1/* SPDX-License-Identifier: GPL-2.0 */ 2/* 3 * fs/f2fs/segment.h 4 * 5 * Copyright (c) 2012 Samsung Electronics Co., Ltd. 6 * http://www.samsung.com/ 7 */ 8#include <linux/blkdev.h> 9#include <linux/backing-dev.h> 10 11/* constant macro */ 12#define NULL_SEGNO ((unsigned int)(~0)) 13#define NULL_SECNO ((unsigned int)(~0)) 14 15#define DEF_RECLAIM_PREFREE_SEGMENTS 5 /* 5% over total segments */ 16#define DEF_MAX_RECLAIM_PREFREE_SEGMENTS 4096 /* 8GB in maximum */ 17 18#define F2FS_MIN_SEGMENTS 9 /* SB + 2 (CP + SIT + NAT) + SSA + MAIN */ 19#define F2FS_MIN_META_SEGMENTS 8 /* SB + 2 (CP + SIT + NAT) + SSA */ 20 21/* L: Logical segment # in volume, R: Relative segment # in main area */ 22#define GET_L2R_SEGNO(free_i, segno) ((segno) - (free_i)->start_segno) 23#define GET_R2L_SEGNO(free_i, segno) ((segno) + (free_i)->start_segno) 24 25#define IS_DATASEG(t) ((t) <= CURSEG_COLD_DATA) 26#define IS_NODESEG(t) ((t) >= CURSEG_HOT_NODE && (t) <= CURSEG_COLD_NODE) 27#define SE_PAGETYPE(se) ((IS_NODESEG((se)->type) ? NODE : DATA)) 28 29static inline void sanity_check_seg_type(struct f2fs_sb_info *sbi, 30 unsigned short seg_type) 31{ 32 f2fs_bug_on(sbi, seg_type >= NR_PERSISTENT_LOG); 33} 34 35#define IS_HOT(t) ((t) == CURSEG_HOT_NODE || (t) == CURSEG_HOT_DATA) 36#define IS_WARM(t) ((t) == CURSEG_WARM_NODE || (t) == CURSEG_WARM_DATA) 37#define IS_COLD(t) ((t) == CURSEG_COLD_NODE || (t) == CURSEG_COLD_DATA) 38 39#define IS_CURSEG(sbi, seg) \ 40 (((seg) == CURSEG_I(sbi, CURSEG_HOT_DATA)->segno) || \ 41 ((seg) == CURSEG_I(sbi, CURSEG_WARM_DATA)->segno) || \ 42 ((seg) == CURSEG_I(sbi, CURSEG_COLD_DATA)->segno) || \ 43 ((seg) == CURSEG_I(sbi, CURSEG_HOT_NODE)->segno) || \ 44 ((seg) == CURSEG_I(sbi, CURSEG_WARM_NODE)->segno) || \ 45 ((seg) == CURSEG_I(sbi, CURSEG_COLD_NODE)->segno) || \ 46 ((seg) == CURSEG_I(sbi, CURSEG_COLD_DATA_PINNED)->segno) || \ 47 ((seg) == CURSEG_I(sbi, CURSEG_ALL_DATA_ATGC)->segno)) 48 49#define IS_CURSEC(sbi, secno) \ 50 (((secno) == CURSEG_I(sbi, CURSEG_HOT_DATA)->segno / \ 51 (sbi)->segs_per_sec) || \ 52 ((secno) == CURSEG_I(sbi, CURSEG_WARM_DATA)->segno / \ 53 (sbi)->segs_per_sec) || \ 54 ((secno) == CURSEG_I(sbi, CURSEG_COLD_DATA)->segno / \ 55 (sbi)->segs_per_sec) || \ 56 ((secno) == CURSEG_I(sbi, CURSEG_HOT_NODE)->segno / \ 57 (sbi)->segs_per_sec) || \ 58 ((secno) == CURSEG_I(sbi, CURSEG_WARM_NODE)->segno / \ 59 (sbi)->segs_per_sec) || \ 60 ((secno) == CURSEG_I(sbi, CURSEG_COLD_NODE)->segno / \ 61 (sbi)->segs_per_sec) || \ 62 ((secno) == CURSEG_I(sbi, CURSEG_COLD_DATA_PINNED)->segno / \ 63 (sbi)->segs_per_sec) || \ 64 ((secno) == CURSEG_I(sbi, CURSEG_ALL_DATA_ATGC)->segno / \ 65 (sbi)->segs_per_sec)) 66 67#define MAIN_BLKADDR(sbi) \ 68 (SM_I(sbi) ? SM_I(sbi)->main_blkaddr : \ 69 le32_to_cpu(F2FS_RAW_SUPER(sbi)->main_blkaddr)) 70#define SEG0_BLKADDR(sbi) \ 71 (SM_I(sbi) ? SM_I(sbi)->seg0_blkaddr : \ 72 le32_to_cpu(F2FS_RAW_SUPER(sbi)->segment0_blkaddr)) 73 74#define MAIN_SEGS(sbi) (SM_I(sbi)->main_segments) 75#define MAIN_SECS(sbi) ((sbi)->total_sections) 76 77#define TOTAL_SEGS(sbi) \ 78 (SM_I(sbi) ? SM_I(sbi)->segment_count : \ 79 le32_to_cpu(F2FS_RAW_SUPER(sbi)->segment_count)) 80#define TOTAL_BLKS(sbi) (TOTAL_SEGS(sbi) << (sbi)->log_blocks_per_seg) 81 82#define MAX_BLKADDR(sbi) (SEG0_BLKADDR(sbi) + TOTAL_BLKS(sbi)) 83#define SEGMENT_SIZE(sbi) (1ULL << ((sbi)->log_blocksize + \ 84 (sbi)->log_blocks_per_seg)) 85 86#define START_BLOCK(sbi, segno) (SEG0_BLKADDR(sbi) + \ 87 (GET_R2L_SEGNO(FREE_I(sbi), segno) << (sbi)->log_blocks_per_seg)) 88 89#define NEXT_FREE_BLKADDR(sbi, curseg) \ 90 (START_BLOCK(sbi, (curseg)->segno) + (curseg)->next_blkoff) 91 92#define GET_SEGOFF_FROM_SEG0(sbi, blk_addr) ((blk_addr) - SEG0_BLKADDR(sbi)) 93#define GET_SEGNO_FROM_SEG0(sbi, blk_addr) \ 94 (GET_SEGOFF_FROM_SEG0(sbi, blk_addr) >> (sbi)->log_blocks_per_seg) 95#define GET_BLKOFF_FROM_SEG0(sbi, blk_addr) \ 96 (GET_SEGOFF_FROM_SEG0(sbi, blk_addr) & ((sbi)->blocks_per_seg - 1)) 97 98#define GET_SEGNO(sbi, blk_addr) \ 99 ((!