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 /* L: Logical segment # in volume, R: Relative segment # in main area */
  22 #define GET_L2R_SEGNO(free_i, segno)    (segno - free_i->start_segno)
  23 #define GET_R2L_SEGNO(free_i, segno)    (segno + free_i->start_segno)
  24
  25 #define IS_DATASEG(t)   (t <= CURSEG_COLD_DATA)
  26 #define IS_NODESEG(t)   (t >= CURSEG_HOT_NODE)
  27
  28 #define IS_CURSEG(sbi, seg)                                             \
  29         ((seg == CURSEG_I(sbi, CURSEG_HOT_DATA)->segno) ||      \
  30          (seg == CURSEG_I(sbi, CURSEG_WARM_DATA)->segno) ||     \
  31          (seg == CURSEG_I(sbi, CURSEG_COLD_DATA)->segno) ||     \
  32          (seg == CURSEG_I(sbi, CURSEG_HOT_NODE)->segno) ||      \
  33          (seg == CURSEG_I(sbi, CURSEG_WARM_NODE)->segno) ||     \
  34          (seg == CURSEG_I(sbi, CURSEG_COLD_NODE)->segno))
  35
  36 #define IS_CURSEC(sbi, secno)                                           \
  37         ((secno == CURSEG_I(sbi, CURSEG_HOT_DATA)->segno /              \
  38           sbi->segs_per_sec) || \
  39          (secno == CURSEG_I(sbi, CURSEG_WARM_DATA)->segno /             \
  40           sbi->segs_per_sec) || \
  41          (secno == CURSEG_I(sbi, CURSEG_COLD_DATA)->segno /             \
  42           sbi->segs_per_sec) || \
  43          (secno == CURSEG_I(sbi, CURSEG_HOT_NODE)->segno /              \
  44           sbi->segs_per_sec) || \
  45          (secno == CURSEG_I(sbi, CURSEG_WARM_NODE)->segno /             \
  46           sbi->segs_per_sec) || \
  47          (secno == CURSEG_I(sbi, CURSEG_COLD_NODE)->segno /             \
  48           sbi->segs_per_sec))   \
  49
  50 #define MAIN_BLKADDR(sbi)       (SM_I(sbi)->main_blkaddr)
  51 #define SEG0_BLKADDR(sbi)       (SM_I(sbi)->seg0_blkaddr)
  52
  53 #define MAIN_SEGS(sbi)  (SM_I(sbi)->main_segments)
  54 #define MAIN_SECS(sbi)  (sbi->total_sections)
  55
  56 #define TOTAL_SEGS(sbi) (SM_I(sbi)->segment_count)
  57 #define TOTAL_BLKS(sbi) (TOTAL_SEGS(sbi) << sbi->log_blocks_per_seg)
  58
  59 #define MAX_BLKADDR(sbi)        (SEG0_BLKADDR(sbi) + TOTAL_BLKS(sbi))
  60 #define SEGMENT_SIZE(sbi)       (1ULL << (sbi->log_blocksize +          \
  61                                         sbi->log_blocks_per_seg))
  62
  63 #define START_BLOCK(sbi, segno) (SEG0_BLKADDR(sbi) +                    \
  64          (GET_R2L_SEGNO(FREE_I(sbi), segno) << sbi->log_blocks_per_seg))
  65
  66 #define NEXT_FREE_BLKADDR(sbi, curseg)                                  \
  67         (START_BLOCK(sbi, curseg->segno) + curseg->next_blkoff)
  68
  69 #define GET_SEGOFF_FROM_SEG0(sbi, blk_addr)     ((blk_addr) - SEG0_BLKADDR(sbi))
  70 #define GET_SEGNO_FROM_SEG0(sbi, blk_addr)                              \
  71         (GET_SEGOFF_FROM_SEG0(sbi, blk_addr) >> sbi->log_blocks_per_seg)
  72 #define GET_BLKOFF_FROM_SEG0(sbi, blk_addr)                             \
  73         (GET_SEGOFF_FROM_SEG0(sbi, blk_addr) & (sbi->blocks_per_seg - 1))
  74
  75 #define GET_SEGNO(sbi, blk_addr)                                        \
  76         (((blk_addr == NULL_ADDR) || (blk_addr == NEW_ADDR)) ?          \
  77         NULL_SEGNO : GET_L2R_SEGNO(FREE_I(sbi),                 \
  78                 GET_SEGNO_FROM_SEG0(sbi, blk_addr)))
