blob: 5d016a1edf21bdc66182912e3c0ed5d7588e66c8 [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>
Tejun Heo66114ca2015-05-22 17:13:32 -040012#include <linux/backing-dev.h>
Jaegeuk Kimac5d1562013-04-29 16:58:39 +090013
Jaegeuk Kim39a53e02012-11-28 13:37:31 +090014/* constant macro */
15#define NULL_SEGNO ((unsigned int)(~0))
Jaegeuk Kim5ec4e492013-03-31 13:26:03 +090016#define NULL_SECNO ((unsigned int)(~0))
Jaegeuk Kim39a53e02012-11-28 13:37:31 +090017
Jaegeuk Kim58c41032014-03-19 14:17:21 +090018#define DEF_RECLAIM_PREFREE_SEGMENTS 5 /* 5% over total segments */
Jaegeuk Kim81eb8d62013-10-24 13:31:34 +090019
Namjae Jeon6224da82013-04-06 14:44:32 +090020/* L: Logical segment # in volume, R: Relative segment # in main area */
Jaegeuk Kim39a53e02012-11-28 13:37:31 +090021#define GET_L2R_SEGNO(free_i, segno) (segno - free_i->start_segno)
22#define GET_R2L_SEGNO(free_i, segno) (segno + free_i->start_segno)
23
Changman Lee61ae45c2013-11-21 20:04:21 +090024#define IS_DATASEG(t) (t <= CURSEG_COLD_DATA)
25#define IS_NODESEG(t) (t >= CURSEG_HOT_NODE)
Jaegeuk Kim39a53e02012-11-28 13:37:31 +090026
Jaegeuk Kim5c773ba2013-03-31 12:30:04 +090027#define IS_CURSEG(sbi, seg) \
28 ((seg == CURSEG_I(sbi, CURSEG_HOT_DATA)->segno) || \
29 (seg == CURSEG_I(sbi, CURSEG_WARM_DATA)->segno) || \
30 (seg == CURSEG_I(sbi, CURSEG_COLD_DATA)->segno) || \
31 (seg == CURSEG_I(sbi, CURSEG_HOT_NODE)->segno) || \
32 (seg == CURSEG_I(sbi, CURSEG_WARM_NODE)->segno) || \
33 (seg == CURSEG_I(sbi, CURSEG_COLD_NODE)->segno))
Jaegeuk Kim39a53e02012-11-28 13:37:31 +090034
35#define IS_CURSEC(sbi, secno) \
36 ((secno == CURSEG_I(sbi, CURSEG_HOT_DATA)->segno / \
37 sbi->segs_per_sec) || \
38 (secno == CURSEG_I(sbi, CURSEG_WARM_DATA)->segno / \
39 sbi->segs_per_sec) || \
40 (secno == CURSEG_I(sbi, CURSEG_COLD_DATA)->segno / \
41 sbi->segs_per_sec) || \
42 (secno == CURSEG_I(sbi, CURSEG_HOT_NODE)->segno / \
43 sbi->segs_per_sec) || \
44 (secno == CURSEG_I(sbi, CURSEG_WARM_NODE)->segno / \
45 sbi->segs_per_sec) || \
46 (secno == CURSEG_I(sbi, CURSEG_COLD_NODE)->segno / \
47 sbi->segs_per_sec)) \
48
Jaegeuk Kim7cd85582014-09-23 11:23:01 -070049#define MAIN_BLKADDR(sbi) (SM_I(sbi)->main_blkaddr)
50#define SEG0_BLKADDR(sbi) (SM_I(sbi)->seg0_blkaddr)
51
52#define MAIN_SEGS(sbi) (SM_I(sbi)->main_segments)
53#define MAIN_SECS(sbi) (sbi->total_sections)
54
55#define TOTAL_SEGS(sbi) (SM_I(sbi)->segment_count)
56#define TOTAL_BLKS(sbi) (TOTAL_SEGS(sbi) << sbi->log_blocks_per_seg)
57
58#define MAX_BLKADDR(sbi) (SEG0_BLKADDR(sbi) + TOTAL_BLKS(sbi))
Dan Carpenter8a219842014-09-25 14:39:17 +030059#define SEGMENT_SIZE(sbi) (1ULL << (sbi->log_blocksize + \
Jaegeuk Kim7cd85582014-09-23 11:23:01 -070060 sbi->log_blocks_per_seg))
61
62#define START_BLOCK(sbi, segno) (SEG0_BLKADDR(sbi) + \
Jaegeuk Kim39a53e02012-11-28 13:37:31 +090063 (GET_R2L_SEGNO(FREE_I(sbi), segno) << sbi->log_blocks_per_seg))
Jaegeuk Kim7cd85582014-09-23 11:23:01 -070064
Jaegeuk Kim39a53e02012-11-28 13:37:31 +090065#define NEXT_FREE_BLKADDR(sbi, curseg) \
66 (START_BLOCK(sbi, curseg->segno) + curseg->next_blkoff)
67
Jaegeuk Kim7cd85582014-09-23 11:23:01 -070068#define GET_SEGOFF_FROM_SEG0(sbi, blk_addr) ((blk_addr) - SEG0_BLKADDR(sbi))
Jaegeuk Kim39a53e02012-11-28 13:37:31 +090069#define GET_SEGNO_FROM_SEG0(sbi, blk_addr) \
70 (GET_SEGOFF_FROM_SEG0(sbi, blk_addr) >> sbi->log_blocks_per_seg)
Jaegeuk Kim491c0852014-02-04 13:01:10 +090071#define GET_BLKOFF_FROM_SEG0(sbi, blk_addr) \
72 (GET_SEGOFF_FROM_SEG0(sbi, blk_addr) & (sbi->blocks_per_seg - 1))
73
Jaegeuk Kim39a53e02012-11-28 13:37:31 +090074#define GET_SEGNO(sbi, blk_addr) \
75 (((blk_addr == NULL_ADDR) || (blk_addr == NEW_ADDR)) ? \
76 NULL_SEGNO : GET_L2R_SEGNO(FREE_I(sbi), \
77 GET_SEGNO_FROM_SEG0(sbi, blk_addr)))
78#define GET_SECNO(sbi, segno) \
