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_HOT(t) ((t) == CURSEG_HOT_NODE || (t) == CURSEG_HOT_DATA)
31#define IS_WARM(t) ((t) == CURSEG_WARM_NODE || (t) == CURSEG_WARM_DATA)
32#define IS_COLD(t) ((t) == CURSEG_COLD_NODE || (t) == CURSEG_COLD_DATA)
33
34#define IS_CURSEG(sbi, seg) \
35 (((seg) == CURSEG_I(sbi, CURSEG_HOT_DATA)->segno) || \
36 ((seg) == CURSEG_I(sbi, CURSEG_WARM_DATA)->segno) || \
37 ((seg) == CURSEG_I(sbi, CURSEG_COLD_DATA)->segno) || \
38 ((seg) == CURSEG_I(sbi, CURSEG_HOT_NODE)->segno) || \
39 ((seg) == CURSEG_I(sbi, CURSEG_WARM_NODE)->segno) || \
40 ((seg) == CURSEG_I(sbi, CURSEG_COLD_NODE)->segno))
41
42#define IS_CURSEC(sbi, secno) \
43 (((secno) == CURSEG_I(sbi, CURSEG_HOT_DATA)->segno / \
44 (sbi)->segs_per_sec) || \
45 ((secno) == CURSEG_I(sbi, CURSEG_WARM_DATA)->segno / \
46 (sbi)->segs_per_sec) || \
47 ((secno) == CURSEG_I(sbi, CURSEG_COLD_DATA)->segno / \
48 (sbi)->segs_per_sec) || \
49 ((secno) == CURSEG_I(sbi, CURSEG_HOT_NODE)->segno / \
50 (sbi)->segs_per_sec) || \
51 ((secno) == CURSEG_I(sbi, CURSEG_WARM_NODE)->segno / \
52 (sbi)->segs_per_sec) || \
53 ((secno) == CURSEG_I(sbi, CURSEG_COLD_NODE)->segno / \
54 (sbi)->segs_per_sec)) \
55
56#define MAIN_BLKADDR(sbi) \
57 (SM_I(sbi) ? SM_I(sbi)->main_blkaddr : \
58 le32_to_cpu(F2FS_RAW_SUPER(sbi)->main_blkaddr))
59#define SEG0_BLKADDR(sbi) \
60 (SM_I(sbi) ? SM_I(sbi)->seg0_blkaddr : \
61 le32_to_cpu(F2FS_RAW_SUPER(sbi)->segment0_blkaddr))
62
63#define MAIN_SEGS(sbi) (SM_I(sbi)->main_segments)
64#define MAIN_SECS(sbi) ((sbi)->total_sections)
65
66#define TOTAL_SEGS(sbi) \
67 (SM_I(sbi) ? SM_I(sbi)->segment_count : \
68 le32_to_cpu(F2FS_RAW_SUPER(sbi)->segment_count))
69#define TOTAL_BLKS(sbi) (TOTAL_SEGS(sbi) << (sbi)->log_blocks_per_seg)
70
71#define MAX_BLKADDR(sbi) (SEG0_BLKADDR(sbi) + TOTAL_BLKS(sbi))
72#define SEGMENT_SIZE(sbi) (1ULL << ((sbi)->log_blocksize + \
73 (sbi)->log_blocks_per_seg))
74
75#define START_BLOCK(sbi, segno) (SEG0_BLKADDR(sbi) + \
76 (GET_R2L_SEGNO(FREE_I(sbi), segno) << (sbi)->log_blocks_per_seg))
77
78#define NEXT_FREE_BLKADDR(sbi, curseg) \
79 (START_BLOCK(sbi, (curseg)->segno) + (curseg)->next_blkoff)
80
81#define GET_SEGOFF_FROM_SEG0(sbi, blk_addr) ((blk_addr) - SEG0_BLKADDR(sbi))
82#define GET_SEGNO_FROM_SEG0(sbi, blk_addr) \
83 (GET_SEGOFF_FROM_SEG0(sbi, blk_addr) >> (sbi)->log_blocks_per_seg)
84#define GET_BLKOFF_FROM_SEG0(sbi, blk_addr) \
85 (GET_SEGOFF_FROM_SEG0(sbi, blk_addr) & ((sbi)->blocks_per_seg - 1))
86
87#define GET_SEGNO(sbi, blk_addr) \
88 ((((blk_addr) == NULL_ADDR) || ((blk_addr) == NEW_ADDR)) ? \
89 NULL_SEGNO : GET_L2R_SEGNO(FREE_I(sbi), \
90 GET_SEGNO_FROM_SEG0(sbi, blk_addr)))
91#define BLKS_PER_SEC(sbi) \
92 ((sbi)->segs_per_sec * (sbi)->blocks_per_seg)