__is_valid_data_blkaddr(blk_addr)) ? \ 100 NULL_SEGNO : GET_L2R_SEGNO(FREE_I(sbi), \ 101 GET_SEGNO_FROM_SEG0(sbi, blk_addr))) 102#define BLKS_PER_SEC(sbi) \ 103 ((sbi)->segs_per_sec * (sbi)->blocks_per_seg) 104#define CAP_BLKS_PER_SEC(sbi) \ 105 ((sbi)->segs_per_sec * (sbi)->blocks_per_seg - \ 106 (sbi)->unusable_blocks_per_sec) 107#define CAP_SEGS_PER_SEC(sbi) \ 108 ((sbi)->segs_per_sec - ((sbi)->unusable_blocks_per_sec >>\ 109 (sbi)->log_blocks_per_seg)) 110#define GET_SEC_FROM_SEG(sbi, segno) \ 111 (((segno) == -1) ? -1: (segno) / (sbi)->segs_per_sec) 112#define GET_SEG_FROM_SEC(sbi, secno) \ 113 ((secno) * (sbi)->segs_per_sec) 114#define GET_ZONE_FROM_SEC(sbi, secno) \ 115 (((secno) == -1) ? -1: (secno) / (sbi)->secs_per_zone) 116#define GET_ZONE_FROM_SEG(sbi, segno) \ 117 GET_ZONE_FROM_SEC(sbi, GET_SEC_FROM_SEG(sbi, segno)) 118 119#define GET_SUM_BLOCK(sbi, segno) \ 120 ((sbi)->sm_info->ssa_blkaddr + (segno)) 121 122#define GET_SUM_TYPE(footer) ((footer)->entry_type) 123#define SET_SUM_TYPE(footer, type) ((footer)->entry_type = (type)) 124 125#define SIT_ENTRY_OFFSET(sit_i, segno) \ 126 ((segno) % (sit_i)->sents_per_block) 127#define SIT_BLOCK_OFFSET(segno) \ 128 ((segno) / SIT_ENTRY_PER_BLOCK) 129#define START_SEGNO(segno) \ 130 (SIT_BLOCK_OFFSET(segno) * SIT_ENTRY_PER_BLOCK) 131#define SIT_BLK_CNT(sbi) \ 132 DIV_ROUND_UP(MAIN_SEGS(sbi), SIT_ENTRY_PER_BLOCK) 133#define f2fs_bitmap_size(nr) \ 134 (BITS_TO_LONGS(nr) * sizeof(unsigned long)) 135 136#define SECTOR_FROM_BLOCK(blk_addr) \ 137 (((sector_t)blk_addr) << F2FS_LOG_SECTORS_PER_BLOCK) 138#define SECTOR_TO_BLOCK(sectors) \ 139 ((sectors) >> F2FS_LOG_SECTORS_PER_BLOCK) 140 141/* 142 * indicate a block allocation direction: RIGHT and LEFT. 143 * RIGHT means allocating new sections towards the end of volume. 144 * LEFT means the opposite direction. 145 */ 146enum { 147 ALLOC_RIGHT = 0, 148 ALLOC_LEFT 149}; 150 151/* 152 * In the victim_sel_policy->alloc_mode, there are three block allocation modes. 153 * LFS writes data sequentially with cleaning operations. 154 * SSR (Slack Space Recycle) reuses obsolete space without cleaning operations. 155 * AT_SSR (Age Threshold based Slack Space Recycle) merges fragments into 156 * fragmented segment which has similar aging degree. 157 */ 158enum { 159 LFS = 0, 160 SSR, 161 AT_SSR, 162}; 163 164/* 165 * In the victim_sel_policy->gc_mode, there are three gc, aka cleaning, modes. 166 * GC_CB is based on cost-benefit algorithm. 167 * GC_GREEDY is based on greedy algorithm. 168 * GC_AT is based on age-threshold algorithm. 169 */ 170enum { 171 GC_CB = 0, 172 GC_GREEDY, 173 GC_AT, 174 ALLOC_NEXT, 175 FLUSH_DEVICE, 176 MAX_GC_POLICY, 177}; 178 179/* 180 * BG_GC means the background cleaning job. 181 * FG_GC means the on-demand cleaning job. 182 */ 183enum { 184 BG_GC = 0, 185 FG_GC, 186}; 187 188/* for a function parameter to select a victim segment */ 189struct victim_sel_policy { 190 int alloc_mode; /* LFS or SSR */ 191 int gc_mode; /* GC_CB or GC_GREEDY */ 192 unsigned long *dirty_bitmap; /* dirty segment/section bitmap */ 193 unsigned int max_search; /* 194 * maximum # of segments/sections 195 * to search 196 */ 197 unsigned int offset; /* last scanned bitmap offset */ 198 unsigned int ofs_unit; /* bitmap search unit */ 199 unsigned int min_cost; /* minimum cost */ 200 unsigned long long oldest_age; /* oldest age of segments having the same min cost */ 201 unsigned int min_segno; /* segment # having min. cost */ 202 unsigned long long age; /* mtime of GCed section*/ 203 unsigned long long age_threshold;/* age threshold */ 204}; 205 206struct seg_entry { 207 unsigned int type:6; /* segment type like CURSEG_XXX_TYPE */ 208 unsigned int valid_blocks:10; /* # of valid blocks */ 209 unsigned int ckpt_valid_blocks:10; /* # of valid blocks last cp */ 210 unsigned int padding:6; /* padding */ 211 unsigned char *cur_valid_map; /* validity bitmap of blocks */ 212#ifdef CONFIG_F2FS_CHECK_FS 213 unsigned char *cur_valid_map_mir; /* mirror of current valid bitmap */ 214#endif 215 /* 216 * # of valid blocks and the validity bitmap stored in the last 217 * checkpoint pack. This information is used by the