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 SUMS_PER_BLOCK (F2FS_BLKSIZE / F2FS_SUM_BLKSIZE) 94 #define GET_SUM_BLOCK(sbi, segno) \ 95 (SM_I(sbi)->ssa_blkaddr + (segno / SUMS_PER_BLOCK)) 96 #define GET_SUM_BLKOFF(segno) (segno % SUMS_PER_BLOCK) 97 #define SUM_BLK_PAGE_ADDR(folio, segno) \ 98 (folio_address(folio) + GET_SUM_BLKOFF(segno) * F2FS_SUM_BLKSIZE) 99 100 #define GET_SUM_TYPE(footer) ((footer)->entry_type) 101 #define SET_SUM_TYPE(footer, type) ((footer)->entry_type = (type)) 102 103 #define SIT_ENTRY_OFFSET(sit_i, segno) \ 104 ((segno) % (sit_i)->sents_per_block) 105 #define SIT_BLOCK_OFFSET(segno) \ 106 ((segno) / SIT_ENTRY_PER_BLOCK) 107 #define START_SEGNO(segno) \ 108 (SIT_BLOCK_OFFSET(segno) * SIT_ENTRY_PER_BLOCK) 109 #define SIT_BLK_CNT(sbi) \ 110 DIV_ROUND_UP(MAIN_SEGS(sbi), SIT_ENTRY_PER_BLOCK) 111 #define f2fs_bitmap_size(nr) \ 112 (BITS_TO_LONGS(nr) * sizeof(unsigned long)) 113 114 #define SECTOR_FROM_BLOCK(blk_addr) \ 115 (((sector_t)blk_addr) << F2FS_LOG_SECTORS_PER_BLOCK) 116 #define SECTOR_TO_BLOCK(sectors) \ 117 ((sectors) >> F2FS_LOG_SECTORS_PER_BLOCK) 118 119 /* 120 * In the victim_sel_policy->alloc_mode, there are three block allocation modes. 121 * LFS writes data sequentially with cleaning operations. 122 * SSR (Slack Space Recycle) reuses obsolete space without cleaning operations. 123 * AT_SSR (Age Threshold based Slack Space Recycle) merges fragments into 124 * fragmented segment which has similar aging degree. 125 */ 126 enum { 127 LFS = 0, 128 SSR, 129 AT_SSR, 130 }; 131 132 /* 133 * In the victim_sel_policy->gc_mode, there are three gc, aka cleaning, modes. 134 * GC_CB is based on cost-benefit algorithm. 135 * GC_GREEDY is based on greedy algorithm. 136 * GC_AT is based on age-threshold algorithm. 137 */ 138 enum { 139 GC_CB = 0, 140 GC_GREEDY, 141 GC_AT, 142 ALLOC_NEXT, 143 FLUSH_DEVICE, 144 MAX_GC_POLICY, 145 }; 146 147 /* 148 * BG_GC means the background cleaning job. 149 * FG_GC means the on-demand cleaning job. 150 */ 151 enum { 152 BG_GC = 0, 153 FG_GC, 154 }; 155 156 /* for a function parameter to select a victim segment */ 157 struct victim_sel_policy { 158 int alloc_mode; /* LFS or SSR */ 159 int gc_mode; /* GC_CB or GC_GREEDY */ 160 unsigned long *dirty_bitmap; /* dirty segment/section bitmap */ 161 unsigned int max_search; /* 162 * maximum # of segments/sections 163 * to search 164 */ 165 unsigned int offset; /* last scanned bitmap offset */ 166 unsigned int ofs_unit; /* bitmap search unit */ 167 unsigned int min_cost; /* minimum cost */ 168 unsigned long long oldest_age; /* oldest age of segments having the same min cost */ 169 unsigned int min_segno; /* segment # having min. cost */ 170 unsigned long long age; /* mtime of GCed section*/ 171 unsigned long long age_threshold;/* age threshold */ 172 bool one_time_gc; /* one time GC */ 173 }; 174 175 struct seg_entry { 176 unsigned int type:6; /* segment type like CURSEG_XXX_TYPE */ 177 unsigned int valid_blocks:10; /* # of valid blocks */ 178 unsigned int ckpt_valid_blocks:10; /* # of valid blocks last cp */ 179 unsigned int padding:6; /* padding */ 180 unsigned char *cur_valid_map; /* validity bitmap of blocks */ 181 #ifdef CONFIG_F2FS_CHECK_FS 182 unsigned char *cur_valid_map_mir; /* mirror of current valid bitmap */ 183 #endif 184 /* 185 * # of valid blocks and the validity bitmap stored in the last 186 * checkpoint pack. This information is used by the SSR mode. 187 */ 188 unsigned char *ckpt_valid_map; /* validity bitmap of blocks last cp */ 189 unsigned char *discard_map; 190 unsigned long long mtime; /* modification time of the segment */ 191 }; 192 193 struct sec_entry { 194 unsigned int valid_blocks; /* # of valid blocks in a section */ 195 unsigned int ckpt_valid_blocks; /* # of valid blocks last cp in a section */ 196 }; 197 198 #define MAX_SKIP_GC_COUNT 16 199 200 struct revoke_entry { 201 struct list_head list; 202 block_t old_addr; /* for revoking when