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             0x0000ffff
 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 };
 195
 196 struct sit_info {
 197         const struct segment_allocation *s_ops;
 198
 199         block_t sit_base_addr;          /* start block address of SIT area */
 200         block_t sit_blocks;             /* # of blocks used by SIT area */
 201         block_t written_valid_blocks;   /* # of valid blocks in main area */
 202         char *sit_bitmap;               /* SIT bitmap pointer */
 203         unsigned int bitmap_size;       /* SIT bitmap size */
 204
 205         unsigned long *tmp_map;                 /* bitmap for temporal use */
 206         unsigned long *dirty_sentries_bitmap;   /* bitmap for dirty sentries */
 207         unsigned int dirty_sentries;            /* # of dirty sentries */
 208         unsigned int sents_per_block;           /* # of SIT entries per block */
 209         struct mutex sentry_lock;               /* to protect SIT cache */
 210         struct seg_entry *sentries;             /* SIT segment-level cache */
 211         struct sec_entry *sec_entries;          /* SIT section-level cache */
 212
 213         /* for cost-benefit algorithm in cleaning procedure */
 214         unsigned long long elapsed_time;        /* elapsed time after mount */
 215         unsigned long long mounted_time;        /* mount time */
 216         unsigned long long min_mtime;           /* min. modification time */
 217         unsigned long long max_mtime;           /* max. modification time */
 218 };
 219
 220 struct free_segmap_info {
 221         unsigned int start_segno;       /* start segment number logically */
 222         unsigned int free_segments;     /* # of free segments */
 223         unsigned int free_sections;     /* # of free sections */
 224         spinlock_t segmap_lock;         /* free segmap lock */
 225         unsigned long *free_segmap;     /* free segment bitmap */
 226         unsigned long *free_secmap;     /* free section bitmap */
 227 };
 228
 229 /* Notice: The order of dirty type is same with CURSEG_XXX in f2fs.h */
 230 enum dirty_type {
 231         DIRTY_HOT_DATA,         /* dirty segments assigned as hot data logs */
 232         DIRTY_WARM_DATA,        /* dirty segments assigned as warm data logs */
 233         DIRTY_COLD_DATA,        /* dirty segments assigned as cold data logs */
 234         DIRTY_HOT_NODE,         /* dirty segments assigned as hot node logs */
 235         DIRTY_WARM_NODE,        /* dirty segments assigned as warm node logs */
 236         DIRTY_COLD_NODE,        /* dirty segments assigned as cold node logs */
 237         DIRTY,                  /* to count # of dirty segments */
 238         PRE,                    /* to count # of entirely obsolete segments */
 239         NR_DIRTY_TYPE
 240 };
 241
 242 struct dirty_seglist_info {
 243         const struct victim_selection *v_ops;   /* victim selction operation */
 244         unsigned long *dirty_segmap[NR_DIRTY_TYPE];
 245         struct mutex seglist_lock;              /* lock for segment bitmaps */
 246         int nr_dirty[NR_DIRTY_TYPE];            /* # of dirty segments */
 247         unsigned long *victim_secmap;           /* background GC victims */
 248 };
 249
 250 /* victim selection function for cleaning and SSR */
 251 struct victim_selection {
 252         int (*get_victim)(struct f2fs_sb_info *, unsigned int *,
 253                                                         int, int, char);
 254 };
 255
 256 /* for active log information */
 257 struct curseg_info {
 258         struct mutex curseg_mutex;              /* lock for consistency */
 259         struct f2fs_summary_block *sum_blk;     /* cached summary block */
 260         unsigned char alloc_type;               /* current allocation type */
 261         unsigned int segno;                     /* current segment number */
 262         unsigned short next_blkoff;             /* next block offset to write */
 263         unsigned int zone;                      /* current zone number */
 264         unsigned int next_segno;                /* preallocated segment */
 265 };
 266
 267 struct sit_entry_set {
 268         struct list_head set_list;      /* link with all sit sets */
 269         unsigned int start_segno;       /* start segno of sits in set */
 270         unsigned int entry_cnt;         /* the # of sit entries in set */
 271 };
 272
 273 /*
 274  * inline functions
 275  */
 276 static inline struct curseg_info *CURSEG_I(struct f2fs_sb_info *sbi, int type)
 277 {
 278         return (struct curseg_info *)(SM_I(sbi)->curseg_array + type);
 279 }
 280
 281 static inline struct seg_entry *get_seg_entry(struct f2fs_sb_info *sbi,
 282                                                 unsigned int segno)
 283 {
 284         struct sit_info *sit_i = SIT_I(sbi);
 285         return &sit_i->sentries[segno];
 286 }
 287
 288 static inline struct sec_entry *get_sec_entry(struct f2fs_sb_info *sbi,
 289                                                 unsigned int segno)
 290 {
 291         struct sit_info *sit_i = SIT_I(sbi);
 292         return &sit_i->sec_entries[GET_SECNO(sbi, segno)];
 293 }
 294
 295 static inline unsigned int get_valid_blocks(struct f2fs_sb_info *sbi,
 296                                 unsigned int segno, int section)
 297 {
 298         /*
 299          * In order to get # of valid blocks in a section instantly from many
 300          * segments, f2fs manages two counting structures separately.
