blob: 08f1455c812c1a49c027a47d0223dff4adfb44bd [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
Jaegeuk Kim2040fce82016-12-05 13:56:04 -080021#define F2FS_MIN_SEGMENTS 9 /* SB + 2 (CP + SIT + NAT) + SSA + MAIN */
22
Namjae Jeon6224da82013-04-06 14:44:32 +090023/* L: Logical segment # in volume, R: Relative segment # in main area */
Jaegeuk Kim39a53e02012-11-28 13:37:31 +090024#define GET_L2R_SEGNO(free_i, segno) (segno - free_i->start_segno)
25#define GET_R2L_SEGNO(free_i, segno) (segno + free_i->start_segno)
26
Changman Lee61ae45c2013-11-21 20:04:21 +090027#define IS_DATASEG(t) (t <= CURSEG_COLD_DATA)
28#define IS_NODESEG(t) (t >= CURSEG_HOT_NODE)
Jaegeuk Kim39a53e02012-11-28 13:37:31 +090029
Jaegeuk Kim5c773ba2013-03-31 12:30:04 +090030#define IS_CURSEG(sbi, seg) \
31 ((seg == CURSEG_I(sbi, CURSEG_HOT_DATA)->segno) || \
32 (seg == CURSEG_I(sbi, CURSEG_WARM_DATA)->segno) || \
33 (seg == CURSEG_I(sbi, CURSEG_COLD_DATA)->segno) || \
34 (seg == CURSEG_I(sbi, CURSEG_HOT_NODE)->segno) || \
35 (seg == CURSEG_I(sbi, CURSEG_WARM_NODE)->segno) || \
36 (seg == CURSEG_I(sbi, CURSEG_COLD_NODE)->segno))
Jaegeuk Kim39a53e02012-11-28 13:37:31 +090037
38#define IS_CURSEC(sbi, secno) \
39 ((secno == CURSEG_I(sbi, CURSEG_HOT_DATA)->segno / \
40 sbi->segs_per_sec) || \
41 (secno == CURSEG_I(sbi, CURSEG_WARM_DATA)->segno / \
42 sbi->segs_per_sec) || \
43 (secno == CURSEG_I(sbi, CURSEG_COLD_DATA)->segno / \
44 sbi->segs_per_sec) || \
45 (secno == CURSEG_I(sbi, CURSEG_HOT_NODE)->segno / \
46 sbi->segs_per_sec) || \
47 (secno == CURSEG_I(sbi, CURSEG_WARM_NODE)->segno / \
48 sbi->segs_per_sec) || \
49 (secno == CURSEG_I(sbi, CURSEG_COLD_NODE)->segno / \
50 sbi->segs_per_sec)) \
51
Jaegeuk Kim7cd85582014-09-23 11:23:01 -070052#define MAIN_BLKADDR(sbi) (SM_I(sbi)->main_blkaddr)
53#define SEG0_BLKADDR(sbi) (SM_I(sbi)->seg0_blkaddr)
54
55#define MAIN_SEGS(sbi) (SM_I(sbi)->main_segments)
56#define MAIN_SECS(sbi) (sbi->total_sections)
57
58#define TOTAL_SEGS(sbi) (SM_I(sbi)->segment_count)
59#define TOTAL_BLKS(sbi) (TOTAL_SEGS(sbi) << sbi->log_blocks_per_seg)
60
61#define MAX_BLKADDR(sbi) (SEG0_BLKADDR(sbi) + TOTAL_BLKS(sbi))
Dan Carpenter8a219842014-09-25 14:39:17 +030062#define SEGMENT_SIZE(sbi) (1ULL << (sbi->log_blocksize + \
Jaegeuk Kim7cd85582014-09-23 11:23:01 -070063 sbi->log_blocks_per_seg))
64
65#define START_BLOCK(sbi, segno) (SEG0_BLKADDR(sbi) + \
Jaegeuk Kim39a53e02012-11-28 13:37:31 +090066 (GET_R2L_SEGNO(FREE_I(sbi), segno) << sbi->log_blocks_per_seg))
Jaegeuk Kim7cd85582014-09-23 11:23:01 -070067
Jaegeuk Kim39a53e02012-11-28 13:37:31 +090068#define NEXT_FREE_BLKADDR(sbi, curseg) \
69 (START_BLOCK(sbi, curseg->segno) + curseg->next_blkoff)
70
Jaegeuk Kim7cd85582014-09-23 11:23:01 -070071#define GET_SEGOFF_FROM_SEG0(sbi, blk_addr) ((blk_addr) - SEG0_BLKADDR(sbi))
Jaegeuk Kim39a53e02012-11-28 13:37:31 +090072#define GET_SEGNO_FROM_SEG0(sbi, blk_addr) \
73 (GET_SEGOFF_FROM_SEG0(sbi, blk_addr) >> sbi->log_blocks_per_seg)
Jaegeuk Kim491c0852014-02-04 13:01:10 +090074#define GET_BLKOFF_FROM_SEG0(sbi, blk_addr) \
75 (GET_SEGOFF_FROM_SEG0(sbi, blk_addr) & (sbi->blocks_per_seg - 1))
76
Jaegeuk Kim39a53e02012-11-28 13:37:31 +090077#define GET_SEGNO(sbi, blk_addr) \
78 (((blk_addr == NULL_ADDR) || (blk_addr == NEW_ADDR)) ? \
79 NULL_SEGNO : GET_L2R_SEGNO(FREE_I(sbi), \
80 GET_SEGNO_FROM_SEG0(sbi, blk_addr)))
81#define GET_SECNO(sbi, segno) \
82 ((segno) / sbi->segs_per_sec)
