4 * Copyright (c) 2012 Samsung Electronics Co., Ltd.
5 * http://www.samsung.com/
7 * This program is free software; you can redistribute it and/or modify
8 * it under the terms of the GNU General Public License version 2 as
9 * published by the Free Software Foundation.
11 #include <linux/blkdev.h>
14 #define NULL_SEGNO ((unsigned int)(~0))
15 #define NULL_SECNO ((unsigned int)(~0))
17 /* L: Logical segment # in volume, R: Relative segment # in main area */
18 #define GET_L2R_SEGNO(free_i, segno) (segno - free_i->start_segno)
19 #define GET_R2L_SEGNO(free_i, segno) (segno + free_i->start_segno)
21 #define IS_DATASEG(t) \
22 ((t == CURSEG_HOT_DATA) || (t == CURSEG_COLD_DATA) || \
23 (t == CURSEG_WARM_DATA))
25 #define IS_NODESEG(t) \
26 ((t == CURSEG_HOT_NODE) || (t == CURSEG_COLD_NODE) || \
27 (t == CURSEG_WARM_NODE))
29 #define IS_CURSEG(sbi, seg) \
30 ((seg == CURSEG_I(sbi, CURSEG_HOT_DATA)->segno) || \
31 (seg == CURSEG_I(sbi, CURSEG_WARM_DATA)->segno) || \
32 (seg == CURSEG_I(sbi, CURSEG_COLD_DATA)->segno) || \
33 (seg == CURSEG_I(sbi, CURSEG_HOT_NODE)->segno) || \
34 (seg == CURSEG_I(sbi, CURSEG_WARM_NODE)->segno) || \
35 (seg == CURSEG_I(sbi, CURSEG_COLD_NODE)->segno))
37 #define IS_CURSEC(sbi, secno) \
38 ((secno == CURSEG_I(sbi, CURSEG_HOT_DATA)->segno / \
39 sbi->segs_per_sec) || \
40 (secno == CURSEG_I(sbi, CURSEG_WARM_DATA)->segno / \
41 sbi->segs_per_sec) || \
42 (secno == CURSEG_I(sbi, CURSEG_COLD_DATA)->segno / \
43 sbi->segs_per_sec) || \
44 (secno == CURSEG_I(sbi, CURSEG_HOT_NODE)->segno / \
45 sbi->segs_per_sec) || \
46 (secno == CURSEG_I(sbi, CURSEG_WARM_NODE)->segno / \
47 sbi->segs_per_sec) || \
48 (secno == CURSEG_I(sbi, CURSEG_COLD_NODE)->segno / \
51 #define START_BLOCK(sbi, segno) \
52 (SM_I(sbi)->seg0_blkaddr + \
53 (GET_R2L_SEGNO(FREE_I(sbi), segno) << sbi->log_blocks_per_seg))
54 #define NEXT_FREE_BLKADDR(sbi, curseg) \
55 (START_BLOCK(sbi, curseg->segno) + curseg->next_blkoff)
57 #define MAIN_BASE_BLOCK(sbi) (SM_I(sbi)->main_blkaddr)
59 #define GET_SEGOFF_FROM_SEG0(sbi, blk_addr) \
60 ((blk_addr) - SM_I(sbi)->seg0_blkaddr)
61 #define GET_SEGNO_FROM_SEG0(sbi, blk_addr) \
62 (GET_SEGOFF_FROM_SEG0(sbi, blk_addr) >> sbi->log_blocks_per_seg)
63 #define GET_SEGNO(sbi, blk_addr) \
64 (((blk_addr == NULL_ADDR) || (blk_addr == NEW_ADDR)) ? \
65 NULL_SEGNO : GET_L2R_SEGNO(FREE_I(sbi), \
66 GET_SEGNO_FROM_SEG0(sbi, blk_addr)))
67 #define GET_SECNO(sbi, segno) \
68 ((segno) / sbi->segs_per_sec)
