Loading...
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
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)
23
24#define IS_DATASEG(t) ((t) <= CURSEG_COLD_DATA)
25#define IS_NODESEG(t) ((t) >= CURSEG_HOT_NODE)
26
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)
30
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))
38
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)) \
52
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))
59
60#define MAIN_SEGS(sbi) (SM_I(sbi)->main_segments)
61#define MAIN_SECS(sbi) ((sbi)->total_sections)
62
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)
67
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))
71
72#define START_BLOCK(sbi, segno) (SEG0_BLKADDR(sbi) + \
73 (GET_R2L_SEGNO(FREE_I(sbi), segno) << (sbi)->log_blocks_per_seg))
74
75#define NEXT_FREE_BLKADDR(sbi, curseg) \
76 (START_BLOCK(sbi, (curseg)->segno) + (curseg)->next_blkoff)
77
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))
83
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))
98
99#define GET_SUM_BLOCK(sbi, segno) \
100 ((sbi)->sm_info->ssa_blkaddr + (segno))
101
102#define GET_SUM_TYPE(footer) ((footer)->entry_type)
103#define SET_SUM_TYPE(footer, type) ((footer)->entry_type = (type))
104
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))
115
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)
120
121/*
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.
125 */
126enum {
127 ALLOC_RIGHT = 0,
128 ALLOC_LEFT
129};
130
131/*
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.
135 */
136enum {
137 LFS = 0,
138 SSR
139};
140
141/*
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.
145 */
146enum {
147 GC_CB = 0,
148 GC_GREEDY,
149 ALLOC_NEXT,
150 FLUSH_DEVICE,
151 MAX_GC_POLICY,
152};
153
154/*
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.
158 */
159enum {
160 BG_GC = 0,
161 FG_GC,
162 FORCE_FG_GC,
163};
164
165/* for a function parameter to select a victim segment */
166struct 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
172 * to search
173 */
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 */
178};
179
180struct seg_entry {
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 */
188#endif
189 /*
190 * # of valid blocks and the validity bitmap stored in the last
191 * checkpoint pack. This information is used by the SSR mode.
192 */
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 */
196};
197
198struct sec_entry {
199 unsigned int valid_blocks; /* # of valid blocks in a section */
200};
201
202struct segment_allocation {
203 void (*allocate_segment)(struct f2fs_sb_info *, int, bool);
204};
205
206#define MAX_SKIP_GC_COUNT 16
207
208struct inmem_pages {
209 struct list_head list;
210 struct page *page;
211 block_t old_addr; /* for revoking when fail to commit */
212};
213
214struct sit_info {
215 const struct segment_allocation *s_ops;
216
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 */
224
225 /* bitmap of segments to be ignored by GC in case of errors */
226 unsigned long *invalid_segmap;
227#endif
228 unsigned int bitmap_size; /* SIT bitmap size */
229
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 */
237
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 */
243
244 unsigned int last_victim[MAX_GC_POLICY]; /* last victim segment # */
245};
246
247struct 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 */
254};
255
256/* Notice: The order of dirty type is same with CURSEG_XXX in f2fs.h */
257enum dirty_type {
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 */
266 NR_DIRTY_TYPE
267};
268
269struct 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 */
276};
277
278/* victim selection function for cleaning and SSR */
279struct victim_selection {
280 int (*get_victim)(struct f2fs_sb_info *, unsigned int *,
281 int, int, char);
282};
283
284/* for active log information */
285struct curseg_info {
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 */
295};
296
297struct 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 */
301};
302
303/*
304 * inline functions
305 */
306static inline struct curseg_info *CURSEG_I(struct f2fs_sb_info *sbi, int type)
307{
308 if (type == CURSEG_COLD_DATA_PINNED)
309 type = CURSEG_COLD_DATA;
310 return (struct curseg_info *)(SM_I(sbi)->curseg_array + type);
311}
312
313static inline struct seg_entry *get_seg_entry(struct f2fs_sb_info *sbi,
314 unsigned int segno)
315{
316 struct sit_info *sit_i = SIT_I(sbi);
317 return &sit_i->sentries[segno];
318}
319
320static inline struct sec_entry *get_sec_entry(struct f2fs_sb_info *sbi,
321 unsigned int segno)
322{
323 struct sit_info *sit_i = SIT_I(sbi);
324 return &sit_i->sec_entries[GET_SEC_FROM_SEG(sbi, segno)];
325}
326
327static inline unsigned int get_valid_blocks(struct f2fs_sb_info *sbi,
328 unsigned int segno, bool use_section)
329{
330 /*
331 * In order to get # of valid blocks in a section instantly from many
332 * segments, f2fs manages two counting structures separately.
