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)       (SM_I(sbi)->main_blkaddr)
  53 #define SEG0_BLKADDR(sbi)       (SM_I(sbi)->seg0_blkaddr)
  54
  55 #define MAIN_SEGS(sbi)  (SM_I(sbi)->main_segments)
  56 #define MAIN_SECS(sbi)  (sbi->total_sections)
  57
  58 #define TOTAL_SEGS(sbi) (SM_I(sbi)->segment_count)
  59 #define TOTAL_BLKS(sbi) (TOTAL_SEGS(sbi) << sbi->log_blocks_per_seg)
  60
  61 #define MAX_BLKADDR(sbi)        (SEG0_BLKADDR(sbi) + TOTAL_BLKS(sbi))
  62 #define SEGMENT_SIZE(sbi)       (1ULL << (sbi->log_blocksize +          \
  63                                         sbi->log_blocks_per_seg))
  64
  65 #define START_BLOCK(sbi, segno) (SEG0_BLKADDR(sbi) +                    \
  66          (GET_R2L_SEGNO(FREE_I(sbi), segno) << sbi->log_blocks_per_seg))
  67
  68 #define NEXT_FREE_BLKADDR(sbi, curseg)                                  \
  69         (START_BLOCK(sbi, curseg->segno) + curseg->next_blkoff)
  70
  71 #define GET_SEGOFF_FROM_SEG0(sbi, blk_addr)     ((blk_addr) - SEG0_BLKADDR(sbi))
  72 #define GET_SEGNO_FROM_SEG0(sbi, blk_addr)                              \
  73         (GET_SEGOFF_FROM_SEG0(sbi, blk_addr) >> sbi->log_blocks_per_seg)
  74 #define GET_BLKOFF_FROM_SEG0(sbi, blk_addr)                             \
  75         (GET_SEGOFF_FROM_SEG0(sbi, blk_addr) & (sbi->blocks_per_seg - 1))
  76
  77 #define GET_SEGNO(sbi, blk_addr)                                        \
  78         (((blk_addr == NULL_ADDR) || (blk_addr == NEW_ADDR)) ?          \
  79         NULL_SEGNO : GET_L2R_SEGNO(FREE_I(sbi),                 \
  80                 GET_SEGNO_FROM_SEG0(sbi, blk_addr)))
  81 #define GET_SECNO(sbi, segno)                                   \
  82         ((segno) / sbi->segs_per_sec)
  83 #define GET_ZONENO_FROM_SEGNO(sbi, segno)                               \
  84         ((segno / sbi->segs_per_sec) / sbi->secs_per_zone)
  85
  86 #define GET_SUM_BLOCK(sbi, segno)                               \
  87         ((sbi->sm_info->ssa_blkaddr) + segno)
  88
  89 #define GET_SUM_TYPE(footer) ((footer)->entry_type)
  90 #define SET_SUM_TYPE(footer, type) ((footer)->entry_type = type)
  91
  92 #define SIT_ENTRY_OFFSET(sit_i, segno)                                  \
  93         (segno % sit_i->sents_per_block)
  94 #define SIT_BLOCK_OFFSET(segno)                                 \
  95         (segno / SIT_ENTRY_PER_BLOCK)
  96 #define START_SEGNO(segno)              \
  97         (SIT_BLOCK_OFFSET(segno) * SIT_ENTRY_PER_BLOCK)
  98 #define SIT_BLK_CNT(sbi)                        \
  99         ((MAIN_SEGS(sbi) + SIT_ENTRY_PER_BLOCK - 1) / SIT_ENTRY_PER_BLOCK)
 100 #define f2fs_bitmap_size(nr)                    \
 101         (BITS_TO_LONGS(nr) * sizeof(unsigned long))
 102
 103 #define SECTOR_FROM_BLOCK(blk_addr)                                     \
 104         (((sector_t)blk_addr) << F2FS_LOG_SECTORS_PER_BLOCK)
 105 #define SECTOR_TO_BLOCK(sectors)                                        \
 106         (sectors >> F2FS_LOG_SECTORS_PER_BLOCK)
 107
 108 /*
 109  * indicate a block allocation direction: RIGHT and LEFT.
