fs/f2fs/segment.c

   1 /*
   2  * fs/f2fs/segment.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/f2fs_fs.h>
  13 #include <linux/bio.h>
  14 #include <linux/blkdev.h>
  15 #include <linux/prefetch.h>
  16 #include <linux/vmalloc.h>
  17
  18 #include "f2fs.h"
  19 #include "segment.h"
  20 #include "node.h"
  21 #include <trace/events/f2fs.h>
  22
  23 /*
  24  * This function balances dirty node and dentry pages.
  25  * In addition, it controls garbage collection.
  26  */
  27 void f2fs_balance_fs(struct f2fs_sb_info *sbi)
  28 {
  29         /*
  30          * We should do GC or end up with checkpoint, if there are so many dirty
  31          * dir/node pages without enough free segments.
  32          */
  33         if (has_not_enough_free_secs(sbi, 0)) {
  34                 mutex_lock(&sbi->gc_mutex);
  35                 f2fs_gc(sbi);
  36         }
  37 }
  38
  39 void f2fs_balance_fs_bg(struct f2fs_sb_info *sbi)
  40 {
  41         /* check the # of cached NAT entries and prefree segments */
  42         if (try_to_free_nats(sbi, NAT_ENTRY_PER_BLOCK) ||
  43                                 excess_prefree_segs(sbi))
  44                 f2fs_sync_fs(sbi->sb, true);
  45 }
  46
  47 static void __locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
  48                 enum dirty_type dirty_type)
  49 {
  50         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
  51
  52         /* need not be added */
  53         if (IS_CURSEG(sbi, segno))
  54                 return;
  55
  56         if (!test_and_set_bit(segno, dirty_i->dirty_segmap[dirty_type]))
  57                 dirty_i->nr_dirty[dirty_type]++;
  58
  59         if (dirty_type == DIRTY) {
  60                 struct seg_entry *sentry = get_seg_entry(sbi, segno);
  61                 enum dirty_type t = sentry->type;
  62
  63                 if (!test_and_set_bit(segno, dirty_i->dirty_segmap[t]))
  64                         dirty_i->nr_dirty[t]++;
  65         }
  66 }
  67
  68 static void __remove_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
  69                 enum dirty_type dirty_type)
  70 {
  71         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
  72
  73         if (test_and_clear_bit(segno, dirty_i->dirty_segmap[dirty_type]))
  74                 dirty_i->nr_dirty[dirty_type]--;
  75
  76         if (dirty_type == DIRTY) {
  77                 struct seg_entry *sentry = get_seg_entry(sbi, segno);
  78                 enum dirty_type t = sentry->type;
  79
  80                 if (test_and_clear_bit(segno, dirty_i->dirty_segmap[t]))
  81                         dirty_i->nr_dirty[t]--;
  82
  83                 if (get_valid_blocks(sbi, segno, sbi->segs_per_sec) == 0)
  84                         clear_bit(GET_SECNO(sbi, segno),
  85                                                 dirty_i->victim_secmap);
  86         }
  87 }
  88
  89 /*
  90  * Should not occur error such as -ENOMEM.
  91  * Adding dirty entry into seglist is not critical operation.
  92  * If a given segment is one of current working segments, it won't be added.
  93  */
  94 static void locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno)
  95 {
  96         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
  97         unsigned short valid_blocks;
  98
  99         if (segno == NULL_SEGNO || IS_CURSEG(sbi, segno))
 100                 return;
 101
 102         mutex_lock(&dirty_i->seglist_lock);
 103
 104         valid_blocks = get_valid_blocks(sbi, segno, 0);
 105
 106         if (valid_blocks == 0) {
 107                 __locate_dirty_segment(sbi, segno, PRE);
 108                 __remove_dirty_segment(sbi, segno, DIRTY);
 109         } else if (valid_blocks < sbi->blocks_per_seg) {
 110                 __locate_dirty_segment(sbi, segno, DIRTY);
 111         } else {
 112                 /* Recovery routine with SSR needs this */
 113                 __remove_dirty_segment(sbi, segno, DIRTY);
 114         }
 115
 116         mutex_unlock(&dirty_i->seglist_lock);
 117 }
 118
 119 /*
 120  * Should call clear_prefree_segments after checkpoint is done.
 121  */
 122 static void set_prefree_as_free_segments(struct f2fs_sb_info *sbi)
 123 {
 124         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
 125         unsigned int segno = -1;
 126         unsigned int total_segs = TOTAL_SEGS(sbi);
 127
 128         mutex_lock(&dirty_i->seglist_lock);
 129         while (1) {
 130                 segno = find_next_bit(dirty_i->dirty_segmap[PRE], total_segs,
 131                                 segno + 1);
 132                 if (segno >= total_segs)
 133                         break;
 134                 __set_test_and_free(sbi, segno);
 135         }
 136         mutex_unlock(&dirty_i->seglist_lock);
 137 }
 138
 139 void clear_prefree_segments(struct f2fs_sb_info *sbi)
 140 {
 141         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
 142         unsigned long *prefree_map = dirty_i->dirty_segmap[PRE];
 143         unsigned int total_segs = TOTAL_SEGS(sbi);
 144         unsigned int start = 0, end = -1;
 145
 146         mutex_lock(&dirty_i->seglist_lock);
 147
 148         while (1) {
 149                 int i;
 150                 start = find_next_bit(prefree_map, total_segs, end + 1);
 151                 if (start >= total_segs)
 152                         break;
 153                 end = find_next_zero_bit(prefree_map, total_segs, start + 1);
 154
 155                 for (i = start; i < end; i++)
 156                         clear_bit(i, prefree_map);
 157
 158                 dirty_i->nr_dirty[PRE] -= end - start;
 159
 160                 if (!test_opt(sbi, DISCARD))
 161                         continue;
 162
 163                 blkdev_issue_discard(sbi->sb->s_bdev,
 164                                 START_BLOCK(sbi, start) <<
 165                                 sbi->log_sectors_per_block,
 166                                 (1 << (sbi->log_sectors_per_block +
 167                                 sbi->log_blocks_per_seg)) * (end - start),
 168                                 GFP_NOFS, 0);
 169         }
 170         mutex_unlock(&dirty_i->seglist_lock);
 171 }
 172
 173 static void __mark_sit_entry_dirty(struct f2fs_sb_info *sbi, unsigned int segno)
 174 {
 175         struct sit_info *sit_i = SIT_I(sbi);
 176         if (!__test_and_set_bit(segno, sit_i->dirty_sentries_bitmap))
 177                 sit_i->dirty_sentries++;
 178 }
 179
 180 static void __set_sit_entry_type(struct f2fs_sb_info *sbi, int type,
 181                                         unsigned int segno, int modified)
 182 {
 183         struct seg_entry *se = get_seg_entry(sbi, segno);
 184         se->type = type;
 185         if (modified)
 186                 __mark_sit_entry_dirty(sbi, segno);
 187 }
 188
 189 static void update_sit_entry(struct f2fs_sb_info *sbi, block_t blkaddr, int del)
 190 {
 191         struct seg_entry *se;
 192         unsigned int segno, offset;
 193         long int new_vblocks;
 194
 195         segno = GET_SEGNO(sbi, blkaddr);
 196
 197         se = get_seg_entry(sbi, segno);
 198         new_vblocks = se->valid_blocks + del;
 199         offset = GET_SEGOFF_FROM_SEG0(sbi, blkaddr) & (sbi->blocks_per_seg - 1);
 200
 201         f2fs_bug_on((new_vblocks >> (sizeof(unsigned short) << 3) ||
 202                                 (new_vblocks > sbi->blocks_per_seg)));
 203
 204         se->valid_blocks = new_vblocks;
 205         se->mtime = get_mtime(sbi);
 206         SIT_I(sbi)->max_mtime = se->mtime;
 207
 208         /* Update valid block bitmap */
 209         if (del > 0) {
 210                 if (f2fs_set_bit(offset, se->cur_valid_map))
 211                         BUG();
 212         } else {
 213                 if (!f2fs_clear_bit(offset, se->cur_valid_map))
 214                         BUG();
 215         }
 216         if (!f2fs_test_bit(offset, se->ckpt_valid_map))
 217                 se->ckpt_valid_blocks += del;
 218
 219         __mark_sit_entry_dirty(sbi, segno);
 220
 221         /* update total number of valid blocks to be written in ckpt area */
 222         SIT_I(sbi)->written_valid_blocks += del;
 223
 224         if (sbi->segs_per_sec > 1)
 225                 get_sec_entry(sbi, segno)->valid_blocks += del;
