1 /* SPDX-License-Identifier: GPL-2.0 */
5 * Copyright (c) 2012 Samsung Electronics Co., Ltd.
6 * http://www.samsung.com/
8 #include <linux/blkdev.h>
9 #include <linux/backing-dev.h>
12 #define NULL_SEGNO ((unsigned int)(~0))
13 #define NULL_SECNO ((unsigned int)(~0))
15 #define DEF_RECLAIM_PREFREE_SEGMENTS 5 /* 5% over total segments */
16 #define DEF_MAX_RECLAIM_PREFREE_SEGMENTS 4096 /* 8GB in maximum */
18 #define F2FS_MIN_SEGMENTS 9 /* SB + 2 (CP + SIT + NAT) + SSA + MAIN */
20 /* L: Logical segment # in volume, R: Relative segment # in main area */
21 #define GET_L2R_SEGNO(free_i, segno) ((segno) - (free_i)->start_segno)
22 #define GET_R2L_SEGNO(free_i, segno) ((segno) + (free_i)->start_segno)
24 #define IS_DATASEG(t) ((t) <= CURSEG_COLD_DATA)
25 #define IS_NODESEG(t) ((t) >= CURSEG_HOT_NODE)
27 #define IS_HOT(t) ((t) == CURSEG_HOT_NODE || (t) == CURSEG_HOT_DATA)
28 #define IS_WARM(t) ((t) == CURSEG_WARM_NODE || (t) == CURSEG_WARM_DATA)
29 #define IS_COLD(t) ((t) == CURSEG_COLD_NODE || (t) == CURSEG_COLD_DATA)
31 #define IS_CURSEG(sbi, seg) \
32 (((seg) == CURSEG_I(sbi, CURSEG_HOT_DATA)->segno) || \
33 ((seg) == CURSEG_I(sbi, CURSEG_WARM_DATA)->segno) || \
34 ((seg) == CURSEG_I(sbi, CURSEG_COLD_DATA)->segno) || \
35 ((seg) == CURSEG_I(sbi, CURSEG_HOT_NODE)->segno) || \
36 ((seg) == CURSEG_I(sbi, CURSEG_WARM_NODE)->segno) || \
37 ((seg) == CURSEG_I(sbi, CURSEG_COLD_NODE)->segno))
39 #define IS_CURSEC(sbi, secno) \
40 (((secno) == CURSEG_I(sbi, CURSEG_HOT_DATA)->segno / \
41 (sbi)->segs_per_sec) || \
42 ((secno) == CURSEG_I(sbi, CURSEG_WARM_DATA)->segno / \
43 (sbi)->segs_per_sec) || \
44 ((secno) == CURSEG_I(sbi, CURSEG_COLD_DATA)->segno / \
45 (sbi)->segs_per_sec) || \
46 ((secno) == CURSEG_I(sbi, CURSEG_HOT_NODE)->segno / \
47 (sbi)->segs_per_sec) || \
48 ((secno) == CURSEG_I(sbi, CURSEG_WARM_NODE)->segno / \
49 (sbi)->segs_per_sec) || \
50 ((secno) == CURSEG_I(sbi, CURSEG_COLD_NODE)->segno / \
51 (sbi)->segs_per_sec)) \
53 #define MAIN_BLKADDR(sbi) \
54 (SM_I(sbi) ? SM_I(sbi)->main_blkaddr : \
55 le32_to_cpu(F2FS_RAW_SUPER(sbi)->main_blkaddr))
56 #define SEG0_BLKADDR(sbi) \
57 (SM_I(sbi) ? SM_I(sbi)->seg0_blkaddr : \
58 le32_to_cpu(F2FS_RAW_SUPER(sbi)->segment0_blkaddr))
60 #define MAIN_SEGS(sbi) (SM_I(sbi)->main_segments)
61 #define MAIN_SECS(sbi) ((sbi)->total_sections)
63 #define TOTAL_SEGS(sbi) \
64 (SM_I(sbi) ? SM_I(sbi)->segment_count : \
65 le32_to_cpu(F2FS_RAW_SUPER(sbi)->segment_count))
66 #define TOTAL_BLKS(sbi) (TOTAL_SEGS(sbi) << (sbi)->log_blocks_per_seg)
68 #define MAX_BLKADDR(sbi) (SEG0_BLKADDR(sbi) + TOTAL_BLKS(sbi))
69 #define SEGMENT_SIZE(sbi) (1ULL << ((sbi)->log_blocksize + \
70 (sbi)->log_blocks_per_seg))
