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 static void __locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
  40                 enum dirty_type dirty_type)
  41 {
  42         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
  43
  44         /* need not be added */
  45         if (IS_CURSEG(sbi, segno))
  46                 return;
  47
  48         if (!test_and_set_bit(segno, dirty_i->dirty_segmap[dirty_type]))
  49                 dirty_i->nr_dirty[dirty_type]++;
  50
  51         if (dirty_type == DIRTY) {
  52                 struct seg_entry *sentry = get_seg_entry(sbi, segno);
  53                 enum dirty_type t = DIRTY_HOT_DATA;
  54
  55                 dirty_type = sentry->type;
  56
  57                 if (!test_and_set_bit(segno, dirty_i->dirty_segmap[dirty_type]))
  58                         dirty_i->nr_dirty[dirty_type]++;
  59
  60                 /* Only one bitmap should be set */
  61                 for (; t <= DIRTY_COLD_NODE; t++) {
  62                         if (t == dirty_type)
  63                                 continue;
  64                         if (test_and_clear_bit(segno, dirty_i->dirty_segmap[t]))
  65                                 dirty_i->nr_dirty[t]--;
  66                 }
  67         }
  68 }
  69
  70 static void __remove_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
  71                 enum dirty_type dirty_type)
  72 {
  73         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
  74
  75         if (test_and_clear_bit(segno, dirty_i->dirty_segmap[dirty_type]))
  76                 dirty_i->nr_dirty[dirty_type]--;
  77
  78         if (dirty_type == DIRTY) {
  79                 enum dirty_type t = DIRTY_HOT_DATA;
  80
  81                 /* clear all the bitmaps */
  82                 for (; t <= DIRTY_COLD_NODE; t++)
  83                         if (test_and_clear_bit(segno, dirty_i->dirty_segmap[t]))
  84                                 dirty_i->nr_dirty[t]--;
  85
  86                 if (get_valid_blocks(sbi, segno, sbi->segs_per_sec) == 0)
  87                         clear_bit(GET_SECNO(sbi, segno),
  88                                                 dirty_i->victim_secmap);
  89         }
  90 }
  91
  92 /*
  93  * Should not occur error such as -ENOMEM.
  94  * Adding dirty entry into seglist is not critical operation.
  95  * If a given segment is one of current working segments, it won't be added.
  96  */
  97 static void locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno)
  98 {
  99         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
 100         unsigned short valid_blocks;
 101
 102         if (segno == NULL_SEGNO || IS_CURSEG(sbi, segno))
 103                 return;
 104
 105         mutex_lock(&dirty_i->seglist_lock);
 106
 107         valid_blocks = get_valid_blocks(sbi, segno, 0);
 108
 109         if (valid_blocks == 0) {
 110                 __locate_dirty_segment(sbi, segno, PRE);
 111                 __remove_dirty_segment(sbi, segno, DIRTY);
 112         } else if (valid_blocks < sbi->blocks_per_seg) {
 113                 __locate_dirty_segment(sbi, segno, DIRTY);
 114         } else {
 115                 /* Recovery routine with SSR needs this */
 116                 __remove_dirty_segment(sbi, segno, DIRTY);
 117         }
 118
 119         mutex_unlock(&dirty_i->seglist_lock);
 120         return;
 121 }
 122
 123 /*
 124  * Should call clear_prefree_segments after checkpoint is done.
 125  */
 126 static void set_prefree_as_free_segments(struct f2fs_sb_info *sbi)
 127 {
 128         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
 129         unsigned int segno = -1;
 130         unsigned int total_segs = TOTAL_SEGS(sbi);
 131
 132         mutex_lock(&dirty_i->seglist_lock);
 133         while (1) {
 134                 segno = find_next_bit(dirty_i->dirty_segmap[PRE], total_segs,
 135                                 segno + 1);
 136                 if (segno >= total_segs)
 137                         break;
 138                 __set_test_and_free(sbi, segno);
 139         }
 140         mutex_unlock(&dirty_i->seglist_lock);
 141 }
 142
 143 void clear_prefree_segments(struct f2fs_sb_info *sbi)
 144 {
 145         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
 146         unsigned int segno = -1;
 147         unsigned int total_segs = TOTAL_SEGS(sbi);
 148
 149         mutex_lock(&dirty_i->seglist_lock);
 150         while (1) {
 151                 segno = find_next_bit(dirty_i->dirty_segmap[PRE], total_segs,
 152                                 segno + 1);
 153                 if (segno >= total_segs)
 154                         break;
 155
 156                 if (test_and_clear_bit(segno, dirty_i->dirty_segmap[PRE]))
 157                         dirty_i->nr_dirty[PRE]--;
 158
 159                 /* Let's use trim */
 160                 if (test_opt(sbi, DISCARD))
 161                         blkdev_issue_discard(sbi->sb->s_bdev,
 162                                         START_BLOCK(sbi, segno) <<
 163                                         sbi->log_sectors_per_block,
 164                                         1 << (sbi->log_sectors_per_block +
 165                                                 sbi->log_blocks_per_seg),
 166                                         GFP_NOFS, 0);
 167         }
 168         mutex_unlock(&dirty_i->seglist_lock);
 169 }
 170
 171 static void __mark_sit_entry_dirty(struct f2fs_sb_info *sbi, unsigned int segno)
 172 {
 173         struct sit_info *sit_i = SIT_I(sbi);
 174         if (!__test_and_set_bit(segno, sit_i->dirty_sentries_bitmap))
 175                 sit_i->dirty_sentries++;
 176 }
 177
 178 static void __set_sit_entry_type(struct f2fs_sb_info *sbi, int type,
 179                                         unsigned int segno, int modified)
 180 {
 181         struct seg_entry *se = get_seg_entry(sbi, segno);
 182         se->type = type;
 183         if (modified)
 184                 __mark_sit_entry_dirty(sbi, segno);
 185 }
 186
 187 static void update_sit_entry(struct f2fs_sb_info *sbi, block_t blkaddr, int del)
 188 {
 189         struct seg_entry *se;
 190         unsigned int segno, offset;
 191         long int new_vblocks;
 192
 193         segno = GET_SEGNO(sbi, blkaddr);
 194
 195         se = get_seg_entry(sbi, segno);
 196         new_vblocks = se->valid_blocks + del;
 197         offset = GET_SEGOFF_FROM_SEG0(sbi, blkaddr) & (sbi->blocks_per_seg - 1);
 198
 199         BUG_ON((new_vblocks >> (sizeof(unsigned short) << 3) ||
 200                                 (new_vblocks > sbi->blocks_per_seg)));
 201
 202         se->valid_blocks = new_vblocks;
 203         se->mtime = get_mtime(sbi);
 204         SIT_I(sbi)->max_mtime = se->mtime;
 205
 206         /* Update valid block bitmap */
 207         if (del > 0) {
 208                 if (f2fs_set_bit(offset, se->cur_valid_map))
 209                         BUG();
 210         } else {
 211                 if (!f2fs_clear_bit(offset, se->cur_valid_map))
 212                         BUG();
 213         }
 214         if (!f2fs_test_bit(offset, se->ckpt_valid_map))
 215                 se->ckpt_valid_blocks += del;
 216
 217         __mark_sit_entry_dirty(sbi, segno);
 218
 219         /* update total number of valid blocks to be written in ckpt area */
 220         SIT_I(sbi)->written_valid_blocks += del;
 221
 222         if (sbi->segs_per_sec > 1)
 223                 get_sec_entry(sbi, segno)->valid_blocks += del;
 224 }
 225
 226 static void refresh_sit_entry(struct f2fs_sb_info *sbi,
 227                         block_t old_blkaddr, block_t new_blkaddr)
 228 {
 229         update_sit_entry(sbi, new_blkaddr, 1);
 230         if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO)
