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>
22 static int need_to_flush(struct f2fs_sb_info
*sbi
)
24 unsigned int pages_per_sec
= (1 << sbi
->log_blocks_per_seg
) *
26 int node_secs
= ((get_pages(sbi
, F2FS_DIRTY_NODES
) + pages_per_sec
- 1)
27 >> sbi
->log_blocks_per_seg
) / sbi
->segs_per_sec
;
28 int dent_secs
= ((get_pages(sbi
, F2FS_DIRTY_DENTS
) + pages_per_sec
- 1)
29 >> sbi
->log_blocks_per_seg
) / sbi
->segs_per_sec
;
34 if (free_sections(sbi
) <= (node_secs
+ 2 * dent_secs
+
35 reserved_sections(sbi
)))
41 * This function balances dirty node and dentry pages.
42 * In addition, it controls garbage collection.
44 void f2fs_balance_fs(struct f2fs_sb_info
*sbi
)
46 struct writeback_control wbc
= {
47 .sync_mode
= WB_SYNC_ALL
,
48 .nr_to_write
= LONG_MAX
,
56 * We should do checkpoint when there are so many dirty node pages
57 * with enough free segments. After then, we should do GC.
59 if (need_to_flush(sbi
)) {
60 sync_dirty_dir_inodes(sbi
);
61 sync_node_pages(sbi
, 0, &wbc
);
64 if (has_not_enough_free_secs(sbi
)) {
65 mutex_lock(&sbi
->gc_mutex
);
70 static void __locate_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 /* need not be added */
76 if (IS_CURSEG(sbi
, segno
))
79 if (!test_and_set_bit(segno
, dirty_i
->dirty_segmap
[dirty_type
]))
80 dirty_i
->nr_dirty
[dirty_type
]++;
82 if (dirty_type
== DIRTY
) {
83 struct seg_entry
*sentry
= get_seg_entry(sbi
, segno
);
84 dirty_type
= sentry
->type
;
85 if (!test_and_set_bit(segno
, dirty_i
->dirty_segmap
[dirty_type
]))
86 dirty_i
->nr_dirty
[dirty_type
]++;
90 static void __remove_dirty_segment(struct f2fs_sb_info
*sbi
, unsigned int segno
,
91 enum dirty_type dirty_type
)
93 struct dirty_seglist_info
*dirty_i
= DIRTY_I(sbi
);
95 if (test_and_clear_bit(segno
, dirty_i
->dirty_segmap
[dirty_type
]))
96 dirty_i
->nr_dirty
[dirty_type
]--;
98 if (dirty_type
== DIRTY
) {
99 struct seg_entry
*sentry
= get_seg_entry(sbi
, segno
);
100 dirty_type
= sentry
->type
;
101 if (test_and_clear_bit(segno
,
102 dirty_i
->dirty_segmap
[dirty_type
]))
103 dirty_i
->nr_dirty
[dirty_type
]--;
104 clear_bit(segno
, dirty_i
->victim_segmap
[FG_GC
]);
105 clear_bit(segno
, dirty_i
->victim_segmap
[BG_GC
]);
110 * Should not occur error such as -ENOMEM.
111 * Adding dirty entry into seglist is not critical operation.
112 * If a given segment is one of current working segments, it won't be added.
114 void locate_dirty_segment(struct f2fs_sb_info
*sbi
, unsigned int segno
)
116 struct dirty_seglist_info
*dirty_i
= DIRTY_I(sbi
);
117 unsigned short valid_blocks
;
119 if (segno
== NULL_SEGNO
|| IS_CURSEG(sbi
, segno
))
122 mutex_lock(&dirty_i
->seglist_lock
);
124 valid_blocks
= get_valid_blocks(sbi
, segno
, 0);
126 if (valid_blocks
== 0) {
127 __locate_dirty_segment(sbi
, segno
, PRE
);
128 __remove_dirty_segment(sbi
, segno
, DIRTY
);
129 } else if (valid_blocks
< sbi
->blocks_per_seg
) {
130 __locate_dirty_segment(sbi
, segno
, DIRTY
);
132 /* Recovery routine with SSR needs this */
133 __remove_dirty_segment(sbi
, segno
, DIRTY
);
136 mutex_unlock(&dirty_i
->seglist_lock
);
141 * Should call clear_prefree_segments after checkpoint is done.
