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 #ifdef CONFIG_F2FS_CHECK_FS 181 unsigned char *cur_valid_map_mir; /* mirror of current valid bitmap */ 182 #endif 183 /* 184 * # of valid blocks and the validity bitmap stored in the last 185 * checkpoint pack. This information is used by the SSR mode. 186 */ 187 unsigned char *ckpt_valid_map; /* validity bitmap of blocks last cp */ 188 unsigned char *discard_map; 189 unsigned long long mtime; /* modification time of the segment */ 190 }; 191 192 struct sec_entry { 193 unsigned int valid_blocks; /* # of valid blocks in a section */ 194 unsigned int ckpt_valid_blocks; /* # of valid blocks last cp in a section */ 195 }; 196 197 #define MAX_SKIP_GC_COUNT 16 198 199 struct revoke_entry { 200 struct list_head list; 201 block_t old_addr; /* for revoking when fail to commit */ 202 pgoff_t index; 203 }; 204 205 struct sit_info { 206 block_t sit_base_addr; /* start block address of SIT area */ 207 block_t sit_blocks; /* # of blocks used by SIT area */ 208 block_t written_valid_blocks; /* # of valid blocks in main area */ 209 char *bitmap; /* all bitmaps pointer */ 210 char *sit_bitmap; /* SIT bitmap pointer */ 211 #ifdef CONFIG_F2FS_CHECK_FS 212 char *sit_bitmap_mir; /* SIT bitmap mirror */ 213 214 /* bitmap of segments to be ignored by GC in case of errors */ 215 unsigned long *invalid_segmap; 216 #endif 217 unsigned int bitmap_size; /* SIT bitmap size */ 218 219 unsigned long *tmp_map; /* bitmap for temporal use */ 220 unsigned long *dirty_sentries_bitmap; /* bitmap for dirty sentries */ 221 unsigned int dirty_sentries; /* # of dirty sentries */ 222 unsigned int sents_per_block; /* # of SIT entries per block */ 223 struct rw_semaphore sentry_lock; /* to protect SIT cache */ 224 struct seg_entry *sentries; /* SIT segment-level cache */ 225 struct sec_entry *sec_entries; /* SIT section-level cache */ 226 227 /* for cost-benefit algorithm in cleaning procedure */ 228 unsigned long long elapsed_time; /* elapsed time after mount */ 229 unsigned long long mounted_time; /* mount time */ 230 unsigned long long min_mtime; /* min. modification time */ 231 unsigned long long max_mtime; /* max. modification time */ 232 unsigned long long dirty_min_mtime; /* rerange candidates in GC_AT */ 233 unsigned long long dirty_max_mtime; /* rerange candidates in GC_AT */ 234 235 unsigned int last_victim[MAX_GC_POLICY]; /* last victim segment # */ 236 }; 237 238 struct free_segmap_info { 239 unsigned int start_segno; /* start segment number logically */ 240 unsigned int free_segments; /* # of free segments */ 241 unsigned int free_sections; /* # of free sections */ 242 spinlock_t segmap_lock; /* free segmap lock */ 243 unsigned long *free_segmap; /* free segment bitmap */ 244 unsigned long *free_secmap; /* free section bitmap */ 245 }; 246 247 /* Notice: The order of dirty type is same with CURSEG_XXX in f2fs.h */ 248 enum dirty_type { 249 DIRTY_HOT_DATA, /* dirty segments assigned as hot data logs */ 250 DIRTY_WARM_DATA, /* dirty segments assigned as warm data logs */ 251 DIRTY_COLD_DATA, /* dirty segments assigned as cold data logs */ 252 DIRTY_HOT_NODE, /* dirty segments assigned as hot node logs */ 253 DIRTY_WARM_NODE, /* dirty segments assigned as warm node logs */ 254 DIRTY_COLD_NODE, /* dirty segments assigned as cold node logs */ 255 DIRTY, /* to count # of dirty segments */ 256 PRE, /* to count # of entirely obsolete segments */ 257 NR_DIRTY_TYPE 258 }; 259 260 struct dirty_seglist_info { 261 unsigned long *dirty_segmap[NR_DIRTY_TYPE]; 262 unsigned long *dirty_secmap; 263 struct