 231                 update_sit_entry(sbi, old_blkaddr, -1);
 232 }
 233
 234 void invalidate_blocks(struct f2fs_sb_info *sbi, block_t addr)
 235 {
 236         unsigned int segno = GET_SEGNO(sbi, addr);
 237         struct sit_info *sit_i = SIT_I(sbi);
 238
 239         BUG_ON(addr == NULL_ADDR);
 240         if (addr == NEW_ADDR)
 241                 return;
 242
 243         /* add it into sit main buffer */
 244         mutex_lock(&sit_i->sentry_lock);
 245
 246         update_sit_entry(sbi, addr, -1);
 247
 248         /* add it into dirty seglist */
 249         locate_dirty_segment(sbi, segno);
 250
 251         mutex_unlock(&sit_i->sentry_lock);
 252 }
 253
 254 /*
 255  * This function should be resided under the curseg_mutex lock
 256  */
 257 static void __add_sum_entry(struct f2fs_sb_info *sbi, int type,
 258                                         struct f2fs_summary *sum)
 259 {
 260         struct curseg_info *curseg = CURSEG_I(sbi, type);
 261         void *addr = curseg->sum_blk;
 262         addr += curseg->next_blkoff * sizeof(struct f2fs_summary);
 263         memcpy(addr, sum, sizeof(struct f2fs_summary));
 264         return;
 265 }
 266
 267 /*
 268  * Calculate the number of current summary pages for writing
 269  */
 270 int npages_for_summary_flush(struct f2fs_sb_info *sbi)
 271 {
 272         int total_size_bytes = 0;
 273         int valid_sum_count = 0;
 274         int i, sum_space;
 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         total_size_bytes = valid_sum_count * (SUMMARY_SIZE + 1)
 284                         + sizeof(struct nat_journal) + 2
 285                         + sizeof(struct sit_journal) + 2;
 286         sum_space = PAGE_CACHE_SIZE - SUM_FOOTER_SIZE;
 287         if (total_size_bytes < sum_space)
 288                 return 1;
 289         else if (total_size_bytes < 2 * sum_space)
 290                 return 2;
 291         return 3;
 292 }
 293
 294 /*
 295  * Caller should put this summary page
 296  */
 297 struct page *get_sum_page(struct f2fs_sb_info *sbi, unsigned int segno)
 298 {
 299         return get_meta_page(sbi, GET_SUM_BLOCK(sbi, segno));
 300 }
 301
 302 static void write_sum_page(struct f2fs_sb_info *sbi,
 303                         struct f2fs_summary_block *sum_blk, block_t blk_addr)
 304 {
 305         struct page *page = grab_meta_page(sbi, blk_addr);
 306         void *kaddr = page_address(page);
 307         memcpy(kaddr, sum_blk, PAGE_CACHE_SIZE);
 308         set_page_dirty(page);
 309         f2fs_put_page(page, 1);
 310 }
 311
 312 static int is_next_segment_free(struct f2fs_sb_info *sbi, int type)
 313 {
 314         struct curseg_info *curseg = CURSEG_I(sbi, type);
 315         unsigned int segno = curseg->segno + 1;
 316         struct free_segmap_info *free_i = FREE_I(sbi);
 317
 318         if (segno < TOTAL_SEGS(sbi) && segno % sbi->segs_per_sec)
 319                 return !test_bit(segno, free_i->free_segmap);
 320         return 0;
 321 }
 322
 323 /*
 324  * Find a new segment from the free segments bitmap to right order
 325  * This function should be returned with success, otherwise BUG
 326  */
 327 static void get_new_segment(struct f2fs_sb_info *sbi,
 328                         unsigned int *newseg, bool new_sec, int dir)
 329 {
 330         struct free_segmap_info *free_i = FREE_I(sbi);
 331         unsigned int segno, secno, zoneno;
 332         unsigned int total_zones = TOTAL_SECS(sbi) / sbi->secs_per_zone;
 333         unsigned int hint = *newseg / sbi->segs_per_sec;
 334         unsigned int old_zoneno = GET_ZONENO_FROM_SEGNO(sbi, *newseg);
 335         unsigned int left_start = hint;
 336         bool init = true;
 337         int go_left = 0;
 338         int i;
 339
 340         write_lock(&free_i->segmap_lock);
 341
 342         if (!new_sec && ((*newseg + 1) % sbi->segs_per_sec)) {
 343                 segno = find_next_zero_bit(free_i->free_segmap,
 344                                         TOTAL_SEGS(sbi), *newseg + 1);
 345                 if (segno - *newseg < sbi->segs_per_sec -
 346                                         (*newseg % sbi->segs_per_sec))
 347                         goto got_it;
 348         }
 349 find_other_zone:
 350         secno = find_next_zero_bit(free_i->free_secmap, TOTAL_SECS(sbi), hint);
 351         if (secno >= TOTAL_SECS(sbi)) {
 352                 if (dir == ALLOC_RIGHT) {
 353                         secno = find_next_zero_bit(free_i->free_secmap,
 354                                                         TOTAL_SECS(sbi), 0);
 355                         BUG_ON(secno >= TOTAL_SECS(sbi));
 356                 } else {
 357                         go_left = 1;
 358                         left_start = hint - 1;
 359                 }
 360         }
 361         if (go_left == 0)
 362                 goto skip_left;
 363
 364         while (test_bit(left_start, free_i->free_secmap)) {
 365                 if (left_start > 0) {
 366                         left_start--;
 367                         continue;
 368                 }
 369                 left_start = find_next_zero_bit(free_i->free_secmap,
 370                                                         TOTAL_SECS(sbi), 0);
 371                 BUG_ON(left_start >= TOTAL_SECS(sbi));
 372                 break;
 373         }
 374         secno = left_start;
 375 skip_left:
 376         hint = secno;
 377         segno = secno * sbi->segs_per_sec;
 378         zoneno = secno / sbi->secs_per_zone;
 379
 380         /* give up on finding another zone */
 381         if (!init)
 382                 goto got_it;
 383         if (sbi->secs_per_zone == 1)
 384                 goto got_it;
 385         if (zoneno == old_zoneno)
 386                 goto got_it;
 387         if (dir == ALLOC_LEFT) {
 388                 if (!go_left && zoneno + 1 >= total_zones)
 389                         goto got_it;
 390                 if (go_left && zoneno == 0)
 391                         goto got_it;
 392         }
 393         for (i = 0; i < NR_CURSEG_TYPE; i++)
 394                 if (CURSEG_I(sbi, i)->zone == zoneno)
 395                         break;
 396
 397         if (i < NR_CURSEG_TYPE) {
 398                 /* zone is in user, try another */
 399                 if (go_left)
 400                         hint = zoneno * sbi->secs_per_zone - 1;
 401                 else if (zoneno + 1 >= total_zones)
 402                         hint = 0;
 403                 else
 404                         hint = (zoneno + 1) * sbi->secs_per_zone;
 405                 init = false;
 406                 goto find_other_zone;
 407         }
 408 got_it:
 409         /* set it as dirty segment in free segmap */
 410         BUG_ON(test_bit(segno, free_i->free_segmap));
 411         __set_inuse(sbi, segno);
 412         *newseg = segno;
 413         write_unlock(&free_i->segmap_lock);
 414 }
 415
 416 static void reset_curseg(struct f2fs_sb_info *sbi, int type, int modified)
 417 {
 418         struct curseg_info *curseg = CURSEG_I(sbi, type);
 419         struct summary_footer *sum_footer;
 420
 421         curseg->segno = curseg->next_segno;
 422         curseg->zone = GET_ZONENO_FROM_SEGNO(sbi, curseg->segno);
 423         curseg->next_blkoff = 0;
 424         curseg->next_segno = NULL_SEGNO;
 425
 426         sum_footer = &(curseg->sum_blk->footer);
 427         memset(sum_footer, 0, sizeof(struct summary_footer));
 428         if (IS_DATASEG(type))
 429                 SET_SUM_TYPE(sum_footer, SUM_TYPE_DATA);
 430         if (IS_NODESEG(type))
 431                 SET_SUM_TYPE(sum_footer, SUM_TYPE_NODE);
 432         __set_sit_entry_type(sbi, type, curseg->segno, modified);
 433 }
 434
 435 /*
 436  * Allocate a current working segment.
 437  * This function always allocates a free segment in LFS manner.
