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, 0)) {
  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                 if (get_valid_blocks(sbi, segno, sbi->segs_per_sec) == 0)
  73                         clear_bit(GET_SECNO(sbi, segno),
  74                                                 dirty_i->victim_secmap);
  75         }
  76 }
  77
  78 /*
  79  * Should not occur error such as -ENOMEM.
  80  * Adding dirty entry into seglist is not critical operation.
  81  * If a given segment is one of current working segments, it won't be added.
  82  */
  83 void locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno)
  84 {
  85         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
  86         unsigned short valid_blocks;
  87
  88         if (segno == NULL_SEGNO || IS_CURSEG(sbi, segno))
  89                 return;
  90
  91         mutex_lock(&dirty_i->seglist_lock);
  92
  93         valid_blocks = get_valid_blocks(sbi, segno, 0);
  94
  95         if (valid_blocks == 0) {
  96                 __locate_dirty_segment(sbi, segno, PRE);
  97                 __remove_dirty_segment(sbi, segno, DIRTY);
  98         } else if (valid_blocks < sbi->blocks_per_seg) {
  99                 __locate_dirty_segment(sbi, segno, DIRTY);
 100         } else {
 101                 /* Recovery routine with SSR needs this */
 102                 __remove_dirty_segment(sbi, segno, DIRTY);
 103         }
 104
 105         mutex_unlock(&dirty_i->seglist_lock);
 106         return;
 107 }
 108
 109 /*
 110  * Should call clear_prefree_segments after checkpoint is done.
 111  */
 112 static void set_prefree_as_free_segments(struct f2fs_sb_info *sbi)
 113 {
 114         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
 115         unsigned int segno, offset = 0;
 116         unsigned int total_segs = TOTAL_SEGS(sbi);
 117
 118         mutex_lock(&dirty_i->seglist_lock);
 119         while (1) {
 120                 segno = find_next_bit(dirty_i->dirty_segmap[PRE], total_segs,
 121                                 offset);
 122                 if (segno >= total_segs)
 123                         break;
 124                 __set_test_and_free(sbi, segno);
 125                 offset = segno + 1;
 126         }
 127         mutex_unlock(&dirty_i->seglist_lock);
 128 }
 129
 130 void clear_prefree_segments(struct f2fs_sb_info *sbi)
 131 {
 132         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
 133         unsigned int segno, offset = 0;
 134         unsigned int total_segs = TOTAL_SEGS(sbi);
 135
 136         mutex_lock(&dirty_i->seglist_lock);
 137         while (1) {
 138                 segno = find_next_bit(dirty_i->dirty_segmap[PRE], total_segs,
 139                                 offset);
 140                 if (segno >= total_segs)
 141                         break;
 142
 143                 offset = segno + 1;
 144                 if (test_and_clear_bit(segno, dirty_i->dirty_segmap[PRE]))
 145                         dirty_i->nr_dirty[PRE]--;
 146
 147                 /* Let's use trim */
 148                 if (test_opt(sbi, DISCARD))
 149                         blkdev_issue_discard(sbi->sb->s_bdev,
 150                                         START_BLOCK(sbi, segno) <<
 151                                         sbi->log_sectors_per_block,
 152                                         1 << (sbi->log_sectors_per_block +
 153                                                 sbi->log_blocks_per_seg),
 154                                         GFP_NOFS, 0);
 155         }
 156         mutex_unlock(&dirty_i->seglist_lock);
 157 }
 158
 159 static void __mark_sit_entry_dirty(struct f2fs_sb_info *sbi, unsigned int segno)
 160 {
 161         struct sit_info *sit_i = SIT_I(sbi);
 162         if (!__test_and_set_bit(segno, sit_i->dirty_sentries_bitmap))
 163                 sit_i->dirty_sentries++;
 164 }
 165
 166 static void __set_sit_entry_type(struct f2fs_sb_info *sbi, int type,
 167                                         unsigned int segno, int modified)
 168 {
 169         struct seg_entry *se = get_seg_entry(sbi, segno);
 170         se->type = type;
 171         if (modified)
 172                 __mark_sit_entry_dirty(sbi, segno);
 173 }
 174
 175 static void update_sit_entry(struct f2fs_sb_info *sbi, block_t blkaddr, int del)
 176 {
 177         struct seg_entry *se;
 178         unsigned int segno, offset;
 179         long int new_vblocks;
 180
 181         segno = GET_SEGNO(sbi, blkaddr);
 182
 183         se = get_seg_entry(sbi, segno);
 184         new_vblocks = se->valid_blocks + del;
 185         offset = GET_SEGOFF_FROM_SEG0(sbi, blkaddr) & (sbi->blocks_per_seg - 1);
 186
 187         BUG_ON((new_vblocks >> (sizeof(unsigned short) << 3) ||
 188                                 (new_vblocks > sbi->blocks_per_seg)));
 189
 190         se->valid_blocks = new_vblocks;
 191         se->mtime = get_mtime(sbi);
 192         SIT_I(sbi)->max_mtime = se->mtime;
 193
 194         /* Update valid block bitmap */
 195         if (del > 0) {
 196                 if (f2fs_set_bit(offset, se->cur_valid_map))
 197                         BUG();
 198         } else {
 199                 if (!f2fs_clear_bit(offset, se->cur_valid_map))
 200                         BUG();
 201         }
 202         if (!f2fs_test_bit(offset, se->ckpt_valid_map))
 203                 se->ckpt_valid_blocks += del;
 204
 205         __mark_sit_entry_dirty(sbi, segno);
 206
 207         /* update total number of valid blocks to be written in ckpt area */
 208         SIT_I(sbi)->written_valid_blocks += del;
 209
 210         if (sbi->segs_per_sec > 1)
 211                 get_sec_entry(sbi, segno)->valid_blocks += del;
 212 }
 213
 214 static void refresh_sit_entry(struct f2fs_sb_info *sbi,
 215                         block_t old_blkaddr, block_t new_blkaddr)
 216 {
 217         update_sit_entry(sbi, new_blkaddr, 1);
 218         if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO)
 219                 update_sit_entry(sbi, old_blkaddr, -1);
 220 }
 221
 222 void invalidate_blocks(struct f2fs_sb_info *sbi, block_t addr)
 223 {
 224         unsigned int segno = GET_SEGNO(sbi, addr);
 225         struct sit_info *sit_i = SIT_I(sbi);
 226
 227         BUG_ON(addr == NULL_ADDR);
 228         if (addr == NEW_ADDR)
 229                 return;
 230
 231         /* add it into sit main buffer */
 232         mutex_lock(&sit_i->sentry_lock);
 233
 234         update_sit_entry(sbi, addr, -1);
 235
 236         /* add it into dirty seglist */
 237         locate_dirty_segment(sbi, segno);
 238
 239         mutex_unlock(&sit_i->sentry_lock);
 240 }
 241
 242 /*
 243  * This function should be resided under the curseg_mutex lock
 244  */
 245 static void __add_sum_entry(struct f2fs_sb_info *sbi, int type,
 246                 struct f2fs_summary *sum, unsigned short offset)
 247 {
 248         struct curseg_info *curseg = CURSEG_I(sbi, type);
 249         void *addr = curseg->sum_blk;
 250         addr += offset * sizeof(struct f2fs_summary);
 251         memcpy(addr, sum, sizeof(struct f2fs_summary));
 252         return;
 253 }
 254
 255 /*
 256  * Calculate the number of current summary pages for writing
 257  */
 258 int npages_for_summary_flush(struct f2fs_sb_info *sbi)
 259 {
 260         int total_size_bytes = 0;
 261         int valid_sum_count = 0;
 262         int i, sum_space;
 263
 264         for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
