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 #define INVALID_MTIME ULLONG_MAX /* no valid blocks in a segment/section */ 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 && (t) <= CURSEG_COLD_NODE) 29 #define SE_PAGETYPE(se) ((IS_NODESEG((se)->type) ? NODE : DATA)) 30 31 static inline void sanity_check_seg_type(struct f2fs_sb_info *sbi, 32 unsigned short seg_type) 33 { 34 f2fs_bug_on(sbi, seg_type >= NR_PERSISTENT_LOG); 35 } 36 37 #define MAIN_BLKADDR(sbi) \ 38 (SM_I(sbi) ? SM_I(sbi)->main_blkaddr : \ 39 le32_to_cpu(F2FS_RAW_SUPER(sbi)->main_blkaddr)) 40 #define SEG0_BLKADDR(sbi) \ 41 (SM_I(sbi) ? SM_I(sbi)->seg0_blkaddr : \ 42 le32_to_cpu(F2FS_RAW_SUPER(sbi)->segment0_blkaddr)) 43 44 #define MAIN_SEGS(sbi) (SM_I(sbi)->main_segments) 45 #define MAIN_SECS(sbi) ((sbi)->total_sections) 46 47 #define TOTAL_SEGS(sbi) \ 48 (SM_I(sbi) ? SM_I(sbi)->segment_count : \ 49 le32_to_cpu(F2FS_RAW_SUPER(sbi)->segment_count)) 50 #define TOTAL_BLKS(sbi) (SEGS_TO_BLKS(sbi, TOTAL_SEGS(sbi))) 51 52 #define MAX_BLKADDR(sbi) (SEG0_BLKADDR(sbi) + TOTAL_BLKS(sbi)) 53 #define SEGMENT_SIZE(sbi) (1ULL << ((sbi)->log_blocksize + \ 54 (sbi)->log_blocks_per_seg)) 55 56 #define START_BLOCK(sbi, segno) (SEG0_BLKADDR(sbi) + \ 57 (SEGS_TO_BLKS(sbi, GET_R2L_SEGNO(FREE_I(sbi), segno)))) 58 59 #define NEXT_FREE_BLKADDR(sbi, curseg) \ 60 (START_BLOCK(sbi, (curseg)->segno) + (curseg)->next_blkoff) 61 62 #define GET_SEGOFF_FROM_SEG0(sbi, blk_addr) ((blk_addr) - SEG0_BLKADDR(sbi)) 63 #define GET_SEGNO_FROM_SEG0(sbi, blk_addr) \ 64 (BLKS_TO_SEGS(sbi, GET_SEGOFF_FROM_SEG0(sbi, blk_addr))) 65 #define GET_BLKOFF_FROM_SEG0(sbi, blk_addr) \ 66 (GET_SEGOFF_FROM_SEG0(sbi, blk_addr) & (BLKS_PER_SEG(sbi) - 1)) 67 68 #define GET_SEGNO(sbi, blk_addr) \ 69 ((!__is_valid_data_blkaddr(blk_addr)) ? \ 70 NULL_SEGNO : GET_L2R_SEGNO(FREE_I(sbi), \ 71 GET_SEGNO_FROM_SEG0(sbi, blk_addr))) 72 #ifdef CONFIG_BLK_DEV_ZONED 73 #define CAP_BLKS_PER_SEC(sbi) \ 74 (BLKS_PER_SEC(sbi) - (sbi)->unusable_blocks_per_sec) 75 #define CAP_SEGS_PER_SEC(sbi) \ 76 (SEGS_PER_SEC(sbi) - \ 77 BLKS_TO_SEGS(sbi, (sbi)->unusable_blocks_per_sec)) 78 #else 79 #define CAP_BLKS_PER_SEC(sbi) BLKS_PER_SEC(sbi) 80 #define CAP_SEGS_PER_SEC(sbi) SEGS_PER_SEC(sbi) 81 #endif 82 #define GET_START_SEG_FROM_SEC(sbi, segno) \ 83 (rounddown(segno, SEGS_PER_SEC(sbi))) 84 #define GET_SEC_FROM_SEG(sbi, segno) \ 85 (((segno) == -1) ? -1 : (segno) / SEGS_PER_SEC(sbi)) 86 #define GET_SEG_FROM_SEC(sbi, secno) \ 87 ((secno) * SEGS_PER_SEC(sbi)) 88 #define GET_ZONE_FROM_SEC(sbi, secno) \ 89 (((secno) == -1) ? -1 : (secno) / (sbi)->secs_per_zone) 90 #define GET_ZONE_FROM_SEG(sbi, segno) \ 91 GET_ZONE_FROM_SEC(sbi, GET_SEC_FROM_SEG(sbi, segno)) 92 93 #define GET_SUM_BLOCK(sbi, segno) \ 94 (SM_I(sbi)->ssa_blkaddr + (segno / (sbi)->sums_per_block)) 95 #define GET_SUM_BLKOFF(sbi, segno) (segno % (sbi)->sums_per_block) 96 #define SUM_BLK_PAGE_ADDR(sbi, folio, segno) \ 97 (folio_address(folio) + GET_SUM_BLKOFF(sbi, segno) * (sbi)->sum_blocksize) 98 99 #define GET_SUM_TYPE(footer) ((footer)->entry_type) 100 #define SET_SUM_TYPE(footer, type) ((footer)->entry_type = (type)) 101 102 #define SIT_ENTRY_OFFSET(sit_i, segno) \ 103 ((segno) % (sit_i)->sents_per_block) 104 #define SIT_BLOCK_OFFSET(segno) \ 105 ((segno) / SIT_ENTRY_PER_BLOCK) 106 #define START_SEGNO(segno) \ 107 (SIT_BLOCK_OFFSET(segno) * SIT_ENTRY_PER_BLOCK) 108 #define SIT_BLK_CNT(sbi) \ 109 DIV_ROUND_UP(MAIN_SEGS(sbi), SIT_ENTRY_PER_BLOCK) 110 #define f2fs_bitmap_size(nr) \ 111 (BITS_TO_LONGS(nr) * sizeof(unsigned long)) 112 113 #define SECTOR_FROM_BLOCK(blk_addr) \ 114 (((sector_t)blk_addr) << F2FS_LOG_SECTORS_PER_BLOCK) 115 #define SECTOR_TO_BLOCK(sectors) \ 116 ((sectors) >> F2FS_LOG_SECTORS_PER_BLOCK) 117 118 /* 119 * In the victim_sel_policy->alloc_mode, there are three block allocation modes. 120 * LFS writes data sequentially with cleaning operations. 121 * SSR (Slack Space Recycle) reuses obsolete space without cleaning operations. 122 * AT_SSR (Age Threshold based Slack Space Recycle) merges fragments into 123 * fragmented segment which has similar aging degree. 124 */ 125 enum { 126 LFS = 0, 127 SSR, 128 AT_SSR, 129 }; 130 131 /* 132 * In the victim_sel_policy->gc_mode, there are three gc, aka cleaning, modes. 