Loading...
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
13/* constant macro */
14#define NULL_SEGNO ((unsigned int)(~0))
15#define NULL_SECNO ((unsigned int)(~0))
16
17#define DEF_RECLAIM_PREFREE_SEGMENTS 5 /* 5% over total segments */
18
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)
22
23#define IS_DATASEG(t) (t <= CURSEG_COLD_DATA)
24#define IS_NODESEG(t) (t >= CURSEG_HOT_NODE)
25
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))
33
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 / \
46 sbi->segs_per_sec)) \
47
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)
53
54#define MAIN_BASE_BLOCK(sbi) (SM_I(sbi)->main_blkaddr)
55
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_BLKOFF_FROM_SEG0(sbi, blk_addr) \
61 (GET_SEGOFF_FROM_SEG0(sbi, blk_addr) & (sbi->blocks_per_seg - 1))
62
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)
71
72#define GET_SUM_BLOCK(sbi, segno) \
73 ((sbi->sm_info->ssa_blkaddr) + segno)
74
75#define GET_SUM_TYPE(footer) ((footer)->entry_type)
76#define SET_SUM_TYPE(footer, type) ((footer)->entry_type = type)
77
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 SIT_BLK_CNT(sbi) \
85 ((TOTAL_SEGS(sbi) + SIT_ENTRY_PER_BLOCK - 1) / SIT_ENTRY_PER_BLOCK)
86#define f2fs_bitmap_size(nr) \
87 (BITS_TO_LONGS(nr) * sizeof(unsigned long))
88#define TOTAL_SEGS(sbi) (SM_I(sbi)->main_segments)
89#define TOTAL_SECS(sbi) (sbi->total_sections)
90
91#define SECTOR_FROM_BLOCK(sbi, blk_addr) \
92 (((sector_t)blk_addr) << (sbi)->log_sectors_per_block)
93#define SECTOR_TO_BLOCK(sbi, sectors) \
94 (sectors >> (sbi)->log_sectors_per_block)
95#define MAX_BIO_BLOCKS(max_hw_blocks) \
96 (min((int)max_hw_blocks, BIO_MAX_PAGES))
97
98/*
99 * indicate a block allocation direction: RIGHT and LEFT.
100 * RIGHT means allocating new sections towards the end of volume.
101 * LEFT means the opposite direction.
102 */
103enum {
104 ALLOC_RIGHT = 0,
105 ALLOC_LEFT
106};
107
108/*
109 * In the victim_sel_policy->alloc_mode, there are two block allocation modes.
110 * LFS writes data sequentially with cleaning operations.
111 * SSR (Slack Space Recycle) reuses obsolete space without cleaning operations.
112 */
113enum {
114 LFS = 0,
115 SSR
116};
117
118/*
119 * In the victim_sel_policy->gc_mode, there are two gc, aka cleaning, modes.
120 * GC_CB is based on cost-benefit algorithm.
121 * GC_GREEDY is based on greedy algorithm.
122 */
123enum {
124 GC_CB = 0,
125 GC_GREEDY
126};
127
128/*
129 * BG_GC means the background cleaning job.
130 * FG_GC means the on-demand cleaning job.
131 */
132enum {
133 BG_GC = 0,
134 FG_GC
135};
136
137/* for a function parameter to select a victim segment */
138struct victim_sel_policy {
139 int alloc_mode; /* LFS or SSR */
140 int gc_mode; /* GC_CB or GC_GREEDY */
141 unsigned long *dirty_segmap; /* dirty segment bitmap */
142 unsigned int max_search; /* maximum # of segments to search */
143 unsigned int offset; /* last scanned bitmap offset */
144 unsigned int ofs_unit; /* bitmap search unit */
145 unsigned int min_cost; /* minimum cost */
146 unsigned int min_segno; /* segment # having min. cost */
147};
148
149struct seg_entry {
150 unsigned short valid_blocks; /* # of valid blocks */
151 unsigned char *cur_valid_map; /* validity bitmap of blocks */
152 /*
153 * # of valid blocks and the validity bitmap stored in the the last
154 * checkpoint pack. This information is used by the SSR mode.
