blob: 76274398842692f4238671de311c8d919e1f3ab0 [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 Kim44a83492016-07-13 18:23:35 -070019#define DEF_MAX_RECLAIM_PREFREE_SEGMENTS 4096 /* 8GB in maximum */
Jaegeuk Kim81eb8d62013-10-24 13:31:34 +090020
Namjae Jeon6224da82013-04-06 14:44:32 +090021/* L: Logical segment # in volume, R: Relative segment # in main area */
Jaegeuk Kim39a53e02012-11-28 13:37:31 +090022#define GET_L2R_SEGNO(free_i, segno) (segno - free_i->start_segno)
23#define GET_R2L_SEGNO(free_i, segno) (segno + free_i->start_segno)
24
Changman Lee61ae45c2013-11-21 20:04:21 +090025#define IS_DATASEG(t) (t <= CURSEG_COLD_DATA)
26#define IS_NODESEG(t) (t >= CURSEG_HOT_NODE)
Jaegeuk Kim39a53e02012-11-28 13:37:31 +090027
Jaegeuk Kim5c773ba2013-03-31 12:30:04 +090028#define IS_CURSEG(sbi, seg) \
29 ((seg == CURSEG_I(sbi, CURSEG_HOT_DATA)->segno) || \
30 (seg == CURSEG_I(sbi, CURSEG_WARM_DATA)->segno) || \
31 (seg == CURSEG_I(sbi, CURSEG_COLD_DATA)->segno) || \
32 (seg == CURSEG_I(sbi, CURSEG_HOT_NODE)->segno) || \
33 (seg == CURSEG_I(sbi, CURSEG_WARM_NODE)->segno) || \
34 (seg == CURSEG_I(sbi, CURSEG_COLD_NODE)->segno))
Jaegeuk Kim39a53e02012-11-28 13:37:31 +090035
36#define IS_CURSEC(sbi, secno) \
37 ((secno == CURSEG_I(sbi, CURSEG_HOT_DATA)->segno / \
38 sbi->segs_per_sec) || \
39 (secno == CURSEG_I(sbi, CURSEG_WARM_DATA)->segno / \
40 sbi->segs_per_sec) || \
41 (secno == CURSEG_I(sbi, CURSEG_COLD_DATA)->segno / \
42 sbi->segs_per_sec) || \
43 (secno == CURSEG_I(sbi, CURSEG_HOT_NODE)->segno / \
44 sbi->segs_per_sec) || \
45 (secno == CURSEG_I(sbi, CURSEG_WARM_NODE)->segno / \
46 sbi->segs_per_sec) || \
47 (secno == CURSEG_I(sbi, CURSEG_COLD_NODE)->segno / \
48 sbi->segs_per_sec)) \
49
Jaegeuk Kim7cd85582014-09-23 11:23:01 -070050#define MAIN_BLKADDR(sbi) (SM_I(sbi)->main_blkaddr)
51#define SEG0_BLKADDR(sbi) (SM_I(sbi)->seg0_blkaddr)
52
53#define MAIN_SEGS(sbi) (SM_I(sbi)->main_segments)
54#define MAIN_SECS(sbi) (sbi->total_sections)
55
56#define TOTAL_SEGS(sbi) (SM_I(sbi)->segment_count)
57#define TOTAL_BLKS(sbi) (TOTAL_SEGS(sbi) << sbi->log_blocks_per_seg)
58
59#define MAX_BLKADDR(sbi) (SEG0_BLKADDR(sbi) + TOTAL_BLKS(sbi))
Dan Carpenter8a219842014-09-25 14:39:17 +030060#define SEGMENT_SIZE(sbi) (1ULL << (sbi->log_blocksize + \
Jaegeuk Kim7cd85582014-09-23 11:23:01 -070061 sbi->log_blocks_per_seg))
62
63#define START_BLOCK(sbi, segno) (SEG0_BLKADDR(sbi) + \
Jaegeuk Kim39a53e02012-11-28 13:37:31 +090064 (GET_R2L_SEGNO(FREE_I(sbi), segno) << sbi->log_blocks_per_seg))
Jaegeuk Kim7cd85582014-09-23 11:23:01 -070065
Jaegeuk Kim39a53e02012-11-28 13:37:31 +090066#define NEXT_FREE_BLKADDR(sbi, curseg) \
67 (START_BLOCK(sbi, curseg->segno) + curseg->next_blkoff)
68
Jaegeuk Kim7cd85582014-09-23 11:23:01 -070069#define GET_SEGOFF_FROM_SEG0(sbi, blk_addr) ((blk_addr) - SEG0_BLKADDR(sbi))
Jaegeuk Kim39a53e02012-11-28 13:37:31 +090070#define GET_SEGNO_FROM_SEG0(sbi, blk_addr) \
71 (GET_SEGOFF_FROM_SEG0(sbi, blk_addr) >> sbi->log_blocks_per_seg)
Jaegeuk Kim491c0852014-02-04 13:01:10 +090072#define GET_BLKOFF_FROM_SEG0(sbi, blk_addr) \
73 (GET_SEGOFF_FROM_SEG0(sbi, blk_addr) & (sbi->blocks_per_seg - 1))
74
Jaegeuk Kim39a53e02012-11-28 13:37:31 +090075#define GET_SEGNO(sbi, blk_addr) \
76 (((blk_addr == NULL_ADDR) || (blk_addr == NEW_ADDR)) ? \
77 NULL_SEGNO : GET_L2R_SEGNO(FREE_I(sbi), \
78 GET_SEGNO_FROM_SEG0(sbi, blk_addr)))
79#define GET_SECNO(sbi, segno) \
80 ((segno) / sbi->segs_per_sec)
81#define GET_ZONENO_FROM_SEGNO(sbi, segno) \
82 ((segno / sbi->segs_per_sec) / sbi->secs_per_zone)
83
84#define GET_SUM_BLOCK(sbi, segno) \
