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                 }
 604                 end_page_writeback(page);
 605                 dec_page_count(p->sbi, F2FS_WRITEBACK);
 606         } while (bvec >= bio->bi_io_vec);
 607
 608         if (p->is_sync)
 609                 complete(p->wait);
 610         kfree(p);
 611         bio_put(bio);
 612 }
 613
 614 struct bio *f2fs_bio_alloc(struct block_device *bdev, int npages)
 615 {
 616         struct bio *bio;
 617         struct bio_private *priv;
 618 retry:
 619         priv = kmalloc(sizeof(struct bio_private), GFP_NOFS);
 620         if (!priv) {
 621                 cond_resched();
 622                 goto retry;
 623         }
 624
 625         /* No failure on bio allocation */
 626         bio = bio_alloc(GFP_NOIO, npages);
 627         bio->bi_bdev = bdev;
 628         bio->bi_private = priv;
 629         return bio;
 630 }
 631
 632 static void do_submit_bio(struct f2fs_sb_info *sbi,
 633                                 enum page_type type, bool sync)
 634 {
 635         int rw = sync ? WRITE_SYNC : WRITE;
 636         enum page_type btype = type > META ? META : type;
 637
 638         if (type >= META_FLUSH)
 639                 rw = WRITE_FLUSH_FUA;
 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                 if (type == META_FLUSH) {
 646                         DECLARE_COMPLETION_ONSTACK(wait);
 647                         p->is_sync = true;
 648                         p->wait = &wait;
 649                         submit_bio(rw, sbi->bio[btype]);
 650                         wait_for_completion(&wait);
 651                 } else {
 652                         p->is_sync = false;
 653                         submit_bio(rw, sbi->bio[btype]);
 654                 }
 655                 sbi->bio[btype] = NULL;
 656         }
 657 }
 658
 659 void f2fs_submit_bio(struct f2fs_sb_info *sbi, enum page_type type, bool sync)
 660 {
 661         down_write(&sbi->bio_sem);
 662         do_submit_bio(sbi, type, sync);
 663         up_write(&sbi->bio_sem);
 664 }
 665
 666 static void submit_write_page(struct f2fs_sb_info *sbi, struct page *page,
 667                                 block_t blk_addr, enum page_type type)
 668 {
 669         struct block_device *bdev = sbi->sb->s_bdev;
 670
 671         verify_block_addr(sbi, blk_addr);
 672
 673         down_write(&sbi->bio_sem);
 674
 675         inc_page_count(sbi, F2FS_WRITEBACK);
 676
 677         if (sbi->bio[type] && sbi->last_block_in_bio[type] != blk_addr - 1)
 678                 do_submit_bio(sbi, type, false);
 679 alloc_new:
 680         if (sbi->bio[type] == NULL) {
 681                 sbi->bio[type] = f2fs_bio_alloc(bdev, bio_get_nr_vecs(bdev));
 682                 sbi->bio[type]->bi_sector = SECTOR_FROM_BLOCK(sbi, blk_addr);
 683                 /*
 684                  * The end_io will be assigned at the sumbission phase.
 685                  * Until then, let bio_add_page() merge consecutive IOs as much
 686                  * as possible.
 687                  */
 688         }
 689
 690         if (bio_add_page(sbi->bio[type], page, PAGE_CACHE_SIZE, 0) <
 691                                                         PAGE_CACHE_SIZE) {
 692                 do_submit_bio(sbi, type, false);
 693                 goto alloc_new;
 694         }
 695
 696         sbi->last_block_in_bio[type] = blk_addr;
 697
 698         up_write(&sbi->bio_sem);
 699 }
 700
 701 static bool __has_curseg_space(struct f2fs_sb_info *sbi, int type)
 702 {
 703         struct curseg_info *curseg = CURSEG_I(sbi, type);
 704         if (curseg->next_blkoff < sbi->blocks_per_seg)
 705                 return true;
 706         return false;
 707 }
 708
 709 static int __get_segment_type_2(struct page *page, enum page_type p_type)
 710 {
 711         if (p_type == DATA)
 712                 return CURSEG_HOT_DATA;
 713         else
 714                 return CURSEG_HOT_NODE;
 715 }
 716
 717 static int __get_segment_type_4(struct page *page, enum page_type p_type)
 718 {
 719         if (p_type == DATA) {
 720                 struct inode *inode = page->mapping->host;
 721
 722                 if (S_ISDIR(inode->i_mode))
 723                         return CURSEG_HOT_DATA;
 724                 else
 725                         return CURSEG_COLD_DATA;
 726         } else {
 727                 if (IS_DNODE(page) && !is_cold_node(page))
 728                         return CURSEG_HOT_NODE;
 729                 else
 730                         return CURSEG_COLD_NODE;
 731         }
 732 }
 733
 734 static int __get_segment_type_6(struct page *page, enum page_type p_type)
 735 {
 736         if (p_type == DATA) {
 737                 struct inode *inode = page->mapping->host;
 738
 739                 if (S_ISDIR(inode->i_mode))
 740                         return CURSEG_HOT_DATA;
 741                 else if (is_cold_data(page) || is_cold_file(inode))
 742                         return CURSEG_COLD_DATA;
 743                 else
 744                         return CURSEG_WARM_DATA;
 745         } else {
 746                 if (IS_DNODE(page))
 747                         return is_cold_node(page) ? CURSEG_WARM_NODE :