 265                 if (sbi->ckpt->alloc_type[i] == SSR)
 266                         valid_sum_count += sbi->blocks_per_seg;
 267                 else
 268                         valid_sum_count += curseg_blkoff(sbi, i);
 269         }
 270
 271         total_size_bytes = valid_sum_count * (SUMMARY_SIZE + 1)
 272                         + sizeof(struct nat_journal) + 2
 273                         + sizeof(struct sit_journal) + 2;
 274         sum_space = PAGE_CACHE_SIZE - SUM_FOOTER_SIZE;
 275         if (total_size_bytes < sum_space)
 276                 return 1;
 277         else if (total_size_bytes < 2 * sum_space)
 278                 return 2;
 279         return 3;
 280 }
 281
 282 /*
 283  * Caller should put this summary page
 284  */
 285 struct page *get_sum_page(struct f2fs_sb_info *sbi, unsigned int segno)
 286 {
 287         return get_meta_page(sbi, GET_SUM_BLOCK(sbi, segno));
 288 }
 289
 290 static void write_sum_page(struct f2fs_sb_info *sbi,
 291                         struct f2fs_summary_block *sum_blk, block_t blk_addr)
 292 {
 293         struct page *page = grab_meta_page(sbi, blk_addr);
 294         void *kaddr = page_address(page);
 295         memcpy(kaddr, sum_blk, PAGE_CACHE_SIZE);
 296         set_page_dirty(page);
 297         f2fs_put_page(page, 1);
 298 }
 299
 300 static unsigned int check_prefree_segments(struct f2fs_sb_info *sbi, 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;
 305         unsigned 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, 0))
 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 += sbi->segs_per_sec;
 323
 324         if (segno < TOTAL_SEGS(sbi)) {
 325                 int i;
 326
 327                 /* skip intermediate segments in a section */
 328                 if (segno % sbi->segs_per_sec)
 329                         goto next;
 330
 331                 /* skip if the section is currently used */
 332                 if (sec_usage_check(sbi, GET_SECNO(sbi, segno)))
 333                         goto next;
 334
 335                 /* skip if whole section is not prefree */
 336                 for (i = 1; i < sbi->segs_per_sec; i++)
 337                         if (!test_bit(segno + i, prefree_segmap))
 338                                 goto next;
 339
 340                 /* skip if whole section was not free at the last checkpoint */
 341                 for (i = 0; i < sbi->segs_per_sec; i++)
 342                         if (get_seg_entry(sbi, segno + i)->ckpt_valid_blocks)
 343                                 goto next;
 344
 345                 return segno;
 346         }
 347         return NULL_SEGNO;
 348 }
 349
 350 /*
 351  * Find a new segment from the free segments bitmap to right order
 352  * This function should be returned with success, otherwise BUG
 353  */
 354 static void get_new_segment(struct f2fs_sb_info *sbi,
 355                         unsigned int *newseg, bool new_sec, int dir)
 356 {
 357         struct free_segmap_info *free_i = FREE_I(sbi);
 358         unsigned int segno, secno, zoneno;
 359         unsigned int total_zones = TOTAL_SECS(sbi) / sbi->secs_per_zone;
 360         unsigned int hint = *newseg / sbi->segs_per_sec;
 361         unsigned int old_zoneno = GET_ZONENO_FROM_SEGNO(sbi, *newseg);
 362         unsigned int left_start = hint;
 363         bool init = true;
 364         int go_left = 0;
 365         int i;
 366
 367         write_lock(&free_i->segmap_lock);
 368
 369         if (!new_sec && ((*newseg + 1) % sbi->segs_per_sec)) {
 370                 segno = find_next_zero_bit(free_i->free_segmap,
 371                                         TOTAL_SEGS(sbi), *newseg + 1);
 372                 if (segno - *newseg < sbi->segs_per_sec -
 373                                         (*newseg % sbi->segs_per_sec))
 374                         goto got_it;
 375         }
 376 find_other_zone:
 377         secno = find_next_zero_bit(free_i->free_secmap, TOTAL_SECS(sbi), hint);
 378         if (secno >= TOTAL_SECS(sbi)) {
 379                 if (dir == ALLOC_RIGHT) {
 380                         secno = find_next_zero_bit(free_i->free_secmap,
 381                                                         TOTAL_SECS(sbi), 0);
 382                         BUG_ON(secno >= TOTAL_SECS(sbi));
 383                 } else {
 384                         go_left = 1;
 385                         left_start = hint - 1;
 386                 }
 387         }
 388         if (go_left == 0)
 389                 goto skip_left;
 390
 391         while (test_bit(left_start, free_i->free_secmap)) {
 392                 if (left_start > 0) {
 393                         left_start--;
 394                         continue;
 395                 }
 396                 left_start = find_next_zero_bit(free_i->free_secmap,
 397                                                         TOTAL_SECS(sbi), 0);
 398                 BUG_ON(left_start >= TOTAL_SECS(sbi));
 399                 break;
 400         }
 401         secno = left_start;
 402 skip_left:
 403         hint = secno;
 404         segno = secno * sbi->segs_per_sec;
 405         zoneno = secno / sbi->secs_per_zone;
 406
 407         /* give up on finding another zone */
 408         if (!init)
 409                 goto got_it;
 410         if (sbi->secs_per_zone == 1)
 411                 goto got_it;
 412         if (zoneno == old_zoneno)
 413                 goto got_it;
 414         if (dir == ALLOC_LEFT) {
 415                 if (!go_left && zoneno + 1 >= total_zones)
 416                         goto got_it;
 417                 if (go_left && zoneno == 0)
 418                         goto got_it;
 419         }
 420         for (i = 0; i < NR_CURSEG_TYPE; i++)
 421                 if (CURSEG_I(sbi, i)->zone == zoneno)
 422                         break;
 423
 424         if (i < NR_CURSEG_TYPE) {
 425                 /* zone is in user, try another */
 426                 if (go_left)
 427                         hint = zoneno * sbi->secs_per_zone - 1;
 428                 else if (zoneno + 1 >= total_zones)
 429                         hint = 0;
 430                 else
 431                         hint = (zoneno + 1) * sbi->secs_per_zone;
 432                 init = false;
 433                 goto find_other_zone;
 434         }
 435 got_it:
 436         /* set it as dirty segment in free segmap */
 437         BUG_ON(test_bit(segno, free_i->free_segmap));
 438         __set_inuse(sbi, segno);
 439         *newseg = segno;
 440         write_unlock(&free_i->segmap_lock);
 441 }
 442
 443 static void reset_curseg(struct f2fs_sb_info *sbi, int type, int modified)
 444 {
 445         struct curseg_info *curseg = CURSEG_I(sbi, type);
 446         struct summary_footer *sum_footer;
 447
 448         curseg->segno = curseg->next_segno;
 449         curseg->zone = GET_ZONENO_FROM_SEGNO(sbi, curseg->segno);
 450         curseg->next_blkoff = 0;
 451         curseg->next_segno = NULL_SEGNO;
 452
 453         sum_footer = &(curseg->sum_blk->footer);
 454         memset(sum_footer, 0, sizeof(struct summary_footer));
 455         if (IS_DATASEG(type))
 456                 SET_SUM_TYPE(sum_footer, SUM_TYPE_DATA);
 457         if (IS_NODESEG(type))
 458                 SET_SUM_TYPE(sum_footer, SUM_TYPE_NODE);
 459         __set_sit_entry_type(sbi, type, curseg->segno, modified);
 460 }
 461
 462 /*
 463  * Allocate a current working segment.
 464  * This function always allocates a free segment in LFS manner.