SSR mode. 218 */ 219 unsigned char *ckpt_valid_map; /* validity bitmap of blocks last cp */ 220 unsigned char *discard_map; 221 unsigned long long mtime; /* modification time of the segment */ 222}; 223 224struct sec_entry { 225 unsigned int valid_blocks; /* # of valid blocks in a section */ 226}; 227 228#define MAX_SKIP_GC_COUNT 16 229 230struct revoke_entry { 231 struct list_head list; 232 block_t old_addr; /* for revoking when fail to commit */ 233 pgoff_t index; 234}; 235 236struct sit_info { 237 block_t sit_base_addr; /* start block address of SIT area */ 238 block_t sit_blocks; /* # of blocks used by SIT area */ 239 block_t written_valid_blocks; /* # of valid blocks in main area */ 240 char *bitmap; /* all bitmaps pointer */ 241 char *sit_bitmap; /* SIT bitmap pointer */ 242#ifdef CONFIG_F2FS_CHECK_FS 243 char *sit_bitmap_mir; /* SIT bitmap mirror */ 244 245 /* bitmap of segments to be ignored by GC in case of errors */ 246 unsigned long *invalid_segmap; 247#endif 248 unsigned int bitmap_size; /* SIT bitmap size */ 249 250 unsigned long *tmp_map; /* bitmap for temporal use */ 251 unsigned long *dirty_sentries_bitmap; /* bitmap for dirty sentries */ 252 unsigned int dirty_sentries; /* # of dirty sentries */ 253 unsigned int sents_per_block; /* # of SIT entries per block */ 254 struct rw_semaphore sentry_lock; /* to protect SIT cache */ 255 struct seg_entry *sentries; /* SIT segment-level cache */ 256 struct sec_entry *sec_entries; /* SIT section-level cache */ 257 258 /* for cost-benefit algorithm in cleaning procedure */ 259 unsigned long long elapsed_time; /* elapsed time after mount */ 260 unsigned long long mounted_time; /* mount time */ 261 unsigned long long min_mtime; /* min. modification time */ 262 unsigned long long max_mtime; /* max. modification time */ 263 unsigned long long dirty_min_mtime; /* rerange candidates in GC_AT */ 264 unsigned long long dirty_max_mtime; /* rerange candidates in GC_AT */ 265 266 unsigned int last_victim[MAX_GC_POLICY]; /* last victim segment # */ 267}; 268 269struct free_segmap_info { 270 unsigned int start_segno; /* start segment number logically */ 271 unsigned int free_segments; /* # of free segments */ 272 unsigned int free_sections; /* # of free sections */ 273 spinlock_t segmap_lock; /* free segmap lock */ 274 unsigned long *free_segmap; /* free segment bitmap */ 275 unsigned long *free_secmap; /* free section bitmap */ 276}; 277 278/* Notice: The order of dirty type is same with CURSEG_XXX in f2fs.h */ 279enum dirty_type { 280 DIRTY_HOT_DATA, /* dirty segments assigned as hot data logs */ 281 DIRTY_WARM_DATA, /* dirty segments assigned as warm data logs */ 282 DIRTY_COLD_DATA, /* dirty segments assigned as cold data logs */ 283 DIRTY_HOT_NODE, /* dirty segments assigned as hot node logs */ 284 DIRTY_WARM_NODE, /* dirty segments assigned as warm node logs */ 285 DIRTY_COLD_NODE, /* dirty segments assigned as cold node logs */ 286 DIRTY, /* to count # of dirty segments */ 287 PRE, /* to count # of entirely obsolete segments */ 288 NR_DIRTY_TYPE 289}; 290 291struct dirty_seglist_info { 292 unsigned long *dirty_segmap[NR_DIRTY_TYPE]; 293 unsigned long *dirty_secmap; 294 struct mutex seglist_lock; /* lock for segment bitmaps */ 295 int nr_dirty[NR_DIRTY_TYPE]; /* # of dirty segments */ 296 unsigned long *victim_secmap; /* background GC victims */ 297 unsigned long *pinned_secmap; /* pinned victims from foreground GC */ 298 unsigned int pinned_secmap_cnt; /* count of victims which has pinned data */ 299 bool enable_pin_section; /* enable pinning section */ 300}; 301 302/* for active log information */ 303struct curseg_info { 304 struct mutex curseg_mutex; /* lock for consistency */ 305 struct f2fs_summary_block *sum_blk; /* cached summary block */ 306 struct rw_semaphore journal_rwsem; /* protect journal area */ 307 struct f2fs_journal *journal; /* cached journal info */ 308 unsigned char alloc_type; /* current allocation type */ 309 unsigned short seg_type; /* segment type like CURSEG_XXX_TYPE */ 310 unsigned int segno; /* current segment number */ 311 