333 */
334 if (use_section && __is_large_section(sbi))
335 return get_sec_entry(sbi, segno)->valid_blocks;
336 else
337 return get_seg_entry(sbi, segno)->valid_blocks;
338}
339
340static inline unsigned int get_ckpt_valid_blocks(struct f2fs_sb_info *sbi,
341 unsigned int segno)
342{
343 return get_seg_entry(sbi, segno)->ckpt_valid_blocks;
344}
345
346static inline void seg_info_from_raw_sit(struct seg_entry *se,
347 struct f2fs_sit_entry *rs)
348{
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);
355#endif
356 se->type = GET_SIT_TYPE(rs);
357 se->mtime = le64_to_cpu(rs->mtime);
358}
359
360static inline void __seg_info_to_raw_sit(struct seg_entry *se,
361 struct f2fs_sit_entry *rs)
362{
363 unsigned short raw_vblocks = (se->type << SIT_VBLOCKS_SHIFT) |
364 se->valid_blocks;
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);
368}
369
370static inline void seg_info_to_sit_page(struct f2fs_sb_info *sbi,
371 struct page *page, unsigned int start)
372{
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));
378 int i;
379
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);
386 }
387}
388
389static inline void seg_info_to_raw_sit(struct seg_entry *se,
390 struct f2fs_sit_entry *rs)
391{
392 __seg_info_to_raw_sit(se, rs);
393
394 memcpy(se->ckpt_valid_map, rs->valid_map, SIT_VBLOCK_MAP_SIZE);
395 se->ckpt_valid_blocks = se->valid_blocks;
396}
397
398static inline unsigned int find_next_inuse(struct free_segmap_info *free_i,
399 unsigned int max, unsigned int segno)
400{
401 unsigned int ret;
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);
405 return ret;
406}
407
408static inline void __set_free(struct f2fs_sb_info *sbi, unsigned int segno)
409{
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);
413 unsigned int next;
414
415 spin_lock(&free_i->segmap_lock);
416 clear_bit(segno, free_i->free_segmap);
417 free_i->free_segments++;
418
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++;
424 }
425 spin_unlock(&free_i->segmap_lock);
426}
427
428static inline void __set_inuse(struct f2fs_sb_info *sbi,
429 unsigned int segno)
430{
431 struct free_segmap_info *free_i = FREE_I(sbi);
432 unsigned int secno = GET_SEC_FROM_SEG(sbi, segno);
433
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--;
438}
439
440static inline void __set_test_and_free(struct f2fs_sb_info *sbi,
441 unsigned int segno)
442{
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);
446 unsigned int next;
447
448 spin_lock(&free_i->segmap_lock);
449 if (test_and_clear_bit(segno, free_i->free_segmap)) {
450 free_i->free_segments++;
451
452 if (IS_CURSEC(sbi, secno))
453 goto skip_free;
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++;
459 }
460 }
461skip_free:
462 spin_unlock(&free_i->segmap_lock);
463}
464
465static inline void __set_test_and_inuse(struct f2fs_sb_info *sbi,
466 unsigned int segno)
467{
468 struct free_segmap_info *free_i = FREE_I(sbi);
469 unsigned int secno = GET_SEC_FROM_SEG(sbi, segno);
470
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--;
476 }
477 spin_unlock(&free_i->segmap_lock);
478}
479
480static inline void get_sit_bitmap(struct f2fs_sb_info *sbi,
481 void *dst_addr)
482{
483 struct sit_info *sit_i = SIT_I(sbi);
484
485#ifdef CONFIG_F2FS_CHECK_FS
486 if (memcmp(sit_i->sit_bitmap, sit_i->sit_bitmap_mir,
487 sit_i->bitmap_size))
488 f2fs_bug_on(sbi, 1);
489#endif
490 memcpy(dst_addr, sit_i->sit_bitmap, sit_i->bitmap_size);
491}
492
493static inline block_t written_block_count(struct f2fs_sb_info *sbi)
494{
495 return SIT_I(sbi)->written_valid_blocks;
496}
497
498static inline unsigned int free_segments(struct f2fs_sb_info *sbi)
499{
500 return FREE_I(sbi)->free_segments;
501}
502
503static inline int reserved_segments(struct f2fs_sb_info *sbi)
504{
505 return SM_I(sbi)->reserved_segments;
506}
507
508static inline unsigned int free_sections(struct f2fs_sb_info *sbi)
509{
510 return FREE_I(sbi)->free_sections;
511}
512
513static inline unsigned int prefree_segments(struct f2fs_sb_info *sbi)
514{
515 return DIRTY_I(sbi)->nr_dirty[PRE];
516}
517
518static inline unsigned int dirty_segments(struct f2fs_sb_info *sbi)
519{
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];
526}
527
528static inline int overprovision_segments(struct f2fs_sb_info *sbi)
529{
530 return SM_I(sbi)->ovp_segments;
531}
532
533static inline int reserved_sections(struct f2fs_sb_info *sbi)
534{
535 return GET_SEC_FROM_SEG(sbi, (unsigned int)reserved_segments(sbi));
536}
537
538static inline bool has_curseg_enough_space(struct f2fs_sb_info *sbi)
539{
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;
544 int i;
545
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;
551
552 if (node_blocks > left_blocks)
553 return false;
554 }
555
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)
561 return false;
562 return true;
563}
564
565static inline bool has_not_enough_free_secs(struct f2fs_sb_info *sbi,
566 int freed, int needed)
567{
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);
571
572 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
573 return false;
574
575 if (free_sections(sbi) + freed == reserved_sections(sbi) + needed &&
576 has_curseg_enough_space(sbi))
577 return false;
578 return (free_sections(sbi) + freed) <=
579 (node_secs + 2 * dent_secs + imeta_secs +
580 reserved_sections(sbi) + needed);
581}
582
583static inline bool f2fs_is_checkpoint_ready(struct f2fs_sb_info *sbi)
584{
585 if (likely(!is_sbi_flag_set(sbi, SBI_CP_DISABLED)))
586 return true;
587 if (likely(!has_not_enough_free_secs(sbi, 0, 0)))
588 return true;
589 return false;
590}
591
592static inline bool excess_prefree_segs(struct f2fs_sb_info *sbi)
593{
594 return prefree_segments(sbi) > SM_I(sbi)->rec_prefree_segments;
595}
596
597static inline int utilization(struct f2fs_sb_info *sbi)
598{
599 return div_u64((u64)valid_user_blocks(sbi) * 100,
600 sbi->user_block_count);
601}
602
603/*
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
611 * threashold,