72 #define START_BLOCK(sbi, segno) (SEG0_BLKADDR(sbi) + \
73 (GET_R2L_SEGNO(FREE_I(sbi), segno) << (sbi)->log_blocks_per_seg))
75 #define NEXT_FREE_BLKADDR(sbi, curseg) \
76 (START_BLOCK(sbi, (curseg)->segno) + (curseg)->next_blkoff)
78 #define GET_SEGOFF_FROM_SEG0(sbi, blk_addr) ((blk_addr) - SEG0_BLKADDR(sbi))
79 #define GET_SEGNO_FROM_SEG0(sbi, blk_addr) \
80 (GET_SEGOFF_FROM_SEG0(sbi, blk_addr) >> (sbi)->log_blocks_per_seg)
81 #define GET_BLKOFF_FROM_SEG0(sbi, blk_addr) \
82 (GET_SEGOFF_FROM_SEG0(sbi, blk_addr) & ((sbi)->blocks_per_seg - 1))
84 #define GET_SEGNO(sbi, blk_addr) \
85 ((!__is_valid_data_blkaddr(blk_addr)) ? \
86 NULL_SEGNO : GET_L2R_SEGNO(FREE_I(sbi), \
87 GET_SEGNO_FROM_SEG0(sbi, blk_addr)))
88 #define BLKS_PER_SEC(sbi) \
89 ((sbi)->segs_per_sec * (sbi)->blocks_per_seg)
90 #define GET_SEC_FROM_SEG(sbi, segno) \
91 ((segno) / (sbi)->segs_per_sec)
92 #define GET_SEG_FROM_SEC(sbi, secno) \
93 ((secno) * (sbi)->segs_per_sec)
94 #define GET_ZONE_FROM_SEC(sbi, secno) \
95 ((secno) / (sbi)->secs_per_zone)
96 #define GET_ZONE_FROM_SEG(sbi, segno) \
97 GET_ZONE_FROM_SEC(sbi, GET_SEC_FROM_SEG(sbi, segno))
99 #define GET_SUM_BLOCK(sbi, segno) \
100 ((sbi)->sm_info->ssa_blkaddr + (segno))
102 #define GET_SUM_TYPE(footer) ((footer)->entry_type)
103 #define SET_SUM_TYPE(footer, type) ((footer)->entry_type = (type))
105 #define SIT_ENTRY_OFFSET(sit_i, segno) \
106 ((segno) % (sit_i)->sents_per_block)
107 #define SIT_BLOCK_OFFSET(segno) \
108 ((segno) / SIT_ENTRY_PER_BLOCK)
109 #define START_SEGNO(segno) \
110 (SIT_BLOCK_OFFSET(segno) * SIT_ENTRY_PER_BLOCK)
111 #define SIT_BLK_CNT(sbi) \
112 DIV_ROUND_UP(MAIN_SEGS(sbi), SIT_ENTRY_PER_BLOCK)
113 #define f2fs_bitmap_size(nr) \
114 (BITS_TO_LONGS(nr) * sizeof(unsigned long))
116 #define SECTOR_FROM_BLOCK(blk_addr) \
117 (((sector_t)blk_addr) << F2FS_LOG_SECTORS_PER_BLOCK)
118 #define SECTOR_TO_BLOCK(sectors) \
119 ((sectors) >> F2FS_LOG_SECTORS_PER_BLOCK)
122 * indicate a block allocation direction: RIGHT and LEFT.
123 * RIGHT means allocating new sections towards the end of volume.
124 * LEFT means the opposite direction.
132 * In the victim_sel_policy->alloc_mode, there are two block allocation modes.
133 * LFS writes data sequentially with cleaning operations.
134 * SSR (Slack Space Recycle) reuses obsolete space without cleaning operations.
142 * In the victim_sel_policy->gc_mode, there are two gc, aka cleaning, modes.
143 * GC_CB is based on cost-benefit algorithm.
144 * GC_GREEDY is based on greedy algorithm.
155 * BG_GC means the background cleaning job.
156 * FG_GC means the on-demand cleaning job.
157 * FORCE_FG_GC means on-demand cleaning job in background.