79 ((segno) / sbi->segs_per_sec)
80#define GET_ZONENO_FROM_SEGNO(sbi, segno) \
81 ((segno / sbi->segs_per_sec) / sbi->secs_per_zone)
82
83#define GET_SUM_BLOCK(sbi, segno) \
84 ((sbi->sm_info->ssa_blkaddr) + segno)
85
86#define GET_SUM_TYPE(footer) ((footer)->entry_type)
87#define SET_SUM_TYPE(footer, type) ((footer)->entry_type = type)
88
89#define SIT_ENTRY_OFFSET(sit_i, segno) \
90 (segno % sit_i->sents_per_block)
Chao Yud3a14af2014-09-04 18:11:47 +080091#define SIT_BLOCK_OFFSET(segno) \
Jaegeuk Kim39a53e02012-11-28 13:37:31 +090092 (segno / SIT_ENTRY_PER_BLOCK)
Chao Yud3a14af2014-09-04 18:11:47 +080093#define START_SEGNO(segno) \
94 (SIT_BLOCK_OFFSET(segno) * SIT_ENTRY_PER_BLOCK)
Chao Yu74de5932013-11-22 09:09:59 +080095#define SIT_BLK_CNT(sbi) \
Jaegeuk Kim7cd85582014-09-23 11:23:01 -070096 ((MAIN_SEGS(sbi) + SIT_ENTRY_PER_BLOCK - 1) / SIT_ENTRY_PER_BLOCK)
Jaegeuk Kim39a53e02012-11-28 13:37:31 +090097#define f2fs_bitmap_size(nr) \
98 (BITS_TO_LONGS(nr) * sizeof(unsigned long))
Jaegeuk Kim39a53e02012-11-28 13:37:31 +090099
Chao Yu55cf9cb2014-09-15 18:01:10 +0800100#define SECTOR_FROM_BLOCK(blk_addr) \
101 (((sector_t)blk_addr) << F2FS_LOG_SECTORS_PER_BLOCK)
102#define SECTOR_TO_BLOCK(sectors) \
103 (sectors >> F2FS_LOG_SECTORS_PER_BLOCK)
Jaegeuk Kim90a893c2014-09-22 16:21:07 -0700104#define MAX_BIO_BLOCKS(sbi) \
105 ((int)min((int)max_hw_blocks(sbi), BIO_MAX_PAGES))
Jaegeuk Kim3cd8a232012-12-10 09:26:05 +0900106
Jaegeuk Kim39a53e02012-11-28 13:37:31 +0900107/*
108 * indicate a block allocation direction: RIGHT and LEFT.
109 * RIGHT means allocating new sections towards the end of volume.
110 * LEFT means the opposite direction.
111 */
112enum {
113 ALLOC_RIGHT = 0,
114 ALLOC_LEFT
115};
116
117/*
118 * In the victim_sel_policy->alloc_mode, there are two block allocation modes.
119 * LFS writes data sequentially with cleaning operations.
120 * SSR (Slack Space Recycle) reuses obsolete space without cleaning operations.
121 */
122enum {
123 LFS = 0,
124 SSR
125};
126
127/*
128 * In the victim_sel_policy->gc_mode, there are two gc, aka cleaning, modes.
129 * GC_CB is based on cost-benefit algorithm.
130 * GC_GREEDY is based on greedy algorithm.
131 */
132enum {
133 GC_CB = 0,
134 GC_GREEDY
135};
136
137/*
138 * BG_GC means the background cleaning job.
139 * FG_GC means the on-demand cleaning job.
Jaegeuk Kim6aefd932015-10-05 11:02:54 -0700140 * FORCE_FG_GC means on-demand cleaning job in background.
Jaegeuk Kim39a53e02012-11-28 13:37:31 +0900141 */
142enum {
143 BG_GC = 0,
Jaegeuk Kim6aefd932015-10-05 11:02:54 -0700144 FG_GC,
145 FORCE_FG_GC,
Jaegeuk Kim39a53e02012-11-28 13:37:31 +0900146};
147
148/* for a function parameter to select a victim segment */
149struct victim_sel_policy {
150 int alloc_mode; /* LFS or SSR */
151 int gc_mode; /* GC_CB or GC_GREEDY */
152 unsigned long *dirty_segmap; /* dirty segment bitmap */
Jin Xua26b7c82013-09-05 12:45:26 +0800153 unsigned int max_search; /* maximum # of segments to search */
Jaegeuk Kim39a53e02012-11-28 13:37:31 +0900154 unsigned int offset; /* last scanned bitmap offset */
155 unsigned int ofs_unit; /* bitmap search unit */
156 unsigned int min_cost; /* minimum cost */
157 unsigned int min_segno; /* segment # having min. cost */
158};
159
160struct seg_entry {
Chao Yuf51b4ce2016-05-04 23:19:48 +0800161 unsigned int type:6; /* segment type like CURSEG_XXX_TYPE */
162 unsigned int valid_blocks:10; /* # of valid blocks */
163 unsigned int ckpt_valid_blocks:10; /* # of valid blocks last cp */
164 unsigned int padding:6; /* padding */
Jaegeuk Kim39a53e02012-11-28 13:37:31 +0900165 unsigned char *cur_valid_map; /* validity bitmap of blocks */
166 /*
167 * # of valid blocks and the validity bitmap stored in the the last
168 * checkpoint pack. This information is used by the SSR mode.
