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 #define DEF_RECLAIM_PREFREE_SEGMENTS 100 /* 200MB of prefree segments */
19 /* L: Logical segment # in volume, R: Relative segment # in main area */
20 #define GET_L2R_SEGNO(free_i, segno) (segno - free_i->start_segno)
21 #define GET_R2L_SEGNO(free_i, segno) (segno + free_i->start_segno)
23 #define IS_DATASEG(t) (t <= CURSEG_COLD_DATA)
24 #define IS_NODESEG(t) (t >= CURSEG_HOT_NODE)
26 #define IS_CURSEG(sbi, seg) \
27 ((seg == CURSEG_I(sbi, CURSEG_HOT_DATA)->segno) || \
28 (seg == CURSEG_I(sbi, CURSEG_WARM_DATA)->segno) || \
29 (seg == CURSEG_I(sbi, CURSEG_COLD_DATA)->segno) || \
30 (seg == CURSEG_I(sbi, CURSEG_HOT_NODE)->segno) || \
31 (seg == CURSEG_I(sbi, CURSEG_WARM_NODE)->segno) || \
32 (seg == CURSEG_I(sbi, CURSEG_COLD_NODE)->segno))
34 #define IS_CURSEC(sbi, secno) \
35 ((secno == CURSEG_I(sbi, CURSEG_HOT_DATA)->segno / \
36 sbi->segs_per_sec) || \
37 (secno == CURSEG_I(sbi, CURSEG_WARM_DATA)->segno / \
38 sbi->segs_per_sec) || \
39 (secno == CURSEG_I(sbi, CURSEG_COLD_DATA)->segno / \
40 sbi->segs_per_sec) || \
41 (secno == CURSEG_I(sbi, CURSEG_HOT_NODE)->segno / \
42 sbi->segs_per_sec) || \
43 (secno == CURSEG_I(sbi, CURSEG_WARM_NODE)->segno / \
44 sbi->segs_per_sec) || \
45 (secno == CURSEG_I(sbi, CURSEG_COLD_NODE)->segno / \
48 #define START_BLOCK(sbi, segno) \
49 (SM_I(sbi)->seg0_blkaddr + \
50 (GET_R2L_SEGNO(FREE_I(sbi), segno) << sbi->log_blocks_per_seg))
51 #define NEXT_FREE_BLKADDR(sbi, curseg) \
52 (START_BLOCK(sbi, curseg->segno) + curseg->next_blkoff)
54 #define MAIN_BASE_BLOCK(sbi) (SM_I(sbi)->main_blkaddr)
56 #define GET_SEGOFF_FROM_SEG0(sbi, blk_addr) \
57 ((blk_addr) - SM_I(sbi)->seg0_blkaddr)
58 #define GET_SEGNO_FROM_SEG0(sbi, blk_addr) \
59 (GET_SEGOFF_FROM_SEG0(sbi, blk_addr) >> sbi->log_blocks_per_seg)
60 #define GET_SEGNO(sbi, blk_addr) \
61 (((blk_addr == NULL_ADDR) || (blk_addr == NEW_ADDR)) ? \
62 NULL_SEGNO : GET_L2R_SEGNO(FREE_I(sbi), \
63 GET_SEGNO_FROM_SEG0(sbi, blk_addr)))
64 #define GET_SECNO(sbi, segno) \
65 ((segno) / sbi->segs_per_sec)
66 #define GET_ZONENO_FROM_SEGNO(sbi, segno) \
67 ((segno / sbi->segs_per_sec) / sbi->secs_per_zone)
69 #define GET_SUM_BLOCK(sbi, segno) \
70 ((sbi->sm_info->ssa_blkaddr) + segno)
72 #define GET_SUM_TYPE(footer) ((footer)->entry_type)
73 #define SET_SUM_TYPE(footer, type) ((footer)->entry_type = type)
75 #define SIT_ENTRY_OFFSET(sit_i, segno) \
76 (segno % sit_i->sents_per_block)
77 #define SIT_BLOCK_OFFSET(sit_i, segno) \
78 (segno / SIT_ENTRY_PER_BLOCK)
79 #define START_SEGNO(sit_i, segno) \
80 (SIT_BLOCK_OFFSET(sit_i, segno) * SIT_ENTRY_PER_BLOCK)
81 #define f2fs_bitmap_size(nr) \
82 (BITS_TO_LONGS(nr) * sizeof(unsigned long))
83 #define TOTAL_SEGS(sbi) (SM_I(sbi)->main_segments)
84 #define TOTAL_SECS(sbi) (sbi->total_sections)
86 #define SECTOR_FROM_BLOCK(sbi, blk_addr) \
87 (blk_addr << ((sbi)->log_blocksize - F2FS_LOG_SECTOR_SIZE))
88 #define SECTOR_TO_BLOCK(sbi, sectors) \
89 (sectors >> ((sbi)->log_blocksize - F2FS_LOG_SECTOR_SIZE))
90 #define MAX_BIO_BLOCKS(max_hw_blocks) \
91 (min((int)max_hw_blocks, BIO_MAX_PAGES))
93 /* during checkpoint, bio_private is used to synchronize the last bio */
95 struct f2fs_sb_info *sbi;
101 * indicate a block allocation direction: RIGHT and LEFT.
