linux/fs/f2fs/segment.c
<<
>>
Prefs
   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 */
  27void 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
  39static void __locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
  40                enum dirty_type dirty_type)
  41{
  42        struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
  43
  44        /* need not be added */
  45        if (IS_CURSEG(sbi, segno))
  46                return;
  47
  48        if (!test_and_set_bit(segno, dirty_i->dirty_segmap[dirty_type]))
  49                dirty_i->nr_dirty[dirty_type]++;
  50
  51        if (dirty_type == DIRTY) {
  52                struct seg_entry *sentry = get_seg_entry(sbi, segno);
  53                enum dirty_type t = DIRTY_HOT_DATA;
  54
  55                dirty_type = sentry->type;
  56
  57                if (!test_and_set_bit(segno, dirty_i->dirty_segmap[dirty_type]))
  58                        dirty_i->nr_dirty[dirty_type]++;
  59
  60                /* Only one bitmap should be set */
  61                for (; t <= DIRTY_COLD_NODE; t++) {
  62                        if (t == dirty_type)
  63                                continue;
  64                        if (test_and_clear_bit(segno, dirty_i->dirty_segmap[t]))
  65                                dirty_i->nr_dirty[t]--;
  66                }
  67        }
  68}
  69
  70static void __remove_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
  71                enum dirty_type dirty_type)
  72{
  73        struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
  74
  75        if (test_and_clear_bit(segno, dirty_i->dirty_segmap[dirty_type]))
  76                dirty_i->nr_dirty[dirty_type]--;
  77
  78        if (dirty_type == DIRTY) {
  79                enum dirty_type t = DIRTY_HOT_DATA;
  80
  81                /* clear all the bitmaps */
  82                for (; t <= DIRTY_COLD_NODE; t++)
  83                        if (test_and_clear_bit(segno, dirty_i->dirty_segmap[t]))
  84                                dirty_i->nr_dirty[t]--;
  85
  86                if (get_valid_blocks(sbi, segno, sbi->segs_per_sec) == 0)
  87                        clear_bit(GET_SECNO(sbi, segno),
  88                                                dirty_i->victim_secmap);
  89        }
  90}
  91
  92/*
  93 * Should not occur error such as -ENOMEM.
  94 * Adding dirty entry into seglist is not critical operation.
  95 * If a given segment is one of current working segments, it won't be added.
  96 */
  97static void locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno)
  98{
  99        struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
 100        unsigned short valid_blocks;
 101
 102        if (segno == NULL_SEGNO || IS_CURSEG(sbi, segno))
 103                return;
 104
 105        mutex_lock(&dirty_i->seglist_lock);
 106
 107        valid_blocks = get_valid_blocks(sbi, segno, 0);
 108
 109        if (valid_blocks == 0) {
 110                __locate_dirty_segment(sbi, segno, PRE);
 111                __remove_dirty_segment(sbi, segno, DIRTY);
 112        } else if (valid_blocks < sbi->blocks_per_seg) {
 113                __locate_dirty_segment(sbi, segno, DIRTY);
 114        } else {
 115                /* Recovery routine with SSR needs this */
 116                __remove_dirty_segment(sbi, segno, DIRTY);
 117        }
 118
 119        mutex_unlock(&dirty_i->seglist_lock);
 120}
 121
 122/*
 123 * Should call clear_prefree_segments after checkpoint is done.
 124 */
 125static void set_prefree_as_free_segments(struct f2fs_sb_info *sbi)
 126{
 127        struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
 128        unsigned int segno = -1;
 129        unsigned int total_segs = TOTAL_SEGS(sbi);
 130
 131        mutex_lock(&dirty_i->seglist_lock);
 132        while (1) {
 133                segno = find_next_bit(dirty_i->dirty_segmap[PRE], total_segs,
 134                                segno + 1);
 135                if (segno >= total_segs)
 136                        break;
 137                __set_test_and_free(sbi, segno);
 138        }
 139        mutex_unlock(&dirty_i->seglist_lock);
 140}
 141
 142void clear_prefree_segments(struct f2fs_sb_info *sbi)
 143{
 144        struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
 145        unsigned int segno = -1;
 146        unsigned int total_segs = TOTAL_SEGS(sbi);
 147
 148        mutex_lock(&dirty_i->seglist_lock);
 149        while (1) {
 150                segno = find_next_bit(dirty_i->dirty_segmap[PRE], total_segs,
 151                                segno + 1);
 152                if (segno >= total_segs)
 153                        break;
 154
 155                if (test_and_clear_bit(segno, dirty_i->dirty_segmap[PRE]))
 156                        dirty_i->nr_dirty[PRE]--;
 157
 158                /* Let's use trim */
 159                if (test_opt(sbi, DISCARD))
 160                        blkdev_issue_discard(sbi->sb->s_bdev,
 161                                        START_BLOCK(sbi, segno) <<
 162                                        sbi->log_sectors_per_block,
 163                                        1 << (sbi->log_sectors_per_block +
 164                                                sbi->log_blocks_per_seg),
 165                                        GFP_NOFS, 0);
 166        }
 167        mutex_unlock(&dirty_i->seglist_lock);
 168}
 169
 170static void __mark_sit_entry_dirty(struct f2fs_sb_info *sbi, unsigned int segno)
 171{
 172        struct sit_info *sit_i = SIT_I(sbi);
 173        if (!__test_and_set_bit(segno, sit_i->dirty_sentries_bitmap))
 174                sit_i->dirty_sentries++;
 175}
 176
 177static void __set_sit_entry_type(struct f2fs_sb_info *sbi, int type,
 178                                        unsigned int segno, int modified)
 179{
 180        struct seg_entry *se = get_seg_entry(sbi, segno);
 181        se->type = type;
 182        if (modified)
 183                __mark_sit_entry_dirty(sbi, segno);
 184}
 185
 186static void update_sit_entry(struct f2fs_sb_info *sbi, block_t blkaddr, int del)
 187{
 188        struct seg_entry *se;
 189        unsigned int segno, offset;
 190        long int new_vblocks;
 191
 192        segno = GET_SEGNO(sbi, blkaddr);
 193
 194        se = get_seg_entry(sbi, segno);
 195        new_vblocks = se->valid_blocks + del;
 196        offset = GET_SEGOFF_FROM_SEG0(sbi, blkaddr) & (sbi->blocks_per_seg - 1);
 197
 198        BUG_ON((new_vblocks >> (sizeof(unsigned short) << 3) ||
 199                                (new_vblocks > sbi->blocks_per_seg)));
 200
 201        se->valid_blocks = new_vblocks;
 202        se->mtime = get_mtime(sbi);
 203        SIT_I(sbi)->max_mtime = se->mtime;
 204
 205        /* Update valid block bitmap */
 206        if (del > 0) {
 207                if (f2fs_set_bit(offset, se->cur_valid_map))
 208                        BUG();
 209        } else {
 210                if (!f2fs_clear_bit(offset, se->cur_valid_map))
 211                        BUG();
 212        }
 213        if (!f2fs_test_bit(offset, se->ckpt_valid_map))
 214                se->ckpt_valid_blocks += del;
 215
 216        __mark_sit_entry_dirty(sbi, segno);
 217
 218        /* update total number of valid blocks to be written in ckpt area */
 219        SIT_I(sbi)->written_valid_blocks += del;
 220
 221        if (sbi->segs_per_sec > 1)
 222                get_sec_entry(sbi, segno)->valid_blocks += del;
 223}
 224
 225static void refresh_sit_entry(struct f2fs_sb_info *sbi,
 226                        block_t old_blkaddr, block_t new_blkaddr)
 227{
 228        update_sit_entry(sbi, new_blkaddr, 1);
 229        if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO)
 230                update_sit_entry(sbi, old_blkaddr, -1);
