4 * Copyright (c) 2012 Samsung Electronics Co., Ltd.
5 * http://www.samsung.com/
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.
12 #include <linux/f2fs_fs.h>
13 #include <linux/bio.h>
14 #include <linux/blkdev.h>
15 #include <linux/prefetch.h>
16 #include <linux/vmalloc.h>
21 #include <trace/events/f2fs.h>
24 * This function balances dirty node and dentry pages.
25 * In addition, it controls garbage collection.
27 void f2fs_balance_fs(struct f2fs_sb_info
*sbi
)
30 * We should do GC or end up with checkpoint, if there are so many dirty
31 * dir/node pages without enough free segments.
33 if (has_not_enough_free_secs(sbi
, 0)) {
34 mutex_lock(&sbi
->gc_mutex
);
39 static void __locate_dirty_segment(struct f2fs_sb_info
*sbi
, unsigned int segno
,
40 enum dirty_type dirty_type
)
42 struct dirty_seglist_info
*dirty_i
= DIRTY_I(sbi
);
44 /* need not be added */
45 if (IS_CURSEG(sbi
, segno
))
48 if (!test_and_set_bit(segno
, dirty_i
->dirty_segmap
[dirty_type
]))
49 dirty_i
->nr_dirty
[dirty_type
]++;
51 if (dirty_type
== DIRTY
) {
52 struct seg_entry
*sentry
= get_seg_entry(sbi
, segno
);
53 enum dirty_type t
= DIRTY_HOT_DATA
;
55 dirty_type
= sentry
->type
;
57 if (!test_and_set_bit(segno
, dirty_i
->dirty_segmap
[dirty_type
]))
58 dirty_i
->nr_dirty
[dirty_type
]++;
60 /* Only one bitmap should be set */
61 for (; t
<= DIRTY_COLD_NODE
; t
++) {
64 if (test_and_clear_bit(segno
, dirty_i
->dirty_segmap
[t
]))
65 dirty_i
->nr_dirty
[t
]--;
70 static void __remove_dirty_segment(struct f2fs_sb_info
*sbi
, unsigned int segno
,
71 enum dirty_type dirty_type
)
73 struct dirty_seglist_info
*dirty_i
= DIRTY_I(sbi
);
75 if (test_and_clear_bit(segno
, dirty_i
->dirty_segmap
[dirty_type
]))
76 dirty_i
->nr_dirty
[dirty_type
]--;
78 if (dirty_type
== DIRTY
) {
79 enum dirty_type t
= DIRTY_HOT_DATA
;
81 /* clear all the bitmaps */
82 for (; t
<= DIRTY_COLD_NODE
; t
++)
83 if (test_and_clear_bit(segno
, dirty_i
->dirty_segmap
[t
]))
84 dirty_i
->nr_dirty
[t
]--;
86 if (get_valid_blocks(sbi
, segno
, sbi
->segs_per_sec
) == 0)
87 clear_bit(GET_SECNO(sbi
, segno
),
88 dirty_i
->victim_secmap
);
93 * Should not occur error such as -ENOMEM.
94 * Adding dirty entry into seglist is not critical operation.
95 * If a given segment is one of current working segments, it won't be added.
97 static void locate_dirty_segment(struct f2fs_sb_info
*sbi
, unsigned int segno
)
99 struct dirty_seglist_info
*dirty_i
= DIRTY_I(sbi
);
100 unsigned short valid_blocks
;
102 if (segno
== NULL_SEGNO
|| IS_CURSEG(sbi
, segno
))
105 mutex_lock(&dirty_i
->seglist_lock
);
107 valid_blocks
= get_valid_blocks(sbi
, segno
, 0);
109 if (valid_blocks
== 0) {
110 __locate_dirty_segment(sbi
, segno
, PRE
);
111 __remove_dirty_segment(sbi
, segno
, DIRTY
);
112 } else if (valid_blocks
< sbi
->blocks_per_seg
) {
113 __locate_dirty_segment(sbi
, segno
, DIRTY
);
115 /* Recovery routine with SSR needs this */
116 __remove_dirty_segment(sbi
, segno
, DIRTY
);
119 mutex_unlock(&dirty_i
->seglist_lock
);
124 * Should call clear_prefree_segments after checkpoint is done.
126 static void set_prefree_as_free_segments(struct f2fs_sb_info
*sbi
)
128 struct dirty_seglist_info
*dirty_i
= DIRTY_I(sbi
);
129 unsigned int segno
= -1;
130 unsigned int total_segs
= TOTAL_SEGS(sbi
);
132 mutex_lock(&dirty_i
->seglist_lock
);
134 segno
= find_next_bit(dirty_i
->dirty_segmap
[PRE
], total_segs
,
136 if (segno
>= total_segs
)
138 __set_test_and_free(sbi
, segno
);
140 mutex_unlock(&dirty_i
->seglist_lock
);
143 void clear_prefree_segments(struct f2fs_sb_info
*sbi
)
145 struct dirty_seglist_info
*dirty_i
= DIRTY_I(sbi
);
146 unsigned int segno
= -1;
147 unsigned int total_segs
= TOTAL_SEGS(sbi
);
149 mutex_lock(&dirty_i
->seglist_lock
);
151 segno
= find_next_bit(dirty_i
->dirty_segmap
[PRE
], total_segs
,
153 if (segno
>= total_segs
)
156 if (test_and_clear_bit(segno
, dirty_i
->dirty_segmap
[PRE
]))
157 dirty_i
->nr_dirty
[PRE
]--;
160 if (test_opt(sbi
, DISCARD
))
161 blkdev_issue_discard(sbi
->sb
->s_bdev
,
162 START_BLOCK(sbi
, segno
) <<
163 sbi
->log_sectors_per_block
,
164 1 << (sbi
->log_sectors_per_block
+
165 sbi
->log_blocks_per_seg
),
168 mutex_unlock(&dirty_i
->seglist_lock
);
171 static void __mark_sit_entry_dirty(struct f2fs_sb_info
*sbi
, unsigned int segno
)
173 struct sit_info
*sit_i
= SIT_I(sbi
);
174 if (!__test_and_set_bit(segno
, sit_i
->dirty_sentries_bitmap
))
175 sit_i
->dirty_sentries
++;
178 static void __set_sit_entry_type(struct f2fs_sb_info
*sbi
, int type
,
179 unsigned int segno
, int modified
)
181 struct seg_entry
*se
= get_seg_entry(sbi
, segno
);
184 __mark_sit_entry_dirty(sbi
, segno
);
187 static void update_sit_entry(struct f2fs_sb_info
*sbi
, block_t blkaddr
, int del
)
189 struct seg_entry
*se
;
190 unsigned int segno
, offset
;
191 long int new_vblocks
;
193 segno
= GET_SEGNO(sbi
, blkaddr
);
195 se
= get_seg_entry(sbi
, segno
);
196 new_vblocks
= se
->valid_blocks
+ del
;
197 offset
= GET_SEGOFF_FROM_SEG0(sbi
, blkaddr
) & (sbi
->blocks_per_seg
- 1);
199 BUG_ON((new_vblocks
>> (sizeof(unsigned short) << 3) ||
200 (new_vblocks
> sbi
->blocks_per_seg
)));
202 se
->valid_blocks
= new_vblocks
;
203 se
->mtime
= get_mtime(sbi
);
204 SIT_I(sbi
)->max_mtime
= se
->mtime
;
206 /* Update valid block bitmap */
208 if (f2fs_set_bit(offset
, se
->cur_valid_map
))
211 if (!f2fs_clear_bit(offset
, se
->cur_valid_map
))
214 if (!f2fs_test_bit(offset
, se
->ckpt_valid_map
))
215 se
->ckpt_valid_blocks
+= del
;
217 __mark_sit_entry_dirty(sbi
, segno
);
219 /* update total number of valid blocks to be written in ckpt area */
220 SIT_I(sbi