 301          */
 302         if (section > 1)
 303                 return get_sec_entry(sbi, segno)->valid_blocks;
 304         else
 305                 return get_seg_entry(sbi, segno)->valid_blocks;
 306 }
 307
 308 static inline void seg_info_from_raw_sit(struct seg_entry *se,
 309                                         struct f2fs_sit_entry *rs)
 310 {
 311         se->valid_blocks = GET_SIT_VBLOCKS(rs);
 312         se->ckpt_valid_blocks = GET_SIT_VBLOCKS(rs);
 313         memcpy(se->cur_valid_map, rs->valid_map, SIT_VBLOCK_MAP_SIZE);
 314         memcpy(se->ckpt_valid_map, rs->valid_map, SIT_VBLOCK_MAP_SIZE);
 315         se->type = GET_SIT_TYPE(rs);
 316         se->mtime = le64_to_cpu(rs->mtime);
 317 }
 318
 319 static inline void seg_info_to_raw_sit(struct seg_entry *se,
 320                                         struct f2fs_sit_entry *rs)
 321 {
 322         unsigned short raw_vblocks = (se->type << SIT_VBLOCKS_SHIFT) |
 323                                         se->valid_blocks;
 324         rs->vblocks = cpu_to_le16(raw_vblocks);
 325         memcpy(rs->valid_map, se->cur_valid_map, SIT_VBLOCK_MAP_SIZE);
 326         memcpy(se->ckpt_valid_map, rs->valid_map, SIT_VBLOCK_MAP_SIZE);
 327         se->ckpt_valid_blocks = se->valid_blocks;
 328         rs->mtime = cpu_to_le64(se->mtime);
 329 }
 330
 331 static inline unsigned int find_next_inuse(struct free_segmap_info *free_i,
 332                 unsigned int max, unsigned int segno)
 333 {
 334         unsigned int ret;
 335         spin_lock(&free_i->segmap_lock);
 336         ret = find_next_bit(free_i->free_segmap, max, segno);
 337         spin_unlock(&free_i->segmap_lock);
 338         return ret;
 339 }
 340
 341 static inline void __set_free(struct f2fs_sb_info *sbi, unsigned int segno)
 342 {
 343         struct free_segmap_info *free_i = FREE_I(sbi);
 344         unsigned int secno = segno / sbi->segs_per_sec;
 345         unsigned int start_segno = secno * sbi->segs_per_sec;
 346         unsigned int next;
 347
 348         spin_lock(&free_i->segmap_lock);
 349         clear_bit(segno, free_i->free_segmap);
 350         free_i->free_segments++;
 351
 352         next = find_next_bit(free_i->free_segmap,
 353                         start_segno + sbi->segs_per_sec, start_segno);
 354         if (next >= start_segno + sbi->segs_per_sec) {
 355                 clear_bit(secno, free_i->free_secmap);
 356                 free_i->free_sections++;
 357         }
 358         spin_unlock(&free_i->segmap_lock);
 359 }
 360
 361 static inline void __set_inuse(struct f2fs_sb_info *sbi,
 362                 unsigned int segno)
 363 {
 364         struct free_segmap_info *free_i = FREE_I(sbi);
 365         unsigned int secno = segno / sbi->segs_per_sec;
 366         set_bit(segno, free_i->free_segmap);
 367         free_i->free_segments--;
 368         if (!test_and_set_bit(secno, free_i->free_secmap))
 369                 free_i->free_sections--;
 370 }
 371
 372 static inline void __set_test_and_free(struct f2fs_sb_info *sbi,
 373                 unsigned int segno)
 374 {
 375         struct free_segmap_info *free_i = FREE_I(sbi);
 376         unsigned int secno = segno / sbi->segs_per_sec;
 377         unsigned int start_segno = secno * sbi->segs_per_sec;
 378         unsigned int next;
 379
 380         spin_lock(&free_i->segmap_lock);