unsigned short next_blkoff; /* next block offset to write */ 312 unsigned int zone; /* current zone number */ 313 unsigned int next_segno; /* preallocated segment */ 314 int fragment_remained_chunk; /* remained block size in a chunk for block fragmentation mode */ 315 bool inited; /* indicate inmem log is inited */ 316}; 317 318struct sit_entry_set { 319 struct list_head set_list; /* link with all sit sets */ 320 unsigned int start_segno; /* start segno of sits in set */ 321 unsigned int entry_cnt; /* the # of sit entries in set */ 322}; 323 324/* 325 * inline functions 326 */ 327static inline struct curseg_info *CURSEG_I(struct f2fs_sb_info *sbi, int type) 328{ 329 return (struct curseg_info *)(SM_I(sbi)->curseg_array + type); 330} 331 332static inline struct seg_entry *get_seg_entry(struct f2fs_sb_info *sbi, 333 unsigned int segno) 334{ 335 struct sit_info *sit_i = SIT_I(sbi); 336 return &sit_i->sentries[segno]; 337} 338 339static inline struct sec_entry *get_sec_entry(struct f2fs_sb_info *sbi, 340 unsigned int segno) 341{ 342 struct sit_info *sit_i = SIT_I(sbi); 343 return &sit_i->sec_entries[GET_SEC_FROM_SEG(sbi, segno)]; 344} 345 346static inline unsigned int get_valid_blocks(struct f2fs_sb_info *sbi, 347 unsigned int segno, bool use_section) 348{ 349 /* 350 * In order to get # of valid blocks in a section instantly from many 351 * segments, f2fs manages two counting structures separately. 352 */ 353 if (use_section && __is_large_section(sbi)) 354 return get_sec_entry(sbi, segno)->valid_blocks; 355 else 356 return get_seg_entry(sbi, segno)->valid_blocks; 357} 358 359static inline unsigned int get_ckpt_valid_blocks(struct f2fs_sb_info *sbi, 360 unsigned int segno, bool use_section) 361{ 362 if (use_section && __is_large_section(sbi)) { 363 unsigned int start_segno = START_SEGNO(segno); 364 unsigned int blocks = 0; 365 int i; 366 367 for (i = 0; i < sbi->segs_per_sec; i++, start_segno++) { 368 struct seg_entry *se = get_seg_entry(sbi, start_segno); 369 370 blocks += se->ckpt_valid_blocks; 371 } 372 return blocks; 373 } 374 return get_seg_entry(sbi, segno)->ckpt_valid_blocks; 375} 376 377static inline void seg_info_from_raw_sit(struct seg_entry *se, 378 struct f2fs_sit_entry *rs) 379{ 380 se->valid_blocks = GET_SIT_VBLOCKS(rs); 381 se->ckpt_valid_blocks = GET_SIT_VBLOCKS(rs); 382 memcpy(se->cur_valid_map, rs->valid_map, SIT_VBLOCK_MAP_SIZE); 383 memcpy(se->ckpt_valid_map, rs->valid_map, SIT_VBLOCK_MAP_SIZE); 384#ifdef CONFIG_F2FS_CHECK_FS 385 memcpy(se->cur_valid_map_mir, rs->valid_map, SIT_VBLOCK_MAP_SIZE); 386#endif 387 se->type = GET_SIT_TYPE(rs); 388 se->mtime = le64_to_cpu(rs->mtime); 389} 390 391static inline void __seg_info_to_raw_sit(struct seg_entry *se, 392 struct f2fs_sit_entry *rs) 393{ 394 unsigned short raw_vblocks = (se->type << SIT_VBLOCKS_SHIFT) | 395 se->valid_blocks; 396 rs->vblocks = cpu_to_le16(raw_vblocks); 397 memcpy(rs->valid_map, se->cur_valid_map, SIT_VBLOCK_MAP_SIZE); 398 rs->mtime = cpu_to_le64(se->mtime); 399} 400 401static inline void seg_info_to_sit_page(struct f2fs_sb_info *sbi, 402 struct page *page, unsigned int start) 403{ 404 struct f2fs_sit_block *raw_sit; 405 struct seg_entry *se; 406 struct f2fs_sit_entry *rs; 407 unsigned int end = min(start + SIT_ENTRY_PER_BLOCK, 408 (unsigned long)MAIN_SEGS(sbi)); 409 int i; 410 411 raw_sit = (struct f2fs_sit_block *)page_address(page); 412 memset(raw_sit, 0, PAGE_SIZE); 413 for (i = 0; i < end - start; i++) { 414 rs = &raw_sit->entries[i]; 415 se = get_seg_entry(sbi, start + i); 416 __seg_info_to_raw_sit(se, rs); 417 } 418} 419 420static inline void seg_info_to_raw_sit(struct seg_entry *se, 421 struct f2fs_sit_entry *rs) 422{ 423 __seg_info_to_raw_sit(se, rs); 424 425 memcpy(se->ckpt_valid_map, rs->valid_map, SIT_VBLOCK_MAP_SIZE); 426 se->ckpt_valid_blocks = se->valid_blocks; 427} 428 429static inline unsigned int find_next_inuse(struct free_segmap_info *free_i, 430 unsigned int max, unsigned int segno) 431{ 432 unsigned int ret; 433 spin_lock(&free_i->segmap_lock); 434 ret = find_next_bit(free_i->free_segmap, max, segno); 