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 /* L: Logical segment # in volume, R: Relative segment # in main area */
  21 #define GET_L2R_SEGNO(free_i, segno)    (segno - free_i->start_segno)
  22 #define GET_R2L_SEGNO(free_i, segno)    (segno + free_i->start_segno)
  23
  24 #define IS_DATASEG(t)   (t <= CURSEG_COLD_DATA)
  25 #define IS_NODESEG(t)   (t >= CURSEG_HOT_NODE)
  26
  27 #define IS_CURSEG(sbi, seg)                                             \
  28         ((seg == CURSEG_I(sbi, CURSEG_HOT_DATA)->segno) ||      \
  29          (seg == CURSEG_I(sbi, CURSEG_WARM_DATA)->segno) ||     \
  30          (seg == CURSEG_I(sbi, CURSEG_COLD_DATA)->segno) ||     \
  31          (seg == CURSEG_I(sbi, CURSEG_HOT_NODE)->segno) ||      \
  32          (seg == CURSEG_I(sbi, CURSEG_WARM_NODE)->segno) ||     \
  33          (seg == CURSEG_I(sbi, CURSEG_COLD_NODE)->segno))
  34
  35 #define IS_CURSEC(sbi, secno)                                           \
  36         ((secno == CURSEG_I(sbi, CURSEG_HOT_DATA)->segno /              \
  37           sbi->segs_per_sec) || \
  38          (secno == CURSEG_I(sbi, CURSEG_WARM_DATA)->segno /             \
  39           sbi->segs_per_sec) || \
  40          (secno == CURSEG_I(sbi, CURSEG_COLD_DATA)->segno /             \
  41           sbi->segs_per_sec) || \
  42          (secno == CURSEG_I(sbi, CURSEG_HOT_NODE)->segno /              \
  43           sbi->segs_per_sec) || \
  44          (secno == CURSEG_I(sbi, CURSEG_WARM_NODE)->segno /             \
  45           sbi->segs_per_sec) || \
  46          (secno == CURSEG_I(sbi, CURSEG_COLD_NODE)->segno /             \
  47           sbi->segs_per_sec))   \
  48
  49 #define MAIN_BLKADDR(sbi)       (SM_I(sbi)->main_blkaddr)
  50 #define SEG0_BLKADDR(sbi)       (SM_I(sbi)->seg0_blkaddr)
  51
  52 #define MAIN_SEGS(sbi)  (SM_I(sbi)->main_segments)
  53 #define MAIN_SECS(sbi)  (sbi->total_sections)
  54
  55 #define TOTAL_SEGS(sbi) (SM_I(sbi)->segment_count)
  56 #define TOTAL_BLKS(sbi) (TOTAL_SEGS(sbi) << sbi->log_blocks_per_seg)
  57
  58 #define MAX_BLKADDR(sbi)        (SEG0_BLKADDR(sbi) + TOTAL_BLKS(sbi))
  59 #define SEGMENT_SIZE(sbi)       (1ULL << (sbi->log_blocksize +          \
  60                                         sbi->log_blocks_per_seg))
  61
  62 #define START_BLOCK(sbi, segno) (SEG0_BLKADDR(sbi) +                    \
  63          (GET_R2L_SEGNO(FREE_I(sbi), segno) << sbi->log_blocks_per_seg))
  64
  65 #define NEXT_FREE_BLKADDR(sbi, curseg)                                  \
  66         (START_BLOCK(sbi, curseg->segno) + curseg->next_blkoff)
  67
  68 #define GET_SEGOFF_FROM_SEG0(sbi, blk_addr)     ((blk_addr) - SEG0_BLKADDR(sbi))
  69 #define GET_SEGNO_FROM_SEG0(sbi, blk_addr)                              \
  70         (GET_SEGOFF_FROM_SEG0(sbi, blk_addr) >> sbi->log_blocks_per_seg)
  71 #define GET_BLKOFF_FROM_SEG0(sbi, blk_addr)                             \
  72         (GET_SEGOFF_FROM_SEG0(sbi, blk_addr) & (sbi->blocks_per_seg - 1))
  73
  74 #define GET_SEGNO(sbi, blk_addr)                                        \
  75         (((blk_addr == NULL_ADDR) || (blk_addr == NEW_ADDR)) ?          \
  76         NULL_SEGNO : GET_L2R_SEGNO(FREE_I(sbi),                 \
  77                 GET_SEGNO_FROM_SEG0(sbi, blk_addr)))
  78 #define GET_SECNO(sbi, segno)                                   \
  79         ((segno) / sbi->segs_per_sec)
  80 #define GET_ZONENO_FROM_SEGNO(sbi, segno)                               \