169 */
Chao Yuf51b4ce2016-05-04 23:19:48 +0800170 unsigned char *ckpt_valid_map; /* validity bitmap of blocks last cp */
Jaegeuk Kima66cdd92015-04-30 22:37:50 -0700171 unsigned char *discard_map;
Jaegeuk Kim39a53e02012-11-28 13:37:31 +0900172 unsigned long long mtime; /* modification time of the segment */
173};
174
175struct sec_entry {
176 unsigned int valid_blocks; /* # of valid blocks in a section */
177};
178
179struct segment_allocation {
180 void (*allocate_segment)(struct f2fs_sb_info *, int, bool);
181};
182
Chao Yudecd36b2015-08-07 18:42:09 +0800183/*
184 * this value is set in page as a private data which indicate that
185 * the page is atomically written, and it is in inmem_pages list.
186 */
Jaegeuk Kimd48dfc22016-01-29 16:21:15 -0800187#define ATOMIC_WRITTEN_PAGE ((unsigned long)-1)
Chao Yudecd36b2015-08-07 18:42:09 +0800188
189#define IS_ATOMIC_WRITTEN_PAGE(page) \
190 (page_private(page) == (unsigned long)ATOMIC_WRITTEN_PAGE)
191
Jaegeuk Kim88b88a62014-10-06 17:39:50 -0700192struct inmem_pages {
193 struct list_head list;
194 struct page *page;
Chao Yu28bc1062016-02-06 14:40:34 +0800195 block_t old_addr; /* for revoking when fail to commit */
Jaegeuk Kim88b88a62014-10-06 17:39:50 -0700196};
197
Jaegeuk Kim39a53e02012-11-28 13:37:31 +0900198struct sit_info {
199 const struct segment_allocation *s_ops;
200
201 block_t sit_base_addr; /* start block address of SIT area */
202 block_t sit_blocks; /* # of blocks used by SIT area */
203 block_t written_valid_blocks; /* # of valid blocks in main area */
204 char *sit_bitmap; /* SIT bitmap pointer */
205 unsigned int bitmap_size; /* SIT bitmap size */
206
Jaegeuk Kim60a3b782015-02-10 16:44:29 -0800207 unsigned long *tmp_map; /* bitmap for temporal use */
Jaegeuk Kim39a53e02012-11-28 13:37:31 +0900208 unsigned long *dirty_sentries_bitmap; /* bitmap for dirty sentries */
209 unsigned int dirty_sentries; /* # of dirty sentries */
210 unsigned int sents_per_block; /* # of SIT entries per block */
211 struct mutex sentry_lock; /* to protect SIT cache */
212 struct seg_entry *sentries; /* SIT segment-level cache */
213 struct sec_entry *sec_entries; /* SIT section-level cache */
214
215 /* for cost-benefit algorithm in cleaning procedure */
216 unsigned long long elapsed_time; /* elapsed time after mount */
217 unsigned long long mounted_time; /* mount time */
218 unsigned long long min_mtime; /* min. modification time */
219 unsigned long long max_mtime; /* max. modification time */
220};
221
222struct free_segmap_info {
223 unsigned int start_segno; /* start segment number logically */
224 unsigned int free_segments; /* # of free segments */
225 unsigned int free_sections; /* # of free sections */
Chao Yu1a118cc2015-02-11 18:20:38 +0800226 spinlock_t segmap_lock; /* free segmap lock */
Jaegeuk Kim39a53e02012-11-28 13:37:31 +0900227 unsigned long *free_segmap; /* free segment bitmap */
228 unsigned long *free_secmap; /* free section bitmap */
229};
230
231/* Notice: The order of dirty type is same with CURSEG_XXX in f2fs.h */
232enum dirty_type {
233 DIRTY_HOT_DATA, /* dirty segments assigned as hot data logs */
234 DIRTY_WARM_DATA, /* dirty segments assigned as warm data logs */
235 DIRTY_COLD_DATA, /* dirty segments assigned as cold data logs */
236 DIRTY_HOT_NODE, /* dirty segments assigned as hot node logs */
237 DIRTY_WARM_NODE, /* dirty segments assigned as warm node logs */
238 DIRTY_COLD_NODE, /* dirty segments assigned as cold node logs */
239 DIRTY, /* to count # of dirty segments */
240 PRE, /* to count # of entirely obsolete segments */
241 NR_DIRTY_TYPE
242};
243
244struct dirty_seglist_info {
245 const struct victim_selection *v_ops; /* victim selction operation */
246 unsigned long *dirty_segmap[NR_DIRTY_TYPE];
247 struct mutex seglist_lock; /* lock for segment bitmaps */
248 int nr_dirty[NR_DIRTY_TYPE]; /* # of dirty segments */
