blob: 013aed2f3539fbdee7c532b811dbbe0f2ec124ab [file] [log] [blame]
Jaegeuk Kim0a8165d2012-11-29 13:28:09 +09001/*
Jaegeuk Kim39a53e02012-11-28 13:37:31 +09002 * 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 */
Jaegeuk Kimac5d1562013-04-29 16:58:39 +090011#include <linux/blkdev.h>
12
Jaegeuk Kim39a53e02012-11-28 13:37:31 +090013/* constant macro */
14#define NULL_SEGNO ((unsigned int)(~0))
Jaegeuk Kim5ec4e492013-03-31 13:26:03 +090015#define NULL_SECNO ((unsigned int)(~0))
Jaegeuk Kim39a53e02012-11-28 13:37:31 +090016
Jaegeuk Kim58c41032014-03-19 14:17:21 +090017#define DEF_RECLAIM_PREFREE_SEGMENTS 5 /* 5% over total segments */
Jaegeuk Kim81eb8d62013-10-24 13:31:34 +090018
Namjae Jeon6224da82013-04-06 14:44:32 +090019/* L: Logical segment # in volume, R: Relative segment # in main area */
Jaegeuk Kim39a53e02012-11-28 13:37:31 +090020#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
Changman Lee61ae45c2013-11-21 20:04:21 +090023#define IS_DATASEG(t) (t <= CURSEG_COLD_DATA)
24#define IS_NODESEG(t) (t >= CURSEG_HOT_NODE)
Jaegeuk Kim39a53e02012-11-28 13:37:31 +090025
Jaegeuk Kim5c773ba2013-03-31 12:30:04 +090026#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))
Jaegeuk Kim39a53e02012-11-28 13:37:31 +090033
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)
Jaegeuk Kim491c0852014-02-04 13:01:10 +090060#define GET_BLKOFF_FROM_SEG0(sbi, blk_addr) \
61 (GET_SEGOFF_FROM_SEG0(sbi, blk_addr) & (sbi->blocks_per_seg - 1))
62
Jaegeuk Kim39a53e02012-11-28 13:37:31 +090063#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)
Chao Yud3a14af2014-09-04 18:11:47 +080080#define SIT_BLOCK_OFFSET(segno) \
Jaegeuk Kim39a53e02012-11-28 13:37:31 +090081 (segno / SIT_ENTRY_PER_BLOCK)
Chao Yud3a14af2014-09-04 18:11:47 +080082#define START_SEGNO(segno) \
83 (SIT_BLOCK_OFFSET(segno) * SIT_ENTRY_PER_BLOCK)
Chao Yu74de5932013-11-22 09:09:59 +080084#define SIT_BLK_CNT(sbi) \
85 ((TOTAL_SEGS(sbi) + SIT_ENTRY_PER_BLOCK - 1) / SIT_ENTRY_PER_BLOCK)
Jaegeuk Kim39a53e02012-11-28 13:37:31 +090086#define f2fs_bitmap_size(nr) \
87 (BITS_TO_LONGS(nr) * sizeof(unsigned long))
88#define TOTAL_SEGS(sbi) (SM_I(sbi)->main_segments)
Jaegeuk Kim53cf9522013-03-31 12:39:49 +090089#define TOTAL_SECS(sbi) (sbi->total_sections)
Jaegeuk Kim39a53e02012-11-28 13:37:31 +090090
Jaegeuk Kim3cd8a232012-12-10 09:26:05 +090091#define SECTOR_FROM_BLOCK(sbi, blk_addr) \
Jaegeuk Kimf9a4e6d2013-11-28 12:44:05 +090092 (((sector_t)blk_addr) << (sbi)->log_sectors_per_block)
Jaegeuk Kimac5d1562013-04-29 16:58:39 +090093#define SECTOR_TO_BLOCK(sbi, sectors) \
Jaegeuk Kimf9a4e6d2013-11-28 12:44:05 +090094 (sectors >> (sbi)->log_sectors_per_block)
Chao Yucc7b1bb2013-09-22 15:50:50 +080095#define MAX_BIO_BLOCKS(max_hw_blocks) \
96 (min((int)max_hw_blocks, BIO_MAX_PAGES))
Jaegeuk Kim3cd8a232012-12-10 09:26:05 +090097
Jaegeuk Kim39a53e02012-11-28 13:37:31 +090098/*
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 */
Jin Xua26b7c82013-09-05 12:45:26 +0800142 unsigned int max_search; /* maximum # of segments to search */