102 * RIGHT means allocating new sections towards the end of volume.
103 * LEFT means the opposite direction.
111 * In the victim_sel_policy->alloc_mode, there are two block allocation modes.
112 * LFS writes data sequentially with cleaning operations.
113 * SSR (Slack Space Recycle) reuses obsolete space without cleaning operations.
121 * In the victim_sel_policy->gc_mode, there are two gc, aka cleaning, modes.
122 * GC_CB is based on cost-benefit algorithm.
123 * GC_GREEDY is based on greedy algorithm.
131 * BG_GC means the background cleaning job.
132 * FG_GC means the on-demand cleaning job.
139 /* for a function parameter to select a victim segment */
140 struct victim_sel_policy {
141 int alloc_mode; /* LFS or SSR */
142 int gc_mode; /* GC_CB or GC_GREEDY */
143 unsigned long *dirty_segmap; /* dirty segment bitmap */
144 unsigned int max_search; /* maximum # of segments to search */
145 unsigned int offset; /* last scanned bitmap offset */
146 unsigned int ofs_unit; /* bitmap search unit */
147 unsigned int min_cost; /* minimum cost */
148 unsigned int min_segno; /* segment # having min. cost */
152 unsigned short valid_blocks; /* # of valid blocks */
153 unsigned char *cur_valid_map; /* validity bitmap of blocks */
155 * # of valid blocks and the validity bitmap stored in the the last
156 * checkpoint pack. This information is used by the SSR mode.
158 unsigned short ckpt_valid_blocks;
159 unsigned char *ckpt_valid_map;
160 unsigned char type; /* segment type like CURSEG_XXX_TYPE */
161 unsigned long long mtime; /* modification time of the segment */
165 unsigned int valid_blocks; /* # of valid blocks in a section */
168 struct segment_allocation {
169 void (*allocate_segment)(struct f2fs_sb_info *, int, bool);
173 const struct segment_allocation *s_ops;
175 block_t sit_base_addr; /* start block address of SIT area */
176 block_t sit_blocks; /* # of blocks used by SIT area */
177 block_t written_valid_blocks; /* # of valid blocks in main area */
178 char *sit_bitmap; /* SIT bitmap pointer */
179 unsigned int bitmap_size; /* SIT bitmap size */
181 unsigned long *dirty_sentries_bitmap; /* bitmap for dirty sentries */
182 unsigned int dirty_sentries; /* # of dirty sentries */
183 unsigned int sents_per_block; /* # of SIT entries per block */
184 struct mutex sentry_lock; /* to protect SIT cache */
185 struct seg_entry *sentries; /* SIT segment-level cache */
186 struct sec_entry *sec_entries; /* SIT section-level cache */
188 /* for cost-benefit algorithm in cleaning procedure */
189 unsigned long long elapsed_time; /* elapsed time after mount */
190 unsigned long long mounted_time; /* mount time */
191 unsigned long long min_mtime; /* min. modification time */
192 unsigned long long max_mtime; /* max. modification time */
195 struct free_segmap_info {
196 unsigned int start_segno; /* start segment number logically */
197 unsigned int free_segments; /* # of free segments */
198 unsigned int free_sections; /* # of free sections */
199 rwlock_t segmap_lock; /* free segmap lock */
200 unsigned long *free_segmap; /* free segment bitmap */
201 unsigned long *free_secmap; /* free section bitmap */
