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.
12 #define NULL_SEGNO ((unsigned int)(~0))
14 /* V: Logical segment # in volume, R: Relative segment # in main area */
15 #define GET_L2R_SEGNO(free_i, segno) (segno - free_i->start_segno)
16 #define GET_R2L_SEGNO(free_i, segno) (segno + free_i->start_segno)
18 #define IS_DATASEG(t) \
19 ((t == CURSEG_HOT_DATA) || (t == CURSEG_COLD_DATA) || \
20 (t == CURSEG_WARM_DATA))
22 #define IS_NODESEG(t) \
23 ((t == CURSEG_HOT_NODE) || (t == CURSEG_COLD_NODE) || \
24 (t == CURSEG_WARM_NODE))
26 #define IS_CURSEG(sbi, segno) \
27 ((segno == CURSEG_I(sbi, CURSEG_HOT_DATA)->segno) || \
28 (segno == CURSEG_I(sbi, CURSEG_WARM_DATA)->segno) || \
29 (segno == CURSEG_I(sbi, CURSEG_COLD_DATA)->segno) || \
30 (segno == CURSEG_I(sbi, CURSEG_HOT_NODE)->segno) || \
31 (segno == CURSEG_I(sbi, CURSEG_WARM_NODE)->segno) || \
32 (segno == 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)
85 #define SECTOR_FROM_BLOCK(sbi, blk_addr) \
86 (blk_addr << ((sbi)->log_blocksize - F2FS_LOG_SECTOR_SIZE))
88 /* during checkpoint, bio_private is used to synchronize the last bio */
90 struct f2fs_sb_info *sbi;
96 * indicate a block allocation direction: RIGHT and LEFT.
97 * RIGHT means allocating new sections towards the end of volume.
98 * LEFT means the opposite direction.
106 * In the victim_sel_policy->alloc_mode, there are two block allocation modes.
107 * LFS writes data sequentially with cleaning operations.
108 * SSR (Slack Space Recycle) reuses obsolete space without cleaning operations.
116 * In the victim_sel_policy->gc_mode, there are two gc, aka cleaning, modes.
117 * GC_CB is based on cost-benefit algorithm.
118 * GC_GREEDY is based on greedy algorithm.
126 * BG_GC means the background cleaning job.
127 * FG_GC means the on-demand cleaning job.
134 /* for a function parameter to select a victim segment */
135 struct victim_sel_policy {
136 int alloc_mode; /* LFS or SSR */
137 int gc_mode; /* GC_CB or GC_GREEDY */
138 unsigned long *dirty_segmap; /* dirty segment bitmap */
139 unsigned int offset; /* last scanned bitmap offset */
140 unsigned int ofs_unit; /* bitmap search unit */
141 unsigned int min_cost; /* minimum cost */
142 unsigned int min_segno; /* segment # having min. cost */
146 unsigned short valid_blocks; /* # of valid blocks */
147 unsigned char *cur_valid_map; /* validity bitmap of blocks */
149 * # of valid blocks and the validity bitmap stored in the the last
150 * checkpoint pack. This information is used by the SSR mode.