93#define GET_SEC_FROM_SEG(sbi, segno) \
94 ((segno) / (sbi)->segs_per_sec)
95#define GET_SEG_FROM_SEC(sbi, secno) \
96 ((secno) * (sbi)->segs_per_sec)
97#define GET_ZONE_FROM_SEC(sbi, secno) \
98 ((secno) / (sbi)->secs_per_zone)
99#define GET_ZONE_FROM_SEG(sbi, segno) \
100 GET_ZONE_FROM_SEC(sbi, GET_SEC_FROM_SEG(sbi, segno))
101
102#define GET_SUM_BLOCK(sbi, segno) \
103 ((sbi)->sm_info->ssa_blkaddr + (segno))
104
105#define GET_SUM_TYPE(footer) ((footer)->entry_type)
106#define SET_SUM_TYPE(footer, type) ((footer)->entry_type = (type))
107
108#define SIT_ENTRY_OFFSET(sit_i, segno) \
109 ((segno) % (sit_i)->sents_per_block)
110#define SIT_BLOCK_OFFSET(segno) \
111 ((segno) / SIT_ENTRY_PER_BLOCK)
112#define START_SEGNO(segno) \
113 (SIT_BLOCK_OFFSET(segno) * SIT_ENTRY_PER_BLOCK)
114#define SIT_BLK_CNT(sbi) \
115 ((MAIN_SEGS(sbi) + SIT_ENTRY_PER_BLOCK - 1) / SIT_ENTRY_PER_BLOCK)
116#define f2fs_bitmap_size(nr) \
117 (BITS_TO_LONGS(nr) * sizeof(unsigned long))
118
119#define SECTOR_FROM_BLOCK(blk_addr) \
120 (((sector_t)blk_addr) << F2FS_LOG_SECTORS_PER_BLOCK)
121#define SECTOR_TO_BLOCK(sectors) \
122 ((sectors) >> F2FS_LOG_SECTORS_PER_BLOCK)
123
124/*
125 * indicate a block allocation direction: RIGHT and LEFT.
126 * RIGHT means allocating new sections towards the end of volume.
127 * LEFT means the opposite direction.
128 */
129enum {
130 ALLOC_RIGHT = 0,
131 ALLOC_LEFT
132};
133
134/*
135 * In the victim_sel_policy->alloc_mode, there are two block allocation modes.
136 * LFS writes data sequentially with cleaning operations.
137 * SSR (Slack Space Recycle) reuses obsolete space without cleaning operations.
138 */
139enum {
140 LFS = 0,
141 SSR
142};
143
144/*
145 * In the victim_sel_policy->gc_mode, there are two gc, aka cleaning, modes.
146 * GC_CB is based on cost-benefit algorithm.
147 * GC_GREEDY is based on greedy algorithm.
148 */
149enum {
150 GC_CB = 0,
151 GC_GREEDY,
152 ALLOC_NEXT,
153 FLUSH_DEVICE,
154 MAX_GC_POLICY,
155};
156
157/*
158 * BG_GC means the background cleaning job.
159 * FG_GC means the on-demand cleaning job.
160 * FORCE_FG_GC means on-demand cleaning job in background.
161 */
162enum {
163 BG_GC = 0,
164 FG_GC,
165 FORCE_FG_GC,
166};
167
168/* for a function parameter to select a victim segment */
169struct victim_sel_policy {
170 int alloc_mode; /* LFS or SSR */
171 int gc_mode; /* GC_CB or GC_GREEDY */
172 unsigned long *dirty_segmap; /* dirty segment bitmap */
173 unsigned int max_search; /* maximum # of segments to search */
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 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/*
207 * this value is set in page as a private data which indicate that
208 * the page is atomically written, and it is in inmem_pages list.
