fs/f2fs/segment.c

   1 /*
   2  * fs/f2fs/segment.c
   3  *
   4  * Copyright (c) 2012 Samsung Electronics Co., Ltd.
   5  *             http://www.samsung.com/
   6  *
   7  * This program is free software; you can redistribute it and/or modify
   8  * it under the terms of the GNU General Public License version 2 as
   9  * published by the Free Software Foundation.
  10  */
  11 #include <linux/fs.h>
  12 #include <linux/f2fs_fs.h>
  13 #include <linux/bio.h>
  14 #include <linux/blkdev.h>
  15 #include <linux/prefetch.h>
  16 #include <linux/vmalloc.h>
  17
  18 #include "f2fs.h"
  19 #include "segment.h"
  20 #include "node.h"
  21 #include <trace/events/f2fs.h>
  22
  23 /*
  24  * This function balances dirty node and dentry pages.
  25  * In addition, it controls garbage collection.
  26  */
  27 void f2fs_balance_fs(struct f2fs_sb_info *sbi)
  28 {
  29         /*
  30          * We should do GC or end up with checkpoint, if there are so many dirty
  31          * dir/node pages without enough free segments.
  32          */
  33         if (has_not_enough_free_secs(sbi, 0)) {
  34                 mutex_lock(&sbi->gc_mutex);
  35                 f2fs_gc(sbi);
  36         }
  37 }
  38
  39 void f2fs_balance_fs_bg(struct f2fs_sb_info *sbi)
  40 {
  41         /* check the # of cached NAT entries and prefree segments */
  42         if (try_to_free_nats(sbi, NAT_ENTRY_PER_BLOCK) ||
  43                                 excess_prefree_segs(sbi))
  44                 f2fs_sync_fs(sbi->sb, true);
  45 }
  46
  47 static void __locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
  48                 enum dirty_type dirty_type)
  49 {
  50         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
  51
  52         /* need not be added */
  53         if (IS_CURSEG(sbi, segno))
  54                 return;
  55
  56         if (!test_and_set_bit(segno, dirty_i->dirty_segmap[dirty_type]))
  57                 dirty_i->nr_dirty[dirty_type]++;
  58
  59         if (dirty_type == DIRTY) {
  60                 struct seg_entry *sentry = get_seg_entry(sbi, segno);
  61                 enum dirty_type t = sentry->type;
  62
  63                 if (!test_and_set_bit(segno, dirty_i->dirty_segmap[t]))
  64                         dirty_i->nr_dirty[t]++;
  65         }
  66 }
  67
  68 static void __remove_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
  69                 enum dirty_type dirty_type)
  70 {
  71         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
  72
  73         if (test_and_clear_bit(segno, dirty_i->dirty_segmap[dirty_type]))
  74                 dirty_i->nr_dirty[dirty_type]--;
  75
  76         if (dirty_type == DIRTY) {
  77                 struct seg_entry *sentry = get_seg_entry(sbi, segno);
  78                 enum dirty_type t = sentry->type;
  79
  80                 if (test_and_clear_bit(segno, dirty_i->dirty_segmap[t]))
  81                         dirty_i->nr_dirty[t]--;
  82
  83                 if (get_valid_blocks(sbi, segno, sbi->segs_per_sec) == 0)
  84                         clear_bit(GET_SECNO(sbi, segno),
  85                                                 dirty_i->victim_secmap);
  86         }
  87 }
  88
  89 /*
  90  * Should not occur error such as -ENOMEM.
  91  * Adding dirty entry into seglist is not critical operation.
  92  * If a given segment is one of current working segments, it won't be added.
  93  */
  94 static void locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno)
  95 {
  96         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
  97         unsigned short valid_blocks;
  98
  99         if (segno == NULL_SEGNO || IS_CURSEG(sbi, segno))
 100                 return;
 101
 102         mutex_lock(&dirty_i->seglist_lock);
 103
 104         valid_blocks = get_valid_blocks(sbi, segno, 0);
 105
 106         if (valid_blocks == 0) {
 107                 __locate_dirty_segment(sbi, segno, PRE);
 108                 __remove_dirty_segment(sbi, segno, DIRTY);
 109         } else if (valid_blocks < sbi->blocks_per_seg) {
 110                 __locate_dirty_segment(sbi, segno, DIRTY);
 111         } else {
 112                 /* Recovery routine with SSR needs this */
 113                 __remove_dirty_segment(sbi, segno, DIRTY);
 114         }
 115
 116         mutex_unlock(&dirty_i->seglist_lock);
 117 }
 118
 119 /*
 120  * Should call clear_prefree_segments after checkpoint is done.
 121  */
 122 static void set_prefree_as_free_segments(struct f2fs_sb_info *sbi)
 123 {
 124         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
 125         unsigned int segno = -1;
 126         unsigned int total_segs = TOTAL_SEGS(sbi);
 127
 128         mutex_lock(&dirty_i->seglist_lock);
 129         while (1) {
 130                 segno = find_next_bit(dirty_i->dirty_segmap[PRE], total_segs,
 131                                 segno + 1);
 132                 if (segno >= total_segs)
 133                         break;
 134                 __set_test_and_free(sbi, segno);
 135         }
 136         mutex_unlock(&dirty_i->seglist_lock);
 137 }
 138
 139 void clear_prefree_segments(struct f2fs_sb_info *sbi)
 140 {
 141         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
 142         unsigned int segno = -1;
 143         unsigned int total_segs = TOTAL_SEGS(sbi);
 144
 145         mutex_lock(&dirty_i->seglist_lock);
 146         while (1) {
 147                 segno = find_next_bit(dirty_i->dirty_segmap[PRE], total_segs,
 148                                 segno + 1);
 149                 if (segno >= total_segs)
 150                         break;
 151
 152                 if (test_and_clear_bit(segno, dirty_i->dirty_segmap[PRE]))
 153                         dirty_i->nr_dirty[PRE]--;
 154
 155                 /* Let's use trim */
 156                 if (test_opt(sbi, DISCARD))
 157                         blkdev_issue_discard(sbi->sb->s_bdev,
 158                                         START_BLOCK(sbi, segno) <<
 159                                         sbi->log_sectors_per_block,
 160                                         1 << (sbi->log_sectors_per_block +
 161                                                 sbi->log_blocks_per_seg),
 162                                         GFP_NOFS, 0);
 163         }
 164         mutex_unlock(&dirty_i->seglist_lock);
 165 }
 166
 167 static void __mark_sit_entry_dirty(struct f2fs_sb_info *sbi, unsigned int segno)
 168 {
 169         struct sit_info *sit_i = SIT_I(sbi);
 170         if (!__test_and_set_bit(segno, sit_i->dirty_sentries_bitmap))
 171                 sit_i->dirty_sentries++;
 172 }
 173
 174 static void __set_sit_entry_type(struct f2fs_sb_info *sbi, int type,
 175                                         unsigned int segno, int modified)
 176 {
 177         struct seg_entry *se = get_seg_entry(sbi, segno);
 178         se->type = type;
 179         if (modified)
 180                 __mark_sit_entry_dirty(sbi, segno);
 181 }
 182
 183 static void update_sit_entry(struct f2fs_sb_info *sbi, block_t blkaddr, int del)
 184 {
 185         struct seg_entry *se;
 186         unsigned int segno, offset;
 187         long int new_vblocks;
 188
 189         segno = GET_SEGNO(sbi, blkaddr);
 190
 191         se = get_seg_entry(sbi, segno);
 192         new_vblocks = se->valid_blocks + del;
 193         offset = GET_SEGOFF_FROM_SEG0(sbi, blkaddr) & (sbi->blocks_per_seg - 1);
 194
 195         f2fs_bug_on((new_vblocks >> (sizeof(unsigned short) << 3) ||
 196                                 (new_vblocks > sbi->blocks_per_seg)));
 197
 198         se->valid_blocks = new_vblocks;
 199         se->mtime = get_mtime(sbi);
 200         SIT_I(sbi)->max_mtime = se->mtime;
 201
 202         /* Update valid block bitmap */
 203         if (del > 0) {
 204                 if (f2fs_set_bit(offset, se->cur_valid_map))
 205                         BUG();
 206         } else {
 207                 if (!f2fs_clear_bit(offset, se->cur_valid_map))
 208                         BUG();
 209         }
 210         if (!f2fs_test_bit(offset, se->ckpt_valid_map))
 211                 se->ckpt_valid_blocks += del;
 212
 213         __mark_sit_entry_dirty(sbi, segno);
 214
 215         /* update total number of valid blocks to be written in ckpt area */
 216         SIT_I(sbi)->written_valid_blocks += del;
