fs/f2fs/segment.h

   1 /*
   2  * fs/f2fs/segment.h
   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/blkdev.h>
  12 #include <linux/backing-dev.h>
  13
  14 /* constant macro */
  15 #define NULL_SEGNO                      ((unsigned int)(~0))
  16 #define NULL_SECNO                      ((unsigned int)(~0))
  17
  18 #define DEF_RECLAIM_PREFREE_SEGMENTS    5       /* 5% over total segments */
  19 #define DEF_MAX_RECLAIM_PREFREE_SEGMENTS        4096    /* 8GB in maximum */
  20
  21 #define F2FS_MIN_SEGMENTS       9 /* SB + 2 (CP + SIT + NAT) + SSA + MAIN */
  22
  23 /* L: Logical segment # in volume, R: Relative segment # in main area */
  24 #define GET_L2R_SEGNO(free_i, segno)    (segno - free_i->start_segno)
  25 #define GET_R2L_SEGNO(free_i, segno)    (segno + free_i->start_segno)
  26
  27 #define IS_DATASEG(t)   (t <= CURSEG_COLD_DATA)
  28 #define IS_NODESEG(t)   (t >= CURSEG_HOT_NODE)
  29
  30 #define IS_CURSEG(sbi, seg)                                             \
  31         ((seg == CURSEG_I(sbi, CURSEG_HOT_DATA)->segno) ||      \
  32          (seg == CURSEG_I(sbi, CURSEG_WARM_DATA)->segno) ||     \
  33          (seg == CURSEG_I(sbi, CURSEG_COLD_DATA)->segno) ||     \
  34          (seg == CURSEG_I(sbi, CURSEG_HOT_NODE)->segno) ||      \
  35          (seg == CURSEG_I(sbi, CURSEG_WARM_NODE)->segno) ||     \
  36          (seg == CURSEG_I(sbi, CURSEG_COLD_NODE)->segno))
  37
  38 #define IS_CURSEC(sbi, secno)                                           \
  39         ((secno == CURSEG_I(sbi, CURSEG_HOT_DATA)->segno /              \
  40           sbi->segs_per_sec) || \
  41          (secno == CURSEG_I(sbi, CURSEG_WARM_DATA)->segno /             \
  42           sbi->segs_per_sec) || \
  43          (secno == CURSEG_I(sbi, CURSEG_COLD_DATA)->segno /             \
  44           sbi->segs_per_sec) || \
  45          (secno == CURSEG_I(sbi, CURSEG_HOT_NODE)->segno /              \
  46           sbi->segs_per_sec) || \
  47          (secno == CURSEG_I(sbi, CURSEG_WARM_NODE)->segno /             \
  48           sbi->segs_per_sec) || \
  49          (secno == CURSEG_I(sbi, CURSEG_COLD_NODE)->segno /             \
  50           sbi->segs_per_sec))   \
  51
  52 #define MAIN_BLKADDR(sbi)                                               \
  53         (SM_I(sbi) ? SM_I(sbi)->main_blkaddr :                          \
  54                 le32_to_cpu(F2FS_RAW_SUPER(sbi)->main_blkaddr))
  55 #define SEG0_BLKADDR(sbi)                                               \
  56         (SM_I(sbi) ? SM_I(sbi)->seg0_blkaddr :                          \
  57                 le32_to_cpu(F2FS_RAW_SUPER(sbi)->segment0_blkaddr))
  58
  59 #define MAIN_SEGS(sbi)  (SM_I(sbi)->main_segments)
  60 #define MAIN_SECS(sbi)  (sbi->total_sections)
  61
  62 #define TOTAL_SEGS(sbi)                                                 \
  63         (SM_I(sbi) ? SM_I(sbi)->segment_count :                                 \
  64                 le32_to_cpu(F2FS_RAW_SUPER(sbi)->segment_count))
  65 #define TOTAL_BLKS(sbi) (TOTAL_SEGS(sbi) << sbi->log_blocks_per_seg)
  66
  67 #define MAX_BLKADDR(sbi)        (SEG0_BLKADDR(sbi) + TOTAL_BLKS(sbi))
  68 #define SEGMENT_SIZE(sbi)       (1ULL << (sbi->log_blocksize +          \
  69                                         sbi->log_blocks_per_seg))
  70
  71 #define START_BLOCK(sbi, segno) (SEG0_BLKADDR(sbi) +                    \
  72          (GET_R2L_SEGNO(FREE_I(sbi), segno) << sbi->log_blocks_per_seg))