 231}
 232
 233void invalidate_blocks(struct f2fs_sb_info *sbi, block_t addr)
 234{
 235        unsigned int segno = GET_SEGNO(sbi, addr);
 236        struct sit_info *sit_i = SIT_I(sbi);
 237
 238        BUG_ON(addr == NULL_ADDR);
 239        if (addr == NEW_ADDR)
 240                return;
 241
 242        /* add it into sit main buffer */
 243        mutex_lock(&sit_i->sentry_lock);
 244
 245        update_sit_entry(sbi, addr, -1);
 246
 247        /* add it into dirty seglist */
 248        locate_dirty_segment(sbi, segno);
 249
 250        mutex_unlock(&sit_i->sentry_lock);
 251}
 252
 253/*
 254 * This function should be resided under the curseg_mutex lock
 255 */
 256static void __add_sum_entry(struct f2fs_sb_info *sbi, int type,
 257                                        struct f2fs_summary *sum)
 258{
 259        struct curseg_info *curseg = CURSEG_I(sbi, type);
 260        void *addr = curseg->sum_blk;
 261        addr += curseg->next_blkoff * sizeof(struct f2fs_summary);
 262        memcpy(addr, sum, sizeof(struct f2fs_summary));
 263}
 264
 265/*
 266 * Calculate the number of current summary pages for writing
 267 */
 268int npages_for_summary_flush(struct f2fs_sb_info *sbi)
 269{
 270        int total_size_bytes = 0;
 271        int valid_sum_count = 0;
 272        int i, sum_space;
 273
 274        for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
 275                if (sbi->ckpt->alloc_type[i] == SSR)
 276                        valid_sum_count += sbi->blocks_per_seg;
 277                else
 278                        valid_sum_count += curseg_blkoff(sbi, i);
 279        }
 280
 281        total_size_bytes = valid_sum_count * (SUMMARY_SIZE + 1)
 282                        + sizeof(struct nat_journal) + 2
 283                        + sizeof(struct sit_journal) + 2;
 284        sum_space = PAGE_CACHE_SIZE - SUM_FOOTER_SIZE;
 285        if (total_size_bytes < sum_space)
 286                return 1;
 287        else if (total_size_bytes < 2 * sum_space)
 288                return 2;
 289        return 3;
 290}
 291
 292/*
 293 * Caller should put this summary page
 294 */
 295struct page *get_sum_page(struct f2fs_sb_info *sbi, unsigned int segno)
 296{
 297        return get_meta_page(sbi, GET_SUM_BLOCK(sbi, segno));
 298}
 299
 300static void write_sum_page(struct f2fs_sb_info *sbi,
 301                        struct f2fs_summary_block *sum_blk, block_t blk_addr)
 302{
 303        struct page *page = grab_meta_page(sbi, blk_addr);
 304        void *kaddr = page_address(page);
 305        memcpy(kaddr, sum_blk, PAGE_CACHE_SIZE);
 306        set_page_dirty(page);
 307        f2fs_put_page(page, 1);
 308}
 309
 310static int is_next_segment_free(struct f2fs_sb_info *sbi, int type)
 311{
 312        struct curseg_info *curseg = CURSEG_I(sbi, type);
 313        unsigned int segno = curseg->segno + 1;
 314        struct free_segmap_info *free_i = FREE_I(sbi);
 315
 316        if (segno < TOTAL_SEGS(sbi) && segno % sbi->segs_per_sec)
 317                return !test_bit(segno, free_i->free_segmap);
 318        return 0;
 319}
 320
 321/*
 322 * Find a new segment from the free segments bitmap to right order
 323 * This function should be returned with success, otherwise BUG
 324 */
 325static void get_new_segment(struct f2fs_sb_info *sbi,
 326                        unsigned int *newseg, bool new_sec, int dir)
 327{
 328        struct free_segmap_info *free_i = FREE_I(sbi);
 329        unsigned int segno, secno, zoneno;
 330        unsigned int total_zones = TOTAL_SECS(sbi) / sbi->secs_per_zone;
 331        unsigned int hint = *newseg / sbi->segs_per_sec;
 332        unsigned int old_zoneno = GET_ZONENO_FROM_SEGNO(sbi, *newseg);
 333        unsigned int left_start = hint;
 334        bool init = true;
 335        int go_left = 0;
 336        int i;
 337
 338        write_lock(&free_i->segmap_lock);
 339
 340        if (!new_sec && ((*newseg + 1) % sbi->segs_per_sec)) {
 341                segno = find_next_zero_bit(free_i->free_segmap,
 342                                        TOTAL_SEGS(sbi), *newseg + 1);
 343                if (segno - *newseg < sbi->segs_per_sec -
 344                                        (*newseg % sbi->segs_per_sec))
 345                        goto got_it;
 346        }
 347find_other_zone:
 348        secno = find_next_zero_bit(free_i->free_secmap, TOTAL_SECS(sbi), hint);
 349        if (secno >= TOTAL_SECS(sbi)) {
 350                if (dir == ALLOC_RIGHT) {
 351                        secno = find_next_zero_bit(free_i->free_secmap,
 352                                                        TOTAL_SECS(sbi), 0);
 353                        BUG_ON(secno >= TOTAL_SECS(sbi));
 354                } else {
 355                        go_left = 1;
 356                        left_start = hint - 1;
 357                }
 358        }
 359        if (go_left == 0)
 360                goto skip_left;
 361
 362        while (test_bit(left_start, free_i->free_secmap)) {
 363                if (left_start > 0) {
 364                        left_start--;
 365                        continue;
 366                }
 367                left_start = find_next_zero_bit(free_i->free_secmap,
 368                                                        TOTAL_SECS(sbi), 0);
 369                BUG_ON(left_start >= TOTAL_SECS(sbi));
 370                break;
 371        }
 372        secno = left_start;
 373skip_left:
 374        hint = secno;
 375        segno = secno * sbi->segs_per_sec;
 376        zoneno = secno / sbi->secs_per_zone;
 377
 378        /* give up on finding another zone */
 379        if (!init)
 380                goto got_it;
 381        if (sbi->secs_per_zone == 1)
 382                goto got_it;
 383        if (zoneno == old_zoneno)
 384                goto got_it;
 385        if (dir == ALLOC_LEFT) {
 386                if (!go_left && zoneno + 1 >= total_zones)
 387                        goto got_it;
 388                if (go_left && zoneno == 0)
 389                        goto got_it;
 390        }
 391        for (i = 0; i < NR_CURSEG_TYPE; i++)
 392                if (CURSEG_I(sbi, i)->zone == zoneno)
 393                        break;
 394
 395        if (i < NR_CURSEG_TYPE) {
 396                /* zone is in user, try another */
 397                if (go_left)
 398                        hint = zoneno * sbi->secs_per_zone - 1;
 399                else if (zoneno + 1 >= total_zones)
 400                        hint = 0;
 401                else
 402                        hint = (zoneno + 1) * sbi->secs_per_zone;
 403                init = false;
 404                goto find_other_zone;
 405        }
 406got_it:
 407        /* set it as dirty segment in free segmap */
 408        BUG_ON(test_bit(segno, free_i->free_segmap));
 409        __set_inuse(sbi, segno);
 410        *newseg = segno;
 411        write_unlock(&free_i->segmap_lock);
 412}
 413
 414static void reset_curseg(struct f2fs_sb_info *sbi, int type, int modified)
 415{
 416        struct curseg_info *curseg = CURSEG_I(sbi, type);
 417        struct summary_footer *sum_footer;
 418
 419        curseg->segno = curseg->next_segno;
 420        curseg->zone = GET_ZONENO_FROM_SEGNO(sbi, curseg->segno);
 421        curseg->next_blkoff = 0;
 422        curseg->next_segno = NULL_SEGNO;
 423
 424        sum_footer = &(curseg->sum_blk->footer);
 425        memset(sum_footer, 0, sizeof(struct summary_footer));
 426        if (IS_DATASEG(type))
 427                SET_SUM_TYPE(sum_footer, SUM_TYPE_DATA);
 428        if (IS_NODESEG(type))
 429                SET_SUM_TYPE(sum_footer, SUM_TYPE_NODE);
 430        __set_sit_entry_type(sbi, type, curseg->segno, modified);
 431}
 432
 433/*
 434 * Allocate a current working segment.
