1 /* 2 * fs/f2fs/segment.h 3 * 4 * Copyright (c) 2012 Samsung Electronics Co., Ltd. 5 * http://www.samsung.com/ 6 * 7 * This program is free software; you can redistribute it and/or modify 8 * it under the terms of the GNU General Public License version 2 as 9 * published by the Free Software Foundation. 10 */ 11 #include <linux/blkdev.h> 12 #include <linux/backing-dev.h> 13 14 /* constant macro */ 15 #define NULL_SEGNO ((unsigned int)(~0)) 16 #define NULL_SECNO ((unsigned int)(~0)) 17 18 #define DEF_RECLAIM_PREFREE_SEGMENTS 5 /* 5% over total segments */ 19 #define DEF_MAX_RECLAIM_PREFREE_SEGMENTS 4096 /* 8GB in maximum */ 20 21 #define F2FS_MIN_SEGMENTS 9 /* SB + 2 (CP + SIT + NAT) + SSA + MAIN */ 22 23 /* L: Logical segment # in volume, R: Relative segment # in main area */ 24 #define GET_L2R_SEGNO(free_i, segno) (segno - free_i->start_segno) 25 #define GET_R2L_SEGNO(free_i, segno) (segno + free_i->start_segno) 26 27 #define IS_DATASEG(t) (t <= CURSEG_COLD_DATA) 28 #define IS_NODESEG(t) (t >= CURSEG_HOT_NODE) 29 30 #define IS_CURSEG(sbi, seg) \ 31 ((seg == CURSEG_I(sbi, CURSEG_HOT_DATA)->segno) || \ 32 (seg == CURSEG_I(sbi, CURSEG_WARM_DATA)->segno) || \ 33 (seg == CURSEG_I(sbi, CURSEG_COLD_DATA)->segno) || \ 34 (seg == CURSEG_I(sbi, CURSEG_HOT_NODE)->segno) || \ 35 (seg == CURSEG_I(sbi, CURSEG_WARM_NODE)->segno) || \ 36 (seg == CURSEG_I(sbi, CURSEG_COLD_NODE)->segno)) 37 38 #define IS_CURSEC(sbi, secno) \ 39 ((secno == CURSEG_I(sbi, CURSEG_HOT_DATA)->segno / \ 40 sbi->segs_per_sec) || \ 41 (secno == CURSEG_I(sbi, CURSEG_WARM_DATA)->segno / \ 42 sbi->segs_per_sec) || \ 43 (secno == CURSEG_I(sbi, CURSEG_COLD_DATA)->segno / \ 44 sbi->segs_per_sec) || \ 45 (secno == CURSEG_I(sbi, CURSEG_HOT_NODE)->segno / \ 46 sbi->segs_per_sec) || \ 47 (secno == CURSEG_I(sbi, CURSEG_WARM_NODE)->segno / \ 48 sbi->segs_per_sec) || \ 49 (secno == CURSEG_I(sbi, CURSEG_COLD_NODE)->segno / \ 50 sbi->segs_per_sec)) \ 51 52 #define MAIN_BLKADDR(sbi) (SM_I(sbi)->main_blkaddr) 53 #define SEG0_BLKADDR(sbi) (SM_I(sbi)->seg0_blkaddr) 54 55 #define MAIN_SEGS(sbi) (SM_I(sbi)->main_segments) 56 #define MAIN_SECS(sbi) (sbi->total_sections) 57 58 #define TOTAL_SEGS(sbi) (SM_I(sbi)->segment_count) 59 #define TOTAL_BLKS(sbi) (TOTAL_SEGS(sbi) << sbi->log_blocks_per_seg) 60 61 #define MAX_BLKADDR(sbi) (SEG0_BLKADDR(sbi) + TOTAL_BLKS(sbi)) 62 #define SEGMENT_SIZE(sbi) (1ULL << (sbi->log_blocksize + \ 63 sbi->log_blocks_per_seg)) 64 65 #define START_BLOCK(sbi, segno) (SEG0_BLKADDR(sbi) + \ 66 (GET_R2L_SEGNO(FREE_I(sbi), segno) << sbi->log_blocks_per_seg)) 67 68 #define NEXT_FREE_BLKADDR(sbi, curseg) \ 69 (START_BLOCK(sbi, curseg->segno) + curseg->next_blkoff) 70 71 #define GET_SEGOFF_FROM_SEG0(sbi, blk_addr) ((blk_addr) - SEG0_BLKADDR(sbi)) 72 #define GET_SEGNO_FROM_SEG0(sbi, blk_addr) \ 73 (GET_SEGOFF_FROM_SEG0(sbi, blk_addr) >> sbi->log_blocks_per_seg) 74 #define GET_BLKOFF_FROM_SEG0(sbi, blk_addr) \ 75 (GET_SEGOFF_FROM_SEG0(sbi, blk_addr) & (sbi->blocks_per_seg - 1)) 76 77 #define GET_SEGNO(sbi, blk_addr) \ 78 (((blk_addr == NULL_ADDR) || (blk_addr == NEW_ADDR)) ? \ 79 NULL_SEGNO : GET_L2R_SEGNO(FREE_I(sbi), \ 80 GET_SEGNO_FROM_SEG0(sbi, blk_addr))) 81 #define GET_SECNO(sbi, segno) \ 82 ((segno) / sbi->segs_per_sec) 