  73
  74 #define NEXT_FREE_BLKADDR(sbi, curseg)                                  \
  75         (START_BLOCK(sbi, curseg->segno) + curseg->next_blkoff)
  76
  77 #define GET_SEGOFF_FROM_SEG0(sbi, blk_addr)     ((blk_addr) - SEG0_BLKADDR(sbi))
  78 #define GET_SEGNO_FROM_SEG0(sbi, blk_addr)                              \
  79         (GET_SEGOFF_FROM_SEG0(sbi, blk_addr) >> sbi->log_blocks_per_seg)
  80 #define GET_BLKOFF_FROM_SEG0(sbi, blk_addr)                             \
  81         (GET_SEGOFF_FROM_SEG0(sbi, blk_addr) & (sbi->blocks_per_seg - 1))
  82
  83 #define GET_SEGNO(sbi, blk_addr)                                        \
  84         ((!is_valid_data_blkaddr(sbi, blk_addr)) ?                      \
  85         NULL_SEGNO : GET_L2R_SEGNO(FREE_I(sbi),                 \
  86                 GET_SEGNO_FROM_SEG0(sbi, blk_addr)))
  87 #define GET_SECNO(sbi, segno)                                   \
  88         ((segno) / sbi->segs_per_sec)
  89 #define GET_ZONENO_FROM_SEGNO(sbi, segno)                               \
  90         ((segno / sbi->segs_per_sec) / sbi->secs_per_zone)
  91
  92 #define GET_SUM_BLOCK(sbi, segno)                               \
  93         ((sbi->sm_info->ssa_blkaddr) + segno)
  94
  95 #define GET_SUM_TYPE(footer) ((footer)->entry_type)
  96 #define SET_SUM_TYPE(footer, type) ((footer)->entry_type = type)
  97
  98 #define SIT_ENTRY_OFFSET(sit_i, segno)                                  \
  99         (segno % sit_i->sents_per_block)
 100 #define SIT_BLOCK_OFFSET(segno)                                 \
 101         (segno / SIT_ENTRY_PER_BLOCK)
 102 #define START_SEGNO(segno)              \
 103         (SIT_BLOCK_OFFSET(segno) * SIT_ENTRY_PER_BLOCK)
 104 #define SIT_BLK_CNT(sbi)                        \
 105         ((MAIN_SEGS(sbi) + SIT_ENTRY_PER_BLOCK - 1) / SIT_ENTRY_PER_BLOCK)
 106 #define f2fs_bitmap_size(nr)                    \
 107         (BITS_TO_LONGS(nr) * sizeof(unsigned long))
 108
 109 #define SECTOR_FROM_BLOCK(blk_addr)                                     \
 110         (((sector_t)blk_addr) << F2FS_LOG_SECTORS_PER_BLOCK)
 111 #define SECTOR_TO_BLOCK(sectors)                                        \
 112         (sectors >> F2FS_LOG_SECTORS_PER_BLOCK)
 113 #define MAX_BIO_BLOCKS(sbi)                                             \
 114         ((int)min((int)max_hw_blocks(sbi), BIO_MAX_PAGES))
 115
 116 /*
 117  * indicate a block allocation direction: RIGHT and LEFT.
 118  * RIGHT means allocating new sections towards the end of volume.
 119  * LEFT means the opposite direction.
 120  */
 121 enum {
 122         ALLOC_RIGHT = 0,
 123         ALLOC_LEFT
 124 };
 125
 126 /*
 127  * In the victim_sel_policy->alloc_mode, there are two block allocation modes.
 128  * LFS writes data sequentially with cleaning operations.
 129  * SSR (Slack Space Recycle) reuses obsolete space without cleaning operations.
 130  */
 131 enum {
 132         LFS = 0,
 133         SSR
 134 };
 135
 136 /*
 137  * In the victim_sel_policy->gc_mode, there are two gc, aka cleaning, modes.
 138  * GC_CB is based on cost-benefit algorithm.
 139  * GC_GREEDY is based on greedy algorithm.
 140  */
 141 enum {
 142         GC_CB = 0,
 143         GC_GREEDY
 144 };
 145
 146 /*
 147  * BG_GC means the background cleaning job.
 148  * FG_GC means the on-demand cleaning job.
 149  * FORCE_FG_GC means on-demand cleaning job in background.