 435 * This function always allocates a free segment in LFS manner.
 436 */
 437static void new_curseg(struct f2fs_sb_info *sbi, int type, bool new_sec)
 438{
 439        struct curseg_info *curseg = CURSEG_I(sbi, type);
 440        unsigned int segno = curseg->segno;
 441        int dir = ALLOC_LEFT;
 442
 443        write_sum_page(sbi, curseg->sum_blk,
 444                                GET_SUM_BLOCK(sbi, segno));
 445        if (type == CURSEG_WARM_DATA || type == CURSEG_COLD_DATA)
 446                dir = ALLOC_RIGHT;
 447
 448        if (test_opt(sbi, NOHEAP))
 449                dir = ALLOC_RIGHT;
 450
 451        get_new_segment(sbi, &segno, new_sec, dir);
 452        curseg->next_segno = segno;
 453        reset_curseg(sbi, type, 1);
 454        curseg->alloc_type = LFS;
 455}
 456
 457static void __next_free_blkoff(struct f2fs_sb_info *sbi,
 458                        struct curseg_info *seg, block_t start)
 459{
 460        struct seg_entry *se = get_seg_entry(sbi, seg->segno);
 461        block_t ofs;
 462        for (ofs = start; ofs < sbi->blocks_per_seg; ofs++) {
 463                if (!f2fs_test_bit(ofs, se->ckpt_valid_map)
 464                        && !f2fs_test_bit(ofs, se->cur_valid_map))
 465                        break;
 466        }
 467        seg->next_blkoff = ofs;
 468}
 469
 470/*
 471 * If a segment is written by LFS manner, next block offset is just obtained
 472 * by increasing the current block offset. However, if a segment is written by
 473 * SSR manner, next block offset obtained by calling __next_free_blkoff
 474 */
 475static void __refresh_next_blkoff(struct f2fs_sb_info *sbi,
 476                                struct curseg_info *seg)
 477{
 478        if (seg->alloc_type == SSR)
 479                __next_free_blkoff(sbi, seg, seg->next_blkoff + 1);
 480        else
 481                seg->next_blkoff++;
 482}
 483
 484/*
 485 * This function always allocates a used segment (from dirty seglist) by SSR
 486 * manner, so it should recover the existing segment information of valid blocks
 487 */
 488static void change_curseg(struct f2fs_sb_info *sbi, int type, bool reuse)
 489{
 490        struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
 491        struct curseg_info *curseg = CURSEG_I(sbi, type);
 492        unsigned int new_segno = curseg->next_segno;
 493        struct f2fs_summary_block *sum_node;
 494        struct page *sum_page;
 495
 496        write_sum_page(sbi, curseg->sum_blk,
 497                                GET_SUM_BLOCK(sbi, curseg->segno));
 498        __set_test_and_inuse(sbi, new_segno);
 499
 500        mutex_lock(&dirty_i->seglist_lock);
 501        __remove_dirty_segment(sbi, new_segno, PRE);
 502        __remove_dirty_segment(sbi, new_segno, DIRTY);
 503        mutex_unlock(&dirty_i->seglist_lock);
 504
 505        reset_curseg(sbi, type, 1);
 506        curseg->alloc_type = SSR;
 507        __next_free_blkoff(sbi, curseg, 0);
 508
 509        if (reuse) {
 510                sum_page = get_sum_page(sbi, new_segno);
 511                sum_node = (struct f2fs_summary_block *)page_address(sum_page);
 512                memcpy(curseg->sum_blk, sum_node, SUM_ENTRY_SIZE);
 513                f2fs_put_page(sum_page, 1);
 514        }
 515}
 516
 517static int get_ssr_segment(struct f2fs_sb_info *sbi, int type)
 518{
 519        struct curseg_info *curseg = CURSEG_I(sbi, type);
 520        const struct victim_selection *v_ops = DIRTY_I(sbi)->v_ops;
 521
 522        if (IS_NODESEG(type) || !has_not_enough_free_secs(sbi, 0))
 523                return v_ops->get_victim(sbi,
 524                                &(curseg)->next_segno, BG_GC, type, SSR);
 525
 526        /* For data segments, let's do SSR more intensively */
 527        for (; type >= CURSEG_HOT_DATA; type--)
 528                if (v_ops->get_victim(sbi, &(curseg)->next_segno,
 529                                                BG_GC, type, SSR))
 530                        return 1;
 531        return 0;
 532}
 533
 534/*
 535 * flush out current segment and replace it with new segment
 536 * This function should be returned with success, otherwise BUG
 537 */
 538static void allocate_segment_by_default(struct f2fs_sb_info *sbi,
 539                                                int type, bool force)
 540{
 541        struct curseg_info *curseg = CURSEG_I(sbi, type);
 542
 543        if (force)
 544                new_curseg(sbi, type, true);
 545        else if (type == CURSEG_WARM_NODE)
 546                new_curseg(sbi, type, false);
 547        else if (curseg->alloc_type == LFS && is_next_segment_free(sbi, type))
 548                new_curseg(sbi, type, false);
 549        else if (need_SSR(sbi) && get_ssr_segment(sbi, type))
 550                change_curseg(sbi, type, true);
 551        else
 552                new_curseg(sbi, type, false);
 553#ifdef CONFIG_F2FS_STAT_FS
 554        sbi->segment_count[curseg->alloc_type]++;
 555#endif
 556}
 557
 558void allocate_new_segments(struct f2fs_sb_info *sbi)
 559{
 560        struct curseg_info *curseg;
 561        unsigned int old_curseg;
 562        int i;
 563
 564        for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
 565                curseg = CURSEG_I(sbi, i);
 566                old_curseg = curseg->segno;
 567                SIT_I(sbi)->s_ops->allocate_segment(sbi, i, true);
 568                locate_dirty_segment(sbi, old_curseg);
 569        }
 570}
 571
 572static const struct segment_allocation default_salloc_ops = {
 573        .allocate_segment = allocate_segment_by_default,
 574};
 575
 576static void f2fs_end_io_write(struct bio *bio, int err)
 577{
 578        const int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
 579        struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
 580        struct bio_private *p = bio->bi_private;
 581
 582        do {
 583                struct page *page = bvec->bv_page;
 584
 585                if (--bvec >= bio->bi_io_vec)
 586                        prefetchw(&bvec->bv_page->flags);
 587                if (!uptodate) {
 588                        SetPageError(page);
 589                        if (page->mapping)
 590                                set_bit(AS_EIO, &page->mapping->flags);
 591                        set_ckpt_flags(p->sbi->ckpt, CP_ERROR_FLAG);
 592                        p->sbi->sb->s_flags |= MS_RDONLY;
 593                }
 594                end_page_writeback(page);
 595                dec_page_count(p->sbi, F2FS_WRITEBACK);
 596        } while (bvec >= bio->bi_io_vec);
 597
 598        if (p->is_sync)
 599                complete(p->wait);
 600        kfree(p);
 601        bio_put(bio);
 602}
 603
 604struct bio *f2fs_bio_alloc(struct block_device *bdev, int npages)
 605{
 606        struct bio *bio;
 607
 608        /* No failure on bio allocation */
 609        bio = bio_alloc(GFP_NOIO, npages);
 610        bio->bi_bdev = bdev;
 611        bio->bi_private = NULL;
 612
 613        return bio;
 614}
 615
 616static void do_submit_bio(struct f2fs_sb_info *sbi,
 617                                enum page_type type, bool sync)
 618{
 619        int rw = sync ? WRITE_SYNC : WRITE;
 620        enum page_type btype = type > META ? META : type;
 621
 622        if (type >= META_FLUSH)
 623                rw = WRITE_FLUSH_FUA;
 624
 625        if (btype == META)
 626                rw |= REQ_META;
 627
 628        if (sbi->bio[btype]) {
 629                struct bio_private *p = sbi->bio[btype]->bi_private;