 217
 218         if (sbi->segs_per_sec > 1)
 219                 get_sec_entry(sbi, segno)->valid_blocks += del;
 220 }
 221
 222 static void refresh_sit_entry(struct f2fs_sb_info *sbi,
 223                         block_t old_blkaddr, block_t new_blkaddr)
 224 {
 225         update_sit_entry(sbi, new_blkaddr, 1);
 226         if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO)
 227                 update_sit_entry(sbi, old_blkaddr, -1);
 228 }
 229
 230 void invalidate_blocks(struct f2fs_sb_info *sbi, block_t addr)
 231 {
 232         unsigned int segno = GET_SEGNO(sbi, addr);
 233         struct sit_info *sit_i = SIT_I(sbi);
 234
 235         f2fs_bug_on(addr == NULL_ADDR);
 236         if (addr == NEW_ADDR)
 237                 return;
 238
 239         /* add it into sit main buffer */
 240         mutex_lock(&sit_i->sentry_lock);
 241
 242         update_sit_entry(sbi, addr, -1);
 243
 244         /* add it into dirty seglist */
 245         locate_dirty_segment(sbi, segno);
 246
 247         mutex_unlock(&sit_i->sentry_lock);
 248 }
 249
 250 /*
 251  * This function should be resided under the curseg_mutex lock
 252  */
 253 static void __add_sum_entry(struct f2fs_sb_info *sbi, int type,
 254                                         struct f2fs_summary *sum)
 255 {
 256         struct curseg_info *curseg = CURSEG_I(sbi, type);
 257         void *addr = curseg->sum_blk;
 258         addr += curseg->next_blkoff * sizeof(struct f2fs_summary);
 259         memcpy(addr, sum, sizeof(struct f2fs_summary));
 260 }
 261
 262 /*
 263  * Calculate the number of current summary pages for writing
 264  */
 265 int npages_for_summary_flush(struct f2fs_sb_info *sbi)
 266 {
 267         int valid_sum_count = 0;
 268         int i, sum_in_page;
 269
 270         for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
 271                 if (sbi->ckpt->alloc_type[i] == SSR)
 272                         valid_sum_count += sbi->blocks_per_seg;
 273                 else
 274                         valid_sum_count += curseg_blkoff(sbi, i);
 275         }
 276
 277         sum_in_page = (PAGE_CACHE_SIZE - 2 * SUM_JOURNAL_SIZE -
 278                         SUM_FOOTER_SIZE) / SUMMARY_SIZE;
 279         if (valid_sum_count <= sum_in_page)
 280                 return 1;
 281         else if ((valid_sum_count - sum_in_page) <=
 282                 (PAGE_CACHE_SIZE - SUM_FOOTER_SIZE) / SUMMARY_SIZE)
 283                 return 2;
 284         return 3;
 285 }
 286
 287 /*
 288  * Caller should put this summary page
 289  */
 290 struct page *get_sum_page(struct f2fs_sb_info *sbi, unsigned int segno)
 291 {
 292         return get_meta_page(sbi, GET_SUM_BLOCK(sbi, segno));
 293 }
 294
 295 static void write_sum_page(struct f2fs_sb_info *sbi,
 296                         struct f2fs_summary_block *sum_blk, block_t blk_addr)
 297 {
 298         struct page *page = grab_meta_page(sbi, blk_addr);
 299         void *kaddr = page_address(page);
 300         memcpy(kaddr, sum_blk, PAGE_CACHE_SIZE);
 301         set_page_dirty(page);
 302         f2fs_put_page(page, 1);
 303 }
 304
 305 static int is_next_segment_free(struct f2fs_sb_info *sbi, int type)
 306 {
 307         struct curseg_info *curseg = CURSEG_I(sbi, type);
 308         unsigned int segno = curseg->segno + 1;
 309         struct free_segmap_info *free_i = FREE_I(sbi);
 310
 311         if (segno < TOTAL_SEGS(sbi) && segno % sbi->segs_per_sec)
 312                 return !test_bit(segno, free_i->free_segmap);
 313         return 0;
 314 }
 315
 316 /*
 317  * Find a new segment from the free segments bitmap to right order
 318  * This function should be returned with success, otherwise BUG
 319  */
 320 static void get_new_segment(struct f2fs_sb_info *sbi,
 321                         unsigned int *newseg, bool new_sec, int dir)
 322 {
 323         struct free_segmap_info *free_i = FREE_I(sbi);
 324         unsigned int segno, secno, zoneno;
 325         unsigned int total_zones = TOTAL_SECS(sbi) / sbi->secs_per_zone;
 326         unsigned int hint = *newseg / sbi->segs_per_sec;
 327         unsigned int old_zoneno = GET_ZONENO_FROM_SEGNO(sbi, *newseg);
 328         unsigned int left_start = hint;
 329         bool init = true;
 330         int go_left = 0;
 331         int i;
 332
 333         write_lock(&free_i->segmap_lock);
 334
 335         if (!new_sec && ((*newseg + 1) % sbi->segs_per_sec)) {
 336                 segno = find_next_zero_bit(free_i->free_segmap,
 337                                         TOTAL_SEGS(sbi), *newseg + 1);
 338                 if (segno - *newseg < sbi->segs_per_sec -
 339                                         (*newseg % sbi->segs_per_sec))
 340                         goto got_it;
 341         }
 342 find_other_zone:
 343         secno = find_next_zero_bit(free_i->free_secmap, TOTAL_SECS(sbi), hint);
 344         if (secno >= TOTAL_SECS(sbi)) {
 345                 if (dir == ALLOC_RIGHT) {
 346                         secno = find_next_zero_bit(free_i->free_secmap,
 347                                                         TOTAL_SECS(sbi), 0);
 348                         f2fs_bug_on(secno >= TOTAL_SECS(sbi));
 349                 } else {
 350                         go_left = 1;
 351                         left_start = hint - 1;
 352                 }
 353         }
 354         if (go_left == 0)
 355                 goto skip_left;
 356
 357         while (test_bit(left_start, free_i->free_secmap)) {
 358                 if (left_start > 0) {
 359                         left_start--;
 360                         continue;
 361                 }
 362                 left_start = find_next_zero_bit(free_i->free_secmap,
 363                                                         TOTAL_SECS(sbi), 0);
 364                 f2fs_bug_on(left_start >= TOTAL_SECS(sbi));
 365                 break;
 366         }
 367         secno = left_start;
 368 skip_left:
 369         hint = secno;
 370         segno = secno * sbi->segs_per_sec;
 371         zoneno = secno / sbi->secs_per_zone;
 372
 373         /* give up on finding another zone */
 374         if (!init)
 375                 goto got_it;
 376         if (sbi->secs_per_zone == 1)
 377                 goto got_it;
 378         if (zoneno == old_zoneno)
 379                 goto got_it;
 380         if (dir == ALLOC_LEFT) {
 381                 if (!go_left && zoneno + 1 >= total_zones)
 382                         goto got_it;
 383                 if (go_left && zoneno == 0)
 384                         goto got_it;
 385         }
 386         for (i = 0; i < NR_CURSEG_TYPE; i++)
 387                 if (CURSEG_I(sbi, i)->zone == zoneno)
 388                         break;
 389
 390         if (i < NR_CURSEG_TYPE) {
 391                 /* zone is in user, try another */
 392                 if (go_left)
 393                         hint = zoneno * sbi->secs_per_zone - 1;
 394                 else if (zoneno + 1 >= total_zones)
 395                         hint = 0;
 396                 else
 397                         hint = (zoneno + 1) * sbi->secs_per_zone;
 398                 init = false;
 399                 goto find_other_zone;
 400         }
 401 got_it:
 402         /* set it as dirty segment in free segmap */
 403         f2fs_bug_on(test_bit(segno, free_i->free_segmap));
 404         __set_inuse(sbi, segno);
 405         *newseg = segno;
 406         write_unlock(&free_i->segmap_lock);
 407 }
 408
 409 static void reset_curseg(struct f2fs_sb_info *sbi, int type, int modified)
 410 {
 411         struct curseg_info *curseg = CURSEG_I(sbi, type);
 412         struct summary_footer *sum_footer;
 413
 414         curseg->segno = curseg->next_segno;
 415         curseg->zone = GET_ZONENO_FROM_SEGNO(sbi, curseg->segno);
 416         curseg->next_blkoff = 0;
 417         curseg->next_segno = NULL_SEGNO;
 418
 419         sum_footer = &(curseg->sum_blk->footer);
 420         memset(sum_footer, 0, sizeof(struct summary_footer));
 421         if (IS_DATASEG(type))
 422                 SET_SUM_TYPE(sum_footer, SUM_TYPE_DATA);
 423         if (IS_NODESEG(type))
 424                 SET_SUM_TYPE(sum_footer, SUM_TYPE_NODE);
 425         __set_sit_entry_type(sbi, type, curseg->segno, modified);
 426 }
 427
 428 /*
 429  * Allocate a current working segment.