mutex seglist_lock; /* lock for segment bitmaps */ 264 int nr_dirty[NR_DIRTY_TYPE]; /* # of dirty segments */ 265 unsigned long *victim_secmap; /* background GC victims */ 266 unsigned long *pinned_secmap; /* pinned victims from foreground GC */ 267 unsigned int pinned_secmap_cnt; /* count of victims which has pinned data */ 268 bool enable_pin_section; /* enable pinning section */ 269 }; 270 271 /* for active log information */ 272 struct curseg_info { 273 struct mutex curseg_mutex; /* lock for consistency */ 274 struct f2fs_summary_block *sum_blk; /* cached summary block */ 275 struct rw_semaphore journal_rwsem; /* protect journal area */ 276 struct f2fs_journal *journal; /* cached journal info */ 277 unsigned char alloc_type; /* current allocation type */ 278 unsigned short seg_type; /* segment type like CURSEG_XXX_TYPE */ 279 unsigned int segno; /* current segment number */ 280 unsigned short next_blkoff; /* next block offset to write */ 281 unsigned int zone; /* current zone number */ 282 unsigned int next_segno; /* preallocated segment */ 283 int fragment_remained_chunk; /* remained block size in a chunk for block fragmentation mode */ 284 bool inited; /* indicate inmem log is inited */ 285 }; 286 287 struct sit_entry_set { 288 struct list_head set_list; /* link with all sit sets */ 289 unsigned int start_segno; /* start segno of sits in set */ 290 unsigned int entry_cnt; /* the # of sit entries in set */ 291 }; 292 293 /* 294 * inline functions 295 */ 296 static inline struct curseg_info *CURSEG_I(struct f2fs_sb_info *sbi, int type) 297 { 298 return (struct curseg_info *)(SM_I(sbi)->curseg_array + type); 299 } 300 301 static inline bool is_curseg(struct f2fs_sb_info *sbi, unsigned int segno) 302 { 303 int i; 304 305 for (i = CURSEG_HOT_DATA; i < NO_CHECK_TYPE; i++) { 306 if (segno == CURSEG_I(sbi, i)->segno) 307 return true; 308 } 309 return false; 310 } 311 312 static inline bool is_cursec(struct f2fs_sb_info *sbi, unsigned int secno) 313 { 314 int i; 315 316 for (i = CURSEG_HOT_DATA; i < NO_CHECK_TYPE; i++) { 317 if (secno == GET_SEC_FROM_SEG(sbi, CURSEG_I(sbi, i)->segno)) 318 return true; 319 } 320 return false; 321 } 322 323 static inline struct seg_entry *get_seg_entry(struct f2fs_sb_info *sbi, 324 unsigned int segno) 325 { 326 struct sit_info *sit_i = SIT_I(sbi); 327 return &sit_i->sentries[segno]; 328 } 329 330 static inline struct sec_entry *get_sec_entry(struct f2fs_sb_info *sbi, 331 unsigned int segno) 332 { 333 struct sit_info *sit_i = SIT_I(sbi); 334 return &sit_i->sec_entries[GET_SEC_FROM_SEG(sbi, segno)]; 335 } 336 337 static inline unsigned int get_valid_blocks(struct f2fs_sb_info *sbi, 338 unsigned int segno, bool use_section) 339 { 340 /* 341 * In order to get # of valid blocks in a section instantly from many 342 * segments, f2fs manages two counting structures separately. 343 */ 344 if (use_section && __is_large_section(sbi)) 345 return get_sec_entry(sbi, segno)->valid_blocks; 346 else 347 return get_seg_entry(sbi, segno)->valid_blocks; 348 } 349 350 static inline unsigned int get_ckpt_valid_blocks(struct f2fs_sb_info *sbi, 351 unsigned int segno, bool use_section) 352 { 353 if (use_section && __is_large_section(sbi)) 354 return get_sec_entry(sbi, segno)->ckpt_valid_blocks; 355 else 356 return get_seg_entry(sbi, segno)->ckpt_valid_blocks; 357 } 358 359 static inline void set_ckpt_valid_blocks(struct f2fs_sb_info *sbi, 360 unsigned int segno) 361 { 362 unsigned int secno = GET_SEC_FROM_SEG(sbi, segno); 363 unsigned int start_segno = GET_SEG_FROM_SEC(sbi, secno); 364 unsigned int blocks = 0; 