)->written_valid_blocks
+= del
;
222 if (sbi
->segs_per_sec
> 1)
223 get_sec_entry(sbi
, segno
)->valid_blocks
+= del
;
226 static void refresh_sit_entry(struct f2fs_sb_info
*sbi
,
227 block_t old_blkaddr
, block_t new_blkaddr
)
229 update_sit_entry(sbi
, new_blkaddr
, 1);
230 if (GET_SEGNO(sbi
, old_blkaddr
) != NULL_SEGNO
)
231 update_sit_entry(sbi
, old_blkaddr
, -1);
234 void invalidate_blocks(struct f2fs_sb_info
*sbi
, block_t addr
)
236 unsigned int segno
= GET_SEGNO(sbi
, addr
);
237 struct sit_info
*sit_i
= SIT_I(sbi
);
239 BUG_ON(addr
== NULL_ADDR
);
240 if (addr
== NEW_ADDR
)
243 /* add it into sit main buffer */
244 mutex_lock(&sit_i
->sentry_lock
);
246 update_sit_entry(sbi
, addr
, -1);
248 /* add it into dirty seglist */
249 locate_dirty_segment(sbi
, segno
);
251 mutex_unlock(&sit_i
->sentry_lock
);
255 * This function should be resided under the curseg_mutex lock
257 static void __add_sum_entry(struct f2fs_sb_info
*sbi
, int type
,
258 struct f2fs_summary
*sum
)
260 struct curseg_info
*curseg
= CURSEG_I(sbi
, type
);
261 void *addr
= curseg
->sum_blk
;
262 addr
+= curseg
->next_blkoff
* sizeof(struct f2fs_summary
);
263 memcpy(addr
, sum
, sizeof(struct f2fs_summary
));
268 * Calculate the number of current summary pages for writing
270 int npages_for_summary_flush(struct f2fs_sb_info
*sbi
)
272 int total_size_bytes
= 0;
273 int valid_sum_count
= 0;
276 for (i
= CURSEG_HOT_DATA
; i
<= CURSEG_COLD_DATA
; i
++) {
277 if (sbi
->ckpt
->alloc_type
[i
] == SSR
)
278 valid_sum_count
+= sbi
->blocks_per_seg
;
280 valid_sum_count
+= curseg_blkoff(sbi
, i
);
283 total_size_bytes
= valid_sum_count
* (SUMMARY_SIZE
+ 1)
284 + sizeof(struct nat_journal
) + 2
285 + sizeof(struct sit_journal
) + 2;
286 sum_space
= PAGE_CACHE_SIZE
- SUM_FOOTER_SIZE
;
287 if (total_size_bytes
< sum_space
)
289 else if (total_size_bytes
< 2 * sum_space
)
295 * Caller should put this summary page
297 struct page
*get_sum_page(struct f2fs_sb_info
*sbi
, unsigned int segno
)
299 return get_meta_page(sbi
, GET_SUM_BLOCK(sbi
, segno
));
302 static void write_sum_page(struct f2fs_sb_info
*sbi
,
303 struct f2fs_summary_block
*sum_blk
, block_t blk_addr
)
305 struct page
*page
= grab_meta_page(sbi
, blk_addr
);
306 void *kaddr
= page_address(page
);
307 memcpy(kaddr
, sum_blk
, PAGE_CACHE_SIZE
);
308 set_page_dirty(page
);
309 f2fs_put_page(page
, 1);
312 static int is_next_segment_free(struct f2fs_sb_info
*sbi
, int type
)
314 struct curseg_info
*curseg
= CURSEG_I(sbi
, type
);
315 unsigned int segno
= curseg
->segno
+ 1;
316 struct free_segmap_info
*free_i
= FREE_I(sbi
);
318 if (segno
< TOTAL_SEGS(sbi
) && segno
% sbi
->segs_per_sec
)
319 return !test_bit(segno
, free_i
->free_segmap
);
324 * Find a new segment from the free segments bitmap to right order
325 * This function should be returned with success, otherwise BUG
327 static void get_new_segment(struct f2fs_sb_info
*sbi
,
328 unsigned int *newseg
, bool new_sec
, int dir
)
330 struct free_segmap_info
*free_i
= FREE_I(sbi
);
331 unsigned int segno
, secno
, zoneno
;
332 unsigned int total_zones
= TOTAL_SECS(sbi
) / sbi
->secs_per_zone
;
333 unsigned int hint
= *newseg
/ sbi
->segs_per_sec
;
334 unsigned int old_zoneno
= GET_ZONENO_FROM_SEGNO(sbi
, *newseg
);
335 unsigned int left_start
= hint
;
340 write_lock(&free_i
->segmap_lock
);
342 if (!new_sec
&& ((*newseg
+ 1) % sbi
->segs_per_sec
)) {
343 segno
= find_next_zero_bit(free_i
->free_segmap
,
344 TOTAL_SEGS(sbi
), *newseg
+ 1);
345 if (segno
- *newseg
< sbi
->segs_per_sec
-
346 (*newseg
% sbi
->segs_per_sec
))
350 secno
= find_next_zero_bit(free_i
->free_secmap
, TOTAL_SECS(sbi
), hint
);
351 if (secno
>= TOTAL_SECS(sbi
)) {
352 if (dir
== ALLOC_RIGHT
) {
353 secno
= find_next_zero_bit(free_i
->free_secmap
,
355 BUG_ON(secno
>= TOTAL_SECS(sbi
));
358 left_start
= hint
- 1;
364 while (test_bit(left_start
, free_i
->free_secmap
)) {
365 if (left_start
> 0) {
369 left_start
= find_next_zero_bit(free_i
->free_secmap
,
371 BUG_ON(left_start
>= TOTAL_SECS(sbi
));
377 segno
= secno
* sbi
->segs_per_sec
;
378 zoneno
= secno
/ sbi
->secs_per_zone
;
380 /* give up on finding another zone */
383 if (sbi
->secs_per_zone
== 1)
385 if (zoneno
== old_zoneno
)
387 if (dir
== ALLOC_LEFT
) {
388 if (!go_left
&& zoneno
+ 1 >= total_zones
)
390 if (go_left
&& zoneno
== 0)
393 for (i
= 0; i
< NR_CURSEG_TYPE
; i
++)
394 if (CURSEG_I(sbi
, i
)->zone
== zoneno
)
397 if (i
< NR_CURSEG_TYPE
) {
398 /* zone is in user, try another */
400 hint
= zoneno
* sbi
->secs_per_zone
- 1;
401 else if (zoneno
+ 1 >= total_zones
)
404 hint
= (zoneno
+ 1) * sbi
->secs_per_zone
;
406 goto find_other_zone
;
409 /* set it as dirty segment in free segmap */
410 BUG_ON(test_bit(segno
, free_i
->free_segmap
));
411 __set_inuse(sbi
, segno
);
413 write_unlock(&free_i
->segmap_lock
);
416 static void reset_curseg(struct f2fs_sb_info
*sbi
, int type
, int modified
)
418 struct curseg_info
*curseg
= CURSEG_I(sbi
, type
);
419 struct summary_footer
*sum_footer
;
421 curseg
->segno
= curseg
->next_segno
;
422 curseg
->zone
= GET_ZONENO_FROM_SEGNO(sbi
, curseg
->segno
);
423 curseg
->next_blkoff
= 0;
424 curseg
->next_segno
= NULL_SEGNO
;
426 sum_footer
= &(curseg
->sum_blk
->footer
);
427 memset(sum_footer
, 0, sizeof(struct summary_footer
));
428 if (IS_DATASEG(type
))
429 SET_SUM_TYPE(sum_footer
, SUM_TYPE_DATA
);
430 if (IS_NODESEG(type
))
431 SET_SUM_TYPE(sum_footer
, SUM_TYPE_NODE
);
432 __set_sit_entry_type(sbi
, type
, curseg
->segno
, modified
);
436 * Allocate a current working segment.
437 * This function always allocates a free segment in LFS manner.