 430  * This function always allocates a free segment in LFS manner.
 431  */
 432 static void new_curseg(struct f2fs_sb_info *sbi, int type, bool new_sec)
 433 {
 434         struct curseg_info *curseg = CURSEG_I(sbi, type);
 435         unsigned int segno = curseg->segno;
 436         int dir = ALLOC_LEFT;
 437
 438         write_sum_page(sbi, curseg->sum_blk,
 439                                 GET_SUM_BLOCK(sbi, segno));
 440         if (type == CURSEG_WARM_DATA || type == CURSEG_COLD_DATA)
 441                 dir = ALLOC_RIGHT;
 442
 443         if (test_opt(sbi, NOHEAP))
 444                 dir = ALLOC_RIGHT;
 445
 446         get_new_segment(sbi, &segno, new_sec, dir);
 447         curseg->next_segno = segno;
 448         reset_curseg(sbi, type, 1);
 449         curseg->alloc_type = LFS;
 450 }
 451
 452 static void __next_free_blkoff(struct f2fs_sb_info *sbi,
 453                         struct curseg_info *seg, block_t start)
 454 {
 455         struct seg_entry *se = get_seg_entry(sbi, seg->segno);
 456         block_t ofs;
 457         for (ofs = start; ofs < sbi->blocks_per_seg; ofs++) {
 458                 if (!f2fs_test_bit(ofs, se->ckpt_valid_map)
 459                         && !f2fs_test_bit(ofs, se->cur_valid_map))
 460                         break;
 461         }
 462         seg->next_blkoff = ofs;
 463 }
 464
 465 /*
 466  * If a segment is written by LFS manner, next block offset is just obtained
 467  * by increasing the current block offset. However, if a segment is written by
 468  * SSR manner, next block offset obtained by calling __next_free_blkoff
 469  */
 470 static void __refresh_next_blkoff(struct f2fs_sb_info *sbi,
 471                                 struct curseg_info *seg)
 472 {
 473         if (seg->alloc_type == SSR)
 474                 __next_free_blkoff(sbi, seg, seg->next_blkoff + 1);
 475         else
 476                 seg->next_blkoff++;
 477 }
 478
 479 /*
 480  * This function always allocates a used segment (from dirty seglist) by SSR
 481  * manner, so it should recover the existing segment information of valid blocks
 482  */
 483 static void change_curseg(struct f2fs_sb_info *sbi, int type, bool reuse)
 484 {
 485         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
 486         struct curseg_info *curseg = CURSEG_I(sbi, type);
 487         unsigned int new_segno = curseg->next_segno;
 488         struct f2fs_summary_block *sum_node;
 489         struct page *sum_page;
 490
 491         write_sum_page(sbi, curseg->sum_blk,
 492                                 GET_SUM_BLOCK(sbi, curseg->segno));
 493         __set_test_and_inuse(sbi, new_segno);
 494
 495         mutex_lock(&dirty_i->seglist_lock);
 496         __remove_dirty_segment(sbi, new_segno, PRE);
 497         __remove_dirty_segment(sbi, new_segno, DIRTY);
 498         mutex_unlock(&dirty_i->seglist_lock);
 499
 500         reset_curseg(sbi, type, 1);
 501         curseg->alloc_type = SSR;
 502         __next_free_blkoff(sbi, curseg, 0);
 503
 504         if (reuse) {
 505                 sum_page = get_sum_page(sbi, new_segno);
 506                 sum_node = (struct f2fs_summary_block *)page_address(sum_page);
 507                 memcpy(curseg->sum_blk, sum_node, SUM_ENTRY_SIZE);
 508                 f2fs_put_page(sum_page, 1);
 509         }
 510 }
 511
 512 static int get_ssr_segment(struct f2fs_sb_info *sbi, int type)
 513 {
 514         struct curseg_info *curseg = CURSEG_I(sbi, type);
 515         const struct victim_selection *v_ops = DIRTY_I(sbi)->v_ops;
 516
 517         if (IS_NODESEG(type) || !has_not_enough_free_secs(sbi, 0))
 518                 return v_ops->get_victim(sbi,
 519                                 &(curseg)->next_segno, BG_GC, type, SSR);
 520
 521         /* For data segments, let's do SSR more intensively */
 522         for (; type >= CURSEG_HOT_DATA; type--)
 523                 if (v_ops->get_victim(sbi, &(curseg)->next_segno,
 524                                                 BG_GC, type, SSR))
 525                         return 1;
 526         return 0;
 527 }
 528
 529 /*
 530  * flush out current segment and replace it with new segment
 531  * This function should be returned with success, otherwise BUG
 532  */
 533 static void allocate_segment_by_default(struct f2fs_sb_info *sbi,
 534                                                 int type, bool force)
 535 {
 536         struct curseg_info *curseg = CURSEG_I(sbi, type);
 537
 538         if (force)
 539                 new_curseg(sbi, type, true);
 540         else if (type == CURSEG_WARM_NODE)
 541                 new_curseg(sbi, type, false);
 542         else if (curseg->alloc_type == LFS && is_next_segment_free(sbi, type))
 543                 new_curseg(sbi, type, false);
 544         else if (need_SSR(sbi) && get_ssr_segment(sbi, type))
 545                 change_curseg(sbi, type, true);
 546         else
 547                 new_curseg(sbi, type, false);
 548
 549         stat_inc_seg_type(sbi, curseg);
 550 }
 551
 552 void allocate_new_segments(struct f2fs_sb_info *sbi)
 553 {
 554         struct curseg_info *curseg;
 555         unsigned int old_curseg;
 556         int i;
 557
 558         for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
 559                 curseg = CURSEG_I(sbi, i);
 560                 old_curseg = curseg->segno;
 561                 SIT_I(sbi)->s_ops->allocate_segment(sbi, i, true);
 562                 locate_dirty_segment(sbi, old_curseg);
 563         }
 564 }
 565
 566 static const struct segment_allocation default_salloc_ops = {
 567         .allocate_segment = allocate_segment_by_default,
 568 };
 569
 570 static void f2fs_end_io_write(struct bio *bio, int err)
 571 {
 572         const int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
 573         struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
 574         struct bio_private *p = bio->bi_private;
 575
 576         do {
 577                 struct page *page = bvec->bv_page;
 578
 579                 if (--bvec >= bio->bi_io_vec)
 580                         prefetchw(&bvec->bv_page->flags);
 581                 if (!uptodate) {
 582                         SetPageError(page);
 583                         if (page->mapping)
 584                                 set_bit(AS_EIO, &page->mapping->flags);
 585                         set_ckpt_flags(p->sbi->ckpt, CP_ERROR_FLAG);
 586                         p->sbi->sb->s_flags |= MS_RDONLY;
 587                 }
 588                 end_page_writeback(page);
 589                 dec_page_count(p->sbi, F2FS_WRITEBACK);
 590         } while (bvec >= bio->bi_io_vec);
 591
 592         if (p->is_sync)
 593                 complete(p->wait);
 594
 595         if (!get_pages(p->sbi, F2FS_WRITEBACK) && p->sbi->cp_task)
 596                 wake_up_process(p->sbi->cp_task);
 597
 598         kfree(p);
 599         bio_put(bio);
 600 }
 601
 602 struct bio *f2fs_bio_alloc(struct block_device *bdev, int npages)
 603 {
 604         struct bio *bio;
 605
 606         /* No failure on bio allocation */
 607         bio = bio_alloc(GFP_NOIO, npages);
 608         bio->bi_bdev = bdev;
 609         bio->bi_private = NULL;
 610
 611         return bio;
 612 }
 613
 614 static void do_submit_bio(struct f2fs_sb_info *sbi,
 615                                 enum page_type type, bool sync)
 616 {
 617         int rw = sync ? WRITE_SYNC : WRITE;
 618         enum page_type btype = type > META ? META : type;
 619
 620         if (type >= META_FLUSH)
 621                 rw = WRITE_FLUSH_FUA;
 622
 623         if (btype == META)
 624                 rw |= REQ_META;
 625
 626         if (sbi->bio[btype]) {
 627                 struct bio_private *p = sbi->bio[btype]->bi_private;
 628                 p->sbi = sbi;
 629                 sbi->bio[btype]->bi_end_io = f2fs_end_io_write;
 630
 631                 trace_f2fs_do_submit_bio(sbi->sb, btype, sync, sbi->bio[btype]);
 632
 633                 if (type == META_FLUSH) {
 634                         DECLARE_COMPLETION_ONSTACK(wait);
 635                         p->is_sync = true;
 636                         p->wait = &wait;
 637                         submit_bio(rw, sbi->bio[btype]);
 638                         wait_for_completion(&wait);
 639                 } else {
 640                         p->is_sync = false;
 641                         submit_bio(rw, sbi->bio[btype]);
 642                 }
 643                 sbi->bio[btype] = NULL;
 644         }
 645 }
 646
 647 void f2fs_submit_bio(struct f2fs_sb_info *sbi, enum page_type type, bool sync)
 648 {
 649         down_write(&sbi->bio_sem);
 650         do_submit_bio(sbi, type, sync);
 651         up_write(&sbi->bio_sem);
 652 }
 653
 654 static void submit_write_page(struct f2fs_sb_info *sbi, struct page *page,
 655                                 block_t blk_addr, enum page_type type)
 656 {
 657         struct block_device *bdev = sbi->sb->s_bdev;
 658         int bio_blocks;
 659
 660         verify_block_addr(sbi, blk_addr);
 661
 662         down_write(&sbi->bio_sem);
 663
 664         inc_page_count(sbi, F2FS_WRITEBACK);
 665
 666         if (sbi->bio[type] && sbi->last_block_in_bio[type] != blk_addr - 1)
 667                 do_submit_bio(sbi, type, false);
 668 alloc_new:
 669         if (sbi->bio[type] == NULL) {
 670                 struct bio_private *priv;
 671 retry:
 672                 priv = kmalloc(sizeof(struct bio_private), GFP_NOFS);
 673                 if (!priv) {
 674                         cond_resched();
 675                         goto retry;
 676                 }
 677
 678                 bio_blocks = MAX_BIO_BLOCKS(max_hw_blocks(sbi));
 679                 sbi->bio[type] = f2fs_bio_alloc(bdev, bio_blocks);
 680                 sbi->bio[type]->bi_sector = SECTOR_FROM_BLOCK(sbi, blk_addr);
 681                 sbi->bio[type]->bi_private = priv;
 682                 /*
 683                  * The end_io will be assigned at the sumbission phase.