85 ((sbi->sm_info->ssa_blkaddr) + segno)
86
87#define GET_SUM_TYPE(footer) ((footer)->entry_type)
88#define SET_SUM_TYPE(footer, type) ((footer)->entry_type = type)
89
90#define SIT_ENTRY_OFFSET(sit_i, segno) \
91 (segno % sit_i->sents_per_block)
Chao Yud3a14af2014-09-04 18:11:47 +080092#define SIT_BLOCK_OFFSET(segno) \
Jaegeuk Kim39a53e02012-11-28 13:37:31 +090093 (segno / SIT_ENTRY_PER_BLOCK)
Chao Yud3a14af2014-09-04 18:11:47 +080094#define START_SEGNO(segno) \
95 (SIT_BLOCK_OFFSET(segno) * SIT_ENTRY_PER_BLOCK)
Chao Yu74de5932013-11-22 09:09:59 +080096#define SIT_BLK_CNT(sbi) \
Jaegeuk Kim7cd85582014-09-23 11:23:01 -070097 ((MAIN_SEGS(sbi) + SIT_ENTRY_PER_BLOCK - 1) / SIT_ENTRY_PER_BLOCK)
Jaegeuk Kim39a53e02012-11-28 13:37:31 +090098#define f2fs_bitmap_size(nr) \
99 (BITS_TO_LONGS(nr) * sizeof(unsigned long))
Jaegeuk Kim39a53e02012-11-28 13:37:31 +0900100
Chao Yu55cf9cb2014-09-15 18:01:10 +0800101#define SECTOR_FROM_BLOCK(blk_addr) \
102 (((sector_t)blk_addr) << F2FS_LOG_SECTORS_PER_BLOCK)
103#define SECTOR_TO_BLOCK(sectors) \
104 (sectors >> F2FS_LOG_SECTORS_PER_BLOCK)
Jaegeuk Kim90a893c2014-09-22 16:21:07 -0700105#define MAX_BIO_BLOCKS(sbi) \
106 ((int)min((int)max_hw_blocks(sbi), BIO_MAX_PAGES))
Jaegeuk Kim3cd8a232012-12-10 09:26:05 +0900107
Jaegeuk Kim39a53e02012-11-28 13:37:31 +0900108/*
109 * indicate a block allocation direction: RIGHT and LEFT.
110 * RIGHT means allocating new sections towards the end of volume.
111 * LEFT means the opposite direction.
112 */
113enum {
114 ALLOC_RIGHT = 0,
115 ALLOC_LEFT
116};
117
118/*
119 * In the victim_sel_policy->alloc_mode, there are two block allocation modes.
120 * LFS writes data sequentially with cleaning operations.
121 * SSR (Slack Space Recycle) reuses obsolete space without cleaning operations.
122 */
123enum {
124 LFS = 0,
125 SSR
126};
127
128/*
129 * In the victim_sel_policy->gc_mode, there are two gc, aka cleaning, modes.
130 * GC_CB is based on cost-benefit algorithm.
131 * GC_GREEDY is based on greedy algorithm.
132 */
133enum {
134 GC_CB = 0,
135 GC_GREEDY
136};
137
138/*
139 * BG_GC means the background cleaning job.
140 * FG_GC means the on-demand cleaning job.
Jaegeuk Kim6aefd932015-10-05 11:02:54 -0700141 * FORCE_FG_GC means on-demand cleaning job in background.
Jaegeuk Kim39a53e02012-11-28 13:37:31 +0900142 */
143enum {
144 BG_GC = 0,
Jaegeuk Kim6aefd932015-10-05 11:02:54 -0700145 FG_GC,
146 FORCE_FG_GC,
Jaegeuk Kim39a53e02012-11-28 13:37:31 +0900147};
148
149/* for a function parameter to select a victim segment */
150struct victim_sel_policy {
151 int alloc_mode; /* LFS or SSR */
152 int gc_mode; /* GC_CB or GC_GREEDY */
153 unsigned long *dirty_segmap; /* dirty segment bitmap */
Jin Xua26b7c82013-09-05 12:45:26 +0800154 unsigned int max_search; /* maximum # of segments to search */
Jaegeuk Kim39a53e02012-11-28 13:37:31 +0900155 unsigned int offset; /* last scanned bitmap offset */
156 unsigned int ofs_unit; /* bitmap search unit */
157 unsigned int min_cost; /* minimum cost */
158 unsigned int min_segno; /* segment # having min. cost */
159};
160
161struct seg_entry {
Chao Yuf51b4ce2016-05-04 23:19:48 +0800162 unsigned int type:6; /* segment type like CURSEG_XXX_TYPE */
163 unsigned int valid_blocks:10; /* # of valid blocks */
164 unsigned int ckpt_valid_blocks:10; /* # of valid blocks last cp */
165 unsigned int padding:6; /* padding */
Jaegeuk Kim39a53e02012-11-28 13:37:31 +0900166 unsigned char *cur_valid_map; /* validity bitmap of blocks */
167 /*
168 * # of valid blocks and the validity bitmap stored in the the last
169 * checkpoint pack. This information is used by the SSR mode.