439 static void new_curseg(struct f2fs_sb_info
*sbi
, int type
, bool new_sec
)
441 struct curseg_info
*curseg
= CURSEG_I(sbi
, type
);
442 unsigned int segno
= curseg
->segno
;
443 int dir
= ALLOC_LEFT
;
445 write_sum_page(sbi
, curseg
->sum_blk
,
446 GET_SUM_BLOCK(sbi
, segno
));
447 if (type
== CURSEG_WARM_DATA
|| type
== CURSEG_COLD_DATA
)
450 if (test_opt(sbi
, NOHEAP
))
453 get_new_segment(sbi
, &segno
, new_sec
, dir
);
454 curseg
->next_segno
= segno
;
455 reset_curseg(sbi
, type
, 1);
456 curseg
->alloc_type
= LFS
;
459 static void __next_free_blkoff(struct f2fs_sb_info
*sbi
,
460 struct curseg_info
*seg
, block_t start
)
462 struct seg_entry
*se
= get_seg_entry(sbi
, seg
->segno
);
464 for (ofs
= start
; ofs
< sbi
->blocks_per_seg
; ofs
++) {
465 if (!f2fs_test_bit(ofs
, se
->ckpt_valid_map
)
466 && !f2fs_test_bit(ofs
, se
->cur_valid_map
))
469 seg
->next_blkoff
= ofs
;
473 * If a segment is written by LFS manner, next block offset is just obtained
474 * by increasing the current block offset. However, if a segment is written by
475 * SSR manner, next block offset obtained by calling __next_free_blkoff
477 static void __refresh_next_blkoff(struct f2fs_sb_info
*sbi
,
478 struct curseg_info
*seg
)
480 if (seg
->alloc_type
== SSR
)
481 __next_free_blkoff(sbi
, seg
, seg
->next_blkoff
+ 1);
487 * This function always allocates a used segment (from dirty seglist) by SSR
488 * manner, so it should recover the existing segment information of valid blocks
490 static void change_curseg(struct f2fs_sb_info
*sbi
, int type
, bool reuse
)
492 struct dirty_seglist_info
*dirty_i
= DIRTY_I(sbi
);
493 struct curseg_info
*curseg
= CURSEG_I(sbi
, type
);
494 unsigned int new_segno
= curseg
->next_segno
;
495 struct f2fs_summary_block
*sum_node
;
496 struct page
*sum_page
;
498 write_sum_page(sbi
, curseg
->sum_blk
,
499 GET_SUM_BLOCK(sbi
, curseg
->segno
));
500 __set_test_and_inuse(sbi
, new_segno
);
502 mutex_lock(&dirty_i
->seglist_lock
);
503 __remove_dirty_segment(sbi
, new_segno
, PRE
);
504 __remove_dirty_segment(sbi
, new_segno
, DIRTY
);
505 mutex_unlock(&dirty_i
->seglist_lock
);
507 reset_curseg(sbi
, type
, 1);
508 curseg
->alloc_type
= SSR
;
509 __next_free_blkoff(sbi
, curseg
, 0);
512 sum_page
= get_sum_page(sbi
, new_segno
);
513 sum_node
= (struct f2fs_summary_block
*)page_address(sum_page
);
514 memcpy(curseg
->sum_blk
, sum_node
, SUM_ENTRY_SIZE
);
515 f2fs_put_page(sum_page
, 1);
519 static int get_ssr_segment(struct f2fs_sb_info
*sbi
, int type
)
521 struct curseg_info
*curseg
= CURSEG_I(sbi
, type
);
522 const struct victim_selection
*v_ops
= DIRTY_I(sbi
)->v_ops
;
524 if (IS_NODESEG(type
) || !has_not_enough_free_secs(sbi
, 0))
525 return v_ops
->get_victim(sbi
,
526 &(curseg
)->next_segno
, BG_GC
, type
, SSR
);
528 /* For data segments, let's do SSR more intensively */
529 for (; type
>= CURSEG_HOT_DATA
; type
--)
530 if (v_ops
->get_victim(sbi
, &(curseg
)->next_segno
,
537 * flush out current segment and replace it with new segment
538 * This function should be returned with success, otherwise BUG
540 static void allocate_segment_by_default(struct f2fs_sb_info
*sbi
,
541 int type
, bool force
)
543 struct curseg_info
*curseg
= CURSEG_I(sbi
, type
);
546 new_curseg(sbi
, type
, true);
550 if (type
== CURSEG_WARM_NODE
)
551 new_curseg(sbi
, type
, false);
552 else if (curseg
->alloc_type
== LFS
&& is_next_segment_free(sbi
, type
))
553 new_curseg(sbi
, type
, false);
554 else if (need_SSR(sbi
) && get_ssr_segment(sbi
, type
))
555 change_curseg(sbi
, type
, true);
557 new_curseg(sbi
, type
, false);
559 #ifdef CONFIG_F2FS_STAT_FS
560 sbi
->segment_count
[curseg
->alloc_type
]++;
565 void allocate_new_segments(struct f2fs_sb_info
*sbi
)
567 struct curseg_info
*curseg
;
568 unsigned int old_curseg
;
571 for (i
= CURSEG_HOT_DATA
; i
<= CURSEG_COLD_DATA
; i
++) {
572 curseg
= CURSEG_I(sbi
, i
);
573 old_curseg
= curseg
->segno
;
574 SIT_I(sbi
)->s_ops
->allocate_segment(sbi
, i
, true);
575 locate_dirty_segment(sbi
, old_curseg
);
579 static const struct segment_allocation default_salloc_ops
= {
580 .allocate_segment
= allocate_segment_by_default
,
583 static void f2fs_end_io_write(struct bio
*bio
, int err
)
585 const int uptodate
= test_bit(BIO_UPTODATE
, &bio
->bi_flags
);
586 struct bio_vec
*bvec
= bio
->bi_io_vec
+ bio
->bi_vcnt
- 1;
587 struct bio_private
*p
= bio
->bi_private
;
590 struct page
*page
= bvec
->bv_page
;
592 if (--bvec
>= bio
->bi_io_vec
)
593 prefetchw(&bvec
->bv_page
->flags
);
597 set_bit(AS_EIO
, &page
->mapping
->flags
);
598 set_ckpt_flags(p
->sbi
->ckpt
, CP_ERROR_FLAG
);
599 p
->sbi
->sb
->s_flags
|= MS_RDONLY
;
601 end_page_writeback(page
);
602 dec_page_count(p
->sbi
, F2FS_WRITEBACK
);
603 } while (bvec
>= bio
->bi_io_vec
);
611 struct bio
*f2fs_bio_alloc(struct block_device
*bdev
, int npages
)
614 struct bio_private
*priv
;
616 priv
= kmalloc(sizeof(struct bio_private
), GFP_NOFS
);
622 /* No failure on bio allocation */
623 bio
= bio_alloc(GFP_NOIO
, npages
);
625 bio
->bi_private
= priv
;
629 static void do_submit_bio(struct f2fs_sb_info
*sbi
,
630 enum page_type type
, bool sync
)
632 int rw
= sync
? WRITE_SYNC
: WRITE
;
633 enum page_type btype
= type
> META
? META
: type
;
635 if (type
>= META_FLUSH
)
636 rw
= WRITE_FLUSH_FUA
;
641 if (sbi
->bio
[btype
]) {
642 struct bio_private
*p
= sbi
->bio
[btype
]->bi_private
;
644 sbi
->bio
[btype
]->bi_end_io
= f2fs_end_io_write
;
646 trace_f2fs_do_submit_bio(sbi
->sb
, btype
, sync
, sbi
->bio
[btype
]);
648 if (type
== META_FLUSH
) {
649 DECLARE_COMPLETION_ONSTACK(wait
);
652 submit_bio(rw
, sbi
->bio
[btype
]);
653 wait_for_completion(&wait
);
656 submit_bio(rw
, sbi
->bio
[btype
]);
658 sbi
->bio
[btype
] = NULL
;
662 void f2fs_submit_bio(struct f2fs_sb_info
*sbi
, enum page_type type
, bool sync
)
664 down_write(&sbi
->bio_sem
);
665 do_submit_bio(sbi
, type
, sync
);
666 up_write(&sbi
->bio_sem
);
669 static void submit_write_page(struct f2fs_sb_info
*sbi
, struct page
*page
,
670 block_t blk_addr
, enum page_type type
)
672 struct block_device
*bdev
= sbi
->sb
->s_bdev
;
674 verify_block_addr(sbi
, blk_addr
);
676 down_write(&sbi
->bio_sem
);
678 inc_page_count(sbi
, F2FS_WRITEBACK
);
680 if (sbi
->bio
[type
] && sbi
->last_block_in_bio
[type
] != blk_addr
- 1)
681 do_submit_bio(sbi
, type
, false);
683 if (sbi
->bio
[type
] == NULL
) {
684 sbi
->bio
[type
] = f2fs_bio_alloc(bdev
, max_hw_blocks(sbi
));
685 sbi
->bio
[type
]->bi_sector
= SECTOR_FROM_BLOCK(sbi
, blk_addr
);
687 * The end_io will be assigned at the sumbission phase.