155 */
156 unsigned short ckpt_valid_blocks;
157 unsigned char *ckpt_valid_map;
158 unsigned char type; /* segment type like CURSEG_XXX_TYPE */
159 unsigned long long mtime; /* modification time of the segment */
160};
161
162struct sec_entry {
163 unsigned int valid_blocks; /* # of valid blocks in a section */
164};
165
166struct segment_allocation {
167 void (*allocate_segment)(struct f2fs_sb_info *, int, bool);
168};
169
170struct sit_info {
171 const struct segment_allocation *s_ops;
172
173 block_t sit_base_addr; /* start block address of SIT area */
174 block_t sit_blocks; /* # of blocks used by SIT area */
175 block_t written_valid_blocks; /* # of valid blocks in main area */
176 char *sit_bitmap; /* SIT bitmap pointer */
177 unsigned int bitmap_size; /* SIT bitmap size */
178
179 unsigned long *dirty_sentries_bitmap; /* bitmap for dirty sentries */
180 unsigned int dirty_sentries; /* # of dirty sentries */
181 unsigned int sents_per_block; /* # of SIT entries per block */
182 struct mutex sentry_lock; /* to protect SIT cache */
183 struct seg_entry *sentries; /* SIT segment-level cache */
184 struct sec_entry *sec_entries; /* SIT section-level cache */
185
186 /* for cost-benefit algorithm in cleaning procedure */
187 unsigned long long elapsed_time; /* elapsed time after mount */
188 unsigned long long mounted_time; /* mount time */
189 unsigned long long min_mtime; /* min. modification time */
190 unsigned long long max_mtime; /* max. modification time */
191};
192
193struct free_segmap_info {
194 unsigned int start_segno; /* start segment number logically */
195 unsigned int free_segments; /* # of free segments */
196 unsigned int free_sections; /* # of free sections */
197 rwlock_t segmap_lock; /* free segmap lock */
198 unsigned long *free_segmap; /* free segment bitmap */
199 unsigned long *free_secmap; /* free section bitmap */
200};
201
202/* Notice: The order of dirty type is same with CURSEG_XXX in f2fs.h */
203enum dirty_type {
204 DIRTY_HOT_DATA, /* dirty segments assigned as hot data logs */
205 DIRTY_WARM_DATA, /* dirty segments assigned as warm data logs */
206 DIRTY_COLD_DATA, /* dirty segments assigned as cold data logs */
207 DIRTY_HOT_NODE, /* dirty segments assigned as hot node logs */
208 DIRTY_WARM_NODE, /* dirty segments assigned as warm node logs */
209 DIRTY_COLD_NODE, /* dirty segments assigned as cold node logs */
210 DIRTY, /* to count # of dirty segments */
211 PRE, /* to count # of entirely obsolete segments */
212 NR_DIRTY_TYPE
213};
214
215struct dirty_seglist_info {
216 const struct victim_selection *v_ops; /* victim selction operation */
217 unsigned long *dirty_segmap[NR_DIRTY_TYPE];
218 struct mutex seglist_lock; /* lock for segment bitmaps */
219 int nr_dirty[NR_DIRTY_TYPE]; /* # of dirty segments */
220 unsigned long *victim_secmap; /* background GC victims */
221};
222
223/* victim selection function for cleaning and SSR */
224struct victim_selection {
225 int (*get_victim)(struct f2fs_sb_info *, unsigned int *,
226 int, int, char);
227};
228
229/* for active log information */
230struct curseg_info {
231 struct mutex curseg_mutex; /* lock for consistency */
232 struct f2fs_summary_block *sum_blk; /* cached summary block */
233 unsigned char alloc_type; /* current allocation type */
234 unsigned int segno; /* current segment number */
235 unsigned short next_blkoff; /* next block offset to write */
236 unsigned int zone; /* current zone number */
237 unsigned int next_segno; /* preallocated segment */
238};
239
240/*
241 * inline functions
242 */
243static inline struct curseg_info *CURSEG_I(struct f2fs_sb_info *sbi, int type)
244{
245 return (struct curseg_info *)(SM_I(sbi)->curseg_array + type);
246}
247
248static inline struct seg_entry *get_seg_entry(struct f2fs_sb_info *sbi,
249 unsigned int segno)
250{
251 struct sit_info *sit_i = SIT_I(sbi);
252 return &sit_i->sentries[segno];
253}
254
255static inline struct sec_entry *get_sec_entry(struct f2fs_sb_info *sbi,
256 unsigned int segno)
257{
258 struct sit_info *sit_i = SIT_I(sbi);
259 return &sit_i->sec_entries[GET_SECNO(sbi, segno)];
260}
261
262static inline unsigned int get_valid_blocks(struct f2fs_sb_info *sbi,
263 unsigned int segno, int section)
264{
265 /*
266 * In order to get # of valid blocks in a section instantly from many
267 * segments, f2fs manages two counting structures separately.