fail to commit */ 203 pgoff_t index; 204 }; 205 206 struct sit_info { 207 block_t sit_base_addr; /* start block address of SIT area */ 208 block_t sit_blocks; /* # of blocks used by SIT area */ 209 block_t written_valid_blocks; /* # of valid blocks in main area */ 210 char *bitmap; /* all bitmaps pointer */ 211 char *sit_bitmap; /* SIT bitmap pointer */ 212 #ifdef CONFIG_F2FS_CHECK_FS 213 char *sit_bitmap_mir; /* SIT bitmap mirror */ 214 215 /* bitmap of segments to be ignored by GC in case of errors */ 216 unsigned long *invalid_segmap; 217 #endif 218 unsigned int bitmap_size; /* SIT bitmap size */ 219 220 unsigned long *tmp_map; /* bitmap for temporal use */ 221 unsigned long *dirty_sentries_bitmap; /* bitmap for dirty sentries */ 222 unsigned int dirty_sentries; /* # of dirty sentries */ 223 unsigned int sents_per_block; /* # of SIT entries per block */ 224 struct rw_semaphore sentry_lock; /* to protect SIT cache */ 225 struct seg_entry *sentries; /* SIT segment-level cache */ 226 struct sec_entry *sec_entries; /* SIT section-level cache */ 227 228 /* for cost-benefit algorithm in cleaning procedure */ 229 unsigned long long elapsed_time; /* elapsed time after mount */ 230 unsigned long long mounted_time; /* mount time */ 231 unsigned long long min_mtime; /* min. modification time */ 232 unsigned long long max_mtime; /* max. modification time */ 233 unsigned long long dirty_min_mtime; /* rerange candidates in GC_AT */ 234 unsigned long long dirty_max_mtime; /* rerange candidates in GC_AT */ 235 236 unsigned int last_victim[MAX_GC_POLICY]; /* last victim segment # */ 237 }; 238 239 struct free_segmap_info { 240 unsigned int start_segno; /* start segment number logically */ 241 unsigned int free_segments; /* # of free segments */ 242 unsigned int free_sections; /* # of free sections */ 243 spinlock_t segmap_lock; /* free segmap lock */ 244 unsigned long *free_segmap; /* free segment bitmap */ 245 unsigned long *free_secmap; /* free section bitmap */ 246 }; 247 248 /* Notice: The order of dirty type is same with CURSEG_XXX in f2fs.h */ 249 enum dirty_type { 250 DIRTY_HOT_DATA, /* dirty segments assigned as hot data logs */ 251 DIRTY_WARM_DATA, /* dirty segments assigned as warm data logs */ 252 DIRTY_COLD_DATA, /* dirty segments assigned as cold data logs */ 253 DIRTY_HOT_NODE, /* dirty segments assigned as hot node logs */ 254 DIRTY_WARM_NODE, /* dirty segments assigned as warm node logs */ 255 DIRTY_COLD_NODE, /* dirty segments assigned as cold node logs */ 256 DIRTY, /* to count # of dirty segments */ 257 PRE, /* to count # of entirely obsolete segments */ 258 NR_DIRTY_TYPE 259 }; 260 261 struct dirty_seglist_info { 262 unsigned long *dirty_segmap[NR_DIRTY_TYPE]; 263 unsigned long *dirty_secmap; 264 struct mutex seglist_lock; /* lock for segment bitmaps */ 265 int nr_dirty[NR_DIRTY_TYPE]; /* # of dirty segments */ 266 unsigned long *victim_secmap; /* background GC victims */ 267 unsigned long *pinned_secmap; /* pinned victims from foreground GC */ 268 unsigned int pinned_secmap_cnt; /* count of victims which has pinned data */ 269 bool enable_pin_section; /* enable pinning section */ 270 }; 271 272 /* for active log information */ 273 struct curseg_info { 274 struct mutex curseg_mutex; /* lock for consistency */ 275 struct f2fs_summary_block *sum_blk; /* cached summary block */ 276 struct rw_semaphore journal_rwsem; /* protect journal area */ 277 struct f2fs_journal *journal; /* cached journal info */ 278 unsigned char alloc_type; /* current allocation type */ 279 unsigned short seg_type; /* segment type like CURSEG_XXX_TYPE */ 280 unsigned int segno; /* current segment number */ 281 unsigned short next_blkoff; /* next block offset to write */ 282 unsigned int zone; /* current zone number */ 283 unsigned int next_segno; /* preallocated segment */ 284 int fragment_remained_chunk; /* remained block size in a chunk for block fragmentation mode */ 285 bool inited; /* indicate