133 * GC_CB is based on cost-benefit algorithm. 134 * GC_GREEDY is based on greedy algorithm. 135 * GC_AT is based on age-threshold algorithm. 136 */ 137 enum { 138 GC_CB = 0, 139 GC_GREEDY, 140 GC_AT, 141 ALLOC_NEXT, 142 FLUSH_DEVICE, 143 MAX_GC_POLICY, 144 }; 145 146 /* 147 * BG_GC means the background cleaning job. 148 * FG_GC means the on-demand cleaning job. 149 */ 150 enum { 151 BG_GC = 0, 152 FG_GC, 153 }; 154 155 /* for a function parameter to select a victim segment */ 156 struct victim_sel_policy { 157 int alloc_mode; /* LFS or SSR */ 158 int gc_mode; /* GC_CB or GC_GREEDY */ 159 unsigned long *dirty_bitmap; /* dirty segment/section bitmap */ 160 unsigned int max_search; /* 161 * maximum # of segments/sections 162 * to search 163 */ 164 unsigned int offset; /* last scanned bitmap offset */ 165 unsigned int ofs_unit; /* bitmap search unit */ 166 unsigned int min_cost; /* minimum cost */ 167 unsigned long long oldest_age; /* oldest age of segments having the same min cost */ 168 unsigned int min_segno; /* segment # having min. cost */ 169 unsigned long long age; /* mtime of GCed section*/ 170 unsigned long long age_threshold;/* age threshold */ 171 bool one_time_gc; /* one time GC */ 172 }; 173 174 struct seg_entry { 175 unsigned int type:6; /* segment type like CURSEG_XXX_TYPE */ 176 unsigned int valid_blocks:10; /* # of valid blocks */ 177 unsigned int ckpt_valid_blocks:10; /* # of valid blocks last cp */ 178 unsigned int padding:6; /* padding */ 179 unsigned char *cur_valid_map; /* validity bitmap of blocks */ 180 /* 181 * # of valid blocks and the validity bitmap stored in the last 182 * checkpoint pack. This information is used by the SSR mode. 183 */ 184 unsigned char *ckpt_valid_map; /* validity bitmap of blocks last cp */ 185 unsigned char *discard_map; 186 unsigned long long mtime; /* modification time of the segment */ 187 }; 188 189 struct sec_entry { 190 unsigned int valid_blocks; /* # of valid blocks in a section */ 191 unsigned int ckpt_valid_blocks; /* # of valid blocks last cp in a section */ 192 }; 193 194 #define MAX_SKIP_GC_COUNT 16 195 196 struct revoke_entry { 197 struct list_head list; 198 block_t old_addr; /* for revoking when fail to commit */ 199 pgoff_t index; 200 }; 201 202 struct sit_info { 203 block_t sit_base_addr; /* start block address of SIT area */ 204 block_t sit_blocks; /* # of blocks used by SIT area */ 205 block_t written_valid_blocks; /* # of valid blocks in main area */ 206 char *bitmap; /* all bitmaps pointer */ 207 char *sit_bitmap; /* SIT bitmap pointer */ 208 #ifdef CONFIG_F2FS_CHECK_FS 209 /* bitmap of segments to be ignored by GC in case of errors */ 210 unsigned long *invalid_segmap; 211 #endif 212 unsigned int bitmap_size; /* SIT bitmap size */ 213 214 unsigned long *tmp_map; /* bitmap for temporal use */ 215 unsigned long *dirty_sentries_bitmap; /* bitmap for dirty sentries */ 216 unsigned int dirty_sentries; /* # of dirty sentries */ 217 unsigned int sents_per_block; /* # of SIT entries per block */ 218 struct rw_semaphore sentry_lock; /* to protect SIT cache */ 219 struct seg_entry *sentries; /* SIT segment-level cache */ 220 struct sec_entry *sec_entries; /* SIT section-level cache */ 221 222 /* for cost-benefit algorithm in cleaning procedure */ 223 unsigned long long elapsed_time; /* elapsed time after mount */ 224 unsigned long long mounted_time; /* mount time */ 225 unsigned long long min_mtime; /* min. modification time */ 226 unsigned long long max_mtime; /* max. modification time */ 227 unsigned long long dirty_min_mtime; /* rerange candidates in GC_AT */ 228 unsigned long long dirty_max_mtime; /* rerange candidates in GC_AT */ 229 230 unsigned int last_victim[MAX_GC_POLICY]; /* last victim segment # */ 231 }; 232 233 struct free_segmap_info { 234 unsigned int start_segno; /* start segment number logically */ 235 unsigned int free_segments; /* # of free segments */ 236 unsigned int free_sections; /* # of free sections */ 237 spinlock_t segmap_lock; /* free segmap lock */ 238 unsigned long *free_segmap; /* free segment bitmap */ 239 unsigned long *free_secmap; /* free section bitmap */ 240 }; 241 242 /* Notice: The