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/*
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#define DEF_MAX_RECLAIM_PREFREE_SEGMENTS 4096 /* 8GB in maximum */
20
21#define F2FS_MIN_SEGMENTS 9 /* SB + 2 (CP + SIT + NAT) + SSA + MAIN */
22
23/* L: Logical segment # in volume, R: Relative segment # in main area */
24#define GET_L2R_SEGNO(free_i, segno) (segno - free_i->start_segno)
25#define GET_R2L_SEGNO(free_i, segno) (segno + free_i->start_segno)
26
27#define IS_DATASEG(t) (t <= CURSEG_COLD_DATA)
28#define IS_NODESEG(t) (t >= CURSEG_HOT_NODE)
29
30#define IS_CURSEG(sbi, seg) \
31 ((seg == CURSEG_I(sbi, CURSEG_HOT_DATA)->segno) || \
32 (seg == CURSEG_I(sbi, CURSEG_WARM_DATA)->segno) || \
33 (seg == CURSEG_I(sbi, CURSEG_COLD_DATA)->segno) || \
34 (seg == CURSEG_I(sbi, CURSEG_HOT_NODE)->segno) || \
35 (seg == CURSEG_I(sbi, CURSEG_WARM_NODE)->segno) || \
36 (seg == CURSEG_I(sbi, CURSEG_COLD_NODE)->segno))
37
38#define IS_CURSEC(sbi, secno) \
39 ((secno == CURSEG_I(sbi, CURSEG_HOT_DATA)->segno / \
40 sbi->segs_per_sec) || \
41 (secno == CURSEG_I(sbi, CURSEG_WARM_DATA)->segno / \
42 sbi->segs_per_sec) || \
43 (secno == CURSEG_I(sbi, CURSEG_COLD_DATA)->segno / \
44 sbi->segs_per_sec) || \
45 (secno == CURSEG_I(sbi, CURSEG_HOT_NODE)->segno / \
46 sbi->segs_per_sec) || \
47 (secno == CURSEG_I(sbi, CURSEG_WARM_NODE)->segno / \
48 sbi->segs_per_sec) || \
49 (secno == CURSEG_I(sbi, CURSEG_COLD_NODE)->segno / \
50 sbi->segs_per_sec)) \
51
52#define MAIN_BLKADDR(sbi) (SM_I(sbi)->main_blkaddr)
53#define SEG0_BLKADDR(sbi) (SM_I(sbi)->seg0_blkaddr)
54
55#define MAIN_SEGS(sbi) (SM_I(sbi)->main_segments)
56#define MAIN_SECS(sbi) (sbi->total_sections)
57
58#define TOTAL_SEGS(sbi) (SM_I(sbi)->segment_count)
59#define TOTAL_BLKS(sbi) (TOTAL_SEGS(sbi) << sbi->log_blocks_per_seg)
60
61#define MAX_BLKADDR(sbi) (SEG0_BLKADDR(sbi) + TOTAL_BLKS(sbi))
62#define SEGMENT_SIZE(sbi) (1ULL << (sbi->log_blocksize + \
63 sbi->log_blocks_per_seg))
64
65#define START_BLOCK(sbi, segno) (SEG0_BLKADDR(sbi) + \
66 (GET_R2L_SEGNO(FREE_I(sbi), segno) << sbi->log_blocks_per_seg))
67
68#define NEXT_FREE_BLKADDR(sbi, curseg) \
69 (START_BLOCK(sbi, curseg->segno) + curseg->next_blkoff)
70
71#define GET_SEGOFF_FROM_SEG0(sbi, blk_addr) ((blk_addr) - SEG0_BLKADDR(sbi))
72#define GET_SEGNO_FROM_SEG0(sbi, blk_addr) \
73 (GET_SEGOFF_FROM_SEG0(sbi, blk_addr) >> sbi->log_blocks_per_seg)
74#define GET_BLKOFF_FROM_SEG0(sbi, blk_addr) \
75 (GET_SEGOFF_FROM_SEG0(sbi, blk_addr) & (sbi->blocks_per_seg - 1))
76
77#define GET_SEGNO(sbi, blk_addr) \
78 (((blk_addr == NULL_ADDR) || (blk_addr == NEW_ADDR)) ? \
79 NULL_SEGNO : GET_L2R_SEGNO(FREE_I(sbi), \
80 GET_SEGNO_FROM_SEG0(sbi, blk_addr)))
81#define GET_SECNO(sbi, segno) \
82 ((segno) / sbi->segs_per_sec)
83#define GET_ZONENO_FROM_SEGNO(sbi, segno) \