83#define GET_ZONENO_FROM_SEGNO(sbi, segno) \
84 ((segno / sbi->segs_per_sec) / sbi->secs_per_zone)
85
86#define GET_SUM_BLOCK(sbi, segno) \
87 ((sbi->sm_info->ssa_blkaddr) + segno)
88
89#define GET_SUM_TYPE(footer) ((footer)->entry_type)
90#define SET_SUM_TYPE(footer, type) ((footer)->entry_type = type)
91
92#define SIT_ENTRY_OFFSET(sit_i, segno) \
93 (segno % sit_i->sents_per_block)
Chao Yud3a14af2014-09-04 18:11:47 +080094#define SIT_BLOCK_OFFSET(segno) \
Jaegeuk Kim39a53e02012-11-28 13:37:31 +090095 (segno / SIT_ENTRY_PER_BLOCK)
Chao Yud3a14af2014-09-04 18:11:47 +080096#define START_SEGNO(segno) \
97 (SIT_BLOCK_OFFSET(segno) * SIT_ENTRY_PER_BLOCK)
Chao Yu74de5932013-11-22 09:09:59 +080098#define SIT_BLK_CNT(sbi) \
Jaegeuk Kim7cd85582014-09-23 11:23:01 -070099 ((MAIN_SEGS(sbi) + SIT_ENTRY_PER_BLOCK - 1) / SIT_ENTRY_PER_BLOCK)
Jaegeuk Kim39a53e02012-11-28 13:37:31 +0900100#define f2fs_bitmap_size(nr) \
101 (BITS_TO_LONGS(nr) * sizeof(unsigned long))
Jaegeuk Kim39a53e02012-11-28 13:37:31 +0900102
Chao Yu55cf9cb2014-09-15 18:01:10 +0800103#define SECTOR_FROM_BLOCK(blk_addr) \
104 (((sector_t)blk_addr) << F2FS_LOG_SECTORS_PER_BLOCK)
105#define SECTOR_TO_BLOCK(sectors) \
106 (sectors >> F2FS_LOG_SECTORS_PER_BLOCK)
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)
Jaegeuk Kim0a595eb2016-12-14 10:12:56 -0800189#define DUMMY_WRITTEN_PAGE ((unsigned long)-2)
Chao Yudecd36b2015-08-07 18:42:09 +0800190
191#define IS_ATOMIC_WRITTEN_PAGE(page) \
192 (page_private(page) == (unsigned long)ATOMIC_WRITTEN_PAGE)
Jaegeuk Kim0a595eb2016-12-14 10:12:56 -0800193#define IS_DUMMY_WRITTEN_PAGE(page) \
194 (page_private(page) == (unsigned long)DUMMY_WRITTEN_PAGE)
Chao Yudecd36b2015-08-07 18:42:09 +0800195
Jaegeuk Kim88b88a62014-10-06 17:39:50 -0700196struct inmem_pages {
197 struct list_head list;
198 struct page *page;
Chao Yu28bc1062016-02-06 14:40:34 +0800199 block_t old_addr; /* for revoking when fail to commit */
Jaegeuk Kim88b88a62014-10-06 17:39:50 -0700200};
201
Jaegeuk Kim39a53e02012-11-28 13:37:31 +0900202struct sit_info {
203 const struct segment_allocation *s_ops;
204
205 block_t sit_base_addr; /* start block address of SIT area */
206 block_t sit_blocks; /* # of blocks used by SIT area */
207 block_t written_valid_blocks; /* # of valid blocks in main area */
208 char *sit_bitmap; /* SIT bitmap pointer */
209 unsigned int bitmap_size; /* SIT bitmap size */
210
Jaegeuk Kim60a3b782015-02-10 16:44:29 -0800211 unsigned long *tmp_map; /* bitmap for temporal use */
Jaegeuk Kim39a53e02012-11-28 13:37:31 +0900212 unsigned long *dirty_sentries_bitmap; /* bitmap for dirty sentries */
213 unsigned int dirty_sentries; /* # of dirty sentries */
214 unsigned int sents_per_block; /* # of SIT entries per block */
215 struct mutex sentry_lock; /* to protect SIT cache */
216 struct seg_entry *sentries; /* SIT segment-level cache */
217 struct sec_entry *sec_entries; /* SIT section-level cache */
218
219 /* for cost-benefit algorithm in cleaning procedure */
220 unsigned long long elapsed_time; /* elapsed time after mount */
221 unsigned long long mounted_time; /* mount time */
222 unsigned long long min_mtime; /* min. modification time */
223 unsigned long long max_mtime; /* max. modification time */
224};
225
226struct free_segmap_info {
227 unsigned int start_segno; /* start segment number logically */
228 unsigned int free_segments; /* # of free segments */
