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
  13 /* constant macro */
  14 #define NULL_SEGNO                      ((unsigned int)(~0))
  15 #define NULL_SECNO                      ((unsigned int)(~0))
  16
  17 #define DEF_RECLAIM_PREFREE_SEGMENTS    5       /* 5% over total segments */
  18
  19 /* L: Logical segment # in volume, R: Relative segment # in main area */
  20 #define GET_L2R_SEGNO(free_i, segno)    (segno - free_i->start_segno)
  21 #define GET_R2L_SEGNO(free_i, segno)    (segno + free_i->start_segno)
  22
  23 #define IS_DATASEG(t)   (t <= CURSEG_COLD_DATA)
  24 #define IS_NODESEG(t)   (t >= CURSEG_HOT_NODE)
  25
  26 #define IS_CURSEG(sbi, seg)                                             \
  27         ((seg == CURSEG_I(sbi, CURSEG_HOT_DATA)->segno) ||      \
  28          (seg == CURSEG_I(sbi, CURSEG_WARM_DATA)->segno) ||     \
  29          (seg == CURSEG_I(sbi, CURSEG_COLD_DATA)->segno) ||     \
  30          (seg == CURSEG_I(sbi, CURSEG_HOT_NODE)->segno) ||      \
  31          (seg == CURSEG_I(sbi, CURSEG_WARM_NODE)->segno) ||     \
  32          (seg == CURSEG_I(sbi, CURSEG_COLD_NODE)->segno))
  33
  34 #define IS_CURSEC(sbi, secno)                                           \
  35         ((secno == CURSEG_I(sbi, CURSEG_HOT_DATA)->segno /              \
  36           sbi->segs_per_sec) || \
  37          (secno == CURSEG_I(sbi, CURSEG_WARM_DATA)->segno /             \
  38           sbi->segs_per_sec) || \
  39          (secno == CURSEG_I(sbi, CURSEG_COLD_DATA)->segno /             \
  40           sbi->segs_per_sec) || \
  41          (secno == CURSEG_I(sbi, CURSEG_HOT_NODE)->segno /              \
  42           sbi->segs_per_sec) || \
  43          (secno == CURSEG_I(sbi, CURSEG_WARM_NODE)->segno /             \
  44           sbi->segs_per_sec) || \
  45          (secno == CURSEG_I(sbi, CURSEG_COLD_NODE)->segno /             \
  46           sbi->segs_per_sec))   \
  47
  48 #define MAIN_BLKADDR(sbi)       (SM_I(sbi)->main_blkaddr)
  49 #define SEG0_BLKADDR(sbi)       (SM_I(sbi)->seg0_blkaddr)
  50
  51 #define MAIN_SEGS(sbi)  (SM_I(sbi)->main_segments)
  52 #define MAIN_SECS(sbi)  (sbi->total_sections)
  53
  54 #define TOTAL_SEGS(sbi) (SM_I(sbi)->segment_count)
  55 #define TOTAL_BLKS(sbi) (TOTAL_SEGS(sbi) << sbi->log_blocks_per_seg)
  56
  57 #define MAX_BLKADDR(sbi)        (SEG0_BLKADDR(sbi) + TOTAL_BLKS(sbi))
  58 #define SEGMENT_SIZE(sbi)       (1ULL << (sbi->log_blocksize +          \
  59                                         sbi->log_blocks_per_seg))
  60
  61 #define START_BLOCK(sbi, segno) (SEG0_BLKADDR(sbi) +                    \
  62          (GET_R2L_SEGNO(FREE_I(sbi), segno) << sbi->log_blocks_per_seg))
  63
  64 #define NEXT_FREE_BLKADDR(sbi, curseg)                                  \
  65         (START_BLOCK(sbi, curseg->segno) + curseg->next_blkoff)
  66
  67 #define GET_SEGOFF_FROM_SEG0(sbi, blk_addr)     ((blk_addr) - SEG0_BLKADDR(sbi))
  68 #define GET_SEGNO_FROM_SEG0(sbi, blk_addr)                              \
  69         (GET_SEGOFF_FROM_SEG0(sbi, blk_addr) >> sbi->log_blocks_per_seg)
  70 #define GET_BLKOFF_FROM_SEG0(sbi, blk_addr)                             \
  71         (GET_SEGOFF_FROM_SEG0(sbi, blk_addr) & (sbi->blocks_per_seg - 1))
  72
  73 #define GET_SEGNO(sbi, blk_addr)                                        \
  74         (((blk_addr == NULL_ADDR) || (blk_addr == NEW_ADDR)) ?          \
  75         NULL_SEGNO : GET_L2R_SEGNO(FREE_I(sbi),                 \
  76                 GET_SEGNO_FROM_SEG0(sbi, blk_addr)))