 110  * RIGHT means allocating new sections towards the end of volume.
 111  * LEFT means the opposite direction.
 112  */
 113 enum {
 114         ALLOC_RIGHT = 0,
 115         ALLOC_LEFT
 116 };
 117
 118 /*
 119  * In the victim_sel_policy->alloc_mode, there are two block allocation modes.
 120  * LFS writes data sequentially with cleaning operations.
 121  * SSR (Slack Space Recycle) reuses obsolete space without cleaning operations.
 122  */
 123 enum {
 124         LFS = 0,
 125         SSR
 126 };
 127
 128 /*
 129  * In the victim_sel_policy->gc_mode, there are two gc, aka cleaning, modes.
 130  * GC_CB is based on cost-benefit algorithm.
 131  * GC_GREEDY is based on greedy algorithm.
 132  */
 133 enum {
 134         GC_CB = 0,
 135         GC_GREEDY
 136 };
 137
 138 /*
 139  * BG_GC means the background cleaning job.
 140  * FG_GC means the on-demand cleaning job.
 141  * FORCE_FG_GC means on-demand cleaning job in background.
 142  */
 143 enum {
 144         BG_GC = 0,
 145         FG_GC,
 146         FORCE_FG_GC,
 147 };
 148
 149 /* for a function parameter to select a victim segment */
 150 struct victim_sel_policy {
 151         int alloc_mode;                 /* LFS or SSR */
 152         int gc_mode;                    /* GC_CB or GC_GREEDY */
 153         unsigned long *dirty_segmap;    /* dirty segment bitmap */
 154         unsigned int max_search;        /* maximum # of segments to search */
 155         unsigned int offset;            /* last scanned bitmap offset */
 156         unsigned int ofs_unit;          /* bitmap search unit */
 157         unsigned int min_cost;          /* minimum cost */
 158         unsigned int min_segno;         /* segment # having min. cost */
 159 };
 160
 161 struct seg_entry {
 162         unsigned int type:6;            /* segment type like CURSEG_XXX_TYPE */
 163         unsigned int valid_blocks:10;   /* # of valid blocks */
 164         unsigned int ckpt_valid_blocks:10;      /* # of valid blocks last cp */
 165         unsigned int padding:6;         /* padding */
 166         unsigned char *cur_valid_map;   /* validity bitmap of blocks */
 167         /*
 168          * # of valid blocks and the validity bitmap stored in the the last
 169          * checkpoint pack. This information is used by the SSR mode.
 170          */
 171         unsigned char *ckpt_valid_map;  /* validity bitmap of blocks last cp */
 172         unsigned char *discard_map;
 173         unsigned long long mtime;       /* modification time of the segment */
 174 };
 175
 176 struct sec_entry {
 177         unsigned int valid_blocks;      /* # of valid blocks in a section */
 178 };
 179
 180 struct segment_allocation {
 181         void (*allocate_segment)(struct f2fs_sb_info *, int, bool);
 182 };
 183
 184 /*
 185  * this value is set in page as a private data which indicate that
 186  * the page is atomically written, and it is in inmem_pages list.