268 */
269 if (section > 1)
270 return get_sec_entry(sbi, segno)->valid_blocks;
271 else
272 return get_seg_entry(sbi, segno)->valid_blocks;
273}
274
275static inline void seg_info_from_raw_sit(struct seg_entry *se,
276 struct f2fs_sit_entry *rs)
277{
278 se->valid_blocks = GET_SIT_VBLOCKS(rs);
279 se->ckpt_valid_blocks = GET_SIT_VBLOCKS(rs);
280 memcpy(se->cur_valid_map, rs->valid_map, SIT_VBLOCK_MAP_SIZE);
281 memcpy(se->ckpt_valid_map, rs->valid_map, SIT_VBLOCK_MAP_SIZE);
282 se->type = GET_SIT_TYPE(rs);
283 se->mtime = le64_to_cpu(rs->mtime);
284}
285
286static inline void seg_info_to_raw_sit(struct seg_entry *se,
287 struct f2fs_sit_entry *rs)
288{
289 unsigned short raw_vblocks = (se->type << SIT_VBLOCKS_SHIFT) |
290 se->valid_blocks;
291 rs->vblocks = cpu_to_le16(raw_vblocks);
292 memcpy(rs->valid_map, se->cur_valid_map, SIT_VBLOCK_MAP_SIZE);
293 memcpy(se->ckpt_valid_map, rs->valid_map, SIT_VBLOCK_MAP_SIZE);
294 se->ckpt_valid_blocks = se->valid_blocks;
295 rs->mtime = cpu_to_le64(se->mtime);
296}
297
298static inline unsigned int find_next_inuse(struct free_segmap_info *free_i,
299 unsigned int max, unsigned int segno)
300{
301 unsigned int ret;
302 read_lock(&free_i->segmap_lock);
303 ret = find_next_bit(free_i->free_segmap, max, segno);
304 read_unlock(&free_i->segmap_lock);
305 return ret;
306}
307
308static inline void __set_free(struct f2fs_sb_info *sbi, unsigned int segno)
309{
310 struct free_segmap_info *free_i = FREE_I(sbi);
311 unsigned int secno = segno / sbi->segs_per_sec;
312 unsigned int start_segno = secno * sbi->segs_per_sec;
313 unsigned int next;
314
315 write_lock(&free_i->segmap_lock);
316 clear_bit(segno, free_i->free_segmap);
317 free_i->free_segments++;
318
319 next = find_next_bit(free_i->free_segmap, TOTAL_SEGS(sbi), start_segno);
320 if (next >= start_segno + sbi->segs_per_sec) {
321 clear_bit(secno, free_i->free_secmap);
322 free_i->free_sections++;
323 }
324 write_unlock(&free_i->segmap_lock);
325}
326
327static inline void __set_inuse(struct f2fs_sb_info *sbi,
328 unsigned int segno)
329{
330 struct free_segmap_info *free_i = FREE_I(sbi);
331 unsigned int secno = segno / sbi->segs_per_sec;
332 set_bit(segno, free_i->free_segmap);
333 free_i->free_segments--;
334 if (!test_and_set_bit(secno, free_i->free_secmap))
335 free_i->free_sections--;
336}
337
338static inline void __set_test_and_free(struct f2fs_sb_info *sbi,
339 unsigned int segno)
340{
341 struct free_segmap_info *free_i = FREE_I(sbi);
342 unsigned int secno = segno / sbi->segs_per_sec;
343 unsigned int start_segno = secno * sbi->segs_per_sec;
344 unsigned int next;
345
346 write_lock(&free_i->segmap_lock);
347 if (test_and_clear_bit(segno, free_i->free_segmap)) {
348 free_i->free_segments++;
349
350 next = find_next_bit(free_i->free_segmap, TOTAL_SEGS(sbi),
351 start_segno);
352 if (next >= start_segno + sbi->segs_per_sec) {
353 if (test_and_clear_bit(secno, free_i->free_secmap))
354 free_i->free_sections++;
355 }
356 }
357 write_unlock(&free_i->segmap_lock);
358}
359
360static inline void __set_test_and_inuse(struct f2fs_sb_info *sbi,