165 /* for a function parameter to select a victim segment */
166 struct victim_sel_policy {
167 int alloc_mode; /* LFS or SSR */
168 int gc_mode; /* GC_CB or GC_GREEDY */
169 unsigned long *dirty_bitmap; /* dirty segment/section bitmap */
170 unsigned int max_search; /*
171 * maximum # of segments/sections
174 unsigned int offset; /* last scanned bitmap offset */
175 unsigned int ofs_unit; /* bitmap search unit */
176 unsigned int min_cost; /* minimum cost */
177 unsigned int min_segno; /* segment # having min. cost */
181 unsigned int type:6; /* segment type like CURSEG_XXX_TYPE */
182 unsigned int valid_blocks:10; /* # of valid blocks */
183 unsigned int ckpt_valid_blocks:10; /* # of valid blocks last cp */
184 unsigned int padding:6; /* padding */
185 unsigned char *cur_valid_map; /* validity bitmap of blocks */
186 #ifdef CONFIG_F2FS_CHECK_FS
187 unsigned char *cur_valid_map_mir; /* mirror of current valid bitmap */
190 * # of valid blocks and the validity bitmap stored in the last
191 * checkpoint pack. This information is used by the SSR mode.
193 unsigned char *ckpt_valid_map; /* validity bitmap of blocks last cp */
194 unsigned char *discard_map;
195 unsigned long long mtime; /* modification time of the segment */
199 unsigned int valid_blocks; /* # of valid blocks in a section */
202 struct segment_allocation {
203 void (*allocate_segment)(struct f2fs_sb_info *, int, bool);
206 #define MAX_SKIP_GC_COUNT 16
209 struct list_head list;
211 block_t old_addr; /* for revoking when fail to commit */
215 const struct segment_allocation *s_ops;
217 block_t sit_base_addr; /* start block address of SIT area */
218 block_t sit_blocks; /* # of blocks used by SIT area */
219 block_t written_valid_blocks; /* # of valid blocks in main area */
220 char *bitmap; /* all bitmaps pointer */
221 char *sit_bitmap; /* SIT bitmap pointer */
222 #ifdef CONFIG_F2FS_CHECK_FS
223 char *sit_bitmap_mir; /* SIT bitmap mirror */
225 /* bitmap of segments to be ignored by GC in case of errors */
226 unsigned long *invalid_segmap;
228 unsigned int bitmap_size; /* SIT bitmap size */
230 unsigned long *tmp_map; /* bitmap for temporal use */
231 unsigned long *dirty_sentries_bitmap; /* bitmap for dirty sentries */
232 unsigned int dirty_sentries; /* # of dirty sentries */
233 unsigned int sents_per_block; /* # of SIT entries per block */
234 struct rw_semaphore sentry_lock; /* to protect SIT cache */
235 struct seg_entry *sentries; /* SIT segment-level cache */
236 struct sec_entry *sec_entries; /* SIT section-level cache */
238 /* for cost-benefit algorithm in cleaning procedure */
239 unsigned long long elapsed_time; /* elapsed time after mount */
240 unsigned long long mounted_time; /* mount time */
241 unsigned long long min_mtime; /* min. modification time */
242 unsigned long long max_mtime; /* max. modification time */
244 unsigned int last_victim[MAX_GC_POLICY]; /* last victim segment # */
247 struct free_segmap_info {
248 unsigned int start_segno; /* start segment number logically */
249 unsigned int free_segments; /* # of free segments */
250 unsigned int free_sections; /* # of free sections */
251 spinlock_t segmap_lock; /* free segmap lock */
252 unsigned long *free_segmap; /* free segment bitmap */
253 unsigned long *free_secmap; /* free section bitmap */
256 /* Notice: The order of dirty type is same with CURSEG_XXX in f2fs.h */