83 #define GET_ZONENO_FROM_SEGNO(sbi, segno) \ 84 ((segno / sbi->segs_per_sec) / sbi->secs_per_zone) 85 86 #define GET_SUM_BLOCK(sbi, segno) \ 87 ((sbi->sm_info->ssa_blkaddr) + segno) 88 89 #define GET_SUM_TYPE(footer) ((footer)->entry_type) 90 #define SET_SUM_TYPE(footer, type) ((footer)->entry_type = type) 91 92 #define SIT_ENTRY_OFFSET(sit_i, segno) \ 93 (segno % sit_i->sents_per_block) 94 #define SIT_BLOCK_OFFSET(segno) \ 95 (segno / SIT_ENTRY_PER_BLOCK) 96 #define START_SEGNO(segno) \ 97 (SIT_BLOCK_OFFSET(segno) * SIT_ENTRY_PER_BLOCK) 98 #define SIT_BLK_CNT(sbi) \ 99 ((MAIN_SEGS(sbi) + SIT_ENTRY_PER_BLOCK - 1) / SIT_ENTRY_PER_BLOCK) 100 #define f2fs_bitmap_size(nr) \ 101 (BITS_TO_LONGS(nr) * sizeof(unsigned long)) 102 103 #define SECTOR_FROM_BLOCK(blk_addr) \ 104 (((sector_t)blk_addr) << F2FS_LOG_SECTORS_PER_BLOCK) 105 #define SECTOR_TO_BLOCK(sectors) \ 106 (sectors >> F2FS_LOG_SECTORS_PER_BLOCK) 107 108 /* 109 * indicate a block allocation direction: RIGHT and LEFT. 110 * RIGHT means allocating new sections towards the end of volume. 111 * LEFT means the opposite direction. 112 */ 113 enum { 114 ALLOC_RIGHT = 0, 115 ALLOC_LEFT 116 }; 117 118 /* 119 * In the victim_sel_policy->alloc_mode, there are two block allocation modes. 120 * LFS writes data sequentially with cleaning operations. 121 * SSR (Slack Space Recycle) reuses obsolete space without cleaning operations. 122 */ 123 enum { 124 LFS = 0, 125 SSR 126 }; 127 128 /* 129 * In the victim_sel_policy->gc_mode, there are two gc, aka cleaning, modes. 130 * GC_CB is based on cost-benefit algorithm. 131 * GC_GREEDY is based on greedy algorithm. 132 */ 133 enum { 134 GC_CB = 0, 135 GC_GREEDY 136 }; 137 138 /* 139 * BG_GC means the background cleaning job. 140 * FG_GC means the on-demand cleaning job. 141 * FORCE_FG_GC means on-demand cleaning job in background. 142 */ 143 enum { 144 BG_GC = 0, 145 FG_GC, 146 FORCE_FG_GC, 147 }; 148 149 /* for a function parameter to select a victim segment */ 150 struct victim_sel_policy { 151 int alloc_mode; /* LFS or SSR */ 152 int gc_mode; /* GC_CB or GC_GREEDY */ 153 unsigned long *dirty_segmap; /* dirty segment bitmap */ 154 unsigned int max_search; /* maximum # of segments to search */ 155 unsigned int offset; /* last scanned bitmap offset */ 156 unsigned int ofs_unit; /* bitmap search unit */ 157 unsigned int min_cost; /* minimum cost */ 158 unsigned int min_segno; /* segment # having min. cost */ 159 }; 160 161 struct seg_entry { 162 unsigned int type:6; /* segment type like CURSEG_XXX_TYPE */ 163 unsigned int valid_blocks:10; /* # of valid blocks */ 164 unsigned int ckpt_valid_blocks:10; /* # of valid blocks last cp */ 165 unsigned int padding:6; /* padding */ 166 unsigned char *cur_valid_map; /* validity bitmap of blocks */ 167 #ifdef CONFIG_F2FS_CHECK_FS 168 unsigned char *cur_valid_map_mir; /* mirror of current valid bitmap */ 169 #endif 170 /* 171 * # of valid blocks and the validity bitmap stored in the the last 172 * checkpoint pack. This information is used by the SSR mode. 173 */ 174 unsigned char *ckpt_valid_map; /* validity bitmap of blocks last cp */ 175 unsigned char *discard_map; 176 unsigned long long mtime; /* modification time of the segment */ 177 }; 178 179 