 226 }
 227
 228 static void refresh_sit_entry(struct f2fs_sb_info *sbi,
 229                         block_t old_blkaddr, block_t new_blkaddr)
 230 {
 231         update_sit_entry(sbi, new_blkaddr, 1);
 232         if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO)
 233                 update_sit_entry(sbi, old_blkaddr, -1);
 234 }
 235
 236 void invalidate_blocks(struct f2fs_sb_info *sbi, block_t addr)
 237 {
 238         unsigned int segno = GET_SEGNO(sbi, addr);
 239         struct sit_info *sit_i = SIT_I(sbi);
 240
 241         f2fs_bug_on(addr == NULL_ADDR);
 242         if (addr == NEW_ADDR)
 243                 return;
 244
 245         /* add it into sit main buffer */
 246         mutex_lock(&sit_i->sentry_lock);
 247
 248         update_sit_entry(sbi, addr, -1);
 249
 250         /* add it into dirty seglist */
 251         locate_dirty_segment(sbi, segno);
 252
 253         mutex_unlock(&sit_i->sentry_lock);
 254 }
 255
 256 /*
 257  * This function should be resided under the curseg_mutex lock
 258  */
 259 static void __add_sum_entry(struct f2fs_sb_info *sbi, int type,
 260                                         struct f2fs_summary *sum)
 261 {
 262         struct curseg_info *curseg = CURSEG_I(sbi, type);
 263         void *addr = curseg->sum_blk;
 264         addr += curseg->next_blkoff * sizeof(struct f2fs_summary);
 265         memcpy(addr, sum, sizeof(struct f2fs_summary));
 266 }
 267
 268 /*
 269  * Calculate the number of current summary pages for writing
 270  */
 271 int npages_for_summary_flush(struct f2fs_sb_info *sbi)
 272 {
 273         int valid_sum_count = 0;
 274         int i, sum_in_page;
 275
 276         for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
 277                 if (sbi->ckpt->alloc_type[i] == SSR)
 278                         valid_sum_count += sbi->blocks_per_seg;
 279                 else
 280                         valid_sum_count += curseg_blkoff(sbi, i);
 281         }
 282
 283         sum_in_page = (PAGE_CACHE_SIZE - 2 * SUM_JOURNAL_SIZE -
 284                         SUM_FOOTER_SIZE) / SUMMARY_SIZE;
 285         if (valid_sum_count <= sum_in_page)
 286                 return 1;
 287         else if ((valid_sum_count - sum_in_page) <=
 288                 (PAGE_CACHE_SIZE - SUM_FOOTER_SIZE) / SUMMARY_SIZE)
 289                 return 2;
 290         return 3;
 291 }
 292
 293 /*
 294  * Caller should put this summary page
 295  */
 296 struct page *get_sum_page(struct f2fs_sb_info *sbi, unsigned int segno)
 297 {
 298         return get_meta_page(sbi, GET_SUM_BLOCK(sbi, segno));
 299 }
 300
 301 static void write_sum_page(struct f2fs_sb_info *sbi,
 302                         struct f2fs_summary_block *sum_blk, block_t blk_addr)
 303 {
 304         struct page *page = grab_meta_page(sbi, blk_addr);
 305         void *kaddr = page_address(page);
 306         memcpy(kaddr, sum_blk, PAGE_CACHE_SIZE);
 307         set_page_dirty(page);
 308         f2fs_put_page(page, 1);
 309 }
 310
 311 static int is_next_segment_free(struct f2fs_sb_info *sbi, int type)
 312 {
 313         struct curseg_info *curseg = CURSEG_I(sbi, type);
 314         unsigned int segno = curseg->segno + 1;
 315         struct free_segmap_info *free_i = FREE_I(sbi);
 316
 317         if (segno < TOTAL_SEGS(sbi) && segno % sbi->segs_per_sec)
 318                 return !test_bit(segno, free_i->free_segmap);
 319         return 0;
 320 }
 321
 322 /*
 323  * Find a new segment from the free segments bitmap to right order
 324  * This function should be returned with success, otherwise BUG
 325  */
 326 static void get_new_segment(struct f2fs_sb_info *sbi,
 327                         unsigned int *newseg, bool new_sec, int dir)
 328 {
 329         struct free_segmap_info *free_i = FREE_I(sbi);
 330         unsigned int segno, secno, zoneno;
 331         unsigned int total_zones = TOTAL_SECS(sbi) / sbi->secs_per_zone;
 332         unsigned int hint = *newseg / sbi->segs_per_sec;
 333         unsigned int old_zoneno = GET_ZONENO_FROM_SEGNO(sbi, *newseg);
 334         unsigned int left_start = hint;
 335         bool init = true;
 336         int go_left = 0;
 337         int i;
 338
 339         write_lock(&free_i->segmap_lock);
 340
 341         if (!new_sec && ((*newseg + 1) % sbi->segs_per_sec)) {
 342                 segno = find_next_zero_bit(free_i->free_segmap,
 343                                         TOTAL_SEGS(sbi), *newseg + 1);
 344                 if (segno - *newseg < sbi->segs_per_sec -
 345                                         (*newseg % sbi->segs_per_sec))
 346                         goto got_it;
 347         }
 348 find_other_zone:
 349         secno = find_next_zero_bit(free_i->free_secmap, TOTAL_SECS(sbi), hint);
 350         if (secno >= TOTAL_SECS(sbi)) {
 351                 if (dir == ALLOC_RIGHT) {
 352                         secno = find_next_zero_bit(free_i->free_secmap,
 353                                                         TOTAL_SECS(sbi), 0);
 354                         f2fs_bug_on(secno >= TOTAL_SECS(sbi));
 355                 } else {
 356                         go_left = 1;
 357                         left_start = hint - 1;
 358                 }
 359         }
 360         if (go_left == 0)
 361                 goto skip_left;
 362
 363         while (test_bit(left_start, free_i->free_secmap)) {
 364                 if (left_start > 0) {
 365                         left_start--;
 366                         continue;
 367                 }
 368                 left_start = find_next_zero_bit(free_i->free_secmap,
 369                                                         TOTAL_SECS(sbi), 0);
 370                 f2fs_bug_on(left_start >= TOTAL_SECS(sbi));
 371                 break;
 372         }
 373         secno = left_start;
 374 skip_left:
 375         hint = secno;
 376         segno = secno * sbi->segs_per_sec;
 377         zoneno = secno / sbi->secs_per_zone;
 378
 379         /* give up on finding another zone */
 380         if (!init)
 381                 goto got_it;
 382         if (sbi->secs_per_zone == 1)
 383                 goto got_it;
 384         if (zoneno == old_zoneno)
 385                 goto got_it;
 386         if (dir == ALLOC_LEFT) {
 387                 if (!go_left && zoneno + 1 >= total_zones)
 388                         goto got_it;
 389                 if (go_left && zoneno == 0)
 390                         goto got_it;
 391         }
 392         for (i = 0; i < NR_CURSEG_TYPE; i++)
 393                 if (CURSEG_I(sbi, i)->zone == zoneno)
 394                         break;
 395
 396         if (i < NR_CURSEG_TYPE) {
 397                 /* zone is in user, try another */
 398                 if (go_left)
 399                         hint = zoneno * sbi->secs_per_zone - 1;
 400                 else if (zoneno + 1 >= total_zones)
 401                         hint = 0;
 402                 else
 403                         hint = (zoneno + 1) * sbi->secs_per_zone;
 404                 init = false;
 405                 goto find_other_zone;
 406         }
 407 got_it:
 408         /* set it as dirty segment in free segmap */
 409         f2fs_bug_on(test_bit(segno, free_i->free_segmap));
 410         __set_inuse(sbi, segno);
 411         *newseg = segno;
 412         write_unlock(&free_i->segmap_lock);
 413 }
 414
 415 static void reset_curseg(struct f2fs_sb_info *sbi, int type, int modified)
 416 {
 417         struct curseg_info *curseg = CURSEG_I(sbi, type);
 418         struct summary_footer *sum_footer;
 419
 420         curseg->segno = curseg->next_segno;
 421         curseg->zone = GET_ZONENO_FROM_SEGNO(sbi, curseg->segno);
 422         curseg->next_blkoff = 0;
 423         curseg->next_segno = NULL_SEGNO;
 424
 425         sum_footer = &(curseg->sum_blk->footer);
 426         memset(sum_footer, 0, sizeof(struct summary_footer));
 427         if (IS_DATASEG(type))
 428                 SET_SUM_TYPE(sum_footer, SUM_TYPE_DATA);
 429         if (IS_NODESEG(type))
 430                 SET_SUM_TYPE(sum_footer, SUM_TYPE_NODE);
 431         __set_sit_entry_type(sbi, type, curseg->segno, modified);
 432 }
 433
 434 /*
 435  * Allocate a current working segment.
 436  * This function always allocates a free segment in LFS manner.