 187  */
 188 #define ATOMIC_WRITTEN_PAGE             ((unsigned long)-1)
 189
 190 #define IS_ATOMIC_WRITTEN_PAGE(page)                    \
 191                 (page_private(page) == (unsigned long)ATOMIC_WRITTEN_PAGE)
 192
 193 struct inmem_pages {
 194         struct list_head list;
 195         struct page *page;
 196         block_t old_addr;               /* for revoking when fail to commit */
 197 };
 198
 199 struct sit_info {
 200         const struct segment_allocation *s_ops;
 201
 202         block_t sit_base_addr;          /* start block address of SIT area */
 203         block_t sit_blocks;             /* # of blocks used by SIT area */
 204         block_t written_valid_blocks;   /* # of valid blocks in main area */
 205         char *sit_bitmap;               /* SIT bitmap pointer */
 206         unsigned int bitmap_size;       /* SIT bitmap size */
 207
 208         unsigned long *tmp_map;                 /* bitmap for temporal use */
 209         unsigned long *dirty_sentries_bitmap;   /* bitmap for dirty sentries */
 210         unsigned int dirty_sentries;            /* # of dirty sentries */
 211         unsigned int sents_per_block;           /* # of SIT entries per block */
 212         struct mutex sentry_lock;               /* to protect SIT cache */
 213         struct seg_entry *sentries;             /* SIT segment-level cache */
 214         struct sec_entry *sec_entries;          /* SIT section-level cache */
 215
 216         /* for cost-benefit algorithm in cleaning procedure */
 217         unsigned long long elapsed_time;        /* elapsed time after mount */
 218         unsigned long long mounted_time;        /* mount time */
 219         unsigned long long min_mtime;           /* min. modification time */
 220         unsigned long long max_mtime;           /* max. modification time */
 221 };
 222
 223 struct free_segmap_info {
 224         unsigned int start_segno;       /* start segment number logically */
 225         unsigned int free_segments;     /* # of free segments */
 226         unsigned int free_sections;     /* # of free sections */
 227         spinlock_t segmap_lock;         /* free segmap lock */
 228         unsigned long *free_segmap;     /* free segment bitmap */
 229         unsigned long *free_secmap;     /* free section bitmap */
 230 };
 231
 232 /* Notice: The order of dirty type is same with CURSEG_XXX in f2fs.h */
 233 enum dirty_type {
 234         DIRTY_HOT_DATA,         /* dirty segments assigned as hot data logs */
 235         DIRTY_WARM_DATA,        /* dirty segments assigned as warm data logs */
 236         DIRTY_COLD_DATA,        /* dirty segments assigned as cold data logs */
 237         DIRTY_HOT_NODE,         /* dirty segments assigned as hot node logs */
 238         DIRTY_WARM_NODE,        /* dirty segments assigned as warm node logs */
 239         DIRTY_COLD_NODE,        /* dirty segments assigned as cold node logs */
 240         DIRTY,                  /* to count # of dirty segments */
 241         PRE,                    /* to count # of entirely obsolete segments */
 242         NR_DIRTY_TYPE
 243 };
 244
 245 struct dirty_seglist_info {
 246         const struct victim_selection *v_ops;   /* victim selction operation */
 247         unsigned long *dirty_segmap[NR_DIRTY_TYPE];
 248         struct mutex seglist_lock;              /* lock for segment bitmaps */
 249         int nr_dirty[NR_DIRTY_TYPE];            /* # of dirty segments */
 250         unsigned long *victim_secmap;           /* background GC victims */
 251 };
 252
 253 /* victim selection function for cleaning and SSR */
 254 struct victim_selection {
 255         int (*get_victim)(struct f2fs_sb_info *, unsigned int *,
 256                                                         int, int, char);
 257 };
 258
 259 /* for active log information */
 260 struct curseg_info {
 261         struct mutex curseg_mutex;              /* lock for consistency */
 262         struct f2fs_summary_block *sum_blk;     /* cached summary block */
 263         struct rw_semaphore journal_rwsem;      /* protect journal area */
 264         struct f2fs_journal *journal;           /* cached journal info */
 265         unsigned char alloc_type;               /* current allocation type */
 266         unsigned int segno;                     /* current segment number */
 267         unsigned short next_blkoff;             /* next block offset to write */
 268         unsigned int zone;                      /* current zone number */
 269         unsigned int next_segno;                /* preallocated segment */
 270 };
 271
 272 struct sit_entry_set {
 273         struct list_head set_list;      /* link with all sit sets */
 274         unsigned int start_segno;       /* start segno of sits in set */
 275         unsigned int entry_cnt;         /* the # of sit entries in set */
 276 };
 277
 278 /*
 279  * inline functions
 280  */
 281 static inline struct curseg_info *CURSEG_I(struct f2fs_sb_info *sbi, int type)
 282 {
 283         return (struct curseg_info *)(SM_I(sbi)->curseg_array + type);
 284 }
 285
 286 static inline struct seg_entry *get_seg_entry(struct f2fs_sb_info *sbi,
 287                                                 unsigned int segno)
 288 {
 289         struct sit_info *sit_i = SIT_I(sbi);
 290         return &sit_i->sentries[segno];
 291 }
 292
 293 static inline struct sec_entry *get_sec_entry(struct f2fs_sb_info *sbi,
 294                                                 unsigned int segno)
 295 {
 296         struct sit_info *sit_i = SIT_I(sbi);
 297         return &sit_i->sec_entries[GET_SECNO(sbi, segno)];
 298 }
 299
 300 static inline unsigned int get_valid_blocks(struct f2fs_sb_info *sbi,
 301                                 unsigned int segno, int section)
 302 {
 303         /*
 304          * In order to get # of valid blocks in a section instantly from many
 305          * segments, f2fs manages two counting structures separately.