435 spin_unlock(&free_i->segmap_lock); 436 return ret; 437} 438 439static inline void __set_free(struct f2fs_sb_info *sbi, unsigned int segno) 440{ 441 struct free_segmap_info *free_i = FREE_I(sbi); 442 unsigned int secno = GET_SEC_FROM_SEG(sbi, segno); 443 unsigned int start_segno = GET_SEG_FROM_SEC(sbi, secno); 444 unsigned int next; 445 unsigned int usable_segs = f2fs_usable_segs_in_sec(sbi, segno); 446 447 spin_lock(&free_i->segmap_lock); 448 clear_bit(segno, free_i->free_segmap); 449 free_i->free_segments++; 450 451 next = find_next_bit(free_i->free_segmap, 452 start_segno + sbi->segs_per_sec, start_segno); 453 if (next >= start_segno + usable_segs) { 454 clear_bit(secno, free_i->free_secmap); 455 free_i->free_sections++; 456 } 457 spin_unlock(&free_i->segmap_lock); 458} 459 460static inline void __set_inuse(struct f2fs_sb_info *sbi, 461 unsigned int segno) 462{ 463 struct free_segmap_info *free_i = FREE_I(sbi); 464 unsigned int secno = GET_SEC_FROM_SEG(sbi, segno); 465 466 set_bit(segno, free_i->free_segmap); 467 free_i->free_segments--; 468 if (!test_and_set_bit(secno, free_i->free_secmap)) 469 free_i->free_sections--; 470} 471 472static inline void __set_test_and_free(struct f2fs_sb_info *sbi, 473 unsigned int segno, bool inmem) 474{ 475 struct free_segmap_info *free_i = FREE_I(sbi); 476 unsigned int secno = GET_SEC_FROM_SEG(sbi, segno); 477 unsigned int start_segno = GET_SEG_FROM_SEC(sbi, secno); 478 unsigned int next; 479 unsigned int usable_segs = f2fs_usable_segs_in_sec(sbi, segno); 480 481 spin_lock(&free_i->segmap_lock); 482 if (test_and_clear_bit(segno, free_i->free_segmap)) { 483 free_i->free_segments++; 484 485 if (!inmem && IS_CURSEC(sbi, secno)) 486 goto skip_free; 487 next = find_next_bit(free_i->free_segmap, 488 start_segno + sbi->segs_per_sec, start_segno); 489 if (next >= start_segno + usable_segs) { 490 if (test_and_clear_bit(secno, free_i->free_secmap)) 491 free_i->free_sections++; 492 } 493 } 494skip_free: 495 spin_unlock(&free_i->segmap_lock); 496} 497 498static inline void __set_test_and_inuse(struct f2fs_sb_info *sbi, 499 unsigned int segno) 500{ 501 struct free_segmap_info *free_i = FREE_I(sbi); 502 unsigned int secno = GET_SEC_FROM_SEG(sbi, segno); 503 504 spin_lock(&free_i->segmap_lock); 505 if (!test_and_set_bit(segno, free_i->free_segmap)) { 506 free_i->free_segments--; 507 if (!test_and_set_bit(secno, free_i->free_secmap)) 508 free_i->free_sections--; 509 } 510 spin_unlock(&free_i->segmap_lock); 511} 512 513static inline void get_sit_bitmap(struct f2fs_sb_info *sbi, 514 void *dst_addr) 515{ 516 struct sit_info *sit_i = SIT_I(sbi); 517 518#ifdef CONFIG_F2FS_CHECK_FS 519 if (memcmp(sit_i->sit_bitmap, sit_i->sit_bitmap_mir, 520 sit_i->bitmap_size)) 521 f2fs_bug_on(sbi, 1); 522#endif 523 memcpy(dst_addr, sit_i->sit_bitmap, sit_i->bitmap_size); 524} 525 526static inline block_t written_block_count(struct f2fs_sb_info *sbi) 527{ 528 return SIT_I(sbi)->written_valid_blocks; 529} 530 531static inline unsigned int free_segments(struct f2fs_sb_info *sbi) 532{ 533 return FREE_I(sbi)->free_segments; 534} 535 536static inline unsigned int reserved_segments(struct f2fs_sb_info *sbi) 537{ 538 return SM_I(sbi)->reserved_segments + 539 SM_I(sbi)->additional_reserved_segments; 540} 541 542static inline unsigned int free_sections(struct f2fs_sb_info *sbi) 543{ 544 return FREE_I(sbi)->free_sections; 545} 546 547static inline unsigned int prefree_segments(struct f2fs_sb_info *sbi) 548{ 549 return DIRTY_I(sbi)->nr_dirty[PRE]; 550} 551 552static inline unsigned int dirty_segments(struct f2fs_sb_info *sbi) 553{ 554 return DIRTY_I(sbi)->nr_dirty[DIRTY_HOT_DATA] + 555 DIRTY_I(sbi)->nr_dirty[DIRTY_WARM_DATA] + 556 DIRTY_I(sbi)->nr_dirty[DIRTY_COLD_DATA] + 557 DIRTY_I(sbi)->nr_dirty[DIRTY_HOT_NODE] + 558 DIRTY_I(sbi)->nr_dirty[DIRTY_WARM_NODE] + 559 DIRTY_I(sbi)->nr_dirty[DIRTY_COLD_NODE]; 560} 561 562static inline int overprovision_segments(struct f2fs_sb_info *sbi) 563{ 564 return SM_I(sbi)->ovp_segments; 565} 566 567static inline int reserved_sections(struct f2fs_sb_info *sbi) 568{ 