209 */
210#define ATOMIC_WRITTEN_PAGE ((unsigned long)-1)
211#define DUMMY_WRITTEN_PAGE ((unsigned long)-2)
212
213#define IS_ATOMIC_WRITTEN_PAGE(page) \
214 (page_private(page) == (unsigned long)ATOMIC_WRITTEN_PAGE)
215#define IS_DUMMY_WRITTEN_PAGE(page) \
216 (page_private(page) == (unsigned long)DUMMY_WRITTEN_PAGE)
217
218struct inmem_pages {
219 struct list_head list;
220 struct page *page;
221 block_t old_addr; /* for revoking when fail to commit */
222};
223
224struct sit_info {
225 const struct segment_allocation *s_ops;
226
227 block_t sit_base_addr; /* start block address of SIT area */
228 block_t sit_blocks; /* # of blocks used by SIT area */
229 block_t written_valid_blocks; /* # of valid blocks in main area */
230 char *sit_bitmap; /* SIT bitmap pointer */
231#ifdef CONFIG_F2FS_CHECK_FS
232 char *sit_bitmap_mir; /* SIT bitmap mirror */
233#endif
234 unsigned int bitmap_size; /* SIT bitmap size */
235
236 unsigned long *tmp_map; /* bitmap for temporal use */
237 unsigned long *dirty_sentries_bitmap; /* bitmap for dirty sentries */
238 unsigned int dirty_sentries; /* # of dirty sentries */
239 unsigned int sents_per_block; /* # of SIT entries per block */
240 struct rw_semaphore sentry_lock; /* to protect SIT cache */
241 struct seg_entry *sentries; /* SIT segment-level cache */
242 struct sec_entry *sec_entries; /* SIT section-level cache */
243
244 /* for cost-benefit algorithm in cleaning procedure */
245 unsigned long long elapsed_time; /* elapsed time after mount */
246 unsigned long long mounted_time; /* mount time */
247 unsigned long long min_mtime; /* min. modification time */
248 unsigned long long max_mtime; /* max. modification time */
249
250 unsigned int last_victim[MAX_GC_POLICY]; /* last victim segment # */
251};
252
253struct free_segmap_info {
254 unsigned int start_segno; /* start segment number logically */
255 unsigned int free_segments; /* # of free segments */
256 unsigned int free_sections; /* # of free sections */
257 spinlock_t segmap_lock; /* free segmap lock */
258 unsigned long *free_segmap; /* free segment bitmap */
259 unsigned long *free_secmap; /* free section bitmap */
260};
261
262/* Notice: The order of dirty type is same with CURSEG_XXX in f2fs.h */
263enum dirty_type {
264 DIRTY_HOT_DATA, /* dirty segments assigned as hot data logs */
265 DIRTY_WARM_DATA, /* dirty segments assigned as warm data logs */
266 DIRTY_COLD_DATA, /* dirty segments assigned as cold data logs */
267 DIRTY_HOT_NODE, /* dirty segments assigned as hot node logs */
268 DIRTY_WARM_NODE, /* dirty segments assigned as warm node logs */
269 DIRTY_COLD_NODE, /* dirty segments assigned as cold node logs */
270 DIRTY, /* to count # of dirty segments */
271 PRE, /* to count # of entirely obsolete segments */
272 NR_DIRTY_TYPE
273};
274
275struct dirty_seglist_info {
276 const struct victim_selection *v_ops; /* victim selction operation */
277 unsigned long *dirty_segmap[NR_DIRTY_TYPE];
278 struct mutex seglist_lock; /* lock for segment bitmaps */
279 int nr_dirty[NR_DIRTY_TYPE]; /* # of dirty segments */
280 unsigned long *victim_secmap; /* background GC victims */
281};
282
283/* victim selection function for cleaning and SSR */
284struct victim_selection {
285 int (*get_victim)(struct f2fs_sb_info *, unsigned int *,
286 int, int, char);
287};
288
289/* for active log information */
290struct curseg_info {
291 struct mutex curseg_mutex; /* lock for consistency */
292 struct f2fs_summary_block *sum_blk; /* cached summary block */
293 struct rw_semaphore journal_rwsem; /* protect journal area */
294 struct f2fs_journal *journal; /* cached journal info */
295 unsigned char alloc_type; /* current allocation type */
296 unsigned int segno; /* current segment number */
297 unsigned short next_blkoff; /* next block offset to write */
298 unsigned int zone; /* current zone number */
299 unsigned int next_segno; /* preallocated segment */
300};
301
302struct sit_entry_set {
303 struct list_head set_list; /* link with all sit sets */
304 unsigned int start_segno; /* start segno of sits in set */
305 unsigned int entry_cnt; /* the # of sit entries in set */
306};
307
308/*
309 * inline functions
310 */
311static inline struct curseg_info *CURSEG_I(struct f2fs_sb_info *sbi, int type)
312{
313 return (struct curseg_info *)(SM_I(sbi)->curseg_array + type);
314}
315
316static inline struct seg_entry *get_seg_entry(struct f2fs_sb_info *sbi,
317 unsigned int segno)
318{
319 struct sit_info *sit_i = SIT_I(sbi);
320 return &sit_i->sentries[segno];
321}
322
323static inline struct sec_entry *get_sec_entry(struct f2fs_sb_info *sbi,
324 unsigned int segno)
325{
326 struct sit_info *sit_i = SIT_I(sbi);
327 return &sit_i->sec_entries[GET_SEC_FROM_SEG(sbi, segno)];
328}
329
330static inline unsigned int get_valid_blocks(struct f2fs_sb_info *sbi,
331 unsigned int segno, bool use_section)
332{
333 /*
334 * In order to get # of valid blocks in a section instantly from many
335 * segments, f2fs manages two counting structures separately.