  77 #define GET_SECNO(sbi, segno)                                   \
  78         ((segno) / sbi->segs_per_sec)
  79 #define GET_ZONENO_FROM_SEGNO(sbi, segno)                               \
  80         ((segno / sbi->segs_per_sec) / sbi->secs_per_zone)
  81
  82 #define GET_SUM_BLOCK(sbi, segno)                               \
  83         ((sbi->sm_info->ssa_blkaddr) + segno)
  84
  85 #define GET_SUM_TYPE(footer) ((footer)->entry_type)
  86 #define SET_SUM_TYPE(footer, type) ((footer)->entry_type = type)
  87
  88 #define SIT_ENTRY_OFFSET(sit_i, segno)                                  \
  89         (segno % sit_i->sents_per_block)
  90 #define SIT_BLOCK_OFFSET(segno)                                 \
  91         (segno / SIT_ENTRY_PER_BLOCK)
  92 #define START_SEGNO(segno)              \
  93         (SIT_BLOCK_OFFSET(segno) * SIT_ENTRY_PER_BLOCK)
  94 #define SIT_BLK_CNT(sbi)                        \
  95         ((MAIN_SEGS(sbi) + SIT_ENTRY_PER_BLOCK - 1) / SIT_ENTRY_PER_BLOCK)
  96 #define f2fs_bitmap_size(nr)                    \
  97         (BITS_TO_LONGS(nr) * sizeof(unsigned long))
  98
  99 #define SECTOR_FROM_BLOCK(blk_addr)                                     \
 100         (((sector_t)blk_addr) << F2FS_LOG_SECTORS_PER_BLOCK)
 101 #define SECTOR_TO_BLOCK(sectors)                                        \
 102         (sectors >> F2FS_LOG_SECTORS_PER_BLOCK)
 103 #define MAX_BIO_BLOCKS(sbi)                                             \
 104         ((int)min((int)max_hw_blocks(sbi), BIO_MAX_PAGES))
 105
 106 /*
 107  * indicate a block allocation direction: RIGHT and LEFT.
 108  * RIGHT means allocating new sections towards the end of volume.
 109  * LEFT means the opposite direction.
 110  */
 111 enum {
 112         ALLOC_RIGHT = 0,
 113         ALLOC_LEFT
 114 };
 115
 116 /*
 117  * In the victim_sel_policy->alloc_mode, there are two block allocation modes.
 118  * LFS writes data sequentially with cleaning operations.
 119  * SSR (Slack Space Recycle) reuses obsolete space without cleaning operations.
 120  */
 121 enum {
 122         LFS = 0,
 123         SSR
 124 };
 125
 126 /*
 127  * In the victim_sel_policy->gc_mode, there are two gc, aka cleaning, modes.
 128  * GC_CB is based on cost-benefit algorithm.
 129  * GC_GREEDY is based on greedy algorithm.
 130  */
 131 enum {
 132         GC_CB = 0,
 133         GC_GREEDY
 134 };
 135
 136 /*
 137  * BG_GC means the background cleaning job.
 138  * FG_GC means the on-demand cleaning job.
 139  */
 140 enum {
 141         BG_GC = 0,
 142         FG_GC
 143 };
 144
 145 /* for a function parameter to select a victim segment */
 146 struct victim_sel_policy {
 147         int alloc_mode;                 /* LFS or SSR */
 148         int gc_mode;                    /* GC_CB or GC_GREEDY */
 149         unsigned long *dirty_segmap;    /* dirty segment bitmap */
 150         unsigned int max_search;        /* maximum # of segments to search */
 151         unsigned int offset;            /* last scanned bitmap offset */
 152         unsigned int ofs_unit;          /* bitmap search unit */
 153         unsigned int min_cost;          /* minimum cost */
 154         unsigned int min_segno;         /* segment # having min. cost */
 155 };
 156
 157 struct seg_entry {
 158         unsigned short valid_blocks;    /* # of valid blocks */
 159         unsigned char *cur_valid_map;   /* validity bitmap of blocks */
 160         /*
 161          * # of valid blocks and the validity bitmap stored in the the last
 162          * checkpoint pack. This information is used by the SSR mode.