688 * Until then, let bio_add_page() merge consecutive IOs as much
693 if (bio_add_page(sbi
->bio
[type
], page
, PAGE_CACHE_SIZE
, 0) <
695 do_submit_bio(sbi
, type
, false);
699 sbi
->last_block_in_bio
[type
] = blk_addr
;
701 up_write(&sbi
->bio_sem
);
702 trace_f2fs_submit_write_page(page
, blk_addr
, type
);
705 static bool __has_curseg_space(struct f2fs_sb_info
*sbi
, int type
)
707 struct curseg_info
*curseg
= CURSEG_I(sbi
, type
);
708 if (curseg
->next_blkoff
< sbi
->blocks_per_seg
)
713 static int __get_segment_type_2(struct page
*page
, enum page_type p_type
)
716 return CURSEG_HOT_DATA
;
718 return CURSEG_HOT_NODE
;
721 static int __get_segment_type_4(struct page
*page
, enum page_type p_type
)
723 if (p_type
== DATA
) {
724 struct inode
*inode
= page
->mapping
->host
;
726 if (S_ISDIR(inode
->i_mode
))
727 return CURSEG_HOT_DATA
;
729 return CURSEG_COLD_DATA
;
731 if (IS_DNODE(page
) && !is_cold_node(page
))
732 return CURSEG_HOT_NODE
;
734 return CURSEG_COLD_NODE
;
738 static int __get_segment_type_6(struct page
*page
, enum page_type p_type
)
740 if (p_type
== DATA
) {
741 struct inode
*inode
= page
->mapping
->host
;
743 if (S_ISDIR(inode
->i_mode
))
744 return CURSEG_HOT_DATA
;
745 else if (is_cold_data(page
) || file_is_cold(inode
))
746 return CURSEG_COLD_DATA
;
748 return CURSEG_WARM_DATA
;
751 return is_cold_node(page
) ? CURSEG_WARM_NODE
:
754 return CURSEG_COLD_NODE
;
758 static int __get_segment_type(struct page
*page
, enum page_type p_type
)
760 struct f2fs_sb_info
*sbi
= F2FS_SB(page
->mapping
->host
->i_sb
);
761 switch (sbi
->active_logs
) {
763 return __get_segment_type_2(page
, p_type
);
765 return __get_segment_type_4(page
, p_type
);
767 /* NR_CURSEG_TYPE(6) logs by default */
768 BUG_ON(sbi
->active_logs
!= NR_CURSEG_TYPE
);
769 return __get_segment_type_6(page
, p_type
);
772 static void do_write_page(struct f2fs_sb_info
*sbi
, struct page
*page
,
773 block_t old_blkaddr
, block_t
*new_blkaddr
,
774 struct f2fs_summary
*sum
, enum page_type p_type
)
776 struct sit_info
*sit_i
= SIT_I(sbi
);
777 struct curseg_info
*curseg
;
778 unsigned int old_cursegno
;
781 type
= __get_segment_type(page
, p_type
);
782 curseg
= CURSEG_I(sbi
, type
);
784 mutex_lock(&curseg
->curseg_mutex
);
786 *new_blkaddr
= NEXT_FREE_BLKADDR(sbi
, curseg
);
787 old_cursegno
= curseg
->segno
;
790 * __add_sum_entry should be resided under the curseg_mutex
791 * because, this function updates a summary entry in the
792 * current summary block.
794 __add_sum_entry(sbi
, type
, sum
);
796 mutex_lock(&sit_i
->sentry_lock
);
797 __refresh_next_blkoff(sbi
, curseg
);
798 #ifdef CONFIG_F2FS_STAT_FS
799 sbi
->block_count
[curseg
->alloc_type
]++;
803 * SIT information should be updated before segment allocation,
804 * since SSR needs latest valid block information.
806 refresh_sit_entry(sbi
, old_blkaddr
, *new_blkaddr
);
808 if (!__has_curseg_space(sbi
, type
))
809 sit_i
->s_ops
->allocate_segment(sbi
, type
, false);
811 locate_dirty_segment(sbi
, old_cursegno
);
812 locate_dirty_segment(sbi
, GET_SEGNO(sbi
, old_blkaddr
));
813 mutex_unlock(&sit_i
->sentry_lock
);
816 fill_node_footer_blkaddr(page
, NEXT_FREE_BLKADDR(sbi
, curseg
));
818 /* writeout dirty page into bdev */
819 submit_write_page(sbi
, page
, *new_blkaddr
, p_type
);
821 mutex_unlock(&curseg
->curseg_mutex
);
824 void write_meta_page(struct f2fs_sb_info
*sbi
, struct page
*page
)
826 set_page_writeback(page
);
827 submit_write_page(sbi
, page
, page
->index
, META
);
830 void write_node_page(struct f2fs_sb_info
*sbi
, struct page
*page
,
831 unsigned int nid
, block_t old_blkaddr
, block_t
*new_blkaddr
)
833 struct f2fs_summary sum
;
834 set_summary(&sum
, nid
, 0, 0);
835 do_write_page(sbi
, page
, old_blkaddr
, new_blkaddr
, &sum
, NODE
);
838 void write_data_page(struct inode
*inode
, struct page
*page
,
839 struct dnode_of_data
*dn
, block_t old_blkaddr
,
840 block_t
*new_blkaddr
)
842 struct f2fs_sb_info
*sbi
= F2FS_SB(inode
->i_sb
);
843 struct f2fs_summary sum
;
846 BUG_ON(old_blkaddr
== NULL_ADDR
);
847 get_node_info(sbi
, dn
->nid
, &ni
);
848 set_summary(&sum
, dn
->nid
, dn
->ofs_in_node
, ni
.version
);
850 do_write_page(sbi
, page
, old_blkaddr
,
851 new_blkaddr
, &sum
, DATA
);
854 void rewrite_data_page(struct f2fs_sb_info
*sbi
, struct page
*page
,
855 block_t old_blk_addr
)
857 submit_write_page(sbi
, page
, old_blk_addr
, DATA
);
860 void recover_data_page(struct f2fs_sb_info
*sbi
,
861 struct page
*page
, struct f2fs_summary
*sum
,
862 block_t old_blkaddr
, block_t new_blkaddr
)
864 struct sit_info
*sit_i
= SIT_I(sbi
);
865 struct curseg_info
*curseg
;
866 unsigned int segno
, old_cursegno
;
867 struct seg_entry
*se
;
870 segno
= GET_SEGNO(sbi
, new_blkaddr
);
871 se
= get_seg_entry(sbi
, segno
);
874 if (se
->valid_blocks
== 0 && !IS_CURSEG(sbi
, segno
)) {
875 if (old_blkaddr
== NULL_ADDR
)
876 type
= CURSEG_COLD_DATA
;
878 type
= CURSEG_WARM_DATA
;
880 curseg
= CURSEG_I(sbi
, type
);
882 mutex_lock(&curseg
->curseg_mutex
);
883 mutex_lock(&sit_i
->sentry_lock
);
885 old_cursegno
= curseg
->segno
;
887 /* change the current segment */
888 if (segno
!= curseg
->segno
) {
889 curseg
->next_segno
= segno
;
890 change_curseg(sbi
, type
, true);
893 curseg
->next_blkoff
= GET_SEGOFF_FROM_SEG0(sbi
, new_blkaddr
) &
894 (sbi
->blocks_per_seg
- 1);
895 __add_sum_entry(sbi
, type
, sum
);
897 refresh_sit_entry(sbi
, old_blkaddr
, new_blkaddr
);
899 locate_dirty_segment(sbi
, old_cursegno
);
900 locate_dirty_segment(sbi