inmem log is inited */ 286 }; 287 288 struct sit_entry_set { 289 struct list_head set_list; /* link with all sit sets */ 290 unsigned int start_segno; /* start segno of sits in set */ 291 unsigned int entry_cnt; /* the # of sit entries in set */ 292 }; 293 294 /* 295 * inline functions 296 */ 297 static inline struct curseg_info *CURSEG_I(struct f2fs_sb_info *sbi, int type) 298 { 299 return (struct curseg_info *)(SM_I(sbi)->curseg_array + type); 300 } 301 302 static inline bool is_curseg(struct f2fs_sb_info *sbi, unsigned int segno) 303 { 304 int i; 305 306 for (i = CURSEG_HOT_DATA; i < NO_CHECK_TYPE; i++) { 307 if (segno == CURSEG_I(sbi, i)->segno) 308 return true; 309 } 310 return false; 311 } 312 313 static inline bool is_cursec(struct f2fs_sb_info *sbi, unsigned int secno) 314 { 315 int i; 316 317 for (i = CURSEG_HOT_DATA; i < NO_CHECK_TYPE; i++) { 318 if (secno == GET_SEC_FROM_SEG(sbi, CURSEG_I(sbi, i)->segno)) 319 return true; 320 } 321 return false; 322 } 323 324 static inline struct seg_entry *get_seg_entry(struct f2fs_sb_info *sbi, 325 unsigned int segno) 326 { 327 struct sit_info *sit_i = SIT_I(sbi); 328 return &sit_i->sentries[segno]; 329 } 330 331 static inline struct sec_entry *get_sec_entry(struct f2fs_sb_info *sbi, 332 unsigned int segno) 333 { 334 struct sit_info *sit_i = SIT_I(sbi); 335 return &sit_i->sec_entries[GET_SEC_FROM_SEG(sbi, segno)]; 336 } 337 338 static inline unsigned int get_valid_blocks(struct f2fs_sb_info *sbi, 339 unsigned int segno, bool use_section) 340 { 341 /* 342 * In order to get # of valid blocks in a section instantly from many 343 * segments, f2fs manages two counting structures separately. 344 */ 345 if (use_section && __is_large_section(sbi)) 346 return get_sec_entry(sbi, segno)->valid_blocks; 347 else 348 return get_seg_entry(sbi, segno)->valid_blocks; 349 } 350 351 static inline unsigned int get_ckpt_valid_blocks(struct f2fs_sb_info *sbi, 352 unsigned int segno, bool use_section) 353 { 354 if (use_section && __is_large_section(sbi)) 355 return get_sec_entry(sbi, segno)->ckpt_valid_blocks; 356 else 357 return get_seg_entry(sbi, segno)->ckpt_valid_blocks; 358 } 359 360 static inline void set_ckpt_valid_blocks(struct f2fs_sb_info *sbi, 361 unsigned int segno) 362 { 363 unsigned int secno = GET_SEC_FROM_SEG(sbi, segno); 364 unsigned int start_segno = GET_SEG_FROM_SEC(sbi, secno); 365 unsigned int blocks = 0; 366 int i; 367 368 for (i = 0; i < SEGS_PER_SEC(sbi); i++, start_segno++) { 369 struct seg_entry *se = get_seg_entry(sbi, start_segno); 370 371 blocks += se->ckpt_valid_blocks; 372 } 373 get_sec_entry(sbi, segno)->ckpt_valid_blocks = blocks; 374 } 375 376 #ifdef CONFIG_F2FS_CHECK_FS 377 static inline void sanity_check_valid_blocks(struct f2fs_sb_info *sbi, 378 unsigned int segno) 379 { 380 unsigned int secno = GET_SEC_FROM_SEG(sbi, segno); 381 unsigned int start_segno = GET_SEG_FROM_SEC(sbi, secno); 382 unsigned int blocks = 0; 383 int i; 384 385 for (i = 0; i < SEGS_PER_SEC(sbi); i++, start_segno++) { 386 struct seg_entry *se = get_seg_entry(sbi, start_segno); 387 388 blocks += se->ckpt_valid_blocks; 389 } 390 391 if (blocks != get_sec_entry(sbi, segno)->ckpt_valid_blocks) { 392 f2fs_err(sbi, 393 "Inconsistent ckpt valid blocks: " 394 "seg entry(%d) vs sec entry(%d) at secno %d", 395 blocks, get_sec_entry(sbi, segno)->ckpt_valid_blocks, secno); 396 f2fs_bug_on(sbi, 1); 397 } 398 } 399 #else 400 static inline void sanity_check_valid_blocks(struct f2fs_sb_info *sbi, 401 unsigned int segno) 402 { 403 } 404 #endif 405 static inline void seg_info_from_raw_sit(struct seg_entry *se, 406 struct f2fs_sit_entry *rs) 407 { 408 se->valid_blocks = GET_SIT_VBLOCKS(rs); 409 se->ckpt_valid_blocks = GET_SIT_VBLOCKS(rs); 410 memcpy(se->cur_valid_map, rs->valid_map, SIT_VBLOCK_MAP_SIZE); 411 memcpy(se->ckpt_valid_map, rs->valid_map, SIT_VBLOCK_MAP_SIZE); 412 #ifdef CONFIG_F2FS_CHECK_FS 413 memcpy(se->cur_valid_map_mir, rs->valid_map, SIT_VBLOCK_MAP_SIZE); 