 163          */
 164         unsigned short ckpt_valid_blocks;
 165         unsigned char *ckpt_valid_map;
 166         unsigned char type;             /* segment type like CURSEG_XXX_TYPE */
 167         unsigned long long mtime;       /* modification time of the segment */
 168 };
 169
 170 struct sec_entry {
 171         unsigned int valid_blocks;      /* # of valid blocks in a section */
 172 };
 173
 174 struct segment_allocation {
 175         void (*allocate_segment)(struct f2fs_sb_info *, int, bool);
 176 };
 177
 178 struct inmem_pages {
 179         struct list_head list;
 180         struct page *page;
 181 };
 182
 183 struct sit_info {
 184         const struct segment_allocation *s_ops;
 185
 186         block_t sit_base_addr;          /* start block address of SIT area */
 187         block_t sit_blocks;             /* # of blocks used by SIT area */
 188         block_t written_valid_blocks;   /* # of valid blocks in main area */
 189         char *sit_bitmap;               /* SIT bitmap pointer */
 190         unsigned int bitmap_size;       /* SIT bitmap size */
 191
 192         unsigned long *tmp_map;                 /* bitmap for temporal use */
 193         unsigned long *dirty_sentries_bitmap;   /* bitmap for dirty sentries */
 194         unsigned int dirty_sentries;            /* # of dirty sentries */
 195         unsigned int sents_per_block;           /* # of SIT entries per block */
 196         struct mutex sentry_lock;               /* to protect SIT cache */
 197         struct seg_entry *sentries;             /* SIT segment-level cache */
 198         struct sec_entry *sec_entries;          /* SIT section-level cache */
 199
 200         /* for cost-benefit algorithm in cleaning procedure */
 201         unsigned long long elapsed_time;        /* elapsed time after mount */
 202         unsigned long long mounted_time;        /* mount time */
 203         unsigned long long min_mtime;           /* min. modification time */
 204         unsigned long long max_mtime;           /* max. modification time */
 205 };
 206
 207 struct free_segmap_info {
 208         unsigned int start_segno;       /* start segment number logically */
 209         unsigned int free_segments;     /* # of free segments */
 210         unsigned int free_sections;     /* # of free sections */
 211         spinlock_t segmap_lock;         /* free segmap lock */
 212         unsigned long *free_segmap;     /* free segment bitmap */
 213         unsigned long *free_secmap;     /* free section bitmap */
 214 };
 215
 216 /* Notice: The order of dirty type is same with CURSEG_XXX in f2fs.h */
 217 enum dirty_type {
 218         DIRTY_HOT_DATA,         /* dirty segments assigned as hot data logs */
 219         DIRTY_WARM_DATA,        /* dirty segments assigned as warm data logs */
 220         DIRTY_COLD_DATA,        /* dirty segments assigned as cold data logs */
 221         DIRTY_HOT_NODE,         /* dirty segments assigned as hot node logs */
 222         DIRTY_WARM_NODE,        /* dirty segments assigned as warm node logs */
 223         DIRTY_COLD_NODE,        /* dirty segments assigned as cold node logs */
 224         DIRTY,                  /* to count # of dirty segments */
 225         PRE,                    /* to count # of entirely obsolete segments */
 226         NR_DIRTY_TYPE
 227 };
 228
 229 struct dirty_seglist_info {
 230         const struct victim_selection *v_ops;   /* victim selction operation */
 231         unsigned long *dirty_segmap[NR_DIRTY_TYPE];
 232         struct mutex seglist_lock;              /* lock for segment bitmaps */
 233         int nr_dirty[NR_DIRTY_TYPE];            /* # of dirty segments */
 234         unsigned long *victim_secmap;           /* background GC victims */
 235 };
 236
 237 /* victim selection function for cleaning and SSR */
 238 struct victim_selection {
 239         int (*get_victim)(struct f2fs_sb_info *, unsigned int *,