  79 #define GET_SECNO(sbi, segno)                                   \
  80         ((segno) / sbi->segs_per_sec)
  81 #define GET_ZONENO_FROM_SEGNO(sbi, segno)                               \
  82         ((segno / sbi->segs_per_sec) / sbi->secs_per_zone)
  83
  84 #define GET_SUM_BLOCK(sbi, segno)                               \
  85         ((sbi->sm_info->ssa_blkaddr) + segno)
  86
  87 #define GET_SUM_TYPE(footer) ((footer)->entry_type)
  88 #define SET_SUM_TYPE(footer, type) ((footer)->entry_type = type)
  89
  90 #define SIT_ENTRY_OFFSET(sit_i, segno)                                  \
  91         (segno % sit_i->sents_per_block)
  92 #define SIT_BLOCK_OFFSET(segno)                                 \
  93         (segno / SIT_ENTRY_PER_BLOCK)
  94 #define START_SEGNO(segno)              \
  95         (SIT_BLOCK_OFFSET(segno) * SIT_ENTRY_PER_BLOCK)
  96 #define SIT_BLK_CNT(sbi)                        \
  97         ((MAIN_SEGS(sbi) + SIT_ENTRY_PER_BLOCK - 1) / SIT_ENTRY_PER_BLOCK)
  98 #define f2fs_bitmap_size(nr)                    \
  99         (BITS_TO_LONGS(nr) * sizeof(unsigned long))
 100
 101 #define SECTOR_FROM_BLOCK(blk_addr)                                     \
 102         (((sector_t)blk_addr) << F2FS_LOG_SECTORS_PER_BLOCK)
 103 #define SECTOR_TO_BLOCK(sectors)                                        \
 104         (sectors >> F2FS_LOG_SECTORS_PER_BLOCK)
 105 #define MAX_BIO_BLOCKS(sbi)                                             \
 106         ((int)min((int)max_hw_blocks(sbi), BIO_MAX_PAGES))
 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
 475         if (test_opt(sbi, LFS))
 476                 return false;
 477
 478         return free_sections(sbi) <= (node_secs + 2 * dent_secs +
 479                                                 reserved_sections(sbi) + 1);
 480 }
 481
 482 static inline bool has_not_enough_free_secs(struct f2fs_sb_info *sbi,
 483                                         int freed, int needed)
 484 {
 485         int node_secs = get_blocktype_secs(sbi, F2FS_DIRTY_NODES);
 486         int dent_secs = get_blocktype_secs(sbi, F2FS_DIRTY_DENTS);
 487
 488         node_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 + reserved_sections(sbi) + needed);
 495 }
 496
 497 static inline bool excess_prefree_segs(struct f2fs_sb_info *sbi)
 498 {
 499         return prefree_segments(sbi) > SM_I(sbi)->rec_prefree_segments;
 500 }
 501
 502 static inline int utilization(struct f2fs_sb_info *sbi)
 503 {
 504         return div_u64((u64)valid_user_blocks(sbi) * 100,
 505                                         sbi->user_block_count);
 506 }
 507
 508 /*
 509  * Sometimes f2fs may be better to drop out-of-place update policy.
 510  * And, users can control the policy through sysfs entries.
 511  * There are five policies with triggering conditions as follows.
 512  * F2FS_IPU_FORCE - all the time,
 513  * F2FS_IPU_SSR - if SSR mode is activated,
 514  * F2FS_IPU_UTIL - if FS utilization is over threashold,
 515  * F2FS_IPU_SSR_UTIL - if SSR mode is activated and FS utilization is over
 516  *                     threashold,
 517  * F2FS_IPU_FSYNC - activated in fsync path only for high performance flash
 518  *                     storages. IPU will be triggered only if the # of dirty
 519  *                     pages over min_fsync_blocks.