414 #endif 415 se->type = GET_SIT_TYPE(rs); 416 se->mtime = le64_to_cpu(rs->mtime); 417 } 418 419 static inline void __seg_info_to_raw_sit(struct seg_entry *se, 420 struct f2fs_sit_entry *rs) 421 { 422 unsigned short raw_vblocks = (se->type << SIT_VBLOCKS_SHIFT) | 423 se->valid_blocks; 424 rs->vblocks = cpu_to_le16(raw_vblocks); 425 memcpy(rs->valid_map, se->cur_valid_map, SIT_VBLOCK_MAP_SIZE); 426 rs->mtime = cpu_to_le64(se->mtime); 427 } 428 429 static inline void seg_info_to_sit_folio(struct f2fs_sb_info *sbi, 430 struct folio *folio, unsigned int start) 431 { 432 struct f2fs_sit_block *raw_sit; 433 struct seg_entry *se; 434 struct f2fs_sit_entry *rs; 435 unsigned int end = min(start + SIT_ENTRY_PER_BLOCK, 436 (unsigned long)MAIN_SEGS(sbi)); 437 int i; 438 439 raw_sit = folio_address(folio); 440 memset(raw_sit, 0, PAGE_SIZE); 441 for (i = 0; i < end - start; i++) { 442 rs = &raw_sit->entries[i]; 443 se = get_seg_entry(sbi, start + i); 444 __seg_info_to_raw_sit(se, rs); 445 } 446 } 447 448 static inline void seg_info_to_raw_sit(struct seg_entry *se, 449 struct f2fs_sit_entry *rs) 450 { 451 __seg_info_to_raw_sit(se, rs); 452 453 memcpy(se->ckpt_valid_map, rs->valid_map, SIT_VBLOCK_MAP_SIZE); 454 se->ckpt_valid_blocks = se->valid_blocks; 455 } 456 457 static inline unsigned int find_next_inuse(struct free_segmap_info *free_i, 458 unsigned int max, unsigned int segno) 459 { 460 unsigned int ret; 461 spin_lock(&free_i->segmap_lock); 462 ret = find_next_bit(free_i->free_segmap, max, segno); 463 spin_unlock(&free_i->segmap_lock); 464 return ret; 465 } 466 467 static inline void __set_free(struct f2fs_sb_info *sbi, unsigned int segno) 468 { 469 struct free_segmap_info *free_i = FREE_I(sbi); 470 unsigned int secno = GET_SEC_FROM_SEG(sbi, segno); 471 unsigned int start_segno = GET_SEG_FROM_SEC(sbi, secno); 472 unsigned int next; 473 474 spin_lock(&free_i->segmap_lock); 475 clear_bit(segno, free_i->free_segmap); 476 free_i->free_segments++; 477 478 next = find_next_bit(free_i->free_segmap, 479 start_segno + SEGS_PER_SEC(sbi), start_segno); 480 if (next >= start_segno + f2fs_usable_segs_in_sec(sbi)) { 481 clear_bit(secno, free_i->free_secmap); 482 free_i->free_sections++; 483 } 484 spin_unlock(&free_i->segmap_lock); 485 } 486 487 static inline void __set_inuse(struct f2fs_sb_info *sbi, 488 unsigned int segno) 489 { 490 struct free_segmap_info *free_i = FREE_I(sbi); 491 unsigned int secno = GET_SEC_FROM_SEG(sbi, segno); 492 493 set_bit(segno, free_i->free_segmap); 494 free_i->free_segments--; 495 if (!test_and_set_bit(secno, free_i->free_secmap)) 496 free_i->free_sections--; 497 } 498 499 static inline void __set_test_and_free(struct f2fs_sb_info *sbi, 500 unsigned int segno, bool inmem) 501 { 502 struct free_segmap_info *free_i = FREE_I(sbi); 503 unsigned int secno = GET_SEC_FROM_SEG(sbi, segno); 504 unsigned int start_segno = GET_SEG_FROM_SEC(sbi, secno); 505 unsigned int next; 506 bool ret; 507 508 spin_lock(&free_i->segmap_lock); 509 ret = test_and_clear_bit(segno, free_i->free_segmap); 510 if (!ret) 511 goto unlock_out; 512 513 free_i->free_segments++; 514 515 if (!inmem && is_cursec(sbi, secno)) 516 goto unlock_out; 517 518 /* check large section */ 519 next = find_next_bit(free_i->free_segmap, 520 start_segno + SEGS_PER_SEC(sbi), start_segno); 521 if (next < start_segno + f2fs_usable_segs_in_sec(sbi)) 522 goto unlock_out; 523 524 ret = test_and_clear_bit(secno, free_i->free_secmap); 525 if (!ret) 526 goto unlock_out; 527 528 free_i->free_sections++; 529 530 if (GET_SEC_FROM_SEG(sbi, sbi->next_victim_seg[BG_GC]) == secno) 531 sbi->next_victim_seg[BG_GC] = NULL_SEGNO; 532 if (GET_SEC_FROM_SEG(sbi, sbi->next_victim_seg[FG_GC]) == secno) 533 sbi->next_victim_seg[FG_GC] = NULL_SEGNO; 534 535 unlock_out: 536 spin_unlock(&free_i->segmap_lock); 537 } 538 539 static inline void __set_test_and_inuse(struct f2fs_sb_info *sbi, 540 unsigned int segno) 541 { 542 struct