 150  */
 151 enum {
 152         BG_GC = 0,
 153         FG_GC,
 154         FORCE_FG_GC,
 155 };
 156
 157 /* for a function parameter to select a victim segment */
 158 struct victim_sel_policy {
 159         int alloc_mode;                 /* LFS or SSR */
 160         int gc_mode;                    /* GC_CB or GC_GREEDY */
 161         unsigned long *dirty_segmap;    /* dirty segment bitmap */
 162         unsigned int max_search;        /* maximum # of segments to search */
 163         unsigned int offset;            /* last scanned bitmap offset */
 164         unsigned int ofs_unit;          /* bitmap search unit */
 165         unsigned int min_cost;          /* minimum cost */
 166         unsigned int min_segno;         /* segment # having min. cost */
 167 };
 168
 169 struct seg_entry {
 170         unsigned int type:6;            /* segment type like CURSEG_XXX_TYPE */
 171         unsigned int valid_blocks:10;   /* # of valid blocks */
 172         unsigned int ckpt_valid_blocks:10;      /* # of valid blocks last cp */
 173         unsigned int padding:6;         /* padding */
 174         unsigned char *cur_valid_map;   /* validity bitmap of blocks */
 175         /*
 176          * # of valid blocks and the validity bitmap stored in the the last
 177          * checkpoint pack. This information is used by the SSR mode.
 178          */
 179         unsigned char *ckpt_valid_map;  /* validity bitmap of blocks last cp */
 180         unsigned char *discard_map;
 181         unsigned long long mtime;       /* modification time of the segment */
 182 };
 183
 184 struct sec_entry {
 185         unsigned int valid_blocks;      /* # of valid blocks in a section */
 186 };
 187
 188 struct segment_allocation {
 189         void (*allocate_segment)(struct f2fs_sb_info *, int, bool);
 190 };
 191
 192 /*
 193  * this value is set in page as a private data which indicate that
 194  * the page is atomically written, and it is in inmem_pages list.
 195  */
 196 #define ATOMIC_WRITTEN_PAGE             ((unsigned long)-1)
 197
 198 #define IS_ATOMIC_WRITTEN_PAGE(page)                    \
 199                 (page_private(page) == (unsigned long)ATOMIC_WRITTEN_PAGE)
 200
 201 struct inmem_pages {
 202         struct list_head list;
 203         struct page *page;
 204         block_t old_addr;               /* for revoking when fail to commit */
 205 };
 206
 207 struct sit_info {
 208         const struct segment_allocation *s_ops;
 209
 210         block_t sit_base_addr;          /* start block address of SIT area */
 211         block_t sit_blocks;             /* # of blocks used by SIT area */
 212         block_t written_valid_blocks;   /* # of valid blocks in main area */
 213         char *sit_bitmap;               /* SIT bitmap pointer */
 214         unsigned int bitmap_size;       /* SIT bitmap size */
 215
 216         unsigned long *tmp_map;                 /* bitmap for temporal use */
 217         unsigned long *dirty_sentries_bitmap;   /* bitmap for dirty sentries */
 218         unsigned int dirty_sentries;            /* # of dirty sentries */
 219         unsigned int sents_per_block;           /* # of SIT entries per block */
 220         struct mutex sentry_lock;               /* to protect SIT cache */
 221         struct seg_entry *sentries;             /* SIT segment-level cache */
 222         struct sec_entry *sec_entries;          /* SIT section-level cache */
 223
 224         /* for cost-benefit algorithm in cleaning procedure */
 225         unsigned long long elapsed_time;        /* elapsed time after mount */
 226         unsigned long long mounted_time;        /* mount time */
 227         unsigned long long min_mtime;           /* min. modification time */
 228         unsigned long long max_mtime;           /* max. modification time */
 229 };
 230
 231 struct free_segmap_info {
 232         unsigned int start_segno;       /* start segment number logically */
 233         unsigned int free_segments;     /* # of free segments */
 234         unsigned int free_sections;     /* # of free sections */
 235         spinlock_t segmap_lock;         /* free segmap lock */