order of dirty type is same with CURSEG_XXX in f2fs.h */ 243 enum dirty_type { 244 DIRTY_HOT_DATA, /* dirty segments assigned as hot data logs */ 245 DIRTY_WARM_DATA, /* dirty segments assigned as warm data logs */ 246 DIRTY_COLD_DATA, /* dirty segments assigned as cold data logs */ 247 DIRTY_HOT_NODE, /* dirty segments assigned as hot node logs */ 248 DIRTY_WARM_NODE, /* dirty segments assigned as warm node logs */ 249 DIRTY_COLD_NODE, /* dirty segments assigned as cold node logs */ 250 DIRTY, /* to count # of dirty segments */ 251 PRE, /* to count # of entirely obsolete segments */ 252 NR_DIRTY_TYPE 253 }; 254 255 struct dirty_seglist_info { 256 unsigned long *dirty_segmap[NR_DIRTY_TYPE]; 257 unsigned long *dirty_secmap; 258 struct mutex seglist_lock; /* lock for segment bitmaps */ 259 int nr_dirty[NR_DIRTY_TYPE]; /* # of dirty segments */ 260 unsigned long *victim_secmap; /* background GC victims */ 261 unsigned long *pinned_secmap; /* pinned victims from foreground GC */ 262 unsigned int pinned_secmap_cnt; /* count of victims which has pinned data */ 263 bool enable_pin_section; /* enable pinning section */ 264 }; 265 266 /* for active log information */ 267 struct curseg_info { 268 struct mutex curseg_mutex; /* lock for consistency */ 269 struct f2fs_summary_block *sum_blk; /* cached summary block */ 270 struct rw_semaphore journal_rwsem; /* protect journal area */ 271 struct f2fs_journal *journal; /* cached journal info */ 272 unsigned char alloc_type; /* current allocation type */ 273 unsigned short seg_type; /* segment type like CURSEG_XXX_TYPE */ 274 unsigned int segno; /* current segment number */ 275 unsigned short next_blkoff; /* next block offset to write */ 276 unsigned int zone; /* current zone number */ 277 unsigned int next_segno; /* preallocated segment */ 278 int fragment_remained_chunk; /* remained block size in a chunk for block fragmentation mode */ 279 bool inited; /* indicate inmem log is inited */ 280 }; 281 282 struct sit_entry_set { 283 struct list_head set_list; /* link with all sit sets */ 284 unsigned int start_segno; /* start segno of sits in set */ 285 unsigned int entry_cnt; /* the # of sit entries in set */ 286 }; 287 288 /* 289 * inline functions 290 */ 291 static inline struct curseg_info *CURSEG_I(struct f2fs_sb_info *sbi, int type) 292 { 293 return (struct curseg_info *)(SM_I(sbi)->curseg_array + type); 294 } 295 296 static inline bool is_curseg(struct f2fs_sb_info *sbi, unsigned int segno) 297 { 298 int i; 299 300 for (i = CURSEG_HOT_DATA; i < NO_CHECK_TYPE; i++) { 301 if (segno == CURSEG_I(sbi, i)->segno) 302 return true; 303 } 304 return false; 305 } 306 307 static inline bool is_cursec(struct f2fs_sb_info *sbi, unsigned int secno) 308 { 309 int i; 310 311 for (i = CURSEG_HOT_DATA; i < NO_CHECK_TYPE; i++) { 312 if (secno == GET_SEC_FROM_SEG(sbi, CURSEG_I(sbi, i)->segno)) 313 return true; 314 } 315 return false; 316 } 317 318 static inline struct seg_entry *get_seg_entry(struct f2fs_sb_info *sbi, 319 unsigned int segno) 320 { 321 struct sit_info *sit_i = SIT_I(sbi); 322 return &sit_i->sentries[segno]; 323 } 324 325 static inline struct sec_entry *get_sec_entry(struct f2fs_sb_info *sbi, 326 unsigned int segno) 327 { 328 struct sit_info *sit_i = SIT_I(sbi); 329 return &sit_i->sec_entries[GET_SEC_FROM_SEG(sbi, segno)]; 330 } 331 332 static inline unsigned int get_valid_blocks(struct f2fs_sb_info *sbi, 333 unsigned int segno, bool use_section) 334 { 335 /* 336 * In order to get # of valid blocks in a section instantly from many 337 * segments, f2fs manages two counting structures separately. 