 306          */
 307         if (section > 1)
 308                 return get_sec_entry(sbi, segno)->valid_blocks;
 309         else
 310                 return get_seg_entry(sbi, segno)->valid_blocks;
 311 }
 312
 313 static inline void seg_info_from_raw_sit(struct seg_entry *se,
 314                                         struct f2fs_sit_entry *rs)
 315 {
 316         se->valid_blocks = GET_SIT_VBLOCKS(rs);
 317         se->ckpt_valid_blocks = GET_SIT_VBLOCKS(rs);
 318         memcpy(se->cur_valid_map, rs->valid_map, SIT_VBLOCK_MAP_SIZE);
 319         memcpy(se->ckpt_valid_map, rs->valid_map, SIT_VBLOCK_MAP_SIZE);
 320         se->type = GET_SIT_TYPE(rs);
 321         se->mtime = le64_to_cpu(rs->mtime);
 322 }
 323
 324 static inline void seg_info_to_raw_sit(struct seg_entry *se,
 325                                         struct f2fs_sit_entry *rs)
 326 {
 327         unsigned short raw_vblocks = (se->type << SIT_VBLOCKS_SHIFT) |
 328                                         se->valid_blocks;
 329         rs->vblocks = cpu_to_le16(raw_vblocks);
 330         memcpy(rs->valid_map, se->cur_valid_map, SIT_VBLOCK_MAP_SIZE);
 331         memcpy(se->ckpt_valid_map, rs->valid_map, SIT_VBLOCK_MAP_SIZE);
 332         se->ckpt_valid_blocks = se->valid_blocks;
 333         rs->mtime = cpu_to_le64(se->mtime);
 334 }
 335
 336 static inline unsigned int find_next_inuse(struct free_segmap_info *free_i,
 337                 unsigned int max, unsigned int segno)
 338 {
 339         unsigned int ret;
 340         spin_lock(&free_i->segmap_lock);
 341         ret = find_next_bit(free_i->free_segmap, max, segno);
 342         spin_unlock(&free_i->segmap_lock);
 343         return ret;
 344 }
 345
 346 static inline void __set_free(struct f2fs_sb_info *sbi, unsigned int segno)
 347 {
 348         struct free_segmap_info *free_i = FREE_I(sbi);
 349         unsigned int secno = segno / sbi->segs_per_sec;
 350         unsigned int start_segno = secno * sbi->segs_per_sec;
 351         unsigned int next;
 352
 353         spin_lock(&free_i->segmap_lock);
 354         clear_bit(segno, free_i->free_segmap);
 355         free_i->free_segments++;
 356
 357         next = find_next_bit(free_i->free_segmap,
 358                         start_segno + sbi->segs_per_sec, start_segno);
 359         if (next >= start_segno + sbi->segs_per_sec) {
 360                 clear_bit(secno, free_i->free_secmap);
 361                 free_i->free_sections++;
 362         }
 363         spin_unlock(&free_i->segmap_lock);
 364 }
 365
 366 static inline void __set_inuse(struct f2fs_sb_info *sbi,
 367                 unsigned int segno)
 368 {
 369         struct free_segmap_info *free_i = FREE_I(sbi);
 370         unsigned int secno = segno / sbi->segs_per_sec;
 371         set_bit(segno, free_i->free_segmap);
 372         free_i->free_segments--;
 373         if (!test_and_set_bit(secno, free_i->free_secmap))