  81         ((segno / sbi->segs_per_sec) / sbi->secs_per_zone)
  82
  83 #define GET_SUM_BLOCK(sbi, segno)                               \
  84         ((sbi->sm_info->ssa_blkaddr) + segno)
  85
  86 #define GET_SUM_TYPE(footer) ((footer)->entry_type)
  87 #define SET_SUM_TYPE(footer, type) ((footer)->entry_type = type)
  88
  89 #define SIT_ENTRY_OFFSET(sit_i, segno)                                  \
  90         (segno % sit_i->sents_per_block)
  91 #define SIT_BLOCK_OFFSET(segno)                                 \
  92         (segno / SIT_ENTRY_PER_BLOCK)
  93 #define START_SEGNO(segno)              \
  94         (SIT_BLOCK_OFFSET(segno) * SIT_ENTRY_PER_BLOCK)
  95 #define SIT_BLK_CNT(sbi)                        \
  96         ((MAIN_SEGS(sbi) + SIT_ENTRY_PER_BLOCK - 1) / SIT_ENTRY_PER_BLOCK)
  97 #define f2fs_bitmap_size(nr)                    \
  98         (BITS_TO_LONGS(nr) * sizeof(unsigned long))
  99
 100 #define SECTOR_FROM_BLOCK(blk_addr)                                     \
 101         (((sector_t)blk_addr) << F2FS_LOG_SECTORS_PER_BLOCK)
 102 #define SECTOR_TO_BLOCK(sectors)                                        \
 103         (sectors >> F2FS_LOG_SECTORS_PER_BLOCK)
 104 #define MAX_BIO_BLOCKS(sbi)                                             \
 105         ((int)min((int)max_hw_blocks(sbi), BIO_MAX_PAGES))
 106
 107 /*
 108  * indicate a block allocation direction: RIGHT and LEFT.
 109  * RIGHT means allocating new sections towards the end of volume.
 110  * LEFT means the opposite direction.
 111  */
 112 enum {
 113         ALLOC_RIGHT = 0,
 114         ALLOC_LEFT
 115 };
 116
 117 /*
 118  * In the victim_sel_policy->alloc_mode, there are two block allocation modes.
 119  * LFS writes data sequentially with cleaning operations.
 120  * SSR (Slack Space Recycle) reuses obsolete space without cleaning operations.
 121  */
 122 enum {
 123         LFS = 0,
 124         SSR
 125 };
 126
 127 /*
 128  * In the victim_sel_policy->gc_mode, there are two gc, aka cleaning, modes.
 129  * GC_CB is based on cost-benefit algorithm.
 130  * GC_GREEDY is based on greedy algorithm.
 131  */
 132 enum {
 133         GC_CB = 0,
 134         GC_GREEDY
 135 };
 136
 137 /*
 138  * BG_GC means the background cleaning job.
 139  * FG_GC means the on-demand cleaning job.
 140  * FORCE_FG_GC means on-demand cleaning job in background.
 141  */
 142 enum {
 143         BG_GC = 0,
 144         FG_GC,
 145         FORCE_FG_GC,
 146 };
 147
 148 /* for a function parameter to select a victim segment */
 149 struct victim_sel_policy {
 150         int alloc_mode;                 /* LFS or SSR */
 151         int gc_mode;                    /* GC_CB or GC_GREEDY */
 152         unsigned long *dirty_segmap;    /* dirty segment bitmap */
 153         unsigned int max_search;        /* maximum # of segments to search */
 154         unsigned int offset;            /* last scanned bitmap offset */
 155         unsigned int ofs_unit;          /* bitmap search unit */
 156         unsigned int min_cost;          /* minimum cost */
 157         unsigned int min_segno;         /* segment # having min. cost */
 158 };
 159
 160 struct seg_entry {
 161         unsigned short valid_blocks;    /* # of valid blocks */
 162         unsigned char *cur_valid_map;   /* validity bitmap of blocks */
 163         /*
 164          * # of valid blocks and the validity bitmap stored in the the last
 165          * checkpoint pack. This information is used by the SSR mode.
