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