258 DIRTY_HOT_DATA, /* dirty segments assigned as hot data logs */
259 DIRTY_WARM_DATA, /* dirty segments assigned as warm data logs */
260 DIRTY_COLD_DATA, /* dirty segments assigned as cold data logs */
261 DIRTY_HOT_NODE, /* dirty segments assigned as hot node logs */
262 DIRTY_WARM_NODE, /* dirty segments assigned as warm node logs */
263 DIRTY_COLD_NODE, /* dirty segments assigned as cold node logs */
264 DIRTY, /* to count # of dirty segments */
265 PRE, /* to count # of entirely obsolete segments */
269 struct dirty_seglist_info {
270 const struct victim_selection *v_ops; /* victim selction operation */
271 unsigned long *dirty_segmap[NR_DIRTY_TYPE];
272 unsigned long *dirty_secmap;
273 struct mutex seglist_lock; /* lock for segment bitmaps */
274 int nr_dirty[NR_DIRTY_TYPE]; /* # of dirty segments */
275 unsigned long *victim_secmap; /* background GC victims */
278 /* victim selection function for cleaning and SSR */
279 struct victim_selection {
280 int (*get_victim)(struct f2fs_sb_info *, unsigned int *,
284 /* for active log information */
286 struct mutex curseg_mutex; /* lock for consistency */
287 struct f2fs_summary_block *sum_blk; /* cached summary block */
288 struct rw_semaphore journal_rwsem; /* protect journal area */
289 struct f2fs_journal *journal; /* cached journal info */
290 unsigned char alloc_type; /* current allocation type */
291 unsigned int segno; /* current segment number */
292 unsigned short next_blkoff; /* next block offset to write */
293 unsigned int zone; /* current zone number */
294 unsigned int next_segno; /* preallocated segment */
297 struct sit_entry_set {
298 struct list_head set_list; /* link with all sit sets */
299 unsigned int start_segno; /* start segno of sits in set */
300 unsigned int entry_cnt; /* the # of sit entries in set */
306 static inline struct curseg_info *CURSEG_I(struct f2fs_sb_info *sbi, int type)
308 if (type == CURSEG_COLD_DATA_PINNED)
309 type = CURSEG_COLD_DATA;
310 return (struct curseg_info *)(SM_I(sbi)->curseg_array + type);
313 static inline struct seg_entry *get_seg_entry(struct f2fs_sb_info *sbi,
316 struct sit_info *sit_i = SIT_I(sbi);
317 return &sit_i->sentries[segno];
320 static inline struct sec_entry *get_sec_entry(struct f2fs_sb_info *sbi,
323 struct sit_info *sit_i = SIT_I(sbi);
324 return &sit_i->sec_entries[GET_SEC_FROM_SEG(sbi, segno)];
327 static inline unsigned int get_valid_blocks(struct f2fs_sb_info *sbi,
328 unsigned int segno, bool use_section)
331 * In order to get # of valid blocks in a section instantly from many
332 * segments, f2fs manages two counting structures separately.
334 if (use_section && __is_large_section(sbi))
335 return get_sec_entry(sbi, segno)->valid_blocks;
337 return get_seg_entry(sbi, segno)->valid_blocks;
340 static inline unsigned int get_ckpt_valid_blocks(struct f2fs_sb_info *sbi,
343 return get_seg_entry(sbi, segno)->ckpt_valid_blocks;
346 static inline void seg_info_from_raw_sit(struct seg_entry *se,
347 struct f2fs_sit_entry *rs)
349 se->valid_blocks = GET_SIT_VBLOCKS(rs);
350 se->ckpt_valid_blocks = GET_SIT_VBLOCKS(rs);
351 memcpy(se->cur_valid_map, rs->valid_map, SIT_VBLOCK_MAP_SIZE);
352 memcpy(se->ckpt_valid_map, rs->valid_map, SIT_VBLOCK_MAP_SIZE);
353 #ifdef CONFIG_F2FS_CHECK_FS
354 memcpy(se->cur_valid_map_mir, rs->valid_map, SIT_VBLOCK_MAP_SIZE);
356 se->type = GET_SIT_TYPE(rs);
357 se->mtime = le64_to_cpu(rs->mtime);