338 */ 339 if (use_section && __is_large_section(sbi)) 340 return get_sec_entry(sbi, segno)->valid_blocks; 341 else 342 return get_seg_entry(sbi, segno)->valid_blocks; 343 } 344 345 static inline unsigned int get_ckpt_valid_blocks(struct f2fs_sb_info *sbi, 346 unsigned int segno, bool use_section) 347 { 348 if (use_section && __is_large_section(sbi)) 349 return get_sec_entry(sbi, segno)->ckpt_valid_blocks; 350 else 351 return get_seg_entry(sbi, segno)->ckpt_valid_blocks; 352 } 353 354 static inline void set_ckpt_valid_blocks(struct f2fs_sb_info *sbi, 355 unsigned int segno) 356 { 357 unsigned int secno = GET_SEC_FROM_SEG(sbi, segno); 358 unsigned int start_segno = GET_SEG_FROM_SEC(sbi, secno); 359 unsigned int blocks = 0; 360 int i; 361 362 for (i = 0; i < SEGS_PER_SEC(sbi); i++, start_segno++) { 363 struct seg_entry *se = get_seg_entry(sbi, start_segno); 364 365 blocks += se->ckpt_valid_blocks; 366 } 367 get_sec_entry(sbi, segno)->ckpt_valid_blocks = blocks; 368 } 369 370 #ifdef CONFIG_F2FS_CHECK_FS 371 static inline void sanity_check_valid_blocks(struct f2fs_sb_info *sbi, 372 unsigned int segno) 373 { 374 unsigned int secno = GET_SEC_FROM_SEG(sbi, segno); 375 unsigned int start_segno = GET_SEG_FROM_SEC(sbi, secno); 376 unsigned int blocks = 0; 377 int i; 378 379 for (i = 0; i < SEGS_PER_SEC(sbi); i++, start_segno++) { 380 struct seg_entry *se = get_seg_entry(sbi, start_segno); 381 382 blocks += se->ckpt_valid_blocks; 383 } 384 385 if (blocks != get_sec_entry(sbi, segno)->ckpt_valid_blocks) { 386 f2fs_err(sbi, 387 "Inconsistent ckpt valid blocks: " 388 "seg entry(%d) vs sec entry(%d) at secno %d", 389 blocks, get_sec_entry(sbi, segno)->ckpt_valid_blocks, secno); 390 f2fs_bug_on(sbi, 1); 391 } 392 } 393 #else 394 static inline void sanity_check_valid_blocks(struct f2fs_sb_info *sbi, 395 unsigned int segno) 396 { 397 } 398 #endif 399 static inline void seg_info_from_raw_sit(struct seg_entry *se, 400 struct f2fs_sit_entry *rs) 401 { 402 se->valid_blocks = GET_SIT_VBLOCKS(rs); 403 se->ckpt_valid_blocks = GET_SIT_VBLOCKS(rs); 404 memcpy(se->cur_valid_map, rs->valid_map, SIT_VBLOCK_MAP_SIZE); 405 memcpy(se->ckpt_valid_map, rs->valid_map, SIT_VBLOCK_MAP_SIZE); 406 se->type = GET_SIT_TYPE(rs); 407 se->mtime = le64_to_cpu(rs->mtime); 408 } 409 410 static inline void __seg_info_to_raw_sit(struct seg_entry *se, 411 struct f2fs_sit_entry *rs) 412 { 413 unsigned short raw_vblocks = (se->type << SIT_VBLOCKS_SHIFT) | 414 se->valid_blocks; 415 rs->vblocks = cpu_to_le16(raw_vblocks); 416 memcpy(rs->valid_map, se->cur_valid_map, SIT_VBLOCK_MAP_SIZE); 417 rs->mtime = cpu_to_le64(se->mtime); 418 } 419 420 static inline void seg_info_to_sit_folio(struct f2fs_sb_info *sbi, 421 struct folio *folio, unsigned int start) 422 { 423 struct f2fs_sit_block *raw_sit; 424 struct seg_entry *se; 425 struct f2fs_sit_entry *rs; 426 unsigned int end = min(start + SIT_ENTRY_PER_BLOCK, 427 (unsigned long)MAIN_SEGS(sbi)); 428 int i; 429 430 raw_sit = folio_address(folio); 431 memset(raw_sit, 0, PAGE_SIZE); 432 for (i = 0; i < end - start; i++) { 433 rs = &raw_sit->entries[i]; 434 se = get_seg_entry(sbi, start + i); 435 __seg_info_to_raw_sit(se, rs); 436 } 437 } 438 439 static inline void seg_info_to_raw_sit(struct seg_entry *se, 440 struct f2fs_sit_entry *rs) 441 { 442 __seg_info_to_raw_sit(se, rs); 443 444 memcpy(se->ckpt_valid_map, rs->valid_map, SIT_VBLOCK_MAP_SIZE); 445 se->ckpt_valid_blocks = se->valid_blocks; 446 } 447 448 static inline unsigned int find_next_inuse(struct free_segmap_info *free_i, 449 unsigned int max, unsigned int segno) 450 { 451 unsigned int ret; 452 spin_lock(&free_i->segmap_lock); 453 ret = find_next_bit(free_i->free_segmap, max, segno); 454 spin_unlock(&free_i->segmap_lock); 455 return ret; 456 } 457 458 static inline void __set_free(struct f2fs_sb_info *sbi, unsigned int segno) 459 { 460 struct free_segmap_info *free_i = FREE_I(sbi); 461 unsigned int secno = GET_SEC_FROM_SEG(sbi, segno); 462 unsigned int start_segno = GET_SEG_FROM_SEC(sbi, secno); 463 unsigned int next; 464 465 spin_lock(&free_i->segmap_lock); 466 clear_bit(segno, free_i->free_segmap); 467 free_i->free_segments++; 468 469 next = find_next_bit(free_i->free_segmap, 470 start_segno + SEGS_PER_SEC(sbi), start_segno); 471 if (next >= start_segno + f2fs_usable_segs_in_sec(sbi)) { 472 clear_bit(secno, free_i->free_secmap); 473 free_i->free_sections++; 474 } 475 spin_unlock(&free_i->segmap_lock); 476 } 477 478 static inline void __set_inuse(struct f2fs_sb_info *sbi, 479 unsigned int segno) 480 { 481 struct free_segmap_info *free_i = FREE_I(sbi); 482 unsigned int secno = GET_SEC_FROM_SEG(sbi, segno); 483 484 set_bit(segno, free_i->free_segmap); 485 free_i->free_segments--; 486 if (!test_and_set_bit(secno, free_i->free_secmap)) 487 free_i->free_sections--; 488 } 489 490 static inline