229 unsigned int free_sections; /* # of free sections */
Chao Yu1a118cc2015-02-11 18:20:38 +0800230 spinlock_t segmap_lock; /* free segmap lock */
Jaegeuk Kim39a53e02012-11-28 13:37:31 +0900231 unsigned long *free_segmap; /* free segment bitmap */
232 unsigned long *free_secmap; /* free section bitmap */
233};
234
235/* Notice: The order of dirty type is same with CURSEG_XXX in f2fs.h */
236enum dirty_type {
237 DIRTY_HOT_DATA, /* dirty segments assigned as hot data logs */
238 DIRTY_WARM_DATA, /* dirty segments assigned as warm data logs */
239 DIRTY_COLD_DATA, /* dirty segments assigned as cold data logs */
240 DIRTY_HOT_NODE, /* dirty segments assigned as hot node logs */
241 DIRTY_WARM_NODE, /* dirty segments assigned as warm node logs */
242 DIRTY_COLD_NODE, /* dirty segments assigned as cold node logs */
243 DIRTY, /* to count # of dirty segments */
244 PRE, /* to count # of entirely obsolete segments */
245 NR_DIRTY_TYPE
246};
247
248struct dirty_seglist_info {
249 const struct victim_selection *v_ops; /* victim selction operation */
250 unsigned long *dirty_segmap[NR_DIRTY_TYPE];
251 struct mutex seglist_lock; /* lock for segment bitmaps */
252 int nr_dirty[NR_DIRTY_TYPE]; /* # of dirty segments */
Jaegeuk Kim5ec4e492013-03-31 13:26:03 +0900253 unsigned long *victim_secmap; /* background GC victims */
Jaegeuk Kim39a53e02012-11-28 13:37:31 +0900254};
255
256/* victim selection function for cleaning and SSR */
257struct victim_selection {
258 int (*get_victim)(struct f2fs_sb_info *, unsigned int *,
259 int, int, char);
260};
261
262/* for active log information */
263struct curseg_info {
264 struct mutex curseg_mutex; /* lock for consistency */
265 struct f2fs_summary_block *sum_blk; /* cached summary block */
Chao Yub7ad7512016-02-19 18:08:46 +0800266 struct rw_semaphore journal_rwsem; /* protect journal area */
267 struct f2fs_journal *journal; /* cached journal info */
Jaegeuk Kim39a53e02012-11-28 13:37:31 +0900268 unsigned char alloc_type; /* current allocation type */
269 unsigned int segno; /* current segment number */
270 unsigned short next_blkoff; /* next block offset to write */
271 unsigned int zone; /* current zone number */
272 unsigned int next_segno; /* preallocated segment */
273};
274
Chao Yu184a5cd2014-09-04 18:13:01 +0800275struct sit_entry_set {
276 struct list_head set_list; /* link with all sit sets */
277 unsigned int start_segno; /* start segno of sits in set */
278 unsigned int entry_cnt; /* the # of sit entries in set */
279};
280
Jaegeuk Kim39a53e02012-11-28 13:37:31 +0900281/*
282 * inline functions
283 */
284static inline struct curseg_info *CURSEG_I(struct f2fs_sb_info *sbi, int type)
285{
286 return (struct curseg_info *)(SM_I(sbi)->curseg_array + type);
287}
288
289static inline struct seg_entry *get_seg_entry(struct f2fs_sb_info *sbi,
290 unsigned int segno)
291{
292 struct sit_info *sit_i = SIT_I(sbi);
293 return &sit_i->sentries[segno];
294}
295
296static inline struct sec_entry *get_sec_entry(struct f2fs_sb_info *sbi,
297 unsigned int segno)
298{
299 struct sit_info *sit_i = SIT_I(sbi);
300 return &sit_i->sec_entries[GET_SECNO(sbi, segno)];
301}
302
303static inline unsigned int get_valid_blocks(struct f2fs_sb_info *sbi,
304 unsigned int segno, int section)
305{
306 /*
307 * In order to get # of valid blocks in a section instantly from many
308 * segments, f2fs manages two counting structures separately.