 465  */
 466 static void new_curseg(struct f2fs_sb_info *sbi, int type, bool new_sec)
 467 {
 468         struct curseg_info *curseg = CURSEG_I(sbi, type);
 469         unsigned int segno = curseg->segno;
 470         int dir = ALLOC_LEFT;
 471
 472         write_sum_page(sbi, curseg->sum_blk,
 473                                 GET_SUM_BLOCK(sbi, curseg->segno));
 474         if (type == CURSEG_WARM_DATA || type == CURSEG_COLD_DATA)
 475                 dir = ALLOC_RIGHT;
 476
 477         if (test_opt(sbi, NOHEAP))
 478                 dir = ALLOC_RIGHT;
 479
 480         get_new_segment(sbi, &segno, new_sec, dir);
 481         curseg->next_segno = segno;
 482         reset_curseg(sbi, type, 1);
 483         curseg->alloc_type = LFS;
 484 }
 485
 486 static void __next_free_blkoff(struct f2fs_sb_info *sbi,
 487                         struct curseg_info *seg, block_t start)
 488 {
 489         struct seg_entry *se = get_seg_entry(sbi, seg->segno);
 490         block_t ofs;
 491         for (ofs = start; ofs < sbi->blocks_per_seg; ofs++) {
 492                 if (!f2fs_test_bit(ofs, se->ckpt_valid_map)
 493                         && !f2fs_test_bit(ofs, se->cur_valid_map))
 494                         break;
 495         }
 496         seg->next_blkoff = ofs;
 497 }
 498
 499 /*
 500  * If a segment is written by LFS manner, next block offset is just obtained
 501  * by increasing the current block offset. However, if a segment is written by
 502  * SSR manner, next block offset obtained by calling __next_free_blkoff
 503  */
 504 static void __refresh_next_blkoff(struct f2fs_sb_info *sbi,
 505                                 struct curseg_info *seg)
 506 {
 507         if (seg->alloc_type == SSR)
 508                 __next_free_blkoff(sbi, seg, seg->next_blkoff + 1);
 509         else
 510                 seg->next_blkoff++;
 511 }
 512
 513 /*
 514  * This function always allocates a used segment (from dirty seglist) by SSR
 515  * manner, so it should recover the existing segment information of valid blocks
 516  */
 517 static void change_curseg(struct f2fs_sb_info *sbi, int type, bool reuse)
 518 {
 519         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
 520         struct curseg_info *curseg = CURSEG_I(sbi, type);
 521         unsigned int new_segno = curseg->next_segno;
 522         struct f2fs_summary_block *sum_node;
 523         struct page *sum_page;
 524
 525         write_sum_page(sbi, curseg->sum_blk,
 526                                 GET_SUM_BLOCK(sbi, curseg->segno));
 527         __set_test_and_inuse(sbi, new_segno);
 528
 529         mutex_lock(&dirty_i->seglist_lock);
 530         __remove_dirty_segment(sbi, new_segno, PRE);
 531         __remove_dirty_segment(sbi, new_segno, DIRTY);
 532         mutex_unlock(&dirty_i->seglist_lock);
 533
 534         reset_curseg(sbi, type, 1);
 535         curseg->alloc_type = SSR;
 536         __next_free_blkoff(sbi, curseg, 0);
 537
 538         if (reuse) {
 539                 sum_page = get_sum_page(sbi, new_segno);
 540                 sum_node = (struct f2fs_summary_block *)page_address(sum_page);
 541                 memcpy(curseg->sum_blk, sum_node, SUM_ENTRY_SIZE);
 542                 f2fs_put_page(sum_page, 1);
 543         }
 544 }
 545
 546 static int get_ssr_segment(struct f2fs_sb_info *sbi, int type)
 547 {
 548         struct curseg_info *curseg = CURSEG_I(sbi, type);
 549         const struct victim_selection *v_ops = DIRTY_I(sbi)->v_ops;
 550
 551         if (IS_NODESEG(type) || !has_not_enough_free_secs(sbi, 0))
 552                 return v_ops->get_victim(sbi,
 553                                 &(curseg)->next_segno, BG_GC, type, SSR);
 554
 555         /* For data segments, let's do SSR more intensively */
 556         for (; type >= CURSEG_HOT_DATA; type--)
 557                 if (v_ops->get_victim(sbi, &(curseg)->next_segno,
 558                                                 BG_GC, type, SSR))
 559                         return 1;
 560         return 0;
 561 }
 562
 563 /*
 564  * flush out current segment and replace it with new segment
 565  * This function should be returned with success, otherwise BUG
 566  */
 567 static void allocate_segment_by_default(struct f2fs_sb_info *sbi,
 568                                                 int type, bool force)
 569 {
 570         struct curseg_info *curseg = CURSEG_I(sbi, type);
 571
 572         if (force) {
 573                 new_curseg(sbi, type, true);
 574                 goto out;
 575         }
 576
 577         curseg->next_segno = check_prefree_segments(sbi, type);
 578
 579         if (curseg->next_segno != NULL_SEGNO)
 580                 change_curseg(sbi, type, false);
 581         else if (type == CURSEG_WARM_NODE)
 582                 new_curseg(sbi, type, false);
 583         else if (need_SSR(sbi) && get_ssr_segment(sbi, type))
 584                 change_curseg(sbi, type, true);
 585         else
 586                 new_curseg(sbi, type, false);
 587 out:
 588         sbi->segment_count[curseg->alloc_type]++;
 589 }
 590
 591 void allocate_new_segments(struct f2fs_sb_info *sbi)
 592 {
 593         struct curseg_info *curseg;
 594         unsigned int old_curseg;
 595         int i;
 596
 597         for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
 598                 curseg = CURSEG_I(sbi, i);
 599                 old_curseg = curseg->segno;
 600                 SIT_I(sbi)->s_ops->allocate_segment(sbi, i, true);
 601                 locate_dirty_segment(sbi, old_curseg);
 602         }
 603 }
 604
 605 static const struct segment_allocation default_salloc_ops = {
 606         .allocate_segment = allocate_segment_by_default,
 607 };
 608
 609 static void f2fs_end_io_write(struct bio *bio, int err)
 610 {
 611         const int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
 612         struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
 613         struct bio_private *p = bio->bi_private;
 614
 615         do {
 616                 struct page *page = bvec->bv_page;
 617
 618                 if (--bvec >= bio->bi_io_vec)
 619                         prefetchw(&bvec->bv_page->flags);
 620                 if (!uptodate) {
 621                         SetPageError(page);
 622                         if (page->mapping)
 623                                 set_bit(AS_EIO, &page->mapping->flags);
 624                         set_ckpt_flags(p->sbi->ckpt, CP_ERROR_FLAG);
 625                         p->sbi->sb->s_flags |= MS_RDONLY;
 626                 }
 627                 end_page_writeback(page);
 628                 dec_page_count(p->sbi, F2FS_WRITEBACK);
 629         } while (bvec >= bio->bi_io_vec);
 630
 631         if (p->is_sync)
 632                 complete(p->wait);
 633         kfree(p);
 634         bio_put(bio);
 635 }
 636
 637 struct bio *f2fs_bio_alloc(struct block_device *bdev, int npages)
 638 {
 639         struct bio *bio;
 640         struct bio_private *priv;
 641 retry:
 642         priv = kmalloc(sizeof(struct bio_private), GFP_NOFS);
 643         if (!priv) {
 644                 cond_resched();
 645                 goto retry;
 646         }
 647
 648         /* No failure on bio allocation */
 649         bio = bio_alloc(GFP_NOIO, npages);
 650         bio->bi_bdev = bdev;
 651         bio->bi_private = priv;
 652         return bio;
 653 }
 654
 655 static void do_submit_bio(struct f2fs_sb_info *sbi,
 656                                 enum page_type type, bool sync)
 657 {
 658         int rw = sync ? WRITE_SYNC : WRITE;
 659         enum page_type btype = type > META ? META : type;
 660
 661         if (type >= META_FLUSH)
 662                 rw = WRITE_FLUSH_FUA;
 663
 664         if (sbi->bio[btype]) {
 665                 struct bio_private *p = sbi->bio[btype]->bi_private;
 666                 p->sbi = sbi;
 667                 sbi->bio[btype]->bi_end_io = f2fs_end_io_write;
 668                 if (type == META_FLUSH) {
 669                         DECLARE_COMPLETION_ONSTACK(wait);
 670                         p->is_sync = true;
 671                         p->wait = &wait;
 672                         submit_bio(rw, sbi->bio[btype]);
 673                         wait_for_completion(&wait);
 674                 } else {
 675                         p->is_sync = false;
 676                         submit_bio(rw, sbi->bio[btype]);
 677                 }
 678                 sbi->bio[btype] = NULL;
 679         }
 680 }
 681
 682 void f2fs_submit_bio(struct f2fs_sb_info *sbi, enum page_type type, bool sync)
 683 {
 684         down_write(&sbi->bio_sem);
 685         do_submit_bio(sbi, type, sync);
 686         up_write(&sbi->bio_sem);
 687 }
 688
 689 static void submit_write_page(struct f2fs_sb_info *sbi, struct page *page,
 690                                 block_t blk_addr, enum page_type type)
 691 {
 692         struct block_device *bdev = sbi->sb->s_bdev;
 693
 694         verify_block_addr(sbi, blk_addr);
 695
 696         down_write(&sbi->bio_sem);
 697
 698         inc_page_count(sbi, F2FS_WRITEBACK);
 699
 700         if (sbi->bio[type] && sbi->last_block_in_bio[type] != blk_addr - 1)
 701                 do_submit_bio(sbi, type, false);
 702 alloc_new:
 703         if (sbi->bio[type] == NULL) {
 704                 sbi->bio[type] = f2fs_bio_alloc(bdev, bio_get_nr_vecs(bdev));
 705                 sbi->bio[type]->bi_sector = SECTOR_FROM_BLOCK(sbi, blk_addr);
 706                 /*
 707                  * The end_io will be assigned at the sumbission phase.
 708                  * Until then, let bio_add_page() merge consecutive IOs as much
 709                  * as possible.