170 */
Chao Yuf51b4ce2016-05-04 23:19:48 +0800171 unsigned char *ckpt_valid_map; /* validity bitmap of blocks last cp */
Jaegeuk Kima66cdd92015-04-30 22:37:50 -0700172 unsigned char *discard_map;
Jaegeuk Kim39a53e02012-11-28 13:37:31 +0900173 unsigned long long mtime; /* modification time of the segment */
174};
175
176struct sec_entry {
177 unsigned int valid_blocks; /* # of valid blocks in a section */
178};
179
180struct segment_allocation {
181 void (*allocate_segment)(struct f2fs_sb_info *, int, bool);
182};
183
Chao Yudecd36b2015-08-07 18:42:09 +0800184/*
185 * this value is set in page as a private data which indicate that
186 * the page is atomically written, and it is in inmem_pages list.
187 */
Jaegeuk Kimd48dfc22016-01-29 16:21:15 -0800188#define ATOMIC_WRITTEN_PAGE ((unsigned long)-1)
Chao Yudecd36b2015-08-07 18:42:09 +0800189
190#define IS_ATOMIC_WRITTEN_PAGE(page) \
191 (page_private(page) == (unsigned long)ATOMIC_WRITTEN_PAGE)
192
Jaegeuk Kim88b88a62014-10-06 17:39:50 -0700193struct inmem_pages {
194 struct list_head list;
195 struct page *page;
Chao Yu28bc1062016-02-06 14:40:34 +0800196 block_t old_addr; /* for revoking when fail to commit */
Jaegeuk Kim88b88a62014-10-06 17:39:50 -0700197};
198
Jaegeuk Kim39a53e02012-11-28 13:37:31 +0900199struct sit_info {
200 const struct segment_allocation *s_ops;
201
202 block_t sit_base_addr; /* start block address of SIT area */
203 block_t sit_blocks; /* # of blocks used by SIT area */
204 block_t written_valid_blocks; /* # of valid blocks in main area */
205 char *sit_bitmap; /* SIT bitmap pointer */
206 unsigned int bitmap_size; /* SIT bitmap size */
207
Jaegeuk Kim60a3b782015-02-10 16:44:29 -0800208 unsigned long *tmp_map; /* bitmap for temporal use */
Jaegeuk Kim39a53e02012-11-28 13:37:31 +0900209 unsigned long *dirty_sentries_bitmap; /* bitmap for dirty sentries */
210 unsigned int dirty_sentries; /* # of dirty sentries */
211 unsigned int sents_per_block; /* # of SIT entries per block */
212 struct mutex sentry_lock; /* to protect SIT cache */
213 struct seg_entry *sentries; /* SIT segment-level cache */
214 struct sec_entry *sec_entries; /* SIT section-level cache */
215
216 /* for cost-benefit algorithm in cleaning procedure */
217 unsigned long long elapsed_time; /* elapsed time after mount */
218 unsigned long long mounted_time; /* mount time */
219 unsigned long long min_mtime; /* min. modification time */
220 unsigned long long max_mtime; /* max. modification time */
221};
222
223struct free_segmap_info {
224 unsigned int start_segno; /* start segment number logically */
225 unsigned int free_segments; /* # of free segments */
226 unsigned int free_sections; /* # of free sections */
Chao Yu1a118cc2015-02-11 18:20:38 +0800227 spinlock_t segmap_lock; /* free segmap lock */
Jaegeuk Kim39a53e02012-11-28 13:37:31 +0900228 unsigned long *free_segmap; /* free segment bitmap */
229 unsigned long *free_secmap; /* free section bitmap */
230};
231
232/* Notice: The order of dirty type is same with CURSEG_XXX in f2fs.h */
233enum dirty_type {
234 DIRTY_HOT_DATA, /* dirty segments assigned as hot data logs */
235 DIRTY_WARM_DATA, /* dirty segments assigned as warm data logs */
236 DIRTY_COLD_DATA, /* dirty segments assigned as cold data logs */
237 DIRTY_HOT_NODE, /* dirty segments assigned as hot node logs */
238 DIRTY_WARM_NODE, /* dirty segments assigned as warm node logs */
239 DIRTY_COLD_NODE, /* dirty segments assigned as cold node logs */
240 DIRTY, /* to count # of dirty segments */
241 PRE, /* to count # of entirely obsolete segments */
242 NR_DIRTY_TYPE
243};
244
245struct dirty_seglist_info {
246 const struct victim_selection *v_ops; /* victim selction operation */
247 unsigned long *dirty_segmap[NR_DIRTY_TYPE];
248 struct mutex seglist_lock; /* lock for segment bitmaps */
249 int nr_dirty[NR_DIRTY_TYPE]; /* # of dirty segments */
Jaegeuk Kim5ec4e492013-03-31 13:26:03 +0900250 unsigned long *victim_secmap; /* background GC victims */
Jaegeuk Kim39a53e02012-11-28 13:37:31 +0900251};
252
253/* victim selection function for cleaning and SSR */
254struct victim_selection {
255 int (*get_victim)(struct f2fs_sb_info *, unsigned int *,
256 int, int, char);
257};
258
259/* for active log information */
260struct curseg_info {
261 struct mutex curseg_mutex; /* lock for consistency */
262 struct f2fs_summary_block *sum_blk; /* cached summary block */
Chao Yub7ad7512016-02-19 18:08:46 +0800263 struct rw_semaphore journal_rwsem; /* protect journal area */
264 struct f2fs_journal *journal; /* cached journal info */
Jaegeuk Kim39a53e02012-11-28 13:37:31 +0900265 unsigned char alloc_type; /* current allocation type */
266 unsigned int segno; /* current segment number */
267 unsigned short next_blkoff; /* next block offset to write */
268 unsigned int zone; /* current zone number */
269 unsigned int next_segno; /* preallocated segment */
270};
271
Chao Yu184a5cd2014-09-04 18:13:01 +0800272struct sit_entry_set {
273 struct list_head set_list; /* link with all sit sets */
274 unsigned int start_segno; /* start segno of sits in set */
275 unsigned int entry_cnt; /* the # of sit entries in set */
276};
277
Jaegeuk Kim39a53e02012-11-28 13:37:31 +0900278/*
279 * inline functions
280 */
281static inline struct curseg_info *CURSEG_I(struct f2fs_sb_info *sbi, int type)
282{
283 return (struct curseg_info *)(SM_I(sbi)->curseg_array + type);
284}
285
286static inline struct seg_entry *get_seg_entry(struct f2fs_sb_info *sbi,
287 unsigned int segno)
288{
289 struct sit_info *sit_i = SIT_I(sbi);
290 return &sit_i->sentries[segno];
291}
292
293static inline struct sec_entry *get_sec_entry(struct f2fs_sb_info *sbi,
294 unsigned int segno)
295{
296 struct sit_info *sit_i = SIT_I(sbi);
297 return &sit_i->sec_entries[GET_SECNO(sbi, segno)];
298}
299
300static inline unsigned int get_valid_blocks(struct f2fs_sb_info *sbi,
301 unsigned int segno, int section)
302{
303 /*
304 * In order to get # of valid blocks in a section instantly from many
305 * segments, f2fs manages two counting structures separately.