 748                                                 CURSEG_HOT_NODE;
 749                 else
 750                         return CURSEG_COLD_NODE;
 751         }
 752 }
 753
 754 static int __get_segment_type(struct page *page, enum page_type p_type)
 755 {
 756         struct f2fs_sb_info *sbi = F2FS_SB(page->mapping->host->i_sb);
 757         switch (sbi->active_logs) {
 758         case 2:
 759                 return __get_segment_type_2(page, p_type);
 760         case 4:
 761                 return __get_segment_type_4(page, p_type);
 762         }
 763         /* NR_CURSEG_TYPE(6) logs by default */
 764         BUG_ON(sbi->active_logs != NR_CURSEG_TYPE);
 765         return __get_segment_type_6(page, p_type);
 766 }
 767
 768 static void do_write_page(struct f2fs_sb_info *sbi, struct page *page,
 769                         block_t old_blkaddr, block_t *new_blkaddr,
 770                         struct f2fs_summary *sum, enum page_type p_type)
 771 {
 772         struct sit_info *sit_i = SIT_I(sbi);
 773         struct curseg_info *curseg;
 774         unsigned int old_cursegno;
 775         int type;
 776
 777         type = __get_segment_type(page, p_type);
 778         curseg = CURSEG_I(sbi, type);
 779
 780         mutex_lock(&curseg->curseg_mutex);
 781
 782         *new_blkaddr = NEXT_FREE_BLKADDR(sbi, curseg);
 783         old_cursegno = curseg->segno;
 784
 785         /*
 786          * __add_sum_entry should be resided under the curseg_mutex
 787          * because, this function updates a summary entry in the
 788          * current summary block.
 789          */
 790         __add_sum_entry(sbi, type, sum, curseg->next_blkoff);
 791
 792         mutex_lock(&sit_i->sentry_lock);
 793         __refresh_next_blkoff(sbi, curseg);
 794         sbi->block_count[curseg->alloc_type]++;
 795
 796         /*
 797          * SIT information should be updated before segment allocation,
 798          * since SSR needs latest valid block information.
 799          */
 800         refresh_sit_entry(sbi, old_blkaddr, *new_blkaddr);
 801
 802         if (!__has_curseg_space(sbi, type))
 803                 sit_i->s_ops->allocate_segment(sbi, type, false);
 804
 805         locate_dirty_segment(sbi, old_cursegno);
 806         locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr));
 807         mutex_unlock(&sit_i->sentry_lock);
 808
 809         if (p_type == NODE)
 810                 fill_node_footer_blkaddr(page, NEXT_FREE_BLKADDR(sbi, curseg));
 811
 812         /* writeout dirty page into bdev */
 813         submit_write_page(sbi, page, *new_blkaddr, p_type);
 814
 815         mutex_unlock(&curseg->curseg_mutex);
 816 }
 817
 818 int write_meta_page(struct f2fs_sb_info *sbi, struct page *page,
 819                         struct writeback_control *wbc)
 820 {
 821         if (wbc->for_reclaim)
 822                 return AOP_WRITEPAGE_ACTIVATE;
 823
 824         set_page_writeback(page);
 825         submit_write_page(sbi, page, page->index, META);
 826         return 0;
 827 }
 828
 829 void write_node_page(struct f2fs_sb_info *sbi, struct page *page,
 830                 unsigned int nid, block_t old_blkaddr, block_t *new_blkaddr)
 831 {
 832         struct f2fs_summary sum;
 833         set_summary(&sum, nid, 0, 0);
 834         do_write_page(sbi, page, old_blkaddr, new_blkaddr, &sum, NODE);
 835 }
 836
 837 void write_data_page(struct inode *inode, struct page *page,
 838                 struct dnode_of_data *dn, block_t old_blkaddr,
 839                 block_t *new_blkaddr)
 840 {
 841         struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
 842         struct f2fs_summary sum;
 843         struct node_info ni;
 844
 845         BUG_ON(old_blkaddr == NULL_ADDR);
 846         get_node_info(sbi, dn->nid, &ni);
 847         set_summary(&sum, dn->nid, dn->ofs_in_node, ni.version);
 848
 849         do_write_page(sbi, page, old_blkaddr,
 850                         new_blkaddr, &sum, DATA);
 851 }
 852
 853 void rewrite_data_page(struct f2fs_sb_info *sbi, struct page *page,
 854                                         block_t old_blk_addr)
 855 {
 856         submit_write_page(sbi, page, old_blk_addr, DATA);
 857 }
 858
 859 void recover_data_page(struct f2fs_sb_info *sbi,
 860                         struct page *page, struct f2fs_summary *sum,
 861                         block_t old_blkaddr, block_t new_blkaddr)
 862 {
 863         struct sit_info *sit_i = SIT_I(sbi);
 864         struct curseg_info *curseg;
 865         unsigned int segno, old_cursegno;
 866         struct seg_entry *se;
 867         int type;
 868
 869         segno = GET_SEGNO(sbi, new_blkaddr);
 870         se = get_seg_entry(sbi, segno);
 871         type = se->type;
 872
 873         if (se->valid_blocks == 0 && !IS_CURSEG(sbi, segno)) {
 874                 if (old_blkaddr == NULL_ADDR)
 875                         type = CURSEG_COLD_DATA;
 876                 else
 877                         type = CURSEG_WARM_DATA;
 878         }
 879         curseg = CURSEG_I(sbi, type);
 880
 881         mutex_lock(&curseg->curseg_mutex);
 882         mutex_lock(&sit_i->sentry_lock);
 883
 884         old_cursegno = curseg->segno;
 885
 886         /* change the current segment */