152 unsigned short ckpt_valid_blocks;
153 unsigned char *ckpt_valid_map;
154 unsigned char type; /* segment type like CURSEG_XXX_TYPE */
155 unsigned long long mtime; /* modification time of the segment */
159 unsigned int valid_blocks; /* # of valid blocks in a section */
162 struct segment_allocation {
163 void (*allocate_segment)(struct f2fs_sb_info *, int, bool);
167 const struct segment_allocation *s_ops;
169 block_t sit_base_addr; /* start block address of SIT area */
170 block_t sit_blocks; /* # of blocks used by SIT area */
171 block_t written_valid_blocks; /* # of valid blocks in main area */
172 char *sit_bitmap; /* SIT bitmap pointer */
173 unsigned int bitmap_size; /* SIT bitmap size */
175 unsigned long *dirty_sentries_bitmap; /* bitmap for dirty sentries */
176 unsigned int dirty_sentries; /* # of dirty sentries */
177 unsigned int sents_per_block; /* # of SIT entries per block */
178 struct mutex sentry_lock; /* to protect SIT cache */
179 struct seg_entry *sentries; /* SIT segment-level cache */
180 struct sec_entry *sec_entries; /* SIT section-level cache */
182 /* for cost-benefit algorithm in cleaning procedure */
183 unsigned long long elapsed_time; /* elapsed time after mount */
184 unsigned long long mounted_time; /* mount time */
185 unsigned long long min_mtime; /* min. modification time */
186 unsigned long long max_mtime; /* max. modification time */
189 struct free_segmap_info {
190 unsigned int start_segno; /* start segment number logically */
191 unsigned int free_segments; /* # of free segments */
192 unsigned int free_sections; /* # of free sections */
193 rwlock_t segmap_lock; /* free segmap lock */
194 unsigned long *free_segmap; /* free segment bitmap */
195 unsigned long *free_secmap; /* free section bitmap */
198 /* Notice: The order of dirty type is same with CURSEG_XXX in f2fs.h */
200 DIRTY_HOT_DATA, /* dirty segments assigned as hot data logs */
201 DIRTY_WARM_DATA, /* dirty segments assigned as warm data logs */
202 DIRTY_COLD_DATA, /* dirty segments assigned as cold data logs */
203 DIRTY_HOT_NODE, /* dirty segments assigned as hot node logs */
204 DIRTY_WARM_NODE, /* dirty segments assigned as warm node logs */
205 DIRTY_COLD_NODE, /* dirty segments assigned as cold node logs */
206 DIRTY, /* to count # of dirty segments */
207 PRE, /* to count # of entirely obsolete segments */
211 struct dirty_seglist_info {
212 const struct victim_selection *v_ops; /* victim selction operation */
213 unsigned long *dirty_segmap[NR_DIRTY_TYPE];
214 struct mutex seglist_lock; /* lock for segment bitmaps */
215 int nr_dirty[NR_DIRTY_TYPE]; /* # of dirty segments */
216 unsigned long *victim_segmap[2]; /* BG_GC, FG_GC */
219 /* victim selection function for cleaning and SSR */
220 struct victim_selection {
221 int (*get_victim)(struct f2fs_sb_info *, unsigned int *,
225 /* for active log information */
227 struct mutex curseg_mutex; /* lock for consistency */
228 struct f2fs_summary_block *sum_blk; /* cached summary block */
229 unsigned char alloc_type; /* current allocation type */
230 unsigned int segno; /* current segment number */
231 unsigned short next_blkoff; /* next block offset to write */
232 unsigned int zone; /* current zone number */
233 unsigned int next_segno; /* preallocated segment */
239 static inline struct curseg_info *CURSEG_I(struct f2fs_sb_info *sbi, int type)
241 return (struct curseg_info *)(SM_I(sbi)->curseg_array + type);
244 static inline struct seg_entry *get_seg_entry(struct f2fs_sb_info *sbi,
247 struct sit_info *sit_i = SIT_I(sbi);
248 return &sit_i->sentries[segno];
251 static inline struct sec_entry *get_sec_entry(struct f2fs_sb_info *sbi,
254 struct sit_info *sit_i = SIT_I(sbi);
255 return &sit_i->sec_entries[GET_SECNO(sbi, segno)];
258 static inline unsigned int get_valid_blocks(struct f2fs_sb_info *sbi,
259 unsigned int segno, int section)
262 * In order to get # of valid blocks in a section instantly from many
263 * segments, f2fs manages two counting structures separately.