204 /* Notice: The order of dirty type is same with CURSEG_XXX in f2fs.h */
206 DIRTY_HOT_DATA, /* dirty segments assigned as hot data logs */
207 DIRTY_WARM_DATA, /* dirty segments assigned as warm data logs */
208 DIRTY_COLD_DATA, /* dirty segments assigned as cold data logs */
209 DIRTY_HOT_NODE, /* dirty segments assigned as hot node logs */
210 DIRTY_WARM_NODE, /* dirty segments assigned as warm node logs */
211 DIRTY_COLD_NODE, /* dirty segments assigned as cold node logs */
212 DIRTY, /* to count # of dirty segments */
213 PRE, /* to count # of entirely obsolete segments */
217 struct dirty_seglist_info {
218 const struct victim_selection *v_ops; /* victim selction operation */
219 unsigned long *dirty_segmap[NR_DIRTY_TYPE];
220 struct mutex seglist_lock; /* lock for segment bitmaps */
221 int nr_dirty[NR_DIRTY_TYPE]; /* # of dirty segments */
222 unsigned long *victim_secmap; /* background GC victims */
225 /* victim selection function for cleaning and SSR */
226 struct victim_selection {
227 int (*get_victim)(struct f2fs_sb_info *, unsigned int *,
231 /* for active log information */
233 struct mutex curseg_mutex; /* lock for consistency */
234 struct f2fs_summary_block *sum_blk; /* cached summary block */
235 unsigned char alloc_type; /* current allocation type */
236 unsigned int segno; /* current segment number */
237 unsigned short next_blkoff; /* next block offset to write */
238 unsigned int zone; /* current zone number */
239 unsigned int next_segno; /* preallocated segment */
245 static inline struct curseg_info *CURSEG_I(struct f2fs_sb_info *sbi, int type)
247 return (struct curseg_info *)(SM_I(sbi)->curseg_array + type);
250 static inline struct seg_entry *get_seg_entry(struct f2fs_sb_info *sbi,
253 struct sit_info *sit_i = SIT_I(sbi);
254 return &sit_i->sentries[segno];
257 static inline struct sec_entry *get_sec_entry(struct f2fs_sb_info *sbi,
260 struct sit_info *sit_i = SIT_I(sbi);
261 return &sit_i->sec_entries[GET_SECNO(sbi, segno)];
264 static inline unsigned int get_valid_blocks(struct f2fs_sb_info *sbi,
265 unsigned int segno, int section)
268 * In order to get # of valid blocks in a section instantly from many
269 * segments, f2fs manages two counting structures separately.
272 return get_sec_entry(sbi, segno)->valid_blocks;
274 return get_seg_entry(sbi, segno)->valid_blocks;
277 static inline void seg_info_from_raw_sit(struct seg_entry *se,
278 struct f2fs_sit_entry *rs)
280 se->valid_blocks = GET_SIT_VBLOCKS(rs);
281 se->ckpt_valid_blocks = GET_SIT_VBLOCKS(rs);
282 memcpy(se->cur_valid_map, rs->valid_map, SIT_VBLOCK_MAP_SIZE);
283 memcpy(se->ckpt_valid_map, rs->valid_map, SIT_VBLOCK_MAP_SIZE);
284 se->type = GET_SIT_TYPE(rs);
285 se->mtime = le64_to_cpu(rs->mtime);
288 static inline void seg_info_to_raw_sit(struct seg_entry *se,
289 struct f2fs_sit_entry *rs)
291 unsigned short raw_vblocks = (se->type << SIT_VBLOCKS_SHIFT) |
293 rs->vblocks = cpu_to_le16(raw_vblocks);
294 memcpy(rs->valid_map, se->cur_valid_map, SIT_VBLOCK_MAP_SIZE);
295 memcpy(se->ckpt_valid_map, rs->valid_map, SIT_VBLOCK_MAP_SIZE);
296 se->ckpt_valid_blocks = se->valid_blocks;
297 rs->mtime = cpu_to_le64(se->mtime);
300 static inline unsigned int find_next_inuse(struct free_segmap_info *free_i,