84 ((segno / sbi->segs_per_sec) / sbi->secs_per_zone)
85
86#define GET_SUM_BLOCK(sbi, segno) \
87 ((sbi->sm_info->ssa_blkaddr) + segno)
88
89#define GET_SUM_TYPE(footer) ((footer)->entry_type)
90#define SET_SUM_TYPE(footer, type) ((footer)->entry_type = type)
91
92#define SIT_ENTRY_OFFSET(sit_i, segno) \
93 (segno % sit_i->sents_per_block)
94#define SIT_BLOCK_OFFSET(segno) \
95 (segno / SIT_ENTRY_PER_BLOCK)
96#define START_SEGNO(segno) \
97 (SIT_BLOCK_OFFSET(segno) * SIT_ENTRY_PER_BLOCK)
98#define SIT_BLK_CNT(sbi) \
99 ((MAIN_SEGS(sbi) + SIT_ENTRY_PER_BLOCK - 1) / SIT_ENTRY_PER_BLOCK)
100#define f2fs_bitmap_size(nr) \
101 (BITS_TO_LONGS(nr) * sizeof(unsigned long))
102
103#define SECTOR_FROM_BLOCK(blk_addr) \
104 (((sector_t)blk_addr) << F2FS_LOG_SECTORS_PER_BLOCK)
105#define SECTOR_TO_BLOCK(sectors) \
106 (sectors >> F2FS_LOG_SECTORS_PER_BLOCK)
107
108/*
109 * indicate a block allocation direction: RIGHT and LEFT.
110 * RIGHT means allocating new sections towards the end of volume.
111 * LEFT means the opposite direction.
112 */
113enum {
114 ALLOC_RIGHT = 0,
115 ALLOC_LEFT
116};
117
118/*
119 * In the victim_sel_policy->alloc_mode, there are two block allocation modes.
120 * LFS writes data sequentially with cleaning operations.
121 * SSR (Slack Space Recycle) reuses obsolete space without cleaning operations.
122 */
123enum {
124 LFS = 0,
125 SSR
126};
127
128/*
129 * In the victim_sel_policy->gc_mode, there are two gc, aka cleaning, modes.
130 * GC_CB is based on cost-benefit algorithm.
131 * GC_GREEDY is based on greedy algorithm.
132 */
133enum {
134 GC_CB = 0,
135 GC_GREEDY
136};
137
138/*
139 * BG_GC means the background cleaning job.
140 * FG_GC means the on-demand cleaning job.
141 * FORCE_FG_GC means on-demand cleaning job in background.
142 */
143enum {
144 BG_GC = 0,
145 FG_GC,
146 FORCE_FG_GC,
147};
148
149/* for a function parameter to select a victim segment */
150struct victim_sel_policy {
151 int alloc_mode; /* LFS or SSR */
152 int gc_mode; /* GC_CB or GC_GREEDY */
153 unsigned long *dirty_segmap; /* dirty segment bitmap */
154 unsigned int max_search; /* maximum # of segments to search */
155 unsigned int offset; /* last scanned bitmap offset */
156 unsigned int ofs_unit; /* bitmap search unit */
157 unsigned int min_cost; /* minimum cost */
158 unsigned int min_segno; /* segment # having min. cost */
159};
160
161struct seg_entry {
162 unsigned int type:6; /* segment type like CURSEG_XXX_TYPE */
163 unsigned int valid_blocks:10; /* # of valid blocks */
164 unsigned int ckpt_valid_blocks:10; /* # of valid blocks last cp */
165 unsigned int padding:6; /* padding */
166 unsigned char *cur_valid_map; /* validity bitmap of blocks */
167 /*
168 * # of valid blocks and the validity bitmap stored in the the last
169 * checkpoint pack. This information is used by the SSR mode.