 166          */
 167         unsigned short ckpt_valid_blocks;
 168         unsigned char *ckpt_valid_map;
 169         unsigned char *discard_map;
 170         unsigned char type;             /* segment type like CURSEG_XXX_TYPE */
 171         unsigned long long mtime;       /* modification time of the segment */
 172 };
 173
 174 struct sec_entry {
 175         unsigned int valid_blocks;      /* # of valid blocks in a section */
 176 };
 177
 178 struct segment_allocation {
 179         void (*allocate_segment)(struct f2fs_sb_info *, int, bool);
 180 };
 181
 182 /*
 183  * this value is set in page as a private data which indicate that
 184  * the page is atomically written, and it is in inmem_pages list.
 185  */
 186 #define ATOMIC_WRITTEN_PAGE             ((unsigned long)-1)
 187
 188 #define IS_ATOMIC_WRITTEN_PAGE(page)                    \
 189                 (page_private(page) == (unsigned long)ATOMIC_WRITTEN_PAGE)
 190
 191 struct inmem_pages {
 192         struct list_head list;
 193         struct page *page;
 194         block_t old_addr;               /* for revoking when fail to commit */
 195 };
 196
 197 struct sit_info {
 198         const struct segment_allocation *s_ops;
 199
 200         block_t sit_base_addr;          /* start block address of SIT area */
 201         block_t sit_blocks;             /* # of blocks used by SIT area */
 202         block_t written_valid_blocks;   /* # of valid blocks in main area */
 203         char *sit_bitmap;               /* SIT bitmap pointer */
 204         unsigned int bitmap_size;       /* SIT bitmap size */
 205
 206         unsigned long *tmp_map;                 /* bitmap for temporal use */
 207         unsigned long *dirty_sentries_bitmap;   /* bitmap for dirty sentries */
 208         unsigned int dirty_sentries;            /* # of dirty sentries */
 209         unsigned int sents_per_block;           /* # of SIT entries per block */
 210         struct mutex sentry_lock;               /* to protect SIT cache */
 211         struct seg_entry *sentries;             /* SIT segment-level cache */
 212         struct sec_entry *sec_entries;          /* SIT section-level cache */
 213
 214         /* for cost-benefit algorithm in cleaning procedure */
 215         unsigned long long elapsed_time;        /* elapsed time after mount */
 216         unsigned long long mounted_time;        /* mount time */
 217         unsigned long long min_mtime;           /* min. modification time */
 218         unsigned long long max_mtime;           /* max. modification time */
 219 };
 220
 221 struct free_segmap_info {
 222         unsigned int start_segno;       /* start segment number logically */
 223         unsigned int free_segments;     /* # of free segments */
 224         unsigned int free_sections;     /* # of free sections */
 225         spinlock_t segmap_lock;         /* free segmap lock */
 226         unsigned long *free_segmap;     /* free segment bitmap */
 227         unsigned long *free_secmap;     /* free section bitmap */
 228 };
 229
 230 /* Notice: The order of dirty type is same with CURSEG_XXX in f2fs.h */
 231 enum dirty_type {
 232         DIRTY_HOT_DATA,         /* dirty segments assigned as hot data logs */
 233         DIRTY_WARM_DATA,        /* dirty segments assigned as warm data logs */
 234         DIRTY_COLD_DATA,        /* dirty segments assigned as cold data logs */
 235         DIRTY_HOT_NODE,         /* dirty segments assigned as hot node logs */
 236         DIRTY_WARM_NODE,        /* dirty segments assigned as warm node logs */
 237         DIRTY_COLD_NODE,        /* dirty segments assigned as cold node logs */
 238         DIRTY,                  /* to count # of dirty segments */