Jaegeuk Kim39a53e02012-11-28 13:37:31 +0900143 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 */
Jaegeuk Kim5ec4e492013-03-31 13:26:03 +0900220 unsigned long *victim_secmap; /* background GC victims */
Jaegeuk Kim39a53e02012-11-28 13:37:31 +0900221};
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
Chao Yu184a5cd2014-09-04 18:13:01 +0800240struct sit_entry_set {
241 struct list_head set_list; /* link with all sit sets */
242 unsigned int start_segno; /* start segno of sits in set */
243 unsigned int entry_cnt; /* the # of sit entries in set */
244};
245
Jaegeuk Kim39a53e02012-11-28 13:37:31 +0900246/*
247 * inline functions
248 */
249static inline struct curseg_info *CURSEG_I(struct f2fs_sb_info *sbi, int type)
250{
251 return (struct curseg_info *)(SM_I(sbi)->curseg_array + type);
252}
253
254static inline struct seg_entry *get_seg_entry(struct f2fs_sb_info *sbi,
255 unsigned int segno)
256{
257 struct sit_info *sit_i = SIT_I(sbi);
258 return &sit_i->sentries[segno];
259}
260
261static inline struct sec_entry *get_sec_entry(struct f2fs_sb_info *sbi,
262 unsigned int segno)
263{
264 struct sit_info *sit_i = SIT_I(sbi);
265 return &sit_i->sec_entries[GET_SECNO(sbi, segno)];
266}
267
268static inline unsigned int get_valid_blocks(struct f2fs_sb_info *sbi,
269 unsigned int segno, int section)
270{
271 /*
272 * In order to get # of valid blocks in a section instantly from many
273 * segments, f2fs manages two counting structures separately.
274 */
275 if (section > 1)
276 return get_sec_entry(sbi, segno)->valid_blocks;
277 else
278 return get_seg_entry(sbi, segno)->valid_blocks;
279}
280
281static inline void seg_info_from_raw_sit(struct seg_entry *se,
282 struct f2fs_sit_entry *rs)
283{
284 se->valid_blocks = GET_SIT_VBLOCKS(rs);
285 se->ckpt_valid_blocks = GET_SIT_VBLOCKS(rs);
286 memcpy(se->cur_valid_map, rs->valid_map, SIT_VBLOCK_MAP_SIZE);
287 memcpy(se->ckpt_valid_map, rs->valid_map, SIT_VBLOCK_MAP_SIZE);
288 se->type = GET_SIT_TYPE(rs);
289 se->mtime = le64_to_cpu(rs->mtime);
290}
291
292static inline void seg_info_to_raw_sit(struct seg_entry *se,
293 struct f2fs_sit_entry *rs)
294{
295 unsigned short raw_vblocks = (se->type << SIT_VBLOCKS_SHIFT) |
296 se->valid_blocks;
297 rs->vblocks = cpu_to_le16(raw_vblocks);
298 memcpy(rs->valid_map, se->cur_valid_map, SIT_VBLOCK_MAP_SIZE);
299 memcpy(se->ckpt_valid_map, rs->valid_map, SIT_VBLOCK_MAP_SIZE);
300 se->ckpt_valid_blocks = se->valid_blocks;
301 rs->mtime = cpu_to_le64(se->mtime);
302}
303
304static inline unsigned int find_next_inuse(struct free_segmap_info *free_i,
305 unsigned int max, unsigned int segno)
306{
307 unsigned int ret;
308 read_lock(&free_i->segmap_lock);
309 ret = find_next_bit(free_i->free_segmap, max, segno);
310 read_unlock(&free_i->segmap_lock);
311 return ret;
312}
313
314static inline void __set_free(struct f2fs_sb_info *sbi, unsigned int segno)
315{
316 struct free_segmap_info *free_i = FREE_I(sbi);
317 unsigned int secno = segno / sbi->segs_per_sec;
318 unsigned int start_segno = secno * sbi->segs_per_sec;