 381         if (test_and_clear_bit(segno, free_i->free_segmap)) {
 382                 free_i->free_segments++;
 383
 384                 next = find_next_bit(free_i->free_segmap,
 385                                 start_segno + sbi->segs_per_sec, start_segno);
 386                 if (next >= start_segno + sbi->segs_per_sec) {
 387                         if (test_and_clear_bit(secno, free_i->free_secmap))
 388                                 free_i->free_sections++;
 389                 }
 390         }
 391         spin_unlock(&free_i->segmap_lock);
 392 }
 393
 394 static inline void __set_test_and_inuse(struct f2fs_sb_info *sbi,
 395                 unsigned int segno)
 396 {
 397         struct free_segmap_info *free_i = FREE_I(sbi);
 398         unsigned int secno = segno / sbi->segs_per_sec;
 399         spin_lock(&free_i->segmap_lock);
 400         if (!test_and_set_bit(segno, free_i->free_segmap)) {
 401                 free_i->free_segments--;
 402                 if (!test_and_set_bit(secno, free_i->free_secmap))
 403                         free_i->free_sections--;
 404         }
 405         spin_unlock(&free_i->segmap_lock);
 406 }
 407
 408 static inline void get_sit_bitmap(struct f2fs_sb_info *sbi,
 409                 void *dst_addr)
 410 {
 411         struct sit_info *sit_i = SIT_I(sbi);
 412         memcpy(dst_addr, sit_i->sit_bitmap, sit_i->bitmap_size);
 413 }
 414
 415 static inline block_t written_block_count(struct f2fs_sb_info *sbi)
 416 {
 417         return SIT_I(sbi)->written_valid_blocks;
 418 }
 419
 420 static inline unsigned int free_segments(struct f2fs_sb_info *sbi)
 421 {
 422         return FREE_I(sbi)->free_segments;
 423 }
 424
 425 static inline int reserved_segments(struct f2fs_sb_info *sbi)
 426 {
 427         return SM_I(sbi)->reserved_segments;
 428 }
 429
 430 static inline unsigned int free_sections(struct f2fs_sb_info *sbi)
 431 {
 432         return FREE_I(sbi)->free_sections;
 433 }
 434
 435 static inline unsigned int prefree_segments(struct f2fs_sb_info *sbi)
 436 {
 437         return DIRTY_I(sbi)->nr_dirty[PRE];
 438 }
 439
 440 static inline unsigned int dirty_segments(struct f2fs_sb_info *sbi)
 441 {
 442         return DIRTY_I(sbi)->nr_dirty[DIRTY_HOT_DATA] +
 443                 DIRTY_I(sbi)->nr_dirty[DIRTY_WARM_DATA] +
 444                 DIRTY_I(sbi)->nr_dirty[DIRTY_COLD_DATA] +
 445                 DIRTY_I(sbi)->nr_dirty[DIRTY_HOT_NODE] +
 446                 DIRTY_I(sbi)->nr_dirty[DIRTY_WARM_NODE] +
 447                 DIRTY_I(sbi)->nr_dirty[DIRTY_COLD_NODE];
 448 }
 449
 450 static inline int overprovision_segments(struct f2fs_sb_info *sbi)
 451 {
 452         return SM_I(sbi)->ovp_segments;
 453 }
 454
 455 static inline int overprovision_sections(struct f2fs_sb_info *sbi)
 456 {
 457         return ((unsigned int) overprovision_segments(sbi)) / sbi->segs_per_sec;
 458 }
 459
 460 static inline int reserved_sections(struct f2fs_sb_info *sbi)
 461 {
 462         return ((unsigned int) reserved_segments(sbi)) / sbi->segs_per_sec;
 463 }
 464
 465 static inline bool need_SSR(struct f2fs_sb_info *sbi)
 466 {
 467         int node_secs = get_blocktype_secs(sbi, F2FS_DIRTY_NODES);
 468         int dent_secs = get_blocktype_secs(sbi, F2FS_DIRTY_DENTS);
 469         return free_sections(sbi) <= (node_secs + 2 * dent_secs +
 470                                                 reserved_sections(sbi) + 1);