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)
618 */
619#define DEF_MIN_IPU_UTIL 70
620#define DEF_MIN_FSYNC_BLOCKS 8
621#define DEF_MIN_HOT_BLOCKS 16
622
623#define SMALL_VOLUME_SEGMENTS (16 * 512) /* 16GB */
624
625enum {
626 F2FS_IPU_FORCE,
627 F2FS_IPU_SSR,
628 F2FS_IPU_UTIL,
629 F2FS_IPU_SSR_UTIL,
630 F2FS_IPU_FSYNC,
631 F2FS_IPU_ASYNC,
632 F2FS_IPU_NOCACHE,
633};
634
635static inline unsigned int curseg_segno(struct f2fs_sb_info *sbi,
636 int type)
637{
638 struct curseg_info *curseg = CURSEG_I(sbi, type);
639 return curseg->segno;
640}
641
642static inline unsigned char curseg_alloc_type(struct f2fs_sb_info *sbi,
643 int type)
644{
645 struct curseg_info *curseg = CURSEG_I(sbi, type);
646 return curseg->alloc_type;
647}
648
649static inline unsigned short curseg_blkoff(struct f2fs_sb_info *sbi, int type)
650{
651 struct curseg_info *curseg = CURSEG_I(sbi, type);
652 return curseg->next_blkoff;
653}
654
655static inline void check_seg_range(struct f2fs_sb_info *sbi, unsigned int segno)
656{
657 f2fs_bug_on(sbi, segno > TOTAL_SEGS(sbi) - 1);
658}
659
660static inline void verify_fio_blkaddr(struct f2fs_io_info *fio)
661{
662 struct f2fs_sb_info *sbi = fio->sbi;
663
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);
669}
670
671/*
672 * Summary block is always treated as an invalid block
673 */
674static inline int check_block_count(struct f2fs_sb_info *sbi,
675 int segno, struct f2fs_sit_entry *raw_sit)
676{
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;
680
681 /* check bitmap with valid block count */
682 do {
683 if (is_valid) {
684 next_pos = find_next_zero_bit_le(&raw_sit->valid_map,
685 sbi->blocks_per_seg,
686 cur_pos);
687 valid_blocks += next_pos - cur_pos;
688 } else
689 next_pos = find_next_bit_le(&raw_sit->valid_map,
690 sbi->blocks_per_seg,
691 cur_pos);
692 cur_pos = next_pos;
693 is_valid = !is_valid;
694 } while (cur_pos < sbi->blocks_per_seg);
695
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;
701 }
702
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;
710 }
711 return 0;
712}
713
714static inline pgoff_t current_sit_addr(struct f2fs_sb_info *sbi,
715 unsigned int start)
716{
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;
720
721 check_seg_range(sbi, start);
722
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))
726 f2fs_bug_on(sbi, 1);
727#endif
728
729 /* calculate sit block address */
730 if (f2fs_test_bit(offset, sit_i->sit_bitmap))
731 blk_addr += sit_i->sit_blocks;
732
733 return blk_addr;
734}
735
736static inline pgoff_t next_sit_addr(struct f2fs_sb_info *sbi,
737 pgoff_t block_addr)
738{
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;
743 else
744 block_addr -= sit_i->sit_blocks;
745
746 return block_addr + sit_i->sit_base_addr;
747}
748
749static inline void set_to_next_sit(struct sit_info *sit_i, unsigned int start)
750{
751 unsigned int block_off = SIT_BLOCK_OFFSET(start);
752
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);
756#endif
757}
758
759static inline unsigned long long get_mtime(struct f2fs_sb_info *sbi,
760 bool base_time)
761{
762 struct sit_info *sit_i = SIT_I(sbi);
763 time64_t diff, now = ktime_get_boottime_seconds();
764
765 if (now >= sit_i->mounted_time)
766 return sit_i->elapsed_time + now - sit_i->mounted_time;
767
768 /* system time is set to the past */
769 if (!base_time) {
770 diff = sit_i->mounted_time - now;
771 if (sit_i->elapsed_time >= diff)
772 return sit_i->elapsed_time - diff;
773 return 0;
774 }
775 return sit_i->elapsed_time;
776}
777
778static inline void set_summary(struct f2fs_summary *sum, nid_t nid,
779 unsigned int ofs_in_node, unsigned char version)
780{
781 sum->nid = cpu_to_le32(nid);
782 sum->ofs_in_node = cpu_to_le16(ofs_in_node);
783 sum->version = version;
784}
785
786static inline block_t start_sum_block(struct f2fs_sb_info *sbi)
787{
788 return __start_cp_addr(sbi) +
789 le32_to_cpu(F2FS_CKPT(sbi)->cp_pack_start_sum);
790}
791
792static inline block_t sum_blk_addr(struct f2fs_sb_info *sbi, int base, int type)
793{
794 return __start_cp_addr(sbi) +
795 le32_to_cpu(F2FS_CKPT(sbi)->cp_pack_total_block_count)
796 - (base + 1) + type;
797}
798
799static inline bool sec_usage_check(struct f2fs_sb_info *sbi, unsigned int secno)
800{
801 if (IS_CURSEC(sbi, secno) || (sbi->cur_victim_sec == secno))
802 return true;
803 return false;
804}
805
806/*
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.
812 */
813static inline int nr_pages_to_skip(struct f2fs_sb_info *sbi, int type)
814{
815 if (sbi->sb->s_bdi->wb.dirty_exceeded)
816 return 0;
817
818 if (type == DATA)
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;
824 else
825 return 0;
826}
827
828/*
829 * When writing pages, it'd better align nr_to_write for segment size.
830 */
831static inline long nr_pages_to_write(struct f2fs_sb_info *sbi, int type,
832 struct writeback_control *wbc)
833{
834 long nr_to_write, desired;
835
836 if (wbc->sync_mode != WB_SYNC_NONE)
837 return 0;
838
839 nr_to_write = wbc->nr_to_write;
840 desired = BIO_MAX_PAGES;
841 if (type == NODE)
842 desired <<= 1;
843
844 wbc->nr_to_write = desired;
845 return desired - nr_to_write;
846}
847
848static inline void wake_up_discard_thread(struct f2fs_sb_info *sbi, bool force)
849{
850 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
851 bool wakeup = false;
852 int i;
853
854 if (force)
855 goto wake_up;
856
857 mutex_lock(&dcc->cmd_lock);
858 for (i = MAX_PLIST_NUM - 1; i >= 0; i--) {
859 if (i + 1 < dcc->discard_granularity)
860 break;
861 if (!list_empty(&dcc->pend_list[i])) {
862 wakeup = true;
863 break;
864 }
865 }
866 mutex_unlock(&dcc->cmd_lock);
867 if (!wakeup || !is_idle(sbi, DISCARD_TIME))
868 return;
869wake_up:
870 dcc->discard_wake = 1;
871 wake_up_interruptible_all(&dcc->discard_wait_queue);
872}
1/*
2 * fs/f2fs/segment.h
3 *
4 * Copyright (c) 2012 Samsung Electronics Co., Ltd.
5 * http://www.samsung.com/
6 *
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.