 630                p->sbi = sbi;
 631                sbi->bio[btype]->bi_end_io = f2fs_end_io_write;
 632
 633                trace_f2fs_do_submit_bio(sbi->sb, btype, sync, sbi->bio[btype]);
 634
 635                if (type == META_FLUSH) {
 636                        DECLARE_COMPLETION_ONSTACK(wait);
 637                        p->is_sync = true;
 638                        p->wait = &wait;
 639                        submit_bio(rw, sbi->bio[btype]);
 640                        wait_for_completion(&wait);
 641                } else {
 642                        p->is_sync = false;
 643                        submit_bio(rw, sbi->bio[btype]);
 644                }
 645                sbi->bio[btype] = NULL;
 646        }
 647}
 648
 649void f2fs_submit_bio(struct f2fs_sb_info *sbi, enum page_type type, bool sync)
 650{
 651        down_write(&sbi->bio_sem);
 652        do_submit_bio(sbi, type, sync);
 653        up_write(&sbi->bio_sem);
 654}
 655
 656static void submit_write_page(struct f2fs_sb_info *sbi, struct page *page,
 657                                block_t blk_addr, enum page_type type)
 658{
 659        struct block_device *bdev = sbi->sb->s_bdev;
 660
 661        verify_block_addr(sbi, blk_addr);
 662
 663        down_write(&sbi->bio_sem);
 664
 665        inc_page_count(sbi, F2FS_WRITEBACK);
 666
 667        if (sbi->bio[type] && sbi->last_block_in_bio[type] != blk_addr - 1)
 668                do_submit_bio(sbi, type, false);
 669alloc_new:
 670        if (sbi->bio[type] == NULL) {
 671                struct bio_private *priv;
 672retry:
 673                priv = kmalloc(sizeof(struct bio_private), GFP_NOFS);
 674                if (!priv) {
 675                        cond_resched();
 676                        goto retry;
 677                }
 678
 679                sbi->bio[type] = f2fs_bio_alloc(bdev, max_hw_blocks(sbi));
 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
 701void 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
 711static 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
 719static 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
 727static 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
 744static 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
 764static 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        BUG_ON(sbi->active_logs != NR_CURSEG_TYPE);
 775        return __get_segment_type_6(page, p_type);
 776}
 777
 778static 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#ifdef CONFIG_F2FS_STAT_FS
 805        sbi->block_count[curseg->alloc_type]++;
 806#endif
 807
 808        /*
 809         * SIT information should be updated before segment allocation,
 810         * since SSR needs latest valid block information.
 811         */
 812        refresh_sit_entry(sbi, old_blkaddr, *new_blkaddr);
 813
 814        if (!__has_curseg_space(sbi, type))
 815                sit_i->s_ops->allocate_segment(sbi, type, false);
 816
 817        locate_dirty_segment(sbi, old_cursegno);
 818        locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr));
 819        mutex_unlock(&sit_i->sentry_lock);
 820
 821        if (p_type == NODE)
 822                fill_node_footer_blkaddr(page, NEXT_FREE_BLKADDR(sbi, curseg));
 823
 824        /* writeout dirty page into bdev */
 825        submit_write_page(sbi, page, *new_blkaddr, p_type);
 826
 827        mutex_unlock(&curseg->curseg_mutex);
 828}
 829
 830void write_meta_page(struct f2fs_sb_info *sbi, struct page *page)
 831{
 832        set_page_writeback(page);
 833        submit_write_page(sbi, page, page->index, META);
 834}
 835
 836void write_node_page(struct f2fs_sb_info *sbi, struct page *page,
 837                unsigned int nid, block_t old_blkaddr, block_t *new_blkaddr)
 838{
 839        struct f2fs_summary sum;
 840        set_summary(&sum, nid, 0, 0);
 841        do_write_page(sbi, page, old_blkaddr, new_blkaddr, &sum, NODE);
 842}
 843
 844void write_data_page(struct inode *inode, struct page *page,
 845                struct dnode_of_data *dn, block_t old_blkaddr,
 846                block_t *new_blkaddr)
 847{
 848        struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
 849        struct f2fs_summary sum;
 850        struct node_info ni;
 851
 852        BUG_ON(old_blkaddr == NULL_ADDR);
 853        get_node_info(sbi, dn->nid, &ni);
 854        set_summary(&sum, dn->nid, dn->ofs_in_node, ni.version);
 855
 856        do_write_page(sbi, page, old_blkaddr,
 857                        new_blkaddr, &sum, DATA);
 858}
 859
 860void rewrite_data_page(struct f2fs_sb_info *sbi, struct page *page,
 861                                        block_t old_blk_addr)
 862{
 863        submit_write_page(sbi, page, old_blk_addr, DATA);
 864}
 865
 866void recover_data_page(struct f2fs_sb_info *sbi,
 867                        struct page *page, struct f2fs_summary *sum,
 868                        block_t old_blkaddr, block_t new_blkaddr)
 869{
 870        struct sit_info *sit_i = SIT_I(sbi);
 871        struct curseg_info *curseg;
 872        unsigned int segno, old_cursegno;
 873        struct seg_entry *se;
 874        int type;
 875
 876        segno = GET_SEGNO(sbi, new_blkaddr);
 877        se = get_seg_entry(sbi, segno);
 878        type = se->type;
 879
 880        if (se->valid_blocks == 0 && !IS_CURSEG(sbi, segno)) {
 881                if (old_blkaddr == NULL_ADDR)
 882                        type = CURSEG_COLD_DATA;
 883                else
 884                        type = CURSEG_WARM_DATA;
 885        }
 886        curseg = CURSEG_I(sbi, type);
 887
 888        mutex_lock(&curseg->curseg_mutex);
 889        mutex_lock(&sit_i->sentry_lock);
 890
 891        old_cursegno = curseg->segno;
 892
 893        /* change the current segment */
 894        if (segno != curseg->segno) {
 895                curseg->next_segno = segno;
 896                change_curseg(sbi, type, true);
 897        }
 898
 899        curseg->next_blkoff = GET_SEGOFF_FROM_SEG0(sbi, new_blkaddr) &
 900                                        (sbi->blocks_per_seg - 1);
 901        __add_sum_entry(sbi, type, sum);
 902
 903        refresh_sit_entry(sbi, old_blkaddr, new_blkaddr);
 904
 905        locate_dirty_segment(sbi, old_cursegno);
 906        locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr));
 907
 908        mutex_unlock(&sit_i->sentry_lock);
 909        mutex_unlock(&curseg->curseg_mutex);
 910}
 911
 912void rewrite_node_page(struct f2fs_sb_info *sbi,
 913                        struct page *page, struct f2fs_summary *sum,
 914                        block_t old_blkaddr, block_t new_blkaddr)
 915{
 916        struct sit_info *sit_i = SIT_I(sbi);
 917        int type = CURSEG_WARM_NODE;
 918        struct curseg_info *curseg;
 919        unsigned int segno, old_cursegno;
 920        block_t next_blkaddr = next_blkaddr_of_node(page);
 921        unsigned int next_segno = GET_SEGNO(sbi, next_blkaddr);
 922
 923        curseg = CURSEG_I(sbi, type);
 924
 925        mutex_lock(&curseg->curseg_mutex);
 926        mutex_lock(&sit_i->sentry_lock);
 927
 928        segno = GET_SEGNO(sbi, new_blkaddr);
 929        old_cursegno = curseg->segno;
 930
 931        /* change the current segment */
 932        if (segno != curseg->segno) {
 933                curseg->next_segno = segno;
 934                change_curseg(sbi, type, true);
 935        }
 936        curseg->next_blkoff = GET_SEGOFF_FROM_SEG0(sbi, new_blkaddr) &