 471 }
 472
 473 static inline bool has_not_enough_free_secs(struct f2fs_sb_info *sbi, int freed)
 474 {
 475         int node_secs = get_blocktype_secs(sbi, F2FS_DIRTY_NODES);
 476         int dent_secs = get_blocktype_secs(sbi, F2FS_DIRTY_DENTS);
 477
 478         if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
 479                 return false;
 480
 481         return (free_sections(sbi) + freed) <= (node_secs + 2 * dent_secs +
 482                                                 reserved_sections(sbi));
 483 }
 484
 485 static inline bool excess_prefree_segs(struct f2fs_sb_info *sbi)
 486 {
 487         return prefree_segments(sbi) > SM_I(sbi)->rec_prefree_segments;
 488 }
 489
 490 static inline int utilization(struct f2fs_sb_info *sbi)
 491 {
 492         return div_u64((u64)valid_user_blocks(sbi) * 100,
 493                                         sbi->user_block_count);
 494 }
 495
 496 /*
 497  * Sometimes f2fs may be better to drop out-of-place update policy.
 498  * And, users can control the policy through sysfs entries.
 499  * There are five policies with triggering conditions as follows.
 500  * F2FS_IPU_FORCE - all the time,
 501  * F2FS_IPU_SSR - if SSR mode is activated,
 502  * F2FS_IPU_UTIL - if FS utilization is over threashold,
 503  * F2FS_IPU_SSR_UTIL - if SSR mode is activated and FS utilization is over
 504  *                     threashold,
 505  * F2FS_IPU_FSYNC - activated in fsync path only for high performance flash
 506  *                     storages. IPU will be triggered only if the # of dirty
 507  *                     pages over min_fsync_blocks.
 508  * F2FS_IPUT_DISABLE - disable IPU. (=default option)
 509  */
 510 #define DEF_MIN_IPU_UTIL        70
 511 #define DEF_MIN_FSYNC_BLOCKS    8
 512
 513 enum {
 514         F2FS_IPU_FORCE,
 515         F2FS_IPU_SSR,
 516         F2FS_IPU_UTIL,
 517         F2FS_IPU_SSR_UTIL,
 518         F2FS_IPU_FSYNC,
 519 };
 520
 521 static inline bool need_inplace_update(struct inode *inode)
 522 {
 523         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 524         unsigned int policy = SM_I(sbi)->ipu_policy;
 525
 526         /* IPU can be done only for the user data */
 527         if (S_ISDIR(inode->i_mode) || f2fs_is_atomic_file(inode))
 528                 return false;
 529
 530         if (policy & (0x1 << F2FS_IPU_FORCE))
 531                 return true;
 532         if (policy & (0x1 << F2FS_IPU_SSR) && need_SSR(sbi))
 533                 return true;
 534         if (policy & (0x1 << F2FS_IPU_UTIL) &&
 535                         utilization(sbi) > SM_I(sbi)->min_ipu_util)
 536                 return true;
 537         if (policy & (0x1 << F2FS_IPU_SSR_UTIL) && need_SSR(sbi) &&
 538                         utilization(sbi) > SM_I(sbi)->min_ipu_util)
 539                 return true;
 540
 541         /* this is only set during fdatasync */
 542         if (policy & (0x1 << F2FS_IPU_FSYNC) &&
 543                         is_inode_flag_set(F2FS_I(inode), FI_NEED_IPU))
 544                 return true;
 545
 546         return false;
 547 }
 548
 549 static inline unsigned int curseg_segno(struct f2fs_sb_info *sbi,
 550                 int type)
 551 {
 552         struct curseg_info *curseg = CURSEG_I(sbi, type);
 553         return curseg->segno;
 554 }
 555