10 */
11#include <linux/blkdev.h>
12#include <linux/backing-dev.h>
13
14/* constant macro */
15#define NULL_SEGNO ((unsigned int)(~0))
16#define NULL_SECNO ((unsigned int)(~0))
17
18#define DEF_RECLAIM_PREFREE_SEGMENTS 5 /* 5% over total segments */
19
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)
23
24#define IS_DATASEG(t) (t <= CURSEG_COLD_DATA)
25#define IS_NODESEG(t) (t >= CURSEG_HOT_NODE)
26
27#define IS_CURSEG(sbi, seg) \
28 ((seg == CURSEG_I(sbi, CURSEG_HOT_DATA)->segno) || \
29 (seg == CURSEG_I(sbi, CURSEG_WARM_DATA)->segno) || \
30 (seg == CURSEG_I(sbi, CURSEG_COLD_DATA)->segno) || \
31 (seg == CURSEG_I(sbi, CURSEG_HOT_NODE)->segno) || \
32 (seg == CURSEG_I(sbi, CURSEG_WARM_NODE)->segno) || \
33 (seg == CURSEG_I(sbi, CURSEG_COLD_NODE)->segno))
34
35#define IS_CURSEC(sbi, secno) \
36 ((secno == CURSEG_I(sbi, CURSEG_HOT_DATA)->segno / \
37 sbi->segs_per_sec) || \
38 (secno == CURSEG_I(sbi, CURSEG_WARM_DATA)->segno / \
39 sbi->segs_per_sec) || \
40 (secno == CURSEG_I(sbi, CURSEG_COLD_DATA)->segno / \
41 sbi->segs_per_sec) || \
42 (secno == CURSEG_I(sbi, CURSEG_HOT_NODE)->segno / \
43 sbi->segs_per_sec) || \
44 (secno == CURSEG_I(sbi, CURSEG_WARM_NODE)->segno / \
45 sbi->segs_per_sec) || \
46 (secno == CURSEG_I(sbi, CURSEG_COLD_NODE)->segno / \
47 sbi->segs_per_sec)) \
48
49#define MAIN_BLKADDR(sbi) (SM_I(sbi)->main_blkaddr)
50#define SEG0_BLKADDR(sbi) (SM_I(sbi)->seg0_blkaddr)
51
52#define MAIN_SEGS(sbi) (SM_I(sbi)->main_segments)
53#define MAIN_SECS(sbi) (sbi->total_sections)
54
55#define TOTAL_SEGS(sbi) (SM_I(sbi)->segment_count)
56#define TOTAL_BLKS(sbi) (TOTAL_SEGS(sbi) << sbi->log_blocks_per_seg)
57
58#define MAX_BLKADDR(sbi) (SEG0_BLKADDR(sbi) + TOTAL_BLKS(sbi))
59#define SEGMENT_SIZE(sbi) (1ULL << (sbi->log_blocksize + \
60 sbi->log_blocks_per_seg))
61
62#define START_BLOCK(sbi, segno) (SEG0_BLKADDR(sbi) + \
63 (GET_R2L_SEGNO(FREE_I(sbi), segno) << sbi->log_blocks_per_seg))
64
65#define NEXT_FREE_BLKADDR(sbi, curseg) \
66 (START_BLOCK(sbi, curseg->segno) + curseg->next_blkoff)
67
68#define GET_SEGOFF_FROM_SEG0(sbi, blk_addr) ((blk_addr) - SEG0_BLKADDR(sbi))
69#define GET_SEGNO_FROM_SEG0(sbi, blk_addr) \
70 (GET_SEGOFF_FROM_SEG0(sbi, blk_addr) >> sbi->log_blocks_per_seg)
71#define GET_BLKOFF_FROM_SEG0(sbi, blk_addr) \
72 (GET_SEGOFF_FROM_SEG0(sbi, blk_addr) & (sbi->blocks_per_seg - 1))
73
74#define GET_SEGNO(sbi, blk_addr) \
75 (((blk_addr == NULL_ADDR) || (blk_addr == NEW_ADDR)) ? \
76 NULL_SEGNO : GET_L2R_SEGNO(FREE_I(sbi), \
77 GET_SEGNO_FROM_SEG0(sbi, blk_addr)))
78#define GET_SECNO(sbi, segno) \
79 ((segno) / sbi->segs_per_sec)
80#define GET_ZONENO_FROM_SEGNO(sbi, segno) \
81 ((segno / sbi->segs_per_sec) / sbi->secs_per_zone)
82
83#define GET_SUM_BLOCK(sbi, segno) \
84 ((sbi->sm_info->ssa_blkaddr) + segno)
85
86#define GET_SUM_TYPE(footer) ((footer)->entry_type)
87#define SET_SUM_TYPE(footer, type) ((footer)->entry_type = type)
88
89#define SIT_ENTRY_OFFSET(sit_i, segno) \
90 (segno % sit_i->sents_per_block)
91#define SIT_BLOCK_OFFSET(segno) \
92 (segno / SIT_ENTRY_PER_BLOCK)
93#define START_SEGNO(segno) \
94 (SIT_BLOCK_OFFSET(segno) * SIT_ENTRY_PER_BLOCK)
95#define SIT_BLK_CNT(sbi) \
96 ((MAIN_SEGS(sbi) + SIT_ENTRY_PER_BLOCK - 1) / SIT_ENTRY_PER_BLOCK)
97#define f2fs_bitmap_size(nr) \
98 (BITS_TO_LONGS(nr) * sizeof(unsigned long))
99
100#define SECTOR_FROM_BLOCK(blk_addr) \
101 (((sector_t)blk_addr) << F2FS_LOG_SECTORS_PER_BLOCK)
102#define SECTOR_TO_BLOCK(sectors) \
103 (sectors >> F2FS_LOG_SECTORS_PER_BLOCK)
104#define MAX_BIO_BLOCKS(sbi) \
105 ((int)min((int)max_hw_blocks(sbi), BIO_MAX_PAGES))
106
107/*
108 * indicate a block allocation direction: RIGHT and LEFT.
109 * RIGHT means allocating new sections towards the end of volume.
110 * LEFT means the opposite direction.
111 */
112enum {
113 ALLOC_RIGHT = 0,
114 ALLOC_LEFT
115};
116
117/*
118 * In the victim_sel_policy->alloc_mode, there are two block allocation modes.
119 * LFS writes data sequentially with cleaning operations.