365 int i; 366 367 for (i = 0; i < SEGS_PER_SEC(sbi); i++, start_segno++) { 368 struct seg_entry *se = get_seg_entry(sbi, start_segno); 369 370 blocks += se->ckpt_valid_blocks; 371 } 372 get_sec_entry(sbi, segno)->ckpt_valid_blocks = blocks; 373 } 374 375 #ifdef CONFIG_F2FS_CHECK_FS 376 static inline void sanity_check_valid_blocks(struct f2fs_sb_info *sbi, 377 unsigned int segno) 378 { 379 unsigned int secno = GET_SEC_FROM_SEG(sbi, segno); 380 unsigned int start_segno = GET_SEG_FROM_SEC(sbi, secno); 381 unsigned int blocks = 0; 382 int i; 383 384 for (i = 0; i < SEGS_PER_SEC(sbi); i++, start_segno++) { 385 struct seg_entry *se = get_seg_entry(sbi, start_segno); 386 387 blocks += se->ckpt_valid_blocks; 388 } 389 390 if (blocks != get_sec_entry(sbi, segno)->ckpt_valid_blocks) { 391 f2fs_err(sbi, 392 "Inconsistent ckpt valid blocks: " 393 "seg entry(%d) vs sec entry(%d) at secno %d", 394 blocks, get_sec_entry(sbi, segno)->ckpt_valid_blocks, secno); 395 f2fs_bug_on(sbi, 1); 396 } 397 } 398 #else 399 static inline void sanity_check_valid_blocks(struct f2fs_sb_info *sbi, 400 unsigned int segno) 401 { 402 } 403 #endif 404 static inline void seg_info_from_raw_sit(struct seg_entry *se, 405 struct f2fs_sit_entry *rs) 406 { 407 se->valid_blocks = GET_SIT_VBLOCKS(rs); 408 se->ckpt_valid_blocks = GET_SIT_VBLOCKS(rs); 409 memcpy(se->cur_valid_map, rs->valid_map, SIT_VBLOCK_MAP_SIZE); 410 memcpy(se->ckpt_valid_map, rs->valid_map, SIT_VBLOCK_MAP_SIZE); 411 #ifdef CONFIG_F2FS_CHECK_FS 412 memcpy(se->cur_valid_map_mir, rs->valid_map, SIT_VBLOCK_MAP_SIZE); 413 #endif 414 se->type = GET_SIT_TYPE(rs); 415 se->mtime = le64_to_cpu(rs->mtime); 416 } 417 418 static inline void __seg_info_to_raw_sit(struct seg_entry *se, 419 struct f2fs_sit_entry *rs) 420 { 421 unsigned short raw_vblocks = (se->type << SIT_VBLOCKS_SHIFT) | 422 se->valid_blocks; 423 rs->vblocks = cpu_to_le16(raw_vblocks); 424 memcpy(rs->valid_map, se->cur_valid_map, SIT_VBLOCK_MAP_SIZE); 425 rs->mtime = cpu_to_le64(se->mtime); 426 } 427 428 static inline void seg_info_to_sit_folio(struct f2fs_sb_info *sbi, 429 struct folio *folio, unsigned int start) 430 { 431 struct f2fs_sit_block *raw_sit; 432 struct seg_entry *se; 433 struct f2fs_sit_entry *rs; 434 unsigned int end = min(start + SIT_ENTRY_PER_BLOCK, 435 (unsigned long)MAIN_SEGS(sbi)); 436 int i; 437 438 raw_sit = folio_address(folio); 439 memset(raw_sit, 0, PAGE_SIZE); 440 for (i = 0; i < end - start; i++) { 441 rs = &raw_sit->entries[i]; 442 se = get_seg_entry(sbi, start + i); 443 __seg_info_to_raw_sit(se, rs); 444 } 445 } 446 447 static inline void seg_info_to_raw_sit(struct seg_entry *se, 448 struct f2fs_sit_entry *rs) 449 { 450 __seg_info_to_raw_sit(se, rs); 451 452 memcpy(se->ckpt_valid_map, rs->valid_map, SIT_VBLOCK_MAP_SIZE); 453 se->ckpt_valid_blocks = se->valid_blocks; 454 } 455 456 static inline unsigned int find_next_inuse(struct free_segmap_info *free_i, 457 unsigned int max, unsigned int segno) 458 { 459 unsigned int ret; 460 spin_lock(&free_i->segmap_lock); 461 ret = find_next_bit(free_i->free_segmap, max, segno); 462 spin_unlock(&free_i->segmap_lock); 463 return ret; 464 } 465 466 static inline void __set_free(struct f2fs_sb_info *sbi, unsigned int segno) 467 { 468 struct free_segmap_info *free_i = FREE_I(sbi); 469 unsigned int secno = GET_SEC_FROM_SEG(sbi, segno); 470 unsigned int start_segno = GET_SEG_FROM_SEC(sbi, secno); 471 unsigned int next; 472 473 spin_lock(&free_i->segmap_lock); 474 clear_bit(segno, free_i->free_segmap); 475 free_i->free_segments++; 476 477 next = find_next_bit(free_i->free_segmap, 