143 static void set_prefree_as_free_segments(struct f2fs_sb_info
*sbi
)
145 struct dirty_seglist_info
*dirty_i
= DIRTY_I(sbi
);
146 unsigned int segno
, offset
= 0;
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
)
155 __set_test_and_free(sbi
, segno
);
158 mutex_unlock(&dirty_i
->seglist_lock
);
161 void clear_prefree_segments(struct f2fs_sb_info
*sbi
)
163 struct dirty_seglist_info
*dirty_i
= DIRTY_I(sbi
);
164 unsigned int segno
, offset
= 0;
165 unsigned int total_segs
= TOTAL_SEGS(sbi
);
167 mutex_lock(&dirty_i
->seglist_lock
);
169 segno
= find_next_bit(dirty_i
->dirty_segmap
[PRE
], total_segs
,
171 if (segno
>= total_segs
)
175 if (test_and_clear_bit(segno
, dirty_i
->dirty_segmap
[PRE
]))
176 dirty_i
->nr_dirty
[PRE
]--;
179 if (test_opt(sbi
, DISCARD
))
180 blkdev_issue_discard(sbi
->sb
->s_bdev
,
181 START_BLOCK(sbi
, segno
) <<
182 sbi
->log_sectors_per_block
,
183 1 << (sbi
->log_sectors_per_block
+
184 sbi
->log_blocks_per_seg
),
187 mutex_unlock(&dirty_i
->seglist_lock
);
190 static void __mark_sit_entry_dirty(struct f2fs_sb_info
*sbi
, unsigned int segno
)
192 struct sit_info
*sit_i
= SIT_I(sbi
);
193 if (!__test_and_set_bit(segno
, sit_i
->dirty_sentries_bitmap
))
194 sit_i
->dirty_sentries
++;
197 static void __set_sit_entry_type(struct f2fs_sb_info
*sbi
, int type
,
198 unsigned int segno
, int modified
)
200 struct seg_entry
*se
= get_seg_entry(sbi
, segno
);
203 __mark_sit_entry_dirty(sbi
, segno
);
206 static void update_sit_entry(struct f2fs_sb_info
*sbi
, block_t blkaddr
, int del
)
208 struct seg_entry
*se
;
209 unsigned int segno
, offset
;
210 long int new_vblocks
;
212 segno
= GET_SEGNO(sbi
, blkaddr
);
214 se
= get_seg_entry(sbi
, segno
);
215 new_vblocks
= se
->valid_blocks
+ del
;
216 offset
= GET_SEGOFF_FROM_SEG0(sbi
, blkaddr
) & (sbi
->blocks_per_seg
- 1);
218 BUG_ON((new_vblocks
>> (sizeof(unsigned short) << 3) ||
219 (new_vblocks
> sbi
->blocks_per_seg
)));
221 se
->valid_blocks
= new_vblocks
;
222 se
->mtime
= get_mtime(sbi
);
223 SIT_I(sbi
)->max_mtime
= se
->mtime
;
225 /* Update valid block bitmap */
227 if (f2fs_set_bit(offset
, se
->cur_valid_map
))
230 if (!f2fs_clear_bit(offset
, se
->cur_valid_map
))
233 if (!f2fs_test_bit(offset
, se
->ckpt_valid_map
))
234 se
->ckpt_valid_blocks
+= del
;
236 __mark_sit_entry_dirty(sbi
, segno
);
238 /* update total number of valid blocks to be written in ckpt area */
239 SIT_I(sbi
)->written_valid_blocks
+= del
;
241 if (sbi
->segs_per_sec
> 1)
242 get_sec_entry(sbi
, segno
)->valid_blocks
+= del
;
245 static void refresh_sit_entry(struct f2fs_sb_info
*sbi
,
246 block_t old_blkaddr
, block_t new_blkaddr
)