301 unsigned int max, unsigned int segno)
304 read_lock(&free_i->segmap_lock);
305 ret = find_next_bit(free_i->free_segmap, max, segno);
306 read_unlock(&free_i->segmap_lock);
310 static inline void __set_free(struct f2fs_sb_info *sbi, unsigned int segno)
312 struct free_segmap_info *free_i = FREE_I(sbi);
313 unsigned int secno = segno / sbi->segs_per_sec;
314 unsigned int start_segno = secno * sbi->segs_per_sec;
317 write_lock(&free_i->segmap_lock);
318 clear_bit(segno, free_i->free_segmap);
319 free_i->free_segments++;
321 next = find_next_bit(free_i->free_segmap, TOTAL_SEGS(sbi), start_segno);
322 if (next >= start_segno + sbi->segs_per_sec) {
323 clear_bit(secno, free_i->free_secmap);
324 free_i->free_sections++;
326 write_unlock(&free_i->segmap_lock);
329 static inline void __set_inuse(struct f2fs_sb_info *sbi,
332 struct free_segmap_info *free_i = FREE_I(sbi);
333 unsigned int secno = segno / sbi->segs_per_sec;
334 set_bit(segno, free_i->free_segmap);
335 free_i->free_segments--;
336 if (!test_and_set_bit(secno, free_i->free_secmap))
337 free_i->free_sections--;
340 static inline void __set_test_and_free(struct f2fs_sb_info *sbi,
343 struct free_segmap_info *free_i = FREE_I(sbi);
344 unsigned int secno = segno / sbi->segs_per_sec;
345 unsigned int start_segno = secno * sbi->segs_per_sec;
348 write_lock(&free_i->segmap_lock);
349 if (test_and_clear_bit(segno, free_i->free_segmap)) {
350 free_i->free_segments++;
352 next = find_next_bit(free_i->free_segmap, TOTAL_SEGS(sbi),
354 if (next >= start_segno + sbi->segs_per_sec) {
355 if (test_and_clear_bit(secno, free_i->free_secmap))
356 free_i->free_sections++;
359 write_unlock(&free_i->segmap_lock);
362 static inline void __set_test_and_inuse(struct f2fs_sb_info *sbi,
365 struct free_segmap_info *free_i = FREE_I(sbi);
366 unsigned int secno = segno / sbi->segs_per_sec;
367 write_lock(&free_i->segmap_lock);
368 if (!test_and_set_bit(segno, free_i->free_segmap)) {
369 free_i->free_segments--;
370 if (!test_and_set_bit(secno, free_i->free_secmap))
371 free_i->free_sections--;
373 write_unlock(&free_i->segmap_lock);
376 static inline void get_sit_bitmap(struct f2fs_sb_info *sbi,
379 struct sit_info *sit_i = SIT_I(sbi);
380 memcpy(dst_addr, sit_i->sit_bitmap, sit_i->bitmap_size);
383 static inline block_t written_block_count(struct f2fs_sb_info *sbi)
385 struct sit_info *sit_i = SIT_I(sbi);
388 mutex_lock(&sit_i->sentry_lock);
389 vblocks = sit_i->written_valid_blocks;
390 mutex_unlock(&sit_i->sentry_lock);
395 static inline unsigned int free_segments(struct f2fs_sb_info *sbi)
397 struct free_segmap_info *free_i = FREE_I(sbi);
398 unsigned int free_segs;
400 read_lock(&free_i->segmap_lock);
401 free_segs = free_i->free_segments;
402 read_unlock(&free_i->segmap_lock);
407 static inline int reserved_segments(struct f2fs_sb_info *sbi)
409 return SM_I(sbi)->reserved_segments;
412 static inline unsigned int free_sections(struct f2fs_sb_info *sbi)
414 struct free_segmap_info *free_i = FREE_I(sbi);
415 unsigned int free_secs;
417 read_lock(&free_i->segmap_lock);
418 free_secs = free_i->free_sections;
419 read_unlock(&free_i->segmap_lock);