Jaegeuk Kim5ec4e492013-03-31 13:26:03 +0900249 unsigned long *victim_secmap; /* background GC victims */
Jaegeuk Kim39a53e02012-11-28 13:37:31 +0900250};
251
252/* victim selection function for cleaning and SSR */
253struct victim_selection {
254 int (*get_victim)(struct f2fs_sb_info *, unsigned int *,
255 int, int, char);
256};
257
258/* for active log information */
259struct curseg_info {
260 struct mutex curseg_mutex; /* lock for consistency */
261 struct f2fs_summary_block *sum_blk; /* cached summary block */
Chao Yub7ad7512016-02-19 18:08:46 +0800262 struct rw_semaphore journal_rwsem; /* protect journal area */
263 struct f2fs_journal *journal; /* cached journal info */
Jaegeuk Kim39a53e02012-11-28 13:37:31 +0900264 unsigned char alloc_type; /* current allocation type */
265 unsigned int segno; /* current segment number */
266 unsigned short next_blkoff; /* next block offset to write */
267 unsigned int zone; /* current zone number */
268 unsigned int next_segno; /* preallocated segment */
269};
270
Chao Yu184a5cd2014-09-04 18:13:01 +0800271struct sit_entry_set {
272 struct list_head set_list; /* link with all sit sets */
273 unsigned int start_segno; /* start segno of sits in set */
274 unsigned int entry_cnt; /* the # of sit entries in set */
275};
276
Jaegeuk Kim39a53e02012-11-28 13:37:31 +0900277/*
278 * inline functions
279 */
280static inline struct curseg_info *CURSEG_I(struct f2fs_sb_info *sbi, int type)
281{
282 return (struct curseg_info *)(SM_I(sbi)->curseg_array + type);
283}
284
285static inline struct seg_entry *get_seg_entry(struct f2fs_sb_info *sbi,
286 unsigned int segno)
287{
288 struct sit_info *sit_i = SIT_I(sbi);
289 return &sit_i->sentries[segno];
290}
291
292static inline struct sec_entry *get_sec_entry(struct f2fs_sb_info *sbi,
293 unsigned int segno)
294{
295 struct sit_info *sit_i = SIT_I(sbi);
296 return &sit_i->sec_entries[GET_SECNO(sbi, segno)];
297}
298
299static inline unsigned int get_valid_blocks(struct f2fs_sb_info *sbi,
300 unsigned int segno, int section)
301{
302 /*
303 * In order to get # of valid blocks in a section instantly from many
304 * segments, f2fs manages two counting structures separately.
305 */
306 if (section > 1)
307 return get_sec_entry(sbi, segno)->valid_blocks;
308 else
309 return get_seg_entry(sbi, segno)->valid_blocks;
310}
311
312static inline void seg_info_from_raw_sit(struct seg_entry *se,
313 struct f2fs_sit_entry *rs)
314{
315 se->valid_blocks = GET_SIT_VBLOCKS(rs);
316 se->ckpt_valid_blocks = GET_SIT_VBLOCKS(rs);
317 memcpy(se->cur_valid_map, rs->valid_map, SIT_VBLOCK_MAP_SIZE);
318 memcpy(se->ckpt_valid_map, rs->valid_map, SIT_VBLOCK_MAP_SIZE);
319 se->type = GET_SIT_TYPE(rs);
320 se->mtime = le64_to_cpu(rs->mtime);
321}
322
323static inline void seg_info_to_raw_sit(struct seg_entry *se,
324 struct f2fs_sit_entry *rs)
325{
326 unsigned short raw_vblocks = (se->type << SIT_VBLOCKS_SHIFT) |
327 se->valid_blocks;
328 rs->vblocks = cpu_to_le16(raw_vblocks);
329 memcpy(rs->valid_map, se->cur_valid_map, SIT_VBLOCK_MAP_SIZE);
330 memcpy(se->ckpt_valid_map, rs->valid_map, SIT_VBLOCK_MAP_SIZE);
331 se->ckpt_valid_blocks = se->valid_blocks;
332 rs->mtime = cpu_to_le64(se->mtime);
333}
334
335static inline unsigned int find_next_inuse(struct free_segmap_info *free_i,
336 unsigned int max, unsigned int segno)
337{
338 unsigned int ret;
Chao Yu1a118cc2015-02-11 18:20:38 +0800339 spin_lock(&free_i->segmap_lock);
Jaegeuk Kim39a53e02012-11-28 13:37:31 +0900340 ret = find_next_bit(free_i->free_segmap, max, segno);
Chao Yu1a118cc2015-02-11 18:20:38 +0800341 spin_unlock(&free_i->segmap_lock);
Jaegeuk Kim39a53e02012-11-28 13:37:31 +0900342 return ret;
343}
344
345static inline void __set_free(struct f2fs_sb_info *sbi, 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
Chao Yu1a118cc2015-02-11 18:20:38 +0800352 spin_lock(&free_i->segmap_lock);
Jaegeuk Kim39a53e02012-11-28 13:37:31 +0900353 clear_bit(segno, free_i->free_segmap);
