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
  20 #define F2FS_MIN_SEGMENTS       9 /* SB + 2 (CP + SIT + NAT) + SSA + MAIN */
  21
  22 /* L: Logical segment # in volume, R: Relative segment # in main area */
  23 #define GET_L2R_SEGNO(free_i, segno)    (segno - free_i->start_segno)
  24 #define GET_R2L_SEGNO(free_i, segno)    (segno + free_i->start_segno)
  25
  26 #define IS_DATASEG(t)   (t <= CURSEG_COLD_DATA)
  27 #define IS_NODESEG(t)   (t >= CURSEG_HOT_NODE)
  28
  29 #define IS_CURSEG(sbi, seg)                                             \
  30         ((seg == CURSEG_I(sbi, CURSEG_HOT_DATA)->segno) ||      \
  31          (seg == CURSEG_I(sbi, CURSEG_WARM_DATA)->segno) ||     \
  32          (seg == CURSEG_I(sbi, CURSEG_COLD_DATA)->segno) ||     \
  33          (seg == CURSEG_I(sbi, CURSEG_HOT_NODE)->segno) ||      \
  34          (seg == CURSEG_I(sbi, CURSEG_WARM_NODE)->segno) ||     \
  35          (seg == CURSEG_I(sbi, CURSEG_COLD_NODE)->segno))
  36
  37 #define IS_CURSEC(sbi, secno)                                           \
  38         ((secno == CURSEG_I(sbi, CURSEG_HOT_DATA)->segno /              \
  39           sbi->segs_per_sec) || \
  40          (secno == CURSEG_I(sbi, CURSEG_WARM_DATA)->segno /             \
  41           sbi->segs_per_sec) || \
  42          (secno == CURSEG_I(sbi, CURSEG_COLD_DATA)->segno /             \
  43           sbi->segs_per_sec) || \
  44          (secno == CURSEG_I(sbi, CURSEG_HOT_NODE)->segno /              \
  45           sbi->segs_per_sec) || \
  46          (secno == CURSEG_I(sbi, CURSEG_WARM_NODE)->segno /             \
  47           sbi->segs_per_sec) || \
  48          (secno == CURSEG_I(sbi, CURSEG_COLD_NODE)->segno /             \
  49           sbi->segs_per_sec))   \
  50
  51 #define MAIN_BLKADDR(sbi)                                               \
  52         (SM_I(sbi) ? SM_I(sbi)->main_blkaddr :                          \
  53                 le32_to_cpu(F2FS_RAW_SUPER(sbi)->main_blkaddr))
  54 #define SEG0_BLKADDR(sbi)                                               \
  55         (SM_I(sbi) ? SM_I(sbi)->seg0_blkaddr :                          \
  56                 le32_to_cpu(F2FS_RAW_SUPER(sbi)->segment0_blkaddr))
  57
  58 #define MAIN_SEGS(sbi)  (SM_I(sbi)->main_segments)
  59 #define MAIN_SECS(sbi)  (sbi->total_sections)
  60
  61 #define TOTAL_SEGS(sbi)                                                 \
  62         (SM_I(sbi) ? SM_I(sbi)->segment_count :                                 \
  63                 le32_to_cpu(F2FS_RAW_SUPER(sbi)->segment_count))
  64 #define TOTAL_BLKS(sbi) (TOTAL_SEGS(sbi) << sbi->log_blocks_per_seg)
  65
  66 #define MAX_BLKADDR(sbi)        (SEG0_BLKADDR(sbi) + TOTAL_BLKS(sbi))
  67 #define SEGMENT_SIZE(sbi)       (1ULL << (sbi->log_blocksize +          \
  68                                         sbi->log_blocks_per_seg))
  69
  70 #define START_BLOCK(sbi, segno) (SEG0_BLKADDR(sbi) +                    \
  71          (GET_R2L_SEGNO(FREE_I(sbi), segno) << sbi->log_blocks_per_seg))
  72
  73 #define NEXT_FREE_BLKADDR(sbi, curseg)                                  \
  74         (START_BLOCK(sbi, curseg->segno) + curseg->next_blkoff)
  75
  76 #define GET_SEGOFF_FROM_SEG0(sbi, blk_addr)     ((blk_addr) - SEG0_BLKADDR(sbi))
  77 #define GET_SEGNO_FROM_SEG0(sbi, blk_addr)                              \
  78         (GET_SEGOFF_FROM_SEG0(sbi, blk_addr) >> sbi->log_blocks_per_seg)
  79 #define GET_BLKOFF_FROM_SEG0(sbi, blk_addr)                             \
  80         (GET_SEGOFF_FROM_SEG0(sbi, blk_addr) & (sbi->blocks_per_seg - 1))
  81
  82 #define GET_SEGNO(sbi, blk_addr)                                        \
  83         ((!is_valid_data_blkaddr(sbi, blk_addr)) ?                      \