120 * SSR (Slack Space Recycle) reuses obsolete space without cleaning operations.
121 */
122enum {
123 LFS = 0,
124 SSR
125};
126
127/*
128 * In the victim_sel_policy->gc_mode, there are two gc, aka cleaning, modes.
129 * GC_CB is based on cost-benefit algorithm.
130 * GC_GREEDY is based on greedy algorithm.
131 */
132enum {
133 GC_CB = 0,
134 GC_GREEDY
135};
136
137/*
138 * BG_GC means the background cleaning job.
139 * FG_GC means the on-demand cleaning job.
140 * FORCE_FG_GC means on-demand cleaning job in background.
141 */
142enum {
143 BG_GC = 0,
144 FG_GC,
145 FORCE_FG_GC,
146};
147
148/* for a function parameter to select a victim segment */
149struct victim_sel_policy {
150 int alloc_mode; /* LFS or SSR */
151 int gc_mode; /* GC_CB or GC_GREEDY */
152 unsigned long *dirty_segmap; /* dirty segment bitmap */
153 unsigned int max_search; /* maximum # of segments to search */
154 unsigned int offset; /* last scanned bitmap offset */
155 unsigned int ofs_unit; /* bitmap search unit */
156 unsigned int min_cost; /* minimum cost */
157 unsigned int min_segno; /* segment # having min. cost */
158};
159
160struct seg_entry {
161 unsigned short valid_blocks; /* # of valid blocks */
162 unsigned char *cur_valid_map; /* validity bitmap of blocks */
163 /*
164 * # of valid blocks and the validity bitmap stored in the the last
165 * checkpoint pack. This information is used by the SSR mode.
166 */
167 unsigned short ckpt_valid_blocks;
168 unsigned char *ckpt_valid_map;
169 unsigned char *discard_map;
170 unsigned char type; /* segment type like CURSEG_XXX_TYPE */
171 unsigned long long mtime; /* modification time of the segment */
172};
173
174struct sec_entry {
175 unsigned int valid_blocks; /* # of valid blocks in a section */
176};
177
178struct segment_allocation {
179 void (*allocate_segment)(struct f2fs_sb_info *, int, bool);
180};
181
182/*
183 * this value is set in page as a private data which indicate that
184 * the page is atomically written, and it is in inmem_pages list.
185 */
186#define ATOMIC_WRITTEN_PAGE ((unsigned long)-1)
187
188#define IS_ATOMIC_WRITTEN_PAGE(page) \
189 (page_private(page) == (unsigned long)ATOMIC_WRITTEN_PAGE)
190
191struct inmem_pages {
192 struct list_head list;
193 struct page *page;
194 block_t old_addr; /* for revoking when fail to commit */
195};
196
197struct sit_info {
198 const struct segment_allocation *s_ops;
199
200 block_t sit_base_addr; /* start block address of SIT area */
201 block_t sit_blocks; /* # of blocks used by SIT area */
202 block_t written_valid_blocks; /* # of valid blocks in main area */
203 char *sit_bitmap; /* SIT bitmap pointer */
204 unsigned int bitmap_size; /* SIT bitmap size */
205
206 unsigned long *tmp_map; /* bitmap for temporal use */
207 unsigned long *dirty_sentries_bitmap; /* bitmap for dirty sentries */
208 unsigned int dirty_sentries; /* # of dirty sentries */
209 unsigned int sents_per_block; /* # of SIT entries per block */
210 struct mutex sentry_lock; /* to protect SIT cache */
211 struct seg_entry *sentries; /* SIT segment-level cache */
212 struct sec_entry *sec_entries; /* SIT section-level cache */
213
214 /* for cost-benefit algorithm in cleaning procedure */
215 unsigned long long elapsed_time; /* elapsed time after mount */
216 unsigned long long mounted_time; /* mount time */
217 unsigned long long min_mtime; /* min. modification time */
218 unsigned long long max_mtime; /* max. modification time */
219};
220
221struct free_segmap_info {
222 unsigned int start_segno; /* start segment number logically */
223 unsigned int free_segments; /* # of free segments */
224 unsigned int free_sections; /* # of free sections */
225 spinlock_t segmap_lock; /* free segmap lock */
226 unsigned long *free_segmap; /* free segment bitmap */
227 unsigned long *free_secmap; /* free section bitmap */
228};
229
230/* Notice: The order of dirty type is same with CURSEG_XXX in f2fs.h */
231enum dirty_type {
232 DIRTY_HOT_DATA, /* dirty segments assigned as hot data logs */
233 DIRTY_WARM_DATA, /* dirty segments assigned as warm data logs */
234 DIRTY_COLD_DATA, /* dirty segments assigned as cold data logs */
235 DIRTY_HOT_NODE, /* dirty segments assigned as hot node logs */
236 DIRTY_WARM_NODE, /* dirty segments assigned as warm node logs */
237 DIRTY_COLD_NODE, /* dirty segments assigned as cold node logs */
238 DIRTY, /* to count # of dirty segments */
239 PRE, /* to count # of entirely obsolete segments */
240 NR_DIRTY_TYPE
241};
242
243struct dirty_seglist_info {
244 const struct victim_selection *v_ops; /* victim selction operation */
245 unsigned long *dirty_segmap[NR_DIRTY_TYPE];
246 struct mutex seglist_lock; /* lock for segment bitmaps */
247 int nr_dirty[NR_DIRTY_TYPE]; /* # of dirty segments */
248 unsigned long *victim_secmap; /* background GC victims */
249};
250
251/* victim selection function for cleaning and SSR */
252struct victim_selection {
253 int (*get_victim)(struct f2fs_sb_info *, unsigned int *,
254 int, int, char);
255};
256
257/* for active log information */
258struct curseg_info {
259 struct mutex curseg_mutex; /* lock for consistency */
260 struct f2fs_summary_block *sum_blk; /* cached summary block */
261 struct rw_semaphore journal_rwsem; /* protect journal area */
262 struct f2fs_journal *journal; /* cached journal info */
263 unsigned char alloc_type; /* current allocation type */
264 unsigned int segno; /* current segment number */
265 unsigned short next_blkoff; /* next block offset to write */
266 unsigned int zone; /* current zone number */
267 unsigned int next_segno; /* preallocated segment */
268};
269
270struct sit_entry_set {
271 struct list_head set_list; /* link with all sit sets */
272 unsigned int start_segno; /* start segno of sits in set */
273 unsigned int entry_cnt; /* the # of sit entries in set */
274};
275
276/*
277 * inline functions
278 */
279static inline struct curseg_info *CURSEG_I(struct f2fs_sb_info *sbi, int type)
280{
281 return (struct curseg_info *)(SM_I(sbi)->curseg_array + type);
282}
283
284static inline struct seg_entry *get_seg_entry(struct f2fs_sb_info *sbi,
285 unsigned int segno)
286{
287 struct sit_info *sit_i = SIT_I(sbi);
288 return &sit_i->sentries[segno];
289}
290
291static inline struct sec_entry *get_sec_entry(struct f2fs_sb_info *sbi,
292 unsigned int segno)
293{
294 struct sit_info *sit_i = SIT_I(sbi);
295 return &sit_i->sec_entries[GET_SECNO(sbi, segno)];
296}
297
298static inline unsigned int get_valid_blocks(struct f2fs_sb_info *sbi,
299 unsigned int segno, int section)
300{
301 /*
302 * In order to get # of valid blocks in a section instantly from many
303 * segments, f2fs manages two counting structures separately.