478 start_segno + SEGS_PER_SEC(sbi), start_segno); 479 if (next >= start_segno + f2fs_usable_segs_in_sec(sbi)) { 480 clear_bit(secno, free_i->free_secmap); 481 free_i->free_sections++; 482 } 483 spin_unlock(&free_i->segmap_lock); 484 } 485 486 static inline void __set_inuse(struct f2fs_sb_info *sbi, 487 unsigned int segno) 488 { 489 struct free_segmap_info *free_i = FREE_I(sbi); 490 unsigned int secno = GET_SEC_FROM_SEG(sbi, segno); 491 492 set_bit(segno, free_i->free_segmap); 493 free_i->free_segments--; 494 if (!test_and_set_bit(secno, free_i->free_secmap)) 495 free_i->free_sections--; 496 } 497 498 static inline void __set_test_and_free(struct f2fs_sb_info *sbi, 499 unsigned int segno, bool inmem) 500 { 501 struct free_segmap_info *free_i = FREE_I(sbi); 502 unsigned int secno = GET_SEC_FROM_SEG(sbi, segno); 503 unsigned int start_segno = GET_SEG_FROM_SEC(sbi, secno); 504 unsigned int next; 505 bool ret; 506 507 spin_lock(&free_i->segmap_lock); 508 ret = test_and_clear_bit(segno, free_i->free_segmap); 509 if (!ret) 510 goto unlock_out; 511 512 free_i->free_segments++; 513 514 if (!inmem && is_cursec(sbi, secno)) 515 goto unlock_out; 516 517 /* check large section */ 518 next = find_next_bit(free_i->free_segmap, 519 start_segno + SEGS_PER_SEC(sbi), start_segno); 520 if (next < start_segno + f2fs_usable_segs_in_sec(sbi)) 521 goto unlock_out; 522 523 ret = test_and_clear_bit(secno, free_i->free_secmap); 524 if (!ret) 525 goto unlock_out; 526 527 free_i->free_sections++; 528 529 if (GET_SEC_FROM_SEG(sbi, sbi->next_victim_seg[BG_GC]) == secno) 530 sbi->next_victim_seg[BG_GC] = NULL_SEGNO; 531 if (GET_SEC_FROM_SEG(sbi, sbi->next_victim_seg[FG_GC]) == secno) 532 sbi->next_victim_seg[FG_GC] = NULL_SEGNO; 533 534 unlock_out: 535 spin_unlock(&free_i->segmap_lock); 536 } 537 538 static inline void __set_test_and_inuse(struct f2fs_sb_info *sbi, 539 unsigned int segno) 540 { 541 struct free_segmap_info *free_i = FREE_I(sbi); 542 unsigned int secno = GET_SEC_FROM_SEG(sbi, segno); 543 544 spin_lock(&free_i->segmap_lock); 545 if (!test_and_set_bit(segno, free_i->free_segmap)) { 546 free_i->free_segments--; 547 if (!test_and_set_bit(secno, free_i->free_secmap)) 548 free_i->free_sections--; 549 } 550 spin_unlock(&free_i->segmap_lock); 551 } 552 553 static inline void get_sit_bitmap(struct f2fs_sb_info *sbi, 554 void *dst_addr) 555 { 556 struct sit_info *sit_i = SIT_I(sbi); 557 558 #ifdef CONFIG_F2FS_CHECK_FS 559 if (memcmp(sit_i->sit_bitmap, sit_i->sit_bitmap_mir, 560 sit_i->bitmap_size)) 561 f2fs_bug_on(sbi, 1); 562 #endif 563 memcpy(dst_addr, sit_i->sit_bitmap, sit_i->bitmap_size); 564 } 565 566 static inline block_t written_block_count(struct f2fs_sb_info *sbi) 567 { 568 return SIT_I(sbi)->written_valid_blocks; 569 } 570 571 static inline unsigned int free_segments(struct f2fs_sb_info *sbi) 572 { 573 return FREE_I(sbi)->free_segments; 574 } 575 576 static inline unsigned int reserved_segments(struct f2fs_sb_info *sbi) 577 { 578 return SM_I(sbi)->reserved_segments; 579 } 580 581 static inline unsigned int free_sections(struct f2fs_sb_info *sbi) 582 { 583 return FREE_I(sbi)->free_sections; 584 } 585 586 static inline unsigned int prefree_segments(struct f2fs_sb_info *sbi) 587 { 588 return DIRTY_I(sbi)->nr_dirty[PRE]; 589 } 590 591 static inline unsigned int dirty_segments(struct f2fs_sb_info *sbi) 592 { 593 return DIRTY_I(sbi)->nr_dirty[DIRTY_HOT_DATA] + 594 DIRTY_I(sbi)->nr_dirty[DIRTY_WARM_DATA] + 595 DIRTY_I(sbi)->nr_dirty[DIRTY_COLD_DATA] + 596 DIRTY_I(sbi)->nr_dirty[DIRTY_HOT_NODE] + 597 DIRTY_I(sbi)->nr_dirty[DIRTY_WARM_NODE] + 