 240                                                         int, int, char);
 241 };
 242
 243 /* for active log information */
 244 struct curseg_info {
 245         struct mutex curseg_mutex;              /* lock for consistency */
 246         struct f2fs_summary_block *sum_blk;     /* cached summary block */
 247         unsigned char alloc_type;               /* current allocation type */
 248         unsigned int segno;                     /* current segment number */
 249         unsigned short next_blkoff;             /* next block offset to write */
 250         unsigned int zone;                      /* current zone number */
 251         unsigned int next_segno;                /* preallocated segment */
 252 };
 253
 254 struct sit_entry_set {
 255         struct list_head set_list;      /* link with all sit sets */
 256         unsigned int start_segno;       /* start segno of sits in set */
 257         unsigned int entry_cnt;         /* the # of sit entries in set */
 258 };
 259
 260 /*
 261  * inline functions
 262  */
 263 static inline struct curseg_info *CURSEG_I(struct f2fs_sb_info *sbi, int type)
 264 {
 265         return (struct curseg_info *)(SM_I(sbi)->curseg_array + type);
 266 }
 267
 268 static inline struct seg_entry *get_seg_entry(struct f2fs_sb_info *sbi,
 269                                                 unsigned int segno)
 270 {
 271         struct sit_info *sit_i = SIT_I(sbi);
 272         return &sit_i->sentries[segno];
 273 }
 274
 275 static inline struct sec_entry *get_sec_entry(struct f2fs_sb_info *sbi,
 276                                                 unsigned int segno)
 277 {
 278         struct sit_info *sit_i = SIT_I(sbi);
 279         return &sit_i->sec_entries[GET_SECNO(sbi, segno)];
 280 }
 281
 282 static inline unsigned int get_valid_blocks(struct f2fs_sb_info *sbi,
 283                                 unsigned int segno, int section)
 284 {
 285         /*
 286          * In order to get # of valid blocks in a section instantly from many
 287          * segments, f2fs manages two counting structures separately.
 288          */
 289         if (section > 1)
 290                 return get_sec_entry(sbi, segno)->valid_blocks;
 291         else
 292                 return get_seg_entry(sbi, segno)->valid_blocks;
 293 }
 294
 295 static inline void seg_info_from_raw_sit(struct seg_entry *se,
 296                                         struct f2fs_sit_entry *rs)
 297 {
 298         se->valid_blocks = GET_SIT_VBLOCKS(rs);
 299         se->ckpt_valid_blocks = GET_SIT_VBLOCKS(rs);
 300         memcpy(se->cur_valid_map, rs->valid_map, SIT_VBLOCK_MAP_SIZE);
 301         memcpy(se->ckpt_valid_map, rs->valid_map, SIT_VBLOCK_MAP_SIZE);
 302         se->type = GET_SIT_TYPE(rs);
 303         se->mtime = le64_to_cpu(rs->mtime);
 304 }
 305
 306 static inline void seg_info_to_raw_sit(struct seg_entry *se,
 307                                         struct f2fs_sit_entry *rs)
 308 {
 309         unsigned short raw_vblocks = (se->type << SIT_VBLOCKS_SHIFT) |
 310                                         se->valid_blocks;
 311         rs->vblocks = cpu_to_le16(raw_vblocks);
 312         memcpy(rs->valid_map, se->cur_valid_map, SIT_VBLOCK_MAP_SIZE);
 313         memcpy(se->ckpt_valid_map, rs->valid_map, SIT_VBLOCK_MAP_SIZE);
 314         se->ckpt_valid_blocks = se->valid_blocks;
 315         rs->mtime = cpu_to_le64(se->mtime);
 316 }
 317
 318 static inline unsigned int find_next_inuse(struct free_segmap_info *free_i,
 319                 unsigned int max, unsigned int segno)
 320 {
 321         unsigned int ret;
 322         spin_lock(&free_i->segmap_lock);
 323         ret = find_next_bit(free_i->free_segmap, max, segno);
 324         spin_unlock(&free_i->segmap_lock);
 325         return ret;
 326 }
 327
 328 static inline void __set_free(struct f2fs_sb_info *sbi, unsigned int segno)
 329 {
 330         struct free_segmap_info *free_i = FREE_I(sbi);
 331         unsigned int secno = segno / sbi->segs_per_sec;