266 return get_sec_entry(sbi, segno)->valid_blocks;
268 return get_seg_entry(sbi, segno)->valid_blocks;
271 static inline void seg_info_from_raw_sit(struct seg_entry *se,
272 struct f2fs_sit_entry *rs)
274 se->valid_blocks = GET_SIT_VBLOCKS(rs);
275 se->ckpt_valid_blocks = GET_SIT_VBLOCKS(rs);
276 memcpy(se->cur_valid_map, rs->valid_map, SIT_VBLOCK_MAP_SIZE);
277 memcpy(se->ckpt_valid_map, rs->valid_map, SIT_VBLOCK_MAP_SIZE);
278 se->type = GET_SIT_TYPE(rs);
279 se->mtime = le64_to_cpu(rs->mtime);
282 static inline void seg_info_to_raw_sit(struct seg_entry *se,
283 struct f2fs_sit_entry *rs)
285 unsigned short raw_vblocks = (se->type << SIT_VBLOCKS_SHIFT) |
287 rs->vblocks = cpu_to_le16(raw_vblocks);
288 memcpy(rs->valid_map, se->cur_valid_map, SIT_VBLOCK_MAP_SIZE);
289 memcpy(se->ckpt_valid_map, rs->valid_map, SIT_VBLOCK_MAP_SIZE);
290 se->ckpt_valid_blocks = se->valid_blocks;
291 rs->mtime = cpu_to_le64(se->mtime);
294 static inline unsigned int find_next_inuse(struct free_segmap_info *free_i,
295 unsigned int max, unsigned int segno)
298 read_lock(&free_i->segmap_lock);
299 ret = find_next_bit(free_i->free_segmap, max, segno);
300 read_unlock(&free_i->segmap_lock);
304 static inline void __set_free(struct f2fs_sb_info *sbi, unsigned int segno)
306 struct free_segmap_info *free_i = FREE_I(sbi);
307 unsigned int secno = segno / sbi->segs_per_sec;
308 unsigned int start_segno = secno * sbi->segs_per_sec;
311 write_lock(&free_i->segmap_lock);
312 clear_bit(segno, free_i->free_segmap);
313 free_i->free_segments++;
315 next = find_next_bit(free_i->free_segmap, TOTAL_SEGS(sbi), start_segno);
316 if (next >= start_segno + sbi->segs_per_sec) {
317 clear_bit(secno, free_i->free_secmap);
318 free_i->free_sections++;
320 write_unlock(&free_i->segmap_lock);
323 static inline void __set_inuse(struct f2fs_sb_info *sbi,
326 struct free_segmap_info *free_i = FREE_I(sbi);
327 unsigned int secno = segno / sbi->segs_per_sec;
328 set_bit(segno, free_i->free_segmap);
329 free_i->free_segments--;
330 if (!test_and_set_bit(secno, free_i->free_secmap))
331 free_i->free_sections--;
334 static inline void __set_test_and_free(struct f2fs_sb_info *sbi,
337 struct free_segmap_info *free_i = FREE_I(sbi);
338 unsigned int secno = segno / sbi->segs_per_sec;
339 unsigned int start_segno = secno * sbi->segs_per_sec;
342 write_lock(&free_i->segmap_lock);
343 if (test_and_clear_bit(segno, free_i->free_segmap)) {
344 free_i->free_segments++;
346 next = find_next_bit(free_i->free_segmap, TOTAL_SEGS(sbi),
348 if (next >= start_segno + sbi->segs_per_sec) {
349 if (test_and_clear_bit(secno, free_i->free_secmap))
350 free_i->free_sections++;
353 write_unlock(&free_i->segmap_lock);
356 static inline void __set_test_and_inuse(struct f2fs_sb_info *sbi,
359 struct free_segmap_info *free_i = FREE_I(sbi);
360 unsigned int secno = segno / sbi->segs_per_sec;
361 write_lock(&free_i->segmap_lock);
362 if (!test_and_set_bit(segno, free_i->free_segmap)) {
363 free_i->free_segments--;
364 if (!test_and_set_bit(secno, free_i->free_secmap))
365 free_i->free_sections--;
367 write_unlock(&free_i->segmap_lock);
370 static inline void get_sit_bitmap(struct f2fs_sb_info *sbi,
373 struct sit_info *sit_i = SIT_I(sbi);
374 memcpy(dst_addr, sit_i->sit_bitmap, sit_i->bitmap_size);
377 static inline block_t written_block_count(struct f2fs_sb_info *sbi)
379 struct sit_info *sit_i = SIT_I(sbi);
382 mutex_lock(&sit_i->sentry_lock);
383 vblocks = sit_i->written_valid_blocks;
384 mutex_unlock(&sit_i->sentry_lock);
389 static inline unsigned int free_segments(struct f2fs_sb_info *sbi)
391 struct free_segmap_info *free_i = FREE_I(sbi);