free_segmap_info *free_i = FREE_I(sbi); 543 unsigned int secno = GET_SEC_FROM_SEG(sbi, segno); 544 545 spin_lock(&free_i->segmap_lock); 546 if (!test_and_set_bit(segno, free_i->free_segmap)) { 547 free_i->free_segments--; 548 if (!test_and_set_bit(secno, free_i->free_secmap)) 549 free_i->free_sections--; 550 } 551 spin_unlock(&free_i->segmap_lock); 552 } 553 554 static inline void get_sit_bitmap(struct f2fs_sb_info *sbi, 555 void *dst_addr) 556 { 557 struct sit_info *sit_i = SIT_I(sbi); 558 559 #ifdef CONFIG_F2FS_CHECK_FS 560 if (memcmp(sit_i->sit_bitmap, sit_i->sit_bitmap_mir, 561 sit_i->bitmap_size)) 562 f2fs_bug_on(sbi, 1); 563 #endif 564 memcpy(dst_addr, sit_i->sit_bitmap, sit_i->bitmap_size); 565 } 566 567 static inline block_t written_block_count(struct f2fs_sb_info *sbi) 568 { 569 return SIT_I(sbi)->written_valid_blocks; 570 } 571 572 static inline unsigned int free_segments(struct f2fs_sb_info *sbi) 573 { 574 return FREE_I(sbi)->free_segments; 575 } 576 577 static inline unsigned int reserved_segments(struct f2fs_sb_info *sbi) 578 { 579 return SM_I(sbi)->reserved_segments; 580 } 581 582 static inline unsigned int free_sections(struct f2fs_sb_info *sbi) 583 { 584 return FREE_I(sbi)->free_sections; 585 } 586 587 static inline unsigned int prefree_segments(struct f2fs_sb_info *sbi) 588 { 589 return DIRTY_I(sbi)->nr_dirty[PRE]; 590 } 591 592 static inline unsigned int dirty_segments(struct f2fs_sb_info *sbi) 593 { 594 return DIRTY_I(sbi)->nr_dirty[DIRTY_HOT_DATA] + 595 DIRTY_I(sbi)->nr_dirty[DIRTY_WARM_DATA] + 596 DIRTY_I(sbi)->nr_dirty[DIRTY_COLD_DATA] + 597 DIRTY_I(sbi)->nr_dirty[DIRTY_HOT_NODE] + 598 DIRTY_I(sbi)->nr_dirty[DIRTY_WARM_NODE] + 599 DIRTY_I(sbi)->nr_dirty[DIRTY_COLD_NODE]; 600 } 601 602 static inline int overprovision_segments(struct f2fs_sb_info *sbi) 603 { 604 return SM_I(sbi)->ovp_segments; 605 } 606 607 static inline int reserved_sections(struct f2fs_sb_info *sbi) 608 { 609 return GET_SEC_FROM_SEG(sbi, reserved_segments(sbi)); 610 } 611 612 static inline unsigned int get_left_section_blocks(struct f2fs_sb_info *sbi, 613 enum log_type type, unsigned int segno) 614 { 615 if (f2fs_lfs_mode(sbi)) { 616 unsigned int used_blocks = __is_large_section(sbi) ? SEGS_TO_BLKS(sbi, 617 (segno - GET_START_SEG_FROM_SEC(sbi, segno))) : 0; 618 return CAP_BLKS_PER_SEC(sbi) - used_blocks - 619 CURSEG_I(sbi, type)->next_blkoff; 620 } 621 return CAP_BLKS_PER_SEC(sbi) - get_ckpt_valid_blocks(sbi, segno, true); 622 } 623 624 static inline bool has_curseg_enough_space(struct f2fs_sb_info *sbi, 625 unsigned int node_blocks, unsigned int data_blocks, 626 unsigned int dent_blocks) 627 { 628 unsigned int segno, left_blocks, blocks; 629 int i; 630 631 /* check current data/node sections in the worst case. */ 632 for (i = CURSEG_HOT_DATA; i < NR_PERSISTENT_LOG; i++) { 633 segno = CURSEG_I(sbi, i)->segno; 634 635 if (unlikely(segno == NULL_SEGNO)) 636 return false; 637 638 left_blocks = get_left_section_blocks(sbi, i, segno); 639 640 blocks = i <= CURSEG_COLD_DATA ? data_blocks : node_blocks; 641 if (blocks > left_blocks) 642 return false; 643 } 644 645 /* check current data section for dentry blocks. */ 646 segno = CURSEG_I(sbi, CURSEG_HOT_DATA)->segno; 647 648 if (unlikely(segno == NULL_SEGNO)) 649 return false; 650 651 left_blocks = get_left_section_blocks(sbi, CURSEG_HOT_DATA, segno); 652 653 if (dent_blocks > left_blocks) 654 return false; 655 return true; 656 } 657 658 /* 659 * calculate needed sections for dirty node/dentry and call 660 * has_curseg_enough_space, please note that, it needs to account 661 * dirty data as well in lfs mode when checkpoint is disabled. 