 710                  */
 711         }
 712
 713         if (bio_add_page(sbi->bio[type], page, PAGE_CACHE_SIZE, 0) <
 714                                                         PAGE_CACHE_SIZE) {
 715                 do_submit_bio(sbi, type, false);
 716                 goto alloc_new;
 717         }
 718
 719         sbi->last_block_in_bio[type] = blk_addr;
 720
 721         up_write(&sbi->bio_sem);
 722 }
 723
 724 static bool __has_curseg_space(struct f2fs_sb_info *sbi, int type)
 725 {
 726         struct curseg_info *curseg = CURSEG_I(sbi, type);
 727         if (curseg->next_blkoff < sbi->blocks_per_seg)
 728                 return true;
 729         return false;
 730 }
 731
 732 static int __get_segment_type_2(struct page *page, enum page_type p_type)
 733 {
 734         if (p_type == DATA)
 735                 return CURSEG_HOT_DATA;
 736         else
 737                 return CURSEG_HOT_NODE;
 738 }
 739
 740 static int __get_segment_type_4(struct page *page, enum page_type p_type)
 741 {
 742         if (p_type == DATA) {
 743                 struct inode *inode = page->mapping->host;
 744
 745                 if (S_ISDIR(inode->i_mode))
 746                         return CURSEG_HOT_DATA;
 747                 else
 748                         return CURSEG_COLD_DATA;
 749         } else {
 750                 if (IS_DNODE(page) && !is_cold_node(page))
 751                         return CURSEG_HOT_NODE;
 752                 else
 753                         return CURSEG_COLD_NODE;
 754         }
 755 }
 756
 757 static int __get_segment_type_6(struct page *page, enum page_type p_type)
 758 {
 759         if (p_type == DATA) {
 760                 struct inode *inode = page->mapping->host;
 761
 762                 if (S_ISDIR(inode->i_mode))
 763                         return CURSEG_HOT_DATA;
 764                 else if (is_cold_data(page) || is_cold_file(inode))
 765                         return CURSEG_COLD_DATA;
 766                 else
 767                         return CURSEG_WARM_DATA;
 768         } else {
 769                 if (IS_DNODE(page))
 770                         return is_cold_node(page) ? CURSEG_WARM_NODE :
 771                                                 CURSEG_HOT_NODE;
 772                 else
 773                         return CURSEG_COLD_NODE;
 774         }
 775 }
 776
 777 static int __get_segment_type(struct page *page, enum page_type p_type)
 778 {
 779         struct f2fs_sb_info *sbi = F2FS_SB(page->mapping->host->i_sb);
 780         switch (sbi->active_logs) {
 781         case 2:
 782                 return __get_segment_type_2(page, p_type);
 783         case 4:
 784                 return __get_segment_type_4(page, p_type);
 785         }
 786         /* NR_CURSEG_TYPE(6) logs by default */
 787         BUG_ON(sbi->active_logs != NR_CURSEG_TYPE);
 788         return __get_segment_type_6(page, p_type);
 789 }
 790
 791 static void do_write_page(struct f2fs_sb_info *sbi, struct page *page,
 792                         block_t old_blkaddr, block_t *new_blkaddr,
 793                         struct f2fs_summary *sum, enum page_type p_type)
 794 {
 795         struct sit_info *sit_i = SIT_I(sbi);
 796         struct curseg_info *curseg;
 797         unsigned int old_cursegno;
 798         int type;
 799
 800         type = __get_segment_type(page, p_type);
 801         curseg = CURSEG_I(sbi, type);
 802
 803         mutex_lock(&curseg->curseg_mutex);
 804
 805         *new_blkaddr = NEXT_FREE_BLKADDR(sbi, curseg);
 806         old_cursegno = curseg->segno;
 807
 808         /*
 809          * __add_sum_entry should be resided under the curseg_mutex
 810          * because, this function updates a summary entry in the
 811          * current summary block.
 812          */
 813         __add_sum_entry(sbi, type, sum, curseg->next_blkoff);
 814
 815         mutex_lock(&sit_i->sentry_lock);
 816         __refresh_next_blkoff(sbi, curseg);
 817         sbi->block_count[curseg->alloc_type]++;
 818
 819         /*
 820          * SIT information should be updated before segment allocation,
 821          * since SSR needs latest valid block information.
 822          */
 823         refresh_sit_entry(sbi, old_blkaddr, *new_blkaddr);
 824
 825         if (!__has_curseg_space(sbi, type))
 826                 sit_i->s_ops->allocate_segment(sbi, type, false);
 827
 828         locate_dirty_segment(sbi, old_cursegno);
 829         locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr));
 830         mutex_unlock(&sit_i->sentry_lock);
 831
 832         if (p_type == NODE)
 833                 fill_node_footer_blkaddr(page, NEXT_FREE_BLKADDR(sbi, curseg));
 834
 835         /* writeout dirty page into bdev */
 836         submit_write_page(sbi, page, *new_blkaddr, p_type);
 837
 838         mutex_unlock(&curseg->curseg_mutex);
 839 }
 840
 841 void write_meta_page(struct f2fs_sb_info *sbi, struct page *page)
 842 {
 843         set_page_writeback(page);
 844         submit_write_page(sbi, page, page->index, META);
 845 }
 846
 847 void write_node_page(struct f2fs_sb_info *sbi, struct page *page,
 848                 unsigned int nid, block_t old_blkaddr, block_t *new_blkaddr)
 849 {
 850         struct f2fs_summary sum;
 851         set_summary(&sum, nid, 0, 0);
 852         do_write_page(sbi, page, old_blkaddr, new_blkaddr, &sum, NODE);
 853 }
 854
 855 void write_data_page(struct inode *inode, struct page *page,
 856                 struct dnode_of_data *dn, block_t old_blkaddr,
 857                 block_t *new_blkaddr)
 858 {
 859         struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
 860         struct f2fs_summary sum;
 861         struct node_info ni;
 862
 863         BUG_ON(old_blkaddr == NULL_ADDR);
 864         get_node_info(sbi, dn->nid, &ni);
 865         set_summary(&sum, dn->nid, dn->ofs_in_node, ni.version);
 866
 867         do_write_page(sbi, page, old_blkaddr,
 868                         new_blkaddr, &sum, DATA);
 869 }
 870
 871 void rewrite_data_page(struct f2fs_sb_info *sbi, struct page *page,
 872                                         block_t old_blk_addr)
 873 {
 874         submit_write_page(sbi, page, old_blk_addr, DATA);
 875 }
 876
 877 void recover_data_page(struct f2fs_sb_info *sbi,
 878                         struct page *page, struct f2fs_summary *sum,
 879                         block_t old_blkaddr, block_t new_blkaddr)
 880 {
 881         struct sit_info *sit_i = SIT_I(sbi);
 882         struct curseg_info *curseg;
 883         unsigned int segno, old_cursegno;
 884         struct seg_entry *se;
 885         int type;
 886
 887         segno = GET_SEGNO(sbi, new_blkaddr);
 888         se = get_seg_entry(sbi, segno);
 889         type = se->type;
 890
 891         if (se->valid_blocks == 0 && !IS_CURSEG(sbi, segno)) {
 892                 if (old_blkaddr == NULL_ADDR)
 893                         type = CURSEG_COLD_DATA;
 894                 else
 895                         type = CURSEG_WARM_DATA;
 896         }
 897         curseg = CURSEG_I(sbi, type);
 898
 899         mutex_lock(&curseg->curseg_mutex);
 900         mutex_lock(&sit_i->sentry_lock);
 901
 902         old_cursegno = curseg->segno;
 903
 904         /* change the current segment */
 905         if (segno != curseg->segno) {
 906                 curseg->next_segno = segno;
 907                 change_curseg(sbi, type, true);
 908         }
 909
 910         curseg->next_blkoff = GET_SEGOFF_FROM_SEG0(sbi, new_blkaddr) &
 911                                         (sbi->blocks_per_seg - 1);
 912         __add_sum_entry(sbi, type, sum, curseg->next_blkoff);
 913
 914         refresh_sit_entry(sbi, old_blkaddr, new_blkaddr);
 915
 916         locate_dirty_segment(sbi, old_cursegno);
 917         locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr));
 918
 919         mutex_unlock(&sit_i->sentry_lock);
 920         mutex_unlock(&curseg->curseg_mutex);
 921 }
 922
 923 void rewrite_node_page(struct f2fs_sb_info *sbi,
 924                         struct page *page, struct f2fs_summary *sum,
 925                         block_t old_blkaddr, block_t new_blkaddr)