struct sec_entry { 180 unsigned int valid_blocks; /* # of valid blocks in a section */ 181 }; 182 183 struct segment_allocation { 184 void (*allocate_segment)(struct f2fs_sb_info *, int, bool); 185 }; 186 187 /* 188 * this value is set in page as a private data which indicate that 189 * the page is atomically written, and it is in inmem_pages list. 190 */ 191 #define ATOMIC_WRITTEN_PAGE ((unsigned long)-1) 192 #define DUMMY_WRITTEN_PAGE ((unsigned long)-2) 193 194 #define IS_ATOMIC_WRITTEN_PAGE(page) \ 195 (page_private(page) == (unsigned long)ATOMIC_WRITTEN_PAGE) 196 #define IS_DUMMY_WRITTEN_PAGE(page) \ 197 (page_private(page) == (unsigned long)DUMMY_WRITTEN_PAGE) 198 199 struct inmem_pages { 200 struct list_head list; 201 struct page *page; 202 block_t old_addr; /* for revoking when fail to commit */ 203 }; 204 205 struct sit_info { 206 const struct segment_allocation *s_ops; 207 208 block_t sit_base_addr; /* start block address of SIT area */ 209 block_t sit_blocks; /* # of blocks used by SIT area */ 210 block_t written_valid_blocks; /* # of valid blocks in main area */ 211 char *sit_bitmap; /* SIT bitmap pointer */ 212 #ifdef CONFIG_F2FS_CHECK_FS 213 char *sit_bitmap_mir; /* SIT bitmap mirror */ 214 #endif 215 unsigned int bitmap_size; /* SIT bitmap size */ 216 217 unsigned long *tmp_map; /* bitmap for temporal use */ 218 unsigned long *dirty_sentries_bitmap; /* bitmap for dirty sentries */ 219 unsigned int dirty_sentries; /* # of dirty sentries */ 220 unsigned int sents_per_block; /* # of SIT entries per block */ 221 struct mutex sentry_lock; /* to protect SIT cache */ 222 struct seg_entry *sentries; /* SIT segment-level cache */ 223 struct sec_entry *sec_entries; /* SIT section-level cache */ 224 225 /* for cost-benefit algorithm in cleaning procedure */ 226 unsigned long long elapsed_time; /* elapsed time after mount */ 227 unsigned long long mounted_time; /* mount time */ 228 unsigned long long min_mtime; /* min. modification time */ 229 unsigned long long max_mtime; /* max. modification time */ 230 }; 231 232 struct free_segmap_info { 233 unsigned int start_segno; /* start segment number logically */ 234 unsigned int free_segments; /* # of free segments */ 235 unsigned int free_sections; /* # of free sections */ 236 spinlock_t segmap_lock; /* free segmap lock */ 237 unsigned long *free_segmap; /* free segment bitmap */ 238 unsigned long *free_secmap; /* free section bitmap */ 239 }; 240 241 /* Notice: The order of dirty type is same with CURSEG_XXX in f2fs.h */ 242 enum dirty_type { 243 DIRTY_HOT_DATA, /* dirty segments assigned as hot data logs */ 244 DIRTY_WARM_DATA, /* dirty segments assigned as warm data logs */ 245 DIRTY_COLD_DATA, /* dirty segments assigned as cold data logs */ 246 DIRTY_HOT_NODE, /* dirty segments assigned as hot node logs */ 247 DIRTY_WARM_NODE, /* dirty segments assigned as warm node logs */ 248 DIRTY_COLD_NODE, /* dirty segments assigned as cold node logs */ 249 DIRTY, /* to count # of dirty segments */ 250 PRE, /* to count # of entirely obsolete segments */ 251 NR_DIRTY_TYPE 252 }; 253 254 struct dirty_seglist_info { 255 const struct victim_selection *v_ops; /* victim selction operation */ 256 unsigned long *dirty_segmap[NR_DIRTY_TYPE]; 257 