598 DIRTY_I(sbi)->nr_dirty[DIRTY_COLD_NODE]; 599 } 600 601 static inline int overprovision_segments(struct f2fs_sb_info *sbi) 602 { 603 return SM_I(sbi)->ovp_segments; 604 } 605 606 static inline int reserved_sections(struct f2fs_sb_info *sbi) 607 { 608 return GET_SEC_FROM_SEG(sbi, reserved_segments(sbi)); 609 } 610 611 static inline unsigned int get_left_section_blocks(struct f2fs_sb_info *sbi, 612 enum log_type type, unsigned int segno) 613 { 614 if (f2fs_lfs_mode(sbi)) { 615 unsigned int used_blocks = __is_large_section(sbi) ? SEGS_TO_BLKS(sbi, 616 (segno - GET_START_SEG_FROM_SEC(sbi, segno))) : 0; 617 return CAP_BLKS_PER_SEC(sbi) - used_blocks - 618 CURSEG_I(sbi, type)->next_blkoff; 619 } 620 return CAP_BLKS_PER_SEC(sbi) - get_ckpt_valid_blocks(sbi, segno, true); 621 } 622 623 static inline void get_additional_blocks_required(struct f2fs_sb_info *sbi, 624 unsigned int *total_node_blocks, unsigned int *total_data_blocks, 625 unsigned int *total_dent_blocks, bool separate_dent) 626 { 627 unsigned int segno, left_blocks; 628 int i; 629 unsigned int min_free_node_blocks = CAP_BLKS_PER_SEC(sbi); 630 unsigned int min_free_dent_blocks = CAP_BLKS_PER_SEC(sbi); 631 unsigned int min_free_data_blocks = CAP_BLKS_PER_SEC(sbi); 632 633 /* check current data/node sections in the worst case. */ 634 for (i = CURSEG_HOT_DATA; i < NR_PERSISTENT_LOG; i++) { 635 segno = CURSEG_I(sbi, i)->segno; 636 637 if (unlikely(segno == NULL_SEGNO)) 638 return; 639 640 left_blocks = get_left_section_blocks(sbi, i, segno); 641 642 if (i > CURSEG_COLD_DATA) 643 min_free_node_blocks = min(min_free_node_blocks, left_blocks); 644 else if (i == CURSEG_HOT_DATA && separate_dent) 645 min_free_dent_blocks = left_blocks; 646 else 647 min_free_data_blocks = min(min_free_data_blocks, left_blocks); 648 } 649 650 *total_node_blocks = (*total_node_blocks > min_free_node_blocks) ? 651 *total_node_blocks - min_free_node_blocks : 0; 652 *total_dent_blocks = (*total_dent_blocks > min_free_dent_blocks) ? 653 *total_dent_blocks - min_free_dent_blocks : 0; 654 *total_data_blocks = (*total_data_blocks > min_free_data_blocks) ? 655 *total_data_blocks - min_free_data_blocks : 0; 656 } 657 658 /* 659 * call get_additional_blocks_required to calculate dirty blocks 660 * needing to be placed in free sections, please note that, it 661 * needs to account dirty data as well in lfs mode when checkpoint 662 * is disabled. 663 */ 664 static inline int __get_secs_required(struct f2fs_sb_info *sbi) 665 { 666 unsigned int total_node_blocks = get_pages(sbi, F2FS_DIRTY_NODES) + 667 get_pages(sbi, F2FS_DIRTY_DENTS) + 668 get_pages(sbi, F2FS_DIRTY_IMETA); 669 unsigned int total_dent_blocks = get_pages(sbi, F2FS_DIRTY_DENTS); 670 unsigned int total_data_blocks = 0; 671 bool separate_dent = true; 672 673 if (f2fs_lfs_mode(sbi)) 674 total_data_blocks = get_pages(sbi, F2FS_DIRTY_DATA); 675 676 /* 677 * When active_logs != 4, dentry blocks and data blocks can be 678 * mixed in the same logs, so check their space together. 679 */ 680 if (F2FS_OPTION(sbi).active_logs != 4) { 681 total_data_blocks += total_dent_blocks; 682 total_dent_blocks = 0; 683 separate_dent = false; 684 } 685 686 get_additional_blocks_required(sbi, &total_node_blocks, &total_dent_blocks, 687 &total_data_blocks, separate_dent); 688 689 return DIV_ROUND_UP(total_node_blocks, CAP_BLKS_PER_SEC(sbi)) + 690 DIV_ROUND_UP(total_dent_blocks, CAP_BLKS_PER_SEC(sbi)) + 691 DIV_ROUND_UP(total_data_blocks, CAP_BLKS_PER_SEC(sbi)); 692 } 693 694 static inline bool has_not_enough_free_secs(struct f2fs_sb_info *sbi, 695 int