 926 {
 927         struct sit_info *sit_i = SIT_I(sbi);
 928         int type = CURSEG_WARM_NODE;
 929         struct curseg_info *curseg;
 930         unsigned int segno, old_cursegno;
 931         block_t next_blkaddr = next_blkaddr_of_node(page);
 932         unsigned int next_segno = GET_SEGNO(sbi, next_blkaddr);
 933
 934         curseg = CURSEG_I(sbi, type);
 935
 936         mutex_lock(&curseg->curseg_mutex);
 937         mutex_lock(&sit_i->sentry_lock);
 938
 939         segno = GET_SEGNO(sbi, new_blkaddr);
 940         old_cursegno = curseg->segno;
 941
 942         /* change the current segment */
 943         if (segno != curseg->segno) {
 944                 curseg->next_segno = segno;
 945                 change_curseg(sbi, type, true);
 946         }
 947         curseg->next_blkoff = GET_SEGOFF_FROM_SEG0(sbi, new_blkaddr) &
 948                                         (sbi->blocks_per_seg - 1);
 949         __add_sum_entry(sbi, type, sum, curseg->next_blkoff);
 950
 951         /* change the current log to the next block addr in advance */
 952         if (next_segno != segno) {
 953                 curseg->next_segno = next_segno;
 954                 change_curseg(sbi, type, true);
 955         }
 956         curseg->next_blkoff = GET_SEGOFF_FROM_SEG0(sbi, next_blkaddr) &
 957                                         (sbi->blocks_per_seg - 1);
 958
 959         /* rewrite node page */
 960         set_page_writeback(page);
 961         submit_write_page(sbi, page, new_blkaddr, NODE);
 962         f2fs_submit_bio(sbi, NODE, true);
 963         refresh_sit_entry(sbi, old_blkaddr, new_blkaddr);
 964
 965         locate_dirty_segment(sbi, old_cursegno);
 966         locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr));
 967
 968         mutex_unlock(&sit_i->sentry_lock);
 969         mutex_unlock(&curseg->curseg_mutex);
 970 }
 971
 972 static int read_compacted_summaries(struct f2fs_sb_info *sbi)
 973 {
 974         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
 975         struct curseg_info *seg_i;
 976         unsigned char *kaddr;
 977         struct page *page;
 978         block_t start;
 979         int i, j, offset;
 980
 981         start = start_sum_block(sbi);
 982
 983         page = get_meta_page(sbi, start++);
 984         kaddr = (unsigned char *)page_address(page);
 985
 986         /* Step 1: restore nat cache */
 987         seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
 988         memcpy(&seg_i->sum_blk->n_nats, kaddr, SUM_JOURNAL_SIZE);
 989
 990         /* Step 2: restore sit cache */
 991         seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
 992         memcpy(&seg_i->sum_blk->n_sits, kaddr + SUM_JOURNAL_SIZE,
 993                                                 SUM_JOURNAL_SIZE);
 994         offset = 2 * SUM_JOURNAL_SIZE;
 995
 996         /* Step 3: restore summary entries */
 997         for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
 998                 unsigned short blk_off;
 999                 unsigned int segno;
1000
1001                 seg_i = CURSEG_I(sbi, i);
1002                 segno = le32_to_cpu(ckpt->cur_data_segno[i]);
1003                 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[i]);
1004                 seg_i->next_segno = segno;
1005                 reset_curseg(sbi, i, 0);
1006                 seg_i->alloc_type = ckpt->alloc_type[i];
1007                 seg_i->next_blkoff = blk_off;
1008
1009                 if (seg_i->alloc_type == SSR)
1010                         blk_off = sbi->blocks_per_seg;
1011
1012                 for (j = 0; j < blk_off; j++) {
1013                         struct f2fs_summary *s;
1014                         s = (struct f2fs_summary *)(kaddr + offset);
1015                         seg_i->sum_blk->entries[j] = *s;
1016                         offset += SUMMARY_SIZE;
1017                         if (offset + SUMMARY_SIZE <= PAGE_CACHE_SIZE -
1018                                                 SUM_FOOTER_SIZE)
1019                                 continue;
1020
1021                         f2fs_put_page(page, 1);
1022                         page = NULL;
1023
1024                         page = get_meta_page(sbi, start++);
1025                         kaddr = (unsigned char *)page_address(page);
1026                         offset = 0;
1027                 }
1028         }
1029         f2fs_put_page(page, 1);
1030         return 0;
1031 }
1032
1033 static int read_normal_summaries(struct f2fs_sb_info *sbi, int type)
1034 {
1035         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1036         struct f2fs_summary_block *sum;
1037         struct curseg_info *curseg;
1038         struct page *new;
1039         unsigned short blk_off;
1040         unsigned int segno = 0;
1041         block_t blk_addr = 0;
1042
1043         /* get segment number and block addr */
1044         if (IS_DATASEG(type)) {
1045                 segno = le32_to_cpu(ckpt->cur_data_segno[type]);
1046                 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[type -
1047                                                         CURSEG_HOT_DATA]);
1048                 if (is_set_ckpt_flags(ckpt, CP_UMOUNT_FLAG))
1049                         blk_addr = sum_blk_addr(sbi, NR_CURSEG_TYPE, type);
1050                 else
1051                         blk_addr = sum_blk_addr(sbi, NR_CURSEG_DATA_TYPE, type);
1052         } else {
1053                 segno = le32_to_cpu(ckpt->cur_node_segno[type -
1054                                                         CURSEG_HOT_NODE]);
1055                 blk_off = le16_to_cpu(ckpt->cur_node_blkoff[type -
1056                                                         CURSEG_HOT_NODE]);
1057                 if (is_set_ckpt_flags(ckpt, CP_UMOUNT_FLAG))
1058                         blk_addr = sum_blk_addr(sbi, NR_CURSEG_NODE_TYPE,
1059                                                         type - CURSEG_HOT_NODE);
1060                 else
1061                         blk_addr = GET_SUM_BLOCK(sbi, segno);
1062         }
1063
1064         new = get_meta_page(sbi, blk_addr);
1065         sum = (struct f2fs_summary_block *)page_address(new);
1066
1067         if (IS_NODESEG(type)) {
1068                 if (is_set_ckpt_flags(ckpt, CP_UMOUNT_FLAG)) {
1069                         struct f2fs_summary *ns = &sum->entries[0];
1070                         int i;
1071                         for (i = 0; i < sbi->blocks_per_seg; i++, ns++) {
1072                                 ns->version = 0;
1073                                 ns->ofs_in_node = 0;
1074                         }
1075                 } else {
1076                         if (restore_node_summary(sbi, segno, sum)) {
1077                                 f2fs_put_page(new, 1);
1078                                 return -EINVAL;
1079                         }
1080                 }
1081         }
1082
1083         /* set uncompleted segment to curseg */
1084         curseg = CURSEG_I(sbi, type);
1085         mutex_lock(&curseg->curseg_mutex);
1086         memcpy(curseg->sum_blk, sum, PAGE_CACHE_SIZE);
1087         curseg->next_segno = segno;
1088         reset_curseg(sbi, type, 0);
1089         curseg->alloc_type = ckpt->alloc_type[type];
1090         curseg->next_blkoff = blk_off;
1091         mutex_unlock(&curseg->curseg_mutex);
1092         f2fs_put_page(new, 1);
1093         return 0;
1094 }
1095
1096 static int restore_curseg_summaries(struct f2fs_sb_info *sbi)
1097 {
1098         int type = CURSEG_HOT_DATA;
1099
1100         if (is_set_ckpt_flags(F2FS_CKPT(sbi), CP_COMPACT_SUM_FLAG)) {
1101                 /* restore for compacted data summary */
1102                 if (read_compacted_summaries(sbi))
1103                         return -EINVAL;
1104                 type = CURSEG_HOT_NODE;
1105         }
1106
1107         for (; type <= CURSEG_COLD_NODE; type++)
1108                 if (read_normal_summaries(sbi, type))
1109                         return -EINVAL;
1110         return 0;
1111 }
1112
1113 static void write_compacted_summaries(struct f2fs_sb_info *sbi, block_t blkaddr)