336 */
337 if (use_section && sbi->segs_per_sec > 1)
338 return get_sec_entry(sbi, segno)->valid_blocks;
339 else
340 return get_seg_entry(sbi, segno)->valid_blocks;
341}
342
343static inline void seg_info_from_raw_sit(struct seg_entry *se,
344 struct f2fs_sit_entry *rs)
345{
346 se->valid_blocks = GET_SIT_VBLOCKS(rs);
347 se->ckpt_valid_blocks = GET_SIT_VBLOCKS(rs);
348 memcpy(se->cur_valid_map, rs->valid_map, SIT_VBLOCK_MAP_SIZE);
349 memcpy(se->ckpt_valid_map, rs->valid_map, SIT_VBLOCK_MAP_SIZE);
350#ifdef CONFIG_F2FS_CHECK_FS
351 memcpy(se->cur_valid_map_mir, rs->valid_map, SIT_VBLOCK_MAP_SIZE);
352#endif
353 se->type = GET_SIT_TYPE(rs);
354 se->mtime = le64_to_cpu(rs->mtime);
355}
356
357static inline void __seg_info_to_raw_sit(struct seg_entry *se,
358 struct f2fs_sit_entry *rs)
359{
360 unsigned short raw_vblocks = (se->type << SIT_VBLOCKS_SHIFT) |
361 se->valid_blocks;
362 rs->vblocks = cpu_to_le16(raw_vblocks);
363 memcpy(rs->valid_map, se->cur_valid_map, SIT_VBLOCK_MAP_SIZE);
364 rs->mtime = cpu_to_le64(se->mtime);
365}
366
367static inline void seg_info_to_sit_page(struct f2fs_sb_info *sbi,
368 struct page *page, unsigned int start)
369{
370 struct f2fs_sit_block *raw_sit;
371 struct seg_entry *se;
372 struct f2fs_sit_entry *rs;
373 unsigned int end = min(start + SIT_ENTRY_PER_BLOCK,
374 (unsigned long)MAIN_SEGS(sbi));
375 int i;
376
377 raw_sit = (struct f2fs_sit_block *)page_address(page);
378 for (i = 0; i < end - start; i++) {
379 rs = &raw_sit->entries[i];
380 se = get_seg_entry(sbi, start + i);
381 __seg_info_to_raw_sit(se, rs);
382 }
383}
384
385static inline void seg_info_to_raw_sit(struct seg_entry *se,
386 struct f2fs_sit_entry *rs)
387{
388 __seg_info_to_raw_sit(se, rs);
389
390 memcpy(se->ckpt_valid_map, rs->valid_map, SIT_VBLOCK_MAP_SIZE);
391 se->ckpt_valid_blocks = se->valid_blocks;
392}
393
394static inline unsigned int find_next_inuse(struct free_segmap_info *free_i,
395 unsigned int max, unsigned int segno)
396{
397 unsigned int ret;
398 spin_lock(&free_i->segmap_lock);
399 ret = find_next_bit(free_i->free_segmap, max, segno);
400 spin_unlock(&free_i->segmap_lock);
401 return ret;
402}
403
404static inline void __set_free(struct f2fs_sb_info *sbi, unsigned int segno)
405{
406 struct free_segmap_info *free_i = FREE_I(sbi);
407 unsigned int secno = GET_SEC_FROM_SEG(sbi, segno);
408 unsigned int start_segno = GET_SEG_FROM_SEC(sbi, secno);
409 unsigned int next;
410
411 spin_lock(&free_i->segmap_lock);
412 clear_bit(segno, free_i->free_segmap);
413 free_i->free_segments++;
414
415 next = find_next_bit(free_i->free_segmap,
416 start_segno + sbi->segs_per_sec, start_segno);
417 if (next >= start_segno + sbi->segs_per_sec) {
418 clear_bit(secno, free_i->free_secmap);
419 free_i->free_sections++;
420 }
421 spin_unlock(&free_i->segmap_lock);