360 static inline void __seg_info_to_raw_sit(struct seg_entry *se,
361 struct f2fs_sit_entry *rs)
363 unsigned short raw_vblocks = (se->type << SIT_VBLOCKS_SHIFT) |
365 rs->vblocks = cpu_to_le16(raw_vblocks);
366 memcpy(rs->valid_map, se->cur_valid_map, SIT_VBLOCK_MAP_SIZE);
367 rs->mtime = cpu_to_le64(se->mtime);
370 static inline void seg_info_to_sit_page(struct f2fs_sb_info *sbi,
371 struct page *page, unsigned int start)
373 struct f2fs_sit_block *raw_sit;
374 struct seg_entry *se;
375 struct f2fs_sit_entry *rs;
376 unsigned int end = min(start + SIT_ENTRY_PER_BLOCK,
377 (unsigned long)MAIN_SEGS(sbi));
380 raw_sit = (struct f2fs_sit_block *)page_address(page);
381 memset(raw_sit, 0, PAGE_SIZE);
382 for (i = 0; i < end - start; i++) {
383 rs = &raw_sit->entries[i];
384 se = get_seg_entry(sbi, start + i);
385 __seg_info_to_raw_sit(se, rs);
389 static inline void seg_info_to_raw_sit(struct seg_entry *se,
390 struct f2fs_sit_entry *rs)
392 __seg_info_to_raw_sit(se, rs);
394 memcpy(se->ckpt_valid_map, rs->valid_map, SIT_VBLOCK_MAP_SIZE);
395 se->ckpt_valid_blocks = se->valid_blocks;
398 static inline unsigned int find_next_inuse(struct free_segmap_info *free_i,
399 unsigned int max, unsigned int segno)
402 spin_lock(&free_i->segmap_lock);
403 ret = find_next_bit(free_i->free_segmap, max, segno);
404 spin_unlock(&free_i->segmap_lock);
408 static inline void __set_free(struct f2fs_sb_info *sbi, unsigned int segno)
410 struct free_segmap_info *free_i = FREE_I(sbi);
411 unsigned int secno = GET_SEC_FROM_SEG(sbi, segno);
412 unsigned int start_segno = GET_SEG_FROM_SEC(sbi, secno);
415 spin_lock(&free_i->segmap_lock);
416 clear_bit(segno, free_i->free_segmap);
417 free_i->free_segments++;
419 next = find_next_bit(free_i->free_segmap,
420 start_segno + sbi->segs_per_sec, start_segno);
421 if (next >= start_segno + sbi->segs_per_sec) {
422 clear_bit(secno, free_i->free_secmap);
423 free_i->free_sections++;
425 spin_unlock(&free_i->segmap_lock);
428 static inline void __set_inuse(struct f2fs_sb_info *sbi,
431 struct free_segmap_info *free_i = FREE_I(sbi);
432 unsigned int secno = GET_SEC_FROM_SEG(sbi, segno);
434 set_bit(segno, free_i->free_segmap);
435 free_i->free_segments--;
436 if (!test_and_set_bit(secno, free_i->free_secmap))
437 free_i->free_sections--;
440 static inline void __set_test_and_free(struct f2fs_sb_info *sbi,
443 struct free_segmap_info *free_i = FREE_I(sbi);
444 unsigned int secno = GET_SEC_FROM_SEG(sbi, segno);
445 unsigned int start_segno = GET_SEG_FROM_SEC(sbi, secno);
448 spin_lock(&free_i->segmap_lock);
449 if (test_and_clear_bit(segno, free_i->free_segmap)) {
450 free_i->free_segments++;
452 if (IS_CURSEC(sbi, secno))
454 next = find_next_bit(free_i->free_segmap,
455 start_segno + sbi->segs_per_sec, start_segno);
456 if (next >= start_segno + sbi->segs_per_sec) {
457 if (test_and_clear_bit(secno, free_i->free_secmap))
458 free_i->free_sections++;
462 spin_unlock(&free_i->segmap_lock);
465 static inline void __set_test_and_inuse(struct f2fs_sb_info *sbi,
468 struct free_segmap_info *free_i = FREE_I(sbi);
469 unsigned int secno = GET_SEC_FROM_SEG(sbi, segno);
471 spin_lock(&free_i->segmap_lock);
472 if (!test_and_set_bit(segno, free_i->free_segmap)) {
473 free_i->free_segments--;
474 if (!test_and_set_bit(secno, free_i->free_secmap))
475 free_i->free_sections--;
477 spin_unlock(&free_i->segmap_lock);