319 unsigned int next;
320
321 write_lock(&free_i->segmap_lock);
322 clear_bit(segno, free_i->free_segmap);
323 free_i->free_segments++;
324
325 next = find_next_bit(free_i->free_segmap, TOTAL_SEGS(sbi), start_segno);
326 if (next >= start_segno + sbi->segs_per_sec) {
327 clear_bit(secno, free_i->free_secmap);
328 free_i->free_sections++;
329 }
330 write_unlock(&free_i->segmap_lock);
331}
332
333static inline void __set_inuse(struct f2fs_sb_info *sbi,
334 unsigned int segno)
335{
336 struct free_segmap_info *free_i = FREE_I(sbi);
337 unsigned int secno = segno / sbi->segs_per_sec;
338 set_bit(segno, free_i->free_segmap);
339 free_i->free_segments--;
340 if (!test_and_set_bit(secno, free_i->free_secmap))
341 free_i->free_sections--;
342}
343
344static inline void __set_test_and_free(struct f2fs_sb_info *sbi,
345 unsigned int segno)
346{
347 struct free_segmap_info *free_i = FREE_I(sbi);
348 unsigned int secno = segno / sbi->segs_per_sec;
349 unsigned int start_segno = secno * sbi->segs_per_sec;
350 unsigned int next;
351
352 write_lock(&free_i->segmap_lock);
353 if (test_and_clear_bit(segno, free_i->free_segmap)) {
354 free_i->free_segments++;
355
Chao Yuf1121ab2014-07-14 16:45:15 +0800356 next = find_next_bit(free_i->free_segmap,
357 start_segno + sbi->segs_per_sec, start_segno);
Jaegeuk Kim39a53e02012-11-28 13:37:31 +0900358 if (next >= start_segno + sbi->segs_per_sec) {
359 if (test_and_clear_bit(secno, free_i->free_secmap))
360 free_i->free_sections++;
361 }
362 }
363 write_unlock(&free_i->segmap_lock);
364}
365
366static inline void __set_test_and_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 write_lock(&free_i->segmap_lock);
372 if (!test_and_set_bit(segno, free_i->free_segmap)) {
373 free_i->free_segments--;
374 if (!test_and_set_bit(secno, free_i->free_secmap))
375 free_i->free_sections--;
376 }
377 write_unlock(&free_i->segmap_lock);
378}
379
380static inline void get_sit_bitmap(struct f2fs_sb_info *sbi,
381 void *dst_addr)
382{
383 struct sit_info *sit_i = SIT_I(sbi);
384 memcpy(dst_addr, sit_i->sit_bitmap, sit_i->bitmap_size);
385}
386
387static inline block_t written_block_count(struct f2fs_sb_info *sbi)
388{
Jaegeuk Kim8b8343f2014-02-24 13:00:13 +0900389 return SIT_I(sbi)->written_valid_blocks;
Jaegeuk Kim39a53e02012-11-28 13:37:31 +0900390}
391
392static inline unsigned int free_segments(struct f2fs_sb_info *sbi)
393{
Jaegeuk Kim8b8343f2014-02-24 13:00:13 +0900394 return FREE_I(sbi)->free_segments;
Jaegeuk Kim39a53e02012-11-28 13:37:31 +0900395}
396
397static inline int reserved_segments(struct f2fs_sb_info *sbi)
398{
399 return SM_I(sbi)->reserved_segments;
400}
401
402static inline unsigned int free_sections(struct f2fs_sb_info *sbi)
403{
Jaegeuk Kim8b8343f2014-02-24 13:00:13 +0900404 return FREE_I(sbi)->free_sections;
Jaegeuk Kim39a53e02012-11-28 13:37:31 +0900405}
406
407static inline unsigned int prefree_segments(struct f2fs_sb_info *sbi)
408{
409 return DIRTY_I(sbi)->nr_dirty[PRE];
410}
411