1114 {
1115         struct page *page;
1116         unsigned char *kaddr;
1117         struct f2fs_summary *summary;
1118         struct curseg_info *seg_i;
1119         int written_size = 0;
1120         int i, j;
1121
1122         page = grab_meta_page(sbi, blkaddr++);
1123         kaddr = (unsigned char *)page_address(page);
1124
1125         /* Step 1: write nat cache */
1126         seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
1127         memcpy(kaddr, &seg_i->sum_blk->n_nats, SUM_JOURNAL_SIZE);
1128         written_size += SUM_JOURNAL_SIZE;
1129
1130         /* Step 2: write sit cache */
1131         seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
1132         memcpy(kaddr + written_size, &seg_i->sum_blk->n_sits,
1133                                                 SUM_JOURNAL_SIZE);
1134         written_size += SUM_JOURNAL_SIZE;
1135
1136         set_page_dirty(page);
1137
1138         /* Step 3: write summary entries */
1139         for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
1140                 unsigned short blkoff;
1141                 seg_i = CURSEG_I(sbi, i);
1142                 if (sbi->ckpt->alloc_type[i] == SSR)
1143                         blkoff = sbi->blocks_per_seg;
1144                 else
1145                         blkoff = curseg_blkoff(sbi, i);
1146
1147                 for (j = 0; j < blkoff; j++) {
1148                         if (!page) {
1149                                 page = grab_meta_page(sbi, blkaddr++);
1150                                 kaddr = (unsigned char *)page_address(page);
1151                                 written_size = 0;
1152                         }
1153                         summary = (struct f2fs_summary *)(kaddr + written_size);
1154                         *summary = seg_i->sum_blk->entries[j];
1155                         written_size += SUMMARY_SIZE;
1156                         set_page_dirty(page);
1157
1158                         if (written_size + SUMMARY_SIZE <= PAGE_CACHE_SIZE -
1159                                                         SUM_FOOTER_SIZE)
1160                                 continue;
1161
1162                         f2fs_put_page(page, 1);
1163                         page = NULL;
1164                 }
1165         }
1166         if (page)
1167                 f2fs_put_page(page, 1);
1168 }
1169
1170 static void write_normal_summaries(struct f2fs_sb_info *sbi,
1171                                         block_t blkaddr, int type)
1172 {
1173         int i, end;
1174         if (IS_DATASEG(type))
1175                 end = type + NR_CURSEG_DATA_TYPE;
1176         else
1177                 end = type + NR_CURSEG_NODE_TYPE;
1178
1179         for (i = type; i < end; i++) {
1180                 struct curseg_info *sum = CURSEG_I(sbi, i);
1181                 mutex_lock(&sum->curseg_mutex);
1182                 write_sum_page(sbi, sum->sum_blk, blkaddr + (i - type));
1183                 mutex_unlock(&sum->curseg_mutex);
1184         }
1185 }
1186
1187 void write_data_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
1188 {
1189         if (is_set_ckpt_flags(F2FS_CKPT(sbi), CP_COMPACT_SUM_FLAG))
1190                 write_compacted_summaries(sbi, start_blk);
1191         else
1192                 write_normal_summaries(sbi, start_blk, CURSEG_HOT_DATA);
1193 }
1194
1195 void write_node_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
1196 {
1197         if (is_set_ckpt_flags(F2FS_CKPT(sbi), CP_UMOUNT_FLAG))
1198                 write_normal_summaries(sbi, start_blk, CURSEG_HOT_NODE);
1199         return;
1200 }
1201
1202 int lookup_journal_in_cursum(struct f2fs_summary_block *sum, int type,
1203                                         unsigned int val, int alloc)
1204 {
1205         int i;
1206
1207         if (type == NAT_JOURNAL) {
1208                 for (i = 0; i < nats_in_cursum(sum); i++) {
1209                         if (le32_to_cpu(nid_in_journal(sum, i)) == val)
1210                                 return i;
1211                 }
1212                 if (alloc && nats_in_cursum(sum) < NAT_JOURNAL_ENTRIES)
1213                         return update_nats_in_cursum(sum, 1);
1214         } else if (type == SIT_JOURNAL) {
1215                 for (i = 0; i < sits_in_cursum(sum); i++)
1216                         if (le32_to_cpu(segno_in_journal(sum, i)) == val)
1217                                 return i;
1218                 if (alloc && sits_in_cursum(sum) < SIT_JOURNAL_ENTRIES)
1219                         return update_sits_in_cursum(sum, 1);
1220         }
1221         return -1;
1222 }
1223
1224 static struct page *get_current_sit_page(struct f2fs_sb_info *sbi,
1225                                         unsigned int segno)
1226 {
1227         struct sit_info *sit_i = SIT_I(sbi);
1228         unsigned int offset = SIT_BLOCK_OFFSET(sit_i, segno);
1229         block_t blk_addr = sit_i->sit_base_addr + offset;
1230
1231         check_seg_range(sbi, segno);
1232
1233         /* calculate sit block address */
1234         if (f2fs_test_bit(offset, sit_i->sit_bitmap))
1235                 blk_addr += sit_i->sit_blocks;
1236
1237         return get_meta_page(sbi, blk_addr);
1238 }
1239
1240 static struct page *get_next_sit_page(struct f2fs_sb_info *sbi,
1241                                         unsigned int start)
1242 {
1243         struct sit_info *sit_i = SIT_I(sbi);
1244         struct page *src_page, *dst_page;
1245         pgoff_t src_off, dst_off;
1246         void *src_addr, *dst_addr;
1247
1248         src_off = current_sit_addr(sbi, start);
1249         dst_off = next_sit_addr(sbi, src_off);
1250
1251         /* get current sit block page without lock */
1252         src_page = get_meta_page(sbi, src_off);
1253         dst_page = grab_meta_page(sbi, dst_off);
1254         BUG_ON(PageDirty(src_page));
1255
1256         src_addr = page_address(src_page);
1257         dst_addr = page_address(dst_page);
1258         memcpy(dst_addr, src_addr, PAGE_CACHE_SIZE);
1259
1260         set_page_dirty(dst_page);
1261         f2fs_put_page(src_page, 1);
1262
1263         set_to_next_sit(sit_i, start);
1264
1265         return dst_page;
1266 }
1267
1268 static bool flush_sits_in_journal(struct f2fs_sb_info *sbi)
1269 {
1270         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
1271         struct f2fs_summary_block *sum = curseg->sum_blk;
1272         int i;
1273
1274         /*
1275          * If the journal area in the current summary is full of sit entries,
1276          * all the sit entries will be flushed. Otherwise the sit entries
1277          * are not able to replace with newly hot sit entries.
1278          */
1279         if (sits_in_cursum(sum) >= SIT_JOURNAL_ENTRIES) {
1280                 for (i = sits_in_cursum(sum) - 1; i >= 0; i--) {
1281                         unsigned int segno;
1282                         segno = le32_to_cpu(segno_in_journal(sum, i));
1283                         __mark_sit_entry_dirty(sbi, segno);
1284                 }
1285                 update_sits_in_cursum(sum, -sits_in_cursum(sum));
1286                 return 1;
1287         }
1288         return 0;
1289 }
1290
1291 /*
1292  * CP calls this function, which flushes SIT entries including sit_journal,
1293  * and moves prefree segs to free segs.
1294  */
1295 void flush_sit_entries(struct f2fs_sb_info *sbi)
1296 {
1297         struct sit_info *sit_i = SIT_I(sbi);
1298         unsigned long *bitmap = sit_i->dirty_sentries_bitmap;
1299         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
1300         struct f2fs_summary_block *sum = curseg->sum_blk;
1301         unsigned long nsegs = TOTAL_SEGS(sbi);
1302         struct page *page = NULL;
1303         struct f2fs_sit_block *raw_sit = NULL;
1304         unsigned int start = 0, end = 0;
1305         unsigned int segno = -1;
1306         bool flushed;
1307
1308         mutex_lock(&curseg->curseg_mutex);
1309         mutex_lock(&sit_i->sentry_lock);
1310
1311         /*
1312          * "flushed" indicates whether sit entries in journal are flushed
1313          * to the SIT area or not.