  84         NULL_SEGNO : GET_L2R_SEGNO(FREE_I(sbi),                 \
  85                 GET_SEGNO_FROM_SEG0(sbi, blk_addr)))
  86 #define GET_SECNO(sbi, segno)                                   \
  87         ((segno) / sbi->segs_per_sec)
  88 #define GET_ZONENO_FROM_SEGNO(sbi, segno)                               \
  89         ((segno / sbi->segs_per_sec) / sbi->secs_per_zone)
  90
  91 #define GET_SUM_BLOCK(sbi, segno)                               \
  92         ((sbi->sm_info->ssa_blkaddr) + segno)
  93
  94 #define GET_SUM_TYPE(footer) ((footer)->entry_type)
  95 #define SET_SUM_TYPE(footer, type) ((footer)->entry_type = type)
  96
  97 #define SIT_ENTRY_OFFSET(sit_i, segno)                                  \
  98         (segno % sit_i->sents_per_block)
  99 #define SIT_BLOCK_OFFSET(segno)                                 \
 100         (segno / SIT_ENTRY_PER_BLOCK)
 101 #define START_SEGNO(segno)              \
 102         (SIT_BLOCK_OFFSET(segno) * SIT_ENTRY_PER_BLOCK)
 103 #define SIT_BLK_CNT(sbi)                        \
 104         ((MAIN_SEGS(sbi) + SIT_ENTRY_PER_BLOCK - 1) / SIT_ENTRY_PER_BLOCK)
 105 #define f2fs_bitmap_size(nr)                    \
 106         (BITS_TO_LONGS(nr) * sizeof(unsigned long))
 107
 108 #define SECTOR_FROM_BLOCK(blk_addr)                                     \
 109         (((sector_t)blk_addr) << F2FS_LOG_SECTORS_PER_BLOCK)
 110 #define SECTOR_TO_BLOCK(sectors)                                        \
 111         (sectors >> F2FS_LOG_SECTORS_PER_BLOCK)
 112 #define MAX_BIO_BLOCKS(sbi)                                             \
 113         ((int)min((int)max_hw_blocks(sbi), BIO_MAX_PAGES))
 114
 115 /*
 116  * indicate a block allocation direction: RIGHT and LEFT.
 117  * RIGHT means allocating new sections towards the end of volume.
 118  * LEFT means the opposite direction.
 119  */
 120 enum {
 121         ALLOC_RIGHT = 0,
 122         ALLOC_LEFT
 123 };
 124
 125 /*
 126  * In the victim_sel_policy->alloc_mode, there are two block allocation modes.
 127  * LFS writes data sequentially with cleaning operations.
 128  * SSR (Slack Space Recycle) reuses obsolete space without cleaning operations.
 129  */
 130 enum {
 131         LFS = 0,
 132         SSR
 133 };
 134
 135 /*
 136  * In the victim_sel_policy->gc_mode, there are two gc, aka cleaning, modes.
 137  * GC_CB is based on cost-benefit algorithm.
 138  * GC_GREEDY is based on greedy algorithm.
 139  */
 140 enum {
 141         GC_CB = 0,
 142         GC_GREEDY
 143 };
 144
 145 /*
 146  * BG_GC means the background cleaning job.
 147  * FG_GC means the on-demand cleaning job.
 148  * FORCE_FG_GC means on-demand cleaning job in background.
 149  */
 150 enum {
 151         BG_GC = 0,
 152         FG_GC,
 153         FORCE_FG_GC,
 154 };
 155
 156 /* for a function parameter to select a victim segment */
 157 struct victim_sel_policy {
 158         int alloc_mode;                 /* LFS or SSR */
 159         int gc_mode;                    /* GC_CB or GC_GREEDY */
 160         unsigned long *dirty_segmap;    /* dirty segment bitmap */
 161         unsigned int max_search;        /* maximum # of segments to search */
 162         unsigned int offset;            /* last scanned bitmap offset */
 163         unsigned int ofs_unit;          /* bitmap search unit */
 164         unsigned int min_cost;          /* minimum cost */
 165         unsigned int min_segno;         /* segment # having min. cost */
 166 };
 167
 168 struct seg_entry {
 169         unsigned short valid_blocks;    /* # of valid blocks */
 170         unsigned char *cur_valid_map;   /* validity bitmap of blocks */
 171         /*
 172          * # of valid blocks and the validity bitmap stored in the the last
 173          * checkpoint pack. This information is used by the SSR mode.