248 update_sit_entry(sbi
, new_blkaddr
, 1);
249 if (GET_SEGNO(sbi
, old_blkaddr
) != NULL_SEGNO
)
250 update_sit_entry(sbi
, old_blkaddr
, -1);
253 void invalidate_blocks(struct f2fs_sb_info
*sbi
, block_t addr
)
255 unsigned int segno
= GET_SEGNO(sbi
, addr
);
256 struct sit_info
*sit_i
= SIT_I(sbi
);
258 BUG_ON(addr
== NULL_ADDR
);
259 if (addr
== NEW_ADDR
)
262 /* add it into sit main buffer */
263 mutex_lock(&sit_i
->sentry_lock
);
265 update_sit_entry(sbi
, addr
, -1);
267 /* add it into dirty seglist */
268 locate_dirty_segment(sbi
, segno
);
270 mutex_unlock(&sit_i
->sentry_lock
);
274 * This function should be resided under the curseg_mutex lock
276 static void __add_sum_entry(struct f2fs_sb_info
*sbi
, int type
,
277 struct f2fs_summary
*sum
, unsigned short offset
)
279 struct curseg_info
*curseg
= CURSEG_I(sbi
, type
);
280 void *addr
= curseg
->sum_blk
;
281 addr
+= offset
* sizeof(struct f2fs_summary
);
282 memcpy(addr
, sum
, sizeof(struct f2fs_summary
));
287 * Calculate the number of current summary pages for writing
289 int npages_for_summary_flush(struct f2fs_sb_info
*sbi
)
291 int total_size_bytes
= 0;
292 int valid_sum_count
= 0;
295 for (i
= CURSEG_HOT_DATA
; i
<= CURSEG_COLD_DATA
; i
++) {
296 if (sbi
->ckpt
->alloc_type
[i
] == SSR
)
297 valid_sum_count
+= sbi
->blocks_per_seg
;
299 valid_sum_count
+= curseg_blkoff(sbi
, i
);
302 total_size_bytes
= valid_sum_count
* (SUMMARY_SIZE
+ 1)
303 + sizeof(struct nat_journal
) + 2
304 + sizeof(struct sit_journal
) + 2;
305 sum_space
= PAGE_CACHE_SIZE
- SUM_FOOTER_SIZE
;
306 if (total_size_bytes
< sum_space
)
308 else if (total_size_bytes
< 2 * sum_space
)
314 * Caller should put this summary page
316 struct page
*get_sum_page(struct f2fs_sb_info
*sbi
, unsigned int segno
)
318 return get_meta_page(sbi
, GET_SUM_BLOCK(sbi
, segno
));
321 static void write_sum_page(struct f2fs_sb_info
*sbi
,
322 struct f2fs_summary_block
*sum_blk
, block_t blk_addr
)
324 struct page
*page
= grab_meta_page(sbi
, blk_addr
);
325 void *kaddr
= page_address(page
);
326 memcpy(kaddr
, sum_blk
, PAGE_CACHE_SIZE
);
327 set_page_dirty(page
);
328 f2fs_put_page(page
, 1);
331 static unsigned int check_prefree_segments(struct f2fs_sb_info
*sbi
,
332 int ofs_unit
, int type
)
334 struct dirty_seglist_info
*dirty_i
= DIRTY_I(sbi
);
335 unsigned long *prefree_segmap
= dirty_i
->dirty_segmap
[PRE
];
336 unsigned int segno
, next_segno
, i
;
340 * If there is not enough reserved sections,
341 * we should not reuse prefree segments.
343 if (has_not_enough_free_secs(sbi
))
347 * NODE page should not reuse prefree segment,
348 * since those information is used for SPOR.