480 static inline void get_sit_bitmap(struct f2fs_sb_info *sbi,
483 struct sit_info *sit_i = SIT_I(sbi);
485 #ifdef CONFIG_F2FS_CHECK_FS
486 if (memcmp(sit_i->sit_bitmap, sit_i->sit_bitmap_mir,
490 memcpy(dst_addr, sit_i->sit_bitmap, sit_i->bitmap_size);
493 static inline block_t written_block_count(struct f2fs_sb_info *sbi)
495 return SIT_I(sbi)->written_valid_blocks;
498 static inline unsigned int free_segments(struct f2fs_sb_info *sbi)
500 return FREE_I(sbi)->free_segments;
503 static inline int reserved_segments(struct f2fs_sb_info *sbi)
505 return SM_I(sbi)->reserved_segments;
508 static inline unsigned int free_sections(struct f2fs_sb_info *sbi)
510 return FREE_I(sbi)->free_sections;
513 static inline unsigned int prefree_segments(struct f2fs_sb_info *sbi)
515 return DIRTY_I(sbi)->nr_dirty[PRE];
518 static inline unsigned int dirty_segments(struct f2fs_sb_info *sbi)
520 return DIRTY_I(sbi)->nr_dirty[DIRTY_HOT_DATA] +
521 DIRTY_I(sbi)->nr_dirty[DIRTY_WARM_DATA] +
522 DIRTY_I(sbi)->nr_dirty[DIRTY_COLD_DATA] +
523 DIRTY_I(sbi)->nr_dirty[DIRTY_HOT_NODE] +
524 DIRTY_I(sbi)->nr_dirty[DIRTY_WARM_NODE] +
525 DIRTY_I(sbi)->nr_dirty[DIRTY_COLD_NODE];
528 static inline int overprovision_segments(struct f2fs_sb_info *sbi)
530 return SM_I(sbi)->ovp_segments;
533 static inline int reserved_sections(struct f2fs_sb_info *sbi)
535 return GET_SEC_FROM_SEG(sbi, (unsigned int)reserved_segments(sbi));
538 static inline bool has_curseg_enough_space(struct f2fs_sb_info *sbi)
540 unsigned int node_blocks = get_pages(sbi, F2FS_DIRTY_NODES) +
541 get_pages(sbi, F2FS_DIRTY_DENTS);
542 unsigned int dent_blocks = get_pages(sbi, F2FS_DIRTY_DENTS);
543 unsigned int segno, left_blocks;
546 /* check current node segment */
547 for (i = CURSEG_HOT_NODE; i <= CURSEG_COLD_NODE; i++) {
548 segno = CURSEG_I(sbi, i)->segno;
549 left_blocks = sbi->blocks_per_seg -
550 get_seg_entry(sbi, segno)->ckpt_valid_blocks;
552 if (node_blocks > left_blocks)
556 /* check current data segment */
557 segno = CURSEG_I(sbi, CURSEG_HOT_DATA)->segno;
558 left_blocks = sbi->blocks_per_seg -
559 get_seg_entry(sbi, segno)->ckpt_valid_blocks;
560 if (dent_blocks > left_blocks)
565 static inline bool has_not_enough_free_secs(struct f2fs_sb_info *sbi,
566 int freed, int needed)
568 int node_secs = get_blocktype_secs(sbi, F2FS_DIRTY_NODES);
569 int dent_secs = get_blocktype_secs(sbi, F2FS_DIRTY_DENTS);
570 int imeta_secs = get_blocktype_secs(sbi, F2FS_DIRTY_IMETA);
572 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
575 if (free_sections(sbi) + freed == reserved_sections(sbi) + needed &&
576 has_curseg_enough_space(sbi))
578 return (free_sections(sbi) + freed) <=
579 (node_secs + 2 * dent_secs + imeta_secs +
580 reserved_sections(sbi) + needed);
583 static inline bool f2fs_is_checkpoint_ready(struct f2fs_sb_info *sbi)
585 if (likely(!is_sbi_flag_set(sbi, SBI_CP_DISABLED)))
587 if (likely(!has_not_enough_free_secs(sbi, 0, 0)))
592 static inline bool excess_prefree_segs(struct f2fs_sb_info *sbi)
594 return prefree_segments(sbi) > SM_I(sbi)->rec_prefree_segments;
597 static inline int utilization(struct f2fs_sb_info *sbi)
599 return div_u64((u64)valid_user_blocks(sbi) * 100,
600 sbi->user_block_count);
604 * Sometimes f2fs may be better to drop out-of-place update policy.
605 * And, users can control the policy through sysfs entries.
606 * There are five policies with triggering conditions as follows.