412static inline unsigned int dirty_segments(struct f2fs_sb_info *sbi)
413{
414 return DIRTY_I(sbi)->nr_dirty[DIRTY_HOT_DATA] +
415 DIRTY_I(sbi)->nr_dirty[DIRTY_WARM_DATA] +
416 DIRTY_I(sbi)->nr_dirty[DIRTY_COLD_DATA] +
417 DIRTY_I(sbi)->nr_dirty[DIRTY_HOT_NODE] +
418 DIRTY_I(sbi)->nr_dirty[DIRTY_WARM_NODE] +
419 DIRTY_I(sbi)->nr_dirty[DIRTY_COLD_NODE];
420}
421
422static inline int overprovision_segments(struct f2fs_sb_info *sbi)
423{
424 return SM_I(sbi)->ovp_segments;
425}
426
427static inline int overprovision_sections(struct f2fs_sb_info *sbi)
428{
429 return ((unsigned int) overprovision_segments(sbi)) / sbi->segs_per_sec;
430}
431
432static inline int reserved_sections(struct f2fs_sb_info *sbi)
433{
434 return ((unsigned int) reserved_segments(sbi)) / sbi->segs_per_sec;
435}
436
437static inline bool need_SSR(struct f2fs_sb_info *sbi)
438{
Chris Fries6c311ec2014-01-17 14:44:39 -0600439 return (prefree_segments(sbi) / sbi->segs_per_sec)
440 + free_sections(sbi) < overprovision_sections(sbi);
Jaegeuk Kim39a53e02012-11-28 13:37:31 +0900441}
442
Jaegeuk Kim43727522013-02-04 15:11:17 +0900443static inline bool has_not_enough_free_secs(struct f2fs_sb_info *sbi, int freed)
Jaegeuk Kim39a53e02012-11-28 13:37:31 +0900444{
Namjae Jeon5ac206c2013-02-02 23:52:59 +0900445 int node_secs = get_blocktype_secs(sbi, F2FS_DIRTY_NODES);
446 int dent_secs = get_blocktype_secs(sbi, F2FS_DIRTY_DENTS);
Jaegeuk Kim43727522013-02-04 15:11:17 +0900447
Chao Yucfb271d2013-12-05 17:15:22 +0800448 if (unlikely(sbi->por_doing))
Jaegeuk Kim029cd282012-12-21 17:20:21 +0900449 return false;
450
Chris Fries6c311ec2014-01-17 14:44:39 -0600451 return (free_sections(sbi) + freed) <= (node_secs + 2 * dent_secs +
452 reserved_sections(sbi));
Jaegeuk Kim39a53e02012-11-28 13:37:31 +0900453}
454
Jaegeuk Kim81eb8d62013-10-24 13:31:34 +0900455static inline bool excess_prefree_segs(struct f2fs_sb_info *sbi)
456{
Chris Fries6c311ec2014-01-17 14:44:39 -0600457 return prefree_segments(sbi) > SM_I(sbi)->rec_prefree_segments;
Jaegeuk Kim81eb8d62013-10-24 13:31:34 +0900458}
459
Jaegeuk Kim39a53e02012-11-28 13:37:31 +0900460static inline int utilization(struct f2fs_sb_info *sbi)
461{
Chris Fries6c311ec2014-01-17 14:44:39 -0600462 return div_u64((u64)valid_user_blocks(sbi) * 100,
463 sbi->user_block_count);
Jaegeuk Kim39a53e02012-11-28 13:37:31 +0900464}
465
466/*
467 * Sometimes f2fs may be better to drop out-of-place update policy.
Jaegeuk Kim216fbd62013-11-07 13:13:42 +0900468 * And, users can control the policy through sysfs entries.
469 * There are five policies with triggering conditions as follows.
470 * F2FS_IPU_FORCE - all the time,
471 * F2FS_IPU_SSR - if SSR mode is activated,
472 * F2FS_IPU_UTIL - if FS utilization is over threashold,
473 * F2FS_IPU_SSR_UTIL - if SSR mode is activated and FS utilization is over
474 * threashold,
Jaegeuk Kimc1ce1b02014-09-10 16:53:02 -0700475 * F2FS_IPU_FSYNC - activated in fsync path only for high performance flash
476 * storages. IPU will be triggered only if the # of dirty
477 * pages over min_fsync_blocks.