306 */
307 if (section > 1)
308 return get_sec_entry(sbi, segno)->valid_blocks;
309 else
310 return get_seg_entry(sbi, segno)->valid_blocks;
311}
312
313static inline void seg_info_from_raw_sit(struct seg_entry *se,
314 struct f2fs_sit_entry *rs)
315{
316 se->valid_blocks = GET_SIT_VBLOCKS(rs);
317 se->ckpt_valid_blocks = GET_SIT_VBLOCKS(rs);
318 memcpy(se->cur_valid_map, rs->valid_map, SIT_VBLOCK_MAP_SIZE);
319 memcpy(se->ckpt_valid_map, rs->valid_map, SIT_VBLOCK_MAP_SIZE);
320 se->type = GET_SIT_TYPE(rs);
321 se->mtime = le64_to_cpu(rs->mtime);
322}
323
324static inline void seg_info_to_raw_sit(struct seg_entry *se,
325 struct f2fs_sit_entry *rs)
326{
327 unsigned short raw_vblocks = (se->type << SIT_VBLOCKS_SHIFT) |
328 se->valid_blocks;
329 rs->vblocks = cpu_to_le16(raw_vblocks);
330 memcpy(rs->valid_map, se->cur_valid_map, SIT_VBLOCK_MAP_SIZE);
331 memcpy(se->ckpt_valid_map, rs->valid_map, SIT_VBLOCK_MAP_SIZE);
332 se->ckpt_valid_blocks = se->valid_blocks;
333 rs->mtime = cpu_to_le64(se->mtime);
334}
335
336static inline unsigned int find_next_inuse(struct free_segmap_info *free_i,
337 unsigned int max, unsigned int segno)
338{
339 unsigned int ret;
Chao Yu1a118cc2015-02-11 18:20:38 +0800340 spin_lock(&free_i->segmap_lock);
Jaegeuk Kim39a53e02012-11-28 13:37:31 +0900341 ret = find_next_bit(free_i->free_segmap, max, segno);
Chao Yu1a118cc2015-02-11 18:20:38 +0800342 spin_unlock(&free_i->segmap_lock);
Jaegeuk Kim39a53e02012-11-28 13:37:31 +0900343 return ret;
344}
345
346static inline void __set_free(struct f2fs_sb_info *sbi, unsigned int segno)
347{
348 struct free_segmap_info *free_i = FREE_I(sbi);
349 unsigned int secno = segno / sbi->segs_per_sec;
350 unsigned int start_segno = secno * sbi->segs_per_sec;
351 unsigned int next;
352
Chao Yu1a118cc2015-02-11 18:20:38 +0800353 spin_lock(&free_i->segmap_lock);
Jaegeuk Kim39a53e02012-11-28 13:37:31 +0900354 clear_bit(segno, free_i->free_segmap);
355 free_i->free_segments++;
356
Wanpeng Li7fd97012015-03-06 15:00:55 +0800357 next = find_next_bit(free_i->free_segmap,
358 start_segno + sbi->segs_per_sec, start_segno);
Jaegeuk Kim39a53e02012-11-28 13:37:31 +0900359 if (next >= start_segno + sbi->segs_per_sec) {
360 clear_bit(secno, free_i->free_secmap);
361 free_i->free_sections++;
362 }
Chao Yu1a118cc2015-02-11 18:20:38 +0800363 spin_unlock(&free_i->segmap_lock);
Jaegeuk Kim39a53e02012-11-28 13:37:31 +0900364}
365
366static inline void __set_inuse(struct f2fs_sb_info *sbi,
367 unsigned int segno)
368{
369 struct free_segmap_info *free_i = FREE_I(sbi);
370 unsigned int secno = segno / sbi->segs_per_sec;
371 set_bit(segno, free_i->free_segmap);
372 free_i->free_segments--;
373 if (!test_and_set_bit(secno, free_i->free_secmap))
374 free_i->free_sections--;
375}
376
377static inline void __set_test_and_free(struct f2fs_sb_info *sbi,
378 unsigned int segno)
379{
380 struct free_segmap_info *free_i = FREE_I(sbi);
381 unsigned int secno = segno / sbi->segs_per_sec;
382 unsigned int start_segno = secno * sbi->segs_per_sec;
383 unsigned int next;
384
Chao Yu1a118cc2015-02-11 18:20:38 +0800385 spin_lock(&free_i->segmap_lock);
Jaegeuk Kim39a53e02012-11-28 13:37:31 +0900386 if (test_and_clear_bit(segno, free_i->free_segmap)) {
387 free_i->free_segments++;
388
Chao Yuf1121ab2014-07-14 16:45:15 +0800389 next = find_next_bit(free_i->free_segmap,
390 start_segno + sbi->segs_per_sec, start_segno);
Jaegeuk Kim39a53e02012-11-28 13:37:31 +0900391 if (next >= start_segno + sbi->segs_per_sec) {
392 if (test_and_clear_bit(secno, free_i->free_secmap))
393 free_i->free_sections++;
394 }
395 }
Chao Yu1a118cc2015-02-11 18:20:38 +0800396 spin_unlock(&free_i->segmap_lock);