 556 static inline unsigned char curseg_alloc_type(struct f2fs_sb_info *sbi,
 557                 int type)
 558 {
 559         struct curseg_info *curseg = CURSEG_I(sbi, type);
 560         return curseg->alloc_type;
 561 }
 562
 563 static inline unsigned short curseg_blkoff(struct f2fs_sb_info *sbi, int type)
 564 {
 565         struct curseg_info *curseg = CURSEG_I(sbi, type);
 566         return curseg->next_blkoff;
 567 }
 568
 569 static inline void check_seg_range(struct f2fs_sb_info *sbi, unsigned int segno)
 570 {
 571         f2fs_bug_on(sbi, segno > TOTAL_SEGS(sbi) - 1);
 572 }
 573
 574 static inline void verify_block_addr(struct f2fs_sb_info *sbi, block_t blk_addr)
 575 {
 576         f2fs_bug_on(sbi, blk_addr < SEG0_BLKADDR(sbi)
 577                                         || blk_addr >= MAX_BLKADDR(sbi));
 578 }
 579
 580 /*
 581  * Summary block is always treated as an invalid block
 582  */
 583 static inline void check_block_count(struct f2fs_sb_info *sbi,
 584                 int segno, struct f2fs_sit_entry *raw_sit)
 585 {
 586 #ifdef CONFIG_F2FS_CHECK_FS
 587         bool is_valid  = test_bit_le(0, raw_sit->valid_map) ? true : false;
 588         int valid_blocks = 0;
 589         int cur_pos = 0, next_pos;
 590
 591         /* check bitmap with valid block count */
 592         do {
 593                 if (is_valid) {
 594                         next_pos = find_next_zero_bit_le(&raw_sit->valid_map,
 595                                         sbi->blocks_per_seg,
 596                                         cur_pos);
 597                         valid_blocks += next_pos - cur_pos;
 598                 } else
 599                         next_pos = find_next_bit_le(&raw_sit->valid_map,
 600                                         sbi->blocks_per_seg,
 601                                         cur_pos);
 602                 cur_pos = next_pos;
 603                 is_valid = !is_valid;
 604         } while (cur_pos < sbi->blocks_per_seg);
 605         BUG_ON(GET_SIT_VBLOCKS(raw_sit) != valid_blocks);
 606 #endif
 607         /* check segment usage, and check boundary of a given segment number */
 608         f2fs_bug_on(sbi, GET_SIT_VBLOCKS(raw_sit) > sbi->blocks_per_seg
 609                                         || segno > TOTAL_SEGS(sbi) - 1);
 610 }
 611
 612 static inline pgoff_t current_sit_addr(struct f2fs_sb_info *sbi,
 613                                                 unsigned int start)
 614 {
 615         struct sit_info *sit_i = SIT_I(sbi);
 616         unsigned int offset = SIT_BLOCK_OFFSET(start);
 617         block_t blk_addr = sit_i->sit_base_addr + offset;
 618
 619         check_seg_range(sbi, start);
 620
 621         /* calculate sit block address */
 622         if (f2fs_test_bit(offset, sit_i->sit_bitmap))
 623                 blk_addr += sit_i->sit_blocks;
 624
 625         return blk_addr;
 626 }
 627
 628 static inline pgoff_t next_sit_addr(struct f2fs_sb_info *sbi,
 629                                                 pgoff_t block_addr)
 630 {
 631         struct sit_info *sit_i = SIT_I(sbi);
 632         block_addr -= sit_i->sit_base_addr;
 633         if (block_addr < sit_i->sit_blocks)
 634                 block_addr += sit_i->sit_blocks;
 635         else
 636                 block_addr -= sit_i->sit_blocks;
 637
 638         return block_addr + sit_i->sit_base_addr;
 639 }