 684                  * Until then, let bio_add_page() merge consecutive IOs as much
 685                  * as possible.
 686                  */
 687         }
 688
 689         if (bio_add_page(sbi->bio[type], page, PAGE_CACHE_SIZE, 0) <
 690                                                         PAGE_CACHE_SIZE) {
 691                 do_submit_bio(sbi, type, false);
 692                 goto alloc_new;
 693         }
 694
 695         sbi->last_block_in_bio[type] = blk_addr;
 696
 697         up_write(&sbi->bio_sem);
 698         trace_f2fs_submit_write_page(page, blk_addr, type);
 699 }
 700
 701 void f2fs_wait_on_page_writeback(struct page *page,
 702                                 enum page_type type, bool sync)
 703 {
 704         struct f2fs_sb_info *sbi = F2FS_SB(page->mapping->host->i_sb);
 705         if (PageWriteback(page)) {
 706                 f2fs_submit_bio(sbi, type, sync);
 707                 wait_on_page_writeback(page);
 708         }
 709 }
 710
 711 static bool __has_curseg_space(struct f2fs_sb_info *sbi, int type)
 712 {
 713         struct curseg_info *curseg = CURSEG_I(sbi, type);
 714         if (curseg->next_blkoff < sbi->blocks_per_seg)
 715                 return true;
 716         return false;
 717 }
 718
 719 static int __get_segment_type_2(struct page *page, enum page_type p_type)
 720 {
 721         if (p_type == DATA)
 722                 return CURSEG_HOT_DATA;
 723         else
 724                 return CURSEG_HOT_NODE;
 725 }
 726
 727 static int __get_segment_type_4(struct page *page, enum page_type p_type)
 728 {
 729         if (p_type == DATA) {
 730                 struct inode *inode = page->mapping->host;
 731
 732                 if (S_ISDIR(inode->i_mode))
 733                         return CURSEG_HOT_DATA;
 734                 else
 735                         return CURSEG_COLD_DATA;
 736         } else {
 737                 if (IS_DNODE(page) && !is_cold_node(page))
 738                         return CURSEG_HOT_NODE;
 739                 else
 740                         return CURSEG_COLD_NODE;
 741         }
 742 }
 743
 744 static int __get_segment_type_6(struct page *page, enum page_type p_type)
 745 {
 746         if (p_type == DATA) {
 747                 struct inode *inode = page->mapping->host;
 748
 749                 if (S_ISDIR(inode->i_mode))
 750                         return CURSEG_HOT_DATA;
 751                 else if (is_cold_data(page) || file_is_cold(inode))
 752                         return CURSEG_COLD_DATA;
 753                 else
 754                         return CURSEG_WARM_DATA;
 755         } else {
 756                 if (IS_DNODE(page))
 757                         return is_cold_node(page) ? CURSEG_WARM_NODE :
 758                                                 CURSEG_HOT_NODE;
 759                 else
 760                         return CURSEG_COLD_NODE;
 761         }
 762 }
 763
 764 static int __get_segment_type(struct page *page, enum page_type p_type)
 765 {
 766         struct f2fs_sb_info *sbi = F2FS_SB(page->mapping->host->i_sb);
 767         switch (sbi->active_logs) {
 768         case 2:
 769                 return __get_segment_type_2(page, p_type);
 770         case 4:
 771                 return __get_segment_type_4(page, p_type);
 772         }
 773         /* NR_CURSEG_TYPE(6) logs by default */
 774         f2fs_bug_on(sbi->active_logs != NR_CURSEG_TYPE);
 775         return __get_segment_type_6(page, p_type);
 776 }
 777
 778 static void do_write_page(struct f2fs_sb_info *sbi, struct page *page,
 779                         block_t old_blkaddr, block_t *new_blkaddr,
 780                         struct f2fs_summary *sum, enum page_type p_type)
 781 {
 782         struct sit_info *sit_i = SIT_I(sbi);
 783         struct curseg_info *curseg;
 784         unsigned int old_cursegno;
 785         int type;
 786
 787         type = __get_segment_type(page, p_type);
 788         curseg = CURSEG_I(sbi, type);
 789
 790         mutex_lock(&curseg->curseg_mutex);
 791
 792         *new_blkaddr = NEXT_FREE_BLKADDR(sbi, curseg);
 793         old_cursegno = curseg->segno;
 794
 795         /*
 796          * __add_sum_entry should be resided under the curseg_mutex
 797          * because, this function updates a summary entry in the
 798          * current summary block.
 799          */
 800         __add_sum_entry(sbi, type, sum);
 801
 802         mutex_lock(&sit_i->sentry_lock);
 803         __refresh_next_blkoff(sbi, curseg);
 804
 805         stat_inc_block_count(sbi, curseg);
 806
 807         /*
 808          * SIT information should be updated before segment allocation,
 809          * since SSR needs latest valid block information.