170 */
171 unsigned char *ckpt_valid_map; /* validity bitmap of blocks last cp */
172 unsigned char *discard_map;
173 unsigned long long mtime; /* modification time of the segment */
174};
175
176struct sec_entry {
177 unsigned int valid_blocks; /* # of valid blocks in a section */
178};
179
180struct segment_allocation {
181 void (*allocate_segment)(struct f2fs_sb_info *, int, bool);
182};
183
184/*
185 * this value is set in page as a private data which indicate that
186 * the page is atomically written, and it is in inmem_pages list.
187 */
188#define ATOMIC_WRITTEN_PAGE ((unsigned long)-1)
189
190#define IS_ATOMIC_WRITTEN_PAGE(page) \
191 (page_private(page) == (unsigned long)ATOMIC_WRITTEN_PAGE)
192
193struct inmem_pages {
194 struct list_head list;
195 struct page *page;
196 block_t old_addr; /* for revoking when fail to commit */
197};
198
199struct sit_info {
200 const struct segment_allocation *s_ops;
201
202 block_t sit_base_addr; /* start block address of SIT area */
203 block_t sit_blocks; /* # of blocks used by SIT area */
204 block_t written_valid_blocks; /* # of valid blocks in main area */
205 char *sit_bitmap; /* SIT bitmap pointer */
206 unsigned int bitmap_size; /* SIT bitmap size */
207
208 unsigned long *tmp_map; /* bitmap for temporal use */
209 unsigned long *dirty_sentries_bitmap; /* bitmap for dirty sentries */
210 unsigned int dirty_sentries; /* # of dirty sentries */
211 unsigned int sents_per_block; /* # of SIT entries per block */
212 struct mutex sentry_lock; /* to protect SIT cache */
213 struct seg_entry *sentries; /* SIT segment-level cache */
214 struct sec_entry *sec_entries; /* SIT section-level cache */
215
216 /* for cost-benefit algorithm in cleaning procedure */
217 unsigned long long elapsed_time; /* elapsed time after mount */
218 unsigned long long mounted_time; /* mount time */
219 unsigned long long min_mtime; /* min. modification time */
220 unsigned long long max_mtime; /* max. modification time */
221};
222
223struct free_segmap_info {
224 unsigned int start_segno; /* start segment number logically */
225 unsigned int free_segments; /* # of free segments */
226 unsigned int free_sections; /* # of free sections */
227 spinlock_t segmap_lock; /* free segmap lock */
228 unsigned long *free_segmap; /* free segment bitmap */
229 unsigned long *free_secmap; /* free section bitmap */
230};
231
232/* Notice: The order of dirty type is same with CURSEG_XXX in f2fs.h */
233enum dirty_type {
234 DIRTY_HOT_DATA, /* dirty segments assigned as hot data logs */
235 DIRTY_WARM_DATA, /* dirty segments assigned as warm data logs */
236 DIRTY_COLD_DATA, /* dirty segments assigned as cold data logs */
237 DIRTY_HOT_NODE, /* dirty segments assigned as hot node logs */
238 DIRTY_WARM_NODE, /* dirty segments assigned as warm node logs */
239 DIRTY_COLD_NODE, /* dirty segments assigned as cold node logs */
240 DIRTY, /* to count # of dirty segments */
241 PRE, /* to count # of entirely obsolete segments */
242 NR_DIRTY_TYPE
243};
244
245struct dirty_seglist_info {