struct mutex seglist_lock; /* lock for segment bitmaps */ 258 int nr_dirty[NR_DIRTY_TYPE]; /* # of dirty segments */ 259 unsigned long *victim_secmap; /* background GC victims */ 260 }; 261 262 /* victim selection function for cleaning and SSR */ 263 struct victim_selection { 264 int (*get_victim)(struct f2fs_sb_info *, unsigned int *, 265 int, int, char); 266 }; 267 268 /* for active log information */ 269 struct curseg_info { 270 struct mutex curseg_mutex; /* lock for consistency */ 271 struct f2fs_summary_block *sum_blk; /* cached summary block */ 272 struct rw_semaphore journal_rwsem; /* protect journal area */ 273 struct f2fs_journal *journal; /* cached journal info */ 274 unsigned char alloc_type; /* current allocation type */ 275 unsigned int segno; /* current segment number */ 276 unsigned short next_blkoff; /* next block offset to write */ 277 unsigned int zone; /* current zone number */ 278 unsigned int next_segno; /* preallocated segment */ 279 }; 280 281 struct sit_entry_set { 282 struct list_head set_list; /* link with all sit sets */ 283 unsigned int start_segno; /* start segno of sits in set */ 284 unsigned int entry_cnt; /* the # of sit entries in set */ 285 }; 286 287 /* 288 * inline functions 289 */ 290 static inline struct curseg_info *CURSEG_I(struct f2fs_sb_info *sbi, int type) 291 { 292 return (struct curseg_info *)(SM_I(sbi)->curseg_array + type); 293 } 294 295 static inline struct seg_entry *get_seg_entry(struct f2fs_sb_info *sbi, 296 unsigned int segno) 297 { 298 struct sit_info *sit_i = SIT_I(sbi); 299 return &sit_i->sentries[segno]; 300 } 301 302 static inline struct sec_entry *get_sec_entry(struct f2fs_sb_info *sbi, 303 unsigned int segno) 304 { 305 struct sit_info *sit_i = SIT_I(sbi); 306 return &sit_i->sec_entries[GET_SECNO(sbi, segno)]; 307 } 308 309 static inline unsigned int get_valid_blocks(struct f2fs_sb_info *sbi, 310 unsigned int segno, int section) 311 { 312 /* 313 * In order to get # of valid blocks in a section instantly from many 314 * segments, f2fs manages two counting structures separately. 315 */ 316 if (section > 1) 317 return get_sec_entry(sbi, segno)->valid_blocks; 318 else 319 return get_seg_entry(sbi, segno)->valid_blocks; 320 } 321 322 static inline void seg_info_from_raw_sit(struct seg_entry *se, 323 struct f2fs_sit_entry *rs) 324 { 325 se->valid_blocks = GET_SIT_VBLOCKS(rs); 326 se->ckpt_valid_blocks = GET_SIT_VBLOCKS(rs); 327 memcpy(se->cur_valid_map, rs->valid_map, SIT_VBLOCK_MAP_SIZE); 328 memcpy(se->ckpt_valid_map, rs->valid_map, SIT_VBLOCK_MAP_SIZE); 329 #ifdef CONFIG_F2FS_CHECK_FS 330 memcpy(se->cur_valid_map_mir, rs->valid_map, SIT_VBLOCK_MAP_SIZE); 331 #endif 332 se->type = GET_SIT_TYPE(rs); 333 se->mtime = le64_to_cpu(rs->mtime); 334 } 335 336 static inline void seg_info_to_raw_sit(struct seg_entry *se, 337 struct f2fs_sit_entry *rs) 338 { 339 unsigned short raw_vblocks = (se->type << SIT_VBLOCKS_SHIFT) | 340 se->valid_blocks; 341 rs->vblocks = cpu_to_le16(raw_vblocks); 342 memcpy(rs->valid_map, se->cur_valid_map, SIT_VBLOCK_MAP_SIZE); 343 memcpy(se->ckpt_valid_map, rs->valid_map, SIT_VBLOCK_MAP_SIZE); 344 se->ckpt_valid_blocks = se->valid_blocks; 345 rs->mtime = cpu_to_le64(se->mtime); 346 } 347 348 static inline