 810          */
 811         refresh_sit_entry(sbi, old_blkaddr, *new_blkaddr);
 812
 813         if (!__has_curseg_space(sbi, type))
 814                 sit_i->s_ops->allocate_segment(sbi, type, false);
 815
 816         locate_dirty_segment(sbi, old_cursegno);
 817         locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr));
 818         mutex_unlock(&sit_i->sentry_lock);
 819
 820         if (p_type == NODE)
 821                 fill_node_footer_blkaddr(page, NEXT_FREE_BLKADDR(sbi, curseg));
 822
 823         /* writeout dirty page into bdev */
 824         submit_write_page(sbi, page, *new_blkaddr, p_type);
 825
 826         mutex_unlock(&curseg->curseg_mutex);
 827 }
 828
 829 void write_meta_page(struct f2fs_sb_info *sbi, struct page *page)
 830 {
 831         set_page_writeback(page);
 832         submit_write_page(sbi, page, page->index, META);
 833 }
 834
 835 void write_node_page(struct f2fs_sb_info *sbi, struct page *page,
 836                 unsigned int nid, block_t old_blkaddr, block_t *new_blkaddr)
 837 {
 838         struct f2fs_summary sum;
 839         set_summary(&sum, nid, 0, 0);
 840         do_write_page(sbi, page, old_blkaddr, new_blkaddr, &sum, NODE);
 841 }
 842
 843 void write_data_page(struct inode *inode, struct page *page,
 844                 struct dnode_of_data *dn, block_t old_blkaddr,
 845                 block_t *new_blkaddr)
 846 {
 847         struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
 848         struct f2fs_summary sum;
 849         struct node_info ni;
 850
 851         f2fs_bug_on(old_blkaddr == NULL_ADDR);
 852         get_node_info(sbi, dn->nid, &ni);
 853         set_summary(&sum, dn->nid, dn->ofs_in_node, ni.version);
 854
 855         do_write_page(sbi, page, old_blkaddr,
 856                         new_blkaddr, &sum, DATA);
 857 }
 858
 859 void rewrite_data_page(struct f2fs_sb_info *sbi, struct page *page,
 860                                         block_t old_blk_addr)
 861 {
 862         submit_write_page(sbi, page, old_blk_addr, DATA);
 863 }
 864
 865 void recover_data_page(struct f2fs_sb_info *sbi,
 866                         struct page *page, struct f2fs_summary *sum,
 867                         block_t old_blkaddr, block_t new_blkaddr)
 868 {
 869         struct sit_info *sit_i = SIT_I(sbi);
 870         struct curseg_info *curseg;
 871         unsigned int segno, old_cursegno;
 872         struct seg_entry *se;
 873         int type;
 874
 875         segno = GET_SEGNO(sbi, new_blkaddr);
 876         se = get_seg_entry(sbi, segno);
 877         type = se->type;
 878
 879         if (se->valid_blocks == 0 && !IS_CURSEG(sbi, segno)) {
 880                 if (old_blkaddr == NULL_ADDR)
 881                         type = CURSEG_COLD_DATA;
 882                 else
 883                         type = CURSEG_WARM_DATA;
 884         }
 885         curseg = CURSEG_I(sbi, type);
 886
 887         mutex_lock(&curseg->curseg_mutex);
 888         mutex_lock(&sit_i->sentry_lock);
 889
 890         old_cursegno = curseg->segno;
 891
 892         /* change the current segment */
 893         if (segno != curseg->segno) {
 894                 curseg->next_segno = segno;
 895                 change_curseg(sbi, type, true);
 896         }
 897
 898         curseg->next_blkoff = GET_SEGOFF_FROM_SEG0(sbi, new_blkaddr) &
 899                                         (sbi->blocks_per_seg - 1);
 900         __add_sum_entry(sbi, type, sum);
 901
 902         refresh_sit_entry(sbi, old_blkaddr, new_blkaddr);
 903
 904         locate_dirty_segment(sbi, old_cursegno);
 905         locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr));
 906
 907         mutex_unlock(&sit_i->sentry_lock);
 908         mutex_unlock(&curseg->curseg_mutex);
 909 }
 910
 911 void rewrite_node_page(struct f2fs_sb_info *sbi,
 912                         struct page *page, struct f2fs_summary *sum,
 913                         block_t old_blkaddr, block_t new_blkaddr)
 914 {
 915         struct sit_info *sit_i = SIT_I(sbi);
 916         int type = CURSEG_WARM_NODE;
 917         struct curseg_info *curseg;
 918         unsigned int segno, old_cursegno;
 919         block_t next_blkaddr = next_blkaddr_of_node(page);
 920         unsigned int next_segno = GET_SEGNO(sbi, next_blkaddr);
 921
 922         curseg = CURSEG_I(sbi, type);
 923
 924         mutex_lock(&curseg->curseg_mutex);
 925         mutex_lock(&sit_i->sentry_lock);
 926
 927         segno = GET_SEGNO(sbi, new_blkaddr);
 928         old_cursegno = curseg->segno;
 929
 930         /* change the current segment */
 931         if (segno != curseg->segno) {
 932                 curseg->next_segno = segno;
 933                 change_curseg(sbi, type, true);
 934         }
 935         curseg->next_blkoff = GET_SEGOFF_FROM_SEG0(sbi, new_blkaddr) &
 936                                         (sbi->blocks_per_seg - 1);
 937         __add_sum_entry(sbi, type, sum);
 938
 939         /* change the current log to the next block addr in advance */
 940         if (next_segno != segno) {
 941                 curseg->next_segno = next_segno;
 942                 change_curseg(sbi, type, true);
 943         }
 944         curseg->next_blkoff = GET_SEGOFF_FROM_SEG0(sbi, next_blkaddr) &
 945                                         (sbi->blocks_per_seg - 1);
 946
 947         /* rewrite node page */
 948         set_page_writeback(page);
 949         submit_write_page(sbi, page, new_blkaddr, NODE);
 950         f2fs_submit_bio(sbi, NODE, true);
 951         refresh_sit_entry(sbi, old_blkaddr, new_blkaddr);
 952
 953         locate_dirty_segment(sbi, old_cursegno);
 954         locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr));
 955
 956         mutex_unlock(&sit_i->sentry_lock);
 957         mutex_unlock(&curseg->curseg_mutex);
 958 }
 959
 960 static int read_compacted_summaries(struct f2fs_sb_info *sbi)
 961 {
 962         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
 963         struct curseg_info *seg_i;
 964         unsigned char *kaddr;
 965         struct page *page;
 966         block_t start;
 967         int i, j, offset;
 968
 969         start = start_sum_block(sbi);
 970
 971         page = get_meta_page(sbi, start++);
 972         kaddr = (unsigned char *)page_address(page);
 973
 974         /* Step 1: restore nat cache */
 975         seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
 976         memcpy(&seg_i->sum_blk->n_nats, kaddr, SUM_JOURNAL_SIZE);
 977
 978         /* Step 2: restore sit cache */
 979         seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
 980         memcpy(&seg_i->sum_blk->n_sits, kaddr + SUM_JOURNAL_SIZE,
 981                                                 SUM_JOURNAL_SIZE);
 982         offset = 2 * SUM_JOURNAL_SIZE;
 983
 984         /* Step 3: restore summary entries */
 985         for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
 986                 unsigned short blk_off;
 987                 unsigned int segno;
 988
 989                 seg_i = CURSEG_I(sbi, i);
 990                 segno = le32_to_cpu(ckpt->cur_data_segno[i]);
 991                 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[i]);
 992                 seg_i->next_segno = segno;
 993                 reset_curseg(sbi, i, 0);
 994                 seg_i->alloc_type = ckpt->alloc_type[i];
 995                 seg_i->next_blkoff = blk_off;
 996
 997                 if (seg_i->alloc_type == SSR)
 998                         blk_off = sbi->blocks_per_seg;
 999
1000                 for (j = 0; j < blk_off; j++) {
1001                         struct f2fs_summary *s;
1002                         s = (struct f2fs_summary *)(kaddr + offset);
1003                         seg_i->sum_blk->entries[j] = *s;
1004                         offset += SUMMARY_SIZE;
1005                         if (offset + SUMMARY_SIZE <= PAGE_CACHE_SIZE -
1006                                                 SUM_FOOTER_SIZE)
1007                                 continue;
1008
1009                         f2fs_put_page(page, 1);
1010                         page = NULL;
1011
1012                         page = get_meta_page(sbi, start++);
1013                         kaddr = (unsigned char *)page_address(page);
1014                         offset = 0;
1015                 }
1016         }
1017         f2fs_put_page(page, 1);
1018         return 0;
1019 }
1020
1021 static int read_normal_summaries(struct f2fs_sb_info *sbi, int type)
1022 {
1023         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1024         struct f2fs_summary_block *sum;
1025         struct curseg_info *curseg;
1026         struct page *new;
1027         unsigned short blk_off;
1028         unsigned int segno = 0;
1029         block_t blk_addr = 0;
1030
1031         /* get segment number and block addr */
1032         if (IS_DATASEG(type)) {
1033                 segno = le32_to_cpu(ckpt->cur_data_segno[type]);
1034                 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[type -
1035                                                         CURSEG_HOT_DATA]);
1036                 if (is_set_ckpt_flags(ckpt, CP_UMOUNT_FLAG))
1037                         blk_addr = sum_blk_addr(sbi, NR_CURSEG_TYPE, type);
1038                 else
1039                         blk_addr = sum_blk_addr(sbi, NR_CURSEG_DATA_TYPE, type);
1040         } else {
1041                 segno = le32_to_cpu(ckpt->cur_node_segno[type -
1042                                                         CURSEG_HOT_NODE]);
1043                 blk_off = le16_to_cpu(ckpt->cur_node_blkoff[type -
1044                                                         CURSEG_HOT_NODE]);
