 773                                         start_bidx + ofs_in_node, REQ_RAHEAD,
 774                                         true);
 775                         if (IS_ERR(data_page)) {
 776                                 iput(inode);
 777                                 continue;
 778                         }
 779
 780                         f2fs_put_page(data_page, 0);
 781                         add_gc_inode(gc_list, inode);
 782                         continue;
 783                 }
 784
 785                 /* phase 4 */
 786                 inode = find_gc_inode(gc_list, dni.ino);
 787                 if (inode) {
 788                         struct f2fs_inode_info *fi = F2FS_I(inode);
 789                         bool locked = false;
 790
 791                         if (S_ISREG(inode->i_mode)) {
 792                                 if (!down_write_trylock(&fi->dio_rwsem[READ]))
 793                                         continue;
 794                                 if (!down_write_trylock(
 795                                                 &fi->dio_rwsem[WRITE])) {
 796                                         up_write(&fi->dio_rwsem[READ]);
 797                                         continue;
 798                                 }
 799                                 locked = true;
 800                         }
 801
 802                         start_bidx = start_bidx_of_node(nofs, inode)
 803                                                                 + ofs_in_node;
 804                         if (f2fs_encrypted_inode(inode) && S_ISREG(inode->i_mode))
 805                                 move_encrypted_block(inode, start_bidx, segno, off);
 806                         else
 807                                 move_data_page(inode, start_bidx, gc_type, segno, off);
 808
 809                         if (locked) {
 810                                 up_write(&fi->dio_rwsem[WRITE]);
 811                                 up_write(&fi->dio_rwsem[READ]);
 812                         }
 813
 814                         stat_inc_data_blk_count(sbi, 1, gc_type);
 815                 }
 816         }
 817
 818         if (++phase < 5)
 819                 goto next_step;
 820 }
 821
 822 static int __get_victim(struct f2fs_sb_info *sbi, unsigned int *victim,
 823                         int gc_type)
 824 {
 825         struct sit_info *sit_i = SIT_I(sbi);
 826         int ret;
 827
 828         mutex_lock(&sit_i->sentry_lock);
 829         ret = DIRTY_I(sbi)->v_ops->get_victim(sbi, victim, gc_type,
 830                                               NO_CHECK_TYPE, LFS);
 831         mutex_unlock(&sit_i->sentry_lock);
 832         return ret;
 833 }
 834
 835 static int do_garbage_collect(struct f2fs_sb_info *sbi,
 836                                 unsigned int start_segno,
 837                                 struct gc_inode_list *gc_list, int gc_type)
 838 {
 839         struct page *sum_page;
 840         struct f2fs_summary_block *sum;
 841         struct blk_plug plug;
 842         unsigned int segno = start_segno;
 843         unsigned int end_segno = start_segno + sbi->segs_per_sec;
 844         int sec_freed = 0;
 845         unsigned char type = IS_DATASEG(get_seg_entry(sbi, segno)->type) ?
 846                                                 SUM_TYPE_DATA : SUM_TYPE_NODE;
 847
 848         /* readahead multi ssa blocks those have contiguous address */
 849         if (sbi->segs_per_sec > 1)
 850                 ra_meta_pages(sbi, GET_SUM_BLOCK(sbi, segno),
 851                                         sbi->segs_per_sec, META_SSA, true);
 852
 853         /* reference all summary page */
 854         while (segno < end_segno) {
 855                 sum_page = get_sum_page(sbi, segno++);
 856                 unlock_page(sum_page);
 857         }
 858
 859         blk_start_plug(&plug);
 860
 861         for (segno = start_segno; segno < end_segno; segno++) {
 862
 863                 /* find segment summary of victim */
 864                 sum_page = find_get_page(META_MAPPING(sbi),
 865                                         GET_SUM_BLOCK(sbi, segno));
 866                 f2fs_put_page(sum_page, 0);
 867
 868                 if (get_valid_blocks(sbi, segno, 1) == 0 ||
 869                                 !PageUptodate(sum_page) ||
 870                                 unlikely(f2fs_cp_error(sbi)))