 937                                        (sbi->blocks_per_seg - 1);
 938        __add_sum_entry(sbi, type, sum);
 939
 940        /* change the current log to the next block addr in advance */
 941        if (next_segno != segno) {
 942                curseg->next_segno = next_segno;
 943                change_curseg(sbi, type, true);
 944        }
 945        curseg->next_blkoff = GET_SEGOFF_FROM_SEG0(sbi, next_blkaddr) &
 946                                        (sbi->blocks_per_seg - 1);
 947
 948        /* rewrite node page */
 949        set_page_writeback(page);
 950        submit_write_page(sbi, page, new_blkaddr, NODE);
 951        f2fs_submit_bio(sbi, NODE, true);
 952        refresh_sit_entry(sbi, old_blkaddr, new_blkaddr);
 953
 954        locate_dirty_segment(sbi, old_cursegno);
 955        locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr));
 956
 957        mutex_unlock(&sit_i->sentry_lock);
 958        mutex_unlock(&curseg->curseg_mutex);
 959}
 960
 961static int read_compacted_summaries(struct f2fs_sb_info *sbi)
 962{
 963        struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
 964        struct curseg_info *seg_i;
 965        unsigned char *kaddr;
 966        struct page *page;
 967        block_t start;
 968        int i, j, offset;
 969
 970        start = start_sum_block(sbi);
 971
 972        page = get_meta_page(sbi, start++);
 973        kaddr = (unsigned char *)page_address(page);
 974
 975        /* Step 1: restore nat cache */
 976        seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
 977        memcpy(&seg_i->sum_blk->n_nats, kaddr, SUM_JOURNAL_SIZE);
 978
 979        /* Step 2: restore sit cache */
 980        seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
 981        memcpy(&seg_i->sum_blk->n_sits, kaddr + SUM_JOURNAL_SIZE,
 982                                                SUM_JOURNAL_SIZE);
 983        offset = 2 * SUM_JOURNAL_SIZE;
 984
 985        /* Step 3: restore summary entries */
 986        for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
 987                unsigned short blk_off;
 988                unsigned int segno;
 989
 990                seg_i = CURSEG_I(sbi, i);
 991                segno = le32_to_cpu(ckpt->cur_data_segno[i]);
 992                blk_off = le16_to_cpu(ckpt->cur_data_blkoff[i]);
 993                seg_i->next_segno = segno;
 994                reset_curseg(sbi, i, 0);
 995                seg_i->alloc_type = ckpt->alloc_type[i];
 996                seg_i->next_blkoff = blk_off;
 997
 998                if (seg_i->alloc_type == SSR)
 999                        blk_off = sbi->blocks_per_seg;
1000
1001                for (j = 0; j < blk_off; j++) {
1002                        struct f2fs_summary *s;
1003                        s = (struct f2fs_summary *)(kaddr + offset);
1004                        seg_i->sum_blk->entries[j] = *s;
1005                        offset += SUMMARY_SIZE;
1006                        if (offset + SUMMARY_SIZE <= PAGE_CACHE_SIZE -
1007                                                SUM_FOOTER_SIZE)
1008                                continue;
1009
1010                        f2fs_put_page(page, 1);
1011                        page = NULL;
1012
1013                        page = get_meta_page(sbi, start++);
1014                        kaddr = (unsigned char *)page_address(page);
1015                        offset = 0;
1016                }
1017        }
1018        f2fs_put_page(page, 1);
1019        return 0;
1020}
1021
1022static int read_normal_summaries(struct f2fs_sb_info *sbi, int type)
1023{
1024        struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1025        struct f2fs_summary_block *sum;
1026        struct curseg_info *curseg;
1027        struct page *new;
1028        unsigned short blk_off;
1029        unsigned int segno = 0;
1030        block_t blk_addr = 0;
1031
1032        /* get segment number and block addr */
1033        if (IS_DATASEG(type)) {
1034                segno = le32_to_cpu(ckpt->cur_data_segno[type]);
1035                blk_off = le16_to_cpu(ckpt->cur_data_blkoff[type -
1036                                                        CURSEG_HOT_DATA]);
1037                if (is_set_ckpt_flags(ckpt, CP_UMOUNT_FLAG))
1038                        blk_addr = sum_blk_addr(sbi, NR_CURSEG_TYPE, type);
1039                else
1040                        blk_addr = sum_blk_addr(sbi, NR_CURSEG_DATA_TYPE, type);
1041        } else {
1042                segno = le32_to_cpu(ckpt->cur_node_segno[type -
1043                                                        CURSEG_HOT_NODE]);
1044                blk_off = le16_to_cpu(ckpt->cur_node_blkoff[type -
1045                                                        CURSEG_HOT_NODE]);
1046                if (is_set_ckpt_flags(ckpt, CP_UMOUNT_FLAG))
1047                        blk_addr = sum_blk_addr(sbi, NR_CURSEG_NODE_TYPE,
1048                                                        type - CURSEG_HOT_NODE);
1049                else
1050                        blk_addr = GET_SUM_BLOCK(sbi, segno);
1051        }
1052
1053        new = get_meta_page(sbi, blk_addr);
1054        sum = (struct f2fs_summary_block *)page_address(new);
1055
1056        if (IS_NODESEG(type)) {
1057                if (is_set_ckpt_flags(ckpt, CP_UMOUNT_FLAG)) {
1058                        struct f2fs_summary *ns = &sum->entries[0];
1059                        int i;
1060                        for (i = 0; i < sbi->blocks_per_seg; i++, ns++) {
1061                                ns->version = 0;
1062                                ns->ofs_in_node = 0;
1063                        }
1064                } else {
1065                        if (restore_node_summary(sbi, segno, sum)) {
1066                                f2fs_put_page(new, 1);
1067                                return -EINVAL;
1068                        }
1069                }
1070        }
1071
1072        /* set uncompleted segment to curseg */
1073        curseg = CURSEG_I(sbi, type);
1074        mutex_lock(&curseg->curseg_mutex);
1075        memcpy(curseg->sum_blk, sum, PAGE_CACHE_SIZE);
1076        curseg->next_segno = segno;
1077        reset_curseg(sbi, type, 0);
1078        curseg->alloc_type = ckpt->alloc_type[type];
1079        curseg->next_blkoff = blk_off;
1080        mutex_unlock(&curseg->curseg_mutex);
1081        f2fs_put_page(new, 1);
1082        return 0;
1083}
1084
1085static int restore_curseg_summaries(struct f2fs_sb_info *sbi)
1086{
1087        int type = CURSEG_HOT_DATA;
1088
1089        if (is_set_ckpt_flags(F2FS_CKPT(sbi), CP_COMPACT_SUM_FLAG)) {
1090                /* restore for compacted data summary */
1091                if (read_compacted_summaries(sbi))
1092                        return -EINVAL;
1093                type = CURSEG_HOT_NODE;
1094        }
1095
1096        for (; type <= CURSEG_COLD_NODE; type++)
1097                if (read_normal_summaries(sbi, type))
1098                        return -EINVAL;
1099        return 0;
1100}
1101
1102static void write_compacted_summaries(struct f2fs_sb_info *sbi, block_t blkaddr)
1103{
1104        struct page *page;
1105        unsigned char *kaddr;
1106        struct f2fs_summary *summary;
1107        struct curseg_info *seg_i;
1108        int written_size = 0;
1109        int i, j;
1110
1111        page = grab_meta_page(sbi, blkaddr++);
1112        kaddr = (unsigned char *)page_address(page);
1113
1114        /* Step 1: write nat cache */
1115        seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