 239         PRE,                    /* to count # of entirely obsolete segments */
 240         NR_DIRTY_TYPE
 241 };
 242
 243 struct dirty_seglist_info {
 244         const struct victim_selection *v_ops;   /* victim selction operation */
 245         unsigned long *dirty_segmap[NR_DIRTY_TYPE];
 246         struct mutex seglist_lock;              /* lock for segment bitmaps */
 247         int nr_dirty[NR_DIRTY_TYPE];            /* # of dirty segments */
 248         unsigned long *victim_secmap;           /* background GC victims */
 249 };
 250
 251 /* victim selection function for cleaning and SSR */
 252 struct victim_selection {
 253         int (*get_victim)(struct f2fs_sb_info *, unsigned int *,
 254                                                         int, int, char);
 255 };
 256
 257 /* for active log information */
 258 struct curseg_info {
 259         struct mutex curseg_mutex;              /* lock for consistency */
 260         struct f2fs_summary_block *sum_blk;     /* cached summary block */
 261         struct rw_semaphore journal_rwsem;      /* protect journal area */
 262         struct f2fs_journal *journal;           /* cached journal info */
 263         unsigned char alloc_type;               /* current allocation type */
 264         unsigned int segno;                     /* current segment number */
 265         unsigned short next_blkoff;             /* next block offset to write */
 266         unsigned int zone;                      /* current zone number */
 267         unsigned int next_segno;                /* preallocated segment */
 268 };
 269
 270 struct sit_entry_set {
 271         struct list_head set_list;      /* link with all sit sets */
 272         unsigned int start_segno;       /* start segno of sits in set */
 273         unsigned int entry_cnt;         /* the # of sit entries in set */
 274 };
 275
 276 /*
 277  * inline functions
 278  */
 279 static inline struct curseg_info *CURSEG_I(struct f2fs_sb_info *sbi, int type)
 280 {
 281         return (struct curseg_info *)(SM_I(sbi)->curseg_array + type);
 282 }
 283
 284 static inline struct seg_entry *get_seg_entry(struct f2fs_sb_info *sbi,
 285                                                 unsigned int segno)
 286 {
 287         struct sit_info *sit_i = SIT_I(sbi);
 288         return &sit_i->sentries[segno];
 289 }
 290
 291 static inline struct sec_entry *get_sec_entry(struct f2fs_sb_info *sbi,
 292                                                 unsigned int segno)
 293 {
 294         struct sit_info *sit_i = SIT_I(sbi);
 295         return &sit_i->sec_entries[GET_SECNO(sbi, segno)];
 296 }
 297
 298 static inline unsigned int get_valid_blocks(struct f2fs_sb_info *sbi,
 299                                 unsigned int segno, int section)
 300 {
 301         /*
 302          * In order to get # of valid blocks in a section instantly from many
 303          * segments, f2fs manages two counting structures separately.
 304          */
 305         if (section > 1)
 306                 return get_sec_entry(sbi, segno)->valid_blocks;
 307         else
 308                 return get_seg_entry(sbi, segno)->valid_blocks;
 309 }
 310
 311 static inline void seg_info_from_raw_sit(struct seg_entry *se,
 312                                         struct f2fs_sit_entry *rs)
 313 {
 314         se->valid_blocks = GET_SIT_VBLOCKS(rs);
 315         se->ckpt_valid_blocks = GET_SIT_VBLOCKS(rs);
 316         memcpy(se->cur_valid_map, rs->valid_map, SIT_VBLOCK_MAP_SIZE);
 317         memcpy(se->ckpt_valid_map, rs->valid_map, SIT_VBLOCK_MAP_SIZE);
 318         se->type = GET_SIT_TYPE(rs);
 319         se->mtime = le64_to_cpu(rs->mtime);
 320 }
 321
 322 static inline void seg_info_to_raw_sit(struct seg_entry *se,
 323                                         struct f2fs_sit_entry *rs)