354 free_i->free_segments++;
355
Wanpeng Li7fd97012015-03-06 15:00:55 +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 clear_bit(secno, free_i->free_secmap);
360 free_i->free_sections++;
361 }
Chao Yu1a118cc2015-02-11 18:20:38 +0800362 spin_unlock(&free_i->segmap_lock);
Jaegeuk Kim39a53e02012-11-28 13:37:31 +0900363}
364
365static inline void __set_inuse(struct f2fs_sb_info *sbi,
366 unsigned int segno)
367{
368 struct free_segmap_info *free_i = FREE_I(sbi);
369 unsigned int secno = segno / sbi->segs_per_sec;
370 set_bit(segno, free_i->free_segmap);
371 free_i->free_segments--;
372 if (!test_and_set_bit(secno, free_i->free_secmap))
373 free_i->free_sections--;
374}
375
376static inline void __set_test_and_free(struct f2fs_sb_info *sbi,
377 unsigned int segno)
378{
379 struct free_segmap_info *free_i = FREE_I(sbi);
380 unsigned int secno = segno / sbi->segs_per_sec;
381 unsigned int start_segno = secno * sbi->segs_per_sec;
382 unsigned int next;
383
Chao Yu1a118cc2015-02-11 18:20:38 +0800384 spin_lock(&free_i->segmap_lock);
Jaegeuk Kim39a53e02012-11-28 13:37:31 +0900385 if (test_and_clear_bit(segno, free_i->free_segmap)) {
386 free_i->free_segments++;
387
Chao Yuf1121ab2014-07-14 16:45:15 +0800388 next = find_next_bit(free_i->free_segmap,
389 start_segno + sbi->segs_per_sec, start_segno);
Jaegeuk Kim39a53e02012-11-28 13:37:31 +0900390 if (next >= start_segno + sbi->segs_per_sec) {
391 if (test_and_clear_bit(secno, free_i->free_secmap))
392 free_i->free_sections++;
393 }
394 }
Chao Yu1a118cc2015-02-11 18:20:38 +0800395 spin_unlock(&free_i->segmap_lock);
Jaegeuk Kim39a53e02012-11-28 13:37:31 +0900396}
397
398static inline void __set_test_and_inuse(struct f2fs_sb_info *sbi,
399 unsigned int segno)
400{
401 struct free_segmap_info *free_i = FREE_I(sbi);
402 unsigned int secno = segno / sbi->segs_per_sec;
Chao Yu1a118cc2015-02-11 18:20:38 +0800403 spin_lock(&free_i->segmap_lock);
Jaegeuk Kim39a53e02012-11-28 13:37:31 +0900404 if (!test_and_set_bit(segno, free_i->free_segmap)) {
405 free_i->free_segments--;
406 if (!test_and_set_bit(secno, free_i->free_secmap))
407 free_i->free_sections--;
408 }
Chao Yu1a118cc2015-02-11 18:20:38 +0800409 spin_unlock(&free_i->segmap_lock);
Jaegeuk Kim39a53e02012-11-28 13:37:31 +0900410}
411
412static inline void get_sit_bitmap(struct f2fs_sb_info *sbi,
413 void *dst_addr)
414{
415 struct sit_info *sit_i = SIT_I(sbi);
416 memcpy(dst_addr, sit_i->sit_bitmap, sit_i->bitmap_size);
417}
418
419static inline block_t written_block_count(struct f2fs_sb_info *sbi)
420{
Jaegeuk Kim8b8343f2014-02-24 13:00:13 +0900421 return SIT_I(sbi)->written_valid_blocks;
Jaegeuk Kim39a53e02012-11-28 13:37:31 +0900422}
423
424static inline unsigned int free_segments(struct f2fs_sb_info *sbi)
425{
Jaegeuk Kim8b8343f2014-02-24 13:00:13 +0900426 return FREE_I(sbi)->free_segments;
Jaegeuk Kim39a53e02012-11-28 13:37:31 +0900427}
428
429static inline int reserved_segments(struct f2fs_sb_info *sbi)
430{
431 return SM_I(sbi)->reserved_segments;
432}
433
434static inline unsigned int free_sections(struct f2fs_sb_info *sbi)
435{
Jaegeuk Kim8b8343f2014-02-24 13:00:13 +0900436 return FREE_I(sbi)->free_sections;
Jaegeuk Kim39a53e02012-11-28 13:37:31 +0900437}
438
439static inline unsigned int prefree_segments(struct f2fs_sb_info *sbi)
440{
441 return DIRTY_I(sbi)->nr_dirty[PRE];
442}
443
444static inline unsigned int dirty_segments(struct f2fs_sb_info *sbi)
445{
446 return DIRTY_I(sbi)->nr_dirty[DIRTY_HOT_DATA] +
447 DIRTY_I(sbi)->nr_dirty[DIRTY_WARM_DATA] +
448 DIRTY_I(sbi)->nr_dirty[DIRTY_COLD_DATA] +
449 DIRTY_I(sbi)->nr_dirty[DIRTY_HOT_NODE] +
450 DIRTY_I(sbi)->nr_dirty[DIRTY_WARM_NODE] +
451 DIRTY_I(sbi)->nr_dirty[DIRTY_COLD_NODE];
452}
453
454static inline int overprovision_segments(struct f2fs_sb_info *sbi)
455{
456 return SM_I(sbi)->ovp_segments;
457}
458
459static inline int overprovision_sections(struct f2fs_sb_info *sbi)
460{