1116        memcpy(kaddr, &seg_i->sum_blk->n_nats, SUM_JOURNAL_SIZE);
1117        written_size += SUM_JOURNAL_SIZE;
1118
1119        /* Step 2: write sit cache */
1120        seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
1121        memcpy(kaddr + written_size, &seg_i->sum_blk->n_sits,
1122                                                SUM_JOURNAL_SIZE);
1123        written_size += SUM_JOURNAL_SIZE;
1124
1125        set_page_dirty(page);
1126
1127        /* Step 3: write summary entries */
1128        for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
1129                unsigned short blkoff;
1130                seg_i = CURSEG_I(sbi, i);
1131                if (sbi->ckpt->alloc_type[i] == SSR)
1132                        blkoff = sbi->blocks_per_seg;
1133                else
1134                        blkoff = curseg_blkoff(sbi, i);
1135
1136                for (j = 0; j < blkoff; j++) {
1137                        if (!page) {
1138                                page = grab_meta_page(sbi, blkaddr++);
1139                                kaddr = (unsigned char *)page_address(page);
1140                                written_size = 0;
1141                        }
1142                        summary = (struct f2fs_summary *)(kaddr + written_size);
1143                        *summary = seg_i->sum_blk->entries[j];
1144                        written_size += SUMMARY_SIZE;
1145                        set_page_dirty(page);
1146
1147                        if (written_size + SUMMARY_SIZE <= PAGE_CACHE_SIZE -
1148                                                        SUM_FOOTER_SIZE)
1149                                continue;
1150
1151                        f2fs_put_page(page, 1);
1152                        page = NULL;
1153                }
1154        }
1155        if (page)
1156                f2fs_put_page(page, 1);
1157}
1158
1159static void write_normal_summaries(struct f2fs_sb_info *sbi,
1160                                        block_t blkaddr, int type)
1161{
1162        int i, end;
1163        if (IS_DATASEG(type))
1164                end = type + NR_CURSEG_DATA_TYPE;
1165        else
1166                end = type + NR_CURSEG_NODE_TYPE;
1167
1168        for (i = type; i < end; i++) {
1169                struct curseg_info *sum = CURSEG_I(sbi, i);
1170                mutex_lock(&sum->curseg_mutex);
1171                write_sum_page(sbi, sum->sum_blk, blkaddr + (i - type));
1172                mutex_unlock(&sum->curseg_mutex);
1173        }
1174}
1175
1176void write_data_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
1177{
1178        if (is_set_ckpt_flags(F2FS_CKPT(sbi), CP_COMPACT_SUM_FLAG))
1179                write_compacted_summaries(sbi, start_blk);
1180        else
1181                write_normal_summaries(sbi, start_blk, CURSEG_HOT_DATA);
1182}
1183
1184void write_node_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
1185{
1186        if (is_set_ckpt_flags(F2FS_CKPT(sbi), CP_UMOUNT_FLAG))
1187                write_normal_summaries(sbi, start_blk, CURSEG_HOT_NODE);
1188}
1189
1190int lookup_journal_in_cursum(struct f2fs_summary_block *sum, int type,
1191                                        unsigned int val, int alloc)
1192{
1193        int i;
1194
1195        if (type == NAT_JOURNAL) {
1196                for (i = 0; i < nats_in_cursum(sum); i++) {
1197                        if (le32_to_cpu(nid_in_journal(sum, i)) == val)
1198                                return i;
1199                }
1200                if (alloc && nats_in_cursum(sum) < NAT_JOURNAL_ENTRIES)
1201                        return update_nats_in_cursum(sum, 1);
1202        } else if (type == SIT_JOURNAL) {
1203                for (i = 0; i < sits_in_cursum(sum); i++)
1204                        if (le32_to_cpu(segno_in_journal(sum, i)) == val)
1205                                return i;
1206                if (alloc && sits_in_cursum(sum) < SIT_JOURNAL_ENTRIES)
1207                        return update_sits_in_cursum(sum, 1);
1208        }
1209        return -1;
1210}
1211
1212static struct page *get_current_sit_page(struct f2fs_sb_info *sbi,
1213                                        unsigned int segno)
1214{
1215        struct sit_info *sit_i = SIT_I(sbi);
1216        unsigned int offset = SIT_BLOCK_OFFSET(sit_i, segno);
1217        block_t blk_addr = sit_i->sit_base_addr + offset;
1218
1219        check_seg_range(sbi, segno);
1220
1221        /* calculate sit block address */
1222        if (f2fs_test_bit(offset, sit_i->sit_bitmap))
1223                blk_addr += sit_i->sit_blocks;
1224
1225        return get_meta_page(sbi, blk_addr);
1226}
1227
1228static struct page *get_next_sit_page(struct f2fs_sb_info *sbi,
1229                                        unsigned int start)
1230{
1231        struct sit_info *sit_i = SIT_I(sbi);
1232        struct page *src_page, *dst_page;
1233        pgoff_t src_off, dst_off;
1234        void *src_addr, *dst_addr;
1235
1236        src_off = current_sit_addr(sbi, start);
1237        dst_off = next_sit_addr(sbi, src_off);
1238
1239        /* get current sit block page without lock */
1240        src_page = get_meta_page(sbi, src_off);
1241        dst_page = grab_meta_page(sbi, dst_off);
1242        BUG_ON(PageDirty(src_page));
1243
1244        src_addr = page_address(src_page);
1245        dst_addr = page_address(dst_page);
1246        memcpy(dst_addr, src_addr, PAGE_CACHE_SIZE);
1247
1248        set_page_dirty(dst_page);
1249        f2fs_put_page(src_page, 1);
1250
1251        set_to_next_sit(sit_i, start);
1252
1253        return dst_page;
1254}
1255
1256static bool flush_sits_in_journal(struct f2fs_sb_info *sbi)
1257{
1258        struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
1259        struct f2fs_summary_block *sum = curseg->sum_blk;
1260        int i;
1261
1262        /*
1263         * If the journal area in the current summary is full of sit entries,
1264         * all the sit entries will be flushed. Otherwise the sit entries
1265         * are not able to replace with newly hot sit entries.
1266         */
1267        if (sits_in_cursum(sum) >= SIT_JOURNAL_ENTRIES) {
1268                for (i = sits_in_cursum(sum) - 1; i >= 0; i--) {
1269                        unsigned int segno;
1270                        segno = le32_to_cpu(segno_in_journal(sum, i));
1271                        __mark_sit_entry_dirty(sbi, segno);
1272                }
1273                update_sits_in_cursum(sum, -sits_in_cursum(sum));
1274                return 1;
1275        }
1276        return 0;
1277}
1278
1279/*
1280 * CP calls this function, which flushes SIT entries including sit_journal,
1281 * and moves prefree segs to free segs.
1282 */
1283void flush_sit_entries(struct f2fs_sb_info *sbi)
1284{
1285        struct sit_info *sit_i = SIT_I(sbi);
1286        unsigned long *bitmap = sit_i->dirty_sentries_bitmap;
1287        struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
1288        struct f2fs_summary_block *sum = curseg->sum_blk;
1289        unsigned long nsegs = TOTAL_SEGS(sbi);
1290        struct page *page = NULL;
1291        struct f2fs_sit_block *raw_sit = NULL;
1292        unsigned int start = 0, end = 0;
1293        unsigned int segno = -1;
1294        bool flushed;
1295
1296        mutex_lock(&curseg->curseg_mutex);
1297        mutex_lock(&sit_i->sentry_lock);
1298
1299        /*
1300         * "flushed" indicates whether sit entries in journal are flushed
1301         * to the SIT area or not.