607 * F2FS_IPU_FORCE - all the time,
608 * F2FS_IPU_SSR - if SSR mode is activated,
609 * F2FS_IPU_UTIL - if FS utilization is over threashold,
610 * F2FS_IPU_SSR_UTIL - if SSR mode is activated and FS utilization is over
612 * F2FS_IPU_FSYNC - activated in fsync path only for high performance flash
613 * storages. IPU will be triggered only if the # of dirty
614 * pages over min_fsync_blocks. (=default option)
615 * F2FS_IPU_ASYNC - do IPU given by asynchronous write requests.
616 * F2FS_IPU_NOCACHE - disable IPU bio cache.
617 * F2FS_IPUT_DISABLE - disable IPU. (=default option in LFS mode)
619 #define DEF_MIN_IPU_UTIL 70
620 #define DEF_MIN_FSYNC_BLOCKS 8
621 #define DEF_MIN_HOT_BLOCKS 16
623 #define SMALL_VOLUME_SEGMENTS (16 * 512) /* 16GB */
635 static inline unsigned int curseg_segno(struct f2fs_sb_info *sbi,
638 struct curseg_info *curseg = CURSEG_I(sbi, type);
639 return curseg->segno;
642 static inline unsigned char curseg_alloc_type(struct f2fs_sb_info *sbi,
645 struct curseg_info *curseg = CURSEG_I(sbi, type);
646 return curseg->alloc_type;
649 static inline unsigned short curseg_blkoff(struct f2fs_sb_info *sbi, int type)
651 struct curseg_info *curseg = CURSEG_I(sbi, type);
652 return curseg->next_blkoff;
655 static inline void check_seg_range(struct f2fs_sb_info *sbi, unsigned int segno)
657 f2fs_bug_on(sbi, segno > TOTAL_SEGS(sbi) - 1);
660 static inline void verify_fio_blkaddr(struct f2fs_io_info *fio)
662 struct f2fs_sb_info *sbi = fio->sbi;
664 if (__is_valid_data_blkaddr(fio->old_blkaddr))
665 verify_blkaddr(sbi, fio->old_blkaddr, __is_meta_io(fio) ?
666 META_GENERIC : DATA_GENERIC);
667 verify_blkaddr(sbi, fio->new_blkaddr, __is_meta_io(fio) ?
668 META_GENERIC : DATA_GENERIC_ENHANCE);
672 * Summary block is always treated as an invalid block
674 static inline int check_block_count(struct f2fs_sb_info *sbi,
675 int segno, struct f2fs_sit_entry *raw_sit)
677 bool is_valid = test_bit_le(0, raw_sit->valid_map) ? true : false;
678 int valid_blocks = 0;
679 int cur_pos = 0, next_pos;
681 /* check bitmap with valid block count */
684 next_pos = find_next_zero_bit_le(&raw_sit->valid_map,
687 valid_blocks += next_pos - cur_pos;
689 next_pos = find_next_bit_le(&raw_sit->valid_map,
693 is_valid = !is_valid;
694 } while (cur_pos < sbi->blocks_per_seg);
696 if (unlikely(GET_SIT_VBLOCKS(raw_sit) != valid_blocks)) {
697 f2fs_err(sbi, "Mismatch valid blocks %d vs. %d",
698 GET_SIT_VBLOCKS(raw_sit), valid_blocks);
699 set_sbi_flag(sbi, SBI_NEED_FSCK);
700 return -EFSCORRUPTED;
703 /* check segment usage, and check boundary of a given segment number */
704 if (unlikely(GET_SIT_VBLOCKS(raw_sit) > sbi->blocks_per_seg
705 || segno > TOTAL_SEGS(sbi) - 1)) {
706 f2fs_err(sbi, "Wrong valid blocks %d or segno %u",
707 GET_SIT_VBLOCKS(raw_sit), segno);
708 set_sbi_flag(sbi, SBI_NEED_FSCK);
709 return -EFSCORRUPTED;
714 static inline pgoff_t current_sit_addr(struct f2fs_sb_info *sbi,
717 struct sit_info *sit_i = SIT_I(sbi);
718 unsigned int offset = SIT_BLOCK_OFFSET(start);
719 block_t blk_addr = sit_i->sit_base_addr + offset;
721 check_seg_range(sbi, start);
723 #ifdef CONFIG_F2FS_CHECK_FS
724 if (f2fs_test_bit(offset, sit_i->sit_bitmap) !=
725 f2fs_test_bit(offset, sit_i->sit_bitmap_mir))