304 */
305 if (section > 1)
306 return get_sec_entry(sbi, segno)->valid_blocks;
307 else
308 return get_seg_entry(sbi, segno)->valid_blocks;
309}
310
311static inline void seg_info_from_raw_sit(struct seg_entry *se,
312 struct f2fs_sit_entry *rs)
313{
314 se->valid_blocks = GET_SIT_VBLOCKS(rs);
315 se->ckpt_valid_blocks = GET_SIT_VBLOCKS(rs);
316 memcpy(se->cur_valid_map, rs->valid_map, SIT_VBLOCK_MAP_SIZE);
317 memcpy(se->ckpt_valid_map, rs->valid_map, SIT_VBLOCK_MAP_SIZE);
318 se->type = GET_SIT_TYPE(rs);
319 se->mtime = le64_to_cpu(rs->mtime);
320}
321
322static inline void seg_info_to_raw_sit(struct seg_entry *se,
323 struct f2fs_sit_entry *rs)
324{
325 unsigned short raw_vblocks = (se->type << SIT_VBLOCKS_SHIFT) |
326 se->valid_blocks;
327 rs->vblocks = cpu_to_le16(raw_vblocks);
328 memcpy(rs->valid_map, se->cur_valid_map, SIT_VBLOCK_MAP_SIZE);
329 memcpy(se->ckpt_valid_map, rs->valid_map, SIT_VBLOCK_MAP_SIZE);
330 se->ckpt_valid_blocks = se->valid_blocks;
331 rs->mtime = cpu_to_le64(se->mtime);
332}
333
334static inline unsigned int find_next_inuse(struct free_segmap_info *free_i,
335 unsigned int max, unsigned int segno)
336{
337 unsigned int ret;
338 spin_lock(&free_i->segmap_lock);
339 ret = find_next_bit(free_i->free_segmap, max, segno);
340 spin_unlock(&free_i->segmap_lock);
341 return ret;
342}
343
344static inline void __set_free(struct f2fs_sb_info *sbi, unsigned int segno)
345{
346 struct free_segmap_info *free_i = FREE_I(sbi);
347 unsigned int secno = segno / sbi->segs_per_sec;
348 unsigned int start_segno = secno * sbi->segs_per_sec;
349 unsigned int next;
350
351 spin_lock(&free_i->segmap_lock);
352 clear_bit(segno, free_i->free_segmap);
353 free_i->free_segments++;
354
355 next = find_next_bit(free_i->free_segmap,
356 start_segno + sbi->segs_per_sec, start_segno);
357 if (next >= start_segno + sbi->segs_per_sec) {
358 clear_bit(secno, free_i->free_secmap);
359 free_i->free_sections++;
360 }
361 spin_unlock(&free_i->segmap_lock);
362}
363
364static inline void __set_inuse(struct f2fs_sb_info *sbi,
365 unsigned int segno)
366{
367 struct free_segmap_info *free_i = FREE_I(sbi);
368 unsigned int secno = segno / sbi->segs_per_sec;
369 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--;
373}
374
375static inline void __set_test_and_free(struct f2fs_sb_info *sbi,
376 unsigned int segno)
377{
378 struct free_segmap_info *free_i = FREE_I(sbi);
379 unsigned int secno = segno / sbi->segs_per_sec;
380 unsigned int start_segno = secno * sbi->segs_per_sec;
381 unsigned int next;
382
383 spin_lock(&free_i->segmap_lock);
384 if (test_and_clear_bit(segno, free_i->free_segmap)) {
385 free_i->free_segments++;
386
387 next = find_next_bit(free_i->free_segmap,
388 start_segno + sbi->segs_per_sec, start_segno);
389 if (next >= start_segno + sbi->segs_per_sec) {
390 if (test_and_clear_bit(secno, free_i->free_secmap))
391 free_i->free_sections++;
392 }
393 }
394 spin_unlock(&free_i->segmap_lock);
395}
396
397static inline void __set_test_and_inuse(struct f2fs_sb_info *sbi,
398 unsigned int segno)
399{
400 struct free_segmap_info *free_i = FREE_I(sbi);
401 unsigned int secno = segno / sbi->segs_per_sec;
402 spin_lock(&free_i->segmap_lock);
403 if (!test_and_set_bit(segno, free_i->free_segmap)) {
404 free_i->free_segments--;
405 if (!test_and_set_bit(secno, free_i->free_secmap))
406 free_i->free_sections--;
407 }
408 spin_unlock(&free_i->segmap_lock);
409}
410
411static inline void get_sit_bitmap(struct f2fs_sb_info *sbi,
412 void *dst_addr)
413{
414 struct sit_info *sit_i = SIT_I(sbi);
415 memcpy(dst_addr, sit_i->sit_bitmap, sit_i->bitmap_size);
416}
417
418static inline block_t written_block_count(struct f2fs_sb_info *sbi)
419{
420 return SIT_I(sbi)->written_valid_blocks;
421}
422
423static inline unsigned int free_segments(struct f2fs_sb_info *sbi)
424{
425 return FREE_I(sbi)->free_segments;
426}
427
428static inline int reserved_segments(struct f2fs_sb_info *sbi)
429{
430 return SM_I(sbi)->reserved_segments;
431}
432
433static inline unsigned int free_sections(struct f2fs_sb_info *sbi)
434{
435 return FREE_I(sbi)->free_sections;
436}
437
438static inline unsigned int prefree_segments(struct f2fs_sb_info *sbi)
439{
440 return DIRTY_I(sbi)->nr_dirty[PRE];
441}
442
443static inline unsigned int dirty_segments(struct f2fs_sb_info *sbi)
444{
445 return DIRTY_I(sbi)->nr_dirty[DIRTY_HOT_DATA] +