void __set_test_and_free(struct f2fs_sb_info *sbi, 491 unsigned int segno, bool inmem) 492 { 493 struct free_segmap_info *free_i = FREE_I(sbi); 494 unsigned int secno = GET_SEC_FROM_SEG(sbi, segno); 495 unsigned int start_segno = GET_SEG_FROM_SEC(sbi, secno); 496 unsigned int next; 497 bool ret; 498 499 spin_lock(&free_i->segmap_lock); 500 ret = test_and_clear_bit(segno, free_i->free_segmap); 501 if (!ret) 502 goto unlock_out; 503 504 free_i->free_segments++; 505 506 if (!inmem && is_cursec(sbi, secno)) 507 goto unlock_out; 508 509 /* check large section */ 510 next = find_next_bit(free_i->free_segmap, 511 start_segno + SEGS_PER_SEC(sbi), start_segno); 512 if (next < start_segno + f2fs_usable_segs_in_sec(sbi)) 513 goto unlock_out; 514 515 ret = test_and_clear_bit(secno, free_i->free_secmap); 516 if (!ret) 517 goto unlock_out; 518 519 free_i->free_sections++; 520 521 if (GET_SEC_FROM_SEG(sbi, sbi->next_victim_seg[BG_GC]) == secno) 522 sbi->next_victim_seg[BG_GC] = NULL_SEGNO; 523 if (GET_SEC_FROM_SEG(sbi, sbi->next_victim_seg[FG_GC]) == secno) 524 sbi->next_victim_seg[FG_GC] = NULL_SEGNO; 525 526 unlock_out: 527 spin_unlock(&free_i->segmap_lock); 528 } 529 530 static inline void __set_test_and_inuse(struct f2fs_sb_info *sbi, 531 unsigned int segno) 532 { 533 struct free_segmap_info *free_i = FREE_I(sbi); 534 unsigned int secno = GET_SEC_FROM_SEG(sbi, segno); 535 536 spin_lock(&free_i->segmap_lock); 537 if (!test_and_set_bit(segno, free_i->free_segmap)) { 538 free_i->free_segments--; 539 if (!test_and_set_bit(secno, free_i->free_secmap)) 540 free_i->free_sections--; 541 } 542 spin_unlock(&free_i->segmap_lock); 543 } 544 545 static inline void get_sit_bitmap(struct f2fs_sb_info *sbi, 546 void *dst_addr) 547 { 548 struct sit_info *sit_i = SIT_I(sbi); 549 550 memcpy(dst_addr, sit_i->sit_bitmap, sit_i->bitmap_size); 551 } 552 553 static inline block_t written_block_count(struct f2fs_sb_info *sbi) 554 { 555 return SIT_I(sbi)->written_valid_blocks; 556 } 557 558 static inline unsigned int free_segments(struct f2fs_sb_info *sbi) 559 { 560 return FREE_I(sbi)->free_segments; 561 } 562 563 static inline unsigned int reserved_segments(struct f2fs_sb_info *sbi) 564 { 565 return SM_I(sbi)->reserved_segments; 566 } 567 568 static inline unsigned int free_sections(struct f2fs_sb_info *sbi) 569 { 570 return FREE_I(sbi)->free_sections; 571 } 572 573 static inline unsigned int prefree_segments(struct f2fs_sb_info *sbi) 574 { 575 return DIRTY_I(sbi)->nr_dirty[PRE]; 576 } 577 578 static inline unsigned int dirty_segments(struct f2fs_sb_info *sbi) 579 { 580 return DIRTY_I(sbi)->nr_dirty[DIRTY_HOT_DATA] + 581 DIRTY_I(sbi)->nr_dirty[DIRTY_WARM_DATA] + 582 DIRTY_I(sbi)->nr_dirty[DIRTY_COLD_DATA] + 583 DIRTY_I(sbi)->nr_dirty[DIRTY_HOT_NODE] + 584 DIRTY_I(sbi)->nr_dirty[DIRTY_WARM_NODE] + 585 DIRTY_I(sbi)->nr_dirty[DIRTY_COLD_NODE]; 586 } 587 588 static inline int overprovision_segments(struct f2fs_sb_info *sbi) 589 { 590 return SM_I(sbi)->ovp_segments; 591 } 592 593 static inline int reserved_sections(struct f2fs_sb_info *sbi) 594 { 595 return GET_SEC_FROM_SEG(sbi, reserved_segments(sbi)); 596 } 597 598 static inline unsigned int get_left_section_blocks(struct f2fs_sb_info *sbi, 599 enum log_type type, unsigned int segno) 600 { 601 if (f2fs_lfs_mode(sbi)) { 602 unsigned int used_blocks = __is_large_section(sbi) ? SEGS_TO_BLKS(sbi, 603 (segno - GET_START_SEG_FROM_SEC(sbi, segno))) : 0; 604 return CAP_BLKS_PER_SEC(sbi) - used_blocks - 605 CURSEG_I(sbi, type)->next_blkoff; 606 } 607 return CAP_BLKS_PER_SEC(sbi) - get_ckpt_valid_blocks(sbi, segno, true); 608 } 609 610 static inline void get_additional_blocks_required(struct f2fs_sb_info *sbi, 611 unsigned int *total_node_blocks, unsigned int *total_data_blocks, 612 unsigned int *total_dent_blocks, bool separate_dent) 613 { 614 unsigned int segno, left_blocks; 615 int i; 616 unsigned int min_free_node_blocks = CAP_BLKS_PER_SEC(sbi); 617 unsigned int min_free_dent_blocks = CAP_BLKS_PER_SEC(sbi); 618 unsigned int min_free_data_blocks = CAP_BLKS_PER_SEC(sbi); 619 620 /* check current data/node sections in the worst case. */ 621 for (i = CURSEG_HOT_DATA; i < NR_PERSISTENT_LOG; i++) { 622 segno = CURSEG_I(sbi, i)->segno; 623 624 if (unlikely(segno == NULL_SEGNO)) 625 return; 626 627 left_blocks = get_left_section_blocks(sbi, i, segno); 628 629 if (i > CURSEG_COLD_DATA) 630 min_free_node_blocks = min(min_free_node_blocks, left_blocks); 631 else if (i == CURSEG_HOT_DATA && separate_dent) 632 min_free_dent_blocks = left_blocks; 633 else 634 min_free_data_blocks = min(min_free_data_blocks, left_blocks); 635 } 636 637 *total_node_blocks = (*total_node_blocks > min_free_node_blocks) ? 