 887         if (segno != curseg->segno) {
 888                 curseg->next_segno = segno;
 889                 change_curseg(sbi, type, true);
 890         }
 891
 892         curseg->next_blkoff = GET_SEGOFF_FROM_SEG0(sbi, new_blkaddr) &
 893                                         (sbi->blocks_per_seg - 1);
 894         __add_sum_entry(sbi, type, sum, curseg->next_blkoff);
 895
 896         refresh_sit_entry(sbi, old_blkaddr, new_blkaddr);
 897
 898         locate_dirty_segment(sbi, old_cursegno);
 899         locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr));
 900
 901         mutex_unlock(&sit_i->sentry_lock);
 902         mutex_unlock(&curseg->curseg_mutex);
 903 }
 904
 905 void rewrite_node_page(struct f2fs_sb_info *sbi,
 906                         struct page *page, struct f2fs_summary *sum,
 907                         block_t old_blkaddr, block_t new_blkaddr)
 908 {
 909         struct sit_info *sit_i = SIT_I(sbi);
 910         int type = CURSEG_WARM_NODE;
 911         struct curseg_info *curseg;
 912         unsigned int segno, old_cursegno;
 913         block_t next_blkaddr = next_blkaddr_of_node(page);
 914         unsigned int next_segno = GET_SEGNO(sbi, next_blkaddr);
 915
 916         curseg = CURSEG_I(sbi, type);
 917
 918         mutex_lock(&curseg->curseg_mutex);
 919         mutex_lock(&sit_i->sentry_lock);
 920
 921         segno = GET_SEGNO(sbi, new_blkaddr);
 922         old_cursegno = curseg->segno;
 923
 924         /* change the current segment */
 925         if (segno != curseg->segno) {
 926                 curseg->next_segno = segno;
 927                 change_curseg(sbi, type, true);
 928         }
 929         curseg->next_blkoff = GET_SEGOFF_FROM_SEG0(sbi, new_blkaddr) &
 930                                         (sbi->blocks_per_seg - 1);
 931         __add_sum_entry(sbi, type, sum, curseg->next_blkoff);
 932
 933         /* change the current log to the next block addr in advance */
 934         if (next_segno != segno) {
 935                 curseg->next_segno = next_segno;
 936                 change_curseg(sbi, type, true);
 937         }
 938         curseg->next_blkoff = GET_SEGOFF_FROM_SEG0(sbi, next_blkaddr) &
 939                                         (sbi->blocks_per_seg - 1);
 940
 941         /* rewrite node page */
 942         set_page_writeback(page);
 943         submit_write_page(sbi, page, new_blkaddr, NODE);
 944         f2fs_submit_bio(sbi, NODE, true);
 945         refresh_sit_entry(sbi, old_blkaddr, new_blkaddr);
 946
 947         locate_dirty_segment(sbi, old_cursegno);
 948         locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr));
 949
 950         mutex_unlock(&sit_i->sentry_lock);
 951         mutex_unlock(&curseg->curseg_mutex);
 952 }
 953
 954 static int read_compacted_summaries(struct f2fs_sb_info *sbi)
 955 {
 956         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
 957         struct curseg_info *seg_i;
 958         unsigned char *kaddr;
 959         struct page *page;
 960         block_t start;
 961         int i, j, offset;
 962
 963         start = start_sum_block(sbi);
 964
 965         page = get_meta_page(sbi, start++);
 966         kaddr = (unsigned char *)page_address(page);
 967
 968         /* Step 1: restore nat cache */
 969         seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
 970         memcpy(&seg_i->sum_blk->n_nats, kaddr, SUM_JOURNAL_SIZE);
 971
 972         /* Step 2: restore sit cache */
 973         seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
 974         memcpy(&seg_i->sum_blk->n_sits, kaddr + SUM_JOURNAL_SIZE,
 975                                                 SUM_JOURNAL_SIZE);
 976         offset = 2 * SUM_JOURNAL_SIZE;
 977
 978         /* Step 3: restore summary entries */
 979         for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
 980                 unsigned short blk_off;
 981                 unsigned int segno;
 982
 983                 seg_i = CURSEG_I(sbi, i);
 984                 segno = le32_to_cpu(ckpt->cur_data_segno[i]);
 985                 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[i]);
 986                 seg_i->next_segno = segno;
 987                 reset_curseg(sbi, i, 0);
 988                 seg_i->alloc_type = ckpt->alloc_type[i];
 989                 seg_i->next_blkoff = blk_off;
 990
 991                 if (seg_i->alloc_type == SSR)
 992                         blk_off = sbi->blocks_per_seg;
 993
 994                 for (j = 0; j < blk_off; j++) {
 995                         struct f2fs_summary *s;
 996                         s = (struct f2fs_summary *)(kaddr + offset);
 997                         seg_i->sum_blk->entries[j] = *s;
 998                         offset += SUMMARY_SIZE;
 999                         if (offset + SUMMARY_SIZE <= PAGE_CACHE_SIZE -
1000                                                 SUM_FOOTER_SIZE)
1001                                 continue;
1002
1003                         f2fs_put_page(page, 1);
1004                         page = NULL;
1005
1006                         page = get_meta_page(sbi, start++);
1007                         kaddr = (unsigned char *)page_address(page);
1008                         offset = 0;