, GET_SEGNO(sbi
, old_blkaddr
));
902 mutex_unlock(&sit_i
->sentry_lock
);
903 mutex_unlock(&curseg
->curseg_mutex
);
906 void rewrite_node_page(struct f2fs_sb_info
*sbi
,
907 struct page
*page
, struct f2fs_summary
*sum
,
908 block_t old_blkaddr
, block_t new_blkaddr
)
910 struct sit_info
*sit_i
= SIT_I(sbi
);
911 int type
= CURSEG_WARM_NODE
;
912 struct curseg_info
*curseg
;
913 unsigned int segno
, old_cursegno
;
914 block_t next_blkaddr
= next_blkaddr_of_node(page
);
915 unsigned int next_segno
= GET_SEGNO(sbi
, next_blkaddr
);
917 curseg
= CURSEG_I(sbi
, type
);
919 mutex_lock(&curseg
->curseg_mutex
);
920 mutex_lock(&sit_i
->sentry_lock
);
922 segno
= GET_SEGNO(sbi
, new_blkaddr
);
923 old_cursegno
= curseg
->segno
;
925 /* change the current segment */
926 if (segno
!= curseg
->segno
) {
927 curseg
->next_segno
= segno
;
928 change_curseg(sbi
, type
, true);
930 curseg
->next_blkoff
= GET_SEGOFF_FROM_SEG0(sbi
, new_blkaddr
) &
931 (sbi
->blocks_per_seg
- 1);
932 __add_sum_entry(sbi
, type
, sum
);
934 /* change the current log to the next block addr in advance */
935 if (next_segno
!= segno
) {
936 curseg
->next_segno
= next_segno
;
937 change_curseg(sbi
, type
, true);
939 curseg
->next_blkoff
= GET_SEGOFF_FROM_SEG0(sbi
, next_blkaddr
) &
940 (sbi
->blocks_per_seg
- 1);
942 /* rewrite node page */
943 set_page_writeback(page
);
944 submit_write_page(sbi
, page
, new_blkaddr
, NODE
);
945 f2fs_submit_bio(sbi
, NODE
, true);
946 refresh_sit_entry(sbi
, old_blkaddr
, new_blkaddr
);
948 locate_dirty_segment(sbi
, old_cursegno
);
949 locate_dirty_segment(sbi
, GET_SEGNO(sbi
, old_blkaddr
));
951 mutex_unlock(&sit_i
->sentry_lock
);
952 mutex_unlock(&curseg
->curseg_mutex
);
955 static int read_compacted_summaries(struct f2fs_sb_info
*sbi
)
957 struct f2fs_checkpoint
*ckpt
= F2FS_CKPT(sbi
);
958 struct curseg_info
*seg_i
;
959 unsigned char *kaddr
;
964 start
= start_sum_block(sbi
);
966 page
= get_meta_page(sbi
, start
++);
967 kaddr
= (unsigned char *)page_address(page
);
969 /* Step 1: restore nat cache */
970 seg_i
= CURSEG_I(sbi
, CURSEG_HOT_DATA
);
971 memcpy(&seg_i
->sum_blk
->n_nats
, kaddr
, SUM_JOURNAL_SIZE
);
973 /* Step 2: restore sit cache */
974 seg_i
= CURSEG_I(sbi
, CURSEG_COLD_DATA
);
975 memcpy(&seg_i
->sum_blk
->n_sits
, kaddr
+ SUM_JOURNAL_SIZE
,
977 offset
= 2 * SUM_JOURNAL_SIZE
;
979 /* Step 3: restore summary entries */
980 for (i
= CURSEG_HOT_DATA
; i
<= CURSEG_COLD_DATA
; i
++) {
981 unsigned short blk_off
;
984 seg_i
= CURSEG_I(sbi
, i
);
985 segno
= le32_to_cpu(ckpt
->cur_data_segno
[i
]);
986 blk_off
= le16_to_cpu(ckpt
->cur_data_blkoff
[i
]);
987 seg_i
->next_segno
= segno
;
988 reset_curseg(sbi
, i
, 0);
989 seg_i
->alloc_type
= ckpt
->alloc_type
[i
];
990 seg_i
->next_blkoff
= blk_off
;
992 if (seg_i
->alloc_type
== SSR
)
993 blk_off
= sbi
->blocks_per_seg
;
995 for (j
= 0; j
< blk_off
; j
++) {
996 struct f2fs_summary
*s
;
997 s
= (struct f2fs_summary
*)(kaddr
+ offset
);
998 seg_i
->sum_blk
->entries
[j
] = *s
;
999 offset
+= SUMMARY_SIZE
;
1000 if (offset
+ SUMMARY_SIZE
<= PAGE_CACHE_SIZE
-
1004 f2fs_put_page(page
, 1);
1007 page
= get_meta_page(sbi
, start
++);
1008 kaddr
= (unsigned char *)page_address(page
);
1012 f2fs_put_page(page
, 1);
1016 static int read_normal_summaries(struct f2fs_sb_info
*sbi
, int type
)
1018 struct f2fs_checkpoint
*ckpt
= F2FS_CKPT(sbi
);
1019 struct f2fs_summary_block
*sum
;
1020 struct curseg_info
*curseg
;
1022 unsigned short blk_off
;
1023 unsigned int segno
= 0;
1024 block_t blk_addr
= 0;
1026 /* get segment number and block addr */
1027 if (IS_DATASEG(type
)) {
1028 segno
= le32_to_cpu(ckpt
->cur_data_segno
[type
]);
1029 blk_off
= le16_to_cpu(ckpt
->cur_data_blkoff
[type
-
1031 if (is_set_ckpt_flags(ckpt
, CP_UMOUNT_FLAG
))
1032 blk_addr
= sum_blk_addr(sbi
, NR_CURSEG_TYPE
, type
);
1034 blk_addr
= sum_blk_addr(sbi
, NR_CURSEG_DATA_TYPE
, type
);
1036 segno
= le32_to_cpu(ckpt
->cur_node_segno
[type
-
1038 blk_off
= le16_to_cpu(ckpt
->cur_node_blkoff
[type
-
1040 if (is_set_ckpt_flags(ckpt
, CP_UMOUNT_FLAG
))
1041 blk_addr
= sum_blk_addr(sbi
, NR_CURSEG_NODE_TYPE
,
1042 type
- CURSEG_HOT_NODE
);
1044 blk_addr
= GET_SUM_BLOCK(sbi
, segno
);
1047 new = get_meta_page(sbi
, blk_addr
);
1048 sum
= (struct f2fs_summary_block
*)page_address(new);
1050 if (IS_NODESEG(type
)) {
1051 if (is_set_ckpt_flags(ckpt
, CP_UMOUNT_FLAG
)) {
1052 struct f2fs_summary
*ns
= &sum
->entries
[0];
1054 for (i
= 0; i
< sbi
->blocks_per_seg
; i
++, ns
++) {
1056 ns
->ofs_in_node
= 0;
1059 if (restore_node_summary(sbi
, segno
, sum
)) {
1060 f2fs_put_page(new, 1);
1066 /* set uncompleted segment to curseg */
1067 curseg
= CURSEG_I(sbi
, type
);
1068 mutex_lock(&curseg
->curseg_mutex
);
1069 memcpy(curseg
->sum_blk
, sum
, PAGE_CACHE_SIZE
);
1070 curseg
->next_segno
= segno
;
1071 reset_curseg(sbi
, type
, 0);
1072 curseg
->alloc_type
= ckpt
->alloc_type
[type
];
1073 curseg
->next_blkoff
= blk_off
;
1074 mutex_unlock(&curseg
->curseg_mutex
);
1075 f2fs_put_page(new, 1);
1079 static int restore_curseg_summaries(struct f2fs_sb_info
*sbi
)
1081 int type
= CURSEG_HOT_DATA
;
1083 if (is_set_ckpt_flags(F2FS_CKPT(sbi
), CP_COMPACT_SUM_FLAG
)) {
1084 /* restore for compacted data summary */
1085 if (read_compacted_summaries(sbi
))
1087 type
= CURSEG_HOT_NODE
;
1090 for (; type
<= CURSEG_COLD_NODE
; type
++)
1091 if (read_normal_summaries(sbi
, type
))
1096 static void write_compacted_summaries(struct f2fs_sb_info
*sbi
, block_t blkaddr
)