638 *total_node_blocks - min_free_node_blocks : 0; 639 *total_dent_blocks = (*total_dent_blocks > min_free_dent_blocks) ? 640 *total_dent_blocks - min_free_dent_blocks : 0; 641 *total_data_blocks = (*total_data_blocks > min_free_data_blocks) ? 642 *total_data_blocks - min_free_data_blocks : 0; 643 } 644 645 /* 646 * call get_additional_blocks_required to calculate dirty blocks 647 * needing to be placed in free sections, please note that, it 648 * needs to account dirty data as well in lfs mode when checkpoint 649 * is disabled. 650 */ 651 static inline int __get_secs_required(struct f2fs_sb_info *sbi) 652 { 653 unsigned int total_node_blocks = get_pages(sbi, F2FS_DIRTY_NODES) + 654 get_pages(sbi, F2FS_DIRTY_DENTS) + 655 get_pages(sbi, F2FS_DIRTY_IMETA); 656 unsigned int total_dent_blocks = get_pages(sbi, F2FS_DIRTY_DENTS); 657 unsigned int total_data_blocks = 0; 658 bool separate_dent = true; 659 660 if (f2fs_lfs_mode(sbi)) 661 total_data_blocks = get_pages(sbi, F2FS_DIRTY_DATA); 662 663 /* 664 * When active_logs != 4, dentry blocks and data blocks can be 665 * mixed in the same logs, so check their space together. 666 */ 667 if (F2FS_OPTION(sbi).active_logs != 4) { 668 total_data_blocks += total_dent_blocks; 669 total_dent_blocks = 0; 670 separate_dent = false; 671 } 672 673 get_additional_blocks_required(sbi, &total_node_blocks, &total_dent_blocks, 674 &total_data_blocks, separate_dent); 675 676 return DIV_ROUND_UP(total_node_blocks, CAP_BLKS_PER_SEC(sbi)) + 677 DIV_ROUND_UP(total_dent_blocks, CAP_BLKS_PER_SEC(sbi)) + 678 DIV_ROUND_UP(total_data_blocks, CAP_BLKS_PER_SEC(sbi)); 679 } 680 681 static inline bool has_not_enough_free_secs(struct f2fs_sb_info *sbi, 682 int freed, int needed) 683 { 684 unsigned int free_secs, required_secs; 685 686 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING))) 687 return false; 688 689 free_secs = free_sections(sbi) + freed; 690 required_secs = needed + reserved_sections(sbi) + 691 __get_secs_required(sbi); 692 693 return free_secs < required_secs; 694 } 695 696 static inline bool has_enough_free_secs(struct f2fs_sb_info *sbi, 697 int freed, int needed) 698 { 699 return !has_not_enough_free_secs(sbi, freed, needed); 700 } 701 702 static inline bool has_enough_free_blks(struct f2fs_sb_info *sbi) 703 { 704 unsigned int total_free_blocks = 0; 705 unsigned int avail_user_block_count; 706 707 spin_lock(&sbi->stat_lock); 708 709 avail_user_block_count = get_available_block_count(sbi, NULL, true); 710 total_free_blocks = avail_user_block_count - (unsigned int)valid_user_blocks(sbi); 711 712 spin_unlock(&sbi->stat_lock); 713 714 return total_free_blocks > 0; 715 } 716 717 static inline bool f2fs_is_checkpoint_ready(struct f2fs_sb_info *sbi) 718 { 719 if (likely(!is_sbi_flag_set(sbi, SBI_CP_DISABLED))) 720 return true; 721 if (likely(has_enough_free_secs(sbi, 0, 0))) 722 return true; 723 if (!f2fs_lfs_mode(sbi) && 724 likely(has_enough_free_blks(sbi))) 725 return true; 726 return false; 727 } 728 729 static inline bool excess_prefree_segs(struct f2fs_sb_info *sbi) 730 { 731 return prefree_segments(sbi) > SM_I(sbi)->rec_prefree_segments; 732 } 733 734 static inline int utilization(struct f2fs_sb_info *sbi) 735 { 736 return div_u64((u64)valid_user_blocks(sbi) * 100, 737 sbi->user_block_count); 738 } 739 740 /* 741 * Sometimes f2fs may be better to drop out-of-place update policy. 742 * And, users can control the policy through sysfs entries. 743 * There are five policies with triggering conditions as follows. 