 324 {
 325         unsigned short raw_vblocks = (se->type << SIT_VBLOCKS_SHIFT) |
 326                                         se->valid_blocks;
 327         rs->vblocks = cpu_to_le16(raw_vblocks);
 328         memcpy(rs->valid_map, se->cur_valid_map, SIT_VBLOCK_MAP_SIZE);
 329         memcpy(se->ckpt_valid_map, rs->valid_map, SIT_VBLOCK_MAP_SIZE);
 330         se->ckpt_valid_blocks = se->valid_blocks;
 331         rs->mtime = cpu_to_le64(se->mtime);
 332 }
 333
 334 static inline unsigned int find_next_inuse(struct free_segmap_info *free_i,
 335                 unsigned int max, unsigned int segno)
 336 {
 337         unsigned int ret;
 338         spin_lock(&free_i->segmap_lock);
 339         ret = find_next_bit(free_i->free_segmap, max, segno);
 340         spin_unlock(&free_i->segmap_lock);
 341         return ret;
 342 }
 343
 344 static inline void __set_free(struct f2fs_sb_info *sbi, unsigned int segno)
 345 {
 346         struct free_segmap_info *free_i = FREE_I(sbi);
 347         unsigned int secno = segno / sbi->segs_per_sec;
 348         unsigned int start_segno = secno * sbi->segs_per_sec;
 349         unsigned int next;
 350
 351         spin_lock(&free_i->segmap_lock);
 352         clear_bit(segno, free_i->free_segmap);
 353         free_i->free_segments++;
 354
 355         next = find_next_bit(free_i->free_segmap,
 356                         start_segno + sbi->segs_per_sec, start_segno);
 357         if (next >= start_segno + sbi->segs_per_sec) {
 358                 clear_bit(secno, free_i->free_secmap);
 359                 free_i->free_sections++;
 360         }
 361         spin_unlock(&free_i->segmap_lock);
 362 }
 363
 364 static inline void __set_inuse(struct f2fs_sb_info *sbi,
 365                 unsigned int segno)
 366 {
 367         struct free_segmap_info *free_i = FREE_I(sbi);
 368         unsigned int secno = segno / sbi->segs_per_sec;
 369         set_bit(segno, free_i->free_segmap);
 370         free_i->free_segments--;
 371         if (!test_and_set_bit(secno, free_i->free_secmap))
 372                 free_i->free_sections--;
 373 }
 374
 375 static inline void __set_test_and_free(struct f2fs_sb_info *sbi,
 376                 unsigned int segno)
 377 {
 378         struct free_segmap_info *free_i = FREE_I(sbi);
 379         unsigned int secno = segno / sbi->segs_per_sec;
 380         unsigned int start_segno = secno * sbi->segs_per_sec;
 381         unsigned int next;
 382
 383         spin_lock(&free_i->segmap_lock);
 384         if (test_and_clear_bit(segno, free_i->free_segmap)) {
 385                 free_i->free_segments++;
 386
 387                 next = find_next_bit(free_i->free_segmap,
 388                                 start_segno + sbi->segs_per_sec, start_segno);
 389                 if (next >= start_segno + sbi->segs_per_sec) {
 390                         if (test_and_clear_bit(secno, free_i->free_secmap))
 391                                 free_i->free_sections++;
 392                 }
 393         }
 394         spin_unlock(&free_i->segmap_lock);
 395 }
 396
 397 static inline void __set_test_and_inuse(struct f2fs_sb_info *sbi,
 398                 unsigned int segno)
 399 {
 400         struct free_segmap_info *free_i = FREE_I(sbi);
 401         unsigned int secno = segno / sbi->segs_per_sec;
 402         spin_lock(&free_i->segmap_lock);
 403         if (!test_and_set_bit(segno, free_i->free_segmap)) {
 404                 free_i->free_segments--;
 405                 if (!test_and_set_bit(secno, free_i->free_secmap))
 406                         free_i->free_sections--;
 407         }
 408         spin_unlock(&free_i->segmap_lock);
 409 }
 410
 411 static inline void get_sit_bitmap(struct f2fs_sb_info *sbi,
 412                 void *dst_addr)