1314          */
1315         flushed = flush_sits_in_journal(sbi);
1316
1317         while ((segno = find_next_bit(bitmap, nsegs, segno + 1)) < nsegs) {
1318                 struct seg_entry *se = get_seg_entry(sbi, segno);
1319                 int sit_offset, offset;
1320
1321                 sit_offset = SIT_ENTRY_OFFSET(sit_i, segno);
1322
1323                 if (flushed)
1324                         goto to_sit_page;
1325
1326                 offset = lookup_journal_in_cursum(sum, SIT_JOURNAL, segno, 1);
1327                 if (offset >= 0) {
1328                         segno_in_journal(sum, offset) = cpu_to_le32(segno);
1329                         seg_info_to_raw_sit(se, &sit_in_journal(sum, offset));
1330                         goto flush_done;
1331                 }
1332 to_sit_page:
1333                 if (!page || (start > segno) || (segno > end)) {
1334                         if (page) {
1335                                 f2fs_put_page(page, 1);
1336                                 page = NULL;
1337                         }
1338
1339                         start = START_SEGNO(sit_i, segno);
1340                         end = start + SIT_ENTRY_PER_BLOCK - 1;
1341
1342                         /* read sit block that will be updated */
1343                         page = get_next_sit_page(sbi, start);
1344                         raw_sit = page_address(page);
1345                 }
1346
1347                 /* udpate entry in SIT block */
1348                 seg_info_to_raw_sit(se, &raw_sit->entries[sit_offset]);
1349 flush_done:
1350                 __clear_bit(segno, bitmap);
1351                 sit_i->dirty_sentries--;
1352         }
1353         mutex_unlock(&sit_i->sentry_lock);
1354         mutex_unlock(&curseg->curseg_mutex);
1355
1356         /* writeout last modified SIT block */
1357         f2fs_put_page(page, 1);
1358
1359         set_prefree_as_free_segments(sbi);
1360 }
1361
1362 static int build_sit_info(struct f2fs_sb_info *sbi)
1363 {
1364         struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
1365         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1366         struct sit_info *sit_i;
1367         unsigned int sit_segs, start;
1368         char *src_bitmap, *dst_bitmap;
1369         unsigned int bitmap_size;
1370
1371         /* allocate memory for SIT information */
1372         sit_i = kzalloc(sizeof(struct sit_info), GFP_KERNEL);
1373         if (!sit_i)
1374                 return -ENOMEM;
1375
1376         SM_I(sbi)->sit_info = sit_i;
1377
1378         sit_i->sentries = vzalloc(TOTAL_SEGS(sbi) * sizeof(struct seg_entry));
1379         if (!sit_i->sentries)
1380                 return -ENOMEM;
1381
1382         bitmap_size = f2fs_bitmap_size(TOTAL_SEGS(sbi));
1383         sit_i->dirty_sentries_bitmap = kzalloc(bitmap_size, GFP_KERNEL);
1384         if (!sit_i->dirty_sentries_bitmap)
1385                 return -ENOMEM;
1386
1387         for (start = 0; start < TOTAL_SEGS(sbi); start++) {
1388                 sit_i->sentries[start].cur_valid_map
1389                         = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
1390                 sit_i->sentries[start].ckpt_valid_map
1391                         = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
1392                 if (!sit_i->sentries[start].cur_valid_map
1393                                 || !sit_i->sentries[start].ckpt_valid_map)
1394                         return -ENOMEM;
1395         }
1396
1397         if (sbi->segs_per_sec > 1) {
1398                 sit_i->sec_entries = vzalloc(TOTAL_SECS(sbi) *
1399                                         sizeof(struct sec_entry));
1400                 if (!sit_i->sec_entries)
1401                         return -ENOMEM;
1402         }
1403
1404         /* get information related with SIT */
1405         sit_segs = le32_to_cpu(raw_super->segment_count_sit) >> 1;
1406
1407         /* setup SIT bitmap from ckeckpoint pack */
1408         bitmap_size = __bitmap_size(sbi, SIT_BITMAP);
1409         src_bitmap = __bitmap_ptr(sbi, SIT_BITMAP);
1410
1411         dst_bitmap = kmemdup(src_bitmap, bitmap_size, GFP_KERNEL);
1412         if (!dst_bitmap)
1413                 return -ENOMEM;
1414
1415         /* init SIT information */
1416         sit_i->s_ops = &default_salloc_ops;
1417
1418         sit_i->sit_base_addr = le32_to_cpu(raw_super->sit_blkaddr);
1419         sit_i->sit_blocks = sit_segs << sbi->log_blocks_per_seg;
1420         sit_i->written_valid_blocks = le64_to_cpu(ckpt->valid_block_count);
1421         sit_i->sit_bitmap = dst_bitmap;
1422         sit_i->bitmap_size = bitmap_size;
1423         sit_i->dirty_sentries = 0;
1424         sit_i->sents_per_block = SIT_ENTRY_PER_BLOCK;
1425         sit_i->elapsed_time = le64_to_cpu(sbi->ckpt->elapsed_time);
1426         sit_i->mounted_time = CURRENT_TIME_SEC.tv_sec;
1427         mutex_init(&sit_i->sentry_lock);
1428         return 0;
1429 }
1430
1431 static int build_free_segmap(struct f2fs_sb_info *sbi)
1432 {
1433         struct f2fs_sm_info *sm_info = SM_I(sbi);
1434         struct free_segmap_info *free_i;
1435         unsigned int bitmap_size, sec_bitmap_size;
1436
1437         /* allocate memory for free segmap information */
1438         free_i = kzalloc(sizeof(struct free_segmap_info), GFP_KERNEL);
1439         if (!free_i)
1440                 return -ENOMEM;
1441
1442         SM_I(sbi)->free_info = free_i;
1443
1444         bitmap_size = f2fs_bitmap_size(TOTAL_SEGS(sbi));
1445         free_i->free_segmap = kmalloc(bitmap_size, GFP_KERNEL);
1446         if (!free_i->free_segmap)
1447                 return -ENOMEM;
1448
1449         sec_bitmap_size = f2fs_bitmap_size(TOTAL_SECS(sbi));
1450         free_i->free_secmap = kmalloc(sec_bitmap_size, GFP_KERNEL);
1451         if (!free_i->free_secmap)
1452                 return -ENOMEM;
1453
1454         /* set all segments as dirty temporarily */
1455         memset(free_i->free_segmap, 0xff, bitmap_size);
1456         memset(free_i->free_secmap, 0xff, sec_bitmap_size);
1457
1458         /* init free segmap information */
1459         free_i->start_segno =
1460                 (unsigned int) GET_SEGNO_FROM_SEG0(sbi, sm_info->main_blkaddr);
1461         free_i->free_segments = 0;
1462         free_i->free_sections = 0;
1463         rwlock_init(&free_i->segmap_lock);
1464         return 0;
1465 }
1466
1467 static int build_curseg(struct f2fs_sb_info *sbi)
1468 {
1469         struct curseg_info *array;
1470         int i;
1471
1472         array = kzalloc(sizeof(*array) * NR_CURSEG_TYPE, GFP_KERNEL);
1473         if (!array)
1474                 return -ENOMEM;
1475
1476         SM_I(sbi)->curseg_array = array;
1477
1478         for (i = 0; i < NR_CURSEG_TYPE; i++) {
1479                 mutex_init(&array[i].curseg_mutex);
1480                 array[i].sum_blk = kzalloc(PAGE_CACHE_SIZE, GFP_KERNEL);
1481                 if (!array[i].sum_blk)
1482                         return -ENOMEM;
1483                 array[i].segno = NULL_SEGNO;
1484                 array[i].next_blkoff = 0;
1485         }
1486         return restore_curseg_summaries(sbi);
1487 }
1488
1489 static void build_sit_entries(struct f2fs_sb_info *sbi)
1490 {
1491         struct sit_info *sit_i = SIT_I(sbi);
1492         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
1493         struct f2fs_summary_block *sum = curseg->sum_blk;
1494         unsigned int start;
1495
1496         for (start = 0; start < TOTAL_SEGS(sbi); start++) {
1497                 struct seg_entry *se = &sit_i->sentries[start];
1498                 struct f2fs_sit_block *sit_blk;
1499                 struct f2fs_sit_entry sit;
1500                 struct page *page;
1501                 int i;
1502
1503                 mutex_lock(&curseg->curseg_mutex);
1504                 for (i = 0; i < sits_in_cursum(sum); i++) {
1505                         if (le32_to_cpu(segno_in_journal(sum, i)) == start) {
1506                                 sit = sit_in_journal(sum, i);
1507                                 mutex_unlock(&curseg->curseg_mutex);
1508                                 goto got_it;
1509                         }
1510                 }
1511                 mutex_unlock(&curseg->curseg_mutex);
1512                 page = get_current_sit_page(sbi, start);
1513                 sit_blk = (struct f2fs_sit_block *)page_address(page);
1514                 sit = sit_blk->entries[SIT_ENTRY_OFFSET(sit_i, start)];
1515                 f2fs_put_page(page, 1);
1516 got_it:
1517                 check_block_count(sbi, start, &sit);
1518                 seg_info_from_raw_sit(se, &sit);
1519                 if (sbi->segs_per_sec > 1) {
1520                         struct sec_entry *e = get_sec_entry(sbi, start);