 374                 free_i->free_sections--;
 375 }
 376
 377 static inline void __set_test_and_free(struct f2fs_sb_info *sbi,
 378                 unsigned int segno)
 379 {
 380         struct free_segmap_info *free_i = FREE_I(sbi);
 381         unsigned int secno = segno / sbi->segs_per_sec;
 382         unsigned int start_segno = secno * sbi->segs_per_sec;
 383         unsigned int next;
 384
 385         spin_lock(&free_i->segmap_lock);
 386         if (test_and_clear_bit(segno, free_i->free_segmap)) {
 387                 free_i->free_segments++;
 388
 389                 next = find_next_bit(free_i->free_segmap,
 390                                 start_segno + sbi->segs_per_sec, start_segno);
 391                 if (next >= start_segno + sbi->segs_per_sec) {
 392                         if (test_and_clear_bit(secno, free_i->free_secmap))
 393                                 free_i->free_sections++;
 394                 }
 395         }
 396         spin_unlock(&free_i->segmap_lock);
 397 }
 398
 399 static inline void __set_test_and_inuse(struct f2fs_sb_info *sbi,
 400                 unsigned int segno)
 401 {
 402         struct free_segmap_info *free_i = FREE_I(sbi);
 403         unsigned int secno = segno / sbi->segs_per_sec;
 404         spin_lock(&free_i->segmap_lock);
 405         if (!test_and_set_bit(segno, free_i->free_segmap)) {
 406                 free_i->free_segments--;
 407                 if (!test_and_set_bit(secno, free_i->free_secmap))
 408                         free_i->free_sections--;
 409         }
 410         spin_unlock(&free_i->segmap_lock);
 411 }
 412
 413 static inline void get_sit_bitmap(struct f2fs_sb_info *sbi,
 414                 void *dst_addr)
 415 {
 416         struct sit_info *sit_i = SIT_I(sbi);
 417         memcpy(dst_addr, sit_i->sit_bitmap, sit_i->bitmap_size);
 418 }
 419
 420 static inline block_t written_block_count(struct f2fs_sb_info *sbi)
 421 {
 422         return SIT_I(sbi)->written_valid_blocks;
 423 }
 424
 425 static inline unsigned int free_segments(struct f2fs_sb_info *sbi)
 426 {
 427         return FREE_I(sbi)->free_segments;
 428 }
 429
 430 static inline int reserved_segments(struct f2fs_sb_info *sbi)
 431 {
 432         return SM_I(sbi)->reserved_segments;
 433 }
 434
 435 static inline unsigned int free_sections(struct f2fs_sb_info *sbi)
 436 {
 437         return FREE_I(sbi)->free_sections;
 438 }
 439
 440 static inline unsigned int prefree_segments(struct f2fs_sb_info *sbi)
 441 {
 442         return DIRTY_I(sbi)->nr_dirty[PRE];
 443 }
 444
 445 static inline unsigned int dirty_segments(struct f2fs_sb_info *sbi)
 446 {
 447         return DIRTY_I(sbi)->nr_dirty[DIRTY_HOT_DATA] +
 448                 DIRTY_I(sbi)->nr_dirty[DIRTY_WARM_DATA] +
 449                 DIRTY_I(sbi)->nr_dirty[DIRTY_COLD_DATA] +
 450                 DIRTY_I(sbi)->nr_dirty[DIRTY_HOT_NODE] +