424 static inline unsigned int prefree_segments(struct f2fs_sb_info *sbi)
426 return DIRTY_I(sbi)->nr_dirty[PRE];
429 static inline unsigned int dirty_segments(struct f2fs_sb_info *sbi)
431 return DIRTY_I(sbi)->nr_dirty[DIRTY_HOT_DATA] +
432 DIRTY_I(sbi)->nr_dirty[DIRTY_WARM_DATA] +
433 DIRTY_I(sbi)->nr_dirty[DIRTY_COLD_DATA] +
434 DIRTY_I(sbi)->nr_dirty[DIRTY_HOT_NODE] +
435 DIRTY_I(sbi)->nr_dirty[DIRTY_WARM_NODE] +
436 DIRTY_I(sbi)->nr_dirty[DIRTY_COLD_NODE];
439 static inline int overprovision_segments(struct f2fs_sb_info *sbi)
441 return SM_I(sbi)->ovp_segments;
444 static inline int overprovision_sections(struct f2fs_sb_info *sbi)
446 return ((unsigned int) overprovision_segments(sbi)) / sbi->segs_per_sec;
449 static inline int reserved_sections(struct f2fs_sb_info *sbi)
451 return ((unsigned int) reserved_segments(sbi)) / sbi->segs_per_sec;
454 static inline bool need_SSR(struct f2fs_sb_info *sbi)
456 return ((prefree_segments(sbi) / sbi->segs_per_sec)
457 + free_sections(sbi) < overprovision_sections(sbi));
460 static inline bool has_not_enough_free_secs(struct f2fs_sb_info *sbi, int freed)
462 int node_secs = get_blocktype_secs(sbi, F2FS_DIRTY_NODES);
463 int dent_secs = get_blocktype_secs(sbi, F2FS_DIRTY_DENTS);
468 return ((free_sections(sbi) + freed) <= (node_secs + 2 * dent_secs +
469 reserved_sections(sbi)));
472 static inline bool excess_prefree_segs(struct f2fs_sb_info *sbi)
474 return (prefree_segments(sbi) > SM_I(sbi)->rec_prefree_segments);
477 static inline int utilization(struct f2fs_sb_info *sbi)
479 return div_u64((u64)valid_user_blocks(sbi) * 100, sbi->user_block_count);
483 * Sometimes f2fs may be better to drop out-of-place update policy.
484 * So, if fs utilization is over MIN_IPU_UTIL, then f2fs tries to write
485 * data in the original place likewise other traditional file systems.
486 * But, currently set 100 in percentage, which means it is disabled.
487 * See below need_inplace_update().
489 #define MIN_IPU_UTIL 100
490 static inline bool need_inplace_update(struct inode *inode)
492 struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
493 if (S_ISDIR(inode->i_mode))
495 if (need_SSR(sbi) && utilization(sbi) > MIN_IPU_UTIL)
500 static inline unsigned int curseg_segno(struct f2fs_sb_info *sbi,
503 struct curseg_info *curseg = CURSEG_I(sbi, type);
504 return curseg->segno;
507 static inline unsigned char curseg_alloc_type(struct f2fs_sb_info *sbi,
510 struct curseg_info *curseg = CURSEG_I(sbi, type);
511 return curseg->alloc_type;
514 static inline unsigned short curseg_blkoff(struct f2fs_sb_info *sbi, int type)
516 struct curseg_info *curseg = CURSEG_I(sbi, type);
517 return curseg->next_blkoff;
520 #ifdef CONFIG_F2FS_CHECK_FS
521 static inline void check_seg_range(struct f2fs_sb_info *sbi, unsigned int segno)
523 unsigned int end_segno = SM_I(sbi)->segment_count - 1;
524 BUG_ON(segno > end_segno);
527 static inline void verify_block_addr(struct f2fs_sb_info *sbi, block_t blk_addr)
529 struct f2fs_sm_info *sm_info = SM_I(sbi);
530 block_t total_blks = sm_info->segment_count << sbi->log_blocks_per_seg;
531 block_t start_addr = sm_info->seg0_blkaddr;
532 block_t end_addr = start_addr + total_blks - 1;