 437  */
 438 static void new_curseg(struct f2fs_sb_info *sbi, int type, bool new_sec)
 439 {
 440         struct curseg_info *curseg = CURSEG_I(sbi, type);
 441         unsigned int segno = curseg->segno;
 442         int dir = ALLOC_LEFT;
 443
 444         write_sum_page(sbi, curseg->sum_blk,
 445                                 GET_SUM_BLOCK(sbi, segno));
 446         if (type == CURSEG_WARM_DATA || type == CURSEG_COLD_DATA)
 447                 dir = ALLOC_RIGHT;
 448
 449         if (test_opt(sbi, NOHEAP))
 450                 dir = ALLOC_RIGHT;
 451
 452         get_new_segment(sbi, &segno, new_sec, dir);
 453         curseg->next_segno = segno;
 454         reset_curseg(sbi, type, 1);
 455         curseg->alloc_type = LFS;
 456 }
 457
 458 static void __next_free_blkoff(struct f2fs_sb_info *sbi,
 459                         struct curseg_info *seg, block_t start)
 460 {
 461         struct seg_entry *se = get_seg_entry(sbi, seg->segno);
 462         block_t ofs;
 463         for (ofs = start; ofs < sbi->blocks_per_seg; ofs++) {
 464                 if (!f2fs_test_bit(ofs, se->ckpt_valid_map)
 465                         && !f2fs_test_bit(ofs, se->cur_valid_map))
 466                         break;
 467         }
 468         seg->next_blkoff = ofs;
 469 }
 470
 471 /*
 472  * If a segment is written by LFS manner, next block offset is just obtained
 473  * by increasing the current block offset. However, if a segment is written by
 474  * SSR manner, next block offset obtained by calling __next_free_blkoff
 475  */
 476 static void __refresh_next_blkoff(struct f2fs_sb_info *sbi,
 477                                 struct curseg_info *seg)
 478 {
 479         if (seg->alloc_type == SSR)
 480                 __next_free_blkoff(sbi, seg, seg->next_blkoff + 1);
 481         else
 482                 seg->next_blkoff++;
 483 }
 484
 485 /*
 486  * This function always allocates a used segment (from dirty seglist) by SSR
 487  * manner, so it should recover the existing segment information of valid blocks
 488  */
 489 static void change_curseg(struct f2fs_sb_info *sbi, int type, bool reuse)
 490 {
 491         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
 492         struct curseg_info *curseg = CURSEG_I(sbi, type);
 493         unsigned int new_segno = curseg->next_segno;
 494         struct f2fs_summary_block *sum_node;
 495         struct page *sum_page;
 496
 497         write_sum_page(sbi, curseg->sum_blk,
 498                                 GET_SUM_BLOCK(sbi, curseg->segno));
 499         __set_test_and_inuse(sbi, new_segno);
 500
 501         mutex_lock(&dirty_i->seglist_lock);
 502         __remove_dirty_segment(sbi, new_segno, PRE);
 503         __remove_dirty_segment(sbi, new_segno, DIRTY);
 504         mutex_unlock(&dirty_i->seglist_lock);
 505
 506         reset_curseg(sbi, type, 1);
 507         curseg->alloc_type = SSR;
 508         __next_free_blkoff(sbi, curseg, 0);
 509
 510         if (reuse) {
 511                 sum_page = get_sum_page(sbi, new_segno);
 512                 sum_node = (struct f2fs_summary_block *)page_address(sum_page);
 513                 memcpy(curseg->sum_blk, sum_node, SUM_ENTRY_SIZE);
 514                 f2fs_put_page(sum_page, 1);
 515         }
 516 }
 517
 518 static int get_ssr_segment(struct f2fs_sb_info *sbi, int type)
 519 {
 520         struct curseg_info *curseg = CURSEG_I(sbi, type);
 521         const struct victim_selection *v_ops = DIRTY_I(sbi)->v_ops;
 522
 523         if (IS_NODESEG(type) || !has_not_enough_free_secs(sbi, 0))
 524                 return v_ops->get_victim(sbi,
 525                                 &(curseg)->next_segno, BG_GC, type, SSR);
 526
 527         /* For data segments, let's do SSR more intensively */
 528         for (; type >= CURSEG_HOT_DATA; type--)
 529                 if (v_ops->get_victim(sbi, &(curseg)->next_segno,
 530                                                 BG_GC, type, SSR))
 531                         return 1;
 532         return 0;
 533 }
 534
 535 /*
 536  * flush out current segment and replace it with new segment
 537  * This function should be returned with success, otherwise BUG
 538  */
 539 static void allocate_segment_by_default(struct f2fs_sb_info *sbi,
 540                                                 int type, bool force)
 541 {
 542         struct curseg_info *curseg = CURSEG_I(sbi, type);
 543
 544         if (force)
 545                 new_curseg(sbi, type, true);
 546         else if (type == CURSEG_WARM_NODE)
 547                 new_curseg(sbi, type, false);
 548         else if (curseg->alloc_type == LFS && is_next_segment_free(sbi, type))
 549                 new_curseg(sbi, type, false);
 550         else if (need_SSR(sbi) && get_ssr_segment(sbi, type))
 551                 change_curseg(sbi, type, true);
 552         else
 553                 new_curseg(sbi, type, false);
 554
 555         stat_inc_seg_type(sbi, curseg);
 556 }
 557
 558 void allocate_new_segments(struct f2fs_sb_info *sbi)
 559 {
 560         struct curseg_info *curseg;
 561         unsigned int old_curseg;
 562         int i;
 563
 564         for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
 565                 curseg = CURSEG_I(sbi, i);
 566                 old_curseg = curseg->segno;
 567                 SIT_I(sbi)->s_ops->allocate_segment(sbi, i, true);
 568                 locate_dirty_segment(sbi, old_curseg);
 569         }
 570 }
 571
 572 static const struct segment_allocation default_salloc_ops = {
 573         .allocate_segment = allocate_segment_by_default,
 574 };
 575
 576 static void f2fs_end_io_write(struct bio *bio, int err)
 577 {
 578         struct bio_private *p = bio->bi_private;
 579         struct bio_vec *bvec;
 580         int i;
 581
 582         bio_for_each_segment_all(bvec, bio, i) {
 583                 struct page *page = bvec->bv_page;
 584
 585                 if (err) {
 586                         SetPageError(page);
 587                         if (page->mapping)
 588                                 set_bit(AS_EIO, &page->mapping->flags);
 589                         set_ckpt_flags(p->sbi->ckpt, CP_ERROR_FLAG);
 590                         p->sbi->sb->s_flags |= MS_RDONLY;
 591                 }
 592                 end_page_writeback(page);
 593                 dec_page_count(p->sbi, F2FS_WRITEBACK);
 594         }
 595
 596         if (p->is_sync)
 597                 complete(p->wait);
 598
 599         if (!get_pages(p->sbi, F2FS_WRITEBACK) &&
 600                         !list_empty(&p->sbi->cp_wait.task_list))
 601                 wake_up(&p->sbi->cp_wait);
 602
 603         kfree(p);
 604         bio_put(bio);
 605 }
 606
 607 struct bio *f2fs_bio_alloc(struct block_device *bdev, int npages)
 608 {
 609         struct bio *bio;
 610
 611         /* No failure on bio allocation */
 612         bio = bio_alloc(GFP_NOIO, npages);
 613         bio->bi_bdev = bdev;
 614         bio->bi_private = NULL;
 615
 616         return bio;
 617 }
 618
 619 static void do_submit_bio(struct f2fs_sb_info *sbi,
 620                                 enum page_type type, bool sync)
 621 {
 622         int rw = sync ? WRITE_SYNC : WRITE;
 623         enum page_type btype = type > META ? META : type;
 624
 625         if (type >= META_FLUSH)
 626                 rw = WRITE_FLUSH_FUA;
 627
 628         if (btype == META)
 629                 rw |= REQ_META;
 630
 631         if (sbi->bio[btype]) {
 632                 struct bio_private *p = sbi->bio[btype]->bi_private;
 633                 p->sbi = sbi;
 634                 sbi->bio[btype]->bi_end_io = f2fs_end_io_write;
 635
 636                 trace_f2fs_do_submit_bio(sbi->sb, btype, sync, sbi->bio[btype]);
 637
 638                 if (type == META_FLUSH) {
 639                         DECLARE_COMPLETION_ONSTACK(wait);
 640                         p->is_sync = true;
 641                         p->wait = &wait;
 642                         submit_bio(rw, sbi->bio[btype]);
 643                         wait_for_completion(&wait);
 644                 } else {
 645                         p->is_sync = false;
 646                         submit_bio(rw, sbi->bio[btype]);
 647                 }
 648                 sbi->bio[btype] = NULL;
 649         }
 650 }
 651
 652 void f2fs_submit_bio(struct f2fs_sb_info *sbi, enum page_type type, bool sync)
 653 {
 654         down_write(&sbi->bio_sem);
 655         do_submit_bio(sbi, type, sync);
 656         up_write(&sbi->bio_sem);
 657 }
 658
 659 static void submit_write_page(struct f2fs_sb_info *sbi, struct page *page,
 660                                 block_t blk_addr, enum page_type type)
 661 {
 662         struct block_device *bdev = sbi->sb->s_bdev;
 663         int bio_blocks;
 664
 665         verify_block_addr(sbi, blk_addr);
 666
 667         down_write(&sbi->bio_sem);
 668
 669         inc_page_count(sbi, F2FS_WRITEBACK);
 670
 671         if (sbi->bio[type] && sbi->last_block_in_bio[type] != blk_addr - 1)
 672                 do_submit_bio(sbi, type, false);
 673 alloc_new:
 674         if (sbi->bio[type] == NULL) {
 675                 struct bio_private *priv;
 676 retry:
 677                 priv = kmalloc(sizeof(struct bio_private), GFP_NOFS);
 678                 if (!priv) {
 679                         cond_resched();
 680                         goto retry;
 681                 }
 682
 683                 bio_blocks = MAX_BIO_BLOCKS(max_hw_blocks(sbi));
 684                 sbi->bio[type] = f2fs_bio_alloc(bdev, bio_blocks);
 685                 sbi->bio[type]->bi_iter.bi_sector = SECTOR_FROM_BLOCK(sbi, blk_addr);
 686                 sbi->bio[type]->bi_private = priv;
 687                 /*
 688                  * The end_io will be assigned at the sumbission phase.