662 */ 663 static inline void __get_secs_required(struct f2fs_sb_info *sbi, 664 unsigned int *lower_p, unsigned int *upper_p, bool *curseg_p) 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 unsigned int node_secs = total_node_blocks / CAP_BLKS_PER_SEC(sbi); 672 unsigned int dent_secs = total_dent_blocks / CAP_BLKS_PER_SEC(sbi); 673 unsigned int data_secs = 0; 674 unsigned int node_blocks = total_node_blocks % CAP_BLKS_PER_SEC(sbi); 675 unsigned int dent_blocks = total_dent_blocks % CAP_BLKS_PER_SEC(sbi); 676 unsigned int data_blocks = 0; 677 678 if (f2fs_lfs_mode(sbi)) { 679 total_data_blocks = get_pages(sbi, F2FS_DIRTY_DATA); 680 data_secs = total_data_blocks / CAP_BLKS_PER_SEC(sbi); 681 data_blocks = total_data_blocks % CAP_BLKS_PER_SEC(sbi); 682 } 683 684 if (lower_p) 685 *lower_p = node_secs + dent_secs + data_secs; 686 if (upper_p) 687 *upper_p = node_secs + dent_secs + data_secs + 688 (node_blocks ? 1 : 0) + (dent_blocks ? 1 : 0) + 689 (data_blocks ? 1 : 0); 690 if (curseg_p) 691 *curseg_p = has_curseg_enough_space(sbi, 692 node_blocks, data_blocks, dent_blocks); 693 } 694 695 static inline bool has_not_enough_free_secs(struct f2fs_sb_info *sbi, 696 int freed, int needed) 697 { 698 unsigned int free_secs, lower_secs, upper_secs; 699 bool curseg_space; 700 701 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING))) 702 return false; 703 704 __get_secs_required(sbi, &lower_secs, &upper_secs, &curseg_space); 705 706 free_secs = free_sections(sbi) + freed; 707 lower_secs += needed + reserved_sections(sbi); 708 upper_secs += needed + reserved_sections(sbi); 709 710 if (free_secs > upper_secs) 711 return false; 712 if (free_secs <= lower_secs) 713 return true; 714 return !curseg_space; 715 } 716 717 static inline bool has_enough_free_secs(struct f2fs_sb_info *sbi, 718 int freed, int needed) 719 { 720 return !has_not_enough_free_secs(sbi, freed, needed); 721 } 722 723 static inline bool has_enough_free_blks(struct f2fs_sb_info *sbi) 724 { 725 unsigned int total_free_blocks = 0; 726 unsigned int avail_user_block_count; 727 728 spin_lock(&sbi->stat_lock); 729 730 avail_user_block_count = get_available_block_count(sbi, NULL, true); 731 total_free_blocks = avail_user_block_count - (unsigned int)valid_user_blocks(sbi); 732 733 spin_unlock(&sbi->stat_lock); 734 735 return total_free_blocks > 0; 736 } 737 738 static inline bool f2fs_is_checkpoint_ready(struct f2fs_sb_info *sbi) 739 { 740 if (likely(!is_sbi_flag_set(sbi, SBI_CP_DISABLED))) 741 return true; 742 if (likely(has_enough_free_secs(sbi, 0, 0))) 743 return true; 744 if (!f2fs_lfs_mode(sbi) && 745 likely(has_enough_free_blks(sbi))) 746 return true; 747 return false; 748 } 749 750 static inline bool excess_prefree_segs(struct f2fs_sb_info *sbi) 751 { 752 return prefree_segments(sbi) > SM_I(sbi)->rec_prefree_segments; 753 } 754 755 static inline int utilization(struct f2fs_sb_info *sbi) 756 { 757 return div_u64((u64)valid_user_blocks(sbi) * 100, 758 sbi->user_block_count); 759 } 760 761 /* 762 * Sometimes f2fs may be better to drop out-of-place update policy. 763 * And, users can control the policy through sysfs entries. 764 * There are five policies with triggering conditions as follows. 765 * F2FS_IPU_FORCE - all the time, 766 * F2FS_IPU_SSR - if SSR mode is activated, 767 * F2FS_IPU_UTIL - if FS utilization is over threashold, 768 * F2FS_IPU_SSR_UTIL - if SSR mode is activated and FS utilization is over 769 * threashold, 770 * F2FS_IPU_FSYNC - activated in fsync path only for high performance flash 771 * storages. IPU will be triggered only if the # of dirty 772 * pages over min_fsync_blocks. (=default option) 773 * F2FS_IPU_ASYNC - do IPU given by asynchronous write requests. 774 * F2FS_IPU_NOCACHE - disable IPU bio cache. 