69 #define GET_ZONENO_FROM_SEGNO(sbi, segno) \
70 ((segno / sbi->segs_per_sec) / sbi->secs_per_zone)
72 #define GET_SUM_BLOCK(sbi, segno) \
73 ((sbi->sm_info->ssa_blkaddr) + segno)
75 #define GET_SUM_TYPE(footer) ((footer)->entry_type)
76 #define SET_SUM_TYPE(footer, type) ((footer)->entry_type = type)
78 #define SIT_ENTRY_OFFSET(sit_i, segno) \
79 (segno % sit_i->sents_per_block)
80 #define SIT_BLOCK_OFFSET(sit_i, segno) \
81 (segno / SIT_ENTRY_PER_BLOCK)
82 #define START_SEGNO(sit_i, segno) \
83 (SIT_BLOCK_OFFSET(sit_i, segno) * SIT_ENTRY_PER_BLOCK)
84 #define f2fs_bitmap_size(nr) \
85 (BITS_TO_LONGS(nr) * sizeof(unsigned long))
86 #define TOTAL_SEGS(sbi) (SM_I(sbi)->main_segments)
87 #define TOTAL_SECS(sbi) (sbi->total_sections)
89 #define SECTOR_FROM_BLOCK(sbi, blk_addr) \
90 (blk_addr << ((sbi)->log_blocksize - F2FS_LOG_SECTOR_SIZE))
91 #define SECTOR_TO_BLOCK(sbi, sectors) \
92 (sectors >> ((sbi)->log_blocksize - F2FS_LOG_SECTOR_SIZE))
94 /* during checkpoint, bio_private is used to synchronize the last bio */
96 struct f2fs_sb_info *sbi;
102 * indicate a block allocation direction: RIGHT and LEFT.
103 * RIGHT means allocating new sections towards the end of volume.
104 * LEFT means the opposite direction.
112 * In the victim_sel_policy->alloc_mode, there are two block allocation modes.
113 * LFS writes data sequentially with cleaning operations.
114 * SSR (Slack Space Recycle) reuses obsolete space without cleaning operations.
122 * In the victim_sel_policy->gc_mode, there are two gc, aka cleaning, modes.
123 * GC_CB is based on cost-benefit algorithm.
124 * GC_GREEDY is based on greedy algorithm.
132 * BG_GC means the background cleaning job.
133 * FG_GC means the on-demand cleaning job.
140 /* for a function parameter to select a victim segment */
141 struct victim_sel_policy {
142 int alloc_mode; /* LFS or SSR */
143 int gc_mode; /* GC_CB or GC_GREEDY */
144 unsigned long *dirty_segmap; /* dirty segment bitmap */
145 unsigned int max_search; /* maximum # of segments to search */
146 unsigned int offset; /* last scanned bitmap offset */
147 unsigned int ofs_unit; /* bitmap search unit */
148 unsigned int min_cost; /* minimum cost */
149 unsigned int min_segno; /* segment # having min. cost */
153 unsigned short valid_blocks; /* # of valid blocks */
154 unsigned char *cur_valid_map; /* validity bitmap of blocks */
156 * # of valid blocks and the validity bitmap stored in the the last
157 * checkpoint pack. This information is used by the SSR mode.