freed, int needed) 696 { 697 unsigned int free_secs, required_secs; 698 699 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING))) 700 return false; 701 702 free_secs = free_sections(sbi) + freed; 703 required_secs = needed + reserved_sections(sbi) + 704 __get_secs_required(sbi); 705 706 return free_secs < required_secs; 707 } 708 709 static inline bool has_enough_free_secs(struct f2fs_sb_info *sbi, 710 int freed, int needed) 711 { 712 return !has_not_enough_free_secs(sbi, freed, needed); 713 } 714 715 static inline bool has_enough_free_blks(struct f2fs_sb_info *sbi) 716 { 717 unsigned int total_free_blocks = 0; 718 unsigned int avail_user_block_count; 719 720 spin_lock(&sbi->stat_lock); 721 722 avail_user_block_count = get_available_block_count(sbi, NULL, true); 723 total_free_blocks = avail_user_block_count - (unsigned int)valid_user_blocks(sbi); 724 725 spin_unlock(&sbi->stat_lock); 726 727 return total_free_blocks > 0; 728 } 729 730 static inline bool f2fs_is_checkpoint_ready(struct f2fs_sb_info *sbi) 731 { 732 if (likely(!is_sbi_flag_set(sbi, SBI_CP_DISABLED))) 733 return true; 734 if (likely(has_enough_free_secs(sbi, 0, 0))) 735 return true; 736 if (!f2fs_lfs_mode(sbi) && 737 likely(has_enough_free_blks(sbi))) 738 return true; 739 return false; 740 } 741 742 static inline bool excess_prefree_segs(struct f2fs_sb_info *sbi) 743 { 744 return prefree_segments(sbi) > SM_I(sbi)->rec_prefree_segments; 745 } 746 747 static inline int utilization(struct f2fs_sb_info *sbi) 748 { 749 return div_u64((u64)valid_user_blocks(sbi) * 100, 750 sbi->user_block_count); 751 } 752 753 /* 754 * Sometimes f2fs may be better to drop out-of-place update policy. 755 * And, users can control the policy through sysfs entries. 756 * There are five policies with triggering conditions as follows. 757 * F2FS_IPU_FORCE - all the time, 758 * F2FS_IPU_SSR - if SSR mode is activated, 759 * F2FS_IPU_UTIL - if FS utilization is over threashold, 760 * F2FS_IPU_SSR_UTIL - if SSR mode is activated and FS utilization is over 761 * threashold, 762 * F2FS_IPU_FSYNC - activated in fsync path only for high performance flash 763 * storages. IPU will be triggered only if the # of dirty 764 * pages over min_fsync_blocks. (=default option) 765 * F2FS_IPU_ASYNC - do IPU given by asynchronous write requests. 766 * F2FS_IPU_NOCACHE - disable IPU bio cache. 767 * F2FS_IPU_HONOR_OPU_WRITE - use OPU write prior to IPU write if inode has 768 * FI_OPU_WRITE flag. 769 * F2FS_IPU_DISABLE - disable IPU. (=default option in LFS mode) 770 */ 771 #define DEF_MIN_IPU_UTIL 70 772 #define DEF_MIN_FSYNC_BLOCKS 8 773 #define DEF_MIN_HOT_BLOCKS 16 774 775 #define SMALL_VOLUME_SEGMENTS (16 * 512) /* 16GB */ 776 777 #define F2FS_IPU_DISABLE 0 778 779 /* Modification on enum should be synchronized with ipu_mode_names array */ 780 enum { 781 F2FS_IPU_FORCE, 782 F2FS_IPU_SSR, 783 F2FS_IPU_UTIL, 784 F2FS_IPU_SSR_UTIL, 785 F2FS_IPU_FSYNC, 786 F2FS_IPU_ASYNC, 787 F2FS_IPU_NOCACHE, 788 F2FS_IPU_HONOR_OPU_WRITE, 789 F2FS_IPU_MAX, 790 }; 791 792 static inline bool IS_F2FS_IPU_DISABLE(struct f2fs_sb_info *sbi) 793 { 794 return SM_I(sbi)->ipu_policy == F2FS_IPU_DISABLE; 795 } 796 797 #define F2FS_IPU_POLICY(name) \ 798 static inline bool IS_##name(struct f2fs_sb_info *sbi) \ 799 { \ 800 return SM_I(sbi)->ipu_policy & BIT(name); \ 801 } 802 803 F2FS_IPU_POLICY(F2FS_IPU_FORCE); 804 F2FS_IPU_POLICY(F2FS_IPU_SSR); 805 F2FS_IPU_POLICY(F2FS_IPU_UTIL); 806 F2FS_IPU_POLICY(F2FS_IPU_SSR_UTIL); 807 F2FS_IPU_POLICY(F2FS_IPU_FSYNC); 808 F2FS_IPU_POLICY(F2FS_IPU_ASYNC); 809 F2FS_IPU_POLICY(F2FS_IPU_NOCACHE); 