422}
423
424static inline void __set_inuse(struct f2fs_sb_info *sbi,
425 unsigned int segno)
426{
427 struct free_segmap_info *free_i = FREE_I(sbi);
428 unsigned int secno = GET_SEC_FROM_SEG(sbi, segno);
429
430 set_bit(segno, free_i->free_segmap);
431 free_i->free_segments--;
432 if (!test_and_set_bit(secno, free_i->free_secmap))
433 free_i->free_sections--;
434}
435
436static inline void __set_test_and_free(struct f2fs_sb_info *sbi,
437 unsigned int segno)
438{
439 struct free_segmap_info *free_i = FREE_I(sbi);
440 unsigned int secno = GET_SEC_FROM_SEG(sbi, segno);
441 unsigned int start_segno = GET_SEG_FROM_SEC(sbi, secno);
442 unsigned int next;
443
444 spin_lock(&free_i->segmap_lock);
445 if (test_and_clear_bit(segno, free_i->free_segmap)) {
446 free_i->free_segments++;
447
448 next = find_next_bit(free_i->free_segmap,
449 start_segno + sbi->segs_per_sec, start_segno);
450 if (next >= start_segno + sbi->segs_per_sec) {
451 if (test_and_clear_bit(secno, free_i->free_secmap))
452 free_i->free_sections++;
453 }
454 }
455 spin_unlock(&free_i->segmap_lock);
456}
457
458static inline void __set_test_and_inuse(struct f2fs_sb_info *sbi,
459 unsigned int segno)
460{
461 struct free_segmap_info *free_i = FREE_I(sbi);
462 unsigned int secno = GET_SEC_FROM_SEG(sbi, segno);
463
464 spin_lock(&free_i->segmap_lock);
465 if (!test_and_set_bit(segno, free_i->free_segmap)) {
466 free_i->free_segments--;
467 if (!test_and_set_bit(secno, free_i->free_secmap))
468 free_i->free_sections--;
469 }
470 spin_unlock(&free_i->segmap_lock);
471}
472
473static inline void get_sit_bitmap(struct f2fs_sb_info *sbi,
474 void *dst_addr)
475{
476 struct sit_info *sit_i = SIT_I(sbi);
477
478#ifdef CONFIG_F2FS_CHECK_FS
479 if (memcmp(sit_i->sit_bitmap, sit_i->sit_bitmap_mir,
480 sit_i->bitmap_size))
481 f2fs_bug_on(sbi, 1);
482#endif
483 memcpy(dst_addr, sit_i->sit_bitmap, sit_i->bitmap_size);
484}
485
486static inline block_t written_block_count(struct f2fs_sb_info *sbi)
487{
488 return SIT_I(sbi)->written_valid_blocks;
489}
490
491static inline unsigned int free_segments(struct f2fs_sb_info *sbi)
492{
493 return FREE_I(sbi)->free_segments;
494}
495
496static inline int reserved_segments(struct f2fs_sb_info *sbi)
497{
498 return SM_I(sbi)->reserved_segments;
499}
500
501static inline unsigned int free_sections(struct f2fs_sb_info *sbi)
502{
503 return FREE_I(sbi)->free_sections;
504}
505
506static inline unsigned int prefree_segments(struct f2fs_sb_info *sbi)
507{
508 return DIRTY_I(sbi)->nr_dirty[PRE];
509}
510
511static inline unsigned int dirty_segments(struct f2fs_sb_info *sbi)
512{
513 return DIRTY_I(sbi)->nr_dirty[DIRTY_HOT_DATA] +
514 DIRTY_I(sbi)->nr_dirty[DIRTY_WARM_DATA] +
515 DIRTY_I(sbi)->nr_dirty[DIRTY_COLD_DATA] +
516 DIRTY_I(sbi)->nr_dirty[DIRTY_HOT_NODE] +