159 unsigned short ckpt_valid_blocks;
160 unsigned char *ckpt_valid_map;
161 unsigned char type; /* segment type like CURSEG_XXX_TYPE */
162 unsigned long long mtime; /* modification time of the segment */
166 unsigned int valid_blocks; /* # of valid blocks in a section */
169 struct segment_allocation {
170 void (*allocate_segment)(struct f2fs_sb_info *, int, bool);
174 const struct segment_allocation *s_ops;
176 block_t sit_base_addr; /* start block address of SIT area */
177 block_t sit_blocks; /* # of blocks used by SIT area */
178 block_t written_valid_blocks; /* # of valid blocks in main area */
179 char *sit_bitmap; /* SIT bitmap pointer */
180 unsigned int bitmap_size; /* SIT bitmap size */
182 unsigned long *dirty_sentries_bitmap; /* bitmap for dirty sentries */
183 unsigned int dirty_sentries; /* # of dirty sentries */
184 unsigned int sents_per_block; /* # of SIT entries per block */
185 struct mutex sentry_lock; /* to protect SIT cache */
186 struct seg_entry *sentries; /* SIT segment-level cache */
187 struct sec_entry *sec_entries; /* SIT section-level cache */
189 /* for cost-benefit algorithm in cleaning procedure */
190 unsigned long long elapsed_time; /* elapsed time after mount */
191 unsigned long long mounted_time; /* mount time */
192 unsigned long long min_mtime; /* min. modification time */
193 unsigned long long max_mtime; /* max. modification time */
196 struct free_segmap_info {
197 unsigned int start_segno; /* start segment number logically */
198 unsigned int free_segments; /* # of free segments */
199 unsigned int free_sections; /* # of free sections */
200 rwlock_t segmap_lock; /* free segmap lock */
201 unsigned long *free_segmap; /* free segment bitmap */
202 unsigned long *free_secmap; /* free section bitmap */
205 /* Notice: The order of dirty type is same with CURSEG_XXX in f2fs.h */
207 DIRTY_HOT_DATA, /* dirty segments assigned as hot data logs */
208 DIRTY_WARM_DATA, /* dirty segments assigned as warm data logs */
209 DIRTY_COLD_DATA, /* dirty segments assigned as cold data logs */
210 DIRTY_HOT_NODE, /* dirty segments assigned as hot node logs */
211 DIRTY_WARM_NODE, /* dirty segments assigned as warm node logs */
212 DIRTY_COLD_NODE, /* dirty segments assigned as cold node logs */
213 DIRTY, /* to count # of dirty segments */
214 PRE, /* to count # of entirely obsolete segments */
218 struct dirty_seglist_info {
219 const struct victim_selection *v_ops; /* victim selction operation */
220 unsigned long *dirty_segmap[NR_DIRTY_TYPE];
221 struct mutex seglist_lock; /* lock for segment bitmaps */
222 int nr_dirty[NR_DIRTY_TYPE]; /* # of dirty segments */
223 unsigned long *victim_secmap; /* background GC victims */
226 /* victim selection function for cleaning and SSR */
227 struct victim_selection {
228 int (*get_victim)(struct f2fs_sb_info *, unsigned int *,
232 /* for active log information */
234 struct mutex curseg_mutex; /* lock for consistency */
235 struct f2fs_summary_block *sum_blk; /* cached summary block */
236 unsigned char alloc_type; /* current allocation type */
237 unsigned int segno; /* current segment number */
238 unsigned short next_blkoff; /* next block offset to write */
239 unsigned int zone; /* current zone number */
240 unsigned int next_segno; /* preallocated segment */
246 static inline struct curseg_info *CURSEG_I(struct f2fs_sb_info *sbi, int type)
248 return (struct curseg_info *)(SM_I(sbi)->curseg_array + type);
251 static inline struct seg_entry *get_seg_entry(struct f2fs_sb_info *sbi,
254 struct sit_info *sit_i = SIT_I(sbi);
255 return &sit_i->sentries[segno];
258 static inline struct sec_entry *get_sec_entry(struct f2fs_sb_info *sbi,
261 struct sit_info *sit_i = SIT_I(sbi);
262 return &sit_i->sec_entries[GET_SECNO(sbi, segno)];
265 static inline unsigned int get_valid_blocks(struct f2fs_sb_info *sbi,
266 unsigned int segno, int section)
269 * In order to get # of valid blocks in a section instantly from many
270 * segments, f2fs manages two counting structures separately.