 174          */
 175         unsigned short ckpt_valid_blocks;
 176         unsigned char *ckpt_valid_map;
 177         unsigned char *discard_map;
 178         unsigned char type;             /* segment type like CURSEG_XXX_TYPE */
 179         unsigned long long mtime;       /* modification time of the segment */
 180 };
 181
 182 struct sec_entry {
 183         unsigned int valid_blocks;      /* # of valid blocks in a section */
 184 };
 185
 186 struct segment_allocation {
 187         void (*allocate_segment)(struct f2fs_sb_info *, int, bool);
 188 };
 189
 190 /*
 191  * this value is set in page as a private data which indicate that
 192  * the page is atomically written, and it is in inmem_pages list.
 193  */
 194 #define ATOMIC_WRITTEN_PAGE             0x0000ffff
 195
 196 #define IS_ATOMIC_WRITTEN_PAGE(page)                    \
 197                 (page_private(page) == (unsigned long)ATOMIC_WRITTEN_PAGE)
 198
 199 struct inmem_pages {
 200         struct list_head list;
 201         struct page *page;
 202 };
 203
 204 struct sit_info {
 205         const struct segment_allocation *s_ops;
 206
 207         block_t sit_base_addr;          /* start block address of SIT area */
 208         block_t sit_blocks;             /* # of blocks used by SIT area */
 209         block_t written_valid_blocks;   /* # of valid blocks in main area */
 210         char *sit_bitmap;               /* SIT bitmap pointer */
 211         unsigned int bitmap_size;       /* SIT bitmap size */
 212
 213         unsigned long *tmp_map;                 /* bitmap for temporal use */
 214         unsigned long *dirty_sentries_bitmap;   /* bitmap for dirty sentries */
 215         unsigned int dirty_sentries;            /* # of dirty sentries */
 216         unsigned int sents_per_block;           /* # of SIT entries per block */
 217         struct mutex sentry_lock;               /* to protect SIT cache */
 218         struct seg_entry *sentries;             /* SIT segment-level cache */
 219         struct sec_entry *sec_entries;          /* SIT section-level cache */
 220
 221         /* for cost-benefit algorithm in cleaning procedure */
 222         unsigned long long elapsed_time;        /* elapsed time after mount */
 223         unsigned long long mounted_time;        /* mount time */
 224         unsigned long long min_mtime;           /* min. modification time */
 225         unsigned long long max_mtime;           /* max. modification time */
 226 };
 227
 228 struct free_segmap_info {
 229         unsigned int start_segno;       /* start segment number logically */
 230         unsigned int free_segments;     /* # of free segments */
 231         unsigned int free_sections;     /* # of free sections */
 232         spinlock_t segmap_lock;         /* free segmap lock */
 233         unsigned long *free_segmap;     /* free segment bitmap */
 234         unsigned long *free_secmap;     /* free section bitmap */
 235 };
 236
 237 /* Notice: The order of dirty type is same with CURSEG_XXX in f2fs.h */
 238 enum dirty_type {
 239         DIRTY_HOT_DATA,         /* dirty segments assigned as hot data logs */
 240         DIRTY_WARM_DATA,        /* dirty segments assigned as warm data logs */
 241         DIRTY_COLD_DATA,        /* dirty segments assigned as cold data logs */
 242         DIRTY_HOT_NODE,         /* dirty segments assigned as hot node logs */
 243         DIRTY_WARM_NODE,        /* dirty segments assigned as warm node logs */
 244         DIRTY_COLD_NODE,        /* dirty segments assigned as cold node logs */
 245         DIRTY,                  /* to count # of dirty segments */
 246         PRE,                    /* to count # of entirely obsolete segments */