1302         */
1303        flushed = flush_sits_in_journal(sbi);
1304
1305        while ((segno = find_next_bit(bitmap, nsegs, segno + 1)) < nsegs) {
1306                struct seg_entry *se = get_seg_entry(sbi, segno);
1307                int sit_offset, offset;
1308
1309                sit_offset = SIT_ENTRY_OFFSET(sit_i, segno);
1310
1311                if (flushed)
1312                        goto to_sit_page;
1313
1314                offset = lookup_journal_in_cursum(sum, SIT_JOURNAL, segno, 1);
1315                if (offset >= 0) {
1316                        segno_in_journal(sum, offset) = cpu_to_le32(segno);
1317                        seg_info_to_raw_sit(se, &sit_in_journal(sum, offset));
1318                        goto flush_done;
1319                }
1320to_sit_page:
1321                if (!page || (start > segno) || (segno > end)) {
1322                        if (page) {
1323                                f2fs_put_page(page, 1);
1324                                page = NULL;
1325                        }
1326
1327                        start = START_SEGNO(sit_i, segno);
1328                        end = start + SIT_ENTRY_PER_BLOCK - 1;
1329
1330                        /* read sit block that will be updated */
1331                        page = get_next_sit_page(sbi, start);
1332                        raw_sit = page_address(page);
1333                }
1334
1335                /* udpate entry in SIT block */
1336                seg_info_to_raw_sit(se, &raw_sit->entries[sit_offset]);
1337flush_done:
1338                __clear_bit(segno, bitmap);
1339                sit_i->dirty_sentries--;
1340        }
1341        mutex_unlock(&sit_i->sentry_lock);
1342        mutex_unlock(&curseg->curseg_mutex);
1343
1344        /* writeout last modified SIT block */
1345        f2fs_put_page(page, 1);
1346
1347        set_prefree_as_free_segments(sbi);
1348}
1349
1350static int build_sit_info(struct f2fs_sb_info *sbi)
1351{
1352        struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
1353        struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1354        struct sit_info *sit_i;
1355        unsigned int sit_segs, start;
1356        char *src_bitmap, *dst_bitmap;
1357        unsigned int bitmap_size;
1358
1359        /* allocate memory for SIT information */
1360        sit_i = kzalloc(sizeof(struct sit_info), GFP_KERNEL);
1361        if (!sit_i)
1362                return -ENOMEM;
1363
1364        SM_I(sbi)->sit_info = sit_i;
1365
1366        sit_i->sentries = vzalloc(TOTAL_SEGS(sbi) * sizeof(struct seg_entry));
1367        if (!sit_i->sentries)
1368                return -ENOMEM;
1369
1370        bitmap_size = f2fs_bitmap_size(TOTAL_SEGS(sbi));
1371        sit_i->dirty_sentries_bitmap = kzalloc(bitmap_size, GFP_KERNEL);
1372        if (!sit_i->dirty_sentries_bitmap)
1373                return -ENOMEM;
1374
1375        for (start = 0; start < TOTAL_SEGS(sbi); start++) {
1376                sit_i->sentries[start].cur_valid_map
1377                        = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
1378                sit_i->sentries[start].ckpt_valid_map
1379                        = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
1380                if (!sit_i->sentries[start].cur_valid_map
1381                                || !sit_i->sentries[start].ckpt_valid_map)
1382                        return -ENOMEM;
1383        }
1384
1385        if (sbi->segs_per_sec > 1) {
1386                sit_i->sec_entries = vzalloc(TOTAL_SECS(sbi) *
1387                                        sizeof(struct sec_entry));
1388                if (!sit_i->sec_entries)
1389                        return -ENOMEM;
1390        }
1391
1392        /* get information related with SIT */
1393        sit_segs = le32_to_cpu(raw_super->segment_count_sit) >> 1;
1394
1395        /* setup SIT bitmap from ckeckpoint pack */
1396        bitmap_size = __bitmap_size(sbi, SIT_BITMAP);
1397        src_bitmap = __bitmap_ptr(sbi, SIT_BITMAP);
1398
1399        dst_bitmap = kmemdup(src_bitmap, bitmap_size, GFP_KERNEL);
1400        if (!dst_bitmap)
1401                return -ENOMEM;
1402
1403        /* init SIT information */
1404        sit_i->s_ops = &default_salloc_ops;
1405
1406        sit_i->sit_base_addr = le32_to_cpu(raw_super->sit_blkaddr);
1407        sit_i->sit_blocks = sit_segs << sbi->log_blocks_per_seg;
1408        sit_i->written_valid_blocks = le64_to_cpu(ckpt->valid_block_count);
1409        sit_i->sit_bitmap = dst_bitmap;
1410        sit_i->bitmap_size = bitmap_size;
1411        sit_i->dirty_sentries = 0;
1412        sit_i->sents_per_block = SIT_ENTRY_PER_BLOCK;
1413        sit_i->elapsed_time = le64_to_cpu(sbi->ckpt->elapsed_time);
1414        sit_i->mounted_time = CURRENT_TIME_SEC.tv_sec;
1415        mutex_init(&sit_i->sentry_lock);
1416        return 0;
1417}
1418
1419static int build_free_segmap(struct f2fs_sb_info *sbi)
1420{
1421        struct f2fs_sm_info *sm_info = SM_I(sbi);
1422        struct free_segmap_info *free_i;
1423        unsigned int bitmap_size, sec_bitmap_size;
1424
1425        /* allocate memory for free segmap information */
1426        free_i = kzalloc(sizeof(struct free_segmap_info), GFP_KERNEL);
1427        if (!free_i)
1428                return -ENOMEM;
1429
1430        SM_I(sbi)->free_info = free_i;
1431
1432        bitmap_size = f2fs_bitmap_size(TOTAL_SEGS(sbi));
1433        free_i->free_segmap = kmalloc(bitmap_size, GFP_KERNEL);
1434        if (!free_i->free_segmap)
1435                return -ENOMEM;
1436
1437        sec_bitmap_size = f2fs_bitmap_size(TOTAL_SECS(sbi));
1438        free_i->free_secmap = kmalloc(sec_bitmap_size, GFP_KERNEL);
1439        if (!free_i->free_secmap)
1440                return -ENOMEM;
1441
1442        /* set all segments as dirty temporarily */
1443        memset(free_i->free_segmap, 0xff, bitmap_size);
1444        memset(free_i->free_secmap, 0xff, sec_bitmap_size);
1445
1446        /* init free segmap information */
1447        free_i->start_segno =
1448                (unsigned int) GET_SEGNO_FROM_SEG0(sbi, sm_info->main_blkaddr);
1449        free_i->free_segments = 0;
1450        free_i->free_sections = 0;
1451        rwlock_init(&free_i->segmap_lock);
1452        return 0;
1453}
1454
1455static int build_curseg(struct f2fs_sb_info *sbi)
1456{
1457        struct curseg_info *array;
1458        int i;
1459
1460        array = kzalloc(sizeof(*array) * NR_CURSEG_TYPE, GFP_KERNEL);
1461        if (!array)
1462                return -ENOMEM;
1463
1464        SM_I(sbi)->curseg_array = array;
1465
1466        for (i = 0; i < NR_CURSEG_TYPE; i++) {
1467                mutex_init(&array[i].curseg_mutex);
1468                array[i].sum_blk = kzalloc(PAGE_CACHE_SIZE, GFP_KERNEL);
1469                if (!array[i].sum_blk)
1470                        return -ENOMEM;
1471                array[i].segno = NULL_SEGNO;
1472                array[i].next_blkoff = 0;
1473        }
1474        return restore_curseg_summaries(sbi);
1475}
1476
1477static void build_sit_entries(struct f2fs_sb_info *sbi)
1478{
1479        struct sit_info *sit_i = SIT_I(sbi);
1480        struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