1045                 if (is_set_ckpt_flags(ckpt, CP_UMOUNT_FLAG))
1046                         blk_addr = sum_blk_addr(sbi, NR_CURSEG_NODE_TYPE,
1047                                                         type - CURSEG_HOT_NODE);
1048                 else
1049                         blk_addr = GET_SUM_BLOCK(sbi, segno);
1050         }
1051
1052         new = get_meta_page(sbi, blk_addr);
1053         sum = (struct f2fs_summary_block *)page_address(new);
1054
1055         if (IS_NODESEG(type)) {
1056                 if (is_set_ckpt_flags(ckpt, CP_UMOUNT_FLAG)) {
1057                         struct f2fs_summary *ns = &sum->entries[0];
1058                         int i;
1059                         for (i = 0; i < sbi->blocks_per_seg; i++, ns++) {
1060                                 ns->version = 0;
1061                                 ns->ofs_in_node = 0;
1062                         }
1063                 } else {
1064                         if (restore_node_summary(sbi, segno, sum)) {
1065                                 f2fs_put_page(new, 1);
1066                                 return -EINVAL;
1067                         }
1068                 }
1069         }
1070
1071         /* set uncompleted segment to curseg */
1072         curseg = CURSEG_I(sbi, type);
1073         mutex_lock(&curseg->curseg_mutex);
1074         memcpy(curseg->sum_blk, sum, PAGE_CACHE_SIZE);
1075         curseg->next_segno = segno;
1076         reset_curseg(sbi, type, 0);
1077         curseg->alloc_type = ckpt->alloc_type[type];
1078         curseg->next_blkoff = blk_off;
1079         mutex_unlock(&curseg->curseg_mutex);
1080         f2fs_put_page(new, 1);
1081         return 0;
1082 }
1083
1084 static int restore_curseg_summaries(struct f2fs_sb_info *sbi)
1085 {
1086         int type = CURSEG_HOT_DATA;
1087
1088         if (is_set_ckpt_flags(F2FS_CKPT(sbi), CP_COMPACT_SUM_FLAG)) {
1089                 /* restore for compacted data summary */
1090                 if (read_compacted_summaries(sbi))
1091                         return -EINVAL;
1092                 type = CURSEG_HOT_NODE;
1093         }
1094
1095         for (; type <= CURSEG_COLD_NODE; type++)
1096                 if (read_normal_summaries(sbi, type))
1097                         return -EINVAL;
1098         return 0;
1099 }
1100
1101 static void write_compacted_summaries(struct f2fs_sb_info *sbi, block_t blkaddr)
1102 {
1103         struct page *page;
1104         unsigned char *kaddr;
1105         struct f2fs_summary *summary;
1106         struct curseg_info *seg_i;
1107         int written_size = 0;
1108         int i, j;
1109
1110         page = grab_meta_page(sbi, blkaddr++);
1111         kaddr = (unsigned char *)page_address(page);
1112
1113         /* Step 1: write nat cache */
1114         seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
1115         memcpy(kaddr, &seg_i->sum_blk->n_nats, SUM_JOURNAL_SIZE);
1116         written_size += SUM_JOURNAL_SIZE;
1117
1118         /* Step 2: write sit cache */
1119         seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
1120         memcpy(kaddr + written_size, &seg_i->sum_blk->n_sits,
1121                                                 SUM_JOURNAL_SIZE);
1122         written_size += SUM_JOURNAL_SIZE;
1123
1124         /* Step 3: write summary entries */
1125         for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
1126                 unsigned short blkoff;
1127                 seg_i = CURSEG_I(sbi, i);
1128                 if (sbi->ckpt->alloc_type[i] == SSR)
1129                         blkoff = sbi->blocks_per_seg;
1130                 else
1131                         blkoff = curseg_blkoff(sbi, i);
1132
1133                 for (j = 0; j < blkoff; j++) {
1134                         if (!page) {
1135                                 page = grab_meta_page(sbi, blkaddr++);
1136                                 kaddr = (unsigned char *)page_address(page);
1137                                 written_size = 0;
1138                         }
1139                         summary = (struct f2fs_summary *)(kaddr + written_size);
1140                         *summary = seg_i->sum_blk->entries[j];
1141                         written_size += SUMMARY_SIZE;
1142
1143                         if (written_size + SUMMARY_SIZE <= PAGE_CACHE_SIZE -
1144                                                         SUM_FOOTER_SIZE)
1145                                 continue;
1146
1147                         set_page_dirty(page);
1148                         f2fs_put_page(page, 1);
1149                         page = NULL;
1150                 }
1151         }
1152         if (page) {
1153                 set_page_dirty(page);
1154                 f2fs_put_page(page, 1);
1155         }
1156 }
1157
1158 static void write_normal_summaries(struct f2fs_sb_info *sbi,
1159                                         block_t blkaddr, int type)
1160 {
1161         int i, end;
1162         if (IS_DATASEG(type))
1163                 end = type + NR_CURSEG_DATA_TYPE;
1164         else
1165                 end = type + NR_CURSEG_NODE_TYPE;
1166
1167         for (i = type; i < end; i++) {
1168                 struct curseg_info *sum = CURSEG_I(sbi, i);
1169                 mutex_lock(&sum->curseg_mutex);
1170                 write_sum_page(sbi, sum->sum_blk, blkaddr + (i - type));
1171                 mutex_unlock(&sum->curseg_mutex);
1172         }
1173 }
1174
1175 void write_data_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
1176 {
1177         if (is_set_ckpt_flags(F2FS_CKPT(sbi), CP_COMPACT_SUM_FLAG))
1178                 write_compacted_summaries(sbi, start_blk);
1179         else
1180                 write_normal_summaries(sbi, start_blk, CURSEG_HOT_DATA);
1181 }
1182
1183 void write_node_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
1184 {
1185         if (is_set_ckpt_flags(F2FS_CKPT(sbi), CP_UMOUNT_FLAG))
1186                 write_normal_summaries(sbi, start_blk, CURSEG_HOT_NODE);
1187 }
1188
1189 int lookup_journal_in_cursum(struct f2fs_summary_block *sum, int type,
1190                                         unsigned int val, int alloc)
1191 {
1192         int i;
1193
1194         if (type == NAT_JOURNAL) {
1195                 for (i = 0; i < nats_in_cursum(sum); i++) {
1196                         if (le32_to_cpu(nid_in_journal(sum, i)) == val)
1197                                 return i;
1198                 }
1199                 if (alloc && nats_in_cursum(sum) < NAT_JOURNAL_ENTRIES)
1200                         return update_nats_in_cursum(sum, 1);
1201         } else if (type == SIT_JOURNAL) {
1202                 for (i = 0; i < sits_in_cursum(sum); i++)
1203                         if (le32_to_cpu(segno_in_journal(sum, i)) == val)
1204                                 return i;
1205                 if (alloc && sits_in_cursum(sum) < SIT_JOURNAL_ENTRIES)
1206                         return update_sits_in_cursum(sum, 1);
1207         }
1208         return -1;
1209 }
1210
1211 static struct page *get_current_sit_page(struct f2fs_sb_info *sbi,
1212                                         unsigned int segno)
1213 {
1214         struct sit_info *sit_i = SIT_I(sbi);
1215         unsigned int offset = SIT_BLOCK_OFFSET(sit_i, segno);
1216         block_t blk_addr = sit_i->sit_base_addr + offset;
1217
1218         check_seg_range(sbi, segno);
1219
1220         /* calculate sit block address */
1221         if (f2fs_test_bit(offset, sit_i->sit_bitmap))
1222                 blk_addr += sit_i->sit_blocks;
1223
1224         return get_meta_page(sbi, blk_addr);
1225 }
1226
1227 static struct page *get_next_sit_page(struct f2fs_sb_info *sbi,
1228                                         unsigned int start)
1229 {
1230         struct sit_info *sit_i = SIT_I(sbi);
1231         struct page *src_page, *dst_page;
1232         pgoff_t src_off, dst_off;
1233         void *src_addr, *dst_addr;
1234
1235         src_off = current_sit_addr(sbi, start);
1236         dst_off = next_sit_addr(sbi, src_off);
1237
1238         /* get current sit block page without lock */
1239         src_page = get_meta_page(sbi, src_off);
1240         dst_page = grab_meta_page(sbi, dst_off);
1241         f2fs_bug_on(PageDirty(src_page));
1242
1243         src_addr = page_address(src_page);
1244         dst_addr = page_address(dst_page);
1245         memcpy(dst_addr, src_addr, PAGE_CACHE_SIZE);
1246
1247         set_page_dirty(dst_page);
1248         f2fs_put_page(src_page, 1);
1249
1250         set_to_next_sit(sit_i, start);
1251
1252         return dst_page;
1253 }
1254
1255 static bool flush_sits_in_journal(struct f2fs_sb_info *sbi)
1256 {
1257         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
1258         struct f2fs_summary_block *sum = curseg->sum_blk;
1259         int i;
1260
1261         /*
1262          * If the journal area in the current summary is full of sit entries,
1263          * all the sit entries will be flushed. Otherwise the sit entries
1264          * are not able to replace with newly hot sit entries.
1265          */
1266         if (sits_in_cursum(sum) >= SIT_JOURNAL_ENTRIES) {
1267                 for (i = sits_in_cursum(sum) - 1; i >= 0; i--) {
1268                         unsigned int segno;
1269                         segno = le32_to_cpu(segno_in_journal(sum, i));
1270                         __mark_sit_entry_dirty(sbi, segno);
1271                 }
1272                 update_sits_in_cursum(sum, -sits_in_cursum(sum));
1273                 return true;
1274         }
1275         return false;
1276 }
1277
1278 /*
1279  * CP calls this function, which flushes SIT entries including sit_journal,
1280  * and moves prefree segs to free segs.