 332         unsigned int start_segno = secno * sbi->segs_per_sec;
 333         unsigned int next;
 334
 335         spin_lock(&free_i->segmap_lock);
 336         clear_bit(segno, free_i->free_segmap);
 337         free_i->free_segments++;
 338
 339         next = find_next_bit(free_i->free_segmap,
 340                         start_segno + sbi->segs_per_sec, start_segno);
 341         if (next >= start_segno + sbi->segs_per_sec) {
 342                 clear_bit(secno, free_i->free_secmap);
 343                 free_i->free_sections++;
 344         }
 345         spin_unlock(&free_i->segmap_lock);
 346 }
 347
 348 static inline void __set_inuse(struct f2fs_sb_info *sbi,
 349                 unsigned int segno)
 350 {
 351         struct free_segmap_info *free_i = FREE_I(sbi);
 352         unsigned int secno = segno / sbi->segs_per_sec;
 353         set_bit(segno, free_i->free_segmap);
 354         free_i->free_segments--;
 355         if (!test_and_set_bit(secno, free_i->free_secmap))
 356                 free_i->free_sections--;
 357 }
 358
 359 static inline void __set_test_and_free(struct f2fs_sb_info *sbi,
 360                 unsigned int segno)
 361 {
 362         struct free_segmap_info *free_i = FREE_I(sbi);
 363         unsigned int secno = segno / sbi->segs_per_sec;
 364         unsigned int start_segno = secno * sbi->segs_per_sec;
 365         unsigned int next;
 366
 367         spin_lock(&free_i->segmap_lock);
 368         if (test_and_clear_bit(segno, free_i->free_segmap)) {
 369                 free_i->free_segments++;
 370
 371                 next = find_next_bit(free_i->free_segmap,
 372                                 start_segno + sbi->segs_per_sec, start_segno);
 373                 if (next >= start_segno + sbi->segs_per_sec) {
 374                         if (test_and_clear_bit(secno, free_i->free_secmap))
 375                                 free_i->free_sections++;
 376                 }
 377         }
 378         spin_unlock(&free_i->segmap_lock);
 379 }
 380
 381 static inline void __set_test_and_inuse(struct f2fs_sb_info *sbi,
 382                 unsigned int segno)
 383 {
 384         struct free_segmap_info *free_i = FREE_I(sbi);
 385         unsigned int secno = segno / sbi->segs_per_sec;
 386         spin_lock(&free_i->segmap_lock);
 387         if (!test_and_set_bit(segno, free_i->free_segmap)) {
 388                 free_i->free_segments--;
 389                 if (!test_and_set_bit(secno, free_i->free_secmap))
 390                         free_i->free_sections--;
 391         }
 392         spin_unlock(&free_i->segmap_lock);
 393 }
 394
 395 static inline void get_sit_bitmap(struct f2fs_sb_info *sbi,
 396                 void *dst_addr)
 397 {
 398         struct sit_info *sit_i = SIT_I(sbi);
 399         memcpy(dst_addr, sit_i->sit_bitmap, sit_i->bitmap_size);
 400 }
 401
 402 static inline block_t written_block_count(struct f2fs_sb_info *sbi)
 403 {
 404         return SIT_I(sbi)->written_valid_blocks;
 405 }
 406
 407 static inline unsigned int free_segments(struct f2fs_sb_info *sbi)
 408 {
 409         return FREE_I(sbi)->free_segments;
 410 }
 411
 412 static inline int reserved_segments(struct f2fs_sb_info *sbi)
 413 {
 414         return SM_I(sbi)->reserved_segments;
 415 }
 416
 417 static inline unsigned int free_sections(struct f2fs_sb_info *sbi)
 418 {
 419         return FREE_I(sbi)->free_sections;
 420 }
 421
 422 static inline unsigned int prefree_segments(struct f2fs_sb_info *sbi)
 423 {
 424         return DIRTY_I(sbi)->nr_dirty[PRE];
 425 }
 426
 427 static inline unsigned int dirty_segments(struct f2fs_sb_info *sbi)
 428 {
 429         return DIRTY_I(sbi)->nr_dirty[DIRTY_HOT_DATA] +
 430                 DIRTY_I(sbi)->nr_dirty[DIRTY_WARM_DATA] +
 431                 DIRTY_I(sbi)->nr_dirty[DIRTY_COLD_DATA] +
 432                 DIRTY_I(sbi)->nr_dirty[DIRTY_HOT_NODE] +
 433                 DIRTY_I(sbi)->nr_dirty[DIRTY_WARM_NODE] +