446 DIRTY_I(sbi)->nr_dirty[DIRTY_WARM_DATA] +
447 DIRTY_I(sbi)->nr_dirty[DIRTY_COLD_DATA] +
448 DIRTY_I(sbi)->nr_dirty[DIRTY_HOT_NODE] +
449 DIRTY_I(sbi)->nr_dirty[DIRTY_WARM_NODE] +
450 DIRTY_I(sbi)->nr_dirty[DIRTY_COLD_NODE];
451}
452
453static inline int overprovision_segments(struct f2fs_sb_info *sbi)
454{
455 return SM_I(sbi)->ovp_segments;
456}
457
458static inline int overprovision_sections(struct f2fs_sb_info *sbi)
459{
460 return ((unsigned int) overprovision_segments(sbi)) / sbi->segs_per_sec;
461}
462
463static inline int reserved_sections(struct f2fs_sb_info *sbi)
464{
465 return ((unsigned int) reserved_segments(sbi)) / sbi->segs_per_sec;
466}
467
468static inline bool need_SSR(struct f2fs_sb_info *sbi)
469{
470 int node_secs = get_blocktype_secs(sbi, F2FS_DIRTY_NODES);
471 int dent_secs = get_blocktype_secs(sbi, F2FS_DIRTY_DENTS);
472 return free_sections(sbi) <= (node_secs + 2 * dent_secs +
473 reserved_sections(sbi) + 1);
474}
475
476static inline bool has_not_enough_free_secs(struct f2fs_sb_info *sbi, int freed)
477{
478 int node_secs = get_blocktype_secs(sbi, F2FS_DIRTY_NODES);
479 int dent_secs = get_blocktype_secs(sbi, F2FS_DIRTY_DENTS);
480
481 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
482 return false;
483
484 return (free_sections(sbi) + freed) <= (node_secs + 2 * dent_secs +
485 reserved_sections(sbi));
486}
487
488static inline bool excess_prefree_segs(struct f2fs_sb_info *sbi)
489{
490 return prefree_segments(sbi) > SM_I(sbi)->rec_prefree_segments;
491}
492
493static inline int utilization(struct f2fs_sb_info *sbi)
494{
495 return div_u64((u64)valid_user_blocks(sbi) * 100,
496 sbi->user_block_count);
497}
498
499/*
500 * Sometimes f2fs may be better to drop out-of-place update policy.
501 * And, users can control the policy through sysfs entries.
502 * There are five policies with triggering conditions as follows.
503 * F2FS_IPU_FORCE - all the time,
504 * F2FS_IPU_SSR - if SSR mode is activated,
505 * F2FS_IPU_UTIL - if FS utilization is over threashold,
506 * F2FS_IPU_SSR_UTIL - if SSR mode is activated and FS utilization is over
507 * threashold,
508 * F2FS_IPU_FSYNC - activated in fsync path only for high performance flash
509 * storages. IPU will be triggered only if the # of dirty
510 * pages over min_fsync_blocks.
511 * F2FS_IPUT_DISABLE - disable IPU. (=default option)
512 */
513#define DEF_MIN_IPU_UTIL 70
514#define DEF_MIN_FSYNC_BLOCKS 8
515
516enum {
517 F2FS_IPU_FORCE,
518 F2FS_IPU_SSR,
519 F2FS_IPU_UTIL,
520 F2FS_IPU_SSR_UTIL,
521 F2FS_IPU_FSYNC,
522};
523
524static inline bool need_inplace_update(struct inode *inode)
525{
526 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
527 unsigned int policy = SM_I(sbi)->ipu_policy;
528
529 /* IPU can be done only for the user data */
530 if (S_ISDIR(inode->i_mode) || f2fs_is_atomic_file(inode))
531 return false;
532
533 if (policy & (0x1 << F2FS_IPU_FORCE))
534 return true;
535 if (policy & (0x1 << F2FS_IPU_SSR) && need_SSR(sbi))
536 return true;
537 if (policy & (0x1 << F2FS_IPU_UTIL) &&
538 utilization(sbi) > SM_I(sbi)->min_ipu_util)
539 return true;
540 if (policy & (0x1 << F2FS_IPU_SSR_UTIL) && need_SSR(sbi) &&
541 utilization(sbi) > SM_I(sbi)->min_ipu_util)
542 return true;
543
544 /* this is only set during fdatasync */
545 if (policy & (0x1 << F2FS_IPU_FSYNC) &&
546 is_inode_flag_set(F2FS_I(inode), FI_NEED_IPU))
547 return true;
548
549 return false;
550}
551
552static inline unsigned int curseg_segno(struct f2fs_sb_info *sbi,
553 int type)
554{
555 struct curseg_info *curseg = CURSEG_I(sbi, type);
556 return curseg->segno;
557}
558
559static inline unsigned char curseg_alloc_type(struct f2fs_sb_info *sbi,
560 int type)
561{
562 struct curseg_info *curseg = CURSEG_I(sbi, type);
563 return curseg->alloc_type;
564}
565
566static inline unsigned short curseg_blkoff(struct f2fs_sb_info *sbi, int type)
567{
568 struct curseg_info *curseg = CURSEG_I(sbi, type);
569 return curseg->next_blkoff;
570}
571
572static inline void check_seg_range(struct f2fs_sb_info *sbi, unsigned int segno)
573{
574 f2fs_bug_on(sbi, segno > TOTAL_SEGS(sbi) - 1);
575}
576
577static inline void verify_block_addr(struct f2fs_sb_info *sbi, block_t blk_addr)
578{
579 f2fs_bug_on(sbi, blk_addr < SEG0_BLKADDR(sbi)
580 || blk_addr >= MAX_BLKADDR(sbi));
581}
582
583/*
584 * Summary block is always treated as an invalid block
585 */
586static inline void check_block_count(struct f2fs_sb_info *sbi,