350 if (IS_NODESEG(type
))
353 segno
= find_next_bit(prefree_segmap
, TOTAL_SEGS(sbi
), ofs
++);
354 ofs
= ((segno
/ ofs_unit
) * ofs_unit
) + ofs_unit
;
355 if (segno
< TOTAL_SEGS(sbi
)) {
356 /* skip intermediate segments in a section */
357 if (segno
% ofs_unit
)
360 /* skip if whole section is not prefree */
361 next_segno
= find_next_zero_bit(prefree_segmap
,
362 TOTAL_SEGS(sbi
), segno
+ 1);
363 if (next_segno
- segno
< ofs_unit
)
366 /* skip if whole section was not free at the last checkpoint */
367 for (i
= 0; i
< ofs_unit
; i
++)
368 if (get_seg_entry(sbi
, segno
)->ckpt_valid_blocks
)
376 * Find a new segment from the free segments bitmap to right order
377 * This function should be returned with success, otherwise BUG
379 static void get_new_segment(struct f2fs_sb_info
*sbi
,
380 unsigned int *newseg
, bool new_sec
, int dir
)
382 struct free_segmap_info
*free_i
= FREE_I(sbi
);
383 unsigned int total_secs
= sbi
->total_sections
;
384 unsigned int segno
, secno
, zoneno
;
385 unsigned int total_zones
= sbi
->total_sections
/ sbi
->secs_per_zone
;
386 unsigned int hint
= *newseg
/ sbi
->segs_per_sec
;
387 unsigned int old_zoneno
= GET_ZONENO_FROM_SEGNO(sbi
, *newseg
);
388 unsigned int left_start
= hint
;
393 write_lock(&free_i
->segmap_lock
);
395 if (!new_sec
&& ((*newseg
+ 1) % sbi
->segs_per_sec
)) {
396 segno
= find_next_zero_bit(free_i
->free_segmap
,
397 TOTAL_SEGS(sbi
), *newseg
+ 1);
398 if (segno
< TOTAL_SEGS(sbi
))
402 secno
= find_next_zero_bit(free_i
->free_secmap
, total_secs
, hint
);
403 if (secno
>= total_secs
) {
404 if (dir
== ALLOC_RIGHT
) {
405 secno
= find_next_zero_bit(free_i
->free_secmap
,
407 BUG_ON(secno
>= total_secs
);
410 left_start
= hint
- 1;
416 while (test_bit(left_start
, free_i
->free_secmap
)) {
417 if (left_start
> 0) {
421 left_start
= find_next_zero_bit(free_i
->free_secmap
,
423 BUG_ON(left_start
>= total_secs
);
429 segno
= secno
* sbi
->segs_per_sec
;
430 zoneno
= secno
/ sbi
->secs_per_zone
;
432 /* give up on finding another zone */
435 if (sbi
->secs_per_zone
== 1)
437 if (zoneno
== old_zoneno
)
439 if (dir
== ALLOC_LEFT
) {
440 if (!go_left
&& zoneno
+ 1 >= total_zones
)
442 if (go_left
&& zoneno
== 0)
445 for (i
= 0; i
< NR_CURSEG_TYPE
; i
++)
446 if (CURSEG_I(sbi
, i
)->zone
== zoneno
)
449 if (i
< NR_CURSEG_TYPE
) {
450 /* zone is in user, try another */
452 hint
= zoneno
* sbi
->secs_per_zone
- 1;
453 else if (zoneno
+ 1 >= total_zones
)
456 hint
= (zoneno
+ 1) * sbi
->secs_per_zone
;
458 goto find_other_zone
;
461 /* set it as dirty segment in free segmap */
462 BUG_ON(test_bit(segno
, free_i
->free_segmap
));
463 __set_inuse(sbi
, segno
);
465 write_unlock(&free_i
->segmap_lock
);
468 static void reset_curseg(struct f2fs_sb_info
*sbi
, int type
, int modified
)
470 struct curseg_info
*curseg
= CURSEG_I(sbi
, type
);
471 struct summary_footer
*sum_footer
;
473 curseg
->segno
= curseg
->next_segno
;
474 curseg
->zone
= GET_ZONENO_FROM_SEGNO(sbi
, curseg
->segno
);
475 curseg
->next_blkoff
= 0;
476 curseg
->next_segno
= NULL_SEGNO
;
478 sum_footer
= &(curseg
->sum_blk
->footer
);
479 memset(sum_footer
, 0, sizeof(struct summary_footer
));
480 if (IS_DATASEG(type
))
481 SET_SUM_TYPE(sum_footer
, SUM_TYPE_DATA
);
482 if (IS_NODESEG(type
))
483 SET_SUM_TYPE(sum_footer
, SUM_TYPE_NODE
);
484 __set_sit_entry_type(sbi
, type
, curseg
->segno
, modified
);
488 * Allocate a current working segment.