 438  */
 439 static void new_curseg(struct f2fs_sb_info *sbi, int type, bool new_sec)
 440 {
 441         struct curseg_info *curseg = CURSEG_I(sbi, type);
 442         unsigned int segno = curseg->segno;
 443         int dir = ALLOC_LEFT;
 444
 445         write_sum_page(sbi, curseg->sum_blk,
 446                                 GET_SUM_BLOCK(sbi, segno));
 447         if (type == CURSEG_WARM_DATA || type == CURSEG_COLD_DATA)
 448                 dir = ALLOC_RIGHT;
 449
 450         if (test_opt(sbi, NOHEAP))
 451                 dir = ALLOC_RIGHT;
 452
 453         get_new_segment(sbi, &segno, new_sec, dir);
 454         curseg->next_segno = segno;
 455         reset_curseg(sbi, type, 1);
 456         curseg->alloc_type = LFS;
 457 }
 458
 459 static void __next_free_blkoff(struct f2fs_sb_info *sbi,
 460                         struct curseg_info *seg, block_t start)
 461 {
 462         struct seg_entry *se = get_seg_entry(sbi, seg->segno);
 463         block_t ofs;
 464         for (ofs = start; ofs < sbi->blocks_per_seg; ofs++) {
 465                 if (!f2fs_test_bit(ofs, se->ckpt_valid_map)
 466                         && !f2fs_test_bit(ofs, se->cur_valid_map))
 467                         break;
 468         }
 469         seg->next_blkoff = ofs;
 470 }
 471
 472 /*
 473  * If a segment is written by LFS manner, next block offset is just obtained
 474  * by increasing the current block offset. However, if a segment is written by
 475  * SSR manner, next block offset obtained by calling __next_free_blkoff
 476  */
 477 static void __refresh_next_blkoff(struct f2fs_sb_info *sbi,
 478                                 struct curseg_info *seg)
 479 {
 480         if (seg->alloc_type == SSR)
 481                 __next_free_blkoff(sbi, seg, seg->next_blkoff + 1);
 482         else
 483                 seg->next_blkoff++;
 484 }
 485
 486 /*
 487  * This function always allocates a used segment (from dirty seglist) by SSR
 488  * manner, so it should recover the existing segment information of valid blocks
 489  */
 490 static void change_curseg(struct f2fs_sb_info *sbi, int type, bool reuse)
 491 {
 492         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
 493         struct curseg_info *curseg = CURSEG_I(sbi, type);
 494         unsigned int new_segno = curseg->next_segno;
 495         struct f2fs_summary_block *sum_node;
 496         struct page *sum_page;
 497
 498         write_sum_page(sbi, curseg->sum_blk,
 499                                 GET_SUM_BLOCK(sbi, curseg->segno));
 500         __set_test_and_inuse(sbi, new_segno);
 501
 502         mutex_lock(&dirty_i->seglist_lock);
 503         __remove_dirty_segment(sbi, new_segno, PRE);
 504         __remove_dirty_segment(sbi, new_segno, DIRTY);
 505         mutex_unlock(&dirty_i->seglist_lock);
 506
 507         reset_curseg(sbi, type, 1);
 508         curseg->alloc_type = SSR;
 509         __next_free_blkoff(sbi, curseg, 0);
 510
 511         if (reuse) {
 512                 sum_page = get_sum_page(sbi, new_segno);
 513                 sum_node = (struct f2fs_summary_block *)page_address(sum_page);
 514                 memcpy(curseg->sum_blk, sum_node, SUM_ENTRY_SIZE);
 515                 f2fs_put_page(sum_page, 1);
 516         }
 517 }
 518
 519 static int get_ssr_segment(struct f2fs_sb_info *sbi, int type)
 520 {
 521         struct curseg_info *curseg = CURSEG_I(sbi, type);
 522         const struct victim_selection *v_ops = DIRTY_I(sbi)->v_ops;
 523
 524         if (IS_NODESEG(type) || !has_not_enough_free_secs(sbi, 0))
 525                 return v_ops->get_victim(sbi,
 526                                 &(curseg)->next_segno, BG_GC, type, SSR);
 527
 528         /* For data segments, let's do SSR more intensively */
 529         for (; type >= CURSEG_HOT_DATA; type--)
 530                 if (v_ops->get_victim(sbi, &(curseg)->next_segno,
 531                                                 BG_GC, type, SSR))
 532                         return 1;
 533         return 0;
 534 }
 535
 536 /*
 537  * flush out current segment and replace it with new segment
 538  * This function should be returned with success, otherwise BUG
 539  */
 540 static void allocate_segment_by_default(struct f2fs_sb_info *sbi,
 541                                                 int type, bool force)
 542 {
 543         struct curseg_info *curseg = CURSEG_I(sbi, type);
 544
 545         if (force) {
 546                 new_curseg(sbi, type, true);
 547                 goto out;
 548         }
 549
 550         if (type == CURSEG_WARM_NODE)
 551                 new_curseg(sbi, type, false);
 552         else if (curseg->alloc_type == LFS && is_next_segment_free(sbi, type))
 553                 new_curseg(sbi, type, false);
 554         else if (need_SSR(sbi) && get_ssr_segment(sbi, type))
 555                 change_curseg(sbi, type, true);
 556         else
 557                 new_curseg(sbi, type, false);
 558 out:
 559 #ifdef CONFIG_F2FS_STAT_FS
 560         sbi->segment_count[curseg->alloc_type]++;
 561 #endif
 562         return;
 563 }
 564
 565 void allocate_new_segments(struct f2fs_sb_info *sbi)
 566 {
 567         struct curseg_info *curseg;
 568         unsigned int old_curseg;
 569         int i;
 570
 571         for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
 572                 curseg = CURSEG_I(sbi, i);
 573                 old_curseg = curseg->segno;
 574                 SIT_I(sbi)->s_ops->allocate_segment(sbi, i, true);
 575                 locate_dirty_segment(sbi, old_curseg);
 576         }
 577 }
 578
 579 static const struct segment_allocation default_salloc_ops = {
 580         .allocate_segment = allocate_segment_by_default,
 581 };
 582
 583 static void f2fs_end_io_write(struct bio *bio, int err)
 584 {
 585         const int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
 586         struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
 587         struct bio_private *p = bio->bi_private;
 588
 589         do {
 590                 struct page *page = bvec->bv_page;
 591
 592                 if (--bvec >= bio->bi_io_vec)
 593                         prefetchw(&bvec->bv_page->flags);
 594                 if (!uptodate) {
 595                         SetPageError(page);
 596                         if (page->mapping)
 597                                 set_bit(AS_EIO, &page->mapping->flags);
 598                         set_ckpt_flags(p->sbi->ckpt, CP_ERROR_FLAG);
 599                         p->sbi->sb->s_flags |= MS_RDONLY;
 600                 }
 601                 end_page_writeback(page);
 602                 dec_page_count(p->sbi, F2FS_WRITEBACK);
 603         } while (bvec >= bio->bi_io_vec);
 604
 605         if (p->is_sync)
 606                 complete(p->wait);
 607         kfree(p);
 608         bio_put(bio);
 609 }
 610
 611 struct bio *f2fs_bio_alloc(struct block_device *bdev, int npages)
 612 {
 613         struct bio *bio;
 614         struct bio_private *priv;
 615 retry:
 616         priv = kmalloc(sizeof(struct bio_private), GFP_NOFS);
 617         if (!priv) {
 618                 cond_resched();
 619                 goto retry;
 620         }
 621
 622         /* No failure on bio allocation */
 623         bio = bio_alloc(GFP_NOIO, npages);
 624         bio->bi_bdev = bdev;
 625         bio->bi_private = priv;
 626         return bio;
 627 }
 628
 629 static void do_submit_bio(struct f2fs_sb_info *sbi,
 630                                 enum page_type type, bool sync)
 631 {
 632         int rw = sync ? WRITE_SYNC : WRITE;
 633         enum page_type btype = type > META ? META : type;
 634
 635         if (type >= META_FLUSH)
 636                 rw = WRITE_FLUSH_FUA;
 637
 638         if (btype == META)
 639                 rw |= REQ_META;
 640
 641         if (sbi->bio[btype]) {
 642                 struct bio_private *p = sbi->bio[btype]->bi_private;
 643                 p->sbi = sbi;
 644                 sbi->bio[btype]->bi_end_io = f2fs_end_io_write;
 645
 646                 trace_f2fs_do_submit_bio(sbi->sb, btype, sync, sbi->bio[btype]);
 647
 648                 if (type == META_FLUSH) {
 649                         DECLARE_COMPLETION_ONSTACK(wait);
 650                         p->is_sync = true;
 651                         p->wait = &wait;
 652                         submit_bio(rw, sbi->bio[btype]);
 653                         wait_for_completion(&wait);
 654                 } else {
 655                         p->is_sync = false;
 656                         submit_bio(rw, sbi->bio[btype]);
 657                 }
 658                 sbi->bio[btype] = NULL;
 659         }
 660 }
 661
 662 void f2fs_submit_bio(struct f2fs_sb_info *sbi, enum page_type type, bool sync)
 663 {
 664         down_write(&sbi->bio_sem);
 665         do_submit_bio(sbi, type, sync);
 666         up_write(&sbi->bio_sem);
 667 }
 668
 669 static void submit_write_page(struct f2fs_sb_info *sbi, struct page *page,
 670                                 block_t blk_addr, enum page_type type)
 671 {
 672         struct block_device *bdev = sbi->sb->s_bdev;
 673
 674         verify_block_addr(sbi, blk_addr);
 675
 676         down_write(&sbi->bio_sem);
 677
 678         inc_page_count(sbi, F2FS_WRITEBACK);
 679
 680         if (sbi->bio[type] && sbi->last_block_in_bio[type] != blk_addr - 1)
 681                 do_submit_bio(sbi, type, false);
 682 alloc_new:
 683         if (sbi->bio[type] == NULL) {
 684                 sbi->bio[type] = f2fs_bio_alloc(bdev, max_hw_blocks(sbi));
 685                 sbi->bio[type]->bi_sector = SECTOR_FROM_BLOCK(sbi, blk_addr);
 686                 /*
 687                  * The end_io will be assigned at the sumbission phase.