 236         unsigned long *free_segmap;     /* free segment bitmap */
 237         unsigned long *free_secmap;     /* free section bitmap */
 238 };
 239
 240 /* Notice: The order of dirty type is same with CURSEG_XXX in f2fs.h */
 241 enum dirty_type {
 242         DIRTY_HOT_DATA,         /* dirty segments assigned as hot data logs */
 243         DIRTY_WARM_DATA,        /* dirty segments assigned as warm data logs */
 244         DIRTY_COLD_DATA,        /* dirty segments assigned as cold data logs */
 245         DIRTY_HOT_NODE,         /* dirty segments assigned as hot node logs */
 246         DIRTY_WARM_NODE,        /* dirty segments assigned as warm node logs */
 247         DIRTY_COLD_NODE,        /* dirty segments assigned as cold node logs */
 248         DIRTY,                  /* to count # of dirty segments */
 249         PRE,                    /* to count # of entirely obsolete segments */
 250         NR_DIRTY_TYPE
 251 };
 252
 253 struct dirty_seglist_info {
 254         const struct victim_selection *v_ops;   /* victim selction operation */
 255         unsigned long *dirty_segmap[NR_DIRTY_TYPE];
 256         struct mutex seglist_lock;              /* lock for segment bitmaps */
 257         int nr_dirty[NR_DIRTY_TYPE];            /* # of dirty segments */
 258         unsigned long *victim_secmap;           /* background GC victims */
 259 };
 260
 261 /* victim selection function for cleaning and SSR */
 262 struct victim_selection {
 263         int (*get_victim)(struct f2fs_sb_info *, unsigned int *,
 264                                                         int, int, char);
 265 };
 266
 267 /* for active log information */
 268 struct curseg_info {
 269         struct mutex curseg_mutex;              /* lock for consistency */
 270         struct f2fs_summary_block *sum_blk;     /* cached summary block */
 271         struct rw_semaphore journal_rwsem;      /* protect journal area */
 272         struct f2fs_journal *journal;           /* cached journal info */
 273         unsigned char alloc_type;               /* current allocation type */
 274         unsigned int segno;                     /* current segment number */
 275         unsigned short next_blkoff;             /* next block offset to write */
 276         unsigned int zone;                      /* current zone number */
 277         unsigned int next_segno;                /* preallocated segment */
 278 };
 279
 280 struct sit_entry_set {
 281         struct list_head set_list;      /* link with all sit sets */
 282         unsigned int start_segno;       /* start segno of sits in set */
 283         unsigned int entry_cnt;         /* the # of sit entries in set */
 284 };
 285
 286 /*
 287  * inline functions
 288  */
 289 static inline struct curseg_info *CURSEG_I(struct f2fs_sb_info *sbi, int type)
 290 {
 291         return (struct curseg_info *)(SM_I(sbi)->curseg_array + type);
 292 }
 293
 294 static inline struct seg_entry *get_seg_entry(struct f2fs_sb_info *sbi,
 295                                                 unsigned int segno)
 296 {
 297         struct sit_info *sit_i = SIT_I(sbi);
 298         return &sit_i->sentries[segno];
 299 }
 300
 301 static inline struct sec_entry *get_sec_entry(struct f2fs_sb_info *sbi,
 302                                                 unsigned int segno)
 303 {
 304         struct sit_info *sit_i = SIT_I(sbi);
 305         return &sit_i->sec_entries[GET_SECNO(sbi, segno)];
 306 }
 307
 308 static inline unsigned int get_valid_blocks(struct f2fs_sb_info *sbi,
 309                                 unsigned int segno, int section)
 310 {
 311         /*
 312          * In order to get # of valid blocks in a section instantly from many
 313          * segments, f2fs manages two counting structures separately.
 314          */
 315         if (section > 1)
 316                 return get_sec_entry(sbi, segno)->valid_blocks;
 317         else
 318                 return get_seg_entry(sbi, segno)->valid_blocks;
 319 }
 320
 321 static inline void seg_info_from_raw_sit(struct seg_entry *se,
 322                                         struct f2fs_sit_entry *rs)
 323 {