 247         NR_DIRTY_TYPE
 248 };
 249
 250 struct dirty_seglist_info {
 251         const struct victim_selection *v_ops;   /* victim selction operation */
 252         unsigned long *dirty_segmap[NR_DIRTY_TYPE];
 253         struct mutex seglist_lock;              /* lock for segment bitmaps */
 254         int nr_dirty[NR_DIRTY_TYPE];            /* # of dirty segments */
 255         unsigned long *victim_secmap;           /* background GC victims */
 256 };
 257
 258 /* victim selection function for cleaning and SSR */
 259 struct victim_selection {
 260         int (*get_victim)(struct f2fs_sb_info *, unsigned int *,
 261                                                         int, int, char);
 262 };
 263
 264 /* for active log information */
 265 struct curseg_info {
 266         struct mutex curseg_mutex;              /* lock for consistency */
 267         struct f2fs_summary_block *sum_blk;     /* cached summary block */
 268         unsigned char alloc_type;               /* current allocation type */
 269         unsigned int segno;                     /* current segment number */
 270         unsigned short next_blkoff;             /* next block offset to write */
 271         unsigned int zone;                      /* current zone number */
 272         unsigned int next_segno;                /* preallocated segment */
 273 };
 274
 275 struct sit_entry_set {
 276         struct list_head set_list;      /* link with all sit sets */
 277         unsigned int start_segno;       /* start segno of sits in set */
 278         unsigned int entry_cnt;         /* the # of sit entries in set */
 279 };
 280
 281 /*
 282  * inline functions
 283  */
 284 static inline struct curseg_info *CURSEG_I(struct f2fs_sb_info *sbi, int type)
 285 {
 286         return (struct curseg_info *)(SM_I(sbi)->curseg_array + type);
 287 }
 288
 289 static inline struct seg_entry *get_seg_entry(struct f2fs_sb_info *sbi,
 290                                                 unsigned int segno)
 291 {
 292         struct sit_info *sit_i = SIT_I(sbi);
 293         return &sit_i->sentries[segno];
 294 }
 295
 296 static inline struct sec_entry *get_sec_entry(struct f2fs_sb_info *sbi,
 297                                                 unsigned int segno)
 298 {
 299         struct sit_info *sit_i = SIT_I(sbi);
 300         return &sit_i->sec_entries[GET_SECNO(sbi, segno)];
 301 }
 302
 303 static inline unsigned int get_valid_blocks(struct f2fs_sb_info *sbi,
 304                                 unsigned int segno, int section)
 305 {
 306         /*
 307          * In order to get # of valid blocks in a section instantly from many
 308          * segments, f2fs manages two counting structures separately.
 309          */
 310         if (section > 1)
 311                 return get_sec_entry(sbi, segno)->valid_blocks;
 312         else
 313                 return get_seg_entry(sbi, segno)->valid_blocks;
 314 }
 315
 316 static inline void seg_info_from_raw_sit(struct seg_entry *se,
 317                                         struct f2fs_sit_entry *rs)
 318 {
 319         se->valid_blocks = GET_SIT_VBLOCKS(rs);
 320         se->ckpt_valid_blocks = GET_SIT_VBLOCKS(rs);
 321         memcpy(se->cur_valid_map, rs->valid_map, SIT_VBLOCK_MAP_SIZE);
 322         memcpy(se->ckpt_valid_map, rs->valid_map, SIT_VBLOCK_MAP_SIZE);
 323         se->type = GET_SIT_TYPE(rs);
 324         se->mtime = le64_to_cpu(rs->mtime);
 325 }
 326
 327 static inline void seg_info_to_raw_sit(struct seg_entry *se,
 328                                         struct f2fs_sit_entry *rs)
 329 {
 330         unsigned short raw_vblocks = (se->type << SIT_VBLOCKS_SHIFT) |
 331                                         se->valid_blocks;