569 return GET_SEC_FROM_SEG(sbi, reserved_segments(sbi)); 570} 571 572static inline bool has_curseg_enough_space(struct f2fs_sb_info *sbi, 573 unsigned int node_blocks, unsigned int dent_blocks) 574{ 575 576 unsigned segno, left_blocks; 577 int i; 578 579 /* check current node sections in the worst case. */ 580 for (i = CURSEG_HOT_NODE; i <= CURSEG_COLD_NODE; i++) { 581 segno = CURSEG_I(sbi, i)->segno; 582 left_blocks = CAP_BLKS_PER_SEC(sbi) - 583 get_ckpt_valid_blocks(sbi, segno, true); 584 if (node_blocks > left_blocks) 585 return false; 586 } 587 588 /* check current data section for dentry blocks. */ 589 segno = CURSEG_I(sbi, CURSEG_HOT_DATA)->segno; 590 left_blocks = CAP_BLKS_PER_SEC(sbi) - 591 get_ckpt_valid_blocks(sbi, segno, true); 592 if (dent_blocks > left_blocks) 593 return false; 594 return true; 595} 596 597/* 598 * calculate needed sections for dirty node/dentry 599 * and call has_curseg_enough_space 600 */ 601static inline void __get_secs_required(struct f2fs_sb_info *sbi, 602 unsigned int *lower_p, unsigned int *upper_p, bool *curseg_p) 603{ 604 unsigned int total_node_blocks = get_pages(sbi, F2FS_DIRTY_NODES) + 605 get_pages(sbi, F2FS_DIRTY_DENTS) + 606 get_pages(sbi, F2FS_DIRTY_IMETA); 607 unsigned int total_dent_blocks = get_pages(sbi, F2FS_DIRTY_DENTS); 608 unsigned int node_secs = total_node_blocks / CAP_BLKS_PER_SEC(sbi); 609 unsigned int dent_secs = total_dent_blocks / CAP_BLKS_PER_SEC(sbi); 610 unsigned int node_blocks = total_node_blocks % CAP_BLKS_PER_SEC(sbi); 611 unsigned int dent_blocks = total_dent_blocks % CAP_BLKS_PER_SEC(sbi); 612 613 if (lower_p) 614 *lower_p = node_secs + dent_secs; 615 if (upper_p) 616 *upper_p = node_secs + dent_secs + 617 (node_blocks ? 1 : 0) + (dent_blocks ? 1 : 0); 618 if (curseg_p) 619 *curseg_p = has_curseg_enough_space(sbi, 620 node_blocks, dent_blocks); 621} 622 623static inline bool has_not_enough_free_secs(struct f2fs_sb_info *sbi, 624 int freed, int needed) 625{ 626 unsigned int free_secs, lower_secs, upper_secs; 627 bool curseg_space; 628 629 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING))) 630 return false; 631 632 __get_secs_required(sbi, &lower_secs, &upper_secs, &curseg_space); 633 634 free_secs = free_sections(sbi) + freed; 635 lower_secs += needed + reserved_sections(sbi); 636 upper_secs += needed + reserved_sections(sbi); 637 638 if (free_secs > upper_secs) 639 return false; 640 if (free_secs <= lower_secs) 641 return true; 642 return !curseg_space; 643} 644 645static inline bool has_enough_free_secs(struct f2fs_sb_info *sbi, 646 int freed, int needed) 647{ 648 return !has_not_enough_free_secs(sbi, freed, needed); 649} 650 651static inline bool f2fs_is_checkpoint_ready(struct f2fs_sb_info *sbi) 652{ 653 if (likely(!is_sbi_flag_set(sbi, SBI_CP_DISABLED))) 654 return true; 655 if (likely(has_enough_free_secs(sbi, 0, 0))) 656 return true; 657 return false; 658} 659 660static inline bool excess_prefree_segs(struct f2fs_sb_info *sbi) 661{ 662 return prefree_segments(sbi) > SM_I(sbi)->rec_prefree_segments; 663} 664 665static inline int utilization(struct f2fs_sb_info *sbi) 666{ 667 return div_u64((u64)valid_user_blocks(sbi) * 100, 668 sbi->user_block_count); 669} 670 671/* 672 * Sometimes f2fs may be better to drop out-of-place update policy. 673 * And, users can control the policy through sysfs entries. 674 * There are five policies with triggering conditions as follows. 675 * F2FS_IPU_FORCE - all the time, 676 * F2FS_IPU_SSR - if SSR mode is activated, 677 * F2FS_IPU_UTIL - if FS utilization is over threashold, 678 * F2FS_IPU_SSR_UTIL - if SSR mode is activated and FS utilization is over 679 * threashold, 680 * F2FS_IPU_FSYNC - activated in fsync path only for high performance flash 681 * storages. IPU will be triggered only if the # of dirty 682 * pages over min_fsync_blocks. (=default option) 683 * F2FS_IPU_ASYNC - do IPU given by asynchronous write requests. 684 * F2FS_IPU_NOCACHE - disable IPU bio cache. 