 434                 DIRTY_I(sbi)->nr_dirty[DIRTY_COLD_NODE];
 435 }
 436
 437 static inline int overprovision_segments(struct f2fs_sb_info *sbi)
 438 {
 439         return SM_I(sbi)->ovp_segments;
 440 }
 441
 442 static inline int overprovision_sections(struct f2fs_sb_info *sbi)
 443 {
 444         return ((unsigned int) overprovision_segments(sbi)) / sbi->segs_per_sec;
 445 }
 446
 447 static inline int reserved_sections(struct f2fs_sb_info *sbi)
 448 {
 449         return ((unsigned int) reserved_segments(sbi)) / sbi->segs_per_sec;
 450 }
 451
 452 static inline bool need_SSR(struct f2fs_sb_info *sbi)
 453 {
 454         int node_secs = get_blocktype_secs(sbi, F2FS_DIRTY_NODES);
 455         int dent_secs = get_blocktype_secs(sbi, F2FS_DIRTY_DENTS);
 456         return free_sections(sbi) <= (node_secs + 2 * dent_secs +
 457                                                 reserved_sections(sbi) + 1);
 458 }
 459
 460 static inline bool has_not_enough_free_secs(struct f2fs_sb_info *sbi, int freed)
 461 {
 462         int node_secs = get_blocktype_secs(sbi, F2FS_DIRTY_NODES);
 463         int dent_secs = get_blocktype_secs(sbi, F2FS_DIRTY_DENTS);
 464
 465         if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
 466                 return false;
 467
 468         return (free_sections(sbi) + freed) <= (node_secs + 2 * dent_secs +
 469                                                 reserved_sections(sbi));
 470 }
 471
 472 static inline bool excess_prefree_segs(struct f2fs_sb_info *sbi)
 473 {
 474         return prefree_segments(sbi) > SM_I(sbi)->rec_prefree_segments;
 475 }
 476
 477 static inline int utilization(struct f2fs_sb_info *sbi)
 478 {
 479         return div_u64((u64)valid_user_blocks(sbi) * 100,
 480                                         sbi->user_block_count);
 481 }
 482
 483 /*
 484  * Sometimes f2fs may be better to drop out-of-place update policy.
 485  * And, users can control the policy through sysfs entries.
 486  * There are five policies with triggering conditions as follows.
 487  * F2FS_IPU_FORCE - all the time,
 488  * F2FS_IPU_SSR - if SSR mode is activated,
 489  * F2FS_IPU_UTIL - if FS utilization is over threashold,
 490  * F2FS_IPU_SSR_UTIL - if SSR mode is activated and FS utilization is over
 491  *                     threashold,
 492  * F2FS_IPU_FSYNC - activated in fsync path only for high performance flash
 493  *                     storages. IPU will be triggered only if the # of dirty
 494  *                     pages over min_fsync_blocks.
 495  * F2FS_IPUT_DISABLE - disable IPU. (=default option)
 496  */
 497 #define DEF_MIN_IPU_UTIL        70
 498 #define DEF_MIN_FSYNC_BLOCKS    8
 499
 500 enum {
 501         F2FS_IPU_FORCE,
 502         F2FS_IPU_SSR,
 503         F2FS_IPU_UTIL,
 504         F2FS_IPU_SSR_UTIL,
 505         F2FS_IPU_FSYNC,
 506 };
 507
 508 static inline bool need_inplace_update(struct inode *inode)
 509 {
 510         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 511         unsigned int policy = SM_I(sbi)->ipu_policy;
 512
 513         /* IPU can be done only for the user data */
 514         if (S_ISDIR(inode->i_mode) || f2fs_is_atomic_file(inode))
 515                 return false;
 516
 517         if (policy & (0x1 << F2FS_IPU_FORCE))
 518                 return true;
 519         if (policy & (0x1 << F2FS_IPU_SSR) && need_SSR(sbi))
 520                 return true;
 521         if (policy & (0x1 << F2FS_IPU_UTIL) &&
 522                         utilization(sbi) > SM_I(sbi)->min_ipu_util)
 523                 return true;
 524         if (policy & (0x1 << F2FS_IPU_SSR_UTIL) && need_SSR(sbi) &&
 525                         utilization(sbi) > SM_I(sbi)->min_ipu_util)
 526                 return true;
 527
 528         /* this is only set during fdatasync */
 529         if (policy & (0x1 << F2FS_IPU_FSYNC) &&
 530                         is_inode_flag_set(F2FS_I(inode), FI_NEED_IPU))
 531                 return true;
 532