 324         se->valid_blocks = GET_SIT_VBLOCKS(rs);
 325         se->ckpt_valid_blocks = GET_SIT_VBLOCKS(rs);
 326         memcpy(se->cur_valid_map, rs->valid_map, SIT_VBLOCK_MAP_SIZE);
 327         memcpy(se->ckpt_valid_map, rs->valid_map, SIT_VBLOCK_MAP_SIZE);
 328         se->type = GET_SIT_TYPE(rs);
 329         se->mtime = le64_to_cpu(rs->mtime);
 330 }
 331
 332 static inline void seg_info_to_raw_sit(struct seg_entry *se,
 333                                         struct f2fs_sit_entry *rs)
 334 {
 335         unsigned short raw_vblocks = (se->type << SIT_VBLOCKS_SHIFT) |
 336                                         se->valid_blocks;
 337         rs->vblocks = cpu_to_le16(raw_vblocks);
 338         memcpy(rs->valid_map, se->cur_valid_map, SIT_VBLOCK_MAP_SIZE);
 339         memcpy(se->ckpt_valid_map, rs->valid_map, SIT_VBLOCK_MAP_SIZE);
 340         se->ckpt_valid_blocks = se->valid_blocks;
 341         rs->mtime = cpu_to_le64(se->mtime);
 342 }
 343
 344 static inline unsigned int find_next_inuse(struct free_segmap_info *free_i,
 345                 unsigned int max, unsigned int segno)
 346 {
 347         unsigned int ret;
 348         spin_lock(&free_i->segmap_lock);
 349         ret = find_next_bit(free_i->free_segmap, max, segno);
 350         spin_unlock(&free_i->segmap_lock);
 351         return ret;
 352 }
 353
 354 static inline void __set_free(struct f2fs_sb_info *sbi, unsigned int segno)
 355 {
 356         struct free_segmap_info *free_i = FREE_I(sbi);
 357         unsigned int secno = segno / sbi->segs_per_sec;
 358         unsigned int start_segno = secno * sbi->segs_per_sec;
 359         unsigned int next;
 360
 361         spin_lock(&free_i->segmap_lock);
 362         clear_bit(segno, free_i->free_segmap);
 363         free_i->free_segments++;
 364
 365         next = find_next_bit(free_i->free_segmap,
 366                         start_segno + sbi->segs_per_sec, start_segno);
 367         if (next >= start_segno + sbi->segs_per_sec) {
 368                 clear_bit(secno, free_i->free_secmap);
 369                 free_i->free_sections++;
 370         }
 371         spin_unlock(&free_i->segmap_lock);
 372 }
 373
 374 static inline void __set_inuse(struct f2fs_sb_info *sbi,
 375                 unsigned int segno)
 376 {
 377         struct free_segmap_info *free_i = FREE_I(sbi);
 378         unsigned int secno = segno / sbi->segs_per_sec;
 379         set_bit(segno, free_i->free_segmap);
 380         free_i->free_segments--;
 381         if (!test_and_set_bit(secno, free_i->free_secmap))
 382                 free_i->free_sections--;
 383 }
 384
 385 static inline void __set_test_and_free(struct f2fs_sb_info *sbi,
 386                 unsigned int segno)
 387 {
 388         struct free_segmap_info *free_i = FREE_I(sbi);
 389         unsigned int secno = segno / sbi->segs_per_sec;
 390         unsigned int start_segno = secno * sbi->segs_per_sec;
 391         unsigned int next;
 392
 393         spin_lock(&free_i->segmap_lock);
 394         if (test_and_clear_bit(segno, free_i->free_segmap)) {
 395                 free_i->free_segments++;
 396
 397                 if (IS_CURSEC(sbi, secno))
 398                         goto skip_free;
 399                 next = find_next_bit(free_i->free_segmap,
 400                                 start_segno + sbi->segs_per_sec, start_segno);
 401                 if (next >= start_segno + sbi->segs_per_sec) {
 402                         if (test_and_clear_bit(secno, free_i->free_secmap))
 403                                 free_i->free_sections++;
 404                 }
 405         }
 406 skip_free:
 407         spin_unlock(&free_i->segmap_lock);
 408 }
 409
 410 static inline void __set_test_and_inuse(struct f2fs_sb_info *sbi,
 411                 unsigned int segno)
 412 {
 413         struct free_segmap_info *free_i = FREE_I(sbi);
 414         unsigned int secno = segno / sbi->segs_per_sec;
 415         spin_lock(&free_i->segmap_lock);
 416         if (!test_and_set_bit(segno, free_i->free_segmap)) {
 417                 free_i->free_segments--;
 418                 if (!test_and_set_bit(secno, free_i->free_secmap))