273 return get_sec_entry(sbi, segno)->valid_blocks;
275 return get_seg_entry(sbi, segno)->valid_blocks;
278 static inline void seg_info_from_raw_sit(struct seg_entry *se,
279 struct f2fs_sit_entry *rs)
281 se->valid_blocks = GET_SIT_VBLOCKS(rs);
282 se->ckpt_valid_blocks = GET_SIT_VBLOCKS(rs);
283 memcpy(se->cur_valid_map, rs->valid_map, SIT_VBLOCK_MAP_SIZE);
284 memcpy(se->ckpt_valid_map, rs->valid_map, SIT_VBLOCK_MAP_SIZE);
285 se->type = GET_SIT_TYPE(rs);
286 se->mtime = le64_to_cpu(rs->mtime);
289 static inline void seg_info_to_raw_sit(struct seg_entry *se,
290 struct f2fs_sit_entry *rs)
292 unsigned short raw_vblocks = (se->type << SIT_VBLOCKS_SHIFT) |
294 rs->vblocks = cpu_to_le16(raw_vblocks);
295 memcpy(rs->valid_map, se->cur_valid_map, SIT_VBLOCK_MAP_SIZE);
296 memcpy(se->ckpt_valid_map, rs->valid_map, SIT_VBLOCK_MAP_SIZE);
297 se->ckpt_valid_blocks = se->valid_blocks;
298 rs->mtime = cpu_to_le64(se->mtime);
301 static inline unsigned int find_next_inuse(struct free_segmap_info *free_i,
302 unsigned int max, unsigned int segno)
305 read_lock(&free_i->segmap_lock);
306 ret = find_next_bit(free_i->free_segmap, max, segno);
307 read_unlock(&free_i->segmap_lock);
311 static inline void __set_free(struct f2fs_sb_info *sbi, unsigned int segno)
313 struct free_segmap_info *free_i = FREE_I(sbi);
314 unsigned int secno = segno / sbi->segs_per_sec;
315 unsigned int start_segno = secno * sbi->segs_per_sec;
318 write_lock(&free_i->segmap_lock);
319 clear_bit(segno, free_i->free_segmap);
320 free_i->free_segments++;
322 next = find_next_bit(free_i->free_segmap, TOTAL_SEGS(sbi), start_segno);
323 if (next >= start_segno + sbi->segs_per_sec) {
324 clear_bit(secno, free_i->free_secmap);
325 free_i->free_sections++;
327 write_unlock(&free_i->segmap_lock);
330 static inline void __set_inuse(struct f2fs_sb_info *sbi,
333 struct free_segmap_info *free_i = FREE_I(sbi);
334 unsigned int secno = segno / sbi->segs_per_sec;
335 set_bit(segno, free_i->free_segmap);
336 free_i->free_segments--;
337 if (!test_and_set_bit(secno, free_i->free_secmap))
338 free_i->free_sections--;
341 static inline void __set_test_and_free(struct f2fs_sb_info *sbi,
344 struct free_segmap_info *free_i = FREE_I(sbi);
345 unsigned int secno = segno / sbi->segs_per_sec;
346 unsigned int start_segno = secno * sbi->segs_per_sec;
349 write_lock(&free_i->segmap_lock);
350 if (test_and_clear_bit(segno, free_i->free_segmap)) {
351 free_i->free_segments++;
353 next = find_next_bit(free_i->free_segmap, TOTAL_SEGS(sbi),
355 if (next >= start_segno + sbi->segs_per_sec) {
356 if (test_and_clear_bit(secno, free_i->free_secmap))
357 free_i->free_sections++;
360 write_unlock(&free_i->segmap_lock);
363 static inline void __set_test_and_inuse(struct f2fs_sb_info *sbi,
366 struct free_segmap_info *free_i = FREE_I(sbi);
367 unsigned int secno = segno / sbi->segs_per_sec;
368 write_lock(&free_i->segmap_lock);
369 if (!test_and_set_bit(segno, free_i->free_segmap)) {
370 free_i->free_segments--;
371 if (!test_and_set_bit(secno, free_i->free_secmap))
372 free_i->free_sections--;
374 write_unlock(&free_i->segmap_lock);
377 static inline void get_sit_bitmap(struct f2fs_sb_info *sbi,
380 struct sit_info *sit_i = SIT_I(sbi);
381 memcpy(dst_addr, sit_i->sit_bitmap, sit_i->bitmap_size);
384 static inline block_t written_block_count(struct f2fs_sb_info *sbi)
386 struct sit_info *sit_i = SIT_I(sbi);
389 mutex_lock(&sit_i->sentry_lock);
390 vblocks = sit_i->written_valid_blocks;