Jaegeuk Kim216fbd62013-11-07 13:13:42 +0900478 * F2FS_IPUT_DISABLE - disable IPU. (=default option)
Jaegeuk Kim39a53e02012-11-28 13:37:31 +0900479 */
Jaegeuk Kim216fbd62013-11-07 13:13:42 +0900480#define DEF_MIN_IPU_UTIL 70
Jaegeuk Kimc1ce1b02014-09-10 16:53:02 -0700481#define DEF_MIN_FSYNC_BLOCKS 8
Jaegeuk Kim216fbd62013-11-07 13:13:42 +0900482
483enum {
484 F2FS_IPU_FORCE,
485 F2FS_IPU_SSR,
486 F2FS_IPU_UTIL,
487 F2FS_IPU_SSR_UTIL,
Jaegeuk Kimc1ce1b02014-09-10 16:53:02 -0700488 F2FS_IPU_FSYNC,
Jaegeuk Kim216fbd62013-11-07 13:13:42 +0900489 F2FS_IPU_DISABLE,
490};
491
Jaegeuk Kim39a53e02012-11-28 13:37:31 +0900492static inline bool need_inplace_update(struct inode *inode)
493{
Jaegeuk Kim40813632014-09-02 15:31:18 -0700494 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
Jaegeuk Kim216fbd62013-11-07 13:13:42 +0900495
496 /* IPU can be done only for the user data */
Jaegeuk Kim39a53e02012-11-28 13:37:31 +0900497 if (S_ISDIR(inode->i_mode))
498 return false;
Jaegeuk Kim216fbd62013-11-07 13:13:42 +0900499
500 switch (SM_I(sbi)->ipu_policy) {
501 case F2FS_IPU_FORCE:
Jaegeuk Kim39a53e02012-11-28 13:37:31 +0900502 return true;
Jaegeuk Kim216fbd62013-11-07 13:13:42 +0900503 case F2FS_IPU_SSR:
504 if (need_SSR(sbi))
505 return true;
506 break;
507 case F2FS_IPU_UTIL:
508 if (utilization(sbi) > SM_I(sbi)->min_ipu_util)
509 return true;
510 break;
511 case F2FS_IPU_SSR_UTIL:
512 if (need_SSR(sbi) && utilization(sbi) > SM_I(sbi)->min_ipu_util)
513 return true;
514 break;
Jaegeuk Kimc1ce1b02014-09-10 16:53:02 -0700515 case F2FS_IPU_FSYNC:
516 /* this is only set during fdatasync */
517 if (is_inode_flag_set(F2FS_I(inode), FI_NEED_IPU))
518 return true;
519 break;
Jaegeuk Kim216fbd62013-11-07 13:13:42 +0900520 case F2FS_IPU_DISABLE:
521 break;
522 }
Jaegeuk Kim39a53e02012-11-28 13:37:31 +0900523 return false;
524}
525
526static inline unsigned int curseg_segno(struct f2fs_sb_info *sbi,
527 int type)
528{
529 struct curseg_info *curseg = CURSEG_I(sbi, type);
530 return curseg->segno;
531}
532
533static inline unsigned char curseg_alloc_type(struct f2fs_sb_info *sbi,
534 int type)
535{
536 struct curseg_info *curseg = CURSEG_I(sbi, type);
537 return curseg->alloc_type;
538}
539
540static inline unsigned short curseg_blkoff(struct f2fs_sb_info *sbi, int type)
541{
542 struct curseg_info *curseg = CURSEG_I(sbi, type);
543 return curseg->next_blkoff;
544}
545
Jaegeuk Kim5d56b672013-10-29 15:14:54 +0900546#ifdef CONFIG_F2FS_CHECK_FS
Jaegeuk Kim39a53e02012-11-28 13:37:31 +0900547static inline void check_seg_range(struct f2fs_sb_info *sbi, unsigned int segno)
548{
549 unsigned int end_segno = SM_I(sbi)->segment_count - 1;
550 BUG_ON(segno > end_segno);
551}
552
Jaegeuk Kim39a53e02012-11-28 13:37:31 +0900553static inline void verify_block_addr(struct f2fs_sb_info *sbi, block_t blk_addr)
554{
555 struct f2fs_sm_info *sm_info = SM_I(sbi);
556 block_t total_blks = sm_info->segment_count << sbi->log_blocks_per_seg;
557 block_t start_addr = sm_info->seg0_blkaddr;
558 block_t end_addr = start_addr + total_blks - 1;
559 BUG_ON(blk_addr < start_addr);
560 BUG_ON(blk_addr > end_addr);
561}
562
563/*
arter97e1c42042014-08-06 23:22:50 +0900564 * Summary block is always treated as an invalid block