1281  */
1282 void flush_sit_entries(struct f2fs_sb_info *sbi)
1283 {
1284         struct sit_info *sit_i = SIT_I(sbi);
1285         unsigned long *bitmap = sit_i->dirty_sentries_bitmap;
1286         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
1287         struct f2fs_summary_block *sum = curseg->sum_blk;
1288         unsigned long nsegs = TOTAL_SEGS(sbi);
1289         struct page *page = NULL;
1290         struct f2fs_sit_block *raw_sit = NULL;
1291         unsigned int start = 0, end = 0;
1292         unsigned int segno = -1;
1293         bool flushed;
1294
1295         mutex_lock(&curseg->curseg_mutex);
1296         mutex_lock(&sit_i->sentry_lock);
1297
1298         /*
1299          * "flushed" indicates whether sit entries in journal are flushed
1300          * to the SIT area or not.
1301          */
1302         flushed = flush_sits_in_journal(sbi);
1303
1304         while ((segno = find_next_bit(bitmap, nsegs, segno + 1)) < nsegs) {
1305                 struct seg_entry *se = get_seg_entry(sbi, segno);
1306                 int sit_offset, offset;
1307
1308                 sit_offset = SIT_ENTRY_OFFSET(sit_i, segno);
1309
1310                 if (flushed)
1311                         goto to_sit_page;
1312
1313                 offset = lookup_journal_in_cursum(sum, SIT_JOURNAL, segno, 1);
1314                 if (offset >= 0) {
1315                         segno_in_journal(sum, offset) = cpu_to_le32(segno);
1316                         seg_info_to_raw_sit(se, &sit_in_journal(sum, offset));
1317                         goto flush_done;
1318                 }
1319 to_sit_page:
1320                 if (!page || (start > segno) || (segno > end)) {
1321                         if (page) {
1322                                 f2fs_put_page(page, 1);
1323                                 page = NULL;
1324                         }
1325
1326                         start = START_SEGNO(sit_i, segno);
1327                         end = start + SIT_ENTRY_PER_BLOCK - 1;
1328
1329                         /* read sit block that will be updated */
1330                         page = get_next_sit_page(sbi, start);
1331                         raw_sit = page_address(page);
1332                 }
1333
1334                 /* udpate entry in SIT block */
1335                 seg_info_to_raw_sit(se, &raw_sit->entries[sit_offset]);
1336 flush_done:
1337                 __clear_bit(segno, bitmap);
1338                 sit_i->dirty_sentries--;
1339         }
1340         mutex_unlock(&sit_i->sentry_lock);
1341         mutex_unlock(&curseg->curseg_mutex);
1342
1343         /* writeout last modified SIT block */
1344         f2fs_put_page(page, 1);
1345
1346         set_prefree_as_free_segments(sbi);
1347 }
1348
1349 static int build_sit_info(struct f2fs_sb_info *sbi)
1350 {
1351         struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
1352         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1353         struct sit_info *sit_i;
1354         unsigned int sit_segs, start;
1355         char *src_bitmap, *dst_bitmap;
1356         unsigned int bitmap_size;
1357
1358         /* allocate memory for SIT information */
1359         sit_i = kzalloc(sizeof(struct sit_info), GFP_KERNEL);
1360         if (!sit_i)
1361                 return -ENOMEM;
1362
1363         SM_I(sbi)->sit_info = sit_i;
1364
1365         sit_i->sentries = vzalloc(TOTAL_SEGS(sbi) * sizeof(struct seg_entry));
1366         if (!sit_i->sentries)
1367                 return -ENOMEM;
1368
1369         bitmap_size = f2fs_bitmap_size(TOTAL_SEGS(sbi));
1370         sit_i->dirty_sentries_bitmap = kzalloc(bitmap_size, GFP_KERNEL);
1371         if (!sit_i->dirty_sentries_bitmap)
1372                 return -ENOMEM;
1373
1374         for (start = 0; start < TOTAL_SEGS(sbi); start++) {
1375                 sit_i->sentries[start].cur_valid_map
1376                         = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
1377                 sit_i->sentries[start].ckpt_valid_map
1378                         = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
1379                 if (!sit_i->sentries[start].cur_valid_map
1380                                 || !sit_i->sentries[start].ckpt_valid_map)
1381                         return -ENOMEM;
1382         }
1383
1384         if (sbi->segs_per_sec > 1) {
1385                 sit_i->sec_entries = vzalloc(TOTAL_SECS(sbi) *
1386                                         sizeof(struct sec_entry));
1387                 if (!sit_i->sec_entries)
1388                         return -ENOMEM;
1389         }
1390
1391         /* get information related with SIT */
1392         sit_segs = le32_to_cpu(raw_super->segment_count_sit) >> 1;
1393
1394         /* setup SIT bitmap from ckeckpoint pack */
1395         bitmap_size = __bitmap_size(sbi, SIT_BITMAP);
1396         src_bitmap = __bitmap_ptr(sbi, SIT_BITMAP);
1397
1398         dst_bitmap = kmemdup(src_bitmap, bitmap_size, GFP_KERNEL);
1399         if (!dst_bitmap)
1400                 return -ENOMEM;
1401
1402         /* init SIT information */
1403         sit_i->s_ops = &default_salloc_ops;
1404
1405         sit_i->sit_base_addr = le32_to_cpu(raw_super->sit_blkaddr);
1406         sit_i->sit_blocks = sit_segs << sbi->log_blocks_per_seg;
1407         sit_i->written_valid_blocks = le64_to_cpu(ckpt->valid_block_count);
1408         sit_i->sit_bitmap = dst_bitmap;
1409         sit_i->bitmap_size = bitmap_size;
1410         sit_i->dirty_sentries = 0;
1411         sit_i->sents_per_block = SIT_ENTRY_PER_BLOCK;
1412         sit_i->elapsed_time = le64_to_cpu(sbi->ckpt->elapsed_time);
1413         sit_i->mounted_time = CURRENT_TIME_SEC.tv_sec;
1414         mutex_init(&sit_i->sentry_lock);
1415         return 0;
1416 }
1417
1418 static int build_free_segmap(struct f2fs_sb_info *sbi)
1419 {
1420         struct f2fs_sm_info *sm_info = SM_I(sbi);
1421         struct free_segmap_info *free_i;
1422         unsigned int bitmap_size, sec_bitmap_size;
1423
1424         /* allocate memory for free segmap information */
1425         free_i = kzalloc(sizeof(struct free_segmap_info), GFP_KERNEL);
1426         if (!free_i)
1427                 return -ENOMEM;
1428
1429         SM_I(sbi)->free_info = free_i;
1430
1431         bitmap_size = f2fs_bitmap_size(TOTAL_SEGS(sbi));
1432         free_i->free_segmap = kmalloc(bitmap_size, GFP_KERNEL);
1433         if (!free_i->free_segmap)
1434                 return -ENOMEM;
1435
1436         sec_bitmap_size = f2fs_bitmap_size(TOTAL_SECS(sbi));
1437         free_i->free_secmap = kmalloc(sec_bitmap_size, GFP_KERNEL);
1438         if (!free_i->free_secmap)
1439                 return -ENOMEM;
1440
1441         /* set all segments as dirty temporarily */
1442         memset(free_i->free_segmap, 0xff, bitmap_size);
1443         memset(free_i->free_secmap, 0xff, sec_bitmap_size);
1444
1445         /* init free segmap information */
1446         free_i->start_segno =
1447                 (unsigned int) GET_SEGNO_FROM_SEG0(sbi, sm_info->main_blkaddr);
1448         free_i->free_segments = 0;
1449         free_i->free_sections = 0;
1450         rwlock_init(&free_i->segmap_lock);
1451         return 0;
1452 }
1453
1454 static int build_curseg(struct f2fs_sb_info *sbi)
1455 {
1456         struct curseg_info *array;
1457         int i;
1458
1459         array = kzalloc(sizeof(*array) * NR_CURSEG_TYPE, GFP_KERNEL);
1460         if (!array)
1461                 return -ENOMEM;
1462
1463         SM_I(sbi)->curseg_array = array;
1464
1465         for (i = 0; i < NR_CURSEG_TYPE; i++) {
1466                 mutex_init(&array[i].curseg_mutex);
1467                 array[i].sum_blk = kzalloc(PAGE_CACHE_SIZE, GFP_KERNEL);
1468                 if (!array[i].sum_blk)
1469                         return -ENOMEM;
1470                 array[i].segno = NULL_SEGNO;
1471                 array[i].next_blkoff = 0;
1472         }
1473         return restore_curseg_summaries(sbi);
1474 }
1475
1476 static void build_sit_entries(struct f2fs_sb_info *sbi)
1477 {
1478         struct sit_info *sit_i = SIT_I(sbi);
1479         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
1480         struct f2fs_summary_block *sum = curseg->sum_blk;