517 DIRTY_I(sbi)->nr_dirty[DIRTY_WARM_NODE] +
518 DIRTY_I(sbi)->nr_dirty[DIRTY_COLD_NODE];
519}
520
521static inline int overprovision_segments(struct f2fs_sb_info *sbi)
522{
523 return SM_I(sbi)->ovp_segments;
524}
525
526static inline int reserved_sections(struct f2fs_sb_info *sbi)
527{
528 return GET_SEC_FROM_SEG(sbi, (unsigned int)reserved_segments(sbi));
529}
530
531static inline bool has_curseg_enough_space(struct f2fs_sb_info *sbi)
532{
533 unsigned int node_blocks = get_pages(sbi, F2FS_DIRTY_NODES) +
534 get_pages(sbi, F2FS_DIRTY_DENTS);
535 unsigned int dent_blocks = get_pages(sbi, F2FS_DIRTY_DENTS);
536 unsigned int segno, left_blocks;
537 int i;
538
539 /* check current node segment */
540 for (i = CURSEG_HOT_NODE; i <= CURSEG_COLD_NODE; i++) {
541 segno = CURSEG_I(sbi, i)->segno;
542 left_blocks = sbi->blocks_per_seg -
543 get_seg_entry(sbi, segno)->ckpt_valid_blocks;
544
545 if (node_blocks > left_blocks)
546 return false;
547 }
548
549 /* check current data segment */
550 segno = CURSEG_I(sbi, CURSEG_HOT_DATA)->segno;
551 left_blocks = sbi->blocks_per_seg -
552 get_seg_entry(sbi, segno)->ckpt_valid_blocks;
553 if (dent_blocks > left_blocks)
554 return false;
555 return true;
556}
557
558static inline bool has_not_enough_free_secs(struct f2fs_sb_info *sbi,
559 int freed, int needed)
560{
561 int node_secs = get_blocktype_secs(sbi, F2FS_DIRTY_NODES);
562 int dent_secs = get_blocktype_secs(sbi, F2FS_DIRTY_DENTS);
563 int imeta_secs = get_blocktype_secs(sbi, F2FS_DIRTY_IMETA);
564
565 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
566 return false;
567
568 if (free_sections(sbi) + freed == reserved_sections(sbi) + needed &&
569 has_curseg_enough_space(sbi))
570 return false;
571 return (free_sections(sbi) + freed) <=
572 (node_secs + 2 * dent_secs + imeta_secs +
573 reserved_sections(sbi) + needed);
574}
575
576static inline bool excess_prefree_segs(struct f2fs_sb_info *sbi)
577{
578 return prefree_segments(sbi) > SM_I(sbi)->rec_prefree_segments;
579}
580
581static inline int utilization(struct f2fs_sb_info *sbi)
582{
583 return div_u64((u64)valid_user_blocks(sbi) * 100,
584 sbi->user_block_count);
585}
586
587/*
588 * Sometimes f2fs may be better to drop out-of-place update policy.
589 * And, users can control the policy through sysfs entries.
590 * There are five policies with triggering conditions as follows.
591 * F2FS_IPU_FORCE - all the time,
592 * F2FS_IPU_SSR - if SSR mode is activated,
593 * F2FS_IPU_UTIL - if FS utilization is over threashold,
594 * F2FS_IPU_SSR_UTIL - if SSR mode is activated and FS utilization is over
595 * threashold,
596 * F2FS_IPU_FSYNC - activated in fsync path only for high performance flash
597 * storages. IPU will be triggered only if the # of dirty
598 * pages over min_fsync_blocks.