 689                  * Until then, let bio_add_page() merge consecutive IOs as much
 690                  * as possible.
 691                  */
 692         }
 693
 694         if (bio_add_page(sbi->bio[type], page, PAGE_CACHE_SIZE, 0) <
 695                                                         PAGE_CACHE_SIZE) {
 696                 do_submit_bio(sbi, type, false);
 697                 goto alloc_new;
 698         }
 699
 700         sbi->last_block_in_bio[type] = blk_addr;
 701
 702         up_write(&sbi->bio_sem);
 703         trace_f2fs_submit_write_page(page, blk_addr, type);
 704 }
 705
 706 void f2fs_wait_on_page_writeback(struct page *page,
 707                                 enum page_type type, bool sync)
 708 {
 709         struct f2fs_sb_info *sbi = F2FS_SB(page->mapping->host->i_sb);
 710         if (PageWriteback(page)) {
 711                 f2fs_submit_bio(sbi, type, sync);
 712                 wait_on_page_writeback(page);
 713         }
 714 }
 715
 716 static bool __has_curseg_space(struct f2fs_sb_info *sbi, int type)
 717 {
 718         struct curseg_info *curseg = CURSEG_I(sbi, type);
 719         if (curseg->next_blkoff < sbi->blocks_per_seg)
 720                 return true;
 721         return false;
 722 }
 723
 724 static int __get_segment_type_2(struct page *page, enum page_type p_type)
 725 {
 726         if (p_type == DATA)
 727                 return CURSEG_HOT_DATA;
 728         else
 729                 return CURSEG_HOT_NODE;
 730 }
 731
 732 static int __get_segment_type_4(struct page *page, enum page_type p_type)
 733 {
 734         if (p_type == DATA) {
 735                 struct inode *inode = page->mapping->host;
 736
 737                 if (S_ISDIR(inode->i_mode))
 738                         return CURSEG_HOT_DATA;
 739                 else
 740                         return CURSEG_COLD_DATA;
 741         } else {
 742                 if (IS_DNODE(page) && !is_cold_node(page))
 743                         return CURSEG_HOT_NODE;
 744                 else
 745                         return CURSEG_COLD_NODE;
 746         }
 747 }
 748
 749 static int __get_segment_type_6(struct page *page, enum page_type p_type)
 750 {
 751         if (p_type == DATA) {
 752                 struct inode *inode = page->mapping->host;
 753
 754                 if (S_ISDIR(inode->i_mode))
 755                         return CURSEG_HOT_DATA;
 756                 else if (is_cold_data(page) || file_is_cold(inode))
 757                         return CURSEG_COLD_DATA;
 758                 else
 759                         return CURSEG_WARM_DATA;
 760         } else {
 761                 if (IS_DNODE(page))
 762                         return is_cold_node(page) ? CURSEG_WARM_NODE :
 763                                                 CURSEG_HOT_NODE;
 764                 else
 765                         return CURSEG_COLD_NODE;
 766         }
 767 }
 768
 769 static int __get_segment_type(struct page *page, enum page_type p_type)
 770 {
 771         struct f2fs_sb_info *sbi = F2FS_SB(page->mapping->host->i_sb);
 772         switch (sbi->active_logs) {
 773         case 2:
 774                 return __get_segment_type_2(page, p_type);
 775         case 4:
 776                 return __get_segment_type_4(page, p_type);
 777         }
 778         /* NR_CURSEG_TYPE(6) logs by default */
 779         f2fs_bug_on(sbi->active_logs != NR_CURSEG_TYPE);
 780         return __get_segment_type_6(page, p_type);
 781 }
 782
 783 static void do_write_page(struct f2fs_sb_info *sbi, struct page *page,
 784                         block_t old_blkaddr, block_t *new_blkaddr,
 785                         struct f2fs_summary *sum, enum page_type p_type)
 786 {
 787         struct sit_info *sit_i = SIT_I(sbi);
 788         struct curseg_info *curseg;
 789         unsigned int old_cursegno;
 790         int type;
 791
 792         type = __get_segment_type(page, p_type);
 793         curseg = CURSEG_I(sbi, type);
 794
 795         mutex_lock(&curseg->curseg_mutex);
 796
 797         *new_blkaddr = NEXT_FREE_BLKADDR(sbi, curseg);
 798         old_cursegno = curseg->segno;
 799
 800         /*
 801          * __add_sum_entry should be resided under the curseg_mutex
 802          * because, this function updates a summary entry in the
 803          * current summary block.
 804          */
 805         __add_sum_entry(sbi, type, sum);
 806
 807         mutex_lock(&sit_i->sentry_lock);
 808         __refresh_next_blkoff(sbi, curseg);
 809
 810         stat_inc_block_count(sbi, curseg);
 811
 812         /*
 813          * SIT information should be updated before segment allocation,
 814          * since SSR needs latest valid block information.
 815          */
 816         refresh_sit_entry(sbi, old_blkaddr, *new_blkaddr);
 817
 818         if (!__has_curseg_space(sbi, type))
 819                 sit_i->s_ops->allocate_segment(sbi, type, false);
 820
 821         locate_dirty_segment(sbi, old_cursegno);
 822         locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr));
 823         mutex_unlock(&sit_i->sentry_lock);
 824
 825         if (p_type == NODE)
 826                 fill_node_footer_blkaddr(page, NEXT_FREE_BLKADDR(sbi, curseg));
 827
 828         /* writeout dirty page into bdev */
 829         submit_write_page(sbi, page, *new_blkaddr, p_type);
 830
 831         mutex_unlock(&curseg->curseg_mutex);
 832 }
 833
 834 void write_meta_page(struct f2fs_sb_info *sbi, struct page *page)
 835 {
 836         set_page_writeback(page);
 837         submit_write_page(sbi, page, page->index, META);
 838 }
 839
 840 void write_node_page(struct f2fs_sb_info *sbi, struct page *page,
 841                 unsigned int nid, block_t old_blkaddr, block_t *new_blkaddr)
 842 {
 843         struct f2fs_summary sum;
 844         set_summary(&sum, nid, 0, 0);
 845         do_write_page(sbi, page, old_blkaddr, new_blkaddr, &sum, NODE);
 846 }
 847
 848 void write_data_page(struct inode *inode, struct page *page,
 849                 struct dnode_of_data *dn, block_t old_blkaddr,
 850                 block_t *new_blkaddr)
 851 {
 852         struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
 853         struct f2fs_summary sum;
 854         struct node_info ni;
 855
 856         f2fs_bug_on(old_blkaddr == NULL_ADDR);
 857         get_node_info(sbi, dn->nid, &ni);
 858         set_summary(&sum, dn->nid, dn->ofs_in_node, ni.version);
 859
 860         do_write_page(sbi, page, old_blkaddr,
 861                         new_blkaddr, &sum, DATA);
 862 }
 863
 864 void rewrite_data_page(struct f2fs_sb_info *sbi, struct page *page,
 865                                         block_t old_blk_addr)
 866 {
 867         submit_write_page(sbi, page, old_blk_addr, DATA);
 868 }
 869
 870 void recover_data_page(struct f2fs_sb_info *sbi,
 871                         struct page *page, struct f2fs_summary *sum,
 872                         block_t old_blkaddr, block_t new_blkaddr)
 873 {
 874         struct sit_info *sit_i = SIT_I(sbi);
 875         struct curseg_info *curseg;
 876         unsigned int segno, old_cursegno;
 877         struct seg_entry *se;
 878         int type;
 879
 880         segno = GET_SEGNO(sbi, new_blkaddr);
 881         se = get_seg_entry(sbi, segno);
 882         type = se->type;
 883
 884         if (se->valid_blocks == 0 && !IS_CURSEG(sbi, segno)) {
 885                 if (old_blkaddr == NULL_ADDR)
 886                         type = CURSEG_COLD_DATA;
 887                 else
 888                         type = CURSEG_WARM_DATA;
 889         }
 890         curseg = CURSEG_I(sbi, type);
 891
 892         mutex_lock(&curseg->curseg_mutex);
 893         mutex_lock(&sit_i->sentry_lock);
 894
 895         old_cursegno = curseg->segno;
 896
 897         /* change the current segment */
 898         if (segno != curseg->segno) {
 899                 curseg->next_segno = segno;
 900                 change_curseg(sbi, type, true);
 901         }
 902
 903         curseg->next_blkoff = GET_SEGOFF_FROM_SEG0(sbi, new_blkaddr) &
 904                                         (sbi->blocks_per_seg - 1);
 905         __add_sum_entry(sbi, type, sum);
 906
 907         refresh_sit_entry(sbi, old_blkaddr, new_blkaddr);
 908
 909         locate_dirty_segment(sbi, old_cursegno);