1009                 }
1010         }
1011         f2fs_put_page(page, 1);
1012         return 0;
1013 }
1014
1015 static int read_normal_summaries(struct f2fs_sb_info *sbi, int type)
1016 {
1017         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1018         struct f2fs_summary_block *sum;
1019         struct curseg_info *curseg;
1020         struct page *new;
1021         unsigned short blk_off;
1022         unsigned int segno = 0;
1023         block_t blk_addr = 0;
1024
1025         /* get segment number and block addr */
1026         if (IS_DATASEG(type)) {
1027                 segno = le32_to_cpu(ckpt->cur_data_segno[type]);
1028                 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[type -
1029                                                         CURSEG_HOT_DATA]);
1030                 if (is_set_ckpt_flags(ckpt, CP_UMOUNT_FLAG))
1031                         blk_addr = sum_blk_addr(sbi, NR_CURSEG_TYPE, type);
1032                 else
1033                         blk_addr = sum_blk_addr(sbi, NR_CURSEG_DATA_TYPE, type);
1034         } else {
1035                 segno = le32_to_cpu(ckpt->cur_node_segno[type -
1036                                                         CURSEG_HOT_NODE]);
1037                 blk_off = le16_to_cpu(ckpt->cur_node_blkoff[type -
1038                                                         CURSEG_HOT_NODE]);
1039                 if (is_set_ckpt_flags(ckpt, CP_UMOUNT_FLAG))
1040                         blk_addr = sum_blk_addr(sbi, NR_CURSEG_NODE_TYPE,
1041                                                         type - CURSEG_HOT_NODE);
1042                 else
1043                         blk_addr = GET_SUM_BLOCK(sbi, segno);
1044         }
1045
1046         new = get_meta_page(sbi, blk_addr);
1047         sum = (struct f2fs_summary_block *)page_address(new);
1048
1049         if (IS_NODESEG(type)) {
1050                 if (is_set_ckpt_flags(ckpt, CP_UMOUNT_FLAG)) {
1051                         struct f2fs_summary *ns = &sum->entries[0];
1052                         int i;
1053                         for (i = 0; i < sbi->blocks_per_seg; i++, ns++) {
1054                                 ns->version = 0;
1055                                 ns->ofs_in_node = 0;
1056                         }
1057                 } else {
1058                         if (restore_node_summary(sbi, segno, sum)) {
1059                                 f2fs_put_page(new, 1);
1060                                 return -EINVAL;
1061                         }
1062                 }
1063         }
1064
1065         /* set uncompleted segment to curseg */
1066         curseg = CURSEG_I(sbi, type);
1067         mutex_lock(&curseg->curseg_mutex);
1068         memcpy(curseg->sum_blk, sum, PAGE_CACHE_SIZE);
1069         curseg->next_segno = segno;
1070         reset_curseg(sbi, type, 0);
1071         curseg->alloc_type = ckpt->alloc_type[type];
1072         curseg->next_blkoff = blk_off;
1073         mutex_unlock(&curseg->curseg_mutex);
1074         f2fs_put_page(new, 1);
1075         return 0;
1076 }
1077
1078 static int restore_curseg_summaries(struct f2fs_sb_info *sbi)
1079 {
1080         int type = CURSEG_HOT_DATA;
1081
1082         if (is_set_ckpt_flags(F2FS_CKPT(sbi), CP_COMPACT_SUM_FLAG)) {
1083                 /* restore for compacted data summary */
1084                 if (read_compacted_summaries(sbi))
1085                         return -EINVAL;
1086                 type = CURSEG_HOT_NODE;
1087         }
1088
1089         for (; type <= CURSEG_COLD_NODE; type++)
1090                 if (read_normal_summaries(sbi, type))
1091                         return -EINVAL;
1092         return 0;
1093 }
1094
1095 static void write_compacted_summaries(struct f2fs_sb_info *sbi, block_t blkaddr)
1096 {
1097         struct page *page;
1098         unsigned char *kaddr;
1099         struct f2fs_summary *summary;
1100         struct curseg_info *seg_i;
1101         int written_size = 0;
1102         int i, j;
1103
1104         page = grab_meta_page(sbi, blkaddr++);
1105         kaddr = (unsigned char *)page_address(page);
1106
1107         /* Step 1: write nat cache */
1108         seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
1109         memcpy(kaddr, &seg_i->sum_blk->n_nats, SUM_JOURNAL_SIZE);
1110         written_size += SUM_JOURNAL_SIZE;
1111
1112         /* Step 2: write sit cache */
1113         seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
1114         memcpy(kaddr + written_size, &seg_i->sum_blk->n_sits,
1115                                                 SUM_JOURNAL_SIZE);
1116         written_size += SUM_JOURNAL_SIZE;
1117
1118         set_page_dirty(page);
1119
1120         /* Step 3: write summary entries */
1121         for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
1122                 unsigned short blkoff;
1123                 seg_i = CURSEG_I(sbi, i);
1124                 if (sbi->ckpt->alloc_type[i] == SSR)
1125                         blkoff = sbi->blocks_per_seg;
1126                 else
1127                         blkoff = curseg_blkoff(sbi, i);
1128
1129                 for (j = 0; j < blkoff; j++) {
1130                         if (!page) {