 688                  * Until then, let bio_add_page() merge consecutive IOs as much
 689                  * as possible.
 690                  */
 691         }
 692
 693         if (bio_add_page(sbi->bio[type], page, PAGE_CACHE_SIZE, 0) <
 694                                                         PAGE_CACHE_SIZE) {
 695                 do_submit_bio(sbi, type, false);
 696                 goto alloc_new;
 697         }
 698
 699         sbi->last_block_in_bio[type] = blk_addr;
 700
 701         up_write(&sbi->bio_sem);
 702         trace_f2fs_submit_write_page(page, blk_addr, type);
 703 }
 704
 705 static bool __has_curseg_space(struct f2fs_sb_info *sbi, int type)
 706 {
 707         struct curseg_info *curseg = CURSEG_I(sbi, type);
 708         if (curseg->next_blkoff < sbi->blocks_per_seg)
 709                 return true;
 710         return false;
 711 }
 712
 713 static int __get_segment_type_2(struct page *page, enum page_type p_type)
 714 {
 715         if (p_type == DATA)
 716                 return CURSEG_HOT_DATA;
 717         else
 718                 return CURSEG_HOT_NODE;
 719 }
 720
 721 static int __get_segment_type_4(struct page *page, enum page_type p_type)
 722 {
 723         if (p_type == DATA) {
 724                 struct inode *inode = page->mapping->host;
 725
 726                 if (S_ISDIR(inode->i_mode))
 727                         return CURSEG_HOT_DATA;
 728                 else
 729                         return CURSEG_COLD_DATA;
 730         } else {
 731                 if (IS_DNODE(page) && !is_cold_node(page))
 732                         return CURSEG_HOT_NODE;
 733                 else
 734                         return CURSEG_COLD_NODE;
 735         }
 736 }
 737
 738 static int __get_segment_type_6(struct page *page, enum page_type p_type)
 739 {
 740         if (p_type == DATA) {
 741                 struct inode *inode = page->mapping->host;
 742
 743                 if (S_ISDIR(inode->i_mode))
 744                         return CURSEG_HOT_DATA;
 745                 else if (is_cold_data(page) || file_is_cold(inode))
 746                         return CURSEG_COLD_DATA;
 747                 else
 748                         return CURSEG_WARM_DATA;
 749         } else {
 750                 if (IS_DNODE(page))
 751                         return is_cold_node(page) ? CURSEG_WARM_NODE :
 752                                                 CURSEG_HOT_NODE;
 753                 else
 754                         return CURSEG_COLD_NODE;
 755         }
 756 }
 757
 758 static int __get_segment_type(struct page *page, enum page_type p_type)
 759 {
 760         struct f2fs_sb_info *sbi = F2FS_SB(page->mapping->host->i_sb);
 761         switch (sbi->active_logs) {
 762         case 2:
 763                 return __get_segment_type_2(page, p_type);
 764         case 4:
 765                 return __get_segment_type_4(page, p_type);
 766         }
 767         /* NR_CURSEG_TYPE(6) logs by default */
 768         BUG_ON(sbi->active_logs != NR_CURSEG_TYPE);
 769         return __get_segment_type_6(page, p_type);
 770 }
 771
 772 static void do_write_page(struct f2fs_sb_info *sbi, struct page *page,
 773                         block_t old_blkaddr, block_t *new_blkaddr,
 774                         struct f2fs_summary *sum, enum page_type p_type)
 775 {
 776         struct sit_info *sit_i = SIT_I(sbi);
 777         struct curseg_info *curseg;
 778         unsigned int old_cursegno;
 779         int type;
 780
 781         type = __get_segment_type(page, p_type);
 782         curseg = CURSEG_I(sbi, type);
 783
 784         mutex_lock(&curseg->curseg_mutex);
 785
 786         *new_blkaddr = NEXT_FREE_BLKADDR(sbi, curseg);
 787         old_cursegno = curseg->segno;
 788
 789         /*
 790          * __add_sum_entry should be resided under the curseg_mutex
 791          * because, this function updates a summary entry in the
 792          * current summary block.
 793          */
 794         __add_sum_entry(sbi, type, sum);
 795
 796         mutex_lock(&sit_i->sentry_lock);
 797         __refresh_next_blkoff(sbi, curseg);
 798 #ifdef CONFIG_F2FS_STAT_FS
 799         sbi->block_count[curseg->alloc_type]++;
 800 #endif
 801
 802         /*
 803          * SIT information should be updated before segment allocation,
 804          * since SSR needs latest valid block information.
 805          */
 806         refresh_sit_entry(sbi, old_blkaddr, *new_blkaddr);
 807
 808         if (!__has_curseg_space(sbi, type))
 809                 sit_i->s_ops->allocate_segment(sbi, type, false);
 810
 811         locate_dirty_segment(sbi, old_cursegno);
 812         locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr));
 813         mutex_unlock(&sit_i->sentry_lock);
 814
 815         if (p_type == NODE)
 816                 fill_node_footer_blkaddr(page, NEXT_FREE_BLKADDR(sbi, curseg));
 817
 818         /* writeout dirty page into bdev */
 819         submit_write_page(sbi, page, *new_blkaddr, p_type);
 820
 821         mutex_unlock(&curseg->curseg_mutex);
 822 }
 823
 824 void write_meta_page(struct f2fs_sb_info *sbi, struct page *page)
 825 {
 826         set_page_writeback(page);
 827         submit_write_page(sbi, page, page->index, META);
 828 }
 829
 830 void write_node_page(struct f2fs_sb_info *sbi, struct page *page,
 831                 unsigned int nid, block_t old_blkaddr, block_t *new_blkaddr)
 832 {
 833         struct f2fs_summary sum;
 834         set_summary(&sum, nid, 0, 0);
 835         do_write_page(sbi, page, old_blkaddr, new_blkaddr, &sum, NODE);
 836 }
 837
 838 void write_data_page(struct inode *inode, struct page *page,
 839                 struct dnode_of_data *dn, block_t old_blkaddr,
 840                 block_t *new_blkaddr)
 841 {
 842         struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
 843         struct f2fs_summary sum;
 844         struct node_info ni;
 845
 846         BUG_ON(old_blkaddr == NULL_ADDR);
 847         get_node_info(sbi, dn->nid, &ni);
 848         set_summary(&sum, dn->nid, dn->ofs_in_node, ni.version);
 849
 850         do_write_page(sbi, page, old_blkaddr,
 851                         new_blkaddr, &sum, DATA);
 852 }
 853
 854 void rewrite_data_page(struct f2fs_sb_info *sbi, struct page *page,
 855                                         block_t old_blk_addr)
 856 {
 857         submit_write_page(sbi, page, old_blk_addr, DATA);
 858 }
 859
 860 void recover_data_page(struct f2fs_sb_info *sbi,
 861                         struct page *page, struct f2fs_summary *sum,
 862                         block_t old_blkaddr, block_t new_blkaddr)
 863 {
 864         struct sit_info *sit_i = SIT_I(sbi);
 865         struct curseg_info *curseg;
 866         unsigned int segno, old_cursegno;
 867         struct seg_entry *se;
 868         int type;
 869
 870         segno = GET_SEGNO(sbi, new_blkaddr);
 871         se = get_seg_entry(sbi, segno);
 872         type = se->type;
 873
 874         if (se->valid_blocks == 0 && !IS_CURSEG(sbi, segno)) {
 875                 if (old_blkaddr == NULL_ADDR)
 876                         type = CURSEG_COLD_DATA;
 877                 else
 878                         type = CURSEG_WARM_DATA;
 879         }
 880         curseg = CURSEG_I(sbi, type);
 881
 882         mutex_lock(&curseg->curseg_mutex);
 883         mutex_lock(&sit_i->sentry_lock);
 884
 885         old_cursegno = curseg->segno;
 886
 887         /* change the current segment */
 888         if (segno != curseg->segno) {
 889                 curseg->next_segno = segno;
 890                 change_curseg(sbi, type, true);
 891         }
 892
 893         curseg->next_blkoff = GET_SEGOFF_FROM_SEG0(sbi, new_blkaddr) &
 894                                         (sbi->blocks_per_seg - 1);
 895         __add_sum_entry(sbi, type, sum);
 896
 897         refresh_sit_entry(sbi, old_blkaddr, new_blkaddr);
 898
 899         locate_dirty_segment(sbi, old_cursegno);
 900         locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr));
 901
 902         mutex_unlock(&sit_i->sentry_lock);
 903         mutex_unlock(&curseg->curseg_mutex);
 904 }
 905
 906 void rewrite_node_page(struct f2fs_sb_info *sbi,