unsigned int find_next_inuse(struct free_segmap_info *free_i, 349 unsigned int max, unsigned int segno) 350 { 351 unsigned int ret; 352 spin_lock(&free_i->segmap_lock); 353 ret = find_next_bit(free_i->free_segmap, max, segno); 354 spin_unlock(&free_i->segmap_lock); 355 return ret; 356 } 357 358 static inline void __set_free(struct f2fs_sb_info *sbi, unsigned int segno) 359 { 360 struct free_segmap_info *free_i = FREE_I(sbi); 361 unsigned int secno = segno / sbi->segs_per_sec; 362 unsigned int start_segno = secno * sbi->segs_per_sec; 363 unsigned int next; 364 365 spin_lock(&free_i->segmap_lock); 366 clear_bit(segno, free_i->free_segmap); 367 free_i->free_segments++; 368 369 next = find_next_bit(free_i->free_segmap, 370 start_segno + sbi->segs_per_sec, start_segno); 371 if (next >= start_segno + sbi->segs_per_sec) { 372 clear_bit(secno, free_i->free_secmap); 373 free_i->free_sections++; 374 } 375 spin_unlock(&free_i->segmap_lock); 376 } 377 378 static inline void __set_inuse(struct f2fs_sb_info *sbi, 379 unsigned int segno) 380 { 381 struct free_segmap_info *free_i = FREE_I(sbi); 382 unsigned int secno = segno / sbi->segs_per_sec; 383 set_bit(segno, free_i->free_segmap); 384 free_i->free_segments--; 385 if (!test_and_set_bit(secno, free_i->free_secmap)) 386 free_i->free_sections--; 387 } 388 389 static inline void __set_test_and_free(struct f2fs_sb_info *sbi, 390 unsigned int segno) 391 { 392 struct free_segmap_info *free_i = FREE_I(sbi); 393 unsigned int secno = segno / sbi->segs_per_sec; 394 unsigned int start_segno = secno * sbi->segs_per_sec; 395 unsigned int next; 396 397 spin_lock(&free_i->segmap_lock); 398 if (test_and_clear_bit(segno, free_i->free_segmap)) { 399 free_i->free_segments++; 400 401 next = find_next_bit(free_i->free_segmap, 402 start_segno + sbi->segs_per_sec, start_segno); 403 if (next >= start_segno + sbi->segs_per_sec) { 404 if (test_and_clear_bit(secno, free_i->free_secmap)) 405 free_i->free_sections++; 406 } 407 } 408 spin_unlock(&free_i->segmap_lock); 409 } 410 411 static inline void __set_test_and_inuse(struct f2fs_sb_info *sbi, 412 unsigned int segno) 413 { 414 struct free_segmap_info *free_i = FREE_I(sbi); 415 unsigned int secno = segno / sbi->segs_per_sec; 416 spin_lock(&free_i->segmap_lock); 417 if (!test_and_set_bit(segno, free_i->free_segmap)) { 418 free_i->free_segments--; 419 if (!test_and_set_bit(secno, free_i->free_secmap)) 420 free_i->free_sections--; 421 } 422 spin_unlock(&free_i->segmap_lock); 423 } 424 425 static inline void get_sit_bitmap(struct f2fs_sb_info *sbi, 426 void *dst_addr) 427 { 428 struct sit_info *sit_i = SIT_I(sbi); 429 430 #ifdef CONFIG_F2FS_CHECK_FS 431 if (memcmp(sit_i->sit_bitmap, sit_i->sit_bitmap_mir, 432 sit_i->bitmap_size)) 433 f2fs_bug_on(sbi, 1); 434 #endif 435 memcpy(dst_addr, sit_i->sit_bitmap, sit_i->bitmap_size); 436 } 437 438 static inline block_t written_block_count(struct f2fs_sb_info *sbi) 439 { 440 return SIT_I(sbi)->written_valid_blocks; 441 } 442 443 static inline unsigned int free_segments(struct f2fs_sb_info *sbi) 444 { 445 return FREE_I(sbi)->free_segments; 446 } 447 448 static inline int reserved_segments(struct f2fs_sb_info *sbi) 449 { 450 return SM_I(sbi)->reserved_segments; 451 } 452 453 static