 520  * F2FS_IPUT_DISABLE - disable IPU. (=default option)
 521  */
 522 #define DEF_MIN_IPU_UTIL        70
 523 #define DEF_MIN_FSYNC_BLOCKS    8
 524
 525 enum {
 526         F2FS_IPU_FORCE,
 527         F2FS_IPU_SSR,
 528         F2FS_IPU_UTIL,
 529         F2FS_IPU_SSR_UTIL,
 530         F2FS_IPU_FSYNC,
 531 };
 532
 533 static inline bool need_inplace_update(struct inode *inode)
 534 {
 535         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 536         unsigned int policy = SM_I(sbi)->ipu_policy;
 537
 538         /* IPU can be done only for the user data */
 539         if (S_ISDIR(inode->i_mode) || f2fs_is_atomic_file(inode))
 540                 return false;
 541
 542         if (test_opt(sbi, LFS))
 543                 return false;
 544
 545         if (policy & (0x1 << F2FS_IPU_FORCE))
 546                 return true;
 547         if (policy & (0x1 << F2FS_IPU_SSR) && need_SSR(sbi))
 548                 return true;
 549         if (policy & (0x1 << F2FS_IPU_UTIL) &&
 550                         utilization(sbi) > SM_I(sbi)->min_ipu_util)
 551                 return true;
 552         if (policy & (0x1 << F2FS_IPU_SSR_UTIL) && need_SSR(sbi) &&
 553                         utilization(sbi) > SM_I(sbi)->min_ipu_util)
 554                 return true;
 555
 556         /* this is only set during fdatasync */
 557         if (policy & (0x1 << F2FS_IPU_FSYNC) &&
 558                         is_inode_flag_set(inode, FI_NEED_IPU))
 559                 return true;
 560
 561         return false;
 562 }
 563
 564 static inline unsigned int curseg_segno(struct f2fs_sb_info *sbi,
 565                 int type)
 566 {
 567         struct curseg_info *curseg = CURSEG_I(sbi, type);
 568         return curseg->segno;
 569 }
 570
 571 static inline unsigned char curseg_alloc_type(struct f2fs_sb_info *sbi,
 572                 int type)
 573 {
 574         struct curseg_info *curseg = CURSEG_I(sbi, type);
 575         return curseg->alloc_type;
 576 }
 577
 578 static inline unsigned short curseg_blkoff(struct f2fs_sb_info *sbi, int type)
 579 {
 580         struct curseg_info *curseg = CURSEG_I(sbi, type);
 581         return curseg->next_blkoff;
 582 }
 583
 584 static inline void check_seg_range(struct f2fs_sb_info *sbi, unsigned int segno)
 585 {
 586         f2fs_bug_on(sbi, segno > TOTAL_SEGS(sbi) - 1);
 587 }
 588
 589 static inline void verify_block_addr(struct f2fs_sb_info *sbi, block_t blk_addr)
 590 {
 591         BUG_ON(blk_addr < SEG0_BLKADDR(sbi)
 592                         || blk_addr >= MAX_BLKADDR(sbi));
 593 }
 594
 595 /*
 596  * Summary block is always treated as an invalid block
 597  */
 598 static inline void check_block_count(struct f2fs_sb_info *sbi,
 599                 int segno, struct f2fs_sit_entry *raw_sit)
 600 {
 601 #ifdef CONFIG_F2FS_CHECK_FS
 602         bool is_valid  = test_bit_le(0, raw_sit->valid_map) ? true : false;
 603         int valid_blocks = 0;
 604         int cur_pos = 0, next_pos;
 605
 606         /* check bitmap with valid block count */
 607         do {
 608                 if (is_valid) {
 609                         next_pos = find_next_zero_bit_le(&raw_sit->valid_map,
 610                                         sbi->blocks_per_seg,
 611                                         cur_pos);
 612                         valid_blocks += next_pos - cur_pos;
 613                 } else
 614                         next_pos = find_next_bit_le(&raw_sit->valid_map,
 615                                         sbi->blocks_per_seg,
 616                                         cur_pos);
 617                 cur_pos = next_pos;
 618                 is_valid = !is_valid;
 619         } while (cur_pos < sbi->blocks_per_seg);
 620         BUG_ON(GET_SIT_VBLOCKS(raw_sit) != valid_blocks);
 621 #endif
 622         /* check segment usage, and check boundary of a given segment number */
 623         f2fs_bug_on(sbi, GET_SIT_VBLOCKS(raw_sit) > sbi->blocks_per_seg
 624                                         || segno > TOTAL_SEGS(sbi) - 1);
 625 }
 626
 627 static inline pgoff_t current_sit_addr(struct f2fs_sb_info *sbi,
 628                                                 unsigned int start)
 629 {