 910         locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr));
 911
 912         mutex_unlock(&sit_i->sentry_lock);
 913         mutex_unlock(&curseg->curseg_mutex);
 914 }
 915
 916 void rewrite_node_page(struct f2fs_sb_info *sbi,
 917                         struct page *page, struct f2fs_summary *sum,
 918                         block_t old_blkaddr, block_t new_blkaddr)
 919 {
 920         struct sit_info *sit_i = SIT_I(sbi);
 921         int type = CURSEG_WARM_NODE;
 922         struct curseg_info *curseg;
 923         unsigned int segno, old_cursegno;
 924         block_t next_blkaddr = next_blkaddr_of_node(page);
 925         unsigned int next_segno = GET_SEGNO(sbi, next_blkaddr);
 926
 927         curseg = CURSEG_I(sbi, type);
 928
 929         mutex_lock(&curseg->curseg_mutex);
 930         mutex_lock(&sit_i->sentry_lock);
 931
 932         segno = GET_SEGNO(sbi, new_blkaddr);
 933         old_cursegno = curseg->segno;
 934
 935         /* change the current segment */
 936         if (segno != curseg->segno) {
 937                 curseg->next_segno = segno;
 938                 change_curseg(sbi, type, true);
 939         }
 940         curseg->next_blkoff = GET_SEGOFF_FROM_SEG0(sbi, new_blkaddr) &
 941                                         (sbi->blocks_per_seg - 1);
 942         __add_sum_entry(sbi, type, sum);
 943
 944         /* change the current log to the next block addr in advance */
 945         if (next_segno != segno) {
 946                 curseg->next_segno = next_segno;
 947                 change_curseg(sbi, type, true);
 948         }
 949         curseg->next_blkoff = GET_SEGOFF_FROM_SEG0(sbi, next_blkaddr) &
 950                                         (sbi->blocks_per_seg - 1);
 951
 952         /* rewrite node page */
 953         set_page_writeback(page);
 954         submit_write_page(sbi, page, new_blkaddr, NODE);
 955         f2fs_submit_bio(sbi, NODE, true);
 956         refresh_sit_entry(sbi, old_blkaddr, new_blkaddr);
 957
 958         locate_dirty_segment(sbi, old_cursegno);
 959         locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr));
 960
 961         mutex_unlock(&sit_i->sentry_lock);
 962         mutex_unlock(&curseg->curseg_mutex);
 963 }
 964
 965 static int read_compacted_summaries(struct f2fs_sb_info *sbi)
 966 {
 967         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
 968         struct curseg_info *seg_i;
 969         unsigned char *kaddr;
 970         struct page *page;
 971         block_t start;
 972         int i, j, offset;
 973
 974         start = start_sum_block(sbi);
 975
 976         page = get_meta_page(sbi, start++);
 977         kaddr = (unsigned char *)page_address(page);
 978
 979         /* Step 1: restore nat cache */
 980         seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
 981         memcpy(&seg_i->sum_blk->n_nats, kaddr, SUM_JOURNAL_SIZE);
 982
 983         /* Step 2: restore sit cache */
 984         seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
 985         memcpy(&seg_i->sum_blk->n_sits, kaddr + SUM_JOURNAL_SIZE,
 986                                                 SUM_JOURNAL_SIZE);
 987         offset = 2 * SUM_JOURNAL_SIZE;
 988
 989         /* Step 3: restore summary entries */
 990         for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
 991                 unsigned short blk_off;
 992                 unsigned int segno;
 993
 994                 seg_i = CURSEG_I(sbi, i);
 995                 segno = le32_to_cpu(ckpt->cur_data_segno[i]);
 996                 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[i]);
 997                 seg_i->next_segno = segno;
 998                 reset_curseg(sbi, i, 0);
 999                 seg_i->alloc_type = ckpt->alloc_type[i];
1000                 seg_i->next_blkoff = blk_off;
1001
1002                 if (seg_i->alloc_type == SSR)
1003                         blk_off = sbi->blocks_per_seg;
1004
1005                 for (j = 0; j < blk_off; j++) {
1006                         struct f2fs_summary *s;
1007                         s = (struct f2fs_summary *)(kaddr + offset);
1008                         seg_i->sum_blk->entries[j] = *s;
1009                         offset += SUMMARY_SIZE;
1010                         if (offset + SUMMARY_SIZE <= PAGE_CACHE_SIZE -
1011                                                 SUM_FOOTER_SIZE)
1012                                 continue;
1013
1014                         f2fs_put_page(page, 1);
1015                         page = NULL;
1016
1017                         page = get_meta_page(sbi, start++);
1018                         kaddr = (unsigned char *)page_address(page);
1019                         offset = 0;
1020                 }
1021         }
1022         f2fs_put_page(page, 1);
1023         return 0;
1024 }
1025
1026 static int read_normal_summaries(struct f2fs_sb_info *sbi, int type)
1027 {
1028         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1029         struct f2fs_summary_block *sum;
1030         struct curseg_info *curseg;
1031         struct page *new;
1032         unsigned short blk_off;
1033         unsigned int segno = 0;
1034         block_t blk_addr = 0;
1035
1036         /* get segment number and block addr */
1037         if (IS_DATASEG(type)) {
1038                 segno = le32_to_cpu(ckpt->cur_data_segno[type]);
1039                 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[type -
1040                                                         CURSEG_HOT_DATA]);
1041                 if (is_set_ckpt_flags(ckpt, CP_UMOUNT_FLAG))
1042                         blk_addr = sum_blk_addr(sbi, NR_CURSEG_TYPE, type);
1043                 else
1044                         blk_addr = sum_blk_addr(sbi, NR_CURSEG_DATA_TYPE, type);
1045         } else {
1046                 segno = le32_to_cpu(ckpt->cur_node_segno[type -
1047                                                         CURSEG_HOT_NODE]);
1048                 blk_off = le16_to_cpu(ckpt->cur_node_blkoff[type -
1049                                                         CURSEG_HOT_NODE]);
1050                 if (is_set_ckpt_flags(ckpt, CP_UMOUNT_FLAG))
1051                         blk_addr = sum_blk_addr(sbi, NR_CURSEG_NODE_TYPE,
1052                                                         type - CURSEG_HOT_NODE);
1053                 else
1054                         blk_addr = GET_SUM_BLOCK(sbi, segno);
1055         }
1056
1057         new = get_meta_page(sbi, blk_addr);
1058         sum = (struct f2fs_summary_block *)page_address(new);
1059
1060         if (IS_NODESEG(type)) {
1061                 if (is_set_ckpt_flags(ckpt, CP_UMOUNT_FLAG)) {
1062                         struct f2fs_summary *ns = &sum->entries[0];
1063                         int i;
1064                         for (i = 0; i < sbi->blocks_per_seg; i++, ns++) {
1065                                 ns->version = 0;
1066                                 ns->ofs_in_node = 0;
1067                         }
1068                 } else {
1069                         if (restore_node_summary(sbi, segno, sum)) {
1070                                 f2fs_put_page(new, 1);
1071                                 return -EINVAL;
1072                         }
1073                 }
1074         }
1075
1076         /* set uncompleted segment to curseg */
1077         curseg = CURSEG_I(sbi, type);
1078         mutex_lock(&curseg->curseg_mutex);
1079         memcpy(curseg->sum_blk, sum, PAGE_CACHE_SIZE);
1080         curseg->next_segno = segno;
1081         reset_curseg(sbi, type, 0);
1082         curseg->alloc_type = ckpt->alloc_type[type];
1083         curseg->next_blkoff = blk_off;
1084         mutex_unlock(&curseg->curseg_mutex);
1085         f2fs_put_page(new, 1);
1086         return 0;
1087 }
1088
1089 static int restore_curseg_summaries(struct f2fs_sb_info *sbi)
1090 {
1091         int type = CURSEG_HOT_DATA;
1092
1093         if (is_set_ckpt_flags(F2FS_CKPT(sbi), CP_COMPACT_SUM_FLAG)) {
1094                 /* restore for compacted data summary */
1095                 if (read_compacted_summaries(sbi))
1096                         return -EINVAL;
1097                 type = CURSEG_HOT_NODE;
1098         }
1099
1100         for (; type <= CURSEG_COLD_NODE; type++)
1101                 if (read_normal_summaries(sbi, type))