392 unsigned int free_segs;
394 read_lock(&free_i->segmap_lock);
395 free_segs = free_i->free_segments;
396 read_unlock(&free_i->segmap_lock);
401 static inline int reserved_segments(struct f2fs_sb_info *sbi)
403 return SM_I(sbi)->reserved_segments;
406 static inline unsigned int free_sections(struct f2fs_sb_info *sbi)
408 struct free_segmap_info *free_i = FREE_I(sbi);
409 unsigned int free_secs;
411 read_lock(&free_i->segmap_lock);
412 free_secs = free_i->free_sections;
413 read_unlock(&free_i->segmap_lock);
418 static inline unsigned int prefree_segments(struct f2fs_sb_info *sbi)
420 return DIRTY_I(sbi)->nr_dirty[PRE];
423 static inline unsigned int dirty_segments(struct f2fs_sb_info *sbi)
425 return DIRTY_I(sbi)->nr_dirty[DIRTY_HOT_DATA] +
426 DIRTY_I(sbi)->nr_dirty[DIRTY_WARM_DATA] +
427 DIRTY_I(sbi)->nr_dirty[DIRTY_COLD_DATA] +
428 DIRTY_I(sbi)->nr_dirty[DIRTY_HOT_NODE] +
429 DIRTY_I(sbi)->nr_dirty[DIRTY_WARM_NODE] +
430 DIRTY_I(sbi)->nr_dirty[DIRTY_COLD_NODE];
433 static inline int overprovision_segments(struct f2fs_sb_info *sbi)
435 return SM_I(sbi)->ovp_segments;
438 static inline int overprovision_sections(struct f2fs_sb_info *sbi)
440 return ((unsigned int) overprovision_segments(sbi)) / sbi->segs_per_sec;
443 static inline int reserved_sections(struct f2fs_sb_info *sbi)
445 return ((unsigned int) reserved_segments(sbi)) / sbi->segs_per_sec;
448 static inline bool need_SSR(struct f2fs_sb_info *sbi)
450 return (free_sections(sbi) < overprovision_sections(sbi));
453 static inline int get_ssr_segment(struct f2fs_sb_info *sbi, int type)
455 struct curseg_info *curseg = CURSEG_I(sbi, type);
456 return DIRTY_I(sbi)->v_ops->get_victim(sbi,
457 &(curseg)->next_segno, BG_GC, type, SSR);
460 static inline bool has_not_enough_free_secs(struct f2fs_sb_info *sbi)
462 unsigned int pages_per_sec = (1 << sbi->log_blocks_per_seg) *
464 int node_secs = ((get_pages(sbi, F2FS_DIRTY_NODES) + pages_per_sec - 1)
465 >> sbi->log_blocks_per_seg) / sbi->segs_per_sec;
466 int dent_secs = ((get_pages(sbi, F2FS_DIRTY_DENTS) + pages_per_sec - 1)
467 >> sbi->log_blocks_per_seg) / sbi->segs_per_sec;
472 if (free_sections(sbi) <= (node_secs + 2 * dent_secs +
473 reserved_sections(sbi)))
478 static inline int utilization(struct f2fs_sb_info *sbi)
480 return (long int)valid_user_blocks(sbi) * 100 /
481 (long int)sbi->user_block_count;
485 * Sometimes f2fs may be better to drop out-of-place update policy.
486 * So, if fs utilization is over MIN_IPU_UTIL, then f2fs tries to write
487 * data in the original place likewise other traditional file systems.
488 * But, currently set 100 in percentage, which means it is disabled.
489 * See below need_inplace_update().
491 #define MIN_IPU_UTIL 100
492 static inline bool need_inplace_update(struct inode *inode)
494 struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
495 if (S_ISDIR(inode->i_mode))
497 if (need_SSR(sbi) && utilization(sbi) > MIN_IPU_UTIL)
502 static inline unsigned int curseg_segno(struct f2fs_sb_info *sbi,
505 struct curseg_info *curseg = CURSEG_I(sbi, type);
506 return curseg->segno;
509 static inline unsigned char curseg_alloc_type(struct f2fs_sb_info *sbi,
512 struct curseg_info *curseg = CURSEG_I(sbi, type);
513 return curseg->alloc_type;
516 static inline unsigned short curseg_blkoff(struct f2fs_sb_info *sbi, int type)
518 struct curseg_info *curseg = CURSEG_I(sbi, type);
519 return curseg->next_blkoff;
522 static inline void check_seg_range(struct f2fs_sb_info *sbi, unsigned int segno)
524 unsigned int end_segno = SM_I(sbi)->segment_count - 1;
525 BUG_ON(segno > end_segno);
529 * This function is used for only debugging.