 451                 DIRTY_I(sbi)->nr_dirty[DIRTY_WARM_NODE] +
 452                 DIRTY_I(sbi)->nr_dirty[DIRTY_COLD_NODE];
 453 }
 454
 455 static inline int overprovision_segments(struct f2fs_sb_info *sbi)
 456 {
 457         return SM_I(sbi)->ovp_segments;
 458 }
 459
 460 static inline int overprovision_sections(struct f2fs_sb_info *sbi)
 461 {
 462         return ((unsigned int) overprovision_segments(sbi)) / sbi->segs_per_sec;
 463 }
 464
 465 static inline int reserved_sections(struct f2fs_sb_info *sbi)
 466 {
 467         return ((unsigned int) reserved_segments(sbi)) / sbi->segs_per_sec;
 468 }
 469
 470 static inline bool need_SSR(struct f2fs_sb_info *sbi)
 471 {
 472         int node_secs = get_blocktype_secs(sbi, F2FS_DIRTY_NODES);
 473         int dent_secs = get_blocktype_secs(sbi, F2FS_DIRTY_DENTS);
 474         int imeta_secs = get_blocktype_secs(sbi, F2FS_DIRTY_IMETA);
 475
 476         if (test_opt(sbi, LFS))
 477                 return false;
 478
 479         return free_sections(sbi) <= (node_secs + 2 * dent_secs + imeta_secs +
 480                                                 reserved_sections(sbi) + 1);
 481 }
 482
 483 static inline bool has_not_enough_free_secs(struct f2fs_sb_info *sbi,
 484                                         int freed, int needed)
 485 {
 486         int node_secs = get_blocktype_secs(sbi, F2FS_DIRTY_NODES);
 487         int dent_secs = get_blocktype_secs(sbi, F2FS_DIRTY_DENTS);
 488         int imeta_secs = get_blocktype_secs(sbi, F2FS_DIRTY_IMETA);
 489
 490         if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
 491                 return false;
 492
 493         return (free_sections(sbi) + freed) <=
 494                 (node_secs + 2 * dent_secs + imeta_secs +
 495                 reserved_sections(sbi) + needed);
 496 }
 497
 498 static inline bool excess_prefree_segs(struct f2fs_sb_info *sbi)
 499 {
 500         return prefree_segments(sbi) > SM_I(sbi)->rec_prefree_segments;
 501 }
 502
 503 static inline int utilization(struct f2fs_sb_info *sbi)
 504 {
 505         return div_u64((u64)valid_user_blocks(sbi) * 100,
 506                                         sbi->user_block_count);
 507 }
 508
 509 /*
 510  * Sometimes f2fs may be better to drop out-of-place update policy.
 511  * And, users can control the policy through sysfs entries.
 512  * There are five policies with triggering conditions as follows.
 513  * F2FS_IPU_FORCE - all the time,
 514  * F2FS_IPU_SSR - if SSR mode is activated,
 515  * F2FS_IPU_UTIL - if FS utilization is over threashold,
 516  * F2FS_IPU_SSR_UTIL - if SSR mode is activated and FS utilization is over
 517  *                     threashold,
 518  * F2FS_IPU_FSYNC - activated in fsync path only for high performance flash
 519  *                     storages. IPU will be triggered only if the # of dirty
 520  *                     pages over min_fsync_blocks.