Jaegeuk Kim39a53e02012-11-28 13:37:31 +0900397}
398
399static inline void __set_test_and_inuse(struct f2fs_sb_info *sbi,
400 unsigned int segno)
401{
402 struct free_segmap_info *free_i = FREE_I(sbi);
403 unsigned int secno = segno / sbi->segs_per_sec;
Chao Yu1a118cc2015-02-11 18:20:38 +0800404 spin_lock(&free_i->segmap_lock);
Jaegeuk Kim39a53e02012-11-28 13:37:31 +0900405 if (!test_and_set_bit(segno, free_i->free_segmap)) {
406 free_i->free_segments--;
407 if (!test_and_set_bit(secno, free_i->free_secmap))
408 free_i->free_sections--;
409 }
Chao Yu1a118cc2015-02-11 18:20:38 +0800410 spin_unlock(&free_i->segmap_lock);
Jaegeuk Kim39a53e02012-11-28 13:37:31 +0900411}
412
413static inline void get_sit_bitmap(struct f2fs_sb_info *sbi,
414 void *dst_addr)
415{
416 struct sit_info *sit_i = SIT_I(sbi);
417 memcpy(dst_addr, sit_i->sit_bitmap, sit_i->bitmap_size);
418}
419
420static inline block_t written_block_count(struct f2fs_sb_info *sbi)
421{
Jaegeuk Kim8b8343f2014-02-24 13:00:13 +0900422 return SIT_I(sbi)->written_valid_blocks;
Jaegeuk Kim39a53e02012-11-28 13:37:31 +0900423}
424
425static inline unsigned int free_segments(struct f2fs_sb_info *sbi)
426{
Jaegeuk Kim8b8343f2014-02-24 13:00:13 +0900427 return FREE_I(sbi)->free_segments;
Jaegeuk Kim39a53e02012-11-28 13:37:31 +0900428}
429
430static inline int reserved_segments(struct f2fs_sb_info *sbi)
431{
432 return SM_I(sbi)->reserved_segments;
433}
434
435static inline unsigned int free_sections(struct f2fs_sb_info *sbi)
436{
Jaegeuk Kim8b8343f2014-02-24 13:00:13 +0900437 return FREE_I(sbi)->free_sections;
Jaegeuk Kim39a53e02012-11-28 13:37:31 +0900438}
439
440static inline unsigned int prefree_segments(struct f2fs_sb_info *sbi)
441{
442 return DIRTY_I(sbi)->nr_dirty[PRE];
443}
444
445static inline unsigned int dirty_segments(struct f2fs_sb_info *sbi)
446{
447 return DIRTY_I(sbi)->nr_dirty[DIRTY_HOT_DATA] +
448 DIRTY_I(sbi)->nr_dirty[DIRTY_WARM_DATA] +
449 DIRTY_I(sbi)->nr_dirty[DIRTY_COLD_DATA] +
450 DIRTY_I(sbi)->nr_dirty[DIRTY_HOT_NODE] +
451 DIRTY_I(sbi)->nr_dirty[DIRTY_WARM_NODE] +
452 DIRTY_I(sbi)->nr_dirty[DIRTY_COLD_NODE];
453}
454
455static inline int overprovision_segments(struct f2fs_sb_info *sbi)
456{
457 return SM_I(sbi)->ovp_segments;
458}
459
460static inline int overprovision_sections(struct f2fs_sb_info *sbi)
461{
462 return ((unsigned int) overprovision_segments(sbi)) / sbi->segs_per_sec;
463}
464
465static inline int reserved_sections(struct f2fs_sb_info *sbi)
466{
467 return ((unsigned int) reserved_segments(sbi)) / sbi->segs_per_sec;
468}
469
470static inline bool need_SSR(struct f2fs_sb_info *sbi)
471{
Jaegeuk Kim95dd8972014-09-17 17:52:58 -0700472 int node_secs = get_blocktype_secs(sbi, F2FS_DIRTY_NODES);
473 int dent_secs = get_blocktype_secs(sbi, F2FS_DIRTY_DENTS);
Jaegeuk Kimb9610bd2016-10-14 13:28:05 -0700474 int imeta_secs = get_blocktype_secs(sbi, F2FS_DIRTY_IMETA);
Jaegeuk Kim36abef42016-06-03 19:29:38 -0700475
476 if (test_opt(sbi, LFS))
477 return false;
478
Jaegeuk Kimb9610bd2016-10-14 13:28:05 -0700479 return free_sections(sbi) <= (node_secs + 2 * dent_secs + imeta_secs +
Jaegeuk Kim95dd8972014-09-17 17:52:58 -0700480 reserved_sections(sbi) + 1);
Jaegeuk Kim39a53e02012-11-28 13:37:31 +0900481}
482
Jaegeuk Kim7f3037a2016-09-01 12:02:51 -0700483static inline bool has_not_enough_free_secs(struct f2fs_sb_info *sbi,
484 int freed, int needed)
Jaegeuk Kim39a53e02012-11-28 13:37:31 +0900485{
Namjae Jeon5ac206c2013-02-02 23:52:59 +0900486 int node_secs = get_blocktype_secs(sbi, F2FS_DIRTY_NODES);
487 int dent_secs = get_blocktype_secs(sbi, F2FS_DIRTY_DENTS);
Jaegeuk Kimb9610bd2016-10-14 13:28:05 -0700488 int imeta_secs = get_blocktype_secs(sbi, F2FS_DIRTY_IMETA);