309 */
310 if (section > 1)
311 return get_sec_entry(sbi, segno)->valid_blocks;
312 else
313 return get_seg_entry(sbi, segno)->valid_blocks;
314}
315
316static inline void seg_info_from_raw_sit(struct seg_entry *se,
317 struct f2fs_sit_entry *rs)
318{
319 se->valid_blocks = GET_SIT_VBLOCKS(rs);
320 se->ckpt_valid_blocks = GET_SIT_VBLOCKS(rs);
321 memcpy(se->cur_valid_map, rs->valid_map, SIT_VBLOCK_MAP_SIZE);
322 memcpy(se->ckpt_valid_map, rs->valid_map, SIT_VBLOCK_MAP_SIZE);
323 se->type = GET_SIT_TYPE(rs);
324 se->mtime = le64_to_cpu(rs->mtime);
325}
326
327static inline void seg_info_to_raw_sit(struct seg_entry *se,
328 struct f2fs_sit_entry *rs)
329{
330 unsigned short raw_vblocks = (se->type << SIT_VBLOCKS_SHIFT) |
331 se->valid_blocks;
332 rs->vblocks = cpu_to_le16(raw_vblocks);
333 memcpy(rs->valid_map, se->cur_valid_map, SIT_VBLOCK_MAP_SIZE);
334 memcpy(se->ckpt_valid_map, rs->valid_map, SIT_VBLOCK_MAP_SIZE);
335 se->ckpt_valid_blocks = se->valid_blocks;
336 rs->mtime = cpu_to_le64(se->mtime);
337}
338
339static inline unsigned int find_next_inuse(struct free_segmap_info *free_i,
340 unsigned int max, unsigned int segno)
341{
342 unsigned int ret;
Chao Yu1a118cc2015-02-11 18:20:38 +0800343 spin_lock(&free_i->segmap_lock);
Jaegeuk Kim39a53e02012-11-28 13:37:31 +0900344 ret = find_next_bit(free_i->free_segmap, max, segno);
Chao Yu1a118cc2015-02-11 18:20:38 +0800345 spin_unlock(&free_i->segmap_lock);
Jaegeuk Kim39a53e02012-11-28 13:37:31 +0900346 return ret;
347}
348
349static inline void __set_free(struct f2fs_sb_info *sbi, unsigned int segno)
350{
351 struct free_segmap_info *free_i = FREE_I(sbi);
352 unsigned int secno = segno / sbi->segs_per_sec;
353 unsigned int start_segno = secno * sbi->segs_per_sec;
354 unsigned int next;
355
Chao Yu1a118cc2015-02-11 18:20:38 +0800356 spin_lock(&free_i->segmap_lock);
Jaegeuk Kim39a53e02012-11-28 13:37:31 +0900357 clear_bit(segno, free_i->free_segmap);
358 free_i->free_segments++;
359
Wanpeng Li7fd97012015-03-06 15:00:55 +0800360 next = find_next_bit(free_i->free_segmap,
361 start_segno + sbi->segs_per_sec, start_segno);
Jaegeuk Kim39a53e02012-11-28 13:37:31 +0900362 if (next >= start_segno + sbi->segs_per_sec) {
363 clear_bit(secno, free_i->free_secmap);
364 free_i->free_sections++;
365 }
Chao Yu1a118cc2015-02-11 18:20:38 +0800366 spin_unlock(&free_i->segmap_lock);
Jaegeuk Kim39a53e02012-11-28 13:37:31 +0900367}
368
369static inline void __set_inuse(struct f2fs_sb_info *sbi,
370 unsigned int segno)
371{
372 struct free_segmap_info *free_i = FREE_I(sbi);
373 unsigned int secno = segno / sbi->segs_per_sec;
374 set_bit(segno, free_i->free_segmap);
375 free_i->free_segments--;
376 if (!test_and_set_bit(secno, free_i->free_secmap))
377 free_i->free_sections--;
378}
379
380static inline void __set_test_and_free(struct f2fs_sb_info *sbi,
381 unsigned int segno)
382{
383 struct free_segmap_info *free_i = FREE_I(sbi);
384 unsigned int secno = segno / sbi->segs_per_sec;
385 unsigned int start_segno = secno * sbi->segs_per_sec;
386 unsigned int next;
387
Chao Yu1a118cc2015-02-11 18:20:38 +0800388 spin_lock(&free_i->segmap_lock);
Jaegeuk Kim39a53e02012-11-28 13:37:31 +0900389 if (test_and_clear_bit(segno, free_i->free_segmap)) {
390 free_i->free_segments++;
391
Chao Yuf1121ab2014-07-14 16:45:15 +0800392 next = find_next_bit(free_i->free_segmap,
393 start_segno + sbi->segs_per_sec, start_segno);
Jaegeuk Kim39a53e02012-11-28 13:37:31 +0900394 if (next >= start_segno + sbi->segs_per_sec) {
395 if (test_and_clear_bit(secno, free_i->free_secmap))
396 free_i->free_sections++;
397 }
398 }