461 return ((unsigned int) overprovision_segments(sbi)) / sbi->segs_per_sec;
462}
463
464static inline int reserved_sections(struct f2fs_sb_info *sbi)
465{
466 return ((unsigned int) reserved_segments(sbi)) / sbi->segs_per_sec;
467}
468
469static inline bool need_SSR(struct f2fs_sb_info *sbi)
470{
Jaegeuk Kim95dd8972014-09-17 17:52:58 -0700471 int node_secs = get_blocktype_secs(sbi, F2FS_DIRTY_NODES);
472 int dent_secs = get_blocktype_secs(sbi, F2FS_DIRTY_DENTS);
473 return free_sections(sbi) <= (node_secs + 2 * dent_secs +
474 reserved_sections(sbi) + 1);
Jaegeuk Kim39a53e02012-11-28 13:37:31 +0900475}
476
Jaegeuk Kim43727522013-02-04 15:11:17 +0900477static inline bool has_not_enough_free_secs(struct f2fs_sb_info *sbi, int freed)
Jaegeuk Kim39a53e02012-11-28 13:37:31 +0900478{
Namjae Jeon5ac206c2013-02-02 23:52:59 +0900479 int node_secs = get_blocktype_secs(sbi, F2FS_DIRTY_NODES);
480 int dent_secs = get_blocktype_secs(sbi, F2FS_DIRTY_DENTS);
Jaegeuk Kim43727522013-02-04 15:11:17 +0900481
Jaegeuk Kim0f18b462016-05-20 11:10:10 -0700482 node_secs += get_blocktype_secs(sbi, F2FS_DIRTY_IMETA);
483
Chao Yucaf00472015-01-28 17:48:42 +0800484 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
Jaegeuk Kim029cd282012-12-21 17:20:21 +0900485 return false;
486
Chris Fries6c311ec2014-01-17 14:44:39 -0600487 return (free_sections(sbi) + freed) <= (node_secs + 2 * dent_secs +
488 reserved_sections(sbi));
Jaegeuk Kim39a53e02012-11-28 13:37:31 +0900489}
490
Jaegeuk Kim81eb8d62013-10-24 13:31:34 +0900491static inline bool excess_prefree_segs(struct f2fs_sb_info *sbi)
492{
Chris Fries6c311ec2014-01-17 14:44:39 -0600493 return prefree_segments(sbi) > SM_I(sbi)->rec_prefree_segments;
Jaegeuk Kim81eb8d62013-10-24 13:31:34 +0900494}
495
Jaegeuk Kim39a53e02012-11-28 13:37:31 +0900496static inline int utilization(struct f2fs_sb_info *sbi)
497{
Chris Fries6c311ec2014-01-17 14:44:39 -0600498 return div_u64((u64)valid_user_blocks(sbi) * 100,
499 sbi->user_block_count);
Jaegeuk Kim39a53e02012-11-28 13:37:31 +0900500}
501
502/*
503 * Sometimes f2fs may be better to drop out-of-place update policy.
Jaegeuk Kim216fbd62013-11-07 13:13:42 +0900504 * And, users can control the policy through sysfs entries.
505 * There are five policies with triggering conditions as follows.
506 * F2FS_IPU_FORCE - all the time,
507 * F2FS_IPU_SSR - if SSR mode is activated,
508 * F2FS_IPU_UTIL - if FS utilization is over threashold,
509 * F2FS_IPU_SSR_UTIL - if SSR mode is activated and FS utilization is over
510 * threashold,
Jaegeuk Kimc1ce1b02014-09-10 16:53:02 -0700511 * F2FS_IPU_FSYNC - activated in fsync path only for high performance flash
512 * storages. IPU will be triggered only if the # of dirty
513 * pages over min_fsync_blocks.
Jaegeuk Kim216fbd62013-11-07 13:13:42 +0900514 * F2FS_IPUT_DISABLE - disable IPU. (=default option)
Jaegeuk Kim39a53e02012-11-28 13:37:31 +0900515 */
Jaegeuk Kim216fbd62013-11-07 13:13:42 +0900516#define DEF_MIN_IPU_UTIL 70
Jaegeuk Kimc1ce1b02014-09-10 16:53:02 -0700517#define DEF_MIN_FSYNC_BLOCKS 8
Jaegeuk Kim216fbd62013-11-07 13:13:42 +0900518
519enum {
520 F2FS_IPU_FORCE,
521 F2FS_IPU_SSR,
522 F2FS_IPU_UTIL,
523 F2FS_IPU_SSR_UTIL,
Jaegeuk Kimc1ce1b02014-09-10 16:53:02 -0700524 F2FS_IPU_FSYNC,
Jaegeuk Kim216fbd62013-11-07 13:13:42 +0900525};
526
Jaegeuk Kim39a53e02012-11-28 13:37:31 +0900527static inline bool need_inplace_update(struct inode *inode)
528{
Jaegeuk Kim40813632014-09-02 15:31:18 -0700529 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
Jaegeuk Kim9b5f1362014-09-16 18:30:54 -0700530 unsigned int policy = SM_I(sbi)->ipu_policy;
Jaegeuk Kim216fbd62013-11-07 13:13:42 +0900531
532 /* IPU can be done only for the user data */
Jaegeuk Kim88b88a62014-10-06 17:39:50 -0700533 if (S_ISDIR(inode->i_mode) || f2fs_is_atomic_file(inode))
Jaegeuk Kim39a53e02012-11-28 13:37:31 +0900534 return false;
Jaegeuk Kim216fbd62013-11-07 13:13:42 +0900535