1131                                 page = grab_meta_page(sbi, blkaddr++);
1132                                 kaddr = (unsigned char *)page_address(page);
1133                                 written_size = 0;
1134                         }
1135                         summary = (struct f2fs_summary *)(kaddr + written_size);
1136                         *summary = seg_i->sum_blk->entries[j];
1137                         written_size += SUMMARY_SIZE;
1138                         set_page_dirty(page);
1139
1140                         if (written_size + SUMMARY_SIZE <= PAGE_CACHE_SIZE -
1141                                                         SUM_FOOTER_SIZE)
1142                                 continue;
1143
1144                         f2fs_put_page(page, 1);
1145                         page = NULL;
1146                 }
1147         }
1148         if (page)
1149                 f2fs_put_page(page, 1);
1150 }
1151
1152 static void write_normal_summaries(struct f2fs_sb_info *sbi,
1153                                         block_t blkaddr, int type)
1154 {
1155         int i, end;
1156         if (IS_DATASEG(type))
1157                 end = type + NR_CURSEG_DATA_TYPE;
1158         else
1159                 end = type + NR_CURSEG_NODE_TYPE;
1160
1161         for (i = type; i < end; i++) {
1162                 struct curseg_info *sum = CURSEG_I(sbi, i);
1163                 mutex_lock(&sum->curseg_mutex);
1164                 write_sum_page(sbi, sum->sum_blk, blkaddr + (i - type));
1165                 mutex_unlock(&sum->curseg_mutex);
1166         }
1167 }
1168
1169 void write_data_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
1170 {
1171         if (is_set_ckpt_flags(F2FS_CKPT(sbi), CP_COMPACT_SUM_FLAG))
1172                 write_compacted_summaries(sbi, start_blk);
1173         else
1174                 write_normal_summaries(sbi, start_blk, CURSEG_HOT_DATA);
1175 }
1176
1177 void write_node_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
1178 {
1179         if (is_set_ckpt_flags(F2FS_CKPT(sbi), CP_UMOUNT_FLAG))
1180                 write_normal_summaries(sbi, start_blk, CURSEG_HOT_NODE);
1181         return;
1182 }
1183
1184 int lookup_journal_in_cursum(struct f2fs_summary_block *sum, int type,
1185                                         unsigned int val, int alloc)
1186 {
1187         int i;
1188
1189         if (type == NAT_JOURNAL) {
1190                 for (i = 0; i < nats_in_cursum(sum); i++) {
1191                         if (le32_to_cpu(nid_in_journal(sum, i)) == val)
1192                                 return i;
1193                 }
1194                 if (alloc && nats_in_cursum(sum) < NAT_JOURNAL_ENTRIES)
1195                         return update_nats_in_cursum(sum, 1);
1196         } else if (type == SIT_JOURNAL) {
1197                 for (i = 0; i < sits_in_cursum(sum); i++)
1198                         if (le32_to_cpu(segno_in_journal(sum, i)) == val)
1199                                 return i;
1200                 if (alloc && sits_in_cursum(sum) < SIT_JOURNAL_ENTRIES)
1201                         return update_sits_in_cursum(sum, 1);
1202         }
1203         return -1;
1204 }
1205
1206 static struct page *get_current_sit_page(struct f2fs_sb_info *sbi,
1207                                         unsigned int segno)
1208 {
1209         struct sit_info *sit_i = SIT_I(sbi);
1210         unsigned int offset = SIT_BLOCK_OFFSET(sit_i, segno);
1211         block_t blk_addr = sit_i->sit_base_addr + offset;
1212
1213         check_seg_range(sbi, segno);
1214
1215         /* calculate sit block address */
1216         if (f2fs_test_bit(offset, sit_i->sit_bitmap))
1217                 blk_addr += sit_i->sit_blocks;
1218
1219         return get_meta_page(sbi, blk_addr);
1220 }
1221
1222 static struct page *get_next_sit_page(struct f2fs_sb_info *sbi,
1223                                         unsigned int start)
1224 {
1225         struct sit_info *sit_i = SIT_I(sbi);
1226         struct page *src_page, *dst_page;
1227         pgoff_t src_off, dst_off;
1228         void *src_addr, *dst_addr;
1229
1230         src_off = current_sit_addr(sbi, start);
1231         dst_off = next_sit_addr(sbi, src_off);
1232
1233         /* get current sit block page without lock */
1234         src_page = get_meta_page(sbi, src_off);
1235         dst_page = grab_meta_page(sbi, dst_off);
1236         BUG_ON(PageDirty(src_page));
1237
1238         src_addr = page_address(src_page);
1239         dst_addr = page_address(dst_page);
1240         memcpy(dst_addr, src_addr, PAGE_CACHE_SIZE);
1241
1242         set_page_dirty(dst_page);
1243         f2fs_put_page(src_page, 1);
1244
1245         set_to_next_sit(sit_i, start);
1246
1247         return dst_page;
1248 }
1249
1250 static bool flush_sits_in_journal(struct f2fs_sb_info *sbi)
1251 {
1252         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
1253         struct f2fs_summary_block *sum = curseg->sum_blk;
1254         int i;
1255
1256         /*
1257          * If the journal area in the current summary is full of sit entries,
1258          * all the sit entries will be flushed. Otherwise the sit entries
1259          * are not able to replace with newly hot sit entries.