 419                         free_i->free_sections--;
 420         }
 421         spin_unlock(&free_i->segmap_lock);
 422 }
 423
 424 static inline void get_sit_bitmap(struct f2fs_sb_info *sbi,
 425                 void *dst_addr)
 426 {
 427         struct sit_info *sit_i = SIT_I(sbi);
 428         memcpy(dst_addr, sit_i->sit_bitmap, sit_i->bitmap_size);
 429 }
 430
 431 static inline block_t written_block_count(struct f2fs_sb_info *sbi)
 432 {
 433         return SIT_I(sbi)->written_valid_blocks;
 434 }
 435
 436 static inline unsigned int free_segments(struct f2fs_sb_info *sbi)
 437 {
 438         return FREE_I(sbi)->free_segments;
 439 }
 440
 441 static inline int reserved_segments(struct f2fs_sb_info *sbi)
 442 {
 443         return SM_I(sbi)->reserved_segments;
 444 }
 445
 446 static inline unsigned int free_sections(struct f2fs_sb_info *sbi)
 447 {
 448         return FREE_I(sbi)->free_sections;
 449 }
 450
 451 static inline unsigned int prefree_segments(struct f2fs_sb_info *sbi)
 452 {
 453         return DIRTY_I(sbi)->nr_dirty[PRE];
 454 }
 455
 456 static inline unsigned int dirty_segments(struct f2fs_sb_info *sbi)
 457 {
 458         return DIRTY_I(sbi)->nr_dirty[DIRTY_HOT_DATA] +
 459                 DIRTY_I(sbi)->nr_dirty[DIRTY_WARM_DATA] +
 460                 DIRTY_I(sbi)->nr_dirty[DIRTY_COLD_DATA] +
 461                 DIRTY_I(sbi)->nr_dirty[DIRTY_HOT_NODE] +
 462                 DIRTY_I(sbi)->nr_dirty[DIRTY_WARM_NODE] +
 463                 DIRTY_I(sbi)->nr_dirty[DIRTY_COLD_NODE];
 464 }
 465
 466 static inline int overprovision_segments(struct f2fs_sb_info *sbi)
 467 {
 468         return SM_I(sbi)->ovp_segments;
 469 }
 470
 471 static inline int overprovision_sections(struct f2fs_sb_info *sbi)
 472 {
 473         return ((unsigned int) overprovision_segments(sbi)) / sbi->segs_per_sec;
 474 }
 475
 476 static inline int reserved_sections(struct f2fs_sb_info *sbi)
 477 {
 478         return ((unsigned int) reserved_segments(sbi)) / sbi->segs_per_sec;
 479 }
 480
 481 static inline bool need_SSR(struct f2fs_sb_info *sbi)
 482 {
 483         int node_secs = get_blocktype_secs(sbi, F2FS_DIRTY_NODES);
 484         int dent_secs = get_blocktype_secs(sbi, F2FS_DIRTY_DENTS);
 485
 486         if (test_opt(sbi, LFS))
 487                 return false;
 488
 489         return free_sections(sbi) <= (node_secs + 2 * dent_secs +
 490                                                 reserved_sections(sbi) + 1);
 491 }
 492
 493 static inline bool has_not_enough_free_secs(struct f2fs_sb_info *sbi,
 494                                         int freed, int needed)
 495 {
 496         int node_secs = get_blocktype_secs(sbi, F2FS_DIRTY_NODES);
 497         int dent_secs = get_blocktype_secs(sbi, F2FS_DIRTY_DENTS);
 498
 499         node_secs += get_blocktype_secs(sbi, F2FS_DIRTY_IMETA);
 500
 501         if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
 502                 return false;
 503
 504         return (free_sections(sbi) + freed) <=
 505                 (node_secs + 2 * dent_secs + reserved_sections(sbi) + needed);
 506 }
 507
 508 static inline bool excess_prefree_segs(struct f2fs_sb_info *sbi)
 509 {
 510         return prefree_segments(sbi) > SM_I(sbi)->rec_prefree_segments;
 511 }
 512
 513 static inline int utilization(struct f2fs_sb_info *sbi)
 514 {
 515         return div_u64((u64)valid_user_blocks(sbi) * 100,
 516                                         sbi->user_block_count);
 517 }
 518
 519 /*
 520  * Sometimes f2fs may be better to drop out-of-place update policy.
 521  * And, users can control the policy through sysfs entries.
 522  * There are five policies with triggering conditions as follows.
 523  * F2FS_IPU_FORCE - all the time,
 524  * F2FS_IPU_SSR - if SSR mode is activated,
 525  * F2FS_IPU_UTIL - if FS utilization is over threashold,
 526  * F2FS_IPU_SSR_UTIL - if SSR mode is activated and FS utilization is over
 527  *                     threashold,
 528  * F2FS_IPU_FSYNC - activated in fsync path only for high performance flash
 529  *                     storages. IPU will be triggered only if the # of dirty
 530  *                     pages over min_fsync_blocks.