Jaegeuk Kim39a53e02012-11-28 13:37:31 +0900565 */
566static inline void check_block_count(struct f2fs_sb_info *sbi,
567 int segno, struct f2fs_sit_entry *raw_sit)
568{
569 struct f2fs_sm_info *sm_info = SM_I(sbi);
570 unsigned int end_segno = sm_info->segment_count - 1;
Chao Yu44c60bf2013-10-29 14:50:40 +0800571 bool is_valid = test_bit_le(0, raw_sit->valid_map) ? true : false;
Jaegeuk Kim39a53e02012-11-28 13:37:31 +0900572 int valid_blocks = 0;
Chao Yu44c60bf2013-10-29 14:50:40 +0800573 int cur_pos = 0, next_pos;
Jaegeuk Kim39a53e02012-11-28 13:37:31 +0900574
575 /* check segment usage */
576 BUG_ON(GET_SIT_VBLOCKS(raw_sit) > sbi->blocks_per_seg);
577
578 /* check boundary of a given segment number */
579 BUG_ON(segno > end_segno);
580
581 /* check bitmap with valid block count */
Chao Yu44c60bf2013-10-29 14:50:40 +0800582 do {
583 if (is_valid) {
584 next_pos = find_next_zero_bit_le(&raw_sit->valid_map,
585 sbi->blocks_per_seg,
586 cur_pos);
587 valid_blocks += next_pos - cur_pos;
588 } else
589 next_pos = find_next_bit_le(&raw_sit->valid_map,
590 sbi->blocks_per_seg,
591 cur_pos);
592 cur_pos = next_pos;
593 is_valid = !is_valid;
594 } while (cur_pos < sbi->blocks_per_seg);
Jaegeuk Kim39a53e02012-11-28 13:37:31 +0900595 BUG_ON(GET_SIT_VBLOCKS(raw_sit) != valid_blocks);
596}
Jaegeuk Kim5d56b672013-10-29 15:14:54 +0900597#else
Jaegeuk Kim05796762014-09-02 16:05:00 -0700598static inline void check_seg_range(struct f2fs_sb_info *sbi, unsigned int segno)
599{
600 unsigned int end_segno = SM_I(sbi)->segment_count - 1;
601
602 if (segno > end_segno)
603 sbi->need_fsck = true;
604}
605
606static inline void verify_block_addr(struct f2fs_sb_info *sbi, block_t blk_addr)
607{
608 struct f2fs_sm_info *sm_info = SM_I(sbi);
609 block_t total_blks = sm_info->segment_count << sbi->log_blocks_per_seg;
610 block_t start_addr = sm_info->seg0_blkaddr;
611 block_t end_addr = start_addr + total_blks - 1;
612
613 if (blk_addr < start_addr || blk_addr > end_addr)
614 sbi->need_fsck = true;
615}
616
617/*
618 * Summary block is always treated as an invalid block
619 */
620static inline void check_block_count(struct f2fs_sb_info *sbi,
621 int segno, struct f2fs_sit_entry *raw_sit)
622{
623 unsigned int end_segno = SM_I(sbi)->segment_count - 1;
624
625 /* check segment usage */
626 if (GET_SIT_VBLOCKS(raw_sit) > sbi->blocks_per_seg)
627 sbi->need_fsck = true;
628
629 /* check boundary of a given segment number */
630 if (segno > end_segno)
631 sbi->need_fsck = true;
632}
Jaegeuk Kim5d56b672013-10-29 15:14:54 +0900633#endif
Jaegeuk Kim39a53e02012-11-28 13:37:31 +0900634
635static inline pgoff_t current_sit_addr(struct f2fs_sb_info *sbi,
636 unsigned int start)
637{
638 struct sit_info *sit_i = SIT_I(sbi);
Chao Yud3a14af2014-09-04 18:11:47 +0800639 unsigned int offset = SIT_BLOCK_OFFSET(start);
Jaegeuk Kim39a53e02012-11-28 13:37:31 +0900640 block_t blk_addr = sit_i->sit_base_addr + offset;
641
642 check_seg_range(sbi, start);
643
644 /* calculate sit block address */
645 if (f2fs_test_bit(offset, sit_i->sit_bitmap))
646 blk_addr += sit_i->sit_blocks;
647
648 return blk_addr;
649}
650
651static inline pgoff_t next_sit_addr(struct f2fs_sb_info *sbi,
652 pgoff_t block_addr)
653{
654 struct sit_info *sit_i = SIT_I(sbi);