1260          */
1261         if (sits_in_cursum(sum) >= SIT_JOURNAL_ENTRIES) {
1262                 for (i = sits_in_cursum(sum) - 1; i >= 0; i--) {
1263                         unsigned int segno;
1264                         segno = le32_to_cpu(segno_in_journal(sum, i));
1265                         __mark_sit_entry_dirty(sbi, segno);
1266                 }
1267                 update_sits_in_cursum(sum, -sits_in_cursum(sum));
1268                 return 1;
1269         }
1270         return 0;
1271 }
1272
1273 /*
1274  * CP calls this function, which flushes SIT entries including sit_journal,
1275  * and moves prefree segs to free segs.
1276  */
1277 void flush_sit_entries(struct f2fs_sb_info *sbi)
1278 {
1279         struct sit_info *sit_i = SIT_I(sbi);
1280         unsigned long *bitmap = sit_i->dirty_sentries_bitmap;
1281         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
1282         struct f2fs_summary_block *sum = curseg->sum_blk;
1283         unsigned long nsegs = TOTAL_SEGS(sbi);
1284         struct page *page = NULL;
1285         struct f2fs_sit_block *raw_sit = NULL;
1286         unsigned int start = 0, end = 0;
1287         unsigned int segno = -1;
1288         bool flushed;
1289
1290         mutex_lock(&curseg->curseg_mutex);
1291         mutex_lock(&sit_i->sentry_lock);
1292
1293         /*
1294          * "flushed" indicates whether sit entries in journal are flushed
1295          * to the SIT area or not.
1296          */
1297         flushed = flush_sits_in_journal(sbi);
1298
1299         while ((segno = find_next_bit(bitmap, nsegs, segno + 1)) < nsegs) {
1300                 struct seg_entry *se = get_seg_entry(sbi, segno);
1301                 int sit_offset, offset;
1302
1303                 sit_offset = SIT_ENTRY_OFFSET(sit_i, segno);
1304
1305                 if (flushed)
1306                         goto to_sit_page;
1307
1308                 offset = lookup_journal_in_cursum(sum, SIT_JOURNAL, segno, 1);
1309                 if (offset >= 0) {
1310                         segno_in_journal(sum, offset) = cpu_to_le32(segno);
1311                         seg_info_to_raw_sit(se, &sit_in_journal(sum, offset));
1312                         goto flush_done;
1313                 }
1314 to_sit_page:
1315                 if (!page || (start > segno) || (segno > end)) {
1316                         if (page) {
1317                                 f2fs_put_page(page, 1);
1318                                 page = NULL;
1319                         }
1320
1321                         start = START_SEGNO(sit_i, segno);
1322                         end = start + SIT_ENTRY_PER_BLOCK - 1;
1323
1324                         /* read sit block that will be updated */
1325                         page = get_next_sit_page(sbi, start);
1326                         raw_sit = page_address(page);
1327                 }
1328
1329                 /* udpate entry in SIT block */
1330                 seg_info_to_raw_sit(se, &raw_sit->entries[sit_offset]);
1331 flush_done:
1332                 __clear_bit(segno, bitmap);
1333                 sit_i->dirty_sentries--;
1334         }
1335         mutex_unlock(&sit_i->sentry_lock);
1336         mutex_unlock(&curseg->curseg_mutex);
1337
1338         /* writeout last modified SIT block */
1339         f2fs_put_page(page, 1);
1340
1341         set_prefree_as_free_segments(sbi);
1342 }
1343
1344 static int build_sit_info(struct f2fs_sb_info *sbi)
1345 {
1346         struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
1347         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1348         struct sit_info *sit_i;
1349         unsigned int sit_segs, start;
1350         char *src_bitmap, *dst_bitmap;
1351         unsigned int bitmap_size;
1352
1353         /* allocate memory for SIT information */
1354         sit_i = kzalloc(sizeof(struct sit_info), GFP_KERNEL);
1355         if (!sit_i)
1356                 return -ENOMEM;
1357
1358         SM_I(sbi)->sit_info = sit_i;
1359
1360         sit_i->sentries = vzalloc(TOTAL_SEGS(sbi) * sizeof(struct seg_entry));
1361         if (!sit_i->sentries)
1362                 return -ENOMEM;
1363
1364         bitmap_size = f2fs_bitmap_size(TOTAL_SEGS(sbi));
1365         sit_i->dirty_sentries_bitmap = kzalloc(bitmap_size, GFP_KERNEL);
1366         if (!sit_i->dirty_sentries_bitmap)
1367                 return -ENOMEM;
1368
1369         for (start = 0; start < TOTAL_SEGS(sbi); start++) {