246 const struct victim_selection *v_ops; /* victim selction operation */
247 unsigned long *dirty_segmap[NR_DIRTY_TYPE];
248 struct mutex seglist_lock; /* lock for segment bitmaps */
249 int nr_dirty[NR_DIRTY_TYPE]; /* # of dirty segments */
250 unsigned long *victim_secmap; /* background GC victims */
251};
252
253/* victim selection function for cleaning and SSR */
254struct victim_selection {
255 int (*get_victim)(struct f2fs_sb_info *, unsigned int *,
256 int, int, char);
257};
258
259/* for active log information */
260struct curseg_info {
261 struct mutex curseg_mutex; /* lock for consistency */
262 struct f2fs_summary_block *sum_blk; /* cached summary block */
263 struct rw_semaphore journal_rwsem; /* protect journal area */
264 struct f2fs_journal *journal; /* cached journal info */
265 unsigned char alloc_type; /* current allocation type */
266 unsigned int segno; /* current segment number */
267 unsigned short next_blkoff; /* next block offset to write */
268 unsigned int zone; /* current zone number */
269 unsigned int next_segno; /* preallocated segment */
270};
271
272struct sit_entry_set {
273 struct list_head set_list; /* link with all sit sets */
274 unsigned int start_segno; /* start segno of sits in set */
275 unsigned int entry_cnt; /* the # of sit entries in set */
276};
277
278/*
279 * inline functions
280 */
281static inline struct curseg_info *CURSEG_I(struct f2fs_sb_info *sbi, int type)
282{
283 return (struct curseg_info *)(SM_I(sbi)->curseg_array + type);
284}
285
286static inline struct seg_entry *get_seg_entry(struct f2fs_sb_info *sbi,
287 unsigned int segno)
288{
289 struct sit_info *sit_i = SIT_I(sbi);
290 return &sit_i->sentries[segno];
291}
292
293static inline struct sec_entry *get_sec_entry(struct f2fs_sb_info *sbi,
294 unsigned int segno)
295{
296 struct sit_info *sit_i = SIT_I(sbi);
297 return &sit_i->sec_entries[GET_SECNO(sbi, segno)];
298}
299
300static inline unsigned int get_valid_blocks(struct f2fs_sb_info *sbi,
301 unsigned int segno, int section)
302{
303 /*
304 * In order to get # of valid blocks in a section instantly from many
305 * segments, f2fs manages two counting structures separately.
306 */
307 if (section > 1)
308 return get_sec_entry(sbi, segno)->valid_blocks;
309 else
310 return get_seg_entry(sbi, segno)->valid_blocks;
311}
312
313static inline void seg_info_from_raw_sit(struct seg_entry *se,
314 struct f2fs_sit_entry *rs)
315{
316 se->valid_blocks = GET_SIT_VBLOCKS(rs);
317 se->ckpt_valid_blocks = GET_SIT_VBLOCKS(rs);
318 memcpy(se->cur_valid_map, rs->valid_map, SIT_VBLOCK_MAP_SIZE);
319 memcpy(se->ckpt_valid_map, rs->valid_map, SIT_VBLOCK_MAP_SIZE);
320 se->type = GET_SIT_TYPE(rs);
321 se->mtime = le64_to_cpu(rs->mtime);
322}
323
324static inline void seg_info_to_raw_sit(struct seg_entry *se,
325 struct f2fs_sit_entry *rs)
326{
327 unsigned short raw_vblocks = (se->type << SIT_VBLOCKS_SHIFT) |
328 se->valid_blocks;
329 rs->vblocks = cpu_to_le16(raw_vblocks);
330 memcpy(rs->valid_map, se->cur_valid_map, SIT_VBLOCK_MAP_SIZE);