1102                         return -EINVAL;
1103         return 0;
1104 }
1105
1106 static void write_compacted_summaries(struct f2fs_sb_info *sbi, block_t blkaddr)
1107 {
1108         struct page *page;
1109         unsigned char *kaddr;
1110         struct f2fs_summary *summary;
1111         struct curseg_info *seg_i;
1112         int written_size = 0;
1113         int i, j;
1114
1115         page = grab_meta_page(sbi, blkaddr++);
1116         kaddr = (unsigned char *)page_address(page);
1117
1118         /* Step 1: write nat cache */
1119         seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
1120         memcpy(kaddr, &seg_i->sum_blk->n_nats, SUM_JOURNAL_SIZE);
1121         written_size += SUM_JOURNAL_SIZE;
1122
1123         /* Step 2: write sit cache */
1124         seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
1125         memcpy(kaddr + written_size, &seg_i->sum_blk->n_sits,
1126                                                 SUM_JOURNAL_SIZE);
1127         written_size += SUM_JOURNAL_SIZE;
1128
1129         /* Step 3: write summary entries */
1130         for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
1131                 unsigned short blkoff;
1132                 seg_i = CURSEG_I(sbi, i);
1133                 if (sbi->ckpt->alloc_type[i] == SSR)
1134                         blkoff = sbi->blocks_per_seg;
1135                 else
1136                         blkoff = curseg_blkoff(sbi, i);
1137
1138                 for (j = 0; j < blkoff; j++) {
1139                         if (!page) {
1140                                 page = grab_meta_page(sbi, blkaddr++);
1141                                 kaddr = (unsigned char *)page_address(page);
1142                                 written_size = 0;
1143                         }
1144                         summary = (struct f2fs_summary *)(kaddr + written_size);
1145                         *summary = seg_i->sum_blk->entries[j];
1146                         written_size += SUMMARY_SIZE;
1147
1148                         if (written_size + SUMMARY_SIZE <= PAGE_CACHE_SIZE -
1149                                                         SUM_FOOTER_SIZE)
1150                                 continue;
1151
1152                         set_page_dirty(page);
1153                         f2fs_put_page(page, 1);
1154                         page = NULL;
1155                 }
1156         }
1157         if (page) {
1158                 set_page_dirty(page);
1159                 f2fs_put_page(page, 1);
1160         }
1161 }
1162
1163 static void write_normal_summaries(struct f2fs_sb_info *sbi,
1164                                         block_t blkaddr, int type)
1165 {
1166         int i, end;
1167         if (IS_DATASEG(type))
1168                 end = type + NR_CURSEG_DATA_TYPE;
1169         else
1170                 end = type + NR_CURSEG_NODE_TYPE;
1171
1172         for (i = type; i < end; i++) {
1173                 struct curseg_info *sum = CURSEG_I(sbi, i);
1174                 mutex_lock(&sum->curseg_mutex);
1175                 write_sum_page(sbi, sum->sum_blk, blkaddr + (i - type));
1176                 mutex_unlock(&sum->curseg_mutex);
1177         }
1178 }
1179
1180 void write_data_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
1181 {
1182         if (is_set_ckpt_flags(F2FS_CKPT(sbi), CP_COMPACT_SUM_FLAG))
1183                 write_compacted_summaries(sbi, start_blk);
1184         else
1185                 write_normal_summaries(sbi, start_blk, CURSEG_HOT_DATA);
1186 }
1187
1188 void write_node_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
1189 {
1190         if (is_set_ckpt_flags(F2FS_CKPT(sbi), CP_UMOUNT_FLAG))
1191                 write_normal_summaries(sbi, start_blk, CURSEG_HOT_NODE);
1192 }
1193
1194 int lookup_journal_in_cursum(struct f2fs_summary_block *sum, int type,
1195                                         unsigned int val, int alloc)
1196 {
1197         int i;
1198
1199         if (type == NAT_JOURNAL) {
1200                 for (i = 0; i < nats_in_cursum(sum); i++) {
1201                         if (le32_to_cpu(nid_in_journal(sum, i)) == val)
1202                                 return i;
1203                 }
1204                 if (alloc && nats_in_cursum(sum) < NAT_JOURNAL_ENTRIES)
1205                         return update_nats_in_cursum(sum, 1);
1206         } else if (type == SIT_JOURNAL) {
1207                 for (i = 0; i < sits_in_cursum(sum); i++)
1208                         if (le32_to_cpu(segno_in_journal(sum, i)) == val)
1209                                 return i;
1210                 if (alloc && sits_in_cursum(sum) < SIT_JOURNAL_ENTRIES)
1211                         return update_sits_in_cursum(sum, 1);
1212         }
1213         return -1;
1214 }
1215
1216 static struct page *get_current_sit_page(struct f2fs_sb_info *sbi,
1217                                         unsigned int segno)
1218 {
1219         struct sit_info *sit_i = SIT_I(sbi);
1220         unsigned int offset = SIT_BLOCK_OFFSET(sit_i, segno);
1221         block_t blk_addr = sit_i->sit_base_addr + offset;
1222
1223         check_seg_range(sbi, segno);
1224
1225         /* calculate sit block address */
1226         if (f2fs_test_bit(offset, sit_i->sit_bitmap))
1227                 blk_addr += sit_i->sit_blocks;
1228
1229         return get_meta_page(sbi, blk_addr);
1230 }
1231
1232 static struct page *get_next_sit_page(struct f2fs_sb_info *sbi,
1233                                         unsigned int start)
1234 {
1235         struct sit_info *sit_i = SIT_I(sbi);
1236         struct page *src_page, *dst_page;
1237         pgoff_t src_off, dst_off;
1238         void *src_addr, *dst_addr;
1239
1240         src_off = current_sit_addr(sbi, start);
1241         dst_off = next_sit_addr(sbi, src_off);
1242
1243         /* get current sit block page without lock */
1244         src_page = get_meta_page(sbi, src_off);
1245         dst_page = grab_meta_page(sbi, dst_off);
1246         f2fs_bug_on(PageDirty(src_page));
1247
1248         src_addr = page_address(src_page);
1249         dst_addr = page_address(dst_page);
1250         memcpy(dst_addr, src_addr, PAGE_CACHE_SIZE);
1251
1252         set_page_dirty(dst_page);
1253         f2fs_put_page(src_page, 1);
1254
1255         set_to_next_sit(sit_i, start);
1256
1257         return dst_page;
1258 }
1259
1260 static bool flush_sits_in_journal(struct f2fs_sb_info *sbi)
1261 {
1262         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
1263         struct f2fs_summary_block *sum = curseg->sum_blk;
1264         int i;
1265
1266         /*
1267          * If the journal area in the current summary is full of sit entries,
1268          * all the sit entries will be flushed. Otherwise the sit entries
1269          * are not able to replace with newly hot sit entries.
1270          */
1271         if (sits_in_cursum(sum) >= SIT_JOURNAL_ENTRIES) {
1272                 for (i = sits_in_cursum(sum) - 1; i >= 0; i--) {
1273                         unsigned int segno;
1274                         segno = le32_to_cpu(segno_in_journal(sum, i));
1275                         __mark_sit_entry_dirty(sbi, segno);
1276                 }
1277                 update_sits_in_cursum(sum, -sits_in_cursum(sum));
1278                 return true;
1279         }
1280         return false;
1281 }
1282
1283 /*
1284  * CP calls this function, which flushes SIT entries including sit_journal,
1285  * and moves prefree segs to free segs.
1286  */
1287 void flush_sit_entries(struct f2fs_sb_info *sbi)
1288 {
1289         struct sit_info *sit_i = SIT_I(sbi);
1290         unsigned long *bitmap = sit_i->dirty_sentries_bitmap;
1291         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
1292         struct f2fs_summary_block *sum = curseg->sum_blk;
1293         unsigned long nsegs = TOTAL_SEGS(sbi);
1294         struct page *page = NULL;
1295         struct f2fs_sit_block *raw_sit = NULL;
1296         unsigned int start = 0, end = 0;
1297         unsigned int segno = -1;
1298         bool flushed;
1299
1300         mutex_lock(&curseg->curseg_mutex);
1301         mutex_lock(&sit_i->sentry_lock);
1302
1303         /*
1304          * "flushed" indicates whether sit entries in journal are flushed
1305          * to the SIT area or not.