 521  * F2FS_IPUT_DISABLE - disable IPU. (=default option)
 522  */
 523 #define DEF_MIN_IPU_UTIL        70
 524 #define DEF_MIN_FSYNC_BLOCKS    8
 525
 526 enum {
 527         F2FS_IPU_FORCE,
 528         F2FS_IPU_SSR,
 529         F2FS_IPU_UTIL,
 530         F2FS_IPU_SSR_UTIL,
 531         F2FS_IPU_FSYNC,
 532 };
 533
 534 static inline bool need_inplace_update(struct inode *inode)
 535 {
 536         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 537         unsigned int policy = SM_I(sbi)->ipu_policy;
 538
 539         /* IPU can be done only for the user data */
 540         if (S_ISDIR(inode->i_mode) || f2fs_is_atomic_file(inode))
 541                 return false;
 542
 543         if (test_opt(sbi, LFS))
 544                 return false;
 545
 546         if (policy & (0x1 << F2FS_IPU_FORCE))
 547                 return true;
 548         if (policy & (0x1 << F2FS_IPU_SSR) && need_SSR(sbi))
 549                 return true;
 550         if (policy & (0x1 << F2FS_IPU_UTIL) &&
 551                         utilization(sbi) > SM_I(sbi)->min_ipu_util)
 552                 return true;
 553         if (policy & (0x1 << F2FS_IPU_SSR_UTIL) && need_SSR(sbi) &&
 554                         utilization(sbi) > SM_I(sbi)->min_ipu_util)
 555                 return true;
 556
 557         /* this is only set during fdatasync */
 558         if (policy & (0x1 << F2FS_IPU_FSYNC) &&
 559                         is_inode_flag_set(inode, FI_NEED_IPU))
 560                 return true;
 561
 562         return false;
 563 }
 564
 565 static inline unsigned int curseg_segno(struct f2fs_sb_info *sbi,
 566                 int type)
 567 {
 568         struct curseg_info *curseg = CURSEG_I(sbi, type);
 569         return curseg->segno;
 570 }
 571
 572 static inline unsigned char curseg_alloc_type(struct f2fs_sb_info *sbi,
 573                 int type)
 574 {
 575         struct curseg_info *curseg = CURSEG_I(sbi, type);
 576         return curseg->alloc_type;
 577 }
 578
 579 static inline unsigned short curseg_blkoff(struct f2fs_sb_info *sbi, int type)
 580 {
 581         struct curseg_info *curseg = CURSEG_I(sbi, type);
 582         return curseg->next_blkoff;
 583 }
 584
 585 static inline void check_seg_range(struct f2fs_sb_info *sbi, unsigned int segno)
 586 {
 587         f2fs_bug_on(sbi, segno > TOTAL_SEGS(sbi) - 1);
 588 }
 589
 590 static inline void verify_block_addr(struct f2fs_sb_info *sbi, block_t blk_addr)
 591 {
 592         BUG_ON(blk_addr < SEG0_BLKADDR(sbi)
 593                         || blk_addr >= MAX_BLKADDR(sbi));
 594 }
 595
 596 /*
 597  * Summary block is always treated as an invalid block
 598  */
 599 static inline void check_block_count(struct f2fs_sb_info *sbi,
 600                 int segno, struct f2fs_sit_entry *raw_sit)
 601 {
 602 #ifdef CONFIG_F2FS_CHECK_FS
 603         bool is_valid  = test_bit_le(0, raw_sit->valid_map) ? true : false;
 604         int valid_blocks = 0;
 605         int cur_pos = 0, next_pos;
 606
 607         /* check bitmap with valid block count */
 608         do {
 609                 if (is_valid) {
 610                         next_pos = find_next_zero_bit_le(&raw_sit->valid_map,
 611                                         sbi->blocks_per_seg,
 612                                         cur_pos);
 613                         valid_blocks += next_pos - cur_pos;
 614                 } else
 615                         next_pos = find_next_bit_le(&raw_sit->valid_map,
 616                                         sbi->blocks_per_seg,
 617                                         cur_pos);
 618                 cur_pos = next_pos;
 619                 is_valid = !is_valid;
 620         } while (cur_pos < sbi->blocks_per_seg);
 621         BUG_ON(GET_SIT_VBLOCKS(raw_sit) != valid_blocks);
 622 #endif
 623         /* check segment usage, and check boundary of a given segment number */
 624         f2fs_bug_on(sbi, GET_SIT_VBLOCKS(raw_sit) > sbi->blocks_per_seg
 625                                         || segno > TOTAL_SEGS(sbi) - 1);