inline unsigned int free_sections(struct f2fs_sb_info *sbi) 454 { 455 return FREE_I(sbi)->free_sections; 456 } 457 458 static inline unsigned int prefree_segments(struct f2fs_sb_info *sbi) 459 { 460 return DIRTY_I(sbi)->nr_dirty[PRE]; 461 } 462 463 static inline unsigned int dirty_segments(struct f2fs_sb_info *sbi) 464 { 465 return DIRTY_I(sbi)->nr_dirty[DIRTY_HOT_DATA] + 466 DIRTY_I(sbi)->nr_dirty[DIRTY_WARM_DATA] + 467 DIRTY_I(sbi)->nr_dirty[DIRTY_COLD_DATA] + 468 DIRTY_I(sbi)->nr_dirty[DIRTY_HOT_NODE] + 469 DIRTY_I(sbi)->nr_dirty[DIRTY_WARM_NODE] + 470 DIRTY_I(sbi)->nr_dirty[DIRTY_COLD_NODE]; 471 } 472 473 static inline int overprovision_segments(struct f2fs_sb_info *sbi) 474 { 475 return SM_I(sbi)->ovp_segments; 476 } 477 478 static inline int overprovision_sections(struct f2fs_sb_info *sbi) 479 { 480 return ((unsigned int) overprovision_segments(sbi)) / sbi->segs_per_sec; 481 } 482 483 static inline int reserved_sections(struct f2fs_sb_info *sbi) 484 { 485 return ((unsigned int) reserved_segments(sbi)) / sbi->segs_per_sec; 486 } 487 488 static inline bool need_SSR(struct f2fs_sb_info *sbi) 489 { 490 int node_secs = get_blocktype_secs(sbi, F2FS_DIRTY_NODES); 491 int dent_secs = get_blocktype_secs(sbi, F2FS_DIRTY_DENTS); 492 int imeta_secs = get_blocktype_secs(sbi, F2FS_DIRTY_IMETA); 493 494 if (test_opt(sbi, LFS)) 495 return false; 496 497 return free_sections(sbi) <= (node_secs + 2 * dent_secs + imeta_secs + 498 reserved_sections(sbi) + 1); 499 } 500 501 static inline bool has_not_enough_free_secs(struct f2fs_sb_info *sbi, 502 int freed, int needed) 503 { 504 int node_secs = get_blocktype_secs(sbi, F2FS_DIRTY_NODES); 505 int dent_secs = get_blocktype_secs(sbi, F2FS_DIRTY_DENTS); 506 int imeta_secs = get_blocktype_secs(sbi, F2FS_DIRTY_IMETA); 507 508 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING))) 509 return false; 510 511 return (free_sections(sbi) + freed) <= 512 (node_secs + 2 * dent_secs + imeta_secs + 513 reserved_sections(sbi) + needed); 514 } 515 516 static inline bool excess_prefree_segs(struct f2fs_sb_info *sbi) 517 { 518 return prefree_segments(sbi) > SM_I(sbi)->rec_prefree_segments; 519 } 520 521 static inline int utilization(struct f2fs_sb_info *sbi) 522 { 523 return div_u64((u64)valid_user_blocks(sbi) * 100, 524 sbi->user_block_count); 525 } 526 527 /* 528 * Sometimes f2fs may be better to drop out-of-place update policy. 529 * And, users can control the policy through sysfs entries. 530 * There are five policies with triggering conditions as follows. 531 * F2FS_IPU_FORCE - all the time, 532 * F2FS_IPU_SSR - if SSR mode is activated, 533 * F2FS_IPU_UTIL - if FS utilization is over threashold, 534 * F2FS_IPU_SSR_UTIL - if SSR mode is activated and FS utilization is over 535 * threashold, 536 * F2FS_IPU_FSYNC - activated in fsync path only for high performance flash 537 * storages. IPU will be triggered only if the # of dirty 538 * pages over min_fsync_blocks. 