391 mutex_unlock(&sit_i->sentry_lock);
396 static inline unsigned int free_segments(struct f2fs_sb_info *sbi)
398 struct free_segmap_info *free_i = FREE_I(sbi);
399 unsigned int free_segs;
401 read_lock(&free_i->segmap_lock);
402 free_segs = free_i->free_segments;
403 read_unlock(&free_i->segmap_lock);
408 static inline int reserved_segments(struct f2fs_sb_info *sbi)
410 return SM_I(sbi)->reserved_segments;
413 static inline unsigned int free_sections(struct f2fs_sb_info *sbi)
415 struct free_segmap_info *free_i = FREE_I(sbi);
416 unsigned int free_secs;
418 read_lock(&free_i->segmap_lock);
419 free_secs = free_i->free_sections;
420 read_unlock(&free_i->segmap_lock);
425 static inline unsigned int prefree_segments(struct f2fs_sb_info *sbi)
427 return DIRTY_I(sbi)->nr_dirty[PRE];
430 static inline unsigned int dirty_segments(struct f2fs_sb_info *sbi)
432 return DIRTY_I(sbi)->nr_dirty[DIRTY_HOT_DATA] +
433 DIRTY_I(sbi)->nr_dirty[DIRTY_WARM_DATA] +
434 DIRTY_I(sbi)->nr_dirty[DIRTY_COLD_DATA] +
435 DIRTY_I(sbi)->nr_dirty[DIRTY_HOT_NODE] +
436 DIRTY_I(sbi)->nr_dirty[DIRTY_WARM_NODE] +
437 DIRTY_I(sbi)->nr_dirty[DIRTY_COLD_NODE];
440 static inline int overprovision_segments(struct f2fs_sb_info *sbi)
442 return SM_I(sbi)->ovp_segments;
445 static inline int overprovision_sections(struct f2fs_sb_info *sbi)
447 return ((unsigned int) overprovision_segments(sbi)) / sbi->segs_per_sec;
450 static inline int reserved_sections(struct f2fs_sb_info *sbi)
452 return ((unsigned int) reserved_segments(sbi)) / sbi->segs_per_sec;
455 static inline bool need_SSR(struct f2fs_sb_info *sbi)
457 return ((prefree_segments(sbi) / sbi->segs_per_sec)
458 + free_sections(sbi) < overprovision_sections(sbi));
461 static inline bool has_not_enough_free_secs(struct f2fs_sb_info *sbi, int freed)
463 int node_secs = get_blocktype_secs(sbi, F2FS_DIRTY_NODES);
464 int dent_secs = get_blocktype_secs(sbi, F2FS_DIRTY_DENTS);
469 return ((free_sections(sbi) + freed) <= (node_secs + 2 * dent_secs +
470 reserved_sections(sbi)));
473 static inline int utilization(struct f2fs_sb_info *sbi)
475 return div_u64((u64)valid_user_blocks(sbi) * 100, sbi->user_block_count);
479 * Sometimes f2fs may be better to drop out-of-place update policy.
480 * So, if fs utilization is over MIN_IPU_UTIL, then f2fs tries to write
481 * data in the original place likewise other traditional file systems.
482 * But, currently set 100 in percentage, which means it is disabled.
483 * See below need_inplace_update().
485 #define MIN_IPU_UTIL 100
486 static inline bool need_inplace_update(struct inode *inode)
488 struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
489 if (S_ISDIR(inode->i_mode))
491 if (need_SSR(sbi) && utilization(sbi) > MIN_IPU_UTIL)
496 static inline unsigned int curseg_segno(struct f2fs_sb_info *sbi,
499 struct curseg_info *curseg = CURSEG_I(sbi, type);
500 return curseg->segno;
503 static inline unsigned char curseg_alloc_type(struct f2fs_sb_info *sbi,
506 struct curseg_info *curseg = CURSEG_I(sbi, type);
507 return curseg->alloc_type;
510 static inline unsigned short curseg_blkoff(struct f2fs_sb_info *sbi, int type)
512 struct curseg_info *curseg = CURSEG_I(sbi, type);
513 return curseg->next_blkoff;
516 static inline void check_seg_range(struct f2fs_sb_info *sbi, unsigned int segno)
518 unsigned int end_segno = SM_I(sbi)->segment_count - 1;
519 BUG_ON(segno > end_segno);
523 * This function is used for only debugging.
524 * NOTE: In future, we have to remove this function.