 907                         struct page *page, struct f2fs_summary *sum,
 908                         block_t old_blkaddr, block_t new_blkaddr)
 909 {
 910         struct sit_info *sit_i = SIT_I(sbi);
 911         int type = CURSEG_WARM_NODE;
 912         struct curseg_info *curseg;
 913         unsigned int segno, old_cursegno;
 914         block_t next_blkaddr = next_blkaddr_of_node(page);
 915         unsigned int next_segno = GET_SEGNO(sbi, next_blkaddr);
 916
 917         curseg = CURSEG_I(sbi, type);
 918
 919         mutex_lock(&curseg->curseg_mutex);
 920         mutex_lock(&sit_i->sentry_lock);
 921
 922         segno = GET_SEGNO(sbi, new_blkaddr);
 923         old_cursegno = curseg->segno;
 924
 925         /* change the current segment */
 926         if (segno != curseg->segno) {
 927                 curseg->next_segno = segno;
 928                 change_curseg(sbi, type, true);
 929         }
 930         curseg->next_blkoff = GET_SEGOFF_FROM_SEG0(sbi, new_blkaddr) &
 931                                         (sbi->blocks_per_seg - 1);
 932         __add_sum_entry(sbi, type, sum);
 933
 934         /* change the current log to the next block addr in advance */
 935         if (next_segno != segno) {
 936                 curseg->next_segno = next_segno;
 937                 change_curseg(sbi, type, true);
 938         }
 939         curseg->next_blkoff = GET_SEGOFF_FROM_SEG0(sbi, next_blkaddr) &
 940                                         (sbi->blocks_per_seg - 1);
 941
 942         /* rewrite node page */
 943         set_page_writeback(page);
 944         submit_write_page(sbi, page, new_blkaddr, NODE);
 945         f2fs_submit_bio(sbi, NODE, true);
 946         refresh_sit_entry(sbi, old_blkaddr, new_blkaddr);
 947
 948         locate_dirty_segment(sbi, old_cursegno);
 949         locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr));
 950
 951         mutex_unlock(&sit_i->sentry_lock);
 952         mutex_unlock(&curseg->curseg_mutex);
 953 }
 954
 955 static int read_compacted_summaries(struct f2fs_sb_info *sbi)
 956 {
 957         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
 958         struct curseg_info *seg_i;
 959         unsigned char *kaddr;
 960         struct page *page;
 961         block_t start;
 962         int i, j, offset;
 963
 964         start = start_sum_block(sbi);
 965
 966         page = get_meta_page(sbi, start++);
 967         kaddr = (unsigned char *)page_address(page);
 968
 969         /* Step 1: restore nat cache */
 970         seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
 971         memcpy(&seg_i->sum_blk->n_nats, kaddr, SUM_JOURNAL_SIZE);
 972
 973         /* Step 2: restore sit cache */
 974         seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
 975         memcpy(&seg_i->sum_blk->n_sits, kaddr + SUM_JOURNAL_SIZE,
 976                                                 SUM_JOURNAL_SIZE);
 977         offset = 2 * SUM_JOURNAL_SIZE;
 978
 979         /* Step 3: restore summary entries */
 980         for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
 981                 unsigned short blk_off;
 982                 unsigned int segno;
 983
 984                 seg_i = CURSEG_I(sbi, i);
 985                 segno = le32_to_cpu(ckpt->cur_data_segno[i]);
 986                 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[i]);
 987                 seg_i->next_segno = segno;
 988                 reset_curseg(sbi, i, 0);
 989                 seg_i->alloc_type = ckpt->alloc_type[i];
 990                 seg_i->next_blkoff = blk_off;
 991
 992                 if (seg_i->alloc_type == SSR)
 993                         blk_off = sbi->blocks_per_seg;
 994
 995                 for (j = 0; j < blk_off; j++) {
 996                         struct f2fs_summary *s;
 997                         s = (struct f2fs_summary *)(kaddr + offset);
 998                         seg_i->sum_blk->entries[j] = *s;
 999                         offset += SUMMARY_SIZE;
1000                         if (offset + SUMMARY_SIZE <= PAGE_CACHE_SIZE -
1001                                                 SUM_FOOTER_SIZE)
1002                                 continue;
1003
1004                         f2fs_put_page(page, 1);
1005                         page = NULL;
1006
1007                         page = get_meta_page(sbi, start++);
1008                         kaddr = (unsigned char *)page_address(page);
1009                         offset = 0;
1010                 }
1011         }
1012         f2fs_put_page(page, 1);
1013         return 0;
1014 }
1015
1016 static int read_normal_summaries(struct f2fs_sb_info *sbi, int type)
1017 {
1018         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1019         struct f2fs_summary_block *sum;
1020         struct curseg_info *curseg;
1021         struct page *new;
1022         unsigned short blk_off;
1023         unsigned int segno = 0;
1024         block_t blk_addr = 0;
1025
1026         /* get segment number and block addr */
1027         if (IS_DATASEG(type)) {
1028                 segno = le32_to_cpu(ckpt->cur_data_segno[type]);
1029                 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[type -
1030                                                         CURSEG_HOT_DATA]);
1031                 if (is_set_ckpt_flags(ckpt, CP_UMOUNT_FLAG))
1032                         blk_addr = sum_blk_addr(sbi, NR_CURSEG_TYPE, type);
1033                 else
1034                         blk_addr = sum_blk_addr(sbi, NR_CURSEG_DATA_TYPE, type);
1035         } else {
1036                 segno = le32_to_cpu(ckpt->cur_node_segno[type -
1037                                                         CURSEG_HOT_NODE]);
1038                 blk_off = le16_to_cpu(ckpt->cur_node_blkoff[type -
1039                                                         CURSEG_HOT_NODE]);
1040                 if (is_set_ckpt_flags(ckpt, CP_UMOUNT_FLAG))
1041                         blk_addr = sum_blk_addr(sbi, NR_CURSEG_NODE_TYPE,
1042                                                         type - CURSEG_HOT_NODE);
1043                 else
1044                         blk_addr = GET_SUM_BLOCK(sbi, segno);
1045         }
1046
1047         new = get_meta_page(sbi, blk_addr);
1048         sum = (struct f2fs_summary_block *)page_address(new);
1049
1050         if (IS_NODESEG(type)) {
1051                 if (is_set_ckpt_flags(ckpt, CP_UMOUNT_FLAG)) {
1052                         struct f2fs_summary *ns = &sum->entries[0];
1053                         int i;
1054                         for (i = 0; i < sbi->blocks_per_seg; i++, ns++) {
1055                                 ns->version = 0;
1056                                 ns->ofs_in_node = 0;
1057                         }
1058                 } else {
1059                         if (restore_node_summary(sbi, segno, sum)) {
1060                                 f2fs_put_page(new, 1);
1061                                 return -EINVAL;
1062                         }
1063                 }
1064         }
1065
1066         /* set uncompleted segment to curseg */
1067         curseg = CURSEG_I(sbi, type);
1068         mutex_lock(&curseg->curseg_mutex);
1069         memcpy(curseg->sum_blk, sum, PAGE_CACHE_SIZE);
1070         curseg->next_segno = segno;
1071         reset_curseg(sbi, type, 0);
1072         curseg->alloc_type = ckpt->alloc_type[type];
1073         curseg->next_blkoff = blk_off;
1074         mutex_unlock(&curseg->curseg_mutex);
1075         f2fs_put_page(new, 1);
1076         return 0;
1077 }
1078
1079 static int restore_curseg_summaries(struct f2fs_sb_info *sbi)
1080 {
1081         int type = CURSEG_HOT_DATA;
1082
1083         if (is_set_ckpt_flags(F2FS_CKPT(sbi), CP_COMPACT_SUM_FLAG)) {
1084                 /* restore for compacted data summary */
1085                 if (read_compacted_summaries(sbi))
1086                         return -EINVAL;
1087                 type = CURSEG_HOT_NODE;
1088         }
1089
1090         for (; type <= CURSEG_COLD_NODE; type++)
1091                 if (read_normal_summaries(sbi, type))
1092                         return -EINVAL;
1093         return 0;
1094 }
1095