744 * F2FS_IPU_FORCE - all the time, 745 * F2FS_IPU_SSR - if SSR mode is activated, 746 * F2FS_IPU_UTIL - if FS utilization is over threashold, 747 * F2FS_IPU_SSR_UTIL - if SSR mode is activated and FS utilization is over 748 * threashold, 749 * F2FS_IPU_FSYNC - activated in fsync path only for high performance flash 750 * storages. IPU will be triggered only if the # of dirty 751 * pages over min_fsync_blocks. (=default option) 752 * F2FS_IPU_ASYNC - do IPU given by asynchronous write requests. 753 * F2FS_IPU_NOCACHE - disable IPU bio cache. 754 * F2FS_IPU_HONOR_OPU_WRITE - use OPU write prior to IPU write if inode has 755 * FI_OPU_WRITE flag. 756 * F2FS_IPU_DISABLE - disable IPU. (=default option in LFS mode) 757 */ 758 #define DEF_MIN_IPU_UTIL 70 759 #define DEF_MIN_FSYNC_BLOCKS 8 760 #define DEF_MIN_HOT_BLOCKS 16 761 762 #define SMALL_VOLUME_SEGMENTS (16 * 512) /* 16GB */ 763 764 #define F2FS_IPU_DISABLE 0 765 766 /* Modification on enum should be synchronized with ipu_mode_names array */ 767 enum { 768 F2FS_IPU_FORCE, 769 F2FS_IPU_SSR, 770 F2FS_IPU_UTIL, 771 F2FS_IPU_SSR_UTIL, 772 F2FS_IPU_FSYNC, 773 F2FS_IPU_ASYNC, 774 F2FS_IPU_NOCACHE, 775 F2FS_IPU_HONOR_OPU_WRITE, 776 F2FS_IPU_MAX, 777 }; 778 779 static inline bool IS_F2FS_IPU_DISABLE(struct f2fs_sb_info *sbi) 780 { 781 return SM_I(sbi)->ipu_policy == F2FS_IPU_DISABLE; 782 } 783 784 #define F2FS_IPU_POLICY(name) \ 785 static inline bool IS_##name(struct f2fs_sb_info *sbi) \ 786 { \ 787 return SM_I(sbi)->ipu_policy & BIT(name); \ 788 } 789 790 F2FS_IPU_POLICY(F2FS_IPU_FORCE); 791 F2FS_IPU_POLICY(F2FS_IPU_SSR); 792 F2FS_IPU_POLICY(F2FS_IPU_UTIL); 793 F2FS_IPU_POLICY(F2FS_IPU_SSR_UTIL); 794 F2FS_IPU_POLICY(F2FS_IPU_FSYNC); 795 F2FS_IPU_POLICY(F2FS_IPU_ASYNC); 796 F2FS_IPU_POLICY(F2FS_IPU_NOCACHE); 797 F2FS_IPU_POLICY(F2FS_IPU_HONOR_OPU_WRITE); 798 799 static inline unsigned int curseg_segno(struct f2fs_sb_info *sbi, 800 int type) 801 { 802 struct curseg_info *curseg = CURSEG_I(sbi, type); 803 return curseg->segno; 804 } 805 806 static inline unsigned char curseg_alloc_type(struct f2fs_sb_info *sbi, 807 int type) 808 { 809 struct curseg_info *curseg = CURSEG_I(sbi, type); 810 return curseg->alloc_type; 811 } 812 813 static inline bool valid_main_segno(struct f2fs_sb_info *sbi, 814 unsigned int segno) 815 { 816 return segno <= (MAIN_SEGS(sbi) - 1); 817 } 818 819 static inline void verify_fio_blkaddr(struct f2fs_io_info *fio) 820 { 821 struct f2fs_sb_info *sbi = fio->sbi; 822 823 if (__is_valid_data_blkaddr(fio->old_blkaddr)) 824 verify_blkaddr(sbi, fio->old_blkaddr, __is_meta_io(fio) ? 825 META_GENERIC : DATA_GENERIC); 826 verify_blkaddr(sbi, fio->new_blkaddr, __is_meta_io(fio) ? 827 META_GENERIC : DATA_GENERIC_ENHANCE); 828 } 829 830 /* 831 * Summary block is always treated as an invalid block 832 */ 833 static inline int check_block_count(struct f2fs_sb_info *sbi, 834 int segno, struct f2fs_sit_entry *raw_sit) 835 { 836 bool is_valid = test_bit_le(0, raw_sit->valid_map) ? true : false; 837 int valid_blocks = 0; 838 int cur_pos = 0, next_pos; 839 unsigned int usable_blks_per_seg = f2fs_usable_blks_in_seg(sbi, segno); 840 841 /* check bitmap with valid block count */ 842 do { 843 if (is_valid) { 844 next_pos = find_next_zero_bit_le(&raw_sit->valid_map, 845 usable_blks_per_seg, 846 cur_pos); 847 valid_blocks += next_pos - cur_pos; 848 } else 849 next_pos = find_next_bit_le(&raw_sit->valid_map, 850 usable_blks_per_seg, 851 cur_pos); 852 cur_pos = next_pos; 853 is_valid = !is_valid; 854 } while (cur_pos < usable_blks_per_seg); 855 856 if (unlikely(GET_SIT_VBLOCKS(raw_sit) != valid_blocks)) { 857 f2fs_err(sbi, "Mismatch valid blocks %d vs. %d", 858 GET_SIT_VBLOCKS(raw_sit), valid_blocks); 859 set_sbi_flag(sbi, SBI_NEED_FSCK); 860 f2fs_handle_error(sbi, ERROR_INCONSISTENT_SIT); 861 return -EFSCORRUPTED; 862 } 863 864 if (usable_blks_per_seg < BLKS_PER_SEG(sbi)) 865 f2fs_bug_on(sbi, find_next_bit_le(&raw_sit->valid_map, 866 BLKS_PER_SEG(sbi), 867 usable_blks_per_seg) != BLKS_PER_SEG(sbi)); 868 869 /* check segment usage, and check boundary of a given segment number */ 870 if (unlikely(GET_SIT_VBLOCKS(raw_sit) > usable_blks_per_seg 871 || !valid_main_segno(sbi, segno))) { 872 f2fs_err(sbi, "Wrong valid blocks %d or segno %u", 873 GET_SIT_VBLOCKS(raw_sit), segno); 874 set_sbi_flag(sbi, SBI_NEED_FSCK); 875 f2fs_handle_error(sbi, ERROR_INCONSISTENT_SIT); 876 return -EFSCORRUPTED; 877 } 878 return 0; 879 } 880 