 640
 641 static inline void set_to_next_sit(struct sit_info *sit_i, unsigned int start)
 642 {
 643         unsigned int block_off = SIT_BLOCK_OFFSET(start);
 644
 645         f2fs_change_bit(block_off, sit_i->sit_bitmap);
 646 }
 647
 648 static inline unsigned long long get_mtime(struct f2fs_sb_info *sbi)
 649 {
 650         struct sit_info *sit_i = SIT_I(sbi);
 651         return sit_i->elapsed_time + CURRENT_TIME_SEC.tv_sec -
 652                                                 sit_i->mounted_time;
 653 }
 654
 655 static inline void set_summary(struct f2fs_summary *sum, nid_t nid,
 656                         unsigned int ofs_in_node, unsigned char version)
 657 {
 658         sum->nid = cpu_to_le32(nid);
 659         sum->ofs_in_node = cpu_to_le16(ofs_in_node);
 660         sum->version = version;
 661 }
 662
 663 static inline block_t start_sum_block(struct f2fs_sb_info *sbi)
 664 {
 665         return __start_cp_addr(sbi) +
 666                 le32_to_cpu(F2FS_CKPT(sbi)->cp_pack_start_sum);
 667 }
 668
 669 static inline block_t sum_blk_addr(struct f2fs_sb_info *sbi, int base, int type)
 670 {
 671         return __start_cp_addr(sbi) +
 672                 le32_to_cpu(F2FS_CKPT(sbi)->cp_pack_total_block_count)
 673                                 - (base + 1) + type;
 674 }
 675
 676 static inline bool sec_usage_check(struct f2fs_sb_info *sbi, unsigned int secno)
 677 {
 678         if (IS_CURSEC(sbi, secno) || (sbi->cur_victim_sec == secno))
 679                 return true;
 680         return false;
 681 }
 682
 683 static inline unsigned int max_hw_blocks(struct f2fs_sb_info *sbi)
 684 {
 685         struct block_device *bdev = sbi->sb->s_bdev;
 686         struct request_queue *q = bdev_get_queue(bdev);
 687         return SECTOR_TO_BLOCK(queue_max_sectors(q));
 688 }
 689
 690 /*
 691  * It is very important to gather dirty pages and write at once, so that we can
 692  * submit a big bio without interfering other data writes.
 693  * By default, 512 pages for directory data,
 694  * 512 pages (2MB) * 3 for three types of nodes, and
 695  * max_bio_blocks for meta are set.
 696  */
 697 static inline int nr_pages_to_skip(struct f2fs_sb_info *sbi, int type)
 698 {
 699         if (sbi->sb->s_bdi->wb.dirty_exceeded)
 700                 return 0;
 701
 702         if (type == DATA)
 703                 return sbi->blocks_per_seg;
 704         else if (type == NODE)
 705                 return 3 * sbi->blocks_per_seg;
 706         else if (type == META)
 707                 return MAX_BIO_BLOCKS(sbi);
 708         else
 709                 return 0;
 710 }
 711
 712 /*
 713  * When writing pages, it'd better align nr_to_write for segment size.
 714  */
 715 static inline long nr_pages_to_write(struct f2fs_sb_info *sbi, int type,
 716                                         struct writeback_control *wbc)
 717 {
 718         long nr_to_write, desired;
 719
 720         if (wbc->sync_mode != WB_SYNC_NONE)
 721                 return 0;
 722
 723         nr_to_write = wbc->nr_to_write;
 724
 725         if (type == DATA)
 726                 desired = 4096;
 727         else if (type == NODE)
 728                 desired = 3 * max_hw_blocks(sbi);
 729         else
 730                 desired = MAX_BIO_BLOCKS(sbi);
 731
 732         wbc->nr_to_write = desired;
 733         return desired - nr_to_write;
 734 }