1481         unsigned int start;
1482
1483         for (start = 0; start < TOTAL_SEGS(sbi); start++) {
1484                 struct seg_entry *se = &sit_i->sentries[start];
1485                 struct f2fs_sit_block *sit_blk;
1486                 struct f2fs_sit_entry sit;
1487                 struct page *page;
1488                 int i;
1489
1490                 mutex_lock(&curseg->curseg_mutex);
1491                 for (i = 0; i < sits_in_cursum(sum); i++) {
1492                         if (le32_to_cpu(segno_in_journal(sum, i)) == start) {
1493                                 sit = sit_in_journal(sum, i);
1494                                 mutex_unlock(&curseg->curseg_mutex);
1495                                 goto got_it;
1496                         }
1497                 }
1498                 mutex_unlock(&curseg->curseg_mutex);
1499                 page = get_current_sit_page(sbi, start);
1500                 sit_blk = (struct f2fs_sit_block *)page_address(page);
1501                 sit = sit_blk->entries[SIT_ENTRY_OFFSET(sit_i, start)];
1502                 f2fs_put_page(page, 1);
1503 got_it:
1504                 check_block_count(sbi, start, &sit);
1505                 seg_info_from_raw_sit(se, &sit);
1506                 if (sbi->segs_per_sec > 1) {
1507                         struct sec_entry *e = get_sec_entry(sbi, start);
1508                         e->valid_blocks += se->valid_blocks;
1509                 }
1510         }
1511 }
1512
1513 static void init_free_segmap(struct f2fs_sb_info *sbi)
1514 {
1515         unsigned int start;
1516         int type;
1517
1518         for (start = 0; start < TOTAL_SEGS(sbi); start++) {
1519                 struct seg_entry *sentry = get_seg_entry(sbi, start);
1520                 if (!sentry->valid_blocks)
1521                         __set_free(sbi, start);
1522         }
1523
1524         /* set use the current segments */
1525         for (type = CURSEG_HOT_DATA; type <= CURSEG_COLD_NODE; type++) {
1526                 struct curseg_info *curseg_t = CURSEG_I(sbi, type);
1527                 __set_test_and_inuse(sbi, curseg_t->segno);
1528         }
1529 }
1530
1531 static void init_dirty_segmap(struct f2fs_sb_info *sbi)
1532 {
1533         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
1534         struct free_segmap_info *free_i = FREE_I(sbi);
1535         unsigned int segno = 0, offset = 0, total_segs = TOTAL_SEGS(sbi);
1536         unsigned short valid_blocks;
1537
1538         while (1) {
1539                 /* find dirty segment based on free segmap */
1540                 segno = find_next_inuse(free_i, total_segs, offset);
1541                 if (segno >= total_segs)
1542                         break;
1543                 offset = segno + 1;
1544                 valid_blocks = get_valid_blocks(sbi, segno, 0);
1545                 if (valid_blocks >= sbi->blocks_per_seg || !valid_blocks)
1546                         continue;
1547                 mutex_lock(&dirty_i->seglist_lock);
1548                 __locate_dirty_segment(sbi, segno, DIRTY);
1549                 mutex_unlock(&dirty_i->seglist_lock);
1550         }
1551 }
1552
1553 static int init_victim_secmap(struct f2fs_sb_info *sbi)
1554 {
1555         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
1556         unsigned int bitmap_size = f2fs_bitmap_size(TOTAL_SECS(sbi));
1557
1558         dirty_i->victim_secmap = kzalloc(bitmap_size, GFP_KERNEL);
1559         if (!dirty_i->victim_secmap)
1560                 return -ENOMEM;
1561         return 0;
1562 }
1563
1564 static int build_dirty_segmap(struct f2fs_sb_info *sbi)
1565 {
1566         struct dirty_seglist_info *dirty_i;
1567         unsigned int bitmap_size, i;
1568
1569         /* allocate memory for dirty segments list information */
1570         dirty_i = kzalloc(sizeof(struct dirty_seglist_info), GFP_KERNEL);
1571         if (!dirty_i)
1572                 return -ENOMEM;
1573
1574         SM_I(sbi)->dirty_info = dirty_i;
1575         mutex_init(&dirty_i->seglist_lock);
1576
1577         bitmap_size = f2fs_bitmap_size(TOTAL_SEGS(sbi));
1578
1579         for (i = 0; i < NR_DIRTY_TYPE; i++) {
1580                 dirty_i->dirty_segmap[i] = kzalloc(bitmap_size, GFP_KERNEL);
1581                 if (!dirty_i->dirty_segmap[i])
1582                         return -ENOMEM;
1583         }
1584
1585         init_dirty_segmap(sbi);
1586         return init_victim_secmap(sbi);
1587 }
1588
1589 /*
1590  * Update min, max modified time for cost-benefit GC algorithm
1591  */
1592 static void init_min_max_mtime(struct f2fs_sb_info *sbi)
1593 {
1594         struct sit_info *sit_i = SIT_I(sbi);
1595         unsigned int segno;
1596
1597         mutex_lock(&sit_i->sentry_lock);
1598
1599         sit_i->min_mtime = LLONG_MAX;
1600
1601         for (segno = 0; segno < TOTAL_SEGS(sbi); segno += sbi->segs_per_sec) {
1602                 unsigned int i;
1603                 unsigned long long mtime = 0;
1604
1605                 for (i = 0; i < sbi->segs_per_sec; i++)
1606                         mtime += get_seg_entry(sbi, segno + i)->mtime;
1607
1608                 mtime = div_u64(mtime, sbi->segs_per_sec);
1609
1610                 if (sit_i->min_mtime > mtime)
1611                         sit_i->min_mtime = mtime;
1612         }
1613         sit_i->max_mtime = get_mtime(sbi);
1614         mutex_unlock(&sit_i->sentry_lock);
1615 }
1616
1617 int build_segment_manager(struct f2fs_sb_info *sbi)
1618 {
1619         struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
1620         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1621         struct f2fs_sm_info *sm_info;
1622         int err;
1623
1624         sm_info = kzalloc(sizeof(struct f2fs_sm_info), GFP_KERNEL);
1625         if (!sm_info)
1626                 return -ENOMEM;
1627
1628         /* init sm info */
1629         sbi->sm_info = sm_info;
1630         INIT_LIST_HEAD(&sm_info->wblist_head);
1631         spin_lock_init(&sm_info->wblist_lock);
1632         sm_info->seg0_blkaddr = le32_to_cpu(raw_super->segment0_blkaddr);
1633         sm_info->main_blkaddr = le32_to_cpu(raw_super->main_blkaddr);
1634         sm_info->segment_count = le32_to_cpu(raw_super->segment_count);
1635         sm_info->reserved_segments = le32_to_cpu(ckpt->rsvd_segment_count);
1636         sm_info->ovp_segments = le32_to_cpu(ckpt->overprov_segment_count);
1637         sm_info->main_segments = le32_to_cpu(raw_super->segment_count_main);
1638         sm_info->ssa_blkaddr = le32_to_cpu(raw_super->ssa_blkaddr);
1639         sm_info->rec_prefree_segments = DEF_RECLAIM_PREFREE_SEGMENTS;
1640
1641         err = build_sit_info(sbi);
1642         if (err)
1643                 return err;
1644         err = build_free_segmap(sbi);
1645         if (err)
1646                 return err;
1647         err = build_curseg(sbi);
1648         if (err)
1649                 return err;
1650
1651         /* reinit free segmap based on SIT */
1652         build_sit_entries(sbi);
1653
1654         init_free_segmap(sbi);
1655         err = build_dirty_segmap(sbi);
1656         if (err)
1657                 return err;
1658
1659         init_min_max_mtime(sbi);
1660         return 0;
1661 }
1662
1663 static void discard_dirty_segmap(struct f2fs_sb_info *sbi,
1664                 enum dirty_type dirty_type)
1665 {
1666         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
1667
1668         mutex_lock(&dirty_i->seglist_lock);
1669         kfree(dirty_i->dirty_segmap[dirty_type]);
1670         dirty_i->nr_dirty[dirty_type] = 0;
1671         mutex_unlock(&dirty_i->seglist_lock);
1672 }
1673
1674 static void destroy_victim_secmap(struct f2fs_sb_info *sbi)
1675 {
1676         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
1677         kfree(dirty_i->victim_secmap);
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_secmap(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         if (!sm_info)
1748                 return;
1749         destroy_dirty_segmap(sbi);
1750         destroy_curseg(sbi);
1751         destroy_free_segmap(sbi);
1752         destroy_sit_info(sbi);
1753         sbi->sm_info = NULL;
1754         kfree(sm_info);
1755 }