 332         rs->vblocks = cpu_to_le16(raw_vblocks);
 333         memcpy(rs->valid_map, se->cur_valid_map, SIT_VBLOCK_MAP_SIZE);
 334         memcpy(se->ckpt_valid_map, rs->valid_map, SIT_VBLOCK_MAP_SIZE);
 335         se->ckpt_valid_blocks = se->valid_blocks;
 336         rs->mtime = cpu_to_le64(se->mtime);
 337 }
 338
 339 static inline unsigned int find_next_inuse(struct free_segmap_info *free_i,
 340                 unsigned int max, unsigned int segno)
 341 {
 342         unsigned int ret;
 343         spin_lock(&free_i->segmap_lock);
 344         ret = find_next_bit(free_i->free_segmap, max, segno);
 345         spin_unlock(&free_i->segmap_lock);
 346         return ret;
 347 }
 348
 349 static inline void __set_free(struct f2fs_sb_info *sbi, unsigned int segno)
 350 {
 351         struct free_segmap_info *free_i = FREE_I(sbi);
 352         unsigned int secno = segno / sbi->segs_per_sec;
 353         unsigned int start_segno = secno * sbi->segs_per_sec;
 354         unsigned int next;
 355
 356         spin_lock(&free_i->segmap_lock);
 357         clear_bit(segno, free_i->free_segmap);
 358         free_i->free_segments++;
 359
 360         next = find_next_bit(free_i->free_segmap,
 361                         start_segno + sbi->segs_per_sec, start_segno);
 362         if (next >= start_segno + sbi->segs_per_sec) {
 363                 clear_bit(secno, free_i->free_secmap);
 364                 free_i->free_sections++;
 365         }
 366         spin_unlock(&free_i->segmap_lock);
 367 }
 368
 369 static inline void __set_inuse(struct f2fs_sb_info *sbi,
 370                 unsigned int segno)
 371 {
 372         struct free_segmap_info *free_i = FREE_I(sbi);
 373         unsigned int secno = segno / sbi->segs_per_sec;
 374         set_bit(segno, free_i->free_segmap);
 375         free_i->free_segments--;
 376         if (!test_and_set_bit(secno, free_i->free_secmap))
 377                 free_i->free_sections--;
 378 }
 379
 380 static inline void __set_test_and_free(struct f2fs_sb_info *sbi,
 381                 unsigned int segno)
 382 {
 383         struct free_segmap_info *free_i = FREE_I(sbi);
 384         unsigned int secno = segno / sbi->segs_per_sec;
 385         unsigned int start_segno = secno * sbi->segs_per_sec;
 386         unsigned int next;
 387
 388         spin_lock(&free_i->segmap_lock);
 389         if (test_and_clear_bit(segno, free_i->free_segmap)) {
 390                 free_i->free_segments++;
 391
 392                 if (IS_CURSEC(sbi, secno))
 393                         goto skip_free;
 394                 next = find_next_bit(free_i->free_segmap,
 395                                 start_segno + sbi->segs_per_sec, start_segno);
 396                 if (next >= start_segno + sbi->segs_per_sec) {
 397                         if (test_and_clear_bit(secno, free_i->free_secmap))
 398                                 free_i->free_sections++;
 399                 }
 400         }
 401 skip_free:
 402         spin_unlock(&free_i->segmap_lock);
 403 }
 404
 405 static inline void __set_test_and_inuse(struct f2fs_sb_info *sbi,
 406                 unsigned int segno)
 407 {
 408         struct free_segmap_info *free_i = FREE_I(sbi);
 409         unsigned int secno = segno / sbi->segs_per_sec;
 410         spin_lock(&free_i->segmap_lock);
 411         if (!test_and_set_bit(segno, free_i->free_segmap)) {
 412                 free_i->free_segments--;
 413                 if (!test_and_set_bit(secno, free_i->free_secmap))
 414                         free_i->free_sections--;
 415         }
 416         spin_unlock(&free_i->segmap_lock);
 417 }
 418
 419 static inline void get_sit_bitmap(struct f2fs_sb_info *sbi,
 420                 void *dst_addr)
 421 {