 531  * F2FS_IPUT_DISABLE - disable IPU. (=default option)
 532  */
 533 #define DEF_MIN_IPU_UTIL        70
 534 #define DEF_MIN_FSYNC_BLOCKS    8
 535
 536 enum {
 537         F2FS_IPU_FORCE,
 538         F2FS_IPU_SSR,
 539         F2FS_IPU_UTIL,
 540         F2FS_IPU_SSR_UTIL,
 541         F2FS_IPU_FSYNC,
 542 };
 543
 544 static inline bool need_inplace_update(struct inode *inode)
 545 {
 546         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 547         unsigned int policy = SM_I(sbi)->ipu_policy;
 548
 549         /* IPU can be done only for the user data */
 550         if (S_ISDIR(inode->i_mode) || f2fs_is_atomic_file(inode))
 551                 return false;
 552
 553         if (test_opt(sbi, LFS))
 554                 return false;
 555
 556         if (policy & (0x1 << F2FS_IPU_FORCE))
 557                 return true;
 558         if (policy & (0x1 << F2FS_IPU_SSR) && need_SSR(sbi))
 559                 return true;
 560         if (policy & (0x1 << F2FS_IPU_UTIL) &&
 561                         utilization(sbi) > SM_I(sbi)->min_ipu_util)
 562                 return true;
 563         if (policy & (0x1 << F2FS_IPU_SSR_UTIL) && need_SSR(sbi) &&
 564                         utilization(sbi) > SM_I(sbi)->min_ipu_util)
 565                 return true;
 566
 567         /* this is only set during fdatasync */
 568         if (policy & (0x1 << F2FS_IPU_FSYNC) &&
 569                         is_inode_flag_set(inode, FI_NEED_IPU))
 570                 return true;
 571
 572         return false;
 573 }
 574
 575 static inline unsigned int curseg_segno(struct f2fs_sb_info *sbi,
 576                 int type)
 577 {
 578         struct curseg_info *curseg = CURSEG_I(sbi, type);
 579         return curseg->segno;
 580 }
 581
 582 static inline unsigned char curseg_alloc_type(struct f2fs_sb_info *sbi,
 583                 int type)
 584 {
 585         struct curseg_info *curseg = CURSEG_I(sbi, type);
 586         return curseg->alloc_type;
 587 }
 588
 589 static inline unsigned short curseg_blkoff(struct f2fs_sb_info *sbi, int type)
 590 {
 591         struct curseg_info *curseg = CURSEG_I(sbi, type);
 592         return curseg->next_blkoff;
 593 }
 594
 595 static inline void check_seg_range(struct f2fs_sb_info *sbi, unsigned int segno)
 596 {
 597         f2fs_bug_on(sbi, segno > TOTAL_SEGS(sbi) - 1);
 598 }
 599
 600 static inline void verify_block_addr(struct f2fs_io_info *fio, block_t blk_addr)
 601 {
 602         struct f2fs_sb_info *sbi = fio->sbi;
 603
 604         if (__is_meta_io(fio))
 605                 verify_blkaddr(sbi, blk_addr, META_GENERIC);
 606         else
 607                 verify_blkaddr(sbi, blk_addr, DATA_GENERIC);
 608 }
 609
 610 /*
 611  * Summary block is always treated as an invalid block
 612  */
 613 static inline int check_block_count(struct f2fs_sb_info *sbi,
 614                 int segno, struct f2fs_sit_entry *raw_sit)
 615 {
 616         bool is_valid  = test_bit_le(0, raw_sit->valid_map) ? true : false;
 617         int valid_blocks = 0;
 618         int cur_pos = 0, next_pos;
 619
 620         /* check bitmap with valid block count */
 621         do {
 622                 if (is_valid) {
 623                         next_pos = find_next_zero_bit_le(&raw_sit->valid_map,
 624                                         sbi->blocks_per_seg,
 625                                         cur_pos);
 626                         valid_blocks += next_pos - cur_pos;
 627                 } else
 628                         next_pos = find_next_bit_le(&raw_sit->valid_map,
 629                                         sbi->blocks_per_seg,
 630                                         cur_pos);
 631                 cur_pos = next_pos;
 632                 is_valid = !is_valid;
 633         } while (cur_pos < sbi->blocks_per_seg);
 634
 635         if (unlikely(GET_SIT_VBLOCKS(raw_sit) != valid_blocks)) {
 636                 f2fs_msg(sbi->sb, KERN_ERR,
 637                                 "Mismatch valid blocks %d vs. %d",
 638                                         GET_SIT_VBLOCKS(raw_sit), valid_blocks);
 639                 set_sbi_flag(sbi, SBI_NEED_FSCK);
 640                 return -EINVAL;
 641         }
 642
 643         /* check segment usage, and check boundary of a given segment number */
 644         if (unlikely(GET_SIT_VBLOCKS(raw_sit) > sbi->blocks_per_seg
 645                                         || segno > TOTAL_SEGS(sbi) - 1)) {
 646                 f2fs_msg(sbi->sb, KERN_ERR,
 647                                 "Wrong valid blocks %d or segno %u",
 648                                         GET_SIT_VBLOCKS(raw_sit), segno);
 649                 set_sbi_flag(sbi, SBI_NEED_FSCK);
 650                 return -EINVAL;
 651         }