Chao Yu1a118cc2015-02-11 18:20:38 +0800399 spin_unlock(&free_i->segmap_lock);
Jaegeuk Kim39a53e02012-11-28 13:37:31 +0900400}
401
402static inline void __set_test_and_inuse(struct f2fs_sb_info *sbi,
403 unsigned int segno)
404{
405 struct free_segmap_info *free_i = FREE_I(sbi);
406 unsigned int secno = segno / sbi->segs_per_sec;
Chao Yu1a118cc2015-02-11 18:20:38 +0800407 spin_lock(&free_i->segmap_lock);
Jaegeuk Kim39a53e02012-11-28 13:37:31 +0900408 if (!test_and_set_bit(segno, free_i->free_segmap)) {
409 free_i->free_segments--;
410 if (!test_and_set_bit(secno, free_i->free_secmap))
411 free_i->free_sections--;
412 }
Chao Yu1a118cc2015-02-11 18:20:38 +0800413 spin_unlock(&free_i->segmap_lock);
Jaegeuk Kim39a53e02012-11-28 13:37:31 +0900414}
415
416static inline void get_sit_bitmap(struct f2fs_sb_info *sbi,
417 void *dst_addr)
418{
419 struct sit_info *sit_i = SIT_I(sbi);
420 memcpy(dst_addr, sit_i->sit_bitmap, sit_i->bitmap_size);
421}
422
423static inline block_t written_block_count(struct f2fs_sb_info *sbi)
424{
Jaegeuk Kim8b8343f2014-02-24 13:00:13 +0900425 return SIT_I(sbi)->written_valid_blocks;
Jaegeuk Kim39a53e02012-11-28 13:37:31 +0900426}
427
428static inline unsigned int free_segments(struct f2fs_sb_info *sbi)
429{
Jaegeuk Kim8b8343f2014-02-24 13:00:13 +0900430 return FREE_I(sbi)->free_segments;
Jaegeuk Kim39a53e02012-11-28 13:37:31 +0900431}
432
433static inline int reserved_segments(struct f2fs_sb_info *sbi)
434{
435 return SM_I(sbi)->reserved_segments;
436}
437
438static inline unsigned int free_sections(struct f2fs_sb_info *sbi)
439{
Jaegeuk Kim8b8343f2014-02-24 13:00:13 +0900440 return FREE_I(sbi)->free_sections;
Jaegeuk Kim39a53e02012-11-28 13:37:31 +0900441}
442
443static inline unsigned int prefree_segments(struct f2fs_sb_info *sbi)
444{
445 return DIRTY_I(sbi)->nr_dirty[PRE];
446}
447
448static inline unsigned int dirty_segments(struct f2fs_sb_info *sbi)
449{
450 return DIRTY_I(sbi)->nr_dirty[DIRTY_HOT_DATA] +
451 DIRTY_I(sbi)->nr_dirty[DIRTY_WARM_DATA] +
452 DIRTY_I(sbi)->nr_dirty[DIRTY_COLD_DATA] +
453 DIRTY_I(sbi)->nr_dirty[DIRTY_HOT_NODE] +
454 DIRTY_I(sbi)->nr_dirty[DIRTY_WARM_NODE] +
455 DIRTY_I(sbi)->nr_dirty[DIRTY_COLD_NODE];
456}
457
458static inline int overprovision_segments(struct f2fs_sb_info *sbi)
459{
460 return SM_I(sbi)->ovp_segments;
461}
462
463static inline int overprovision_sections(struct f2fs_sb_info *sbi)
464{
465 return ((unsigned int) overprovision_segments(sbi)) / sbi->segs_per_sec;
466}
467
468static inline int reserved_sections(struct f2fs_sb_info *sbi)
469{
470 return ((unsigned int) reserved_segments(sbi)) / sbi->segs_per_sec;
471}
472
473static inline bool need_SSR(struct f2fs_sb_info *sbi)
474{
Jaegeuk Kim95dd8972014-09-17 17:52:58 -0700475 int node_secs = get_blocktype_secs(sbi, F2FS_DIRTY_NODES);
476 int dent_secs = get_blocktype_secs(sbi, F2FS_DIRTY_DENTS);
Jaegeuk Kimb9610bd2016-10-14 13:28:05 -0700477 int imeta_secs = get_blocktype_secs(sbi, F2FS_DIRTY_IMETA);
Jaegeuk Kim36abef42016-06-03 19:29:38 -0700478
479 if (test_opt(sbi, LFS))
480 return false;
481
Jaegeuk Kimb9610bd2016-10-14 13:28:05 -0700482 return free_sections(sbi) <= (node_secs + 2 * dent_secs + imeta_secs +
Jaegeuk Kim95dd8972014-09-17 17:52:58 -0700483 reserved_sections(sbi) + 1);
Jaegeuk Kim39a53e02012-11-28 13:37:31 +0900484}
485
Jaegeuk Kim7f3037a2016-09-01 12:02:51 -0700486static inline bool has_not_enough_free_secs(struct f2fs_sb_info *sbi,
487 int freed, int needed)
Jaegeuk Kim39a53e02012-11-28 13:37:31 +0900488{
Namjae Jeon5ac206c2013-02-02 23:52:59 +0900489 int node_secs = get_blocktype_secs(sbi, F2FS_DIRTY_NODES);