 422         struct sit_info *sit_i = SIT_I(sbi);
 423         memcpy(dst_addr, sit_i->sit_bitmap, sit_i->bitmap_size);
 424 }
 425
 426 static inline block_t written_block_count(struct f2fs_sb_info *sbi)
 427 {
 428         return SIT_I(sbi)->written_valid_blocks;
 429 }
 430
 431 static inline unsigned int free_segments(struct f2fs_sb_info *sbi)
 432 {
 433         return FREE_I(sbi)->free_segments;
 434 }
 435
 436 static inline int reserved_segments(struct f2fs_sb_info *sbi)
 437 {
 438         return SM_I(sbi)->reserved_segments;
 439 }
 440
 441 static inline unsigned int free_sections(struct f2fs_sb_info *sbi)
 442 {
 443         return FREE_I(sbi)->free_sections;
 444 }
 445
 446 static inline unsigned int prefree_segments(struct f2fs_sb_info *sbi)
 447 {
 448         return DIRTY_I(sbi)->nr_dirty[PRE];
 449 }
 450
 451 static inline unsigned int dirty_segments(struct f2fs_sb_info *sbi)
 452 {
 453         return DIRTY_I(sbi)->nr_dirty[DIRTY_HOT_DATA] +
 454                 DIRTY_I(sbi)->nr_dirty[DIRTY_WARM_DATA] +
 455                 DIRTY_I(sbi)->nr_dirty[DIRTY_COLD_DATA] +
 456                 DIRTY_I(sbi)->nr_dirty[DIRTY_HOT_NODE] +
 457                 DIRTY_I(sbi)->nr_dirty[DIRTY_WARM_NODE] +
 458                 DIRTY_I(sbi)->nr_dirty[DIRTY_COLD_NODE];
 459 }
 460
 461 static inline int overprovision_segments(struct f2fs_sb_info *sbi)
 462 {
 463         return SM_I(sbi)->ovp_segments;
 464 }
 465
 466 static inline int overprovision_sections(struct f2fs_sb_info *sbi)
 467 {
 468         return ((unsigned int) overprovision_segments(sbi)) / sbi->segs_per_sec;
 469 }
 470
 471 static inline int reserved_sections(struct f2fs_sb_info *sbi)
 472 {
 473         return ((unsigned int) reserved_segments(sbi)) / sbi->segs_per_sec;
 474 }
 475
 476 static inline bool need_SSR(struct f2fs_sb_info *sbi)
 477 {
 478         int node_secs = get_blocktype_secs(sbi, F2FS_DIRTY_NODES);
 479         int dent_secs = get_blocktype_secs(sbi, F2FS_DIRTY_DENTS);
 480         return free_sections(sbi) <= (node_secs + 2 * dent_secs +
 481                                                 reserved_sections(sbi) + 1);
 482 }
 483
 484 static inline bool has_not_enough_free_secs(struct f2fs_sb_info *sbi, int freed)
 485 {
 486         int node_secs = get_blocktype_secs(sbi, F2FS_DIRTY_NODES);
 487         int dent_secs = get_blocktype_secs(sbi, F2FS_DIRTY_DENTS);
 488
 489         if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
 490                 return false;
 491
 492         return (free_sections(sbi) + freed) <= (node_secs + 2 * dent_secs +
 493                                                 reserved_sections(sbi));
 494 }
 495
 496 static inline bool excess_prefree_segs(struct f2fs_sb_info *sbi)
 497 {
 498         return prefree_segments(sbi) > SM_I(sbi)->rec_prefree_segments;
 499 }
 500
 501 static inline int utilization(struct f2fs_sb_info *sbi)
 502 {
 503         return div_u64((u64)valid_user_blocks(sbi) * 100,
 504                                         sbi->user_block_count);
 505 }
 506
 507 /*
 508  * Sometimes f2fs may be better to drop out-of-place update policy.
 509  * And, users can control the policy through sysfs entries.
 510  * There are five policies with triggering conditions as follows.
 511  * F2FS_IPU_FORCE - all the time,
 512  * F2FS_IPU_SSR - if SSR mode is activated,
 513  * F2FS_IPU_UTIL - if FS utilization is over threashold,
 514  * F2FS_IPU_SSR_UTIL - if SSR mode is activated and FS utilization is over
 515  *                     threashold,
 516  * F2FS_IPU_FSYNC - activated in fsync path only for high performance flash
 517  *                     storages. IPU will be triggered only if the # of dirty
 518  *                     pages over min_fsync_blocks.