1306          */
1307         flushed = flush_sits_in_journal(sbi);
1308
1309         while ((segno = find_next_bit(bitmap, nsegs, segno + 1)) < nsegs) {
1310                 struct seg_entry *se = get_seg_entry(sbi, segno);
1311                 int sit_offset, offset;
1312
1313                 sit_offset = SIT_ENTRY_OFFSET(sit_i, segno);
1314
1315                 if (flushed)
1316                         goto to_sit_page;
1317
1318                 offset = lookup_journal_in_cursum(sum, SIT_JOURNAL, segno, 1);
1319                 if (offset >= 0) {
1320                         segno_in_journal(sum, offset) = cpu_to_le32(segno);
1321                         seg_info_to_raw_sit(se, &sit_in_journal(sum, offset));
1322                         goto flush_done;
1323                 }
1324 to_sit_page:
1325                 if (!page || (start > segno) || (segno > end)) {
1326                         if (page) {
1327                                 f2fs_put_page(page, 1);
1328                                 page = NULL;
1329                         }
1330
1331                         start = START_SEGNO(sit_i, segno);
1332                         end = start + SIT_ENTRY_PER_BLOCK - 1;
1333
1334                         /* read sit block that will be updated */
1335                         page = get_next_sit_page(sbi, start);
1336                         raw_sit = page_address(page);
1337                 }
1338
1339                 /* udpate entry in SIT block */
1340                 seg_info_to_raw_sit(se, &raw_sit->entries[sit_offset]);
1341 flush_done:
1342                 __clear_bit(segno, bitmap);
1343                 sit_i->dirty_sentries--;
1344         }
1345         mutex_unlock(&sit_i->sentry_lock);
1346         mutex_unlock(&curseg->curseg_mutex);
1347
1348         /* writeout last modified SIT block */
1349         f2fs_put_page(page, 1);
1350
1351         set_prefree_as_free_segments(sbi);
1352 }
1353
1354 static int build_sit_info(struct f2fs_sb_info *sbi)
1355 {
1356         struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
1357         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1358         struct sit_info *sit_i;
1359         unsigned int sit_segs, start;
1360         char *src_bitmap, *dst_bitmap;
1361         unsigned int bitmap_size;
1362
1363         /* allocate memory for SIT information */
1364         sit_i = kzalloc(sizeof(struct sit_info), GFP_KERNEL);
1365         if (!sit_i)
1366                 return -ENOMEM;
1367
1368         SM_I(sbi)->sit_info = sit_i;
1369
1370         sit_i->sentries = vzalloc(TOTAL_SEGS(sbi) * sizeof(struct seg_entry));
1371         if (!sit_i->sentries)
1372                 return -ENOMEM;
1373
1374         bitmap_size = f2fs_bitmap_size(TOTAL_SEGS(sbi));
1375         sit_i->dirty_sentries_bitmap = kzalloc(bitmap_size, GFP_KERNEL);
1376         if (!sit_i->dirty_sentries_bitmap)
1377                 return -ENOMEM;
1378
1379         for (start = 0; start < TOTAL_SEGS(sbi); start++) {
1380                 sit_i->sentries[start].cur_valid_map
1381                         = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
1382                 sit_i->sentries[start].ckpt_valid_map
1383                         = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
1384                 if (!sit_i->sentries[start].cur_valid_map
1385                                 || !sit_i->sentries[start].ckpt_valid_map)
1386                         return -ENOMEM;
1387         }
1388
1389         if (sbi->segs_per_sec > 1) {
1390                 sit_i->sec_entries = vzalloc(TOTAL_SECS(sbi) *
1391                                         sizeof(struct sec_entry));
1392                 if (!sit_i->sec_entries)
1393                         return -ENOMEM;
1394         }
1395
1396         /* get information related with SIT */
1397         sit_segs = le32_to_cpu(raw_super->segment_count_sit) >> 1;
1398
1399         /* setup SIT bitmap from ckeckpoint pack */
1400         bitmap_size = __bitmap_size(sbi, SIT_BITMAP);
1401         src_bitmap = __bitmap_ptr(sbi, SIT_BITMAP);
1402
1403         dst_bitmap = kmemdup(src_bitmap, bitmap_size, GFP_KERNEL);
1404         if (!dst_bitmap)
1405                 return -ENOMEM;
1406
1407         /* init SIT information */
1408         sit_i->s_ops = &default_salloc_ops;
1409
1410         sit_i->sit_base_addr = le32_to_cpu(raw_super->sit_blkaddr);
1411         sit_i->sit_blocks = sit_segs << sbi->log_blocks_per_seg;
1412         sit_i->written_valid_blocks = le64_to_cpu(ckpt->valid_block_count);
1413         sit_i->sit_bitmap = dst_bitmap;
1414         sit_i->bitmap_size = bitmap_size;
1415         sit_i->dirty_sentries = 0;
1416         sit_i->sents_per_block = SIT_ENTRY_PER_BLOCK;
1417         sit_i->elapsed_time = le64_to_cpu(sbi->ckpt->elapsed_time);
1418         sit_i->mounted_time = CURRENT_TIME_SEC.tv_sec;
1419         mutex_init(&sit_i->sentry_lock);
1420         return 0;
1421 }
1422
1423 static int build_free_segmap(struct f2fs_sb_info *sbi)
1424 {
1425         struct f2fs_sm_info *sm_info = SM_I(sbi);
1426         struct free_segmap_info *free_i;
1427         unsigned int bitmap_size, sec_bitmap_size;
1428
1429         /* allocate memory for free segmap information */
1430         free_i = kzalloc(sizeof(struct free_segmap_info), GFP_KERNEL);
1431         if (!free_i)
1432                 return -ENOMEM;
1433
1434         SM_I(sbi)->free_info = free_i;
1435
1436         bitmap_size = f2fs_bitmap_size(TOTAL_SEGS(sbi));
1437         free_i->free_segmap = kmalloc(bitmap_size, GFP_KERNEL);
1438         if (!free_i->free_segmap)
1439                 return -ENOMEM;
1440
1441         sec_bitmap_size = f2fs_bitmap_size(TOTAL_SECS(sbi));
1442         free_i->free_secmap = kmalloc(sec_bitmap_size, GFP_KERNEL);
1443         if (!free_i->free_secmap)
1444                 return -ENOMEM;
1445
1446         /* set all segments as dirty temporarily */
1447         memset(free_i->free_segmap, 0xff, bitmap_size);
1448         memset(free_i->free_secmap, 0xff, sec_bitmap_size);
1449
1450         /* init free segmap information */
1451         free_i->start_segno =
1452                 (unsigned int) GET_SEGNO_FROM_SEG0(sbi, sm_info->main_blkaddr);
1453         free_i->free_segments = 0;
1454         free_i->free_sections = 0;
1455         rwlock_init(&free_i->segmap_lock);
1456         return 0;
1457 }
1458
1459 static int build_curseg(struct f2fs_sb_info *sbi)
1460 {
1461         struct curseg_info *array;
1462         int i;
1463
1464         array = kzalloc(sizeof(*array) * NR_CURSEG_TYPE, GFP_KERNEL);
1465         if (!array)
1466                 return -ENOMEM;
1467
1468         SM_I(sbi)->curseg_array = array;
1469
1470         for (i = 0; i < NR_CURSEG_TYPE; i++) {
1471                 mutex_init(&array[i].curseg_mutex);
1472                 array[i].sum_blk = kzalloc(PAGE_CACHE_SIZE, GFP_KERNEL);
1473                 if (!array[i].sum_blk)
1474                         return -ENOMEM;
1475                 array[i].segno = NULL_SEGNO;
1476                 array[i].next_blkoff = 0;
1477         }
1478         return restore_curseg_summaries(sbi);
1479 }
1480
1481 static void build_sit_entries(struct f2fs_sb_info *sbi)
1482 {
1483         struct sit_info *sit_i = SIT_I(sbi);
1484         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
1485         struct f2fs_summary_block *sum = curseg->sum_blk;
1486         unsigned int start;
1487
1488         for (start = 0; start < TOTAL_SEGS(sbi); start++) {
1489                 struct seg_entry *se = &sit_i->sentries[start];
1490                 struct f2fs_sit_block *sit_blk;
1491                 struct f2fs_sit_entry sit;
1492                 struct page *page;
1493                 int i;
1494
1495                 mutex_lock(&curseg->curseg_mutex);
1496                 for (i = 0; i < sits_in_cursum(sum); i++) {
1497                         if (le32_to_cpu(segno_in_journal(sum, i)) == start) {
1498                                 sit = sit_in_journal(sum, i);
1499                                 mutex_unlock(&curseg->curseg_mutex);
1500                                 goto got_it;
1501                         }
1502                 }
1503                 mutex_unlock(&curseg->curseg_mutex);
1504                 page = get_current_sit_page(sbi, start);
1505                 sit_blk = (struct f2fs_sit_block *)page_address(page);
1506                 sit = sit_blk->entries[SIT_ENTRY_OFFSET(sit_i, start)];
1507                 f2fs_put_page(page, 1);
1508 got_it:
1509                 check_block_count(sbi, start, &sit);
1510                 seg_info_from_raw_sit(se, &sit);
1511                 if (sbi->segs_per_sec > 1) {
1512                         struct sec_entry *e = get_sec_entry(sbi, start);
1513                         e->valid_blocks += se->valid_blocks;