490 int dent_secs = get_blocktype_secs(sbi, F2FS_DIRTY_DENTS);
Jaegeuk Kimb9610bd2016-10-14 13:28:05 -0700491 int imeta_secs = get_blocktype_secs(sbi, F2FS_DIRTY_IMETA);
Jaegeuk Kim0f18b462016-05-20 11:10:10 -0700492
Chao Yucaf00472015-01-28 17:48:42 +0800493 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
Jaegeuk Kim029cd282012-12-21 17:20:21 +0900494 return false;
495
Jaegeuk Kim7f3037a2016-09-01 12:02:51 -0700496 return (free_sections(sbi) + freed) <=
Jaegeuk Kimb9610bd2016-10-14 13:28:05 -0700497 (node_secs + 2 * dent_secs + imeta_secs +
498 reserved_sections(sbi) + needed);
Jaegeuk Kim39a53e02012-11-28 13:37:31 +0900499}
500
Jaegeuk Kim81eb8d62013-10-24 13:31:34 +0900501static inline bool excess_prefree_segs(struct f2fs_sb_info *sbi)
502{
Chris Fries6c311ec2014-01-17 14:44:39 -0600503 return prefree_segments(sbi) > SM_I(sbi)->rec_prefree_segments;
Jaegeuk Kim81eb8d62013-10-24 13:31:34 +0900504}
505
Jaegeuk Kim39a53e02012-11-28 13:37:31 +0900506static inline int utilization(struct f2fs_sb_info *sbi)
507{
Chris Fries6c311ec2014-01-17 14:44:39 -0600508 return div_u64((u64)valid_user_blocks(sbi) * 100,
509 sbi->user_block_count);
Jaegeuk Kim39a53e02012-11-28 13:37:31 +0900510}
511
512/*
513 * Sometimes f2fs may be better to drop out-of-place update policy.
Jaegeuk Kim216fbd62013-11-07 13:13:42 +0900514 * And, users can control the policy through sysfs entries.
515 * There are five policies with triggering conditions as follows.
516 * F2FS_IPU_FORCE - all the time,
517 * F2FS_IPU_SSR - if SSR mode is activated,
518 * F2FS_IPU_UTIL - if FS utilization is over threashold,
519 * F2FS_IPU_SSR_UTIL - if SSR mode is activated and FS utilization is over
520 * threashold,
Jaegeuk Kimc1ce1b02014-09-10 16:53:02 -0700521 * F2FS_IPU_FSYNC - activated in fsync path only for high performance flash
522 * storages. IPU will be triggered only if the # of dirty
523 * pages over min_fsync_blocks.
Jaegeuk Kim216fbd62013-11-07 13:13:42 +0900524 * F2FS_IPUT_DISABLE - disable IPU. (=default option)
Jaegeuk Kim39a53e02012-11-28 13:37:31 +0900525 */
Jaegeuk Kim216fbd62013-11-07 13:13:42 +0900526#define DEF_MIN_IPU_UTIL 70
Jaegeuk Kimc1ce1b02014-09-10 16:53:02 -0700527#define DEF_MIN_FSYNC_BLOCKS 8
Jaegeuk Kim216fbd62013-11-07 13:13:42 +0900528
529enum {
530 F2FS_IPU_FORCE,
531 F2FS_IPU_SSR,
532 F2FS_IPU_UTIL,
533 F2FS_IPU_SSR_UTIL,
Jaegeuk Kimc1ce1b02014-09-10 16:53:02 -0700534 F2FS_IPU_FSYNC,
Jaegeuk Kim216fbd62013-11-07 13:13:42 +0900535};
536
Jaegeuk Kim39a53e02012-11-28 13:37:31 +0900537static inline bool need_inplace_update(struct inode *inode)
538{
Jaegeuk Kim40813632014-09-02 15:31:18 -0700539 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
Jaegeuk Kim9b5f1362014-09-16 18:30:54 -0700540 unsigned int policy = SM_I(sbi)->ipu_policy;
Jaegeuk Kim216fbd62013-11-07 13:13:42 +0900541
542 /* IPU can be done only for the user data */
Jaegeuk Kim88b88a62014-10-06 17:39:50 -0700543 if (S_ISDIR(inode->i_mode) || f2fs_is_atomic_file(inode))
Jaegeuk Kim39a53e02012-11-28 13:37:31 +0900544 return false;
Jaegeuk Kim216fbd62013-11-07 13:13:42 +0900545
Jaegeuk Kim36abef42016-06-03 19:29:38 -0700546 if (test_opt(sbi, LFS))
547 return false;
548
Jaegeuk Kim9b5f1362014-09-16 18:30:54 -0700549 if (policy & (0x1 << F2FS_IPU_FORCE))
Jaegeuk Kim39a53e02012-11-28 13:37:31 +0900550 return true;
Jaegeuk Kim9b5f1362014-09-16 18:30:54 -0700551 if (policy & (0x1 << F2FS_IPU_SSR) && need_SSR(sbi))
552 return true;
553 if (policy & (0x1 << F2FS_IPU_UTIL) &&
554 utilization(sbi) > SM_I(sbi)->min_ipu_util)
555 return true;