1096 static void write_compacted_summaries(struct f2fs_sb_info *sbi, block_t blkaddr)
1097 {
1098         struct page *page;
1099         unsigned char *kaddr;
1100         struct f2fs_summary *summary;
1101         struct curseg_info *seg_i;
1102         int written_size = 0;
1103         int i, j;
1104
1105         page = grab_meta_page(sbi, blkaddr++);
1106         kaddr = (unsigned char *)page_address(page);
1107
1108         /* Step 1: write nat cache */
1109         seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
1110         memcpy(kaddr, &seg_i->sum_blk->n_nats, SUM_JOURNAL_SIZE);
1111         written_size += SUM_JOURNAL_SIZE;
1112
1113         /* Step 2: write sit cache */
1114         seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
1115         memcpy(kaddr + written_size, &seg_i->sum_blk->n_sits,
1116                                                 SUM_JOURNAL_SIZE);
1117         written_size += SUM_JOURNAL_SIZE;
1118
1119         set_page_dirty(page);
1120
1121         /* Step 3: write summary entries */
1122         for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
1123                 unsigned short blkoff;
1124                 seg_i = CURSEG_I(sbi, i);
1125                 if (sbi->ckpt->alloc_type[i] == SSR)
1126                         blkoff = sbi->blocks_per_seg;
1127                 else
1128                         blkoff = curseg_blkoff(sbi, i);
1129
1130                 for (j = 0; j < blkoff; j++) {
1131                         if (!page) {
1132                                 page = grab_meta_page(sbi, blkaddr++);
1133                                 kaddr = (unsigned char *)page_address(page);
1134                                 written_size = 0;
1135                         }
1136                         summary = (struct f2fs_summary *)(kaddr + written_size);
1137                         *summary = seg_i->sum_blk->entries[j];
1138                         written_size += SUMMARY_SIZE;
1139                         set_page_dirty(page);
1140
1141                         if (written_size + SUMMARY_SIZE <= PAGE_CACHE_SIZE -
1142                                                         SUM_FOOTER_SIZE)
1143                                 continue;
1144
1145                         f2fs_put_page(page, 1);
1146                         page = NULL;
1147                 }
1148         }
1149         if (page)
1150                 f2fs_put_page(page, 1);
1151 }
1152
1153 static void write_normal_summaries(struct f2fs_sb_info *sbi,
1154                                         block_t blkaddr, int type)
1155 {
1156         int i, end;
1157         if (IS_DATASEG(type))
1158                 end = type + NR_CURSEG_DATA_TYPE;
1159         else
1160                 end = type + NR_CURSEG_NODE_TYPE;
1161
1162         for (i = type; i < end; i++) {
1163                 struct curseg_info *sum = CURSEG_I(sbi, i);
1164                 mutex_lock(&sum->curseg_mutex);
1165                 write_sum_page(sbi, sum->sum_blk, blkaddr + (i - type));
1166                 mutex_unlock(&sum->curseg_mutex);
1167         }
1168 }
1169
1170 void write_data_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
1171 {
1172         if (is_set_ckpt_flags(F2FS_CKPT(sbi), CP_COMPACT_SUM_FLAG))
1173                 write_compacted_summaries(sbi, start_blk);
1174         else
1175                 write_normal_summaries(sbi, start_blk, CURSEG_HOT_DATA);
1176 }
1177
1178 void write_node_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
1179 {
1180         if (is_set_ckpt_flags(F2FS_CKPT(sbi), CP_UMOUNT_FLAG))
1181                 write_normal_summaries(sbi, start_blk, CURSEG_HOT_NODE);
1182         return;
1183 }
1184
1185 int lookup_journal_in_cursum(struct f2fs_summary_block *sum, int type,
1186                                         unsigned int val, int alloc)
1187 {
1188         int i;
1189
1190         if (type == NAT_JOURNAL) {
1191                 for (i = 0; i < nats_in_cursum(sum); i++) {
1192                         if (le32_to_cpu(nid_in_journal(sum, i)) == val)
1193                                 return i;
1194                 }
1195                 if (alloc && nats_in_cursum(sum) < NAT_JOURNAL_ENTRIES)
1196                         return update_nats_in_cursum(sum, 1);
1197         } else if (type == SIT_JOURNAL) {
1198                 for (i = 0; i < sits_in_cursum(sum); i++)
1199                         if (le32_to_cpu(segno_in_journal(sum, i)) == val)
1200                                 return i;
1201                 if (alloc && sits_in_cursum(sum) < SIT_JOURNAL_ENTRIES)
1202                         return update_sits_in_cursum(sum, 1);
1203         }
1204         return -1;
1205 }
1206
1207 static struct page *get_current_sit_page(struct f2fs_sb_info *sbi,
1208                                         unsigned int segno)
1209 {
1210         struct sit_info *sit_i = SIT_I(sbi);
1211         unsigned int offset = SIT_BLOCK_OFFSET(sit_i, segno);
1212         block_t blk_addr = sit_i->sit_base_addr + offset;
1213
1214         check_seg_range(sbi, segno);
1215
1216         /* calculate sit block address */
1217         if (f2fs_test_bit(offset, sit_i->sit_bitmap))
1218                 blk_addr += sit_i->sit_blocks;
1219
1220         return get_meta_page(sbi, blk_addr);
1221 }
1222
1223 static struct page *get_next_sit_page(struct f2fs_sb_info *sbi,
1224                                         unsigned int start)
1225 {
1226         struct sit_info *sit_i = SIT_I(sbi);
1227         struct page *src_page, *dst_page;
1228         pgoff_t src_off, dst_off;
1229         void *src_addr, *dst_addr;
1230
1231         src_off = current_sit_addr(sbi, start);
1232         dst_off = next_sit_addr(sbi, src_off);
1233
1234         /* get current sit block page without lock */
1235         src_page = get_meta_page(sbi, src_off);
1236         dst_page = grab_meta_page(sbi, dst_off);
1237         BUG_ON(PageDirty(src_page));
1238
1239         src_addr = page_address(src_page);
1240         dst_addr = page_address(dst_page);
1241         memcpy(dst_addr, src_addr, PAGE_CACHE_SIZE);
1242
1243         set_page_dirty(dst_page);
1244         f2fs_put_page(src_page, 1);
1245
1246         set_to_next_sit(sit_i, start);
1247
1248         return dst_page;
1249 }
1250
1251 static bool flush_sits_in_journal(struct f2fs_sb_info *sbi)
1252 {
1253         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
1254         struct f2fs_summary_block *sum = curseg->sum_blk;
1255         int i;
1256
1257         /*
1258          * If the journal area in the current summary is full of sit entries,
1259          * all the sit entries will be flushed. Otherwise the sit entries
1260          * are not able to replace with newly hot sit entries.
1261          */
1262         if (sits_in_cursum(sum) >= SIT_JOURNAL_ENTRIES) {
1263                 for (i = sits_in_cursum(sum) - 1; i >= 0; i--) {
1264                         unsigned int segno;
1265                         segno = le32_to_cpu(segno_in_journal(sum, i));
1266                         __mark_sit_entry_dirty(sbi, segno);
1267                 }
1268                 update_sits_in_cursum(sum, -sits_in_cursum(sum));
1269                 return 1;
1270         }
1271         return 0;
1272 }
1273
1274 /*
1275  * CP calls this function, which flushes SIT entries including sit_journal,
1276  * and moves prefree segs to free segs.
1277  */
1278 void flush_sit_entries(struct f2fs_sb_info *sbi)
1279 {
1280         struct sit_info *sit_i = SIT_I(sbi);
1281         unsigned long *bitmap = sit_i->dirty_sentries_bitmap;
1282         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
1283         struct f2fs_summary_block *sum = curseg->sum_blk;
1284         unsigned long nsegs = TOTAL_SEGS(sbi);
1285         struct page *page = NULL;
1286         struct f2fs_sit_block *raw_sit = NULL;
1287         unsigned int start = 0, end = 0;
1288         unsigned int segno = -1;
1289         bool flushed;
1290
1291         mutex_lock(&curseg->curseg_mutex);
1292         mutex_lock(&sit_i->sentry_lock);
1293
1294         /*
1295          * "flushed" indicates whether sit entries in journal are flushed
1296          * to the SIT area or not.