 533         return false;
 534 }
 535
 536 static inline unsigned int curseg_segno(struct f2fs_sb_info *sbi,
 537                 int type)
 538 {
 539         struct curseg_info *curseg = CURSEG_I(sbi, type);
 540         return curseg->segno;
 541 }
 542
 543 static inline unsigned char curseg_alloc_type(struct f2fs_sb_info *sbi,
 544                 int type)
 545 {
 546         struct curseg_info *curseg = CURSEG_I(sbi, type);
 547         return curseg->alloc_type;
 548 }
 549
 550 static inline unsigned short curseg_blkoff(struct f2fs_sb_info *sbi, int type)
 551 {
 552         struct curseg_info *curseg = CURSEG_I(sbi, type);
 553         return curseg->next_blkoff;
 554 }
 555
 556 #ifdef CONFIG_F2FS_CHECK_FS
 557 static inline void check_seg_range(struct f2fs_sb_info *sbi, unsigned int segno)
 558 {
 559         BUG_ON(segno > TOTAL_SEGS(sbi) - 1);
 560 }
 561
 562 static inline void verify_block_addr(struct f2fs_sb_info *sbi, block_t blk_addr)
 563 {
 564         BUG_ON(blk_addr < SEG0_BLKADDR(sbi));
 565         BUG_ON(blk_addr >= MAX_BLKADDR(sbi));
 566 }
 567
 568 /*
 569  * Summary block is always treated as an invalid block
 570  */
 571 static inline void check_block_count(struct f2fs_sb_info *sbi,
 572                 int segno, struct f2fs_sit_entry *raw_sit)
 573 {
 574         bool is_valid  = test_bit_le(0, raw_sit->valid_map) ? true : false;
 575         int valid_blocks = 0;
 576         int cur_pos = 0, next_pos;
 577
 578         /* check segment usage */
 579         BUG_ON(GET_SIT_VBLOCKS(raw_sit) > sbi->blocks_per_seg);
 580
 581         /* check boundary of a given segment number */
 582         BUG_ON(segno > TOTAL_SEGS(sbi) - 1);
 583
 584         /* check bitmap with valid block count */
 585         do {
 586                 if (is_valid) {
 587                         next_pos = find_next_zero_bit_le(&raw_sit->valid_map,
 588                                         sbi->blocks_per_seg,
 589                                         cur_pos);
 590                         valid_blocks += next_pos - cur_pos;
 591                 } else
 592                         next_pos = find_next_bit_le(&raw_sit->valid_map,
 593                                         sbi->blocks_per_seg,
 594                                         cur_pos);
 595                 cur_pos = next_pos;
 596                 is_valid = !is_valid;
 597         } while (cur_pos < sbi->blocks_per_seg);
 598         BUG_ON(GET_SIT_VBLOCKS(raw_sit) != valid_blocks);
 599 }
 600 #else
 601 static inline void check_seg_range(struct f2fs_sb_info *sbi, unsigned int segno)
 602 {
 603         if (segno > TOTAL_SEGS(sbi) - 1)
 604                 set_sbi_flag(sbi, SBI_NEED_FSCK);
 605 }
 606
 607 static inline void verify_block_addr(struct f2fs_sb_info *sbi, block_t blk_addr)
 608 {
 609         if (blk_addr < SEG0_BLKADDR(sbi) || blk_addr >= MAX_BLKADDR(sbi))
 610                 set_sbi_flag(sbi, SBI_NEED_FSCK);
 611 }
 612
 613 /*
 614  * Summary block is always treated as an invalid block
 615  */
 616 static inline void check_block_count(struct f2fs_sb_info *sbi,
 617                 int segno, struct f2fs_sit_entry *raw_sit)
 618 {
 619         /* check segment usage */
 620         if (GET_SIT_VBLOCKS(raw_sit) > sbi->blocks_per_seg)
 621                 set_sbi_flag(sbi, SBI_NEED_FSCK);
 622
 623         /* check boundary of a given segment number */
 624         if (segno > TOTAL_SEGS(sbi) - 1)
 625                 set_sbi_flag(sbi, SBI_NEED_FSCK);
 626 }
 627 #endif
 628
 629 static inline pgoff_t current_sit_addr(struct f2fs_sb_info *sbi,
 630                                                 unsigned int start)
 631 {
 632         struct sit_info *sit_i = SIT_I(sbi);
 633         unsigned int offset = SIT_BLOCK_OFFSET(start);
 634         block_t blk_addr = sit_i->sit_base_addr + offset;
 635
 636         check_seg_range(sbi, start);
 637
 638         /* calculate sit block address */
 639         if (f2fs_test_bit(offset, sit_i->sit_bitmap))