 413 {
 414         struct sit_info *sit_i = SIT_I(sbi);
 415         memcpy(dst_addr, sit_i->sit_bitmap, sit_i->bitmap_size);
 416 }
 417
 418 static inline block_t written_block_count(struct f2fs_sb_info *sbi)
 419 {
 420         return SIT_I(sbi)->written_valid_blocks;
 421 }
 422
 423 static inline unsigned int free_segments(struct f2fs_sb_info *sbi)
 424 {
 425         return FREE_I(sbi)->free_segments;
 426 }
 427
 428 static inline int reserved_segments(struct f2fs_sb_info *sbi)
 429 {
 430         return SM_I(sbi)->reserved_segments;
 431 }
 432
 433 static inline unsigned int free_sections(struct f2fs_sb_info *sbi)
 434 {
 435         return FREE_I(sbi)->free_sections;
 436 }
 437
 438 static inline unsigned int prefree_segments(struct f2fs_sb_info *sbi)
 439 {
 440         return DIRTY_I(sbi)->nr_dirty[PRE];
 441 }
 442
 443 static inline unsigned int dirty_segments(struct f2fs_sb_info *sbi)
 444 {
 445         return DIRTY_I(sbi)->nr_dirty[DIRTY_HOT_DATA] +
 446                 DIRTY_I(sbi)->nr_dirty[DIRTY_WARM_DATA] +
 447                 DIRTY_I(sbi)->nr_dirty[DIRTY_COLD_DATA] +
 448                 DIRTY_I(sbi)->nr_dirty[DIRTY_HOT_NODE] +
 449                 DIRTY_I(sbi)->nr_dirty[DIRTY_WARM_NODE] +
 450                 DIRTY_I(sbi)->nr_dirty[DIRTY_COLD_NODE];
 451 }
 452
 453 static inline int overprovision_segments(struct f2fs_sb_info *sbi)
 454 {
 455         return SM_I(sbi)->ovp_segments;
 456 }
 457
 458 static inline int overprovision_sections(struct f2fs_sb_info *sbi)
 459 {
 460         return ((unsigned int) overprovision_segments(sbi)) / sbi->segs_per_sec;
 461 }
 462
 463 static inline int reserved_sections(struct f2fs_sb_info *sbi)
 464 {
 465         return ((unsigned int) reserved_segments(sbi)) / sbi->segs_per_sec;
 466 }
 467
 468 static inline bool need_SSR(struct f2fs_sb_info *sbi)
 469 {
 470         int node_secs = get_blocktype_secs(sbi, F2FS_DIRTY_NODES);
 471         int dent_secs = get_blocktype_secs(sbi, F2FS_DIRTY_DENTS);
 472         return free_sections(sbi) <= (node_secs + 2 * dent_secs +
 473                                                 reserved_sections(sbi) + 1);
 474 }
 475
 476 static inline bool has_not_enough_free_secs(struct f2fs_sb_info *sbi, int freed)
 477 {
 478         int node_secs = get_blocktype_secs(sbi, F2FS_DIRTY_NODES);
 479         int dent_secs = get_blocktype_secs(sbi, F2FS_DIRTY_DENTS);
 480
 481         if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
 482                 return false;
 483
 484         return (free_sections(sbi) + freed) <= (node_secs + 2 * dent_secs +
 485                                                 reserved_sections(sbi));
 486 }
 487
 488 static inline bool excess_prefree_segs(struct f2fs_sb_info *sbi)
 489 {
 490         return prefree_segments(sbi) > SM_I(sbi)->rec_prefree_segments;
 491 }
 492
 493 static inline int utilization(struct f2fs_sb_info *sbi)
 494 {
 495         return div_u64((u64)valid_user_blocks(sbi) * 100,
 496                                         sbi->user_block_count);
 497 }
 498
 499 /*
 500  * Sometimes f2fs may be better to drop out-of-place update policy.
 501  * And, users can control the policy through sysfs entries.
 502  * There are five policies with triggering conditions as follows.
 503  * F2FS_IPU_FORCE - all the time,
 504  * F2FS_IPU_SSR - if SSR mode is activated,
 505  * F2FS_IPU_UTIL - if FS utilization is over threashold,
 506  * F2FS_IPU_SSR_UTIL - if SSR mode is activated and FS utilization is over
 507  *                     threashold,
 508  * F2FS_IPU_FSYNC - activated in fsync path only for high performance flash
 509  *                     storages. IPU will be triggered only if the # of dirty
 510  *                     pages over min_fsync_blocks.