685 * F2FS_IPU_HONOR_OPU_WRITE - use OPU write prior to IPU write if inode has 686 * FI_OPU_WRITE flag. 687 * F2FS_IPU_DISABLE - disable IPU. (=default option in LFS mode) 688 */ 689#define DEF_MIN_IPU_UTIL 70 690#define DEF_MIN_FSYNC_BLOCKS 8 691#define DEF_MIN_HOT_BLOCKS 16 692 693#define SMALL_VOLUME_SEGMENTS (16 * 512) /* 16GB */ 694 695#define F2FS_IPU_DISABLE 0 696 697/* Modification on enum should be synchronized with ipu_mode_names array */ 698enum { 699 F2FS_IPU_FORCE, 700 F2FS_IPU_SSR, 701 F2FS_IPU_UTIL, 702 F2FS_IPU_SSR_UTIL, 703 F2FS_IPU_FSYNC, 704 F2FS_IPU_ASYNC, 705 F2FS_IPU_NOCACHE, 706 F2FS_IPU_HONOR_OPU_WRITE, 707 F2FS_IPU_MAX, 708}; 709 710static inline bool IS_F2FS_IPU_DISABLE(struct f2fs_sb_info *sbi) 711{ 712 return SM_I(sbi)->ipu_policy == F2FS_IPU_DISABLE; 713} 714 715#define F2FS_IPU_POLICY(name) \ 716static inline bool IS_##name(struct f2fs_sb_info *sbi) \ 717{ \ 718 return SM_I(sbi)->ipu_policy & BIT(name); \ 719} 720 721F2FS_IPU_POLICY(F2FS_IPU_FORCE); 722F2FS_IPU_POLICY(F2FS_IPU_SSR); 723F2FS_IPU_POLICY(F2FS_IPU_UTIL); 724F2FS_IPU_POLICY(F2FS_IPU_SSR_UTIL); 725F2FS_IPU_POLICY(F2FS_IPU_FSYNC); 726F2FS_IPU_POLICY(F2FS_IPU_ASYNC); 727F2FS_IPU_POLICY(F2FS_IPU_NOCACHE); 728F2FS_IPU_POLICY(F2FS_IPU_HONOR_OPU_WRITE); 729 730static inline unsigned int curseg_segno(struct f2fs_sb_info *sbi, 731 int type) 732{ 733 struct curseg_info *curseg = CURSEG_I(sbi, type); 734 return curseg->segno; 735} 736 737static inline unsigned char curseg_alloc_type(struct f2fs_sb_info *sbi, 738 int type) 739{ 740 struct curseg_info *curseg = CURSEG_I(sbi, type); 741 return curseg->alloc_type; 742} 743 744static inline bool valid_main_segno(struct f2fs_sb_info *sbi, 745 unsigned int segno) 746{ 747 return segno <= (MAIN_SEGS(sbi) - 1); 748} 749 750static inline void verify_fio_blkaddr(struct f2fs_io_info *fio) 751{ 752 struct f2fs_sb_info *sbi = fio->sbi; 753 754 if (__is_valid_data_blkaddr(fio->old_blkaddr)) 755 verify_blkaddr(sbi, fio->old_blkaddr, __is_meta_io(fio) ? 756 META_GENERIC : DATA_GENERIC); 757 verify_blkaddr(sbi, fio->new_blkaddr, __is_meta_io(fio) ? 758 META_GENERIC : DATA_GENERIC_ENHANCE); 759} 760 761/* 762 * Summary block is always treated as an invalid block 763 */ 764static inline int check_block_count(struct f2fs_sb_info *sbi, 765 int segno, struct f2fs_sit_entry *raw_sit) 766{ 767 bool is_valid = test_bit_le(0, raw_sit->valid_map) ? true : false; 768 int valid_blocks = 0; 769 int cur_pos = 0, next_pos; 770 unsigned int usable_blks_per_seg = f2fs_usable_blks_in_seg(sbi, segno); 771 772 /* check bitmap with valid block count */ 773 do { 774 if (is_valid) { 775 next_pos = find_next_zero_bit_le(&raw_sit->valid_map, 776 usable_blks_per_seg, 777 cur_pos); 778 valid_blocks += next_pos - cur_pos; 779 } else 780 next_pos = find_next_bit_le(&raw_sit->valid_map, 781 usable_blks_per_seg, 782 cur_pos); 783 cur_pos = next_pos; 784 is_valid = !is_valid; 785 } while (cur_pos < usable_blks_per_seg); 786 787 if (unlikely(GET_SIT_VBLOCKS(raw_sit) != valid_blocks)) { 788 f2fs_err(sbi, "Mismatch valid blocks %d vs. %d", 789 GET_SIT_VBLOCKS(raw_sit), valid_blocks); 790 set_sbi_flag(sbi, SBI_NEED_FSCK); 791 f2fs_handle_error(sbi, ERROR_INCONSISTENT_SIT); 792 return -EFSCORRUPTED; 793 } 794 795 if (usable_blks_per_seg < sbi->blocks_per_seg) 796 f2fs_bug_on(sbi, find_next_bit_le(&raw_sit->valid_map, 797 sbi->blocks_per_seg, 798 usable_blks_per_seg) != sbi->blocks_per_seg); 799 800 /* check segment usage, and check boundary of a given segment number */ 801 if (unlikely(GET_SIT_VBLOCKS(raw_sit) > usable_blks_per_seg 802 || !valid_main_segno(sbi, segno))) { 803 f2fs_err(sbi, "Wrong valid blocks %d or segno %u", 804 GET_SIT_VBLOCKS(raw_sit), segno); 805 set_sbi_flag(sbi, SBI_NEED_FSCK); 806 f2fs_handle_error(sbi, ERROR_INCONSISTENT_SIT); 807 return -EFSCORRUPTED; 808 } 809 return 0; 810} 811 812static inline pgoff_t current_sit_addr(struct f2fs_sb_info *sbi, 813 unsigned int start) 814{ 815 struct sit_info *sit_i = SIT_I(sbi); 816 unsigned int offset = SIT_BLOCK_OFFSET(start); 817 block_t blk_addr = sit_i->sit_base_addr + offset; 818 819 f2fs_bug_on(sbi, !valid_main_segno(sbi, start)); 820 821#ifdef CONFIG_F2FS_CHECK_FS 