Jaegeuk Kim9b5f1362014-09-16 18:30:54 -0700536 if (policy & (0x1 << F2FS_IPU_FORCE))
Jaegeuk Kim39a53e02012-11-28 13:37:31 +0900537 return true;
Jaegeuk Kim9b5f1362014-09-16 18:30:54 -0700538 if (policy & (0x1 << F2FS_IPU_SSR) && need_SSR(sbi))
539 return true;
540 if (policy & (0x1 << F2FS_IPU_UTIL) &&
541 utilization(sbi) > SM_I(sbi)->min_ipu_util)
542 return true;
543 if (policy & (0x1 << F2FS_IPU_SSR_UTIL) && need_SSR(sbi) &&
544 utilization(sbi) > SM_I(sbi)->min_ipu_util)
545 return true;
546
547 /* this is only set during fdatasync */
548 if (policy & (0x1 << F2FS_IPU_FSYNC) &&
Jaegeuk Kim91942322016-05-20 10:13:22 -0700549 is_inode_flag_set(inode, FI_NEED_IPU))
Jaegeuk Kim9b5f1362014-09-16 18:30:54 -0700550 return true;
551
Jaegeuk Kim39a53e02012-11-28 13:37:31 +0900552 return false;
553}
554
555static inline unsigned int curseg_segno(struct f2fs_sb_info *sbi,
556 int type)
557{
558 struct curseg_info *curseg = CURSEG_I(sbi, type);
559 return curseg->segno;
560}
561
562static inline unsigned char curseg_alloc_type(struct f2fs_sb_info *sbi,
563 int type)
564{
565 struct curseg_info *curseg = CURSEG_I(sbi, type);
566 return curseg->alloc_type;
567}
568
569static inline unsigned short curseg_blkoff(struct f2fs_sb_info *sbi, int type)
570{
571 struct curseg_info *curseg = CURSEG_I(sbi, type);
572 return curseg->next_blkoff;
573}
574
575static inline void check_seg_range(struct f2fs_sb_info *sbi, unsigned int segno)
576{
Liu Xue7a04f642015-07-27 10:17:59 +0000577 f2fs_bug_on(sbi, segno > TOTAL_SEGS(sbi) - 1);
Jaegeuk Kim39a53e02012-11-28 13:37:31 +0900578}
579
Jaegeuk Kim39a53e02012-11-28 13:37:31 +0900580static inline void verify_block_addr(struct f2fs_sb_info *sbi, block_t blk_addr)
581{
Liu Xue7a04f642015-07-27 10:17:59 +0000582 f2fs_bug_on(sbi, blk_addr < SEG0_BLKADDR(sbi)
583 || blk_addr >= MAX_BLKADDR(sbi));
Jaegeuk Kim39a53e02012-11-28 13:37:31 +0900584}
585
586/*
arter97e1c42042014-08-06 23:22:50 +0900587 * Summary block is always treated as an invalid block
Jaegeuk Kim39a53e02012-11-28 13:37:31 +0900588 */
589static inline void check_block_count(struct f2fs_sb_info *sbi,
590 int segno, struct f2fs_sit_entry *raw_sit)
591{
Jaegeuk Kim4c278392015-08-11 16:01:30 -0700592#ifdef CONFIG_F2FS_CHECK_FS
Chao Yu44c60bf2013-10-29 14:50:40 +0800593 bool is_valid = test_bit_le(0, raw_sit->valid_map) ? true : false;
Jaegeuk Kim39a53e02012-11-28 13:37:31 +0900594 int valid_blocks = 0;
Chao Yu44c60bf2013-10-29 14:50:40 +0800595 int cur_pos = 0, next_pos;
Jaegeuk Kim39a53e02012-11-28 13:37:31 +0900596
Jaegeuk Kim39a53e02012-11-28 13:37:31 +0900597 /* check bitmap with valid block count */
Chao Yu44c60bf2013-10-29 14:50:40 +0800598 do {
599 if (is_valid) {
600 next_pos = find_next_zero_bit_le(&raw_sit->valid_map,
601 sbi->blocks_per_seg,
602 cur_pos);
603 valid_blocks += next_pos - cur_pos;
604 } else
605 next_pos = find_next_bit_le(&raw_sit->valid_map,
606 sbi->blocks_per_seg,
607 cur_pos);
608 cur_pos = next_pos;
609 is_valid = !is_valid;
610 } while (cur_pos < sbi->blocks_per_seg);
Jaegeuk Kim39a53e02012-11-28 13:37:31 +0900611 BUG_ON(GET_SIT_VBLOCKS(raw_sit) != valid_blocks);
Jaegeuk Kim5d56b672013-10-29 15:14:54 +0900612#endif
Jaegeuk Kim4c278392015-08-11 16:01:30 -0700613 /* check segment usage, and check boundary of a given segment number */
614 f2fs_bug_on(sbi, GET_SIT_VBLOCKS(raw_sit) > sbi->blocks_per_seg
615 || segno > TOTAL_SEGS(sbi) - 1);
Liu Xue7a04f642015-07-27 10:17:59 +0000616}
Jaegeuk Kim39a53e02012-11-28 13:37:31 +0900617
618static inline pgoff_t current_sit_addr(struct f2fs_sb_info *sbi,
619 unsigned int start)
620{
621 struct sit_info *sit_i = SIT_I(sbi);
Chao Yud3a14af2014-09-04 18:11:47 +0800622 unsigned int offset = SIT_BLOCK_OFFSET(start);
Jaegeuk Kim39a53e02012-11-28 13:37:31 +0900623 block_t blk_addr = sit_i->sit_base_addr + offset;
624
625 check_seg_range(sbi, start);
626
627 /* calculate sit block address */