1514                 }
1515         }
1516 }
1517
1518 static void init_free_segmap(struct f2fs_sb_info *sbi)
1519 {
1520         unsigned int start;
1521         int type;
1522
1523         for (start = 0; start < TOTAL_SEGS(sbi); start++) {
1524                 struct seg_entry *sentry = get_seg_entry(sbi, start);
1525                 if (!sentry->valid_blocks)
1526                         __set_free(sbi, start);
1527         }
1528
1529         /* set use the current segments */
1530         for (type = CURSEG_HOT_DATA; type <= CURSEG_COLD_NODE; type++) {
1531                 struct curseg_info *curseg_t = CURSEG_I(sbi, type);
1532                 __set_test_and_inuse(sbi, curseg_t->segno);
1533         }
1534 }
1535
1536 static void init_dirty_segmap(struct f2fs_sb_info *sbi)
1537 {
1538         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
1539         struct free_segmap_info *free_i = FREE_I(sbi);
1540         unsigned int segno = 0, offset = 0, total_segs = TOTAL_SEGS(sbi);
1541         unsigned short valid_blocks;
1542
1543         while (1) {
1544                 /* find dirty segment based on free segmap */
1545                 segno = find_next_inuse(free_i, total_segs, offset);
1546                 if (segno >= total_segs)
1547                         break;
1548                 offset = segno + 1;
1549                 valid_blocks = get_valid_blocks(sbi, segno, 0);
1550                 if (valid_blocks >= sbi->blocks_per_seg || !valid_blocks)
1551                         continue;
1552                 mutex_lock(&dirty_i->seglist_lock);
1553                 __locate_dirty_segment(sbi, segno, DIRTY);
1554                 mutex_unlock(&dirty_i->seglist_lock);
1555         }
1556 }
1557
1558 static int init_victim_secmap(struct f2fs_sb_info *sbi)
1559 {
1560         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
1561         unsigned int bitmap_size = f2fs_bitmap_size(TOTAL_SECS(sbi));
1562
1563         dirty_i->victim_secmap = kzalloc(bitmap_size, GFP_KERNEL);
1564         if (!dirty_i->victim_secmap)
1565                 return -ENOMEM;
1566         return 0;
1567 }
1568
1569 static int build_dirty_segmap(struct f2fs_sb_info *sbi)
1570 {
1571         struct dirty_seglist_info *dirty_i;
1572         unsigned int bitmap_size, i;
1573
1574         /* allocate memory for dirty segments list information */
1575         dirty_i = kzalloc(sizeof(struct dirty_seglist_info), GFP_KERNEL);
1576         if (!dirty_i)
1577                 return -ENOMEM;
1578
1579         SM_I(sbi)->dirty_info = dirty_i;
1580         mutex_init(&dirty_i->seglist_lock);
1581
1582         bitmap_size = f2fs_bitmap_size(TOTAL_SEGS(sbi));
1583
1584         for (i = 0; i < NR_DIRTY_TYPE; i++) {
1585                 dirty_i->dirty_segmap[i] = kzalloc(bitmap_size, GFP_KERNEL);
1586                 if (!dirty_i->dirty_segmap[i])
1587                         return -ENOMEM;
1588         }
1589
1590         init_dirty_segmap(sbi);
1591         return init_victim_secmap(sbi);
1592 }
1593
1594 /*
1595  * Update min, max modified time for cost-benefit GC algorithm
1596  */
1597 static void init_min_max_mtime(struct f2fs_sb_info *sbi)
1598 {
1599         struct sit_info *sit_i = SIT_I(sbi);
1600         unsigned int segno;
1601
1602         mutex_lock(&sit_i->sentry_lock);
1603
1604         sit_i->min_mtime = LLONG_MAX;
1605
1606         for (segno = 0; segno < TOTAL_SEGS(sbi); segno += sbi->segs_per_sec) {
1607                 unsigned int i;
1608                 unsigned long long mtime = 0;
1609
1610                 for (i = 0; i < sbi->segs_per_sec; i++)
1611                         mtime += get_seg_entry(sbi, segno + i)->mtime;
1612
1613                 mtime = div_u64(mtime, sbi->segs_per_sec);
1614
1615                 if (sit_i->min_mtime > mtime)
1616                         sit_i->min_mtime = mtime;
1617         }
1618         sit_i->max_mtime = get_mtime(sbi);
1619         mutex_unlock(&sit_i->sentry_lock);
1620 }
1621
1622 int build_segment_manager(struct f2fs_sb_info *sbi)
1623 {
1624         struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
1625         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1626         struct f2fs_sm_info *sm_info;
1627         int err;
1628
1629         sm_info = kzalloc(sizeof(struct f2fs_sm_info), GFP_KERNEL);
1630         if (!sm_info)
1631                 return -ENOMEM;
1632
1633         /* init sm info */
1634         sbi->sm_info = sm_info;
1635         INIT_LIST_HEAD(&sm_info->wblist_head);
1636         spin_lock_init(&sm_info->wblist_lock);
1637         sm_info->seg0_blkaddr = le32_to_cpu(raw_super->segment0_blkaddr);
1638         sm_info->main_blkaddr = le32_to_cpu(raw_super->main_blkaddr);
1639         sm_info->segment_count = le32_to_cpu(raw_super->segment_count);
1640         sm_info->reserved_segments = le32_to_cpu(ckpt->rsvd_segment_count);
1641         sm_info->ovp_segments = le32_to_cpu(ckpt->overprov_segment_count);
1642         sm_info->main_segments = le32_to_cpu(raw_super->segment_count_main);
1643         sm_info->ssa_blkaddr = le32_to_cpu(raw_super->ssa_blkaddr);
1644         sm_info->rec_prefree_segments = DEF_RECLAIM_PREFREE_SEGMENTS;
1645
1646         err = build_sit_info(sbi);
1647         if (err)
1648                 return err;
1649         err = build_free_segmap(sbi);
1650         if (err)
1651                 return err;
1652         err = build_curseg(sbi);
1653         if (err)
1654                 return err;
1655
1656         /* reinit free segmap based on SIT */
1657         build_sit_entries(sbi);
1658
1659         init_free_segmap(sbi);
1660         err = build_dirty_segmap(sbi);
1661         if (err)
1662                 return err;
1663
1664         init_min_max_mtime(sbi);
1665         return 0;
1666 }
1667
1668 static void discard_dirty_segmap(struct f2fs_sb_info *sbi,
1669                 enum dirty_type dirty_type)
1670 {
1671         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
1672
1673         mutex_lock(&dirty_i->seglist_lock);
1674         kfree(dirty_i->dirty_segmap[dirty_type]);
1675         dirty_i->nr_dirty[dirty_type] = 0;
1676         mutex_unlock(&dirty_i->seglist_lock);
1677 }
1678
1679 static void destroy_victim_secmap(struct f2fs_sb_info *sbi)
1680 {
1681         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
1682         kfree(dirty_i->victim_secmap);
1683 }
1684
1685 static void destroy_dirty_segmap(struct f2fs_sb_info *sbi)
1686 {
1687         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
1688         int i;
1689
1690         if (!dirty_i)
1691                 return;
1692
1693         /* discard pre-free/dirty segments list */
1694         for (i = 0; i < NR_DIRTY_TYPE; i++)
1695                 discard_dirty_segmap(sbi, i);
1696
1697         destroy_victim_secmap(sbi);
1698         SM_I(sbi)->dirty_info = NULL;
1699         kfree(dirty_i);
1700 }
1701
1702 static void destroy_curseg(struct f2fs_sb_info *sbi)
1703 {
1704         struct curseg_info *array = SM_I(sbi)->curseg_array;
1705         int i;
1706
1707         if (!array)
1708                 return;
1709         SM_I(sbi)->curseg_array = NULL;
1710         for (i = 0; i < NR_CURSEG_TYPE; i++)
1711                 kfree(array[i].sum_blk);
1712         kfree(array);
1713 }
1714
1715 static void destroy_free_segmap(struct f2fs_sb_info *sbi)
1716 {
1717         struct free_segmap_info *free_i = SM_I(sbi)->free_info;
1718         if (!free_i)
1719                 return;
1720         SM_I(sbi)->free_info = NULL;
1721         kfree(free_i->free_segmap);
1722         kfree(free_i->free_secmap);
1723         kfree(free_i);
1724 }
1725
1726 static void destroy_sit_info(struct f2fs_sb_info *sbi)
1727 {
1728         struct sit_info *sit_i = SIT_I(sbi);
1729         unsigned int start;
1730
1731         if (!sit_i)
1732                 return;
1733
1734         if (sit_i->sentries) {
1735                 for (start = 0; start < TOTAL_SEGS(sbi); start++) {
1736                         kfree(sit_i->sentries[start].cur_valid_map);
1737                         kfree(sit_i->sentries[start].ckpt_valid_map);
1738                 }
1739         }
1740         vfree(sit_i->sentries);
1741         vfree(sit_i->sec_entries);
1742         kfree(sit_i->dirty_sentries_bitmap);
1743
1744         SM_I(sbi)->sit_info = NULL;
1745         kfree(sit_i->sit_bitmap);
1746         kfree(sit_i);
1747 }
1748
1749 void destroy_segment_manager(struct f2fs_sb_info *sbi)
1750 {
1751         struct f2fs_sm_info *sm_info = SM_I(sbi);
1752         if (!sm_info)
1753                 return;
1754         destroy_dirty_segmap(sbi);
1755         destroy_curseg(sbi);
1756         destroy_free_segmap(sbi);
1757         destroy_sit_info(sbi);
1758         sbi->sm_info = NULL;
1759         kfree(sm_info);
1760 }