822 if (f2fs_test_bit(offset, sit_i->sit_bitmap) != 823 f2fs_test_bit(offset, sit_i->sit_bitmap_mir)) 824 f2fs_bug_on(sbi, 1); 825#endif 826 827 /* calculate sit block address */ 828 if (f2fs_test_bit(offset, sit_i->sit_bitmap)) 829 blk_addr += sit_i->sit_blocks; 830 831 return blk_addr; 832} 833 834static inline pgoff_t next_sit_addr(struct f2fs_sb_info *sbi, 835 pgoff_t block_addr) 836{ 837 struct sit_info *sit_i = SIT_I(sbi); 838 block_addr -= sit_i->sit_base_addr; 839 if (block_addr < sit_i->sit_blocks) 840 block_addr += sit_i->sit_blocks; 841 else 842 block_addr -= sit_i->sit_blocks; 843 844 return block_addr + sit_i->sit_base_addr; 845} 846 847static inline void set_to_next_sit(struct sit_info *sit_i, unsigned int start) 848{ 849 unsigned int block_off = SIT_BLOCK_OFFSET(start); 850 851 f2fs_change_bit(block_off, sit_i->sit_bitmap); 852#ifdef CONFIG_F2FS_CHECK_FS 853 f2fs_change_bit(block_off, sit_i->sit_bitmap_mir); 854#endif 855} 856 857static inline unsigned long long get_mtime(struct f2fs_sb_info *sbi, 858 bool base_time) 859{ 860 struct sit_info *sit_i = SIT_I(sbi); 861 time64_t diff, now = ktime_get_boottime_seconds(); 862 863 if (now >= sit_i->mounted_time) 864 return sit_i->elapsed_time + now - sit_i->mounted_time; 865 866 /* system time is set to the past */ 867 if (!base_time) { 868 diff = sit_i->mounted_time - now; 869 if (sit_i->elapsed_time >= diff) 870 return sit_i->elapsed_time - diff; 871 return 0; 872 } 873 return sit_i->elapsed_time; 874} 875 876static inline void set_summary(struct f2fs_summary *sum, nid_t nid, 877 unsigned int ofs_in_node, unsigned char version) 878{ 879 sum->nid = cpu_to_le32(nid); 880 sum->ofs_in_node = cpu_to_le16(ofs_in_node); 881 sum->version = version; 882} 883 884static inline block_t start_sum_block(struct f2fs_sb_info *sbi) 885{ 886 return __start_cp_addr(sbi) + 887 le32_to_cpu(F2FS_CKPT(sbi)->cp_pack_start_sum); 888} 889 890static inline block_t sum_blk_addr(struct f2fs_sb_info *sbi, int base, int type) 891{ 892 return __start_cp_addr(sbi) + 893 le32_to_cpu(F2FS_CKPT(sbi)->cp_pack_total_block_count) 894 - (base + 1) + type; 895} 896 897static inline bool sec_usage_check(struct f2fs_sb_info *sbi, unsigned int secno) 898{ 899 if (IS_CURSEC(sbi, secno) || (sbi->cur_victim_sec == secno)) 900 return true; 901 return false; 902} 903 904/* 905 * It is very important to gather dirty pages and write at once, so that we can 906 * submit a big bio without interfering other data writes. 907 * By default, 512 pages for directory data, 908 * 512 pages (2MB) * 8 for nodes, and 909 * 256 pages * 8 for meta are set. 910 */ 911static inline int nr_pages_to_skip(struct f2fs_sb_info *sbi, int type) 912{ 913 if (sbi->sb->s_bdi->wb.dirty_exceeded) 914 return 0; 915 916 if (type == DATA) 917 return sbi->blocks_per_seg; 918 else if (type == NODE) 919 return 8 * sbi->blocks_per_seg; 920 else if (type == META) 921 return 8 * BIO_MAX_VECS; 922 else 923 return 0; 924} 925 926/* 927 * When writing pages, it'd better align nr_to_write for segment size. 928 */ 929static inline long nr_pages_to_write(struct f2fs_sb_info *sbi, int type, 930 struct writeback_control *wbc) 931{ 932 long nr_to_write, desired; 933 934 if (wbc->sync_mode != WB_SYNC_NONE) 935 return 0; 936 937 nr_to_write = wbc->nr_to_write; 938 desired = BIO_MAX_VECS; 939 if (type == NODE) 940 desired <<= 1; 941 942 wbc->nr_to_write = desired; 943 return desired - nr_to_write; 944} 945 946static inline void wake_up_discard_thread(struct f2fs_sb_info *sbi, bool force) 947{ 948 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 949 bool wakeup = false; 950 int i; 951 952 if (force) 953 goto wake_up; 954 955 mutex_lock(&dcc->cmd_lock); 956 for (i = MAX_PLIST_NUM - 1; i >= 0; i--) { 957 if (i + 1 < dcc->discard_granularity) 958 break; 959 if (!list_empty(&dcc->pend_list[i])) { 960 wakeup = true; 961 break; 962 } 963 } 964 mutex_unlock(&dcc->cmd_lock); 965 if (!wakeup || !is_idle(sbi, DISCARD_TIME)) 966 return; 967wake_up: 968 dcc->discard_wake = true; 969 wake_up_interruptible_all(&dcc->discard_wait_queue); 970} 971