1099 unsigned char *kaddr
;
1100 struct f2fs_summary
*summary
;
1101 struct curseg_info
*seg_i
;
1102 int written_size
= 0;
1105 page
= grab_meta_page(sbi
, blkaddr
++);
1106 kaddr
= (unsigned char *)page_address(page
);
1108 /* Step 1: write nat cache */
1109 seg_i
= CURSEG_I(sbi
, CURSEG_HOT_DATA
);
1110 memcpy(kaddr
, &seg_i
->sum_blk
->n_nats
, SUM_JOURNAL_SIZE
);
1111 written_size
+= SUM_JOURNAL_SIZE
;
1113 /* Step 2: write sit cache */
1114 seg_i
= CURSEG_I(sbi
, CURSEG_COLD_DATA
);
1115 memcpy(kaddr
+ written_size
, &seg_i
->sum_blk
->n_sits
,
1117 written_size
+= SUM_JOURNAL_SIZE
;
1119 set_page_dirty(page
);
1121 /* Step 3: write summary entries */
1122 for (i
= CURSEG_HOT_DATA
; i
<= CURSEG_COLD_DATA
; i
++) {
1123 unsigned short blkoff
;
1124 seg_i
= CURSEG_I(sbi
, i
);
1125 if (sbi
->ckpt
->alloc_type
[i
] == SSR
)
1126 blkoff
= sbi
->blocks_per_seg
;
1128 blkoff
= curseg_blkoff(sbi
, i
);
1130 for (j
= 0; j
< blkoff
; j
++) {
1132 page
= grab_meta_page(sbi
, blkaddr
++);
1133 kaddr
= (unsigned char *)page_address(page
);
1136 summary
= (struct f2fs_summary
*)(kaddr
+ written_size
);
1137 *summary
= seg_i
->sum_blk
->entries
[j
];
1138 written_size
+= SUMMARY_SIZE
;
1139 set_page_dirty(page
);
1141 if (written_size
+ SUMMARY_SIZE
<= PAGE_CACHE_SIZE
-
1145 f2fs_put_page(page
, 1);
1150 f2fs_put_page(page
, 1);
1153 static void write_normal_summaries(struct f2fs_sb_info
*sbi
,
1154 block_t blkaddr
, int type
)
1157 if (IS_DATASEG(type
))
1158 end
= type
+ NR_CURSEG_DATA_TYPE
;
1160 end
= type
+ NR_CURSEG_NODE_TYPE
;
1162 for (i
= type
; i
< end
; i
++) {
1163 struct curseg_info
*sum
= CURSEG_I(sbi
, i
);
1164 mutex_lock(&sum
->curseg_mutex
);
1165 write_sum_page(sbi
, sum
->sum_blk
, blkaddr
+ (i
- type
));
1166 mutex_unlock(&sum
->curseg_mutex
);
1170 void write_data_summaries(struct f2fs_sb_info
*sbi
, block_t start_blk
)
1172 if (is_set_ckpt_flags(F2FS_CKPT(sbi
), CP_COMPACT_SUM_FLAG
))
1173 write_compacted_summaries(sbi
, start_blk
);
1175 write_normal_summaries(sbi
, start_blk
, CURSEG_HOT_DATA
);
1178 void write_node_summaries(struct f2fs_sb_info
*sbi
, block_t start_blk
)
1180 if (is_set_ckpt_flags(F2FS_CKPT(sbi
), CP_UMOUNT_FLAG
))
1181 write_normal_summaries(sbi
, start_blk
, CURSEG_HOT_NODE
);
1185 int lookup_journal_in_cursum(struct f2fs_summary_block
*sum
, int type
,
1186 unsigned int val
, int alloc
)
1190 if (type
== NAT_JOURNAL
) {
1191 for (i
= 0; i
< nats_in_cursum(sum
); i
++) {
1192 if (le32_to_cpu(nid_in_journal(sum
, i
)) == val
)
1195 if (alloc
&& nats_in_cursum(sum
) < NAT_JOURNAL_ENTRIES
)
1196 return update_nats_in_cursum(sum
, 1);
1197 } else if (type
== SIT_JOURNAL
) {
1198 for (i
= 0; i
< sits_in_cursum(sum
); i
++)
1199 if (le32_to_cpu(segno_in_journal(sum
, i
)) == val
)
1201 if (alloc
&& sits_in_cursum(sum
) < SIT_JOURNAL_ENTRIES
)
1202 return update_sits_in_cursum(sum
, 1);
1207 static struct page
*get_current_sit_page(struct f2fs_sb_info
*sbi
,
1210 struct sit_info
*sit_i
= SIT_I(sbi
);
1211 unsigned int offset
= SIT_BLOCK_OFFSET(sit_i
, segno
);
1212 block_t blk_addr
= sit_i
->sit_base_addr
+ offset
;
1214 check_seg_range(sbi
, segno
);
1216 /* calculate sit block address */
1217 if (f2fs_test_bit(offset
, sit_i
->sit_bitmap
))
1218 blk_addr
+= sit_i
->sit_blocks
;
1220 return get_meta_page(sbi
, blk_addr
);
1223 static struct page
*get_next_sit_page(struct f2fs_sb_info
*sbi
,
1226 struct sit_info
*sit_i
= SIT_I(sbi
);
1227 struct page
*src_page
, *dst_page
;
1228 pgoff_t src_off
, dst_off
;
1229 void *src_addr
, *dst_addr
;
1231 src_off
= current_sit_addr(sbi
, start
);
1232 dst_off
= next_sit_addr(sbi
, src_off
);
1234 /* get current sit block page without lock */
1235 src_page
= get_meta_page(sbi
, src_off
);
1236 dst_page
= grab_meta_page(sbi
, dst_off
);
1237 BUG_ON(PageDirty(src_page
));
1239 src_addr
= page_address(src_page
);
1240 dst_addr
= page_address(dst_page
);
1241 memcpy(dst_addr
, src_addr
, PAGE_CACHE_SIZE
);
1243 set_page_dirty(dst_page
);
1244 f2fs_put_page(src_page
, 1);
1246 set_to_next_sit(sit_i
, start
);
1251 static bool flush_sits_in_journal(struct f2fs_sb_info
*sbi
)
1253 struct curseg_info
*curseg
= CURSEG_I(sbi
, CURSEG_COLD_DATA
);
1254 struct f2fs_summary_block
*sum
= curseg
->sum_blk
;
1258 * If the journal area in the current summary is full of sit entries,
1259 * all the sit entries will be flushed. Otherwise the sit entries
1260 * are not able to replace with newly hot sit entries.
1262 if (sits_in_cursum(sum
) >= SIT_JOURNAL_ENTRIES
) {
1263 for (i
= sits_in_cursum(sum
) - 1; i
>= 0; i
--) {
1265 segno
= le32_to_cpu(segno_in_journal(sum
, i
));
1266 __mark_sit_entry_dirty(sbi
, segno
);
1268 update_sits_in_cursum(sum
, -sits_in_cursum(sum
));
1275 * CP calls this function, which flushes SIT entries including sit_journal,
1276 * and moves prefree segs to free segs.
1278 void flush_sit_entries(struct f2fs_sb_info
*sbi
)
1280 struct sit_info
*sit_i
= SIT_I(sbi
);
1281 unsigned long *bitmap
= sit_i
->dirty_sentries_bitmap
;
1282 struct curseg_info
*curseg
= CURSEG_I(sbi
, CURSEG_COLD_DATA
);
1283 struct f2fs_summary_block
*sum
= curseg
->sum_blk
;
1284 unsigned long nsegs
= TOTAL_SEGS(sbi
);
1285 struct page
*page
= NULL
;
1286 struct f2fs_sit_block
*raw_sit
= NULL
;
1287 unsigned int start
= 0, end
= 0;
1288 unsigned int segno
= -1;
1291 mutex_lock(&curseg
->curseg_mutex
);
1292 mutex_lock(&sit_i
->sentry_lock
);
1295 * "flushed" indicates whether sit entries in journal are flushed
1296 * to the SIT area or not.