587 int segno, struct f2fs_sit_entry *raw_sit)
588{
589#ifdef CONFIG_F2FS_CHECK_FS
590 bool is_valid = test_bit_le(0, raw_sit->valid_map) ? true : false;
591 int valid_blocks = 0;
592 int cur_pos = 0, next_pos;
593
594 /* check bitmap with valid block count */
595 do {
596 if (is_valid) {
597 next_pos = find_next_zero_bit_le(&raw_sit->valid_map,
598 sbi->blocks_per_seg,
599 cur_pos);
600 valid_blocks += next_pos - cur_pos;
601 } else
602 next_pos = find_next_bit_le(&raw_sit->valid_map,
603 sbi->blocks_per_seg,
604 cur_pos);
605 cur_pos = next_pos;
606 is_valid = !is_valid;
607 } while (cur_pos < sbi->blocks_per_seg);
608 BUG_ON(GET_SIT_VBLOCKS(raw_sit) != valid_blocks);
609#endif
610 /* check segment usage, and check boundary of a given segment number */
611 f2fs_bug_on(sbi, GET_SIT_VBLOCKS(raw_sit) > sbi->blocks_per_seg
612 || segno > TOTAL_SEGS(sbi) - 1);
613}
614
615static inline pgoff_t current_sit_addr(struct f2fs_sb_info *sbi,
616 unsigned int start)
617{
618 struct sit_info *sit_i = SIT_I(sbi);
619 unsigned int offset = SIT_BLOCK_OFFSET(start);
620 block_t blk_addr = sit_i->sit_base_addr + offset;
621
622 check_seg_range(sbi, start);
623
624 /* calculate sit block address */
625 if (f2fs_test_bit(offset, sit_i->sit_bitmap))
626 blk_addr += sit_i->sit_blocks;
627
628 return blk_addr;
629}
630
631static inline pgoff_t next_sit_addr(struct f2fs_sb_info *sbi,
632 pgoff_t block_addr)
633{
634 struct sit_info *sit_i = SIT_I(sbi);
635 block_addr -= sit_i->sit_base_addr;
636 if (block_addr < sit_i->sit_blocks)
637 block_addr += sit_i->sit_blocks;
638 else
639 block_addr -= sit_i->sit_blocks;
640
641 return block_addr + sit_i->sit_base_addr;
642}
643
644static inline void set_to_next_sit(struct sit_info *sit_i, unsigned int start)
645{
646 unsigned int block_off = SIT_BLOCK_OFFSET(start);
647
648 f2fs_change_bit(block_off, sit_i->sit_bitmap);
649}
650
651static inline unsigned long long get_mtime(struct f2fs_sb_info *sbi)
652{
653 struct sit_info *sit_i = SIT_I(sbi);
654 return sit_i->elapsed_time + CURRENT_TIME_SEC.tv_sec -
655 sit_i->mounted_time;
656}
657
658static inline void set_summary(struct f2fs_summary *sum, nid_t nid,
659 unsigned int ofs_in_node, unsigned char version)
660{
661 sum->nid = cpu_to_le32(nid);
662 sum->ofs_in_node = cpu_to_le16(ofs_in_node);
663 sum->version = version;
664}
665
666static inline block_t start_sum_block(struct f2fs_sb_info *sbi)
667{
668 return __start_cp_addr(sbi) +
669 le32_to_cpu(F2FS_CKPT(sbi)->cp_pack_start_sum);
670}
671
672static inline block_t sum_blk_addr(struct f2fs_sb_info *sbi, int base, int type)
673{
674 return __start_cp_addr(sbi) +
675 le32_to_cpu(F2FS_CKPT(sbi)->cp_pack_total_block_count)
676 - (base + 1) + type;
677}
678
679static inline bool sec_usage_check(struct f2fs_sb_info *sbi, unsigned int secno)
680{
681 if (IS_CURSEC(sbi, secno) || (sbi->cur_victim_sec == secno))
682 return true;
683 return false;
684}
685
686static inline unsigned int max_hw_blocks(struct f2fs_sb_info *sbi)
687{
688 struct block_device *bdev = sbi->sb->s_bdev;
689 struct request_queue *q = bdev_get_queue(bdev);
690 return SECTOR_TO_BLOCK(queue_max_sectors(q));
691}
692
693/*
694 * It is very important to gather dirty pages and write at once, so that we can
695 * submit a big bio without interfering other data writes.
696 * By default, 512 pages for directory data,
697 * 512 pages (2MB) * 3 for three types of nodes, and
698 * max_bio_blocks for meta are set.
699 */
700static inline int nr_pages_to_skip(struct f2fs_sb_info *sbi, int type)
701{
702 if (sbi->sb->s_bdi->wb.dirty_exceeded)
703 return 0;
704
705 if (type == DATA)
706 return sbi->blocks_per_seg;
707 else if (type == NODE)
708 return 3 * sbi->blocks_per_seg;
709 else if (type == META)
710 return MAX_BIO_BLOCKS(sbi);
711 else
712 return 0;
713}
714
715/*
716 * When writing pages, it'd better align nr_to_write for segment size.
717 */
718static inline long nr_pages_to_write(struct f2fs_sb_info *sbi, int type,
719 struct writeback_control *wbc)
720{
721 long nr_to_write, desired;
722
723 if (wbc->sync_mode != WB_SYNC_NONE)
724 return 0;
725
726 nr_to_write = wbc->nr_to_write;
727
728 if (type == DATA)
729 desired = 4096;
730 else if (type == NODE)
731 desired = 3 * max_hw_blocks(sbi);
732 else
733 desired = MAX_BIO_BLOCKS(sbi);
734
735 wbc->nr_to_write = desired;
736 return desired - nr_to_write;
737}