599 * F2FS_IPUT_DISABLE - disable IPU. (=default option)
600 */
601#define DEF_MIN_IPU_UTIL 70
602#define DEF_MIN_FSYNC_BLOCKS 8
603#define DEF_MIN_HOT_BLOCKS 16
604
605#define SMALL_VOLUME_SEGMENTS (16 * 512) /* 16GB */
606
607enum {
608 F2FS_IPU_FORCE,
609 F2FS_IPU_SSR,
610 F2FS_IPU_UTIL,
611 F2FS_IPU_SSR_UTIL,
612 F2FS_IPU_FSYNC,
613 F2FS_IPU_ASYNC,
614};
615
616static inline unsigned int curseg_segno(struct f2fs_sb_info *sbi,
617 int type)
618{
619 struct curseg_info *curseg = CURSEG_I(sbi, type);
620 return curseg->segno;
621}
622
623static inline unsigned char curseg_alloc_type(struct f2fs_sb_info *sbi,
624 int type)
625{
626 struct curseg_info *curseg = CURSEG_I(sbi, type);
627 return curseg->alloc_type;
628}
629
630static inline unsigned short curseg_blkoff(struct f2fs_sb_info *sbi, int type)
631{
632 struct curseg_info *curseg = CURSEG_I(sbi, type);
633 return curseg->next_blkoff;
634}
635
636static inline void check_seg_range(struct f2fs_sb_info *sbi, unsigned int segno)
637{
638 f2fs_bug_on(sbi, segno > TOTAL_SEGS(sbi) - 1);
639}
640
641static inline void verify_block_addr(struct f2fs_io_info *fio, block_t blk_addr)
642{
643 struct f2fs_sb_info *sbi = fio->sbi;
644
645 if (PAGE_TYPE_OF_BIO(fio->type) == META &&
646 (!is_read_io(fio->op) || fio->is_meta))
647 BUG_ON(blk_addr < SEG0_BLKADDR(sbi) ||
648 blk_addr >= MAIN_BLKADDR(sbi));
649 else
650 BUG_ON(blk_addr < MAIN_BLKADDR(sbi) ||
651 blk_addr >= MAX_BLKADDR(sbi));
652}
653
654/*
655 * Summary block is always treated as an invalid block
656 */
657static inline int check_block_count(struct f2fs_sb_info *sbi,
658 int segno, struct f2fs_sit_entry *raw_sit)
659{
660#ifdef CONFIG_F2FS_CHECK_FS
661 bool is_valid = test_bit_le(0, raw_sit->valid_map) ? true : false;
662 int valid_blocks = 0;
663 int cur_pos = 0, next_pos;
664
665 /* check bitmap with valid block count */
666 do {
667 if (is_valid) {
668 next_pos = find_next_zero_bit_le(&raw_sit->valid_map,
669 sbi->blocks_per_seg,
670 cur_pos);
671 valid_blocks += next_pos - cur_pos;
672 } else
673 next_pos = find_next_bit_le(&raw_sit->valid_map,
674 sbi->blocks_per_seg,
675 cur_pos);
676 cur_pos = next_pos;
677 is_valid = !is_valid;
678 } while (cur_pos < sbi->blocks_per_seg);
679
680 if (unlikely(GET_SIT_VBLOCKS(raw_sit) != valid_blocks)) {
681 f2fs_msg(sbi->sb, KERN_ERR,
682 "Mismatch valid blocks %d vs. %d",
683 GET_SIT_VBLOCKS(raw_sit), valid_blocks);
684 set_sbi_flag(sbi, SBI_NEED_FSCK);
685 return -EINVAL;
686 }
687#endif
688 /* check segment usage, and check boundary of a given segment number */
689 if (unlikely(GET_SIT_VBLOCKS(raw_sit) > sbi->blocks_per_seg
690 || segno > TOTAL_SEGS(sbi) - 1)) {
691 f2fs_msg(sbi->sb, KERN_ERR,
692 "Wrong valid blocks %d or segno %u",
693 GET_SIT_VBLOCKS(raw_sit), segno);
694 set_sbi_flag(sbi, SBI_NEED_FSCK);
695 return -EINVAL;
696 }
697 return 0;
698}
699
700static inline pgoff_t current_sit_addr(struct f2fs_sb_info *sbi,
701 unsigned int start)
702{
703 struct sit_info *sit_i = SIT_I(sbi);
704 unsigned int offset = SIT_BLOCK_OFFSET(start);
705 block_t blk_addr = sit_i->sit_base_addr + offset;
706
707 check_seg_range(sbi, start);
708
709#ifdef CONFIG_F2FS_CHECK_FS
710 if (f2fs_test_bit(offset, sit_i->sit_bitmap) !=
711 f2fs_test_bit(offset, sit_i->sit_bitmap_mir))
712 f2fs_bug_on(sbi, 1);
713#endif
714
715 /* calculate sit block address */
716 if (f2fs_test_bit(offset, sit_i->sit_bitmap))
717 blk_addr += sit_i->sit_blocks;
718
719 return blk_addr;
720}
721
722static inline pgoff_t next_sit_addr(struct f2fs_sb_info *sbi,
723 pgoff_t block_addr)
724{