556 if (policy & (0x1 << F2FS_IPU_SSR_UTIL) && need_SSR(sbi) &&
557 utilization(sbi) > SM_I(sbi)->min_ipu_util)
558 return true;
559
560 /* this is only set during fdatasync */
561 if (policy & (0x1 << F2FS_IPU_FSYNC) &&
Jaegeuk Kim91942322016-05-20 10:13:22 -0700562 is_inode_flag_set(inode, FI_NEED_IPU))
Jaegeuk Kim9b5f1362014-09-16 18:30:54 -0700563 return true;
564
Jaegeuk Kim39a53e02012-11-28 13:37:31 +0900565 return false;
566}
567
568static inline unsigned int curseg_segno(struct f2fs_sb_info *sbi,
569 int type)
570{
571 struct curseg_info *curseg = CURSEG_I(sbi, type);
572 return curseg->segno;
573}
574
575static inline unsigned char curseg_alloc_type(struct f2fs_sb_info *sbi,
576 int type)
577{
578 struct curseg_info *curseg = CURSEG_I(sbi, type);
579 return curseg->alloc_type;
580}
581
582static inline unsigned short curseg_blkoff(struct f2fs_sb_info *sbi, int type)
583{
584 struct curseg_info *curseg = CURSEG_I(sbi, type);
585 return curseg->next_blkoff;
586}
587
588static inline void check_seg_range(struct f2fs_sb_info *sbi, unsigned int segno)
589{
Liu Xue7a04f642015-07-27 10:17:59 +0000590 f2fs_bug_on(sbi, segno > TOTAL_SEGS(sbi) - 1);
Jaegeuk Kim39a53e02012-11-28 13:37:31 +0900591}
592
Jaegeuk Kim39a53e02012-11-28 13:37:31 +0900593static inline void verify_block_addr(struct f2fs_sb_info *sbi, block_t blk_addr)
594{
Yunlei Hebb413d62016-07-28 12:12:38 +0800595 BUG_ON(blk_addr < SEG0_BLKADDR(sbi)
596 || blk_addr >= MAX_BLKADDR(sbi));
Jaegeuk Kim39a53e02012-11-28 13:37:31 +0900597}
598
599/*
arter97e1c42042014-08-06 23:22:50 +0900600 * Summary block is always treated as an invalid block
Jaegeuk Kim39a53e02012-11-28 13:37:31 +0900601 */
602static inline void check_block_count(struct f2fs_sb_info *sbi,
603 int segno, struct f2fs_sit_entry *raw_sit)
604{
Jaegeuk Kim4c278392015-08-11 16:01:30 -0700605#ifdef CONFIG_F2FS_CHECK_FS
Chao Yu44c60bf2013-10-29 14:50:40 +0800606 bool is_valid = test_bit_le(0, raw_sit->valid_map) ? true : false;
Jaegeuk Kim39a53e02012-11-28 13:37:31 +0900607 int valid_blocks = 0;
Chao Yu44c60bf2013-10-29 14:50:40 +0800608 int cur_pos = 0, next_pos;
Jaegeuk Kim39a53e02012-11-28 13:37:31 +0900609
Jaegeuk Kim39a53e02012-11-28 13:37:31 +0900610 /* check bitmap with valid block count */
Chao Yu44c60bf2013-10-29 14:50:40 +0800611 do {
612 if (is_valid) {
613 next_pos = find_next_zero_bit_le(&raw_sit->valid_map,
614 sbi->blocks_per_seg,
615 cur_pos);
616 valid_blocks += next_pos - cur_pos;
617 } else
618 next_pos = find_next_bit_le(&raw_sit->valid_map,
619 sbi->blocks_per_seg,
620 cur_pos);
621 cur_pos = next_pos;
622 is_valid = !is_valid;
623 } while (cur_pos < sbi->blocks_per_seg);
Jaegeuk Kim39a53e02012-11-28 13:37:31 +0900624 BUG_ON(GET_SIT_VBLOCKS(raw_sit) != valid_blocks);
Jaegeuk Kim5d56b672013-10-29 15:14:54 +0900625#endif
Jaegeuk Kim4c278392015-08-11 16:01:30 -0700626 /* check segment usage, and check boundary of a given segment number */
627 f2fs_bug_on(sbi, GET_SIT_VBLOCKS(raw_sit) > sbi->blocks_per_seg
628 || segno > TOTAL_SEGS(sbi) - 1);
Liu Xue7a04f642015-07-27 10:17:59 +0000629}
Jaegeuk Kim39a53e02012-11-28 13:37:31 +0900630
631static inline pgoff_t current_sit_addr(struct f2fs_sb_info *sbi,
632 unsigned int start)
633{
634 struct sit_info *sit_i = SIT_I(sbi);
Chao Yud3a14af2014-09-04 18:11:47 +0800635 unsigned int offset = SIT_BLOCK_OFFSET(start);
Jaegeuk Kim39a53e02012-11-28 13:37:31 +0900636 block_t blk_addr = sit_i->sit_base_addr + offset;
637
638 check_seg_range(sbi, start);
639
640 /* calculate sit block address */
641 if (f2fs_test_bit(offset, sit_i->sit_bitmap))
642 blk_addr += sit_i->sit_blocks;