 626 }
 627
 628 static inline pgoff_t current_sit_addr(struct f2fs_sb_info *sbi,
 629                                                 unsigned int start)
 630 {
 631         struct sit_info *sit_i = SIT_I(sbi);
 632         unsigned int offset = SIT_BLOCK_OFFSET(start);
 633         block_t blk_addr = sit_i->sit_base_addr + offset;
 634
 635         check_seg_range(sbi, start);
 636
 637         /* calculate sit block address */
 638         if (f2fs_test_bit(offset, sit_i->sit_bitmap))
 639                 blk_addr += sit_i->sit_blocks;
 640
 641         return blk_addr;
 642 }
 643
 644 static inline pgoff_t next_sit_addr(struct f2fs_sb_info *sbi,
 645                                                 pgoff_t block_addr)
 646 {
 647         struct sit_info *sit_i = SIT_I(sbi);
 648         block_addr -= sit_i->sit_base_addr;
 649         if (block_addr < sit_i->sit_blocks)
 650                 block_addr += sit_i->sit_blocks;
 651         else
 652                 block_addr -= sit_i->sit_blocks;
 653
 654         return block_addr + sit_i->sit_base_addr;
 655 }
 656
 657 static inline void set_to_next_sit(struct sit_info *sit_i, unsigned int start)
 658 {
 659         unsigned int block_off = SIT_BLOCK_OFFSET(start);
 660
 661         f2fs_change_bit(block_off, sit_i->sit_bitmap);
 662 }
 663
 664 static inline unsigned long long get_mtime(struct f2fs_sb_info *sbi)
 665 {
 666         struct sit_info *sit_i = SIT_I(sbi);
 667         return sit_i->elapsed_time + CURRENT_TIME_SEC.tv_sec -
 668                                                 sit_i->mounted_time;
 669 }
 670
 671 static inline void set_summary(struct f2fs_summary *sum, nid_t nid,
 672                         unsigned int ofs_in_node, unsigned char version)
 673 {
 674         sum->nid = cpu_to_le32(nid);
 675         sum->ofs_in_node = cpu_to_le16(ofs_in_node);
 676         sum->version = version;
 677 }
 678
 679 static inline block_t start_sum_block(struct f2fs_sb_info *sbi)
 680 {
 681         return __start_cp_addr(sbi) +
 682                 le32_to_cpu(F2FS_CKPT(sbi)->cp_pack_start_sum);
 683 }
 684
 685 static inline block_t sum_blk_addr(struct f2fs_sb_info *sbi, int base, int type)
 686 {
 687         return __start_cp_addr(sbi) +
 688                 le32_to_cpu(F2FS_CKPT(sbi)->cp_pack_total_block_count)
 689                                 - (base + 1) + type;
 690 }
 691
 692 static inline bool sec_usage_check(struct f2fs_sb_info *sbi, unsigned int secno)
 693 {
 694         if (IS_CURSEC(sbi, secno) || (sbi->cur_victim_sec == secno))
 695                 return true;
 696         return false;
 697 }
 698
 699 /*
 700  * It is very important to gather dirty pages and write at once, so that we can
 701  * submit a big bio without interfering other data writes.
 702  * By default, 512 pages for directory data,
 703  * 512 pages (2MB) * 3 for three types of nodes, and
 704  * max_bio_blocks for meta are set.
 705  */
 706 static inline int nr_pages_to_skip(struct f2fs_sb_info *sbi, int type)
 707 {
 708         if (sbi->sb->s_bdi->wb.dirty_exceeded)
 709                 return 0;
 710
 711         if (type == DATA)
 712                 return sbi->blocks_per_seg;
 713         else if (type == NODE)
 714                 return 8 * sbi->blocks_per_seg;
 715         else if (type == META)
 716                 return 8 * BIO_MAX_PAGES;
 717         else
 718                 return 0;
 719 }
 720
 721 /*
 722  * When writing pages, it'd better align nr_to_write for segment size.
 723  */
 724 static inline long nr_pages_to_write(struct f2fs_sb_info *sbi, int type,
 725                                         struct writeback_control *wbc)
 726 {
 727         long nr_to_write, desired;
 728
 729         if (wbc->sync_mode != WB_SYNC_NONE)
 730                 return 0;
 731
 732         nr_to_write = wbc->nr_to_write;
 733         desired = BIO_MAX_PAGES;
 734         if (type == NODE)
 735                 desired <<= 1;
 736
 737         wbc->nr_to_write = desired;
 738         return desired - nr_to_write;
 739 }