361 unsigned int segno)
362{
363 struct free_segmap_info *free_i = FREE_I(sbi);
364 unsigned int secno = segno / sbi->segs_per_sec;
365 write_lock(&free_i->segmap_lock);
366 if (!test_and_set_bit(segno, free_i->free_segmap)) {
367 free_i->free_segments--;
368 if (!test_and_set_bit(secno, free_i->free_secmap))
369 free_i->free_sections--;
370 }
371 write_unlock(&free_i->segmap_lock);
372}
373
374static inline void get_sit_bitmap(struct f2fs_sb_info *sbi,
375 void *dst_addr)
376{
377 struct sit_info *sit_i = SIT_I(sbi);
378 memcpy(dst_addr, sit_i->sit_bitmap, sit_i->bitmap_size);
379}
380
381static inline block_t written_block_count(struct f2fs_sb_info *sbi)
382{
383 return SIT_I(sbi)->written_valid_blocks;
384}
385
386static inline unsigned int free_segments(struct f2fs_sb_info *sbi)
387{
388 return FREE_I(sbi)->free_segments;
389}
390
391static inline int reserved_segments(struct f2fs_sb_info *sbi)
392{
393 return SM_I(sbi)->reserved_segments;
394}
395
396static inline unsigned int free_sections(struct f2fs_sb_info *sbi)
397{
398 return FREE_I(sbi)->free_sections;
399}
400
401static inline unsigned int prefree_segments(struct f2fs_sb_info *sbi)
402{
403 return DIRTY_I(sbi)->nr_dirty[PRE];
404}
405
406static inline unsigned int dirty_segments(struct f2fs_sb_info *sbi)
407{
408 return DIRTY_I(sbi)->nr_dirty[DIRTY_HOT_DATA] +
409 DIRTY_I(sbi)->nr_dirty[DIRTY_WARM_DATA] +
410 DIRTY_I(sbi)->nr_dirty[DIRTY_COLD_DATA] +
411 DIRTY_I(sbi)->nr_dirty[DIRTY_HOT_NODE] +
412 DIRTY_I(sbi)->nr_dirty[DIRTY_WARM_NODE] +
413 DIRTY_I(sbi)->nr_dirty[DIRTY_COLD_NODE];
414}
415
416static inline int overprovision_segments(struct f2fs_sb_info *sbi)
417{
418 return SM_I(sbi)->ovp_segments;
419}
420
421static inline int overprovision_sections(struct f2fs_sb_info *sbi)
422{
423 return ((unsigned int) overprovision_segments(sbi)) / sbi->segs_per_sec;
424}
425
426static inline int reserved_sections(struct f2fs_sb_info *sbi)
427{
428 return ((unsigned int) reserved_segments(sbi)) / sbi->segs_per_sec;
429}
430
431static inline bool need_SSR(struct f2fs_sb_info *sbi)
432{
433 return (prefree_segments(sbi) / sbi->segs_per_sec)
434 + free_sections(sbi) < overprovision_sections(sbi);
435}
436
437static inline bool has_not_enough_free_secs(struct f2fs_sb_info *sbi, int freed)
438{
439 int node_secs = get_blocktype_secs(sbi, F2FS_DIRTY_NODES);
440 int dent_secs = get_blocktype_secs(sbi, F2FS_DIRTY_DENTS);
441
442 if (unlikely(sbi->por_doing))
443 return false;
444
445 return (free_sections(sbi) + freed) <= (node_secs + 2 * dent_secs +
446 reserved_sections(sbi));
447}
448
449static inline bool excess_prefree_segs(struct f2fs_sb_info *sbi)
450{
451 return prefree_segments(sbi) > SM_I(sbi)->rec_prefree_segments;
452}
453
454static inline int utilization(struct f2fs_sb_info *sbi)
455{
456 return div_u64((u64)valid_user_blocks(sbi) * 100,
457 sbi->user_block_count);
458}
459
460/*
461 * Sometimes f2fs may be better to drop out-of-place update policy.
462 * And, users can control the policy through sysfs entries.
463 * There are five policies with triggering conditions as follows.