 652         return 0;
 653 }
 654
 655 static inline pgoff_t current_sit_addr(struct f2fs_sb_info *sbi,
 656                                                 unsigned int start)
 657 {
 658         struct sit_info *sit_i = SIT_I(sbi);
 659         unsigned int offset = SIT_BLOCK_OFFSET(start);
 660         block_t blk_addr = sit_i->sit_base_addr + offset;
 661
 662         check_seg_range(sbi, start);
 663
 664         /* calculate sit block address */
 665         if (f2fs_test_bit(offset, sit_i->sit_bitmap))
 666                 blk_addr += sit_i->sit_blocks;
 667
 668         return blk_addr;
 669 }
 670
 671 static inline pgoff_t next_sit_addr(struct f2fs_sb_info *sbi,
 672                                                 pgoff_t block_addr)
 673 {
 674         struct sit_info *sit_i = SIT_I(sbi);
 675         block_addr -= sit_i->sit_base_addr;
 676         if (block_addr < sit_i->sit_blocks)
 677                 block_addr += sit_i->sit_blocks;
 678         else
 679                 block_addr -= sit_i->sit_blocks;
 680
 681         return block_addr + sit_i->sit_base_addr;
 682 }
 683
 684 static inline void set_to_next_sit(struct sit_info *sit_i, unsigned int start)
 685 {
 686         unsigned int block_off = SIT_BLOCK_OFFSET(start);
 687
 688         f2fs_change_bit(block_off, sit_i->sit_bitmap);
 689 }
 690
 691 static inline unsigned long long get_mtime(struct f2fs_sb_info *sbi)
 692 {
 693         struct sit_info *sit_i = SIT_I(sbi);
 694         return sit_i->elapsed_time + CURRENT_TIME_SEC.tv_sec -
 695                                                 sit_i->mounted_time;
 696 }
 697
 698 static inline void set_summary(struct f2fs_summary *sum, nid_t nid,
 699                         unsigned int ofs_in_node, unsigned char version)
 700 {
 701         sum->nid = cpu_to_le32(nid);
 702         sum->ofs_in_node = cpu_to_le16(ofs_in_node);
 703         sum->version = version;
 704 }
 705
 706 static inline block_t start_sum_block(struct f2fs_sb_info *sbi)
 707 {
 708         return __start_cp_addr(sbi) +
 709                 le32_to_cpu(F2FS_CKPT(sbi)->cp_pack_start_sum);
 710 }
 711
 712 static inline block_t sum_blk_addr(struct f2fs_sb_info *sbi, int base, int type)
 713 {
 714         return __start_cp_addr(sbi) +
 715                 le32_to_cpu(F2FS_CKPT(sbi)->cp_pack_total_block_count)
 716                                 - (base + 1) + type;
 717 }
 718
 719 static inline bool no_fggc_candidate(struct f2fs_sb_info *sbi,
 720                                                 unsigned int secno)
 721 {
 722         if (get_valid_blocks(sbi, secno, sbi->segs_per_sec) >=
 723                                                 sbi->fggc_threshold)
 724                 return true;
 725         return false;
 726 }
 727
 728 static inline bool sec_usage_check(struct f2fs_sb_info *sbi, unsigned int secno)
 729 {
 730         if (IS_CURSEC(sbi, secno) || (sbi->cur_victim_sec == secno))
 731                 return true;
 732         return false;
 733 }
 734
 735 static inline unsigned int max_hw_blocks(struct f2fs_sb_info *sbi)
 736 {
 737         struct block_device *bdev = sbi->sb->s_bdev;
 738         struct request_queue *q = bdev_get_queue(bdev);
 739         return SECTOR_TO_BLOCK(queue_max_sectors(q));
 740 }
 741
 742 /*
 743  * It is very important to gather dirty pages and write at once, so that we can
 744  * submit a big bio without interfering other data writes.
 745  * By default, 512 pages for directory data,
 746  * 512 pages (2MB) * 3 for three types of nodes, and
 747  * max_bio_blocks for meta are set.
 748  */
 749 static inline int nr_pages_to_skip(struct f2fs_sb_info *sbi, int type)
 750 {
 751         if (sbi->sb->s_bdi->wb.dirty_exceeded)
 752                 return 0;
 753
 754         if (type == DATA)
 755                 return sbi->blocks_per_seg;
 756         else if (type == NODE)
 757                 return 8 * sbi->blocks_per_seg;
 758         else if (type == META)
 759                 return 8 * MAX_BIO_BLOCKS(sbi);
 760         else
 761                 return 0;
 762 }
 763
 764 /*
 765  * When writing pages, it'd better align nr_to_write for segment size.
 766  */
 767 static inline long nr_pages_to_write(struct f2fs_sb_info *sbi, int type,
 768                                         struct writeback_control *wbc)
 769 {
 770         long nr_to_write, desired;
 771
 772         if (wbc->sync_mode != WB_SYNC_NONE)
 773                 return 0;
 774
 775         nr_to_write = wbc->nr_to_write;
 776
 777         if (type == NODE)
 778                 desired = 2 * max_hw_blocks(sbi);
 779         else
 780                 desired = MAX_BIO_BLOCKS(sbi);
 781
 782         wbc->nr_to_write = desired;
 783         return desired - nr_to_write;
 784 }