643
644 return blk_addr;
645}
646
647static inline pgoff_t next_sit_addr(struct f2fs_sb_info *sbi,
648 pgoff_t block_addr)
649{
650 struct sit_info *sit_i = SIT_I(sbi);
651 block_addr -= sit_i->sit_base_addr;
652 if (block_addr < sit_i->sit_blocks)
653 block_addr += sit_i->sit_blocks;
654 else
655 block_addr -= sit_i->sit_blocks;
656
657 return block_addr + sit_i->sit_base_addr;
658}
659
660static inline void set_to_next_sit(struct sit_info *sit_i, unsigned int start)
661{
Chao Yud3a14af2014-09-04 18:11:47 +0800662 unsigned int block_off = SIT_BLOCK_OFFSET(start);
Jaegeuk Kim39a53e02012-11-28 13:37:31 +0900663
Gu Zhengc6ac4c02014-10-20 17:45:50 +0800664 f2fs_change_bit(block_off, sit_i->sit_bitmap);
Jaegeuk Kim39a53e02012-11-28 13:37:31 +0900665}
666
667static inline unsigned long long get_mtime(struct f2fs_sb_info *sbi)
668{
669 struct sit_info *sit_i = SIT_I(sbi);
670 return sit_i->elapsed_time + CURRENT_TIME_SEC.tv_sec -
671 sit_i->mounted_time;
672}
673
674static inline void set_summary(struct f2fs_summary *sum, nid_t nid,
675 unsigned int ofs_in_node, unsigned char version)
676{
677 sum->nid = cpu_to_le32(nid);
678 sum->ofs_in_node = cpu_to_le16(ofs_in_node);
679 sum->version = version;
680}
681
682static inline block_t start_sum_block(struct f2fs_sb_info *sbi)
683{
684 return __start_cp_addr(sbi) +
685 le32_to_cpu(F2FS_CKPT(sbi)->cp_pack_start_sum);
686}
687
688static inline block_t sum_blk_addr(struct f2fs_sb_info *sbi, int base, int type)
689{
690 return __start_cp_addr(sbi) +
691 le32_to_cpu(F2FS_CKPT(sbi)->cp_pack_total_block_count)
692 - (base + 1) + type;
693}
Jaegeuk Kim5ec4e492013-03-31 13:26:03 +0900694
695static inline bool sec_usage_check(struct f2fs_sb_info *sbi, unsigned int secno)
696{
697 if (IS_CURSEC(sbi, secno) || (sbi->cur_victim_sec == secno))
698 return true;
699 return false;
700}
Jaegeuk Kimac5d1562013-04-29 16:58:39 +0900701
Jaegeuk Kim87d6f892014-03-18 12:40:49 +0900702/*
703 * It is very important to gather dirty pages and write at once, so that we can
704 * submit a big bio without interfering other data writes.
705 * By default, 512 pages for directory data,
706 * 512 pages (2MB) * 3 for three types of nodes, and
707 * max_bio_blocks for meta are set.
708 */
709static inline int nr_pages_to_skip(struct f2fs_sb_info *sbi, int type)
710{
Tejun Heoa88a3412015-05-22 17:13:28 -0400711 if (sbi->sb->s_bdi->wb.dirty_exceeded)
Jaegeuk Kim510184c2014-11-06 17:23:08 -0800712 return 0;
713
Jaegeuk Kima1257022015-10-08 10:40:07 -0700714 if (type == DATA)
715 return sbi->blocks_per_seg;
716 else if (type == NODE)
Jaegeuk Kim2c237eb2016-06-16 16:44:11 -0700717 return 8 * sbi->blocks_per_seg;
Jaegeuk Kim87d6f892014-03-18 12:40:49 +0900718 else if (type == META)
Jaegeuk Kim664ba972016-10-18 11:07:45 -0700719 return 8 * BIO_MAX_PAGES;
Jaegeuk Kim87d6f892014-03-18 12:40:49 +0900720 else
721 return 0;
722}
Jaegeuk Kim50c8cdb2014-03-18 13:47:11 +0900723
724/*
725 * When writing pages, it'd better align nr_to_write for segment size.
726 */
727static inline long nr_pages_to_write(struct f2fs_sb_info *sbi, int type,
728 struct writeback_control *wbc)
729{
730 long nr_to_write, desired;
731
732 if (wbc->sync_mode != WB_SYNC_NONE)
733 return 0;
734
735 nr_to_write = wbc->nr_to_write;
Jaegeuk Kim664ba972016-10-18 11:07:45 -0700736 desired = BIO_MAX_PAGES;
Jaegeuk Kim28ea6162016-05-25 17:17:56 -0700737 if (type == NODE)
Jaegeuk Kim664ba972016-10-18 11:07:45 -0700738 desired <<= 1;
Jaegeuk Kim50c8cdb2014-03-18 13:47:11 +0900739
740 wbc->nr_to_write = desired;
741 return desired - nr_to_write;
742}