 511  * F2FS_IPUT_DISABLE - disable IPU. (=default option)
 512  */
 513 #define DEF_MIN_IPU_UTIL        70
 514 #define DEF_MIN_FSYNC_BLOCKS    8
 515
 516 enum {
 517         F2FS_IPU_FORCE,
 518         F2FS_IPU_SSR,
 519         F2FS_IPU_UTIL,
 520         F2FS_IPU_SSR_UTIL,
 521         F2FS_IPU_FSYNC,
 522 };
 523
 524 static inline bool need_inplace_update(struct inode *inode)
 525 {
 526         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 527         unsigned int policy = SM_I(sbi)->ipu_policy;
 528
 529         /* IPU can be done only for the user data */
 530         if (S_ISDIR(inode->i_mode) || f2fs_is_atomic_file(inode))
 531                 return false;
 532
 533         if (policy & (0x1 << F2FS_IPU_FORCE))
 534                 return true;
 535         if (policy & (0x1 << F2FS_IPU_SSR) && need_SSR(sbi))
 536                 return true;
 537         if (policy & (0x1 << F2FS_IPU_UTIL) &&
 538                         utilization(sbi) > SM_I(sbi)->min_ipu_util)
 539                 return true;
 540         if (policy & (0x1 << F2FS_IPU_SSR_UTIL) && need_SSR(sbi) &&
 541                         utilization(sbi) > SM_I(sbi)->min_ipu_util)
 542                 return true;
 543
 544         /* this is only set during fdatasync */
 545         if (policy & (0x1 << F2FS_IPU_FSYNC) &&
 546                         is_inode_flag_set(F2FS_I(inode), FI_NEED_IPU))
 547                 return true;
 548
 549         return false;
 550 }
 551
 552 static inline unsigned int curseg_segno(struct f2fs_sb_info *sbi,
 553                 int type)
 554 {
 555         struct curseg_info *curseg = CURSEG_I(sbi, type);
 556         return curseg->segno;
 557 }
 558
 559 static inline unsigned char curseg_alloc_type(struct f2fs_sb_info *sbi,
 560                 int type)
 561 {
 562         struct curseg_info *curseg = CURSEG_I(sbi, type);
 563         return curseg->alloc_type;
 564 }
 565
 566 static inline unsigned short curseg_blkoff(struct f2fs_sb_info *sbi, int type)
 567 {
 568         struct curseg_info *curseg = CURSEG_I(sbi, type);
 569         return curseg->next_blkoff;
 570 }
 571
 572 static inline void check_seg_range(struct f2fs_sb_info *sbi, unsigned int segno)
 573 {
 574         f2fs_bug_on(sbi, segno > TOTAL_SEGS(sbi) - 1);
 575 }
 576
 577 static inline void verify_block_addr(struct f2fs_sb_info *sbi, block_t blk_addr)
 578 {
 579         f2fs_bug_on(sbi, blk_addr < SEG0_BLKADDR(sbi)
 580                                         || blk_addr >= MAX_BLKADDR(sbi));
 581 }
 582
 583 /*
 584  * Summary block is always treated as an invalid block
 585  */
 586 static inline void check_block_count(struct f2fs_sb_info *sbi,
 587                 int segno, struct f2fs_sit_entry *raw_sit)
 588 {
 589 #ifdef CONFIG_F2FS_CHECK_FS
 590         bool is_valid  = test_bit_le(0, raw_sit->valid_map) ? true : false;
 591         int valid_blocks = 0;
 592         int cur_pos = 0, next_pos;
 593
 594         /* check bitmap with valid block count */
 595         do {
 596                 if (is_valid) {
 597                         next_pos = find_next_zero_bit_le(&raw_sit->valid_map,
 598                                         sbi->blocks_per_seg,
 599                                         cur_pos);
 600                         valid_blocks += next_pos - cur_pos;
 601                 } else
 602                         next_pos = find_next_bit_le(&raw_sit->valid_map,
 603                                         sbi->blocks_per_seg,
 604                                         cur_pos);
 605                 cur_pos = next_pos;
 606                 is_valid = !is_valid;
 607         } while (cur_pos < sbi->blocks_per_seg);
 608         BUG_ON(GET_SIT_VBLOCKS(raw_sit) != valid_blocks);
 609 #endif
 610         /* check segment usage, and check boundary of a given segment number */
 611         f2fs_bug_on(sbi, GET_SIT_VBLOCKS(raw_sit) > sbi->blocks_per_seg
 612                                         || segno > TOTAL_SEGS(sbi) - 1);
 613 }
 614
 615 static inline pgoff_t current_sit_addr(struct f2fs_sb_info *sbi,
 616                                                 unsigned int start)
 617 {
 618         struct sit_info *sit_i = SIT_I(sbi);