Jaegeuk Kim0f18b462016-05-20 11:10:10 -0700489
Chao Yucaf00472015-01-28 17:48:42 +0800490 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
Jaegeuk Kim029cd282012-12-21 17:20:21 +0900491 return false;
492
Jaegeuk Kim7f3037a2016-09-01 12:02:51 -0700493 return (free_sections(sbi) + freed) <=
Jaegeuk Kimb9610bd2016-10-14 13:28:05 -0700494 (node_secs + 2 * dent_secs + imeta_secs +
495 reserved_sections(sbi) + needed);
Jaegeuk Kim39a53e02012-11-28 13:37:31 +0900496}
497
Jaegeuk Kim81eb8d62013-10-24 13:31:34 +0900498static inline bool excess_prefree_segs(struct f2fs_sb_info *sbi)
499{
Chris Fries6c311ec2014-01-17 14:44:39 -0600500 return prefree_segments(sbi) > SM_I(sbi)->rec_prefree_segments;
Jaegeuk Kim81eb8d62013-10-24 13:31:34 +0900501}
502
Jaegeuk Kim39a53e02012-11-28 13:37:31 +0900503static inline int utilization(struct f2fs_sb_info *sbi)
504{
Chris Fries6c311ec2014-01-17 14:44:39 -0600505 return div_u64((u64)valid_user_blocks(sbi) * 100,
506 sbi->user_block_count);
Jaegeuk Kim39a53e02012-11-28 13:37:31 +0900507}
508
509/*
510 * Sometimes f2fs may be better to drop out-of-place update policy.
Jaegeuk Kim216fbd62013-11-07 13:13:42 +0900511 * And, users can control the policy through sysfs entries.
512 * There are five policies with triggering conditions as follows.
513 * F2FS_IPU_FORCE - all the time,
514 * F2FS_IPU_SSR - if SSR mode is activated,
515 * F2FS_IPU_UTIL - if FS utilization is over threashold,
516 * F2FS_IPU_SSR_UTIL - if SSR mode is activated and FS utilization is over
517 * threashold,
Jaegeuk Kimc1ce1b02014-09-10 16:53:02 -0700518 * F2FS_IPU_FSYNC - activated in fsync path only for high performance flash
519 * storages. IPU will be triggered only if the # of dirty
520 * pages over min_fsync_blocks.
Jaegeuk Kim216fbd62013-11-07 13:13:42 +0900521 * F2FS_IPUT_DISABLE - disable IPU. (=default option)
Jaegeuk Kim39a53e02012-11-28 13:37:31 +0900522 */
Jaegeuk Kim216fbd62013-11-07 13:13:42 +0900523#define DEF_MIN_IPU_UTIL 70
Jaegeuk Kimc1ce1b02014-09-10 16:53:02 -0700524#define DEF_MIN_FSYNC_BLOCKS 8
Jaegeuk Kim216fbd62013-11-07 13:13:42 +0900525
526enum {
527 F2FS_IPU_FORCE,
528 F2FS_IPU_SSR,
529 F2FS_IPU_UTIL,
530 F2FS_IPU_SSR_UTIL,
Jaegeuk Kimc1ce1b02014-09-10 16:53:02 -0700531 F2FS_IPU_FSYNC,
Jaegeuk Kim216fbd62013-11-07 13:13:42 +0900532};
533
Jaegeuk Kim39a53e02012-11-28 13:37:31 +0900534static inline bool need_inplace_update(struct inode *inode)
535{
Jaegeuk Kim40813632014-09-02 15:31:18 -0700536 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
Jaegeuk Kim9b5f1362014-09-16 18:30:54 -0700537 unsigned int policy = SM_I(sbi)->ipu_policy;
Jaegeuk Kim216fbd62013-11-07 13:13:42 +0900538
539 /* IPU can be done only for the user data */
Jaegeuk Kim88b88a62014-10-06 17:39:50 -0700540 if (S_ISDIR(inode->i_mode) || f2fs_is_atomic_file(inode))
Jaegeuk Kim39a53e02012-11-28 13:37:31 +0900541 return false;
Jaegeuk Kim216fbd62013-11-07 13:13:42 +0900542
Jaegeuk Kim36abef42016-06-03 19:29:38 -0700543 if (test_opt(sbi, LFS))
544 return false;
545
Jaegeuk Kim9b5f1362014-09-16 18:30:54 -0700546 if (policy & (0x1 << F2FS_IPU_FORCE))
Jaegeuk Kim39a53e02012-11-28 13:37:31 +0900547 return true;
Jaegeuk Kim9b5f1362014-09-16 18:30:54 -0700548 if (policy & (0x1 << F2FS_IPU_SSR) && need_SSR(sbi))
549 return true;
550 if (policy & (0x1 << F2FS_IPU_UTIL) &&
551 utilization(sbi) > SM_I(sbi)->min_ipu_util)
552 return true;
553 if (policy & (0x1 << F2FS_IPU_SSR_UTIL) && need_SSR(sbi) &&
554 utilization(sbi) > SM_I(sbi)->min_ipu_util)
555 return true;
556
557 /* this is only set during fdatasync */
558 if (policy & (0x1 << F2FS_IPU_FSYNC) &&