775 * F2FS_IPU_HONOR_OPU_WRITE - use OPU write prior to IPU write if inode has 776 * FI_OPU_WRITE flag. 777 * F2FS_IPU_DISABLE - disable IPU. (=default option in LFS mode) 778 */ 779 #define DEF_MIN_IPU_UTIL 70 780 #define DEF_MIN_FSYNC_BLOCKS 8 781 #define DEF_MIN_HOT_BLOCKS 16 782 783 #define SMALL_VOLUME_SEGMENTS (16 * 512) /* 16GB */ 784 785 #define F2FS_IPU_DISABLE 0 786 787 /* Modification on enum should be synchronized with ipu_mode_names array */ 788 enum { 789 F2FS_IPU_FORCE, 790 F2FS_IPU_SSR, 791 F2FS_IPU_UTIL, 792 F2FS_IPU_SSR_UTIL, 793 F2FS_IPU_FSYNC, 794 F2FS_IPU_ASYNC, 795 F2FS_IPU_NOCACHE, 796 F2FS_IPU_HONOR_OPU_WRITE, 797 F2FS_IPU_MAX, 798 }; 799 800 static inline bool IS_F2FS_IPU_DISABLE(struct f2fs_sb_info *sbi) 801 { 802 return SM_I(sbi)->ipu_policy == F2FS_IPU_DISABLE; 803 } 804 805 #define F2FS_IPU_POLICY(name) \ 806 static inline bool IS_##name(struct f2fs_sb_info *sbi) \ 807 { \ 808 return SM_I(sbi)->ipu_policy & BIT(name); \ 809 } 810 811 F2FS_IPU_POLICY(F2FS_IPU_FORCE); 812 F2FS_IPU_POLICY(F2FS_IPU_SSR); 813 F2FS_IPU_POLICY(F2FS_IPU_UTIL); 814 F2FS_IPU_POLICY(F2FS_IPU_SSR_UTIL); 815 F2FS_IPU_POLICY(F2FS_IPU_FSYNC); 816 F2FS_IPU_POLICY(F2FS_IPU_ASYNC); 817 F2FS_IPU_POLICY(F2FS_IPU_NOCACHE); 818 F2FS_IPU_POLICY(F2FS_IPU_HONOR_OPU_WRITE); 819 820 static inline unsigned int curseg_segno(struct f2fs_sb_info *sbi, 821 int type) 822 { 823 struct curseg_info *curseg = CURSEG_I(sbi, type); 824 return curseg->segno; 825 } 826 827 static inline unsigned char curseg_alloc_type(struct f2fs_sb_info *sbi, 828 int type) 829 { 830 struct curseg_info *curseg = CURSEG_I(sbi, type); 831 return curseg->alloc_type; 832 } 833 834 static inline bool valid_main_segno(struct f2fs_sb_info *sbi, 835 unsigned int segno) 836 { 837 return segno <= (MAIN_SEGS(sbi) - 1); 838 } 839 840 static inline void verify_fio_blkaddr(struct f2fs_io_info *fio) 841 { 842 struct f2fs_sb_info *sbi = fio->sbi; 843 844 if (__is_valid_data_blkaddr(fio->old_blkaddr)) 845 verify_blkaddr(sbi, fio->old_blkaddr, __is_meta_io(fio) ? 846 META_GENERIC : DATA_GENERIC); 847 verify_blkaddr(sbi, fio->new_blkaddr, __is_meta_io(fio) ? 848 META_GENERIC : DATA_GENERIC_ENHANCE); 849 } 850 851 /* 852 * Summary block is always treated as an invalid block 853 */ 854 static inline int check_block_count(struct f2fs_sb_info *sbi, 855 int segno, struct f2fs_sit_entry *raw_sit) 856 { 857 bool is_valid = test_bit_le(0, raw_sit->valid_map) ? true : false; 858 int valid_blocks = 0; 859 int cur_pos = 0, next_pos; 860 unsigned int usable_blks_per_seg = f2fs_usable_blks_in_seg(sbi, segno); 861 862 /* check bitmap with valid block count */ 863 do { 864 if (is_valid) { 865 next_pos = find_next_zero_bit_le(&raw_sit->valid_map, 866 usable_blks_per_seg, 867 cur_pos); 868 valid_blocks += next_pos - cur_pos; 869 } else 870 next_pos = find_next_bit_le(&raw_sit->valid_map, 871 usable_blks_per_seg, 872 cur_pos); 873 cur_pos = next_pos; 874 is_valid = !is_valid; 875 } while (cur_pos < usable_blks_per_seg); 876 877 if (unlikely(GET_SIT_VBLOCKS(raw_sit) != valid_blocks)) { 878 f2fs_err(sbi, "Mismatch valid blocks %d vs. %d", 879 GET_SIT_VBLOCKS(raw_sit), valid_blocks); 880 set_sbi_flag(sbi, SBI_NEED_FSCK); 881 f2fs_handle_error(sbi, ERROR_INCONSISTENT_SIT); 882 return -EFSCORRUPTED; 883 } 884 885 if (usable_blks_per_seg < BLKS_PER_SEG(sbi)) 886 f2fs_bug_on(sbi, find_next_bit_le(&raw_sit->valid_map, 887 BLKS_PER_SEG(sbi), 888 usable_blks_per_seg) != BLKS_PER_SEG(sbi)); 889 890 /* check segment usage, and check boundary of a given segment number */ 891 if (unlikely(GET_SIT_VBLOCKS(raw_sit) > usable_blks_per_seg 892 || !valid_main_segno(sbi, segno))) { 893 f2fs_err(sbi, "Wrong valid blocks %d or segno %u", 894 GET_SIT_VBLOCKS(raw_sit), segno); 895 set_sbi_flag(sbi, SBI_NEED_FSCK); 896 f2fs_handle_error(sbi, ERROR_INCONSISTENT_SIT); 897 return -EFSCORRUPTED; 898 } 899 return 0; 900 } 901 902 static inline pgoff_t current_sit_addr(struct f2fs_sb_info *sbi, 903 unsigned int start) 904 { 905 struct sit_info *sit_i = SIT_I(sbi); 906 unsigned int offset = SIT_BLOCK_OFFSET(start); 907 block_t blk_addr = sit_i->sit_base_addr + offset; 908 909 f2fs_bug_on(sbi, !valid_main_segno(sbi, start)); 910 911 #ifdef CONFIG_F2FS_CHECK_FS 912 if (f2fs_test_bit(offset, sit_i->sit_bitmap) != 913 f2fs_test_bit(offset, sit_i->sit_bitmap_mir)) 914 f2fs_bug_on(sbi, 1); 915 #endif 916 917 /* calculate sit block address */ 918 if (f2fs_test_bit(offset, sit_i->sit_bitmap)) 