464 * F2FS_IPU_FORCE - all the time,
465 * F2FS_IPU_SSR - if SSR mode is activated,
466 * F2FS_IPU_UTIL - if FS utilization is over threashold,
467 * F2FS_IPU_SSR_UTIL - if SSR mode is activated and FS utilization is over
468 * threashold,
469 * F2FS_IPUT_DISABLE - disable IPU. (=default option)
470 */
471#define DEF_MIN_IPU_UTIL 70
472
473enum {
474 F2FS_IPU_FORCE,
475 F2FS_IPU_SSR,
476 F2FS_IPU_UTIL,
477 F2FS_IPU_SSR_UTIL,
478 F2FS_IPU_DISABLE,
479};
480
481static inline bool need_inplace_update(struct inode *inode)
482{
483 struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
484
485 /* IPU can be done only for the user data */
486 if (S_ISDIR(inode->i_mode))
487 return false;
488
489 switch (SM_I(sbi)->ipu_policy) {
490 case F2FS_IPU_FORCE:
491 return true;
492 case F2FS_IPU_SSR:
493 if (need_SSR(sbi))
494 return true;
495 break;
496 case F2FS_IPU_UTIL:
497 if (utilization(sbi) > SM_I(sbi)->min_ipu_util)
498 return true;
499 break;
500 case F2FS_IPU_SSR_UTIL:
501 if (need_SSR(sbi) && utilization(sbi) > SM_I(sbi)->min_ipu_util)
502 return true;
503 break;
504 case F2FS_IPU_DISABLE:
505 break;
506 }
507 return false;
508}
509
510static inline unsigned int curseg_segno(struct f2fs_sb_info *sbi,
511 int type)
512{
513 struct curseg_info *curseg = CURSEG_I(sbi, type);
514 return curseg->segno;
515}
516
517static inline unsigned char curseg_alloc_type(struct f2fs_sb_info *sbi,
518 int type)
519{
520 struct curseg_info *curseg = CURSEG_I(sbi, type);
521 return curseg->alloc_type;
522}
523
524static inline unsigned short curseg_blkoff(struct f2fs_sb_info *sbi, int type)
525{
526 struct curseg_info *curseg = CURSEG_I(sbi, type);
527 return curseg->next_blkoff;
528}
529
530#ifdef CONFIG_F2FS_CHECK_FS
531static inline void check_seg_range(struct f2fs_sb_info *sbi, unsigned int segno)
532{
533 unsigned int end_segno = SM_I(sbi)->segment_count - 1;
534 BUG_ON(segno > end_segno);
535}
536
537static inline void verify_block_addr(struct f2fs_sb_info *sbi, block_t blk_addr)
538{
539 struct f2fs_sm_info *sm_info = SM_I(sbi);
540 block_t total_blks = sm_info->segment_count << sbi->log_blocks_per_seg;
541 block_t start_addr = sm_info->seg0_blkaddr;
542 block_t end_addr = start_addr + total_blks - 1;
543 BUG_ON(blk_addr < start_addr);
544 BUG_ON(blk_addr > end_addr);
545}
546
547/*
548 * Summary block is always treated as invalid block
549 */
550static inline void check_block_count(struct f2fs_sb_info *sbi,
551 int segno, struct f2fs_sit_entry *raw_sit)
552{
553 struct f2fs_sm_info *sm_info = SM_I(sbi);
554 unsigned int end_segno = sm_info->segment_count - 1;
555 bool is_valid = test_bit_le(0, raw_sit->valid_map) ? true : false;
556 int valid_blocks = 0;
557 int cur_pos = 0, next_pos;
558
559 /* check segment usage */
560 BUG_ON(GET_SIT_VBLOCKS(raw_sit) > sbi->blocks_per_seg);
561
562 /* check boundary of a given segment number */
563 BUG_ON(segno > end_segno);
564
565 /* check bitmap with valid block count */
566 do {
567 if (is_valid) {
568 next_pos = find_next_zero_bit_le(&raw_sit->valid_map,
569 sbi->blocks_per_seg,
570 cur_pos);
571 valid_blocks += next_pos - cur_pos;
572 } else
573 next_pos = find_next_bit_le(&raw_sit->valid_map,
574 sbi->blocks_per_seg,
575 cur_pos);
576 cur_pos = next_pos;
577 is_valid = !is_valid;
578 } while (cur_pos < sbi->blocks_per_seg);
579 BUG_ON(GET_SIT_VBLOCKS(raw_sit) != valid_blocks);
580}
581#else
582#define check_seg_range(sbi, segno)
583#define verify_block_addr(sbi, blk_addr)
584#define check_block_count(sbi, segno, raw_sit)
585#endif
586
587static inline pgoff_t current_sit_addr(struct f2fs_sb_info *sbi,