1521                         e->valid_blocks += se->valid_blocks;
1522                 }
1523         }
1524 }
1525
1526 static void init_free_segmap(struct f2fs_sb_info *sbi)
1527 {
1528         unsigned int start;
1529         int type;
1530
1531         for (start = 0; start < TOTAL_SEGS(sbi); start++) {
1532                 struct seg_entry *sentry = get_seg_entry(sbi, start);
1533                 if (!sentry->valid_blocks)
1534                         __set_free(sbi, start);
1535         }
1536
1537         /* set use the current segments */
1538         for (type = CURSEG_HOT_DATA; type <= CURSEG_COLD_NODE; type++) {
1539                 struct curseg_info *curseg_t = CURSEG_I(sbi, type);
1540                 __set_test_and_inuse(sbi, curseg_t->segno);
1541         }
1542 }
1543
1544 static void init_dirty_segmap(struct f2fs_sb_info *sbi)
1545 {
1546         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
1547         struct free_segmap_info *free_i = FREE_I(sbi);
1548         unsigned int segno = 0, offset = 0;
1549         unsigned short valid_blocks;
1550
1551         while (segno < TOTAL_SEGS(sbi)) {
1552                 /* find dirty segment based on free segmap */
1553                 segno = find_next_inuse(free_i, TOTAL_SEGS(sbi), offset);
1554                 if (segno >= TOTAL_SEGS(sbi))
1555                         break;
1556                 offset = segno + 1;
1557                 valid_blocks = get_valid_blocks(sbi, segno, 0);
1558                 if (valid_blocks >= sbi->blocks_per_seg || !valid_blocks)
1559                         continue;
1560                 mutex_lock(&dirty_i->seglist_lock);
1561                 __locate_dirty_segment(sbi, segno, DIRTY);
1562                 mutex_unlock(&dirty_i->seglist_lock);
1563         }
1564 }
1565
1566 static int init_victim_secmap(struct f2fs_sb_info *sbi)
1567 {
1568         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
1569         unsigned int bitmap_size = f2fs_bitmap_size(TOTAL_SECS(sbi));
1570
1571         dirty_i->victim_secmap = kzalloc(bitmap_size, GFP_KERNEL);
1572         if (!dirty_i->victim_secmap)
1573                 return -ENOMEM;
1574         return 0;
1575 }
1576
1577 static int build_dirty_segmap(struct f2fs_sb_info *sbi)
1578 {
1579         struct dirty_seglist_info *dirty_i;
1580         unsigned int bitmap_size, i;
1581
1582         /* allocate memory for dirty segments list information */
1583         dirty_i = kzalloc(sizeof(struct dirty_seglist_info), GFP_KERNEL);
1584         if (!dirty_i)
1585                 return -ENOMEM;
1586
1587         SM_I(sbi)->dirty_info = dirty_i;
1588         mutex_init(&dirty_i->seglist_lock);
1589
1590         bitmap_size = f2fs_bitmap_size(TOTAL_SEGS(sbi));
1591
1592         for (i = 0; i < NR_DIRTY_TYPE; i++) {
1593                 dirty_i->dirty_segmap[i] = kzalloc(bitmap_size, GFP_KERNEL);
1594                 if (!dirty_i->dirty_segmap[i])
1595                         return -ENOMEM;
1596         }
1597
1598         init_dirty_segmap(sbi);
1599         return init_victim_secmap(sbi);
1600 }
1601
1602 /*
1603  * Update min, max modified time for cost-benefit GC algorithm
1604  */
1605 static void init_min_max_mtime(struct f2fs_sb_info *sbi)
1606 {
1607         struct sit_info *sit_i = SIT_I(sbi);
1608         unsigned int segno;
1609
1610         mutex_lock(&sit_i->sentry_lock);
1611
1612         sit_i->min_mtime = LLONG_MAX;
1613
1614         for (segno = 0; segno < TOTAL_SEGS(sbi); segno += sbi->segs_per_sec) {
1615                 unsigned int i;
1616                 unsigned long long mtime = 0;
1617
1618                 for (i = 0; i < sbi->segs_per_sec; i++)
1619                         mtime += get_seg_entry(sbi, segno + i)->mtime;
1620
1621                 mtime = div_u64(mtime, sbi->segs_per_sec);
1622
1623                 if (sit_i->min_mtime > mtime)
1624                         sit_i->min_mtime = mtime;
1625         }
1626         sit_i->max_mtime = get_mtime(sbi);
1627         mutex_unlock(&sit_i->sentry_lock);
1628 }
1629
1630 int build_segment_manager(struct f2fs_sb_info *sbi)
1631 {
1632         struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
1633         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1634         struct f2fs_sm_info *sm_info;
1635         int err;
1636
1637         sm_info = kzalloc(sizeof(struct f2fs_sm_info), GFP_KERNEL);
1638         if (!sm_info)
1639                 return -ENOMEM;
1640
1641         /* init sm info */
1642         sbi->sm_info = sm_info;
1643         INIT_LIST_HEAD(&sm_info->wblist_head);
1644         spin_lock_init(&sm_info->wblist_lock);
1645         sm_info->seg0_blkaddr = le32_to_cpu(raw_super->segment0_blkaddr);
1646         sm_info->main_blkaddr = le32_to_cpu(raw_super->main_blkaddr);
1647         sm_info->segment_count = le32_to_cpu(raw_super->segment_count);
1648         sm_info->reserved_segments = le32_to_cpu(ckpt->rsvd_segment_count);
1649         sm_info->ovp_segments = le32_to_cpu(ckpt->overprov_segment_count);
1650         sm_info->main_segments = le32_to_cpu(raw_super->segment_count_main);
1651         sm_info->ssa_blkaddr = le32_to_cpu(raw_super->ssa_blkaddr);
1652
1653         err = build_sit_info(sbi);
1654         if (err)
1655                 return err;
1656         err = build_free_segmap(sbi);
1657         if (err)
1658                 return err;
1659         err = build_curseg(sbi);
1660         if (err)
1661                 return err;
1662
1663         /* reinit free segmap based on SIT */
1664         build_sit_entries(sbi);
1665
1666         init_free_segmap(sbi);
1667         err = build_dirty_segmap(sbi);
1668         if (err)
1669                 return err;
1670
1671         init_min_max_mtime(sbi);
1672         return 0;
1673 }
1674
1675 static void discard_dirty_segmap(struct f2fs_sb_info *sbi,
1676                 enum dirty_type dirty_type)
1677 {
1678         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
1679
1680         mutex_lock(&dirty_i->seglist_lock);
1681         kfree(dirty_i->dirty_segmap[dirty_type]);
1682         dirty_i->nr_dirty[dirty_type] = 0;
1683         mutex_unlock(&dirty_i->seglist_lock);
1684 }
1685
1686 static void destroy_victim_secmap(struct f2fs_sb_info *sbi)
1687 {
1688         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
1689         kfree(dirty_i->victim_secmap);
1690 }
1691
1692 static void destroy_dirty_segmap(struct f2fs_sb_info *sbi)
1693 {
1694         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
1695         int i;
1696
1697         if (!dirty_i)
1698                 return;
1699
1700         /* discard pre-free/dirty segments list */
1701         for (i = 0; i < NR_DIRTY_TYPE; i++)
1702                 discard_dirty_segmap(sbi, i);
1703
1704         destroy_victim_secmap(sbi);
1705         SM_I(sbi)->dirty_info = NULL;
1706         kfree(dirty_i);
1707 }
1708
1709 static void destroy_curseg(struct f2fs_sb_info *sbi)
1710 {
1711         struct curseg_info *array = SM_I(sbi)->curseg_array;
1712         int i;
1713
1714         if (!array)
1715                 return;
1716         SM_I(sbi)->curseg_array = NULL;
1717         for (i = 0; i < NR_CURSEG_TYPE; i++)
1718                 kfree(array[i].sum_blk);
1719         kfree(array);
1720 }
1721
1722 static void destroy_free_segmap(struct f2fs_sb_info *sbi)
1723 {
1724         struct free_segmap_info *free_i = SM_I(sbi)->free_info;
1725         if (!free_i)
1726                 return;
1727         SM_I(sbi)->free_info = NULL;
1728         kfree(free_i->free_segmap);
1729         kfree(free_i->free_secmap);
1730         kfree(free_i);
1731 }
1732
1733 static void destroy_sit_info(struct f2fs_sb_info *sbi)
1734 {
1735         struct sit_info *sit_i = SIT_I(sbi);
1736         unsigned int start;
1737
1738         if (!sit_i)
1739                 return;
1740
1741         if (sit_i->sentries) {
1742                 for (start = 0; start < TOTAL_SEGS(sbi); start++) {
1743                         kfree(sit_i->sentries[start].cur_valid_map);
1744                         kfree(sit_i->sentries[start].ckpt_valid_map);
1745                 }
1746         }
1747         vfree(sit_i->sentries);
1748         vfree(sit_i->sec_entries);
1749         kfree(sit_i->dirty_sentries_bitmap);
1750
1751         SM_I(sbi)->sit_info = NULL;
1752         kfree(sit_i->sit_bitmap);
1753         kfree(sit_i);
1754 }
1755
1756 void destroy_segment_manager(struct f2fs_sb_info *sbi)
1757 {
1758         struct f2fs_sm_info *sm_info = SM_I(sbi);
1759         destroy_dirty_segmap(sbi);
1760         destroy_curseg(sbi);
1761         destroy_free_segmap(sbi);
1762         destroy_sit_info(sbi);
1763         sbi->sm_info = NULL;
1764         kfree(sm_info);
1765 }