919 blk_addr += sit_i->sit_blocks; 920 921 return blk_addr; 922 } 923 924 static inline pgoff_t next_sit_addr(struct f2fs_sb_info *sbi, 925 pgoff_t block_addr) 926 { 927 struct sit_info *sit_i = SIT_I(sbi); 928 block_addr -= sit_i->sit_base_addr; 929 if (block_addr < sit_i->sit_blocks) 930 block_addr += sit_i->sit_blocks; 931 else 932 block_addr -= sit_i->sit_blocks; 933 934 return block_addr + sit_i->sit_base_addr; 935 } 936 937 static inline void set_to_next_sit(struct sit_info *sit_i, unsigned int start) 938 { 939 unsigned int block_off = SIT_BLOCK_OFFSET(start); 940 941 f2fs_change_bit(block_off, sit_i->sit_bitmap); 942 #ifdef CONFIG_F2FS_CHECK_FS 943 f2fs_change_bit(block_off, sit_i->sit_bitmap_mir); 944 #endif 945 } 946 947 static inline unsigned long long get_mtime(struct f2fs_sb_info *sbi, 948 bool base_time) 949 { 950 struct sit_info *sit_i = SIT_I(sbi); 951 time64_t diff, now = ktime_get_boottime_seconds(); 952 953 if (now >= sit_i->mounted_time) 954 return sit_i->elapsed_time + now - sit_i->mounted_time; 955 956 /* system time is set to the past */ 957 if (!base_time) { 958 diff = sit_i->mounted_time - now; 959 if (sit_i->elapsed_time >= diff) 960 return sit_i->elapsed_time - diff; 961 return 0; 962 } 963 return sit_i->elapsed_time; 964 } 965 966 static inline void set_summary(struct f2fs_summary *sum, nid_t nid, 967 unsigned int ofs_in_node, unsigned char version) 968 { 969 sum->nid = cpu_to_le32(nid); 970 sum->ofs_in_node = cpu_to_le16(ofs_in_node); 971 sum->version = version; 972 } 973 974 static inline block_t start_sum_block(struct f2fs_sb_info *sbi) 975 { 976 return __start_cp_addr(sbi) + 977 le32_to_cpu(F2FS_CKPT(sbi)->cp_pack_start_sum); 978 } 979 980 static inline block_t sum_blk_addr(struct f2fs_sb_info *sbi, int base, int type) 981 { 982 return __start_cp_addr(sbi) + 983 le32_to_cpu(F2FS_CKPT(sbi)->cp_pack_total_block_count) 984 - (base + 1) + type; 985 } 986 987 static inline bool sec_usage_check(struct f2fs_sb_info *sbi, unsigned int secno) 988 { 989 if (is_cursec(sbi, secno) || (sbi->cur_victim_sec == secno)) 990 return true; 991 return false; 992 } 993 994 /* 995 * It is very important to gather dirty pages and write at once, so that we can 996 * submit a big bio without interfering other data writes. 997 * By default, 512 pages for directory data, 998 * 512 pages (2MB) * 8 for nodes, and 999 * 256 pages * 8 for meta are set. 1000 */ 1001 static inline int nr_pages_to_skip(struct f2fs_sb_info *sbi, int type) 1002 { 1003 if (sbi->sb->s_bdi->wb.dirty_exceeded) 1004 return 0; 1005 1006 if (type == DATA) 1007 return BLKS_PER_SEG(sbi); 1008 else if (type == NODE) 1009 return SEGS_TO_BLKS(sbi, 8); 1010 else if (type == META) 1011 return 8 * BIO_MAX_VECS; 1012 else 1013 return 0; 1014 } 1015 1016 /* 1017 * When writing pages, it'd better align nr_to_write for segment size. 1018 */ 1019 static inline long nr_pages_to_write(struct f2fs_sb_info *sbi, int type, 1020 struct writeback_control *wbc) 1021 { 1022 long nr_to_write, desired; 1023 1024 if (wbc->sync_mode != WB_SYNC_NONE) 1025 return 0; 1026 1027 nr_to_write = wbc->nr_to_write; 1028 desired = BIO_MAX_VECS; 1029 if (type == NODE) 1030 desired <<= 1; 1031 1032 wbc->nr_to_write = desired; 1033 return desired - nr_to_write; 1034 } 1035 1036 static inline void wake_up_discard_thread(struct f2fs_sb_info *sbi, bool force) 1037 { 1038 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info; 1039 bool wakeup = false; 1040 int i; 1041 1042 if (force) 1043 goto wake_up; 1044 1045 mutex_lock(&dcc->cmd_lock); 1046 for (i = MAX_PLIST_NUM - 1; i >= 0; i--) { 1047 if (i + 1 < dcc->discard_granularity) 1048 break; 1049 if (!list_empty(&dcc->pend_list[i])) { 1050 wakeup = true; 1051 break; 1052 } 1053 } 1054 mutex_unlock(&dcc->cmd_lock); 1055 if (!wakeup || !is_idle(sbi, DISCARD_TIME)) 1056 return; 1057 wake_up: 1058 dcc->discard_wake = true; 1059 wake_up_interruptible_all(&dcc->discard_wait_queue); 1060 } 1061