655 block_addr -= sit_i->sit_base_addr;
656 if (block_addr < sit_i->sit_blocks)
657 block_addr += sit_i->sit_blocks;
658 else
659 block_addr -= sit_i->sit_blocks;
660
661 return block_addr + sit_i->sit_base_addr;
662}
663
664static inline void set_to_next_sit(struct sit_info *sit_i, unsigned int start)
665{
Chao Yud3a14af2014-09-04 18:11:47 +0800666 unsigned int block_off = SIT_BLOCK_OFFSET(start);
Jaegeuk Kim39a53e02012-11-28 13:37:31 +0900667
668 if (f2fs_test_bit(block_off, sit_i->sit_bitmap))
669 f2fs_clear_bit(block_off, sit_i->sit_bitmap);
670 else
671 f2fs_set_bit(block_off, sit_i->sit_bitmap);
672}
673
674static inline unsigned long long get_mtime(struct f2fs_sb_info *sbi)
675{
676 struct sit_info *sit_i = SIT_I(sbi);
677 return sit_i->elapsed_time + CURRENT_TIME_SEC.tv_sec -
678 sit_i->mounted_time;
679}
680
681static inline void set_summary(struct f2fs_summary *sum, nid_t nid,
682 unsigned int ofs_in_node, unsigned char version)
683{
684 sum->nid = cpu_to_le32(nid);
685 sum->ofs_in_node = cpu_to_le16(ofs_in_node);
686 sum->version = version;
687}
688
689static inline block_t start_sum_block(struct f2fs_sb_info *sbi)
690{
691 return __start_cp_addr(sbi) +
692 le32_to_cpu(F2FS_CKPT(sbi)->cp_pack_start_sum);
693}
694
695static inline block_t sum_blk_addr(struct f2fs_sb_info *sbi, int base, int type)
696{
697 return __start_cp_addr(sbi) +
698 le32_to_cpu(F2FS_CKPT(sbi)->cp_pack_total_block_count)
699 - (base + 1) + type;
700}
Jaegeuk Kim5ec4e492013-03-31 13:26:03 +0900701
702static inline bool sec_usage_check(struct f2fs_sb_info *sbi, unsigned int secno)
703{
704 if (IS_CURSEC(sbi, secno) || (sbi->cur_victim_sec == secno))
705 return true;
706 return false;
707}
Jaegeuk Kimac5d1562013-04-29 16:58:39 +0900708
709static inline unsigned int max_hw_blocks(struct f2fs_sb_info *sbi)
710{
711 struct block_device *bdev = sbi->sb->s_bdev;
712 struct request_queue *q = bdev_get_queue(bdev);
713 return SECTOR_TO_BLOCK(sbi, queue_max_sectors(q));
714}
Jaegeuk Kim87d6f892014-03-18 12:40:49 +0900715
716/*
717 * It is very important to gather dirty pages and write at once, so that we can
718 * submit a big bio without interfering other data writes.
719 * By default, 512 pages for directory data,
720 * 512 pages (2MB) * 3 for three types of nodes, and
721 * max_bio_blocks for meta are set.
722 */
723static inline int nr_pages_to_skip(struct f2fs_sb_info *sbi, int type)
724{
725 if (type == DATA)
726 return sbi->blocks_per_seg;
727 else if (type == NODE)
728 return 3 * sbi->blocks_per_seg;
729 else if (type == META)
730 return MAX_BIO_BLOCKS(max_hw_blocks(sbi));
731 else
732 return 0;
733}
Jaegeuk Kim50c8cdb2014-03-18 13:47:11 +0900734
735/*
736 * When writing pages, it'd better align nr_to_write for segment size.
737 */
738static inline long nr_pages_to_write(struct f2fs_sb_info *sbi, int type,
739 struct writeback_control *wbc)
740{
741 long nr_to_write, desired;
742
743 if (wbc->sync_mode != WB_SYNC_NONE)
744 return 0;
745
746 nr_to_write = wbc->nr_to_write;
747
748 if (type == DATA)
749 desired = 4096;
750 else if (type == NODE)
751 desired = 3 * max_hw_blocks(sbi);
752 else
753 desired = MAX_BIO_BLOCKS(max_hw_blocks(sbi));
754
755 wbc->nr_to_write = desired;
756 return desired - nr_to_write;
757}