1297          */
1298         flushed = flush_sits_in_journal(sbi);
1299
1300         while ((segno = find_next_bit(bitmap, nsegs, segno + 1)) < nsegs) {
1301                 struct seg_entry *se = get_seg_entry(sbi, segno);
1302                 int sit_offset, offset;
1303
1304                 sit_offset = SIT_ENTRY_OFFSET(sit_i, segno);
1305
1306                 if (flushed)
1307                         goto to_sit_page;
1308
1309                 offset = lookup_journal_in_cursum(sum, SIT_JOURNAL, segno, 1);
1310                 if (offset >= 0) {
1311                         segno_in_journal(sum, offset) = cpu_to_le32(segno);
1312                         seg_info_to_raw_sit(se, &sit_in_journal(sum, offset));
1313                         goto flush_done;
1314                 }
1315 to_sit_page:
1316                 if (!page || (start > segno) || (segno > end)) {
1317                         if (page) {
1318                                 f2fs_put_page(page, 1);
1319                                 page = NULL;
1320                         }
1321
1322                         start = START_SEGNO(sit_i, segno);
1323                         end = start + SIT_ENTRY_PER_BLOCK - 1;
1324
1325                         /* read sit block that will be updated */
1326                         page = get_next_sit_page(sbi, start);
1327                         raw_sit = page_address(page);
1328                 }
1329
1330                 /* udpate entry in SIT block */
1331                 seg_info_to_raw_sit(se, &raw_sit->entries[sit_offset]);
1332 flush_done:
1333                 __clear_bit(segno, bitmap);
1334                 sit_i->dirty_sentries--;
1335         }
1336         mutex_unlock(&sit_i->sentry_lock);
1337         mutex_unlock(&curseg->curseg_mutex);
1338
1339         /* writeout last modified SIT block */
1340         f2fs_put_page(page, 1);
1341
1342         set_prefree_as_free_segments(sbi);
1343 }
1344
1345 static int build_sit_info(struct f2fs_sb_info *sbi)
1346 {
1347         struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
1348         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1349         struct sit_info *sit_i;
1350         unsigned int sit_segs, start;
1351         char *src_bitmap, *dst_bitmap;
1352         unsigned int bitmap_size;
1353
1354         /* allocate memory for SIT information */
1355         sit_i = kzalloc(sizeof(struct sit_info), GFP_KERNEL);
1356         if (!sit_i)
1357                 return -ENOMEM;
1358
1359         SM_I(sbi)->sit_info = sit_i;
1360
1361         sit_i->sentries = vzalloc(TOTAL_SEGS(sbi) * sizeof(struct seg_entry));
1362         if (!sit_i->sentries)
1363                 return -ENOMEM;
1364
1365         bitmap_size = f2fs_bitmap_size(TOTAL_SEGS(sbi));
1366         sit_i->dirty_sentries_bitmap = kzalloc(bitmap_size, GFP_KERNEL);
1367         if (!sit_i->dirty_sentries_bitmap)
1368                 return -ENOMEM;
1369
1370         for (start = 0; start < TOTAL_SEGS(sbi); start++) {
1371                 sit_i->sentries[start].cur_valid_map
1372                         = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
1373                 sit_i->sentries[start].ckpt_valid_map
1374                         = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
1375                 if (!sit_i->sentries[start].cur_valid_map
1376                                 || !sit_i->sentries[start].ckpt_valid_map)
1377                         return -ENOMEM;
1378         }
1379
1380         if (sbi->segs_per_sec > 1) {
1381                 sit_i->sec_entries = vzalloc(TOTAL_SECS(sbi) *
1382                                         sizeof(struct sec_entry));
1383                 if (!sit_i->sec_entries)
1384                         return -ENOMEM;
1385         }
1386
1387         /* get information related with SIT */
1388         sit_segs = le32_to_cpu(raw_super->segment_count_sit) >> 1;
1389
1390         /* setup SIT bitmap from ckeckpoint pack */
1391         bitmap_size = __bitmap_size(sbi, SIT_BITMAP);
1392         src_bitmap = __bitmap_ptr(sbi, SIT_BITMAP);
1393
1394         dst_bitmap = kmemdup(src_bitmap, bitmap_size, GFP_KERNEL);
1395         if (!dst_bitmap)
1396                 return -ENOMEM;
1397
1398         /* init SIT information */
1399         sit_i->s_ops = &default_salloc_ops;
1400
1401         sit_i->sit_base_addr = le32_to_cpu(raw_super->sit_blkaddr);
1402         sit_i->sit_blocks = sit_segs << sbi->log_blocks_per_seg;
1403         sit_i->written_valid_blocks = le64_to_cpu(ckpt->valid_block_count);
1404         sit_i->sit_bitmap = dst_bitmap;
1405         sit_i->bitmap_size = bitmap_size;
1406         sit_i->dirty_sentries = 0;
1407         sit_i->sents_per_block = SIT_ENTRY_PER_BLOCK;
1408         sit_i->elapsed_time = le64_to_cpu(sbi->ckpt->elapsed_time);
1409         sit_i->mounted_time = CURRENT_TIME_SEC.tv_sec;
1410         mutex_init(&sit_i->sentry_lock);
1411         return 0;
1412 }
1413
1414 static int build_free_segmap(struct f2fs_sb_info *sbi)
1415 {
1416         struct f2fs_sm_info *sm_info = SM_I(sbi);
1417         struct free_segmap_info *free_i;
1418         unsigned int bitmap_size, sec_bitmap_size;
1419
1420         /* allocate memory for free segmap information */
1421         free_i = kzalloc(sizeof(struct free_segmap_info), GFP_KERNEL);
1422         if (!free_i)
1423                 return -ENOMEM;
1424
1425         SM_I(sbi)->free_info = free_i;
1426
1427         bitmap_size = f2fs_bitmap_size(TOTAL_SEGS(sbi));
1428         free_i->free_segmap = kmalloc(bitmap_size, GFP_KERNEL);
1429         if (!free_i->free_segmap)
1430                 return -ENOMEM;
1431
1432         sec_bitmap_size = f2fs_bitmap_size(TOTAL_SECS(sbi));
1433         free_i->free_secmap = kmalloc(sec_bitmap_size, GFP_KERNEL);
1434         if (!free_i->free_secmap)
1435                 return -ENOMEM;
1436
1437         /* set all segments as dirty temporarily */
1438         memset(free_i->free_segmap, 0xff, bitmap_size);
1439         memset(free_i->free_secmap, 0xff, sec_bitmap_size);
1440
1441         /* init free segmap information */
1442         free_i->start_segno =
1443                 (unsigned int) GET_SEGNO_FROM_SEG0(sbi, sm_info->main_blkaddr);
1444         free_i->free_segments = 0;
1445         free_i->free_sections = 0;
1446         rwlock_init(&free_i->segmap_lock);
1447         return 0;
1448 }
1449
1450 static int build_curseg(struct f2fs_sb_info *sbi)
1451 {
1452         struct curseg_info *array;
1453         int i;
1454
1455         array = kzalloc(sizeof(*array) * NR_CURSEG_TYPE, GFP_KERNEL);
1456         if (!array)
1457                 return -ENOMEM;
1458
1459         SM_I(sbi)->curseg_array = array;
1460
1461         for (i = 0; i < NR_CURSEG_TYPE; i++) {
1462                 mutex_init(&array[i].curseg_mutex);
1463                 array[i].sum_blk = kzalloc(PAGE_CACHE_SIZE, GFP_KERNEL);
1464                 if (!array[i].sum_blk)
1465                         return -ENOMEM;
1466                 array[i].segno = NULL_SEGNO;
1467                 array[i].next_blkoff = 0;
1468         }
1469         return restore_curseg_summaries(sbi);
1470 }
1471
1472 static void build_sit_entries(struct f2fs_sb_info *sbi)
1473 {
1474         struct sit_info *sit_i = SIT_I(sbi);
1475         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
1476         struct f2fs_summary_block *sum = curseg->sum_blk;
1477         unsigned int start;
1478
1479         for (start = 0; start < TOTAL_SEGS(sbi); start++) {
1480                 struct seg_entry *se = &sit_i->sentries[start];
1481                 struct f2fs_sit_block *sit_blk;
1482                 struct f2fs_sit_entry sit;
1483                 struct page *page;
1484                 int i;
1485
1486                 mutex_lock(&curseg->curseg_mutex);
1487                 for (i = 0; i < sits_in_cursum(sum); i++) {
1488                         if (le32_to_cpu(segno_in_journal(sum, i)) == start) {
1489                                 sit = sit_in_journal(sum, i);
1490                                 mutex_unlock(&curseg->curseg_mutex);
1491                                 goto got_it;
1492                         }
1493                 }
1494                 mutex_unlock(&curseg->curseg_mutex);
1495                 page = get_current_sit_page(sbi, start);
1496                 sit_blk = (struct f2fs_sit_block *)page_address(page);
1497                 sit = sit_blk->entries[SIT_ENTRY_OFFSET(sit_i, start)];
1498                 f2fs_put_page(page, 1);
1499 got_it:
1500                 check_block_count(sbi, start, &sit);
1501                 seg_info_from_raw_sit(se, &sit);
1502                 if (sbi->segs_per_sec > 1) {
1503                         struct sec_entry *e = get_sec_entry(sbi, start);