1481        struct f2fs_summary_block *sum = curseg->sum_blk;
1482        unsigned int start;
1483
1484        for (start = 0; start < TOTAL_SEGS(sbi); start++) {
1485                struct seg_entry *se = &sit_i->sentries[start];
1486                struct f2fs_sit_block *sit_blk;
1487                struct f2fs_sit_entry sit;
1488                struct page *page;
1489                int i;
1490
1491                mutex_lock(&curseg->curseg_mutex);
1492                for (i = 0; i < sits_in_cursum(sum); i++) {
1493                        if (le32_to_cpu(segno_in_journal(sum, i)) == start) {
1494                                sit = sit_in_journal(sum, i);
1495                                mutex_unlock(&curseg->curseg_mutex);
1496                                goto got_it;
1497                        }
1498                }
1499                mutex_unlock(&curseg->curseg_mutex);
1500                page = get_current_sit_page(sbi, start);
1501                sit_blk = (struct f2fs_sit_block *)page_address(page);
1502                sit = sit_blk->entries[SIT_ENTRY_OFFSET(sit_i, start)];
1503                f2fs_put_page(page, 1);
1504got_it:
1505                check_block_count(sbi, start, &sit);
1506                seg_info_from_raw_sit(se, &sit);
1507                if (sbi->segs_per_sec > 1) {
1508                        struct sec_entry *e = get_sec_entry(sbi, start);
1509                        e->valid_blocks += se->valid_blocks;
1510                }
1511        }
1512}
1513
1514static void init_free_segmap(struct f2fs_sb_info *sbi)
1515{
1516        unsigned int start;
1517        int type;
1518
1519        for (start = 0; start < TOTAL_SEGS(sbi); start++) {
1520                struct seg_entry *sentry = get_seg_entry(sbi, start);
1521                if (!sentry->valid_blocks)
1522                        __set_free(sbi, start);
1523        }
1524
1525        /* set use the current segments */
1526        for (type = CURSEG_HOT_DATA; type <= CURSEG_COLD_NODE; type++) {
1527                struct curseg_info *curseg_t = CURSEG_I(sbi, type);
1528                __set_test_and_inuse(sbi, curseg_t->segno);
1529        }
1530}
1531
1532static void init_dirty_segmap(struct f2fs_sb_info *sbi)
1533{
1534        struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
1535        struct free_segmap_info *free_i = FREE_I(sbi);
1536        unsigned int segno = 0, offset = 0, total_segs = TOTAL_SEGS(sbi);
1537        unsigned short valid_blocks;
1538
1539        while (1) {
1540                /* find dirty segment based on free segmap */
1541                segno = find_next_inuse(free_i, total_segs, offset);
1542                if (segno >= total_segs)
1543                        break;
1544                offset = segno + 1;
1545                valid_blocks = get_valid_blocks(sbi, segno, 0);
1546                if (valid_blocks >= sbi->blocks_per_seg || !valid_blocks)
1547                        continue;
1548                mutex_lock(&dirty_i->seglist_lock);
1549                __locate_dirty_segment(sbi, segno, DIRTY);
1550                mutex_unlock(&dirty_i->seglist_lock);
1551        }
1552}
1553
1554static int init_victim_secmap(struct f2fs_sb_info *sbi)
1555{
1556        struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
1557        unsigned int bitmap_size = f2fs_bitmap_size(TOTAL_SECS(sbi));
1558
1559        dirty_i->victim_secmap = kzalloc(bitmap_size, GFP_KERNEL);
1560        if (!dirty_i->victim_secmap)
1561                return -ENOMEM;
1562        return 0;
1563}
1564
1565static int build_dirty_segmap(struct f2fs_sb_info *sbi)
1566{
1567        struct dirty_seglist_info *dirty_i;
1568        unsigned int bitmap_size, i;
1569
1570        /* allocate memory for dirty segments list information */
1571        dirty_i = kzalloc(sizeof(struct dirty_seglist_info), GFP_KERNEL);
1572        if (!dirty_i)
1573                return -ENOMEM;
1574
1575        SM_I(sbi)->dirty_info = dirty_i;
1576        mutex_init(&dirty_i->seglist_lock);
1577
1578        bitmap_size = f2fs_bitmap_size(TOTAL_SEGS(sbi));
1579
1580        for (i = 0; i < NR_DIRTY_TYPE; i++) {
1581                dirty_i->dirty_segmap[i] = kzalloc(bitmap_size, GFP_KERNEL);
1582                if (!dirty_i->dirty_segmap[i])
1583                        return -ENOMEM;
1584        }
1585
1586        init_dirty_segmap(sbi);
1587        return init_victim_secmap(sbi);
1588}
1589
1590/*
1591 * Update min, max modified time for cost-benefit GC algorithm
1592 */
1593static void init_min_max_mtime(struct f2fs_sb_info *sbi)
1594{
1595        struct sit_info *sit_i = SIT_I(sbi);
1596        unsigned int segno;
1597
1598        mutex_lock(&sit_i->sentry_lock);
1599
1600        sit_i->min_mtime = LLONG_MAX;
1601
1602        for (segno = 0; segno < TOTAL_SEGS(sbi); segno += sbi->segs_per_sec) {
1603                unsigned int i;
1604                unsigned long long mtime = 0;
1605
1606                for (i = 0; i < sbi->segs_per_sec; i++)
1607                        mtime += get_seg_entry(sbi, segno + i)->mtime;
1608
1609                mtime = div_u64(mtime, sbi->segs_per_sec);
1610
1611                if (sit_i->min_mtime > mtime)
1612                        sit_i->min_mtime = mtime;
1613        }
1614        sit_i->max_mtime = get_mtime(sbi);
1615        mutex_unlock(&sit_i->sentry_lock);
1616}
1617
1618int build_segment_manager(struct f2fs_sb_info *sbi)
1619{
1620        struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
1621        struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1622        struct f2fs_sm_info *sm_info;
1623        int err;
1624
1625        sm_info = kzalloc(sizeof(struct f2fs_sm_info), GFP_KERNEL);
1626        if (!sm_info)
1627                return -ENOMEM;
1628
1629        /* init sm info */
1630        sbi->sm_info = sm_info;
1631        INIT_LIST_HEAD(&sm_info->wblist_head);
1632        spin_lock_init(&sm_info->wblist_lock);
1633        sm_info->seg0_blkaddr = le32_to_cpu(raw_super->segment0_blkaddr);
1634        sm_info->main_blkaddr = le32_to_cpu(raw_super->main_blkaddr);
1635        sm_info->segment_count = le32_to_cpu(raw_super->segment_count);
1636        sm_info->reserved_segments = le32_to_cpu(ckpt->rsvd_segment_count);
1637        sm_info->ovp_segments = le32_to_cpu(ckpt->overprov_segment_count);
1638        sm_info->main_segments = le32_to_cpu(raw_super->segment_count_main);
1639        sm_info->ssa_blkaddr = le32_to_cpu(raw_super->ssa_blkaddr);
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
1663static 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
1674static 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
1680static 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
1697static 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
1710static 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
1721static 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
1744void destroy_segment_manager(struct f2fs_sb_info *sbi)
1745{
1746        struct f2fs_sm_info *sm_info = SM_I(sbi);
1747        destroy_dirty_segmap(sbi);
1748        destroy_curseg(sbi);
1749        destroy_free_segmap(sbi);
1750        destroy_sit_info(sbi);
1751        sbi->sm_info = NULL;
1752        kfree(sm_info);
1753}
1754