 519  * F2FS_IPUT_DISABLE - disable IPU. (=default option)
 520  */
 521 #define DEF_MIN_IPU_UTIL        70
 522 #define DEF_MIN_FSYNC_BLOCKS    8
 523
 524 enum {
 525         F2FS_IPU_FORCE,
 526         F2FS_IPU_SSR,
 527         F2FS_IPU_UTIL,
 528         F2FS_IPU_SSR_UTIL,
 529         F2FS_IPU_FSYNC,
 530 };
 531
 532 static inline bool need_inplace_update(struct inode *inode)
 533 {
 534         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 535         unsigned int policy = SM_I(sbi)->ipu_policy;
 536
 537         /* IPU can be done only for the user data */
 538         if (S_ISDIR(inode->i_mode) || f2fs_is_atomic_file(inode))
 539                 return false;
 540
 541         if (policy & (0x1 << F2FS_IPU_FORCE))
 542                 return true;
 543         if (policy & (0x1 << F2FS_IPU_SSR) && need_SSR(sbi))
 544                 return true;
 545         if (policy & (0x1 << F2FS_IPU_UTIL) &&
 546                         utilization(sbi) > SM_I(sbi)->min_ipu_util)
 547                 return true;
 548         if (policy & (0x1 << F2FS_IPU_SSR_UTIL) && need_SSR(sbi) &&
 549                         utilization(sbi) > SM_I(sbi)->min_ipu_util)
 550                 return true;
 551
 552         /* this is only set during fdatasync */
 553         if (policy & (0x1 << F2FS_IPU_FSYNC) &&
 554                         is_inode_flag_set(F2FS_I(inode), FI_NEED_IPU))
 555                 return true;
 556
 557         return false;
 558 }
 559
 560 static inline unsigned int curseg_segno(struct f2fs_sb_info *sbi,
 561                 int type)
 562 {
 563         struct curseg_info *curseg = CURSEG_I(sbi, type);
 564         return curseg->segno;
 565 }
 566
 567 static inline unsigned char curseg_alloc_type(struct f2fs_sb_info *sbi,
 568                 int type)
 569 {
 570         struct curseg_info *curseg = CURSEG_I(sbi, type);
 571         return curseg->alloc_type;
 572 }
 573
 574 static inline unsigned short curseg_blkoff(struct f2fs_sb_info *sbi, int type)
 575 {
 576         struct curseg_info *curseg = CURSEG_I(sbi, type);
 577         return curseg->next_blkoff;
 578 }
 579
 580 static inline void check_seg_range(struct f2fs_sb_info *sbi, unsigned int segno)
 581 {
 582         f2fs_bug_on(sbi, segno > TOTAL_SEGS(sbi) - 1);
 583 }
 584
 585 static inline void verify_block_addr(struct f2fs_io_info *fio, block_t blk_addr)
 586 {
 587         struct f2fs_sb_info *sbi = fio->sbi;
 588
 589         if (__is_meta_io(fio))
 590                 verify_blkaddr(sbi, blk_addr, META_GENERIC);
 591         else
 592                 verify_blkaddr(sbi, blk_addr, DATA_GENERIC);
 593 }
 594
 595 /*
 596  * Summary block is always treated as an invalid block
 597  */
 598 static inline int check_block_count(struct f2fs_sb_info *sbi,
 599                 int segno, struct f2fs_sit_entry *raw_sit)
 600 {
 601         bool is_valid  = test_bit_le(0, raw_sit->valid_map) ? true : false;
 602         int valid_blocks = 0;
 603         int cur_pos = 0, next_pos;
 604
 605         /* check bitmap with valid block count */
 606         do {
 607                 if (is_valid) {
 608                         next_pos = find_next_zero_bit_le(&raw_sit->valid_map,
 609                                         sbi->blocks_per_seg,
 610                                         cur_pos);
 611                         valid_blocks += next_pos - cur_pos;
 612                 } else
 613                         next_pos = find_next_bit_le(&raw_sit->valid_map,
 614                                         sbi->blocks_per_seg,
 615                                         cur_pos);
 616                 cur_pos = next_pos;
 617                 is_valid = !is_valid;
 618         } while (cur_pos < sbi->blocks_per_seg);
 619
 620         if (unlikely(GET_SIT_VBLOCKS(raw_sit) != valid_blocks)) {
 621                 f2fs_msg(sbi->sb, KERN_ERR,
 622                                 "Mismatch valid blocks %d vs. %d",
 623                                         GET_SIT_VBLOCKS(raw_sit), valid_blocks);
 624                 set_sbi_flag(sbi, SBI_NEED_FSCK);
 625                 return -EINVAL;
 626         }
 627
 628         /* check segment usage, and check boundary of a given segment number */
 629         if (unlikely(GET_SIT_VBLOCKS(raw_sit) > sbi->blocks_per_seg
 630                                         || segno > TOTAL_SEGS(sbi) - 1)) {
 631                 f2fs_msg(sbi->sb, KERN_ERR,
 632                                 "Wrong valid blocks %d or segno %u",
 633                                         GET_SIT_VBLOCKS(raw_sit), segno);