1370                 sit_i->sentries[start].cur_valid_map
1371                         = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
1372                 sit_i->sentries[start].ckpt_valid_map
1373                         = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
1374                 if (!sit_i->sentries[start].cur_valid_map
1375                                 || !sit_i->sentries[start].ckpt_valid_map)
1376                         return -ENOMEM;
1377         }
1378
1379         if (sbi->segs_per_sec > 1) {
1380                 sit_i->sec_entries = vzalloc(sbi->total_sections *
1381                                         sizeof(struct sec_entry));
1382                 if (!sit_i->sec_entries)
1383                         return -ENOMEM;
1384         }
1385
1386         /* get information related with SIT */
1387         sit_segs = le32_to_cpu(raw_super->segment_count_sit) >> 1;
1388
1389         /* setup SIT bitmap from ckeckpoint pack */
1390         bitmap_size = __bitmap_size(sbi, SIT_BITMAP);
1391         src_bitmap = __bitmap_ptr(sbi, SIT_BITMAP);
1392
1393         dst_bitmap = kzalloc(bitmap_size, GFP_KERNEL);
1394         if (!dst_bitmap)
1395                 return -ENOMEM;
1396         memcpy(dst_bitmap, src_bitmap, bitmap_size);
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(sbi->total_sections);
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;
1532         unsigned short valid_blocks;
1533
1534         while (segno < TOTAL_SEGS(sbi)) {
1535                 /* find dirty segment based on free segmap */
1536                 segno = find_next_inuse(free_i, TOTAL_SEGS(sbi), offset);
1537                 if (segno >= TOTAL_SEGS(sbi))
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_segmap(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_SEGS(sbi));
1553
1554         dirty_i->victim_segmap[FG_GC] = kzalloc(bitmap_size, GFP_KERNEL);
1555         dirty_i->victim_segmap[BG_GC] = kzalloc(bitmap_size, GFP_KERNEL);
1556         if (!dirty_i->victim_segmap[FG_GC] || !dirty_i->victim_segmap[BG_GC])
1557                 return -ENOMEM;
1558         return 0;
1559 }
1560
1561 static int build_dirty_segmap(struct f2fs_sb_info *sbi)
1562 {
1563         struct dirty_seglist_info *dirty_i;
1564         unsigned int bitmap_size, i;
1565
1566         /* allocate memory for dirty segments list information */
1567         dirty_i = kzalloc(sizeof(struct dirty_seglist_info), GFP_KERNEL);
1568         if (!dirty_i)
1569                 return -ENOMEM;
1570
1571         SM_I(sbi)->dirty_info = dirty_i;
1572         mutex_init(&dirty_i->seglist_lock);
1573
1574         bitmap_size = f2fs_bitmap_size(TOTAL_SEGS(sbi));
1575
1576         for (i = 0; i < NR_DIRTY_TYPE; i++) {
1577                 dirty_i->dirty_segmap[i] = kzalloc(bitmap_size, GFP_KERNEL);
1578                 if (!dirty_i->dirty_segmap[i])
1579                         return -ENOMEM;
1580         }
1581
1582         init_dirty_segmap(sbi);
1583         return init_victim_segmap(sbi);
1584 }
1585
1586 /*
1587  * Update min, max modified time for cost-benefit GC algorithm
1588  */
1589 static void init_min_max_mtime(struct f2fs_sb_info *sbi)
1590 {
1591         struct sit_info *sit_i = SIT_I(sbi);
1592         unsigned int segno;
1593
1594         mutex_lock(&sit_i->sentry_lock);
1595
1596         sit_i->min_mtime = LLONG_MAX;
1597
1598         for (segno = 0; segno < TOTAL_SEGS(sbi); segno += sbi->segs_per_sec) {
1599                 unsigned int i;
1600                 unsigned long long mtime = 0;
1601
1602                 for (i = 0; i < sbi->segs_per_sec; i++)
1603                         mtime += get_seg_entry(sbi, segno + i)->mtime;
1604
1605                 mtime = div_u64(mtime, sbi->segs_per_sec);
1606
1607                 if (sit_i->min_mtime > mtime)
1608                         sit_i->min_mtime = mtime;
1609         }
1610         sit_i->max_mtime = get_mtime(sbi);
1611         mutex_unlock(&sit_i->sentry_lock);
1612 }
1613
1614 int build_segment_manager(struct f2fs_sb_info *sbi)
1615 {
1616         struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
1617         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1618         struct f2fs_sm_info *sm_info;
1619         int err;
1620
1621         sm_info = kzalloc(sizeof(struct f2fs_sm_info), GFP_KERNEL);
1622         if (!sm_info)
1623                 return -ENOMEM;
1624
1625         /* init sm info */
1626         sbi->sm_info = sm_info;