331 memcpy(se->ckpt_valid_map, rs->valid_map, SIT_VBLOCK_MAP_SIZE);
332 se->ckpt_valid_blocks = se->valid_blocks;
333 rs->mtime = cpu_to_le64(se->mtime);
334}
335
336static inline unsigned int find_next_inuse(struct free_segmap_info *free_i,
337 unsigned int max, unsigned int segno)
338{
339 unsigned int ret;
340 spin_lock(&free_i->segmap_lock);
341 ret = find_next_bit(free_i->free_segmap, max, segno);
342 spin_unlock(&free_i->segmap_lock);
343 return ret;
344}
345
346static inline void __set_free(struct f2fs_sb_info *sbi, unsigned int segno)
347{
348 struct free_segmap_info *free_i = FREE_I(sbi);
349 unsigned int secno = segno / sbi->segs_per_sec;
350 unsigned int start_segno = secno * sbi->segs_per_sec;
351 unsigned int next;
352
353 spin_lock(&free_i->segmap_lock);
354 clear_bit(segno, free_i->free_segmap);
355 free_i->free_segments++;
356
357 next = find_next_bit(free_i->free_segmap,
358 start_segno + sbi->segs_per_sec, start_segno);
359 if (next >= start_segno + sbi->segs_per_sec) {
360 clear_bit(secno, free_i->free_secmap);
361 free_i->free_sections++;
362 }
363 spin_unlock(&free_i->segmap_lock);
364}
365
366static inline void __set_inuse(struct f2fs_sb_info *sbi,
367 unsigned int segno)
368{
369 struct free_segmap_info *free_i = FREE_I(sbi);
370 unsigned int secno = segno / sbi->segs_per_sec;
371 set_bit(segno, free_i->free_segmap);
372 free_i->free_segments--;
373 if (!test_and_set_bit(secno, free_i->free_secmap))
374 free_i->free_sections--;
375}
376
377static inline void __set_test_and_free(struct f2fs_sb_info *sbi,
378 unsigned int segno)
379{
380 struct free_segmap_info *free_i = FREE_I(sbi);
381 unsigned int secno = segno / sbi->segs_per_sec;
382 unsigned int start_segno = secno * sbi->segs_per_sec;
383 unsigned int next;
384
385 spin_lock(&free_i->segmap_lock);
386 if (test_and_clear_bit(segno, free_i->free_segmap)) {
387 free_i->free_segments++;
388
389 next = find_next_bit(free_i->free_segmap,
390 start_segno + sbi->segs_per_sec, start_segno);
391 if (next >= start_segno + sbi->segs_per_sec) {
392 if (test_and_clear_bit(secno, free_i->free_secmap))
393 free_i->free_sections++;
394 }
395 }
396 spin_unlock(&free_i->segmap_lock);
397}
398
399static inline void __set_test_and_inuse(struct f2fs_sb_info *sbi,
400 unsigned int segno)
401{
402 struct free_segmap_info *free_i = FREE_I(sbi);
403 unsigned int secno = segno / sbi->segs_per_sec;
404 spin_lock(&free_i->segmap_lock);
405 if (!test_and_set_bit(segno, free_i->free_segmap)) {
406 free_i->free_segments--;
407 if (!test_and_set_bit(secno, free_i->free_secmap))
408 free_i->free_sections--;
409 }
410 spin_unlock(&free_i->segmap_lock);
411}
412
413static inline void get_sit_bitmap(struct f2fs_sb_info *sbi,
414 void *dst_addr)
415{
416 struct sit_info *sit_i = SIT_I(sbi);
417 memcpy(dst_addr, sit_i->sit_bitmap, sit_i->bitmap_size);
418}
419
420static inline block_t written_block_count(struct f2fs_sb_info *sbi)
421{
422 return SIT_I(sbi)->written_valid_blocks;
423}
424
425static inline unsigned int free_segments(struct f2fs_sb_info *sbi)
426{
427 return FREE_I(sbi)->free_segments;