 619         unsigned int offset = SIT_BLOCK_OFFSET(start);
 620         block_t blk_addr = sit_i->sit_base_addr + offset;
 621
 622         check_seg_range(sbi, start);
 623
 624         /* calculate sit block address */
 625         if (f2fs_test_bit(offset, sit_i->sit_bitmap))
 626                 blk_addr += sit_i->sit_blocks;
 627
 628         return blk_addr;
 629 }
 630
 631 static inline pgoff_t next_sit_addr(struct f2fs_sb_info *sbi,
 632                                                 pgoff_t block_addr)
 633 {
 634         struct sit_info *sit_i = SIT_I(sbi);
 635         block_addr -= sit_i->sit_base_addr;
 636         if (block_addr < sit_i->sit_blocks)
 637                 block_addr += sit_i->sit_blocks;
 638         else
 639                 block_addr -= sit_i->sit_blocks;
 640
 641         return block_addr + sit_i->sit_base_addr;
 642 }
 643
 644 static inline void set_to_next_sit(struct sit_info *sit_i, unsigned int start)
 645 {
 646         unsigned int block_off = SIT_BLOCK_OFFSET(start);
 647
 648         f2fs_change_bit(block_off, sit_i->sit_bitmap);
 649 }
 650
 651 static inline unsigned long long get_mtime(struct f2fs_sb_info *sbi)
 652 {
 653         struct sit_info *sit_i = SIT_I(sbi);
 654         return sit_i->elapsed_time + CURRENT_TIME_SEC.tv_sec -
 655                                                 sit_i->mounted_time;
 656 }
 657
 658 static inline void set_summary(struct f2fs_summary *sum, nid_t nid,
 659                         unsigned int ofs_in_node, unsigned char version)
 660 {
 661         sum->nid = cpu_to_le32(nid);
 662         sum->ofs_in_node = cpu_to_le16(ofs_in_node);
 663         sum->version = version;
 664 }
 665
 666 static inline block_t start_sum_block(struct f2fs_sb_info *sbi)
 667 {
 668         return __start_cp_addr(sbi) +
 669                 le32_to_cpu(F2FS_CKPT(sbi)->cp_pack_start_sum);
 670 }
 671
 672 static inline block_t sum_blk_addr(struct f2fs_sb_info *sbi, int base, int type)
 673 {
 674         return __start_cp_addr(sbi) +
 675                 le32_to_cpu(F2FS_CKPT(sbi)->cp_pack_total_block_count)
 676                                 - (base + 1) + type;
 677 }
 678
 679 static inline bool sec_usage_check(struct f2fs_sb_info *sbi, unsigned int secno)
 680 {
 681         if (IS_CURSEC(sbi, secno) || (sbi->cur_victim_sec == secno))
 682                 return true;
 683         return false;
 684 }
 685
 686 static inline unsigned int max_hw_blocks(struct f2fs_sb_info *sbi)
 687 {
 688         struct block_device *bdev = sbi->sb->s_bdev;
 689         struct request_queue *q = bdev_get_queue(bdev);
 690         return SECTOR_TO_BLOCK(queue_max_sectors(q));
 691 }
 692
 693 /*
 694  * It is very important to gather dirty pages and write at once, so that we can
 695  * submit a big bio without interfering other data writes.
 696  * By default, 512 pages for directory data,
 697  * 512 pages (2MB) * 3 for three types of nodes, and
 698  * max_bio_blocks for meta are set.
 699  */
 700 static inline int nr_pages_to_skip(struct f2fs_sb_info *sbi, int type)
 701 {
 702         if (sbi->sb->s_bdi->wb.dirty_exceeded)
 703                 return 0;
 704
 705         if (type == DATA)
 706                 return sbi->blocks_per_seg;
 707         else if (type == NODE)
 708                 return 3 * sbi->blocks_per_seg;
 709         else if (type == META)
 710                 return MAX_BIO_BLOCKS(sbi);
 711         else
 712                 return 0;
 713 }
 714
 715 /*
 716  * When writing pages, it'd better align nr_to_write for segment size.
 717  */
 718 static inline long nr_pages_to_write(struct f2fs_sb_info *sbi, int type,
 719                                         struct writeback_control *wbc)
 720 {
 721         long nr_to_write, desired;
 722
 723         if (wbc->sync_mode != WB_SYNC_NONE)
 724                 return 0;
 725
 726         nr_to_write = wbc->nr_to_write;
 727
 728         if (type == DATA)
 729                 desired = 4096;
 730         else if (type == NODE)
 731                 desired = 3 * max_hw_blocks(sbi);
 732         else
 733                 desired = MAX_BIO_BLOCKS(sbi);
 734
 735         wbc->nr_to_write = desired;
 736         return desired - nr_to_write;
 737 }