530 * NOTE: In future, we have to remove this function.
532 static inline void verify_block_addr(struct f2fs_sb_info *sbi, block_t blk_addr)
534 struct f2fs_sm_info *sm_info = SM_I(sbi);
535 block_t total_blks = sm_info->segment_count << sbi->log_blocks_per_seg;
536 block_t start_addr = sm_info->seg0_blkaddr;
537 block_t end_addr = start_addr + total_blks - 1;
538 BUG_ON(blk_addr < start_addr);
539 BUG_ON(blk_addr > end_addr);
543 * Summary block is always treated as invalid block
545 static inline void check_block_count(struct f2fs_sb_info *sbi,
546 int segno, struct f2fs_sit_entry *raw_sit)
548 struct f2fs_sm_info *sm_info = SM_I(sbi);
549 unsigned int end_segno = sm_info->segment_count - 1;
550 int valid_blocks = 0;
553 /* check segment usage */
554 BUG_ON(GET_SIT_VBLOCKS(raw_sit) > sbi->blocks_per_seg);
556 /* check boundary of a given segment number */
557 BUG_ON(segno > end_segno);
559 /* check bitmap with valid block count */
560 for (i = 0; i < sbi->blocks_per_seg; i++)
561 if (f2fs_test_bit(i, raw_sit->valid_map))
563 BUG_ON(GET_SIT_VBLOCKS(raw_sit) != valid_blocks);
566 static inline pgoff_t current_sit_addr(struct f2fs_sb_info *sbi,
569 struct sit_info *sit_i = SIT_I(sbi);
570 unsigned int offset = SIT_BLOCK_OFFSET(sit_i, start);
571 block_t blk_addr = sit_i->sit_base_addr + offset;
573 check_seg_range(sbi, start);
575 /* calculate sit block address */
576 if (f2fs_test_bit(offset, sit_i->sit_bitmap))
577 blk_addr += sit_i->sit_blocks;
582 static inline pgoff_t next_sit_addr(struct f2fs_sb_info *sbi,
585 struct sit_info *sit_i = SIT_I(sbi);
586 block_addr -= sit_i->sit_base_addr;
587 if (block_addr < sit_i->sit_blocks)
588 block_addr += sit_i->sit_blocks;
590 block_addr -= sit_i->sit_blocks;
592 return block_addr + sit_i->sit_base_addr;
595 static inline void set_to_next_sit(struct sit_info *sit_i, unsigned int start)
597 unsigned int block_off = SIT_BLOCK_OFFSET(sit_i, start);
599 if (f2fs_test_bit(block_off, sit_i->sit_bitmap))
600 f2fs_clear_bit(block_off, sit_i->sit_bitmap);
602 f2fs_set_bit(block_off, sit_i->sit_bitmap);
605 static inline unsigned long long get_mtime(struct f2fs_sb_info *sbi)
607 struct sit_info *sit_i = SIT_I(sbi);
608 return sit_i->elapsed_time + CURRENT_TIME_SEC.tv_sec -
612 static inline void set_summary(struct f2fs_summary *sum, nid_t nid,
613 unsigned int ofs_in_node, unsigned char version)
615 sum->nid = cpu_to_le32(nid);
616 sum->ofs_in_node = cpu_to_le16(ofs_in_node);
617 sum->version = version;
620 static inline block_t start_sum_block(struct f2fs_sb_info *sbi)
622 return __start_cp_addr(sbi) +
623 le32_to_cpu(F2FS_CKPT(sbi)->cp_pack_start_sum);
626 static inline block_t sum_blk_addr(struct f2fs_sb_info *sbi, int base, int type)
628 return __start_cp_addr(sbi) +
629 le32_to_cpu(F2FS_CKPT(sbi)->cp_pack_total_block_count)