 634                 set_sbi_flag(sbi, SBI_NEED_FSCK);
 635                 return -EINVAL;
 636         }
 637         return 0;
 638 }
 639
 640 static inline pgoff_t current_sit_addr(struct f2fs_sb_info *sbi,
 641                                                 unsigned int start)
 642 {
 643         struct sit_info *sit_i = SIT_I(sbi);
 644         unsigned int offset = SIT_BLOCK_OFFSET(start);
 645         block_t blk_addr = sit_i->sit_base_addr + offset;
 646
 647         check_seg_range(sbi, start);
 648
 649         /* calculate sit block address */
 650         if (f2fs_test_bit(offset, sit_i->sit_bitmap))
 651                 blk_addr += sit_i->sit_blocks;
 652
 653         return blk_addr;
 654 }
 655
 656 static inline pgoff_t next_sit_addr(struct f2fs_sb_info *sbi,
 657                                                 pgoff_t block_addr)
 658 {
 659         struct sit_info *sit_i = SIT_I(sbi);
 660         block_addr -= sit_i->sit_base_addr;
 661         if (block_addr < sit_i->sit_blocks)
 662                 block_addr += sit_i->sit_blocks;
 663         else
 664                 block_addr -= sit_i->sit_blocks;
 665
 666         return block_addr + sit_i->sit_base_addr;
 667 }
 668
 669 static inline void set_to_next_sit(struct sit_info *sit_i, unsigned int start)
 670 {
 671         unsigned int block_off = SIT_BLOCK_OFFSET(start);
 672
 673         f2fs_change_bit(block_off, sit_i->sit_bitmap);
 674 }
 675
 676 static inline unsigned long long get_mtime(struct f2fs_sb_info *sbi)
 677 {
 678         struct sit_info *sit_i = SIT_I(sbi);
 679         return sit_i->elapsed_time + CURRENT_TIME_SEC.tv_sec -
 680                                                 sit_i->mounted_time;
 681 }
 682
 683 static inline void set_summary(struct f2fs_summary *sum, nid_t nid,
 684                         unsigned int ofs_in_node, unsigned char version)
 685 {
 686         sum->nid = cpu_to_le32(nid);
 687         sum->ofs_in_node = cpu_to_le16(ofs_in_node);
 688         sum->version = version;
 689 }
 690
 691 static inline block_t start_sum_block(struct f2fs_sb_info *sbi)
 692 {
 693         return __start_cp_addr(sbi) +
 694                 le32_to_cpu(F2FS_CKPT(sbi)->cp_pack_start_sum);
 695 }
 696
 697 static inline block_t sum_blk_addr(struct f2fs_sb_info *sbi, int base, int type)
 698 {
 699         return __start_cp_addr(sbi) +
 700                 le32_to_cpu(F2FS_CKPT(sbi)->cp_pack_total_block_count)
 701                                 - (base + 1) + type;
 702 }
 703
 704 static inline bool sec_usage_check(struct f2fs_sb_info *sbi, unsigned int secno)
 705 {
 706         if (IS_CURSEC(sbi, secno) || (sbi->cur_victim_sec == secno))
 707                 return true;
 708         return false;
 709 }
 710
 711 static inline unsigned int max_hw_blocks(struct f2fs_sb_info *sbi)
 712 {
 713         struct block_device *bdev = sbi->sb->s_bdev;
 714         struct request_queue *q = bdev_get_queue(bdev);
 715         return SECTOR_TO_BLOCK(queue_max_sectors(q));
 716 }
 717
 718 /*
 719  * It is very important to gather dirty pages and write at once, so that we can
 720  * submit a big bio without interfering other data writes.
 721  * By default, 512 pages for directory data,
 722  * 512 pages (2MB) * 3 for three types of nodes, and
 723  * max_bio_blocks for meta are set.
 724  */
 725 static inline int nr_pages_to_skip(struct f2fs_sb_info *sbi, int type)
 726 {
 727         if (sbi->sb->s_bdi->wb.dirty_exceeded)
 728                 return 0;
 729
 730         if (type == DATA)
 731                 return sbi->blocks_per_seg;
 732         else if (type == NODE)
 733                 return 3 * sbi->blocks_per_seg;
 734         else if (type == META)
 735                 return MAX_BIO_BLOCKS(sbi);
 736         else
 737                 return 0;
 738 }
 739
 740 /*
 741  * When writing pages, it'd better align nr_to_write for segment size.
 742  */
 743 static inline long nr_pages_to_write(struct f2fs_sb_info *sbi, int type,
 744                                         struct writeback_control *wbc)
 745 {
 746         long nr_to_write, desired;
 747
 748         if (wbc->sync_mode != WB_SYNC_NONE)
 749                 return 0;
 750
 751         nr_to_write = wbc->nr_to_write;
 752
 753         if (type == DATA)
 754                 desired = 4096;
 755         else if (type == NODE)
 756                 desired = 3 * max_hw_blocks(sbi);
 757         else
 758                 desired = MAX_BIO_BLOCKS(sbi);
 759
 760         wbc->nr_to_write = desired;
 761         return desired - nr_to_write;
 762 }