1627         INIT_LIST_HEAD(&sm_info->wblist_head);
1628         spin_lock_init(&sm_info->wblist_lock);
1629         sm_info->seg0_blkaddr = le32_to_cpu(raw_super->segment0_blkaddr);
1630         sm_info->main_blkaddr = le32_to_cpu(raw_super->main_blkaddr);
1631         sm_info->segment_count = le32_to_cpu(raw_super->segment_count);
1632         sm_info->reserved_segments = le32_to_cpu(ckpt->rsvd_segment_count);
1633         sm_info->ovp_segments = le32_to_cpu(ckpt->overprov_segment_count);
1634         sm_info->main_segments = le32_to_cpu(raw_super->segment_count_main);
1635         sm_info->ssa_blkaddr = le32_to_cpu(raw_super->ssa_blkaddr);
1636
1637         err = build_sit_info(sbi);
1638         if (err)
1639                 return err;
1640         err = build_free_segmap(sbi);
1641         if (err)
1642                 return err;
1643         err = build_curseg(sbi);
1644         if (err)
1645                 return err;
1646
1647         /* reinit free segmap based on SIT */
1648         build_sit_entries(sbi);
1649
1650         init_free_segmap(sbi);
1651         err = build_dirty_segmap(sbi);
1652         if (err)
1653                 return err;
1654
1655         init_min_max_mtime(sbi);
1656         return 0;
1657 }
1658
1659 static void discard_dirty_segmap(struct f2fs_sb_info *sbi,
1660                 enum dirty_type dirty_type)
1661 {
1662         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
1663
1664         mutex_lock(&dirty_i->seglist_lock);
1665         kfree(dirty_i->dirty_segmap[dirty_type]);
1666         dirty_i->nr_dirty[dirty_type] = 0;
1667         mutex_unlock(&dirty_i->seglist_lock);
1668 }
1669
1670 void reset_victim_segmap(struct f2fs_sb_info *sbi)
1671 {
1672         unsigned int bitmap_size = f2fs_bitmap_size(TOTAL_SEGS(sbi));
1673         memset(DIRTY_I(sbi)->victim_segmap[FG_GC], 0, bitmap_size);
1674 }
1675
1676 static void destroy_victim_segmap(struct f2fs_sb_info *sbi)
1677 {
1678         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
1679
1680         kfree(dirty_i->victim_segmap[FG_GC]);
1681         kfree(dirty_i->victim_segmap[BG_GC]);
1682 }
1683
1684 static void destroy_dirty_segmap(struct f2fs_sb_info *sbi)
1685 {
1686         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
1687         int i;
1688
1689         if (!dirty_i)
1690                 return;
1691
1692         /* discard pre-free/dirty segments list */
1693         for (i = 0; i < NR_DIRTY_TYPE; i++)
1694                 discard_dirty_segmap(sbi, i);
1695
1696         destroy_victim_segmap(sbi);
1697         SM_I(sbi)->dirty_info = NULL;
1698         kfree(dirty_i);
1699 }
1700
1701 static void destroy_curseg(struct f2fs_sb_info *sbi)
1702 {
1703         struct curseg_info *array = SM_I(sbi)->curseg_array;
1704         int i;
1705
1706         if (!array)
1707                 return;
1708         SM_I(sbi)->curseg_array = NULL;
1709         for (i = 0; i < NR_CURSEG_TYPE; i++)
1710                 kfree(array[i].sum_blk);
1711         kfree(array);
1712 }
1713
1714 static void destroy_free_segmap(struct f2fs_sb_info *sbi)
1715 {
1716         struct free_segmap_info *free_i = SM_I(sbi)->free_info;
1717         if (!free_i)
1718                 return;
1719         SM_I(sbi)->free_info = NULL;
1720         kfree(free_i->free_segmap);
1721         kfree(free_i->free_secmap);
1722         kfree(free_i);
1723 }
1724
1725 static void destroy_sit_info(struct f2fs_sb_info *sbi)
1726 {
1727         struct sit_info *sit_i = SIT_I(sbi);
1728         unsigned int start;
1729
1730         if (!sit_i)
1731                 return;
1732
1733         if (sit_i->sentries) {
1734                 for (start = 0; start < TOTAL_SEGS(sbi); start++) {
1735                         kfree(sit_i->sentries[start].cur_valid_map);
1736                         kfree(sit_i->sentries[start].ckpt_valid_map);
1737                 }
1738         }
1739         vfree(sit_i->sentries);
1740         vfree(sit_i->sec_entries);
1741         kfree(sit_i->dirty_sentries_bitmap);
1742
1743         SM_I(sbi)->sit_info = NULL;
1744         kfree(sit_i->sit_bitmap);
1745         kfree(sit_i);
1746 }
1747
1748 void destroy_segment_manager(struct f2fs_sb_info *sbi)
1749 {
1750         struct f2fs_sm_info *sm_info = SM_I(sbi);
1751         destroy_dirty_segmap(sbi);
1752         destroy_curseg(sbi);
1753         destroy_free_segmap(sbi);
1754         destroy_sit_info(sbi);
1755         sbi->sm_info = NULL;
1756         kfree(sm_info);
1757 }