628 if (f2fs_test_bit(offset, sit_i->sit_bitmap))
629 blk_addr += sit_i->sit_blocks;
630
631 return blk_addr;
632}
633
634static inline pgoff_t next_sit_addr(struct f2fs_sb_info *sbi,
635 pgoff_t block_addr)
636{
637 struct sit_info *sit_i = SIT_I(sbi);
638 block_addr -= sit_i->sit_base_addr;
639 if (block_addr < sit_i->sit_blocks)
640 block_addr += sit_i->sit_blocks;
641 else
642 block_addr -= sit_i->sit_blocks;
643
644 return block_addr + sit_i->sit_base_addr;
645}
646
647static inline void set_to_next_sit(struct sit_info *sit_i, unsigned int start)
648{
Chao Yud3a14af2014-09-04 18:11:47 +0800649 unsigned int block_off = SIT_BLOCK_OFFSET(start);
Jaegeuk Kim39a53e02012-11-28 13:37:31 +0900650
Gu Zhengc6ac4c02014-10-20 17:45:50 +0800651 f2fs_change_bit(block_off, sit_i->sit_bitmap);
Jaegeuk Kim39a53e02012-11-28 13:37:31 +0900652}
653
654static inline unsigned long long get_mtime(struct f2fs_sb_info *sbi)
655{
656 struct sit_info *sit_i = SIT_I(sbi);
657 return sit_i->elapsed_time + CURRENT_TIME_SEC.tv_sec -
658 sit_i->mounted_time;
659}
660
661static inline void set_summary(struct f2fs_summary *sum, nid_t nid,
662 unsigned int ofs_in_node, unsigned char version)
663{
664 sum->nid = cpu_to_le32(nid);
665 sum->ofs_in_node = cpu_to_le16(ofs_in_node);
666 sum->version = version;
667}
668
669static inline block_t start_sum_block(struct f2fs_sb_info *sbi)
670{
671 return __start_cp_addr(sbi) +
672 le32_to_cpu(F2FS_CKPT(sbi)->cp_pack_start_sum);
673}
674
675static inline block_t sum_blk_addr(struct f2fs_sb_info *sbi, int base, int type)
676{
677 return __start_cp_addr(sbi) +
678 le32_to_cpu(F2FS_CKPT(sbi)->cp_pack_total_block_count)
679 - (base + 1) + type;
680}
Jaegeuk Kim5ec4e492013-03-31 13:26:03 +0900681
682static inline bool sec_usage_check(struct f2fs_sb_info *sbi, unsigned int secno)
683{
684 if (IS_CURSEC(sbi, secno) || (sbi->cur_victim_sec == secno))
685 return true;
686 return false;
687}
Jaegeuk Kimac5d1562013-04-29 16:58:39 +0900688
689static inline unsigned int max_hw_blocks(struct f2fs_sb_info *sbi)
690{
691 struct block_device *bdev = sbi->sb->s_bdev;
692 struct request_queue *q = bdev_get_queue(bdev);
Chao Yu55cf9cb2014-09-15 18:01:10 +0800693 return SECTOR_TO_BLOCK(queue_max_sectors(q));
Jaegeuk Kimac5d1562013-04-29 16:58:39 +0900694}
Jaegeuk Kim87d6f892014-03-18 12:40:49 +0900695
696/*
697 * It is very important to gather dirty pages and write at once, so that we can
698 * submit a big bio without interfering other data writes.
699 * By default, 512 pages for directory data,
700 * 512 pages (2MB) * 3 for three types of nodes, and
701 * max_bio_blocks for meta are set.
702 */
703static inline int nr_pages_to_skip(struct f2fs_sb_info *sbi, int type)
704{
Tejun Heoa88a3412015-05-22 17:13:28 -0400705 if (sbi->sb->s_bdi->wb.dirty_exceeded)
Jaegeuk Kim510184c2014-11-06 17:23:08 -0800706 return 0;
707
Jaegeuk Kima1257022015-10-08 10:40:07 -0700708 if (type == DATA)
709 return sbi->blocks_per_seg;
710 else if (type == NODE)
Jaegeuk Kim87d6f892014-03-18 12:40:49 +0900711 return 3 * sbi->blocks_per_seg;
712 else if (type == META)
Jaegeuk Kim90a893c2014-09-22 16:21:07 -0700713 return MAX_BIO_BLOCKS(sbi);
Jaegeuk Kim87d6f892014-03-18 12:40:49 +0900714 else
715 return 0;
716}
Jaegeuk Kim50c8cdb2014-03-18 13:47:11 +0900717
718/*
719 * When writing pages, it'd better align nr_to_write for segment size.
720 */
721static inline long nr_pages_to_write(struct f2fs_sb_info *sbi, int type,
722 struct writeback_control *wbc)
723{
724 long nr_to_write, desired;
725
726 if (wbc->sync_mode != WB_SYNC_NONE)
727 return 0;
728
729 nr_to_write = wbc->nr_to_write;
730
731 if (type == DATA)
732 desired = 4096;
733 else if (type == NODE)
734 desired = 3 * max_hw_blocks(sbi);
735 else
Jaegeuk Kim90a893c2014-09-22 16:21:07 -0700736 desired = MAX_BIO_BLOCKS(sbi);
Jaegeuk Kim50c8cdb2014-03-18 13:47:11 +0900737
738 wbc->nr_to_write = desired;
739 return desired - nr_to_write;
740}