1298 flushed
= flush_sits_in_journal(sbi
);
1300 while ((segno
= find_next_bit(bitmap
, nsegs
, segno
+ 1)) < nsegs
) {
1301 struct seg_entry
*se
= get_seg_entry(sbi
, segno
);
1302 int sit_offset
, offset
;
1304 sit_offset
= SIT_ENTRY_OFFSET(sit_i
, segno
);
1309 offset
= lookup_journal_in_cursum(sum
, SIT_JOURNAL
, segno
, 1);
1311 segno_in_journal(sum
, offset
) = cpu_to_le32(segno
);
1312 seg_info_to_raw_sit(se
, &sit_in_journal(sum
, offset
));
1316 if (!page
|| (start
> segno
) || (segno
> end
)) {
1318 f2fs_put_page(page
, 1);
1322 start
= START_SEGNO(sit_i
, segno
);
1323 end
= start
+ SIT_ENTRY_PER_BLOCK
- 1;
1325 /* read sit block that will be updated */
1326 page
= get_next_sit_page(sbi
, start
);
1327 raw_sit
= page_address(page
);
1330 /* udpate entry in SIT block */
1331 seg_info_to_raw_sit(se
, &raw_sit
->entries
[sit_offset
]);
1333 __clear_bit(segno
, bitmap
);
1334 sit_i
->dirty_sentries
--;
1336 mutex_unlock(&sit_i
->sentry_lock
);
1337 mutex_unlock(&curseg
->curseg_mutex
);
1339 /* writeout last modified SIT block */
1340 f2fs_put_page(page
, 1);
1342 set_prefree_as_free_segments(sbi
);
1345 static int build_sit_info(struct f2fs_sb_info
*sbi
)
1347 struct f2fs_super_block
*raw_super
= F2FS_RAW_SUPER(sbi
);
1348 struct f2fs_checkpoint
*ckpt
= F2FS_CKPT(sbi
);
1349 struct sit_info
*sit_i
;
1350 unsigned int sit_segs
, start
;
1351 char *src_bitmap
, *dst_bitmap
;
1352 unsigned int bitmap_size
;
1354 /* allocate memory for SIT information */
1355 sit_i
= kzalloc(sizeof(struct sit_info
), GFP_KERNEL
);
1359 SM_I(sbi
)->sit_info
= sit_i
;
1361 sit_i
->sentries
= vzalloc(TOTAL_SEGS(sbi
) * sizeof(struct seg_entry
));
1362 if (!sit_i
->sentries
)
1365 bitmap_size
= f2fs_bitmap_size(TOTAL_SEGS(sbi
));
1366 sit_i
->dirty_sentries_bitmap
= kzalloc(bitmap_size
, GFP_KERNEL
);
1367 if (!sit_i
->dirty_sentries_bitmap
)
1370 for (start
= 0; start
< TOTAL_SEGS(sbi
); start
++) {
1371 sit_i
->sentries
[start
].cur_valid_map
1372 = kzalloc(SIT_VBLOCK_MAP_SIZE
, GFP_KERNEL
);
1373 sit_i
->sentries
[start
].ckpt_valid_map
1374 = kzalloc(SIT_VBLOCK_MAP_SIZE
, GFP_KERNEL
);
1375 if (!sit_i
->sentries
[start
].cur_valid_map
1376 || !sit_i
->sentries
[start
].ckpt_valid_map
)
1380 if (sbi
->segs_per_sec
> 1) {
1381 sit_i
->sec_entries
= vzalloc(TOTAL_SECS(sbi
) *
1382 sizeof(struct sec_entry
));
1383 if (!sit_i
->sec_entries
)
1387 /* get information related with SIT */
1388 sit_segs
= le32_to_cpu(raw_super
->segment_count_sit
) >> 1;
1390 /* setup SIT bitmap from ckeckpoint pack */
1391 bitmap_size
= __bitmap_size(sbi
, SIT_BITMAP
);
1392 src_bitmap
= __bitmap_ptr(sbi
, SIT_BITMAP
);
1394 dst_bitmap
= kmemdup(src_bitmap
, bitmap_size
, GFP_KERNEL
);
1398 /* init SIT information */
1399 sit_i
->s_ops
= &default_salloc_ops
;
1401 sit_i
->sit_base_addr
= le32_to_cpu(raw_super
->sit_blkaddr
);
1402 sit_i
->sit_blocks
= sit_segs
<< sbi
->log_blocks_per_seg
;
1403 sit_i
->written_valid_blocks
= le64_to_cpu(ckpt
->valid_block_count
);
1404 sit_i
->sit_bitmap
= dst_bitmap
;
1405 sit_i
->bitmap_size
= bitmap_size
;
1406 sit_i
->dirty_sentries
= 0;
1407 sit_i
->sents_per_block
= SIT_ENTRY_PER_BLOCK
;
1408 sit_i
->elapsed_time
= le64_to_cpu(sbi
->ckpt
->elapsed_time
);
1409 sit_i
->mounted_time
= CURRENT_TIME_SEC
.tv_sec
;
1410 mutex_init(&sit_i
->sentry_lock
);
1414 static int build_free_segmap(struct f2fs_sb_info
*sbi
)
1416 struct f2fs_sm_info
*sm_info
= SM_I(sbi
);
1417 struct free_segmap_info
*free_i
;
1418 unsigned int bitmap_size
, sec_bitmap_size
;
1420 /* allocate memory for free segmap information */
1421 free_i
= kzalloc(sizeof(struct free_segmap_info
), GFP_KERNEL
);
1425 SM_I(sbi
)->free_info
= free_i
;
1427 bitmap_size
= f2fs_bitmap_size(TOTAL_SEGS(sbi
));
1428 free_i
->free_segmap
= kmalloc(bitmap_size
, GFP_KERNEL
);
1429 if (!free_i
->free_segmap
)
1432 sec_bitmap_size
= f2fs_bitmap_size(TOTAL_SECS(sbi
));
1433 free_i
->free_secmap
= kmalloc(sec_bitmap_size
, GFP_KERNEL
);
1434 if (!free_i
->free_secmap
)
1437 /* set all segments as dirty temporarily */
1438 memset(free_i
->free_segmap
, 0xff, bitmap_size
);
1439 memset(free_i
->free_secmap
, 0xff, sec_bitmap_size
);
1441 /* init free segmap information */
1442 free_i
->start_segno
=
1443 (unsigned int) GET_SEGNO_FROM_SEG0(sbi
, sm_info
->main_blkaddr
);
1444 free_i
->free_segments
= 0;
1445 free_i
->free_sections
= 0;
1446 rwlock_init(&free_i
->segmap_lock
);
1450 static int build_curseg(struct f2fs_sb_info
*sbi
)
1452 struct curseg_info
*array
;
1455 array
= kzalloc(sizeof(*array
) * NR_CURSEG_TYPE
, GFP_KERNEL
);
1459 SM_I(sbi
)->curseg_array
= array
;
1461 for (i
= 0; i
< NR_CURSEG_TYPE
; i
++) {
1462 mutex_init(&array
[i
].curseg_mutex
);
1463 array
[i
].sum_blk
= kzalloc(PAGE_CACHE_SIZE
, GFP_KERNEL
);
1464 if (!array
[i
].sum_blk
)
1466 array
[i
].segno
= NULL_SEGNO
;
1467 array
[i
].next_blkoff
= 0;
1469 return restore_curseg_summaries(sbi
);
1472 static void build_sit_entries(struct f2fs_sb_info
*sbi
)
1474 struct sit_info
*sit_i
= SIT_I(sbi
);
1475 struct curseg_info
*curseg
= CURSEG_I(sbi
, CURSEG_COLD_DATA
);
1476 struct f2fs_summary_block
*sum
= curseg
->sum_blk
;
1479 for (start
= 0; start
< TOTAL_SEGS(sbi
); start
++) {
1480 struct seg_entry
*se
= &sit_i
->sentries
[start
];
1481 struct f2fs_sit_block
*sit_blk
;
1482 struct f2fs_sit_entry sit
;
1486 mutex_lock(&curseg
->curseg_mutex
);
1487 for (i
= 0; i
< sits_in_cursum(sum
); i
++) {
1488 if (le32_to_cpu(segno_in_journal(sum
, i
)) == start
) {
1489 sit
= sit_in_journal(sum
, i
);
1490 mutex_unlock(&curseg
->curseg_mutex
);
1494 mutex_unlock(&curseg
->curseg_mutex
);
1495 page
= get_current_sit_page(sbi
, start
);
1496 sit_blk
= (struct f2fs_sit_block
*)page_address(page
);
1497 sit
= sit_blk
->entries
[SIT_ENTRY_OFFSET(sit_i
, start
)];
1498 f2fs_put_page(page
, 1);
1500 check_block_count(sbi
, start
, &sit
);
1501 seg_info_from_raw_sit(se
, &sit
);
1502 if (sbi
->segs_per_sec
> 1) {
1503 struct sec_entry
*e
= get_sec_entry(sbi
, start
);
1504 e