539 * F2FS_IPUT_DISABLE - disable IPU. (=default option) 540 */ 541 #define DEF_MIN_IPU_UTIL 70 542 #define DEF_MIN_FSYNC_BLOCKS 8 543 544 enum { 545 F2FS_IPU_FORCE, 546 F2FS_IPU_SSR, 547 F2FS_IPU_UTIL, 548 F2FS_IPU_SSR_UTIL, 549 F2FS_IPU_FSYNC, 550 }; 551 552 static inline bool need_inplace_update(struct inode *inode) 553 { 554 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 555 unsigned int policy = SM_I(sbi)->ipu_policy; 556 557 /* IPU can be done only for the user data */ 558 if (S_ISDIR(inode->i_mode) || f2fs_is_atomic_file(inode)) 559 return false; 560 561 if (test_opt(sbi, LFS)) 562 return false; 563 564 if (policy & (0x1 << F2FS_IPU_FORCE)) 565 return true; 566 if (policy & (0x1 << F2FS_IPU_SSR) && need_SSR(sbi)) 567 return true; 568 if (policy & (0x1 << F2FS_IPU_UTIL) && 569 utilization(sbi) > SM_I(sbi)->min_ipu_util) 570 return true; 571 if (policy & (0x1 << F2FS_IPU_SSR_UTIL) && need_SSR(sbi) && 572 utilization(sbi) > SM_I(sbi)->min_ipu_util) 573 return true; 574 575 /* this is only set during fdatasync */ 576 if (policy & (0x1 << F2FS_IPU_FSYNC) && 577 is_inode_flag_set(inode, FI_NEED_IPU)) 578 return true; 579 580 return false; 581 } 582 583 static inline unsigned int curseg_segno(struct f2fs_sb_info *sbi, 584 int type) 585 { 586 struct curseg_info *curseg = CURSEG_I(sbi, type); 587 return curseg->segno; 588 } 589 590 static inline unsigned char curseg_alloc_type(struct f2fs_sb_info *sbi, 591 int type) 592 { 593 struct curseg_info *curseg = CURSEG_I(sbi, type); 594 return curseg->alloc_type; 595 } 596 597 static inline unsigned short curseg_blkoff(struct f2fs_sb_info *sbi, int type) 598 { 599 struct curseg_info *curseg = CURSEG_I(sbi, type); 600 return curseg->next_blkoff; 601 } 602 603 static inline void check_seg_range(struct f2fs_sb_info *sbi, unsigned int segno) 604 { 605 f2fs_bug_on(sbi, segno > TOTAL_SEGS(sbi) - 1); 606 } 607 608 static inline void verify_block_addr(struct f2fs_sb_info *sbi, block_t blk_addr) 609 { 610 BUG_ON(blk_addr < SEG0_BLKADDR(sbi) 611 || blk_addr >= MAX_BLKADDR(sbi)); 612 } 613 614 /* 615 * Summary block is always treated as an invalid block 616 */ 617 static inline void check_block_count(struct f2fs_sb_info *sbi, 618 int segno, struct f2fs_sit_entry *raw_sit) 619 { 620 #ifdef CONFIG_F2FS_CHECK_FS 621 bool is_valid = test_bit_le(0, raw_sit->valid_map) ? true : false; 622 int valid_blocks = 0; 623 int cur_pos = 0, next_pos; 624 625 /* check bitmap with valid block count */ 626 do { 627 if (is_valid) { 628 next_pos = find_next_zero_bit_le(&raw_sit->valid_map, 629 sbi->blocks_per_seg, 630 cur_pos); 631 valid_blocks += next_pos - cur_pos; 632 } else 633 next_pos = find_next_bit_le(&raw_sit->valid_map, 634 sbi->blocks_per_seg, 635 cur_pos); 636 cur_pos = next_pos; 637 is_valid = !is_valid; 638 } while (cur_pos < sbi->blocks_per_seg); 639 BUG_ON(GET_SIT_VBLOCKS(raw_sit) != valid_blocks); 640 #endif 641 /* check segment usage, and check boundary of a given segment number */ 642 f2fs_bug_on(sbi, GET_SIT_VBLOCKS(raw_sit) > sbi->blocks_per_seg 643 || segno > TOTAL_SEGS(sbi) - 1); 644 } 645 646 static inline pgoff_t current_sit_addr(struct f2fs_sb_info *sbi, 647 unsigned int start) 648 { 649 struct sit_info *sit_i = SIT_I(sbi); 650 unsigned int offset = SIT_BLOCK_OFFSET(start); 651 block_t blk_addr = sit_i->sit_base_addr + offset; 652 653 check_seg_range(sbi, start); 654 655 #ifdef CONFIG_F2FS_CHECK_FS 656 if (f2fs_test_bit(offset, sit_i->sit_bitmap) != 657 f2fs_test_bit(offset, sit_i->sit_bitmap_mir)) 658 f2fs_bug_on(sbi, 1); 659 #endif 660 661 /* calculate sit block address */ 662 if (f2fs_test_bit(offset, sit_i->sit_bitmap)) 663 blk_addr += sit_i->sit_blocks; 664 665 return blk_addr; 666 } 667 668 static inline pgoff_t next_sit_addr(struct f2fs_sb_info *sbi, 669 pgoff_t block_addr) 670 { 671 struct sit_info *sit_i = SIT_I(sbi); 672 block_addr -= sit_i->sit_base_addr; 673 if (block_addr < sit_i->sit_blocks) 674 block_addr += sit_i->sit_blocks; 675 else 676 block_addr -= sit_i->sit_blocks; 677 678 return block_addr + sit_i->sit_base_addr; 679 } 680 681 static inline void set_to_next_sit(struct sit_info *sit_i, unsigned int start) 682 { 683 unsigned int block_off = SIT_BLOCK_OFFSET(start); 684 685 f2fs_change_bit(block_off, sit_i->sit_bitmap); 686 #ifdef CONFIG_F2FS_CHECK_FS 687 f2fs_change_bit(block_off, sit_i->sit_bitmap_mir); 688 #endif 689 } 690 691 static inline unsigned long long get_mtime(struct f2fs_sb_info *sbi) 692 { 693 struct sit_info *sit_i = SIT_I(sbi); 694 return sit_i->elapsed_time + CURRENT_TIME_SEC.tv_sec - 695 sit_i->mounted_time; 696 } 697 698 static inline void set_summary(struct f2fs_summary *sum, nid_t nid, 699 unsigned int ofs_in_node, unsigned char version) 700 { 701 sum->nid = cpu_to_le32(nid); 702 sum->ofs_in_node = cpu_to_le16(ofs_in_node); 703 sum->version = version; 704 } 705 706 static inline block_t start_sum_block(struct f2fs_sb_info *sbi) 707 { 708 return __start_cp_addr(sbi) + 709 le32_to_cpu(F2FS_CKPT(sbi)->cp_pack_start_sum); 710 } 711 712 static inline block_t sum_blk_addr(struct f2fs_sb_info *sbi, int base, int type) 713 { 714 return __start_cp_addr(sbi) + 715 le32_to_cpu(F2FS_CKPT(sbi)->cp_pack_total_block_count) 716 - (base + 1) + type; 717 } 718 719 static inline bool no_fggc_candidate(struct f2fs_sb_info *sbi, 720 unsigned int secno) 721 { 722 if (get_valid_blocks(sbi, secno, sbi->segs_per_sec) >= 723 sbi->fggc_threshold) 724 return true; 725 return false; 726 } 727 728 static inline bool sec_usage_check(struct f2fs_sb_info *sbi, unsigned int secno) 729 { 730 if (IS_CURSEC(sbi, secno) || (sbi->cur_victim_sec == secno)) 731 return true; 732 return false; 733 } 734 735 /* 736 * It is very important to gather dirty pages and write at once, so that we can 737 * submit a big bio without interfering other data writes. 738 * By default, 512 pages for directory data, 739 * 512 pages (2MB) * 8 for nodes, and 740 * 256 pages * 8 for meta are set. 741 */ 742 static inline int nr_pages_to_skip(struct f2fs_sb_info *sbi, int type) 743 { 744 if (sbi->sb->s_bdi->wb.dirty_exceeded) 745 return 0; 746 747 if (type == DATA) 748 return sbi->blocks_per_seg; 749 else if (type == NODE) 750 return 8 * sbi->blocks_per_seg; 751 else if (type == META) 752 return 8 * BIO_MAX_PAGES; 753 else 754 return 0; 755 } 756 757 /* 758 * When writing pages, it'd better align nr_to_write for segment size. 759 */ 760 static inline long nr_pages_to_write(struct f2fs_sb_info *sbi, int type, 761 struct writeback_control *wbc) 762 { 763 long nr_to_write, desired; 764 765 if (wbc->sync_mode != WB_SYNC_NONE) 766 return 0; 767 768 nr_to_write = wbc->nr_to_write; 769 desired = BIO_MAX_PAGES; 770 if (type == NODE) 771 desired <<= 1; 772 773 wbc->nr_to_write = desired; 774 return desired - nr_to_write; 775 } 776