526 static inline void verify_block_addr(struct f2fs_sb_info *sbi, block_t blk_addr)
528 struct f2fs_sm_info *sm_info = SM_I(sbi);
529 block_t total_blks = sm_info->segment_count << sbi->log_blocks_per_seg;
530 block_t start_addr = sm_info->seg0_blkaddr;
531 block_t end_addr = start_addr + total_blks - 1;
532 BUG_ON(blk_addr < start_addr);
533 BUG_ON(blk_addr > end_addr);
537 * Summary block is always treated as invalid block
539 static inline void check_block_count(struct f2fs_sb_info *sbi,
540 int segno, struct f2fs_sit_entry *raw_sit)
542 struct f2fs_sm_info *sm_info = SM_I(sbi);
543 unsigned int end_segno = sm_info->segment_count - 1;
544 int valid_blocks = 0;
547 /* check segment usage */
548 BUG_ON(GET_SIT_VBLOCKS(raw_sit) > sbi->blocks_per_seg);
550 /* check boundary of a given segment number */
551 BUG_ON(segno > end_segno);
553 /* check bitmap with valid block count */
554 for (i = 0; i < sbi->blocks_per_seg; i++)
555 if (f2fs_test_bit(i, raw_sit->valid_map))
557 BUG_ON(GET_SIT_VBLOCKS(raw_sit) != valid_blocks);
560 static inline pgoff_t current_sit_addr(struct f2fs_sb_info *sbi,
563 struct sit_info *sit_i = SIT_I(sbi);
564 unsigned int offset = SIT_BLOCK_OFFSET(sit_i, start);
565 block_t blk_addr = sit_i->sit_base_addr + offset;
567 check_seg_range(sbi, start);
569 /* calculate sit block address */
570 if (f2fs_test_bit(offset, sit_i->sit_bitmap))
571 blk_addr += sit_i->sit_blocks;
576 static inline pgoff_t next_sit_addr(struct f2fs_sb_info *sbi,
579 struct sit_info *sit_i = SIT_I(sbi);
580 block_addr -= sit_i->sit_base_addr;
581 if (block_addr < sit_i->sit_blocks)
582 block_addr += sit_i->sit_blocks;
584 block_addr -= sit_i->sit_blocks;
586 return block_addr + sit_i->sit_base_addr;
589 static inline void set_to_next_sit(struct sit_info *sit_i, unsigned int start)
591 unsigned int block_off = SIT_BLOCK_OFFSET(sit_i, start);
593 if (f2fs_test_bit(block_off, sit_i->sit_bitmap))
594 f2fs_clear_bit(block_off, sit_i->sit_bitmap);
596 f2fs_set_bit(block_off, sit_i->sit_bitmap);
599 static inline unsigned long long get_mtime(struct f2fs_sb_info *sbi)
601 struct sit_info *sit_i = SIT_I(sbi);
602 return sit_i->elapsed_time + CURRENT_TIME_SEC.tv_sec -
606 static inline void set_summary(struct f2fs_summary *sum, nid_t nid,
607 unsigned int ofs_in_node, unsigned char version)
609 sum->nid = cpu_to_le32(nid);
610 sum->ofs_in_node = cpu_to_le16(ofs_in_node);
611 sum->version = version;
614 static inline block_t start_sum_block(struct f2fs_sb_info *sbi)
616 return __start_cp_addr(sbi) +
617 le32_to_cpu(F2FS_CKPT(sbi)->cp_pack_start_sum);
620 static inline block_t sum_blk_addr(struct f2fs_sb_info *sbi, int base, int type)
622 return __start_cp_addr(sbi) +
623 le32_to_cpu(F2FS_CKPT(sbi)->cp_pack_total_block_count)
627 static inline bool sec_usage_check(struct f2fs_sb_info *sbi, unsigned int secno)
629 if (IS_CURSEC(sbi, secno) || (sbi->cur_victim_sec == secno))
634 static inline unsigned int max_hw_blocks(struct f2fs_sb_info *sbi)
636 struct block_device *bdev = sbi->sb->s_bdev;
637 struct request_queue *q = bdev_get_queue(bdev);
638 return SECTOR_TO_BLOCK(sbi, queue_max_sectors(q));