729 /* calculate sit block address */
730 if (f2fs_test_bit(offset, sit_i->sit_bitmap))
731 blk_addr += sit_i->sit_blocks;
736 static inline pgoff_t next_sit_addr(struct f2fs_sb_info *sbi,
739 struct sit_info *sit_i = SIT_I(sbi);
740 block_addr -= sit_i->sit_base_addr;
741 if (block_addr < sit_i->sit_blocks)
742 block_addr += sit_i->sit_blocks;
744 block_addr -= sit_i->sit_blocks;
746 return block_addr + sit_i->sit_base_addr;
749 static inline void set_to_next_sit(struct sit_info *sit_i, unsigned int start)
751 unsigned int block_off = SIT_BLOCK_OFFSET(start);
753 f2fs_change_bit(block_off, sit_i->sit_bitmap);
754 #ifdef CONFIG_F2FS_CHECK_FS
755 f2fs_change_bit(block_off, sit_i->sit_bitmap_mir);
759 static inline unsigned long long get_mtime(struct f2fs_sb_info *sbi,
762 struct sit_info *sit_i = SIT_I(sbi);
763 time64_t diff, now = ktime_get_boottime_seconds();
765 if (now >= sit_i->mounted_time)
766 return sit_i->elapsed_time + now - sit_i->mounted_time;
768 /* system time is set to the past */
770 diff = sit_i->mounted_time - now;
771 if (sit_i->elapsed_time >= diff)
772 return sit_i->elapsed_time - diff;
775 return sit_i->elapsed_time;
778 static inline void set_summary(struct f2fs_summary *sum, nid_t nid,
779 unsigned int ofs_in_node, unsigned char version)
781 sum->nid = cpu_to_le32(nid);
782 sum->ofs_in_node = cpu_to_le16(ofs_in_node);
783 sum->version = version;
786 static inline block_t start_sum_block(struct f2fs_sb_info *sbi)
788 return __start_cp_addr(sbi) +
789 le32_to_cpu(F2FS_CKPT(sbi)->cp_pack_start_sum);
792 static inline block_t sum_blk_addr(struct f2fs_sb_info *sbi, int base, int type)
794 return __start_cp_addr(sbi) +
795 le32_to_cpu(F2FS_CKPT(sbi)->cp_pack_total_block_count)
799 static inline bool sec_usage_check(struct f2fs_sb_info *sbi, unsigned int secno)
801 if (IS_CURSEC(sbi, secno) || (sbi->cur_victim_sec == secno))
807 * It is very important to gather dirty pages and write at once, so that we can
808 * submit a big bio without interfering other data writes.
809 * By default, 512 pages for directory data,
810 * 512 pages (2MB) * 8 for nodes, and
811 * 256 pages * 8 for meta are set.
813 static inline int nr_pages_to_skip(struct f2fs_sb_info *sbi, int type)
815 if (sbi->sb->s_bdi->wb.dirty_exceeded)
819 return sbi->blocks_per_seg;
820 else if (type == NODE)
821 return 8 * sbi->blocks_per_seg;
822 else if (type == META)
823 return 8 * BIO_MAX_PAGES;
829 * When writing pages, it'd better align nr_to_write for segment size.
831 static inline long nr_pages_to_write(struct f2fs_sb_info *sbi, int type,
832 struct writeback_control *wbc)
834 long nr_to_write, desired;
836 if (wbc->sync_mode != WB_SYNC_NONE)
839 nr_to_write = wbc->nr_to_write;
840 desired = BIO_MAX_PAGES;
844 wbc->nr_to_write = desired;
845 return desired - nr_to_write;
848 static inline void wake_up_discard_thread(struct f2fs_sb_info *sbi, bool force)
850 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
857 mutex_lock(&dcc->cmd_lock);
858 for (i = MAX_PLIST_NUM - 1; i >= 0; i--) {
859 if (i + 1 < dcc->discard_granularity)
861 if (!list_empty(&dcc->pend_list[i])) {
866 mutex_unlock(&dcc->cmd_lock);
867 if (!wakeup || !is_idle(sbi, DISCARD_TIME))
870 dcc->discard_wake = 1;
871 wake_up_interruptible_all(&dcc->discard_wait_queue);