428}
429
430static inline int reserved_segments(struct f2fs_sb_info *sbi)
431{
432 return SM_I(sbi)->reserved_segments;
433}
434
435static inline unsigned int free_sections(struct f2fs_sb_info *sbi)
436{
437 return FREE_I(sbi)->free_sections;
438}
439
440static inline unsigned int prefree_segments(struct f2fs_sb_info *sbi)
441{
442 return DIRTY_I(sbi)->nr_dirty[PRE];
443}
444
445static inline unsigned int dirty_segments(struct f2fs_sb_info *sbi)
446{
447 return DIRTY_I(sbi)->nr_dirty[DIRTY_HOT_DATA] +
448 DIRTY_I(sbi)->nr_dirty[DIRTY_WARM_DATA] +
449 DIRTY_I(sbi)->nr_dirty[DIRTY_COLD_DATA] +
450 DIRTY_I(sbi)->nr_dirty[DIRTY_HOT_NODE] +
451 DIRTY_I(sbi)->nr_dirty[DIRTY_WARM_NODE] +
452 DIRTY_I(sbi)->nr_dirty[DIRTY_COLD_NODE];
453}
454
455static inline int overprovision_segments(struct f2fs_sb_info *sbi)
456{
457 return SM_I(sbi)->ovp_segments;
458}
459
460static inline int overprovision_sections(struct f2fs_sb_info *sbi)
461{
462 return ((unsigned int) overprovision_segments(sbi)) / sbi->segs_per_sec;
463}
464
465static inline int reserved_sections(struct f2fs_sb_info *sbi)
466{
467 return ((unsigned int) reserved_segments(sbi)) / sbi->segs_per_sec;
468}
469
470static inline bool need_SSR(struct f2fs_sb_info *sbi)
471{
472 int node_secs = get_blocktype_secs(sbi, F2FS_DIRTY_NODES);
473 int dent_secs = get_blocktype_secs(sbi, F2FS_DIRTY_DENTS);
474 int imeta_secs = get_blocktype_secs(sbi, F2FS_DIRTY_IMETA);
475
476 if (test_opt(sbi, LFS))
477 return false;
478
479 return free_sections(sbi) <= (node_secs + 2 * dent_secs + imeta_secs +
480 reserved_sections(sbi) + 1);
481}
482
483static inline bool has_not_enough_free_secs(struct f2fs_sb_info *sbi,
484 int freed, int needed)
485{
486 int node_secs = get_blocktype_secs(sbi, F2FS_DIRTY_NODES);
487 int dent_secs = get_blocktype_secs(sbi, F2FS_DIRTY_DENTS);
488 int imeta_secs = get_blocktype_secs(sbi, F2FS_DIRTY_IMETA);
489
490 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
491 return false;
492
493 return (free_sections(sbi) + freed) <=
494 (node_secs + 2 * dent_secs + imeta_secs +
495 reserved_sections(sbi) + needed);
496}
497
498static inline bool excess_prefree_segs(struct f2fs_sb_info *sbi)
499{
500 return prefree_segments(sbi) > SM_I(sbi)->rec_prefree_segments;
501}
502
503static inline int utilization(struct f2fs_sb_info *sbi)
504{
505 return div_u64((u64)valid_user_blocks(sbi) * 100,
506 sbi->user_block_count);
507}
508
509/*
510 * Sometimes f2fs may be better to drop out-of-place update policy.
511 * And, users can control the policy through sysfs entries.
512 * There are five policies with triggering conditions as follows.
513 * F2FS_IPU_FORCE - all the time,
514 * F2FS_IPU_SSR - if SSR mode is activated,
515 * F2FS_IPU_UTIL - if FS utilization is over threashold,
516 * F2FS_IPU_SSR_UTIL - if SSR mode is activated and FS utilization is over
517 * threashold,
518 * F2FS_IPU_FSYNC - activated in fsync path only for high performance flash
519 * storages. IPU will be triggered only if the # of dirty
520 * pages over min_fsync_blocks.