Jaegeuk Kim91942322016-05-20 10:13:22 -0700559 is_inode_flag_set(inode, FI_NEED_IPU))
Jaegeuk Kim9b5f1362014-09-16 18:30:54 -0700560 return true;
561
Jaegeuk Kim39a53e02012-11-28 13:37:31 +0900562 return false;
563}
564
565static inline unsigned int curseg_segno(struct f2fs_sb_info *sbi,
566 int type)
567{
568 struct curseg_info *curseg = CURSEG_I(sbi, type);
569 return curseg->segno;
570}
571
572static inline unsigned char curseg_alloc_type(struct f2fs_sb_info *sbi,
573 int type)
574{
575 struct curseg_info *curseg = CURSEG_I(sbi, type);
576 return curseg->alloc_type;
577}
578
579static inline unsigned short curseg_blkoff(struct f2fs_sb_info *sbi, int type)
580{
581 struct curseg_info *curseg = CURSEG_I(sbi, type);
582 return curseg->next_blkoff;
583}
584
585static inline void check_seg_range(struct f2fs_sb_info *sbi, unsigned int segno)
586{
Liu Xue7a04f642015-07-27 10:17:59 +0000587 f2fs_bug_on(sbi, segno > TOTAL_SEGS(sbi) - 1);
Jaegeuk Kim39a53e02012-11-28 13:37:31 +0900588}
589
Jaegeuk Kim39a53e02012-11-28 13:37:31 +0900590static inline void verify_block_addr(struct f2fs_sb_info *sbi, block_t blk_addr)
591{
Yunlei Hebb413d62016-07-28 12:12:38 +0800592 BUG_ON(blk_addr < SEG0_BLKADDR(sbi)
593 || blk_addr >= MAX_BLKADDR(sbi));
Jaegeuk Kim39a53e02012-11-28 13:37:31 +0900594}
595
596/*
arter97e1c42042014-08-06 23:22:50 +0900597 * Summary block is always treated as an invalid block
Jaegeuk Kim39a53e02012-11-28 13:37:31 +0900598 */
599static inline void check_block_count(struct f2fs_sb_info *sbi,
600 int segno, struct f2fs_sit_entry *raw_sit)
601{
Jaegeuk Kim4c278392015-08-11 16:01:30 -0700602#ifdef CONFIG_F2FS_CHECK_FS
Chao Yu44c60bf2013-10-29 14:50:40 +0800603 bool is_valid = test_bit_le(0, raw_sit->valid_map) ? true : false;
Jaegeuk Kim39a53e02012-11-28 13:37:31 +0900604 int valid_blocks = 0;
Chao Yu44c60bf2013-10-29 14:50:40 +0800605 int cur_pos = 0, next_pos;
Jaegeuk Kim39a53e02012-11-28 13:37:31 +0900606
Jaegeuk Kim39a53e02012-11-28 13:37:31 +0900607 /* check bitmap with valid block count */
Chao Yu44c60bf2013-10-29 14:50:40 +0800608 do {
609 if (is_valid) {
610 next_pos = find_next_zero_bit_le(&raw_sit->valid_map,
611 sbi->blocks_per_seg,
612 cur_pos);
613 valid_blocks += next_pos - cur_pos;
614 } else
615 next_pos = find_next_bit_le(&raw_sit->valid_map,
616 sbi->blocks_per_seg,
617 cur_pos);
618 cur_pos = next_pos;
619 is_valid = !is_valid;
620 } while (cur_pos < sbi->blocks_per_seg);
Jaegeuk Kim39a53e02012-11-28 13:37:31 +0900621 BUG_ON(GET_SIT_VBLOCKS(raw_sit) != valid_blocks);
Jaegeuk Kim5d56b672013-10-29 15:14:54 +0900622#endif
Jaegeuk Kim4c278392015-08-11 16:01:30 -0700623 /* check segment usage, and check boundary of a given segment number */
624 f2fs_bug_on(sbi, GET_SIT_VBLOCKS(raw_sit) > sbi->blocks_per_seg
625 || segno > TOTAL_SEGS(sbi) - 1);
Liu Xue7a04f642015-07-27 10:17:59 +0000626}
Jaegeuk Kim39a53e02012-11-28 13:37:31 +0900627
628static inline pgoff_t current_sit_addr(struct f2fs_sb_info *sbi,
629 unsigned int start)
630{
631 struct sit_info *sit_i = SIT_I(sbi);
Chao Yud3a14af2014-09-04 18:11:47 +0800632 unsigned int offset = SIT_BLOCK_OFFSET(start);
Jaegeuk Kim39a53e02012-11-28 13:37:31 +0900633 block_t blk_addr = sit_i->sit_base_addr + offset;
634
635 check_seg_range(sbi, start);
636
637 /* calculate sit block address */
638 if (f2fs_test_bit(offset, sit_i->sit_bitmap))
639 blk_addr += sit_i->sit_blocks;
640
641 return blk_addr;
642}
643
644static inline pgoff_t next_sit_addr(struct f2fs_sb_info *sbi,
645 pgoff_t block_addr)
646{
647 struct sit_info *sit_i = SIT_I(sbi);
648 block_addr -= sit_i->sit_base_addr;
649 if (block_addr < sit_i->sit_blocks)
650 block_addr += sit_i->sit_blocks;
651 else