 640                 blk_addr += sit_i->sit_blocks;
 641
 642         return blk_addr;
 643 }
 644
 645 static inline pgoff_t next_sit_addr(struct f2fs_sb_info *sbi,
 646                                                 pgoff_t block_addr)
 647 {
 648         struct sit_info *sit_i = SIT_I(sbi);
 649         block_addr -= sit_i->sit_base_addr;
 650         if (block_addr < sit_i->sit_blocks)
 651                 block_addr += sit_i->sit_blocks;
 652         else
 653                 block_addr -= sit_i->sit_blocks;
 654
 655         return block_addr + sit_i->sit_base_addr;
 656 }
 657
 658 static inline void set_to_next_sit(struct sit_info *sit_i, unsigned int start)
 659 {
 660         unsigned int block_off = SIT_BLOCK_OFFSET(start);
 661
 662         f2fs_change_bit(block_off, sit_i->sit_bitmap);
 663 }
 664
 665 static inline unsigned long long get_mtime(struct f2fs_sb_info *sbi)
 666 {
 667         struct sit_info *sit_i = SIT_I(sbi);
 668         return sit_i->elapsed_time + CURRENT_TIME_SEC.tv_sec -
 669                                                 sit_i->mounted_time;
 670 }
 671
 672 static inline void set_summary(struct f2fs_summary *sum, nid_t nid,
 673                         unsigned int ofs_in_node, unsigned char version)
 674 {
 675         sum->nid = cpu_to_le32(nid);
 676         sum->ofs_in_node = cpu_to_le16(ofs_in_node);
 677         sum->version = version;
 678 }
 679
 680 static inline block_t start_sum_block(struct f2fs_sb_info *sbi)
 681 {
 682         return __start_cp_addr(sbi) +
 683                 le32_to_cpu(F2FS_CKPT(sbi)->cp_pack_start_sum);
 684 }
 685
 686 static inline block_t sum_blk_addr(struct f2fs_sb_info *sbi, int base, int type)
 687 {
 688         return __start_cp_addr(sbi) +
 689                 le32_to_cpu(F2FS_CKPT(sbi)->cp_pack_total_block_count)
 690                                 - (base + 1) + type;
 691 }
 692
 693 static inline bool sec_usage_check(struct f2fs_sb_info *sbi, unsigned int secno)
 694 {
 695         if (IS_CURSEC(sbi, secno) || (sbi->cur_victim_sec == secno))
 696                 return true;
 697         return false;
 698 }
 699
 700 static inline unsigned int max_hw_blocks(struct f2fs_sb_info *sbi)
 701 {
 702         struct block_device *bdev = sbi->sb->s_bdev;
 703         struct request_queue *q = bdev_get_queue(bdev);
 704         return SECTOR_TO_BLOCK(queue_max_sectors(q));
 705 }
 706
 707 /*
 708  * It is very important to gather dirty pages and write at once, so that we can
 709  * submit a big bio without interfering other data writes.
 710  * By default, 512 pages for directory data,
 711  * 512 pages (2MB) * 3 for three types of nodes, and
 712  * max_bio_blocks for meta are set.
 713  */
 714 static inline int nr_pages_to_skip(struct f2fs_sb_info *sbi, int type)
 715 {
 716         if (sbi->sb->s_bdi->dirty_exceeded)
 717                 return 0;
 718
 719         if (type == DATA)
 720                 return sbi->blocks_per_seg;
 721         else if (type == NODE)
 722                 return 3 * sbi->blocks_per_seg;
 723         else if (type == META)
 724                 return MAX_BIO_BLOCKS(sbi);
 725         else
 726                 return 0;
 727 }
 728
 729 /*
 730  * When writing pages, it'd better align nr_to_write for segment size.
 731  */
 732 static inline long nr_pages_to_write(struct f2fs_sb_info *sbi, int type,
 733                                         struct writeback_control *wbc)
 734 {
 735         long nr_to_write, desired;
 736
 737         if (wbc->sync_mode != WB_SYNC_NONE)
 738                 return 0;
 739
 740         nr_to_write = wbc->nr_to_write;
 741
 742         if (type == DATA)
 743                 desired = 4096;
 744         else if (type == NODE)
 745                 desired = 3 * max_hw_blocks(sbi);
 746         else
 747                 desired = MAX_BIO_BLOCKS(sbi);
 748
 749         wbc->nr_to_write = desired;
 750         return desired - nr_to_write;
 751 }