533 BUG_ON(blk_addr < start_addr);
534 BUG_ON(blk_addr > end_addr);
538 * Summary block is always treated as invalid block
540 static inline void check_block_count(struct f2fs_sb_info *sbi,
541 int segno, struct f2fs_sit_entry *raw_sit)
543 struct f2fs_sm_info *sm_info = SM_I(sbi);
544 unsigned int end_segno = sm_info->segment_count - 1;
545 bool is_valid = test_bit_le(0, raw_sit->valid_map) ? true : false;
546 int valid_blocks = 0;
547 int cur_pos = 0, next_pos;
549 /* check segment usage */
550 BUG_ON(GET_SIT_VBLOCKS(raw_sit) > sbi->blocks_per_seg);
552 /* check boundary of a given segment number */
553 BUG_ON(segno > end_segno);
555 /* check bitmap with valid block count */
558 next_pos = find_next_zero_bit_le(&raw_sit->valid_map,
561 valid_blocks += next_pos - cur_pos;
563 next_pos = find_next_bit_le(&raw_sit->valid_map,
567 is_valid = !is_valid;
568 } while (cur_pos < sbi->blocks_per_seg);
569 BUG_ON(GET_SIT_VBLOCKS(raw_sit) != valid_blocks);
572 #define check_seg_range(sbi, segno)
573 #define verify_block_addr(sbi, blk_addr)
574 #define check_block_count(sbi, segno, raw_sit)
577 static inline pgoff_t current_sit_addr(struct f2fs_sb_info *sbi,
580 struct sit_info *sit_i = SIT_I(sbi);
581 unsigned int offset = SIT_BLOCK_OFFSET(sit_i, start);
582 block_t blk_addr = sit_i->sit_base_addr + offset;
584 check_seg_range(sbi, start);
586 /* calculate sit block address */
587 if (f2fs_test_bit(offset, sit_i->sit_bitmap))
588 blk_addr += sit_i->sit_blocks;
593 static inline pgoff_t next_sit_addr(struct f2fs_sb_info *sbi,
596 struct sit_info *sit_i = SIT_I(sbi);
597 block_addr -= sit_i->sit_base_addr;
598 if (block_addr < sit_i->sit_blocks)
599 block_addr += sit_i->sit_blocks;
601 block_addr -= sit_i->sit_blocks;
603 return block_addr + sit_i->sit_base_addr;
606 static inline void set_to_next_sit(struct sit_info *sit_i, unsigned int start)
608 unsigned int block_off = SIT_BLOCK_OFFSET(sit_i, start);
610 if (f2fs_test_bit(block_off, sit_i->sit_bitmap))
611 f2fs_clear_bit(block_off, sit_i->sit_bitmap);
613 f2fs_set_bit(block_off, sit_i->sit_bitmap);
616 static inline unsigned long long get_mtime(struct f2fs_sb_info *sbi)
618 struct sit_info *sit_i = SIT_I(sbi);
619 return sit_i->elapsed_time + CURRENT_TIME_SEC.tv_sec -
623 static inline void set_summary(struct f2fs_summary *sum, nid_t nid,
624 unsigned int ofs_in_node, unsigned char version)
626 sum->nid = cpu_to_le32(nid);
627 sum->ofs_in_node = cpu_to_le16(ofs_in_node);
628 sum->version = version;
631 static inline block_t start_sum_block(struct f2fs_sb_info *sbi)
633 return __start_cp_addr(sbi) +
634 le32_to_cpu(F2FS_CKPT(sbi)->cp_pack_start_sum);
637 static inline block_t sum_blk_addr(struct f2fs_sb_info *sbi, int base, int type)
639 return __start_cp_addr(sbi) +
640 le32_to_cpu(F2FS_CKPT(sbi)->cp_pack_total_block_count)
644 static inline bool sec_usage_check(struct f2fs_sb_info *sbi, unsigned int secno)
646 if (IS_CURSEC(sbi, secno) || (sbi->cur_victim_sec == secno))
651 static inline unsigned int max_hw_blocks(struct f2fs_sb_info *sbi)
653 struct block_device *bdev = sbi->sb->s_bdev;
654 struct request_queue *q = bdev_get_queue(bdev);
655 return SECTOR_TO_BLOCK(sbi, queue_max_sectors(q));