725 struct sit_info *sit_i = SIT_I(sbi);
726 block_addr -= sit_i->sit_base_addr;
727 if (block_addr < sit_i->sit_blocks)
728 block_addr += sit_i->sit_blocks;
729 else
730 block_addr -= sit_i->sit_blocks;
731
732 return block_addr + sit_i->sit_base_addr;
733}
734
735static inline void set_to_next_sit(struct sit_info *sit_i, unsigned int start)
736{
737 unsigned int block_off = SIT_BLOCK_OFFSET(start);
738
739 f2fs_change_bit(block_off, sit_i->sit_bitmap);
740#ifdef CONFIG_F2FS_CHECK_FS
741 f2fs_change_bit(block_off, sit_i->sit_bitmap_mir);
742#endif
743}
744
745static inline unsigned long long get_mtime(struct f2fs_sb_info *sbi)
746{
747 struct sit_info *sit_i = SIT_I(sbi);
748 time64_t now = ktime_get_real_seconds();
749
750 return sit_i->elapsed_time + now - sit_i->mounted_time;
751}
752
753static inline void set_summary(struct f2fs_summary *sum, nid_t nid,
754 unsigned int ofs_in_node, unsigned char version)
755{
756 sum->nid = cpu_to_le32(nid);
757 sum->ofs_in_node = cpu_to_le16(ofs_in_node);
758 sum->version = version;
759}
760
761static inline block_t start_sum_block(struct f2fs_sb_info *sbi)
762{
763 return __start_cp_addr(sbi) +
764 le32_to_cpu(F2FS_CKPT(sbi)->cp_pack_start_sum);
765}
766
767static inline block_t sum_blk_addr(struct f2fs_sb_info *sbi, int base, int type)
768{
769 return __start_cp_addr(sbi) +
770 le32_to_cpu(F2FS_CKPT(sbi)->cp_pack_total_block_count)
771 - (base + 1) + type;
772}
773
774static inline bool no_fggc_candidate(struct f2fs_sb_info *sbi,
775 unsigned int secno)
776{
777 if (get_valid_blocks(sbi, GET_SEG_FROM_SEC(sbi, secno), true) >
778 sbi->fggc_threshold)
779 return true;
780 return false;
781}
782
783static inline bool sec_usage_check(struct f2fs_sb_info *sbi, unsigned int secno)
784{
785 if (IS_CURSEC(sbi, secno) || (sbi->cur_victim_sec == secno))
786 return true;
787 return false;
788}
789
790/*
791 * It is very important to gather dirty pages and write at once, so that we can
792 * submit a big bio without interfering other data writes.
793 * By default, 512 pages for directory data,
794 * 512 pages (2MB) * 8 for nodes, and
795 * 256 pages * 8 for meta are set.
796 */
797static inline int nr_pages_to_skip(struct f2fs_sb_info *sbi, int type)
798{
799 if (sbi->sb->s_bdi->wb.dirty_exceeded)
800 return 0;
801
802 if (type == DATA)
803 return sbi->blocks_per_seg;
804 else if (type == NODE)
805 return 8 * sbi->blocks_per_seg;
806 else if (type == META)
807 return 8 * BIO_MAX_PAGES;
808 else
809 return 0;
810}
811
812/*
813 * When writing pages, it'd better align nr_to_write for segment size.
814 */
815static inline long nr_pages_to_write(struct f2fs_sb_info *sbi, int type,
816 struct writeback_control *wbc)
817{
818 long nr_to_write, desired;
819
820 if (wbc->sync_mode != WB_SYNC_NONE)
821 return 0;
822
823 nr_to_write = wbc->nr_to_write;
824 desired = BIO_MAX_PAGES;
825 if (type == NODE)
826 desired <<= 1;
827
828 wbc->nr_to_write = desired;
829 return desired - nr_to_write;
830}
831
832static inline void wake_up_discard_thread(struct f2fs_sb_info *sbi, bool force)
833{
834 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
835 bool wakeup = false;
836 int i;
837
838 if (force)
839 goto wake_up;
840
841 mutex_lock(&dcc->cmd_lock);
842 for (i = MAX_PLIST_NUM - 1; i >= 0; i--) {
843 if (i + 1 < dcc->discard_granularity)
844 break;
845 if (!list_empty(&dcc->pend_list[i])) {
846 wakeup = true;
847 break;
848 }
849 }
850 mutex_unlock(&dcc->cmd_lock);
851 if (!wakeup)
852 return;
853wake_up:
854 dcc->discard_wake = 1;
855 wake_up_interruptible_all(&dcc->discard_wait_queue);
856}