 630         struct sit_info *sit_i = SIT_I(sbi);
 631         unsigned int offset = SIT_BLOCK_OFFSET(start);
 632         block_t blk_addr = sit_i->sit_base_addr + offset;
 633
 634         check_seg_range(sbi, start);
 635
 636         /* calculate sit block address */
 637         if (f2fs_test_bit(offset, sit_i->sit_bitmap))
 638                 blk_addr += sit_i->sit_blocks;
 639
 640         return blk_addr;
 641 }
 642
 643 static inline pgoff_t next_sit_addr(struct f2fs_sb_info *sbi,
 644                                                 pgoff_t block_addr)
 645 {
 646         struct sit_info *sit_i = SIT_I(sbi);
 647         block_addr -= sit_i->sit_base_addr;
 648         if (block_addr < sit_i->sit_blocks)
 649                 block_addr += sit_i->sit_blocks;
 650         else
 651                 block_addr -= sit_i->sit_blocks;
 652
 653         return block_addr + sit_i->sit_base_addr;
 654 }
 655
 656 static inline void set_to_next_sit(struct sit_info *sit_i, unsigned int start)
 657 {
 658         unsigned int block_off = SIT_BLOCK_OFFSET(start);
 659
 660         f2fs_change_bit(block_off, sit_i->sit_bitmap);
 661 }
 662
 663 static inline unsigned long long get_mtime(struct f2fs_sb_info *sbi)
 664 {
 665         struct sit_info *sit_i = SIT_I(sbi);
 666         return sit_i->elapsed_time + CURRENT_TIME_SEC.tv_sec -
 667                                                 sit_i->mounted_time;
 668 }
 669
 670 static inline void set_summary(struct f2fs_summary *sum, nid_t nid,
 671                         unsigned int ofs_in_node, unsigned char version)
 672 {
 673         sum->nid = cpu_to_le32(nid);
 674         sum->ofs_in_node = cpu_to_le16(ofs_in_node);
 675         sum->version = version;
 676 }
 677
 678 static inline block_t start_sum_block(struct f2fs_sb_info *sbi)
 679 {
 680         return __start_cp_addr(sbi) +
 681                 le32_to_cpu(F2FS_CKPT(sbi)->cp_pack_start_sum);
 682 }
 683
 684 static inline block_t sum_blk_addr(struct f2fs_sb_info *sbi, int base, int type)
 685 {
 686         return __start_cp_addr(sbi) +
 687                 le32_to_cpu(F2FS_CKPT(sbi)->cp_pack_total_block_count)
 688                                 - (base + 1) + type;
 689 }
 690
 691 static inline bool sec_usage_check(struct f2fs_sb_info *sbi, unsigned int secno)
 692 {
 693         if (IS_CURSEC(sbi, secno) || (sbi->cur_victim_sec == secno))
 694                 return true;
 695         return false;
 696 }
 697
 698 static inline unsigned int max_hw_blocks(struct f2fs_sb_info *sbi)
 699 {
 700         struct block_device *bdev = sbi->sb->s_bdev;
 701         struct request_queue *q = bdev_get_queue(bdev);
 702         return SECTOR_TO_BLOCK(queue_max_sectors(q));
 703 }
 704
 705 /*
 706  * It is very important to gather dirty pages and write at once, so that we can
 707  * submit a big bio without interfering other data writes.
 708  * By default, 512 pages for directory data,
 709  * 512 pages (2MB) * 3 for three types of nodes, and
 710  * max_bio_blocks for meta are set.
 711  */
 712 static inline int nr_pages_to_skip(struct f2fs_sb_info *sbi, int type)
 713 {
 714         if (sbi->sb->s_bdi->wb.dirty_exceeded)
 715                 return 0;
 716
 717         if (type == DATA)
 718                 return sbi->blocks_per_seg;
 719         else if (type == NODE)
 720                 return 8 * sbi->blocks_per_seg;
 721         else if (type == META)
 722                 return 8 * MAX_BIO_BLOCKS(sbi);
 723         else
 724                 return 0;
 725 }
 726
 727 /*
 728  * When writing pages, it'd better align nr_to_write for segment size.
 729  */
 730 static inline long nr_pages_to_write(struct f2fs_sb_info *sbi, int type,
 731                                         struct writeback_control *wbc)
 732 {
 733         long nr_to_write, desired;
 734
 735         if (wbc->sync_mode != WB_SYNC_NONE)
 736                 return 0;
 737
 738         nr_to_write = wbc->nr_to_write;
 739
 740         if (type == NODE)
 741                 desired = 2 * max_hw_blocks(sbi);
 742         else
 743                 desired = MAX_BIO_BLOCKS(sbi);
 744
 745         wbc->nr_to_write = desired;
 746         return desired - nr_to_write;
 747 }