881 static inline pgoff_t current_sit_addr(struct f2fs_sb_info *sbi, 882 unsigned int start) 883 { 884 struct sit_info *sit_i = SIT_I(sbi); 885 unsigned int offset = SIT_BLOCK_OFFSET(start); 886 block_t blk_addr = sit_i->sit_base_addr + offset; 887 888 f2fs_bug_on(sbi, !valid_main_segno(sbi, start)); 889 890 /* calculate sit block address */ 891 if (f2fs_test_bit(offset, sit_i->sit_bitmap)) 892 blk_addr += sit_i->sit_blocks; 893 894 return blk_addr; 895 } 896 897 static inline pgoff_t next_sit_addr(struct f2fs_sb_info *sbi, 898 pgoff_t block_addr) 899 { 900 struct sit_info *sit_i = SIT_I(sbi); 901 block_addr -= sit_i->sit_base_addr; 902 if (block_addr < sit_i->sit_blocks) 903 block_addr += sit_i->sit_blocks; 904 else 905 block_addr -= sit_i->sit_blocks; 906 907 return block_addr + sit_i->sit_base_addr; 908 } 909 910 static inline void set_to_next_sit(struct sit_info *sit_i, unsigned int start) 911 { 912 unsigned int block_off = SIT_BLOCK_OFFSET(start); 913 914 f2fs_change_bit(block_off, sit_i->sit_bitmap); 915 } 916 917 static inline unsigned long long get_mtime(struct f2fs_sb_info *sbi, 918 bool base_time) 919 { 920 struct sit_info *sit_i = SIT_I(sbi); 921 time64_t diff, now = ktime_get_boottime_seconds(); 922 923 if (now >= sit_i->mounted_time) 924 return sit_i->elapsed_time + now - sit_i->mounted_time; 925 926 /* system time is set to the past */ 927 if (!base_time) { 928 diff = sit_i->mounted_time - now; 929 if (sit_i->elapsed_time >= diff) 930 return sit_i->elapsed_time - diff; 931 return 0; 932 } 933 return sit_i->elapsed_time; 934 } 935 936 static inline void set_summary(struct f2fs_summary *sum, nid_t nid, 937 unsigned int ofs_in_node, unsigned char version) 938 { 939 sum->nid = cpu_to_le32(nid); 940 sum->ofs_in_node = cpu_to_le16(ofs_in_node); 941 sum->version = version; 942 } 943 944 static inline block_t start_sum_block(struct f2fs_sb_info *sbi) 945 { 946 return __start_cp_addr(sbi) + 947 le32_to_cpu(F2FS_CKPT(sbi)->cp_pack_start_sum); 948 } 949 950 static inline block_t sum_blk_addr(struct f2fs_sb_info *sbi, int base, int type) 951 { 952 return __start_cp_addr(sbi) + 953 le32_to_cpu(F2FS_CKPT(sbi)->cp_pack_total_block_count) 954 - (base + 1) + type; 955 } 956 957 static inline bool sec_usage_check(struct f2fs_sb_info *sbi, unsigned int secno) 958 { 959 if (is_cursec(sbi, secno) || (sbi->cur_victim_sec == secno)) 960 return true; 961 return false; 962 } 963 964 /* 965 * It is very important to gather dirty pages and write at once, so that we can 966 * submit a big bio without interfering other data writes. 967 * By default, 512 pages for directory data, 968 * 512 pages (2MB) * 8 for nodes, and 969 * 256 pages * 8 for meta are set. 970 */ 971 static inline int nr_pages_to_skip(struct f2fs_sb_info *sbi, int type) 972 { 973 if (bdi_wb_dirty_exceeded(sbi->sb->s_bdi)) 974 return 0; 975 976 if (type == DATA) 977 return BLKS_PER_SEG(sbi); 978 else if (type == NODE) 979 return SEGS_TO_BLKS(sbi, 8); 980 else if (type == META) 981 return 8 * BIO_MAX_VECS; 982 else 983 return 0; 984 } 985 986 /* 987 * When writing pages, it'd better align nr_to_write for segment size. 988 */ 989 static inline long nr_pages_to_write(struct f2fs_sb_info *sbi, int type, 990 struct writeback_control *wbc) 991 { 992 long nr_to_write, desired; 993 994 if (wbc->sync_mode != WB_SYNC_NONE) 995 return 0; 996 997 nr_to_write = wbc->nr_to_write; 998 desired = BIO_MAX_VECS; 999 if (type == NODE) 1000 desired <<= 1; 1001 1002 wbc->nr_to_write = desired; 1003 return desired - nr_to_write; 1004 } 1005 1006 static inline void wake_up_discard_thread(struct f2fs_sb_info *sbi, bool force) 1007 { 1008 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 1009 bool wakeup = false; 1010 int i; 1011 1012 if (force) 1013 goto wake_up; 1014 1015 mutex_lock(&dcc->cmd_lock); 1016 for (i = MAX_PLIST_NUM - 1; i >= 0; i--) { 1017 if (i + 1 < dcc->discard_granularity) 1018 break; 1019 if (!list_empty(&dcc->pend_list[i])) { 1020 wakeup = true; 1021 break; 1022 } 1023 } 1024 mutex_unlock(&dcc->cmd_lock); 1025 if (!wakeup || !is_idle(sbi, DISCARD_TIME)) 1026 return; 1027 wake_up: 1028 dcc->discard_wake = true; 1029 wake_up_interruptible_all(&dcc->discard_wait_queue); 1030 } 1031