652 block_addr -= sit_i->sit_blocks;
653
654 return block_addr + sit_i->sit_base_addr;
655}
656
657static inline void set_to_next_sit(struct sit_info *sit_i, unsigned int start)
658{
Chao Yud3a14af2014-09-04 18:11:47 +0800659 unsigned int block_off = SIT_BLOCK_OFFSET(start);
Jaegeuk Kim39a53e02012-11-28 13:37:31 +0900660
Gu Zhengc6ac4c02014-10-20 17:45:50 +0800661 f2fs_change_bit(block_off, sit_i->sit_bitmap);
Jaegeuk Kim39a53e02012-11-28 13:37:31 +0900662}
663
664static inline unsigned long long get_mtime(struct f2fs_sb_info *sbi)
665{
666 struct sit_info *sit_i = SIT_I(sbi);
667 return sit_i->elapsed_time + CURRENT_TIME_SEC.tv_sec -
668 sit_i->mounted_time;
669}
670
671static inline void set_summary(struct f2fs_summary *sum, nid_t nid,
672 unsigned int ofs_in_node, unsigned char version)
673{
674 sum->nid = cpu_to_le32(nid);
675 sum->ofs_in_node = cpu_to_le16(ofs_in_node);
676 sum->version = version;
677}
678
679static inline block_t start_sum_block(struct f2fs_sb_info *sbi)
680{
681 return __start_cp_addr(sbi) +
682 le32_to_cpu(F2FS_CKPT(sbi)->cp_pack_start_sum);
683}
684
685static inline block_t sum_blk_addr(struct f2fs_sb_info *sbi, int base, int type)
686{
687 return __start_cp_addr(sbi) +
688 le32_to_cpu(F2FS_CKPT(sbi)->cp_pack_total_block_count)
689 - (base + 1) + type;
690}
Jaegeuk Kim5ec4e492013-03-31 13:26:03 +0900691
692static inline bool sec_usage_check(struct f2fs_sb_info *sbi, unsigned int secno)
693{
694 if (IS_CURSEC(sbi, secno) || (sbi->cur_victim_sec == secno))
695 return true;
696 return false;
697}
Jaegeuk Kimac5d1562013-04-29 16:58:39 +0900698
699static inline unsigned int max_hw_blocks(struct f2fs_sb_info *sbi)
700{
701 struct block_device *bdev = sbi->sb->s_bdev;
702 struct request_queue *q = bdev_get_queue(bdev);
Chao Yu55cf9cb2014-09-15 18:01:10 +0800703 return SECTOR_TO_BLOCK(queue_max_sectors(q));
Jaegeuk Kimac5d1562013-04-29 16:58:39 +0900704}
Jaegeuk Kim87d6f892014-03-18 12:40:49 +0900705
706/*
707 * It is very important to gather dirty pages and write at once, so that we can
708 * submit a big bio without interfering other data writes.
709 * By default, 512 pages for directory data,
710 * 512 pages (2MB) * 3 for three types of nodes, and
711 * max_bio_blocks for meta are set.
712 */
713static inline int nr_pages_to_skip(struct f2fs_sb_info *sbi, int type)
714{
Tejun Heoa88a3412015-05-22 17:13:28 -0400715 if (sbi->sb->s_bdi->wb.dirty_exceeded)
Jaegeuk Kim510184c2014-11-06 17:23:08 -0800716 return 0;
717
Jaegeuk Kima1257022015-10-08 10:40:07 -0700718 if (type == DATA)
719 return sbi->blocks_per_seg;
720 else if (type == NODE)
Jaegeuk Kim2c237eb2016-06-16 16:44:11 -0700721 return 8 * sbi->blocks_per_seg;
Jaegeuk Kim87d6f892014-03-18 12:40:49 +0900722 else if (type == META)
Jaegeuk Kim2c237eb2016-06-16 16:44:11 -0700723 return 8 * MAX_BIO_BLOCKS(sbi);
Jaegeuk Kim87d6f892014-03-18 12:40:49 +0900724 else
725 return 0;
726}
Jaegeuk Kim50c8cdb2014-03-18 13:47:11 +0900727
728/*
729 * When writing pages, it'd better align nr_to_write for segment size.
730 */
731static inline long nr_pages_to_write(struct f2fs_sb_info *sbi, int type,
732 struct writeback_control *wbc)
733{
734 long nr_to_write, desired;
735
736 if (wbc->sync_mode != WB_SYNC_NONE)
737 return 0;
738
739 nr_to_write = wbc->nr_to_write;
740
Jaegeuk Kim28ea6162016-05-25 17:17:56 -0700741 if (type == NODE)
Jaegeuk Kim2c237eb2016-06-16 16:44:11 -0700742 desired = 2 * max_hw_blocks(sbi);
Jaegeuk Kim50c8cdb2014-03-18 13:47:11 +0900743 else
Jaegeuk Kim90a893c2014-09-22 16:21:07 -0700744 desired = MAX_BIO_BLOCKS(sbi);
Jaegeuk Kim50c8cdb2014-03-18 13:47:11 +0900745
746 wbc->nr_to_write = desired;
747 return desired - nr_to_write;
748}