588 unsigned int start)
589{
590 struct sit_info *sit_i = SIT_I(sbi);
591 unsigned int offset = SIT_BLOCK_OFFSET(sit_i, start);
592 block_t blk_addr = sit_i->sit_base_addr + offset;
593
594 check_seg_range(sbi, start);
595
596 /* calculate sit block address */
597 if (f2fs_test_bit(offset, sit_i->sit_bitmap))
598 blk_addr += sit_i->sit_blocks;
599
600 return blk_addr;
601}
602
603static inline pgoff_t next_sit_addr(struct f2fs_sb_info *sbi,
604 pgoff_t block_addr)
605{
606 struct sit_info *sit_i = SIT_I(sbi);
607 block_addr -= sit_i->sit_base_addr;
608 if (block_addr < sit_i->sit_blocks)
609 block_addr += sit_i->sit_blocks;
610 else
611 block_addr -= sit_i->sit_blocks;
612
613 return block_addr + sit_i->sit_base_addr;
614}
615
616static inline void set_to_next_sit(struct sit_info *sit_i, unsigned int start)
617{
618 unsigned int block_off = SIT_BLOCK_OFFSET(sit_i, start);
619
620 if (f2fs_test_bit(block_off, sit_i->sit_bitmap))
621 f2fs_clear_bit(block_off, sit_i->sit_bitmap);
622 else
623 f2fs_set_bit(block_off, sit_i->sit_bitmap);
624}
625
626static inline unsigned long long get_mtime(struct f2fs_sb_info *sbi)
627{
628 struct sit_info *sit_i = SIT_I(sbi);
629 return sit_i->elapsed_time + CURRENT_TIME_SEC.tv_sec -
630 sit_i->mounted_time;
631}
632
633static inline void set_summary(struct f2fs_summary *sum, nid_t nid,
634 unsigned int ofs_in_node, unsigned char version)
635{
636 sum->nid = cpu_to_le32(nid);
637 sum->ofs_in_node = cpu_to_le16(ofs_in_node);
638 sum->version = version;
639}
640
641static inline block_t start_sum_block(struct f2fs_sb_info *sbi)
642{
643 return __start_cp_addr(sbi) +
644 le32_to_cpu(F2FS_CKPT(sbi)->cp_pack_start_sum);
645}
646
647static inline block_t sum_blk_addr(struct f2fs_sb_info *sbi, int base, int type)
648{
649 return __start_cp_addr(sbi) +
650 le32_to_cpu(F2FS_CKPT(sbi)->cp_pack_total_block_count)
651 - (base + 1) + type;
652}
653
654static inline bool sec_usage_check(struct f2fs_sb_info *sbi, unsigned int secno)
655{
656 if (IS_CURSEC(sbi, secno) || (sbi->cur_victim_sec == secno))
657 return true;
658 return false;
659}
660
661static inline unsigned int max_hw_blocks(struct f2fs_sb_info *sbi)
662{
663 struct block_device *bdev = sbi->sb->s_bdev;
664 struct request_queue *q = bdev_get_queue(bdev);
665 return SECTOR_TO_BLOCK(sbi, queue_max_sectors(q));
666}
667
668/*
669 * It is very important to gather dirty pages and write at once, so that we can
670 * submit a big bio without interfering other data writes.
671 * By default, 512 pages for directory data,
672 * 512 pages (2MB) * 3 for three types of nodes, and
673 * max_bio_blocks for meta are set.
674 */
675static inline int nr_pages_to_skip(struct f2fs_sb_info *sbi, int type)
676{
677 if (type == DATA)
678 return sbi->blocks_per_seg;
679 else if (type == NODE)
680 return 3 * sbi->blocks_per_seg;
681 else if (type == META)
682 return MAX_BIO_BLOCKS(max_hw_blocks(sbi));
683 else
684 return 0;
685}
686
687/*
688 * When writing pages, it'd better align nr_to_write for segment size.
689 */
690static inline long nr_pages_to_write(struct f2fs_sb_info *sbi, int type,
691 struct writeback_control *wbc)
692{
693 long nr_to_write, desired;
694
695 if (wbc->sync_mode != WB_SYNC_NONE)
696 return 0;
697
698 nr_to_write = wbc->nr_to_write;
699
700 if (type == DATA)
701 desired = 4096;
702 else if (type == NODE)
703 desired = 3 * max_hw_blocks(sbi);
704 else
705 desired = MAX_BIO_BLOCKS(max_hw_blocks(sbi));
706
707 wbc->nr_to_write = desired;
708 return desired - nr_to_write;
709}
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}