521 * F2FS_IPUT_DISABLE - disable IPU. (=default option)
522 */
523#define DEF_MIN_IPU_UTIL 70
524#define DEF_MIN_FSYNC_BLOCKS 8
525
526enum {
527 F2FS_IPU_FORCE,
528 F2FS_IPU_SSR,
529 F2FS_IPU_UTIL,
530 F2FS_IPU_SSR_UTIL,
531 F2FS_IPU_FSYNC,
532};
533
534static inline bool need_inplace_update(struct inode *inode)
535{
536 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
537 unsigned int policy = SM_I(sbi)->ipu_policy;
538
539 /* IPU can be done only for the user data */
540 if (S_ISDIR(inode->i_mode) || f2fs_is_atomic_file(inode))
541 return false;
542
543 if (test_opt(sbi, LFS))
544 return false;
545
546 if (policy & (0x1 << F2FS_IPU_FORCE))
547 return true;
548 if (policy & (0x1 << F2FS_IPU_SSR) && need_SSR(sbi))
549 return true;
550 if (policy & (0x1 << F2FS_IPU_UTIL) &&
551 utilization(sbi) > SM_I(sbi)->min_ipu_util)
552 return true;
553 if (policy & (0x1 << F2FS_IPU_SSR_UTIL) && need_SSR(sbi) &&
554 utilization(sbi) > SM_I(sbi)->min_ipu_util)
555 return true;
556
557 /* this is only set during fdatasync */
558 if (policy & (0x1 << F2FS_IPU_FSYNC) &&
559 is_inode_flag_set(inode, FI_NEED_IPU))
560 return true;
561
562 return false;
563}
564
565static inline unsigned int curseg_segno(struct f2fs_sb_info *sbi,
566 int type)
567{
568 struct curseg_info *curseg = CURSEG_I(sbi, type);
569 return curseg->segno;
570}
571
572static inline unsigned char curseg_alloc_type(struct f2fs_sb_info *sbi,
573 int type)
574{
575 struct curseg_info *curseg = CURSEG_I(sbi, type);
576 return curseg->alloc_type;
577}
578
579static inline unsigned short curseg_blkoff(struct f2fs_sb_info *sbi, int type)
580{
581 struct curseg_info *curseg = CURSEG_I(sbi, type);
582 return curseg->next_blkoff;
583}
584
585static inline void check_seg_range(struct f2fs_sb_info *sbi, unsigned int segno)
586{
587 f2fs_bug_on(sbi, segno > TOTAL_SEGS(sbi) - 1);
588}
589
590static inline void verify_block_addr(struct f2fs_sb_info *sbi, block_t blk_addr)
591{
592 BUG_ON(blk_addr < SEG0_BLKADDR(sbi)
593 || blk_addr >= MAX_BLKADDR(sbi));
594}
595
596/*
597 * Summary block is always treated as an invalid block
598 */
599static inline void check_block_count(struct f2fs_sb_info *sbi,
600 int segno, struct f2fs_sit_entry *raw_sit)
601{
602#ifdef CONFIG_F2FS_CHECK_FS
603 bool is_valid = test_bit_le(0, raw_sit->valid_map) ? true : false;
604 int valid_blocks = 0;
605 int cur_pos = 0, next_pos;
606
607 /* check bitmap with valid block count */
608 do {
609 if (is_valid) {
610 next_pos = find_next_zero_bit_le(&raw_sit->valid_map,
611 sbi->blocks_per_seg,
612 cur_pos);
613 valid_blocks += next_pos - cur_pos;
614 } else
615 next_pos = find_next_bit_le(&raw_sit->valid_map,
616 sbi->blocks_per_seg,
617 cur_pos);
618 cur_pos = next_pos;
619 is_valid = !is_valid;
620 } while (cur_pos < sbi->blocks_per_seg);
621 BUG_ON(GET_SIT_VBLOCKS(raw_sit) != valid_blocks);
622#endif
623 /* check segment usage, and check boundary of a given segment number */
624 f2fs_bug_on(sbi, GET_SIT_VBLOCKS(raw_sit) > sbi->blocks_per_seg
625 || segno > TOTAL_SEGS(sbi) - 1);
626}
627
628static inline pgoff_t current_sit_addr(struct f2fs_sb_info *sbi,
629 unsigned int start)
630{
631 struct sit_info *sit_i = SIT_I(sbi);
632 unsigned int offset = SIT_BLOCK_OFFSET(start);