810 F2FS_IPU_POLICY(F2FS_IPU_HONOR_OPU_WRITE); 811 812 static inline unsigned int curseg_segno(struct f2fs_sb_info *sbi, 813 int type) 814 { 815 struct curseg_info *curseg = CURSEG_I(sbi, type); 816 return curseg->segno; 817 } 818 819 static inline unsigned char curseg_alloc_type(struct f2fs_sb_info *sbi, 820 int type) 821 { 822 struct curseg_info *curseg = CURSEG_I(sbi, type); 823 return curseg->alloc_type; 824 } 825 826 static inline bool valid_main_segno(struct f2fs_sb_info *sbi, 827 unsigned int segno) 828 { 829 return segno <= (MAIN_SEGS(sbi) - 1); 830 } 831 832 static inline void verify_fio_blkaddr(struct f2fs_io_info *fio) 833 { 834 struct f2fs_sb_info *sbi = fio->sbi; 835 836 if (__is_valid_data_blkaddr(fio->old_blkaddr)) 837 verify_blkaddr(sbi, fio->old_blkaddr, __is_meta_io(fio) ? 838 META_GENERIC : DATA_GENERIC); 839 verify_blkaddr(sbi, fio->new_blkaddr, __is_meta_io(fio) ? 840 META_GENERIC : DATA_GENERIC_ENHANCE); 841 } 842 843 /* 844 * Summary block is always treated as an invalid block 845 */ 846 static inline int check_block_count(struct f2fs_sb_info *sbi, 847 int segno, struct f2fs_sit_entry *raw_sit) 848 { 849 bool is_valid = test_bit_le(0, raw_sit->valid_map) ? true : false; 850 int valid_blocks = 0; 851 int cur_pos = 0, next_pos; 852 unsigned int usable_blks_per_seg = f2fs_usable_blks_in_seg(sbi, segno); 853 854 /* check bitmap with valid block count */ 855 do { 856 if (is_valid) { 857 next_pos = find_next_zero_bit_le(&raw_sit->valid_map, 858 usable_blks_per_seg, 859 cur_pos); 860 valid_blocks += next_pos - cur_pos; 861 } else 862 next_pos = find_next_bit_le(&raw_sit->valid_map, 863 usable_blks_per_seg, 864 cur_pos); 865 cur_pos = next_pos; 866 is_valid = !is_valid; 867 } while (cur_pos < usable_blks_per_seg); 868 869 if (unlikely(GET_SIT_VBLOCKS(raw_sit) != valid_blocks)) { 870 f2fs_err(sbi, "Mismatch valid blocks %d vs. %d", 871 GET_SIT_VBLOCKS(raw_sit), valid_blocks); 872 set_sbi_flag(sbi, SBI_NEED_FSCK); 873 f2fs_handle_error(sbi, ERROR_INCONSISTENT_SIT); 874 return -EFSCORRUPTED; 875 } 876 877 if (usable_blks_per_seg < BLKS_PER_SEG(sbi)) 878 f2fs_bug_on(sbi, find_next_bit_le(&raw_sit->valid_map, 879 BLKS_PER_SEG(sbi), 880 usable_blks_per_seg) != BLKS_PER_SEG(sbi)); 881 882 /* check segment usage, and check boundary of a given segment number */ 883 if (unlikely(GET_SIT_VBLOCKS(raw_sit) > usable_blks_per_seg 884 || !valid_main_segno(sbi, segno))) { 885 f2fs_err(sbi, "Wrong valid blocks %d or segno %u", 886 GET_SIT_VBLOCKS(raw_sit), segno); 887 set_sbi_flag(sbi, SBI_NEED_FSCK); 888 f2fs_handle_error(sbi, ERROR_INCONSISTENT_SIT); 889 return -EFSCORRUPTED; 890 } 891 return 0; 892 } 893 894 static inline pgoff_t current_sit_addr(struct f2fs_sb_info *sbi, 895 unsigned int start) 896 { 897 struct sit_info *sit_i = SIT_I(sbi); 898 unsigned int offset = SIT_BLOCK_OFFSET(start); 899 block_t blk_addr = sit_i->sit_base_addr + offset; 900 901 f2fs_bug_on(sbi, !valid_main_segno(sbi, start)); 902 903 #ifdef CONFIG_F2FS_CHECK_FS 904 if (f2fs_test_bit(offset, sit_i->sit_bitmap) != 905 f2fs_test_bit(offset, sit_i->sit_bitmap_mir)) 906 f2fs_bug_on(sbi, 1); 907 #endif 908 909 /* calculate sit block address */ 910 if (f2fs_test_bit(offset, sit_i->sit_bitmap)) 911 blk_addr += sit_i->sit_blocks; 912 913 return blk_addr; 914 } 915 916 static inline pgoff_t next_sit_addr(struct f2fs_sb_info *sbi, 917 pgoff_t block_addr) 918 { 919 struct sit_info *sit_i = SIT_I(sbi); 920 block_addr -= sit_i->sit_base_addr; 921 if (block_addr < sit_i->sit_blocks) 922 block_addr += sit_i->sit_blocks; 923 else 924 block_addr -= sit_i->sit_blocks; 925 926 return block_addr + sit_i->sit_base_addr; 927 } 928 929 static inline void set_to_next_sit(struct sit_info *sit_i, unsigned int start) 930 { 931 unsigned int block_off = SIT_BLOCK_OFFSET(start); 932 933 f2fs_change_bit(block_off, sit_i->sit_bitmap); 934 #ifdef CONFIG_F2FS_CHECK_FS 935 f2fs_change_bit(block_off, sit_i->sit_bitmap_mir); 936 #endif 937 } 938 939 static inline unsigned long long get_mtime(struct f2fs_sb_info *sbi, 940 bool base_time) 941 { 942 struct sit_info *sit_i = SIT_I(sbi); 943 time64_t diff, now = ktime_get_boottime_seconds(); 944 945 if (now >= sit_i->mounted_time) 946 return sit_i->elapsed_time + now - sit_i->mounted_time; 947 948 /* system time is set to the past */ 949 if (!base_time) { 950 diff = sit_i->mounted_time - now; 951 if (sit_i->elapsed_time >= diff) 952 return sit_i->elapsed_time - diff; 953 return 0; 954 } 955 return sit_i->elapsed_time; 956 } 957 958 static inline void set_summary(struct f2fs_summary *sum, nid_t nid, 959 unsigned int ofs_in_node, unsigned char version) 960 { 961 sum->nid = cpu_to_le32(nid); 962 sum->ofs_in_node = cpu_to_le16(ofs_in_node); 963 sum->version = version; 964 } 965 966 static inline block_t start_sum_block(struct f2fs_sb_info *sbi) 967 { 968 return __start_cp_addr(sbi) + 969 le32_to_cpu(F2FS_CKPT(sbi)->cp_pack_start_sum); 970 } 971 972 static inline block_t sum_blk_addr(struct f2fs_sb_info *sbi, int base, int type) 973 { 974 return __start_cp_addr(sbi) + 975 le32_to_cpu(F2FS_CKPT(sbi)->cp_pack_total_block_count) 976 - (base + 1) + type; 977 } 978 979 static inline bool sec_usage_check(struct f2fs_sb_info *sbi, unsigned int secno) 980 { 981 if (is_cursec(sbi, secno) || (sbi->cur_victim_sec == secno)) 982 return true; 983 return false; 984 } 985 986 /* 987 * It is very important to gather dirty pages and write at once, so that we can 988 * submit a big bio without interfering other data writes. 989 * By default, 512 pages for directory data, 990 * 512 pages (2MB) * 8 for nodes, and 991 * 256 pages * 8 for meta are set. 992 */ 993 static inline int nr_pages_to_skip(struct f2fs_sb_info *sbi, int type) 994 { 995 if (sbi->sb->s_bdi->wb.dirty_exceeded) 996 return 0; 997 998 if (type == DATA) 999 return BLKS_PER_SEG(sbi); 1000 else if (type == NODE) 1001 return SEGS_TO_BLKS(sbi, 8); 1002 else if (type == META) 1003 return 8 * BIO_MAX_VECS; 1004 else 1005 return 0; 1006 } 1007 1008 /* 1009 * When writing pages, it'd better align nr_to_write for segment size. 1010 */ 1011 static inline long nr_pages_to_write(struct f2fs_sb_info *sbi, int type, 1012 struct writeback_control *wbc) 1013 { 1014 long nr_to_write, desired; 1015 1016 if (wbc->sync_mode != WB_SYNC_NONE) 1017 return 0; 1018 1019 nr_to_write = wbc->nr_to_write; 1020 desired = BIO_MAX_VECS; 1021 if (type == NODE) 1022 desired <<= 1; 1023 1024 wbc->nr_to_write = desired; 1025 return desired - nr_to_write; 1026 } 1027 1028 static inline void wake_up_discard_thread(struct f2fs_sb_info *sbi, bool force) 1029 { 1030 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 1031 bool wakeup = false; 1032 int i; 1033 1034 if (force) 1035 goto wake_up; 1036 1037 mutex_lock(&dcc->cmd_lock); 1038 for (i = MAX_PLIST_NUM - 1; i >= 0; i--) { 1039 if (i + 1 < dcc->discard_granularity) 1040 break; 1041 if (!list_empty(&dcc->pend_list[i])) { 1042 wakeup = true; 1043 break; 1044 } 1045 } 1046 mutex_unlock(&dcc->cmd_lock); 1047 if (!wakeup || !is_idle(sbi, DISCARD_TIME)) 1048 return; 1049 wake_up: 1050 dcc->discard_wake = true; 1051 wake_up_interruptible_all(&dcc->discard_wait_queue); 1052 } 1053