 871                         goto next;
 872
 873                 sum = page_address(sum_page);
 874                 f2fs_bug_on(sbi, type != GET_SUM_TYPE((&sum->footer)));
 875
 876                 /*
 877                  * this is to avoid deadlock:
 878                  * - lock_page(sum_page)         - f2fs_replace_block
 879                  *  - check_valid_map()            - mutex_lock(sentry_lock)
 880                  *   - mutex_lock(sentry_lock)     - change_curseg()
 881                  *                                  - lock_page(sum_page)
 882                  */
 883
 884                 if (type == SUM_TYPE_NODE)
 885                         gc_node_segment(sbi, sum->entries, segno, gc_type);
 886                 else
 887                         gc_data_segment(sbi, sum->entries, gc_list, segno,
 888                                                                 gc_type);
 889
 890                 stat_inc_seg_count(sbi, type, gc_type);
 891 next:
 892                 f2fs_put_page(sum_page, 0);
 893         }
 894
 895         if (gc_type == FG_GC)
 896                 f2fs_submit_merged_bio(sbi,
 897                                 (type == SUM_TYPE_NODE) ? NODE : DATA, WRITE);
 898
 899         blk_finish_plug(&plug);
 900
 901         if (gc_type == FG_GC &&
 902                 get_valid_blocks(sbi, start_segno, sbi->segs_per_sec) == 0)
 903                 sec_freed = 1;
 904
 905         stat_inc_call_count(sbi->stat_info);
 906
 907         return sec_freed;
 908 }
 909
 910 int f2fs_gc(struct f2fs_sb_info *sbi, bool sync, bool background)
 911 {
 912         unsigned int segno;
 913         int gc_type = sync ? FG_GC : BG_GC;
 914         int sec_freed = 0;
 915         int ret = -EINVAL;
 916         struct cp_control cpc;
 917         struct gc_inode_list gc_list = {
 918                 .ilist = LIST_HEAD_INIT(gc_list.ilist),
 919                 .iroot = RADIX_TREE_INIT(GFP_NOFS),
 920         };
 921
 922         cpc.reason = __get_cp_reason(sbi);
 923 gc_more:
 924         segno = NULL_SEGNO;
 925
 926         if (unlikely(!(sbi->sb->s_flags & MS_ACTIVE)))
 927                 goto stop;
 928         if (unlikely(f2fs_cp_error(sbi))) {
 929                 ret = -EIO;
 930                 goto stop;
 931         }
 932
 933         if (gc_type == BG_GC && has_not_enough_free_secs(sbi, sec_freed, 0)) {
 934                 gc_type = FG_GC;
 935                 /*
 936                  * If there is no victim and no prefree segment but still not
 937                  * enough free sections, we should flush dent/node blocks and do
 938                  * garbage collections.
 939                  */
 940                 if (__get_victim(sbi, &segno, gc_type) ||
 941                                                 prefree_segments(sbi)) {
 942                         ret = write_checkpoint(sbi, &cpc);
 943                         if (ret)
 944                                 goto stop;
 945                         segno = NULL_SEGNO;
 946                 } else if (has_not_enough_free_secs(sbi, 0, 0)) {
 947                         ret = write_checkpoint(sbi, &cpc);
 948                         if (ret)
 949                                 goto stop;
 950                 }
 951         } else if (gc_type == BG_GC && !background) {
 952                 /* f2fs_balance_fs doesn't need to do BG_GC in critical path. */
 953                 goto stop;
 954         }
 955
 956         if (segno == NULL_SEGNO && !__get_victim(sbi, &segno, gc_type))
 957                 goto stop;
 958         ret = 0;
 959
 960         if (do_garbage_collect(sbi, segno, &gc_list, gc_type) &&
 961                         gc_type == FG_GC)
 962                 sec_freed++;
 963
 964         if (gc_type == FG_GC)
 965                 sbi->cur_victim_sec = NULL_SEGNO;
 966
 967         if (!sync) {
 968                 if (has_not_enough_free_secs(sbi, sec_freed, 0))
 969                         goto gc_more;
 970
 971                 if (gc_type == FG_GC)
 972                         ret = write_checkpoint(sbi, &cpc);
 973         }
 974 stop:
 975         mutex_unlock(&sbi->gc_mutex);
 976
 977         put_gc_inode(&gc_list);
 978
 979         if (sync)
 980                 ret = sec_freed ? 0 : -EAGAIN;
 981         return ret;
 982 }
 983
 984 void build_gc_manager(struct f2fs_sb_info *sbi)
 985 {
 986         DIRTY_I(sbi)->v_ops = &default_v_ops;
 987 }