633 block_t blk_addr = sit_i->sit_base_addr + offset;
634
635 check_seg_range(sbi, start);
636
637 /* calculate sit block address */
638 if (f2fs_test_bit(offset, sit_i->sit_bitmap))
639 blk_addr += sit_i->sit_blocks;
640
641 return blk_addr;
642}
643
644static inline pgoff_t next_sit_addr(struct f2fs_sb_info *sbi,
645 pgoff_t block_addr)
646{
647 struct sit_info *sit_i = SIT_I(sbi);
648 block_addr -= sit_i->sit_base_addr;
649 if (block_addr < sit_i->sit_blocks)
650 block_addr += sit_i->sit_blocks;
651 else
652 block_addr -= sit_i->sit_blocks;
653
654 return block_addr + sit_i->sit_base_addr;
655}
656
657static inline void set_to_next_sit(struct sit_info *sit_i, unsigned int start)
658{
659 unsigned int block_off = SIT_BLOCK_OFFSET(start);
660
661 f2fs_change_bit(block_off, sit_i->sit_bitmap);
662}
663
664static inline unsigned long long get_mtime(struct f2fs_sb_info *sbi)
665{
666 struct sit_info *sit_i = SIT_I(sbi);
667 return sit_i->elapsed_time + CURRENT_TIME_SEC.tv_sec -
668 sit_i->mounted_time;
669}
670
671static inline void set_summary(struct f2fs_summary *sum, nid_t nid,
672 unsigned int ofs_in_node, unsigned char version)
673{
674 sum->nid = cpu_to_le32(nid);
675 sum->ofs_in_node = cpu_to_le16(ofs_in_node);
676 sum->version = version;
677}
678
679static inline block_t start_sum_block(struct f2fs_sb_info *sbi)
680{
681 return __start_cp_addr(sbi) +
682 le32_to_cpu(F2FS_CKPT(sbi)->cp_pack_start_sum);
683}
684
685static inline block_t sum_blk_addr(struct f2fs_sb_info *sbi, int base, int type)
686{
687 return __start_cp_addr(sbi) +
688 le32_to_cpu(F2FS_CKPT(sbi)->cp_pack_total_block_count)
689 - (base + 1) + type;
690}
691
692static inline bool no_fggc_candidate(struct f2fs_sb_info *sbi,
693 unsigned int secno)
694{
695 if (get_valid_blocks(sbi, secno, sbi->segs_per_sec) >=
696 sbi->fggc_threshold)
697 return true;
698 return false;
699}
700
701static inline bool sec_usage_check(struct f2fs_sb_info *sbi, unsigned int secno)
702{
703 if (IS_CURSEC(sbi, secno) || (sbi->cur_victim_sec == secno))
704 return true;
705 return false;
706}
707
708/*
709 * It is very important to gather dirty pages and write at once, so that we can
710 * submit a big bio without interfering other data writes.
711 * By default, 512 pages for directory data,
712 * 512 pages (2MB) * 3 for three types of nodes, and
713 * max_bio_blocks for meta are set.
714 */
715static inline int nr_pages_to_skip(struct f2fs_sb_info *sbi, int type)
716{
717 if (sbi->sb->s_bdi->wb.dirty_exceeded)
718 return 0;
719
720 if (type == DATA)
721 return sbi->blocks_per_seg;
722 else if (type == NODE)
723 return 8 * sbi->blocks_per_seg;
724 else if (type == META)
725 return 8 * BIO_MAX_PAGES;
726 else
727 return 0;
728}
729
730/*
731 * When writing pages, it'd better align nr_to_write for segment size.
732 */
733static inline long nr_pages_to_write(struct f2fs_sb_info *sbi, int type,
734 struct writeback_control *wbc)
735{
736 long nr_to_write, desired;
737
738 if (wbc->sync_mode != WB_SYNC_NONE)
739 return 0;
740
741 nr_to_write = wbc->nr_to_write;
742 desired = BIO_MAX_PAGES;
743 if (type == NODE)
744 desired <<= 1;
745
746 wbc->nr_to_write = desired;
747 return desired - nr_to_write;
748}