->valid_blocks
+= se
->valid_blocks
;
1509 static void init_free_segmap(struct f2fs_sb_info
*sbi
)
1514 for (start
= 0; start
< TOTAL_SEGS(sbi
); start
++) {
1515 struct seg_entry
*sentry
= get_seg_entry(sbi
, start
);
1516 if (!sentry
->valid_blocks
)
1517 __set_free(sbi
, start
);
1520 /* set use the current segments */
1521 for (type
= CURSEG_HOT_DATA
; type
<= CURSEG_COLD_NODE
; type
++) {
1522 struct curseg_info
*curseg_t
= CURSEG_I(sbi
, type
);
1523 __set_test_and_inuse(sbi
, curseg_t
->segno
);
1527 static void init_dirty_segmap(struct f2fs_sb_info
*sbi
)
1529 struct dirty_seglist_info
*dirty_i
= DIRTY_I(sbi
);
1530 struct free_segmap_info
*free_i
= FREE_I(sbi
);
1531 unsigned int segno
= 0, offset
= 0, total_segs
= TOTAL_SEGS(sbi
);
1532 unsigned short valid_blocks
;
1535 /* find dirty segment based on free segmap */
1536 segno
= find_next_inuse(free_i
, total_segs
, offset
);
1537 if (segno
>= total_segs
)
1540 valid_blocks
= get_valid_blocks(sbi
, segno
, 0);
1541 if (valid_blocks
>= sbi
->blocks_per_seg
|| !valid_blocks
)
1543 mutex_lock(&dirty_i
->seglist_lock
);
1544 __locate_dirty_segment(sbi
, segno
, DIRTY
);
1545 mutex_unlock(&dirty_i
->seglist_lock
);
1549 static int init_victim_secmap(struct f2fs_sb_info
*sbi
)
1551 struct dirty_seglist_info
*dirty_i
= DIRTY_I(sbi
);
1552 unsigned int bitmap_size
= f2fs_bitmap_size(TOTAL_SECS(sbi
));
1554 dirty_i
->victim_secmap
= kzalloc(bitmap_size
, GFP_KERNEL
);
1555 if (!dirty_i
->victim_secmap
)
1560 static int build_dirty_segmap(struct f2fs_sb_info
*sbi
)
1562 struct dirty_seglist_info
*dirty_i
;
1563 unsigned int bitmap_size
, i
;
1565 /* allocate memory for dirty segments list information */
1566 dirty_i
= kzalloc(sizeof(struct dirty_seglist_info
), GFP_KERNEL
);
1570 SM_I(sbi
)->dirty_info
= dirty_i
;
1571 mutex_init(&dirty_i
->seglist_lock
);
1573 bitmap_size
= f2fs_bitmap_size(TOTAL_SEGS(sbi
));
1575 for (i
= 0; i
< NR_DIRTY_TYPE
; i
++) {
1576 dirty_i
->dirty_segmap
[i
] = kzalloc(bitmap_size
, GFP_KERNEL
);
1577 if (!dirty_i
->dirty_segmap
[i
])
1581 init_dirty_segmap(sbi
);
1582 return init_victim_secmap(sbi
);
1586 * Update min, max modified time for cost-benefit GC algorithm
1588 static void init_min_max_mtime(struct f2fs_sb_info
*sbi
)
1590 struct sit_info
*sit_i
= SIT_I(sbi
);
1593 mutex_lock(&sit_i
->sentry_lock
);
1595 sit_i
->min_mtime
= LLONG_MAX
;
1597 for (segno
= 0; segno
< TOTAL_SEGS(sbi
); segno
+= sbi
->segs_per_sec
) {
1599 unsigned long long mtime
= 0;
1601 for (i
= 0; i
< sbi
->segs_per_sec
; i
++)
1602 mtime
+= get_seg_entry(sbi
, segno
+ i
)->mtime
;
1604 mtime
= div_u64(mtime
, sbi
->segs_per_sec
);
1606 if (sit_i
->min_mtime
> mtime
)
1607 sit_i
->min_mtime
= mtime
;
1609 sit_i
->max_mtime
= get_mtime(sbi
);
1610 mutex_unlock(&sit_i
->sentry_lock
);
1613 int build_segment_manager(struct f2fs_sb_info
*sbi
)
1615 struct f2fs_super_block
*raw_super
= F2FS_RAW_SUPER(sbi
);
1616 struct f2fs_checkpoint
*ckpt
= F2FS_CKPT(sbi
);
1617 struct f2fs_sm_info
*sm_info
;
1620 sm_info
= kzalloc(sizeof(struct f2fs_sm_info
), GFP_KERNEL
);
1625 sbi
->sm_info
= sm_info
;
1626 INIT_LIST_HEAD(&sm_info
->wblist_head
);
1627 spin_lock_init(&sm_info
->wblist_lock
);
1628 sm_info
->seg0_blkaddr
= le32_to_cpu(raw_super
->segment0_blkaddr
);
1629 sm_info
->main_blkaddr
= le32_to_cpu(raw_super
->main_blkaddr
);
1630 sm_info
->segment_count
= le32_to_cpu(raw_super
->segment_count
);
1631 sm_info
->reserved_segments
= le32_to_cpu(ckpt
->rsvd_segment_count
);
1632 sm_info
->ovp_segments
= le32_to_cpu(ckpt
->overprov_segment_count
);
1633 sm_info
->main_segments
= le32_to_cpu(raw_super
->segment_count_main
);
1634 sm_info
->ssa_blkaddr
= le32_to_cpu(raw_super
->ssa_blkaddr
);
1636 err
= build_sit_info(sbi
);
1639 err
= build_free_segmap(sbi
);
1642 err
= build_curseg(sbi
);
1646 /* reinit free segmap based on SIT */
1647 build_sit_entries(sbi
);
1649 init_free_segmap(sbi
);
1650 err
= build_dirty_segmap(sbi
);
1654 init_min_max_mtime(sbi
);
1658 static void discard_dirty_segmap(struct f2fs_sb_info
*sbi
,
1659 enum dirty_type dirty_type
)
1661 struct dirty_seglist_info
*dirty_i
= DIRTY_I(sbi
);
1663 mutex_lock(&dirty_i
->seglist_lock
);
1664 kfree(dirty_i
->dirty_segmap
[dirty_type
]);
1665 dirty_i
->nr_dirty
[dirty_type
] = 0;
1666 mutex_unlock(&dirty_i
->seglist_lock
);
1669 static void destroy_victim_secmap(struct f2fs_sb_info
*sbi
)
1671 struct dirty_seglist_info
*dirty_i
= DIRTY_I(sbi
);
1672 kfree(dirty_i
->victim_secmap
);
1675 static void destroy_dirty_segmap(struct f2fs_sb_info
*sbi
)
1677 struct dirty_seglist_info
*dirty_i
= DIRTY_I(sbi
);
1683 /* discard pre-free/dirty segments list */
1684 for (i
= 0; i
< NR_DIRTY_TYPE
; i
++)
1685 discard_dirty_segmap(sbi
, i
);
1687 destroy_victim_secmap(sbi
);
1688 SM_I(sbi
)->dirty_info
= NULL
;
1692 static void destroy_curseg(struct f2fs_sb_info
*sbi
)
1694 struct curseg_info
*array
= SM_I(sbi
)->curseg_array
;
1699 SM_I(sbi
)->curseg_array
= NULL
;
1700 for (i
= 0; i
< NR_CURSEG_TYPE
; i
++)
1701 kfree(array
[i
].sum_blk
);
1705 static void destroy_free_segmap(struct f2fs_sb_info
*sbi
)
1707 struct free_segmap_info
*free_i
= SM_I(sbi
)->free_info
;
1710 SM_I(sbi
)->free_info
= NULL
;
1711 kfree(free_i
->free_segmap
);
1712 kfree(free_i
->free_secmap
);
1716 static void destroy_sit_info(struct f2fs_sb_info
*sbi
)
1718 struct sit_info
*sit_i
= SIT_I(sbi
);
1724 if (sit_i
->sentries
) {
1725 for (start
= 0; start
< TOTAL_SEGS(sbi
); start
++) {
1726 kfree(sit_i
->sentries
[start
].cur_valid_map
);
1727 kfree(sit_i
->sentries
[start
].ckpt_valid_map
);
1730 vfree(sit_i
->sentries
);
1731 vfree(sit_i
->sec_entries
);
1732 kfree(sit_i
->dirty_sentries_bitmap
);
1734 SM_I(sbi
)->sit_info
= NULL
;
1735 kfree(sit_i
->sit_bitmap
);
1739 void destroy_segment_manager(struct f2fs_sb_info
*sbi
)
1741 struct f2fs_sm_info
*sm_info
= SM_I(sbi
);
1742 destroy_dirty_segmap(sbi
);
1743 destroy_curseg(sbi
);
1744 destroy_free_segmap(sbi
);
1745 destroy_sit_info(sbi
);
1746 sbi
->sm_info
= NULL
;