1504                         e->valid_blocks += se->valid_blocks;
1505                 }
1506         }
1507 }
1508
1509 static void init_free_segmap(struct f2fs_sb_info *sbi)
1510 {
1511         unsigned int start;
1512         int type;
1513
1514         for (start = 0; start < TOTAL_SEGS(sbi); start++) {
1515                 struct seg_entry *sentry = get_seg_entry(sbi, start);
1516                 if (!sentry->valid_blocks)
1517                         __set_free(sbi, start);
1518         }
1519
1520         /* set use the current segments */
1521         for (type = CURSEG_HOT_DATA; type <= CURSEG_COLD_NODE; type++) {
1522                 struct curseg_info *curseg_t = CURSEG_I(sbi, type);
1523                 __set_test_and_inuse(sbi, curseg_t->segno);
1524         }
1525 }
1526
1527 static void init_dirty_segmap(struct f2fs_sb_info *sbi)
1528 {
1529         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
1530         struct free_segmap_info *free_i = FREE_I(sbi);
1531         unsigned int segno = 0, offset = 0, total_segs = TOTAL_SEGS(sbi);
1532         unsigned short valid_blocks;
1533
1534         while (1) {
1535                 /* find dirty segment based on free segmap */
1536                 segno = find_next_inuse(free_i, total_segs, offset);
1537                 if (segno >= total_segs)
1538                         break;
1539                 offset = segno + 1;
1540                 valid_blocks = get_valid_blocks(sbi, segno, 0);
1541                 if (valid_blocks >= sbi->blocks_per_seg || !valid_blocks)
1542                         continue;
1543                 mutex_lock(&dirty_i->seglist_lock);
1544                 __locate_dirty_segment(sbi, segno, DIRTY);
1545                 mutex_unlock(&dirty_i->seglist_lock);
1546         }
1547 }
1548
1549 static int init_victim_secmap(struct f2fs_sb_info *sbi)
1550 {
1551         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
1552         unsigned int bitmap_size = f2fs_bitmap_size(TOTAL_SECS(sbi));
1553
1554         dirty_i->victim_secmap = kzalloc(bitmap_size, GFP_KERNEL);
1555         if (!dirty_i->victim_secmap)
1556                 return -ENOMEM;
1557         return 0;
1558 }
1559
1560 static int build_dirty_segmap(struct f2fs_sb_info *sbi)
1561 {
1562         struct dirty_seglist_info *dirty_i;
1563         unsigned int bitmap_size, i;
1564
1565         /* allocate memory for dirty segments list information */
1566         dirty_i = kzalloc(sizeof(struct dirty_seglist_info), GFP_KERNEL);
1567         if (!dirty_i)
1568                 return -ENOMEM;
1569
1570         SM_I(sbi)->dirty_info = dirty_i;
1571         mutex_init(&dirty_i->seglist_lock);
1572
1573         bitmap_size = f2fs_bitmap_size(TOTAL_SEGS(sbi));
1574
1575         for (i = 0; i < NR_DIRTY_TYPE; i++) {
1576                 dirty_i->dirty_segmap[i] = kzalloc(bitmap_size, GFP_KERNEL);
1577                 if (!dirty_i->dirty_segmap[i])
1578                         return -ENOMEM;
1579         }
1580
1581         init_dirty_segmap(sbi);
1582         return init_victim_secmap(sbi);
1583 }
1584
1585 /*
1586  * Update min, max modified time for cost-benefit GC algorithm
1587  */
1588 static void init_min_max_mtime(struct f2fs_sb_info *sbi)
1589 {
1590         struct sit_info *sit_i = SIT_I(sbi);
1591         unsigned int segno;
1592
1593         mutex_lock(&sit_i->sentry_lock);
1594
1595         sit_i->min_mtime = LLONG_MAX;
1596
1597         for (segno = 0; segno < TOTAL_SEGS(sbi); segno += sbi->segs_per_sec) {
1598                 unsigned int i;
1599                 unsigned long long mtime = 0;
1600
1601                 for (i = 0; i < sbi->segs_per_sec; i++)
1602                         mtime += get_seg_entry(sbi, segno + i)->mtime;
1603
1604                 mtime = div_u64(mtime, sbi->segs_per_sec);
1605
1606                 if (sit_i->min_mtime > mtime)
1607                         sit_i->min_mtime = mtime;
1608         }
1609         sit_i->max_mtime = get_mtime(sbi);
1610         mutex_unlock(&sit_i->sentry_lock);
1611 }
1612
1613 int build_segment_manager(struct f2fs_sb_info *sbi)
1614 {
1615         struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
1616         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1617         struct f2fs_sm_info *sm_info;
1618         int err;
1619
1620         sm_info = kzalloc(sizeof(struct f2fs_sm_info), GFP_KERNEL);
1621         if (!sm_info)
1622                 return -ENOMEM;
1623
1624         /* init sm info */
1625         sbi->sm_info = sm_info;
1626         INIT_LIST_HEAD(&sm_info->wblist_head);
1627         spin_lock_init(&sm_info->wblist_lock);
1628         sm_info->seg0_blkaddr = le32_to_cpu(raw_super->segment0_blkaddr);
1629         sm_info->main_blkaddr = le32_to_cpu(raw_super->main_blkaddr);
1630         sm_info->segment_count = le32_to_cpu(raw_super->segment_count);
1631         sm_info->reserved_segments = le32_to_cpu(ckpt->rsvd_segment_count);
1632         sm_info->ovp_segments = le32_to_cpu(ckpt->overprov_segment_count);
1633         sm_info->main_segments = le32_to_cpu(raw_super->segment_count_main);
1634         sm_info->ssa_blkaddr = le32_to_cpu(raw_super->ssa_blkaddr);
1635
1636         err = build_sit_info(sbi);
1637         if (err)
1638                 return err;
1639         err = build_free_segmap(sbi);
1640         if (err)
1641                 return err;
1642         err = build_curseg(sbi);
1643         if (err)
1644                 return err;
1645
1646         /* reinit free segmap based on SIT */
1647         build_sit_entries(sbi);
1648
1649         init_free_segmap(sbi);
1650         err = build_dirty_segmap(sbi);
1651         if (err)
1652                 return err;
1653
1654         init_min_max_mtime(sbi);
1655         return 0;
1656 }
1657
1658 static void discard_dirty_segmap(struct f2fs_sb_info *sbi,
1659                 enum dirty_type dirty_type)
1660 {
1661         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
1662
1663         mutex_lock(&dirty_i->seglist_lock);
1664         kfree(dirty_i->dirty_segmap[dirty_type]);
1665         dirty_i->nr_dirty[dirty_type] = 0;
1666         mutex_unlock(&dirty_i->seglist_lock);
1667 }
1668
1669 static void destroy_victim_secmap(struct f2fs_sb_info *sbi)
1670 {
1671         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
1672         kfree(dirty_i->victim_secmap);
1673 }
1674
1675 static void destroy_dirty_segmap(struct f2fs_sb_info *sbi)
1676 {
1677         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
1678         int i;
1679
1680         if (!dirty_i)
1681                 return;
1682
1683         /* discard pre-free/dirty segments list */
1684         for (i = 0; i < NR_DIRTY_TYPE; i++)
1685                 discard_dirty_segmap(sbi, i);
1686
1687         destroy_victim_secmap(sbi);
1688         SM_I(sbi)->dirty_info = NULL;
1689         kfree(dirty_i);
1690 }
1691
1692 static void destroy_curseg(struct f2fs_sb_info *sbi)
1693 {
1694         struct curseg_info *array = SM_I(sbi)->curseg_array;
1695         int i;
1696
1697         if (!array)
1698                 return;
1699         SM_I(sbi)->curseg_array = NULL;
1700         for (i = 0; i < NR_CURSEG_TYPE; i++)
1701                 kfree(array[i].sum_blk);
1702         kfree(array);
1703 }
1704
1705 static void destroy_free_segmap(struct f2fs_sb_info *sbi)
1706 {
1707         struct free_segmap_info *free_i = SM_I(sbi)->free_info;
1708         if (!free_i)
1709                 return;
1710         SM_I(sbi)->free_info = NULL;
1711         kfree(free_i->free_segmap);
1712         kfree(free_i->free_secmap);
1713         kfree(free_i);
1714 }
1715
1716 static void destroy_sit_info(struct f2fs_sb_info *sbi)
1717 {
1718         struct sit_info *sit_i = SIT_I(sbi);
1719         unsigned int start;
1720
1721         if (!sit_i)
1722                 return;
1723
1724         if (sit_i->sentries) {
1725                 for (start = 0; start < TOTAL_SEGS(sbi); start++) {
1726                         kfree(sit_i->sentries[start].cur_valid_map);
1727                         kfree(sit_i->sentries[start].ckpt_valid_map);
1728                 }
1729         }
1730         vfree(sit_i->sentries);
1731         vfree(sit_i->sec_entries);
1732         kfree(sit_i->dirty_sentries_bitmap);
1733
1734         SM_I(sbi)->sit_info = NULL;
1735         kfree(sit_i->sit_bitmap);
1736         kfree(sit_i);
1737 }
1738
1739 void destroy_segment_manager(struct f2fs_sb_info *sbi)
1740 {
1741         struct f2fs_sm_info *sm_info = SM_I(sbi);
1742         destroy_dirty_segmap(sbi);
1743         destroy_curseg(sbi);
1744         destroy_free_segmap(sbi);
1745         destroy_sit_info(sbi);
1746         sbi->sm_info = NULL;
1747         kfree(sm_info);
1748 }