489 * This function always allocates a free segment in LFS manner.
491 static void new_curseg(struct f2fs_sb_info
*sbi
, int type
, bool new_sec
)
493 struct curseg_info
*curseg
= CURSEG_I(sbi
, type
);
494 unsigned int segno
= curseg
->segno
;
495 int dir
= ALLOC_LEFT
;
497 write_sum_page(sbi
, curseg
->sum_blk
,
498 GET_SUM_BLOCK(sbi
, curseg
->segno
));
499 if (type
== CURSEG_WARM_DATA
|| type
== CURSEG_COLD_DATA
)
502 if (test_opt(sbi
, NOHEAP
))
505 get_new_segment(sbi
, &segno
, new_sec
, dir
);
506 curseg
->next_segno
= segno
;
507 reset_curseg(sbi
, type
, 1);
508 curseg
->alloc_type
= LFS
;
511 static void __next_free_blkoff(struct f2fs_sb_info
*sbi
,
512 struct curseg_info
*seg
, block_t start
)
514 struct seg_entry
*se
= get_seg_entry(sbi
, seg
->segno
);
516 for (ofs
= start
; ofs
< sbi
->blocks_per_seg
; ofs
++) {
517 if (!f2fs_test_bit(ofs
, se
->ckpt_valid_map
)
518 && !f2fs_test_bit(ofs
, se
->cur_valid_map
))
521 seg
->next_blkoff
= ofs
;
525 * If a segment is written by LFS manner, next block offset is just obtained
526 * by increasing the current block offset. However, if a segment is written by
527 * SSR manner, next block offset obtained by calling __next_free_blkoff
529 static void __refresh_next_blkoff(struct f2fs_sb_info
*sbi
,
530 struct curseg_info
*seg
)
532 if (seg
->alloc_type
== SSR
)
533 __next_free_blkoff(sbi
, seg
, seg
->next_blkoff
+ 1);
539 * This function always allocates a used segment (from dirty seglist) by SSR
540 * manner, so it should recover the existing segment information of valid blocks
542 static void change_curseg(struct f2fs_sb_info
*sbi
, int type
, bool reuse
)
544 struct dirty_seglist_info
*dirty_i
= DIRTY_I(sbi
);
545 struct curseg_info
*curseg
= CURSEG_I(sbi
, type
);
546 unsigned int new_segno
= curseg
->next_segno
;
547 struct f2fs_summary_block
*sum_node
;
548 struct page
*sum_page
;
550 write_sum_page(sbi
, curseg
->sum_blk
,
551 GET_SUM_BLOCK(sbi
, curseg
->segno
));
552 __set_test_and_inuse(sbi
, new_segno
);
554 mutex_lock(&dirty_i
->seglist_lock
);
555 __remove_dirty_segment(sbi
, new_segno
, PRE
);
556 __remove_dirty_segment(sbi
, new_segno
, DIRTY
);
557 mutex_unlock(&dirty_i
->seglist_lock
);
559 reset_curseg(sbi
, type
, 1);
560 curseg
->alloc_type
= SSR
;
561 __next_free_blkoff(sbi
, curseg
, 0);
564 sum_page
= get_sum_page(sbi
, new_segno
);
565 sum_node
= (struct f2fs_summary_block
*)page_address(sum_page
);
566 memcpy(curseg
->sum_blk
, sum_node
, SUM_ENTRY_SIZE
);
567 f2fs_put_page(sum_page
, 1);
572 * flush out current segment and replace it with new segment
573 * This function should be returned with success, otherwise BUG
575 static void allocate_segment_by_default(struct f2fs_sb_info
*sbi
,
576 int type
, bool force
)
578 struct curseg_info
*curseg
= CURSEG_I(sbi
, type
);
579 unsigned int ofs_unit
;
582 new_curseg(sbi
, type
, true);
586 ofs_unit
= need_SSR(sbi
) ? 1 : sbi
->segs_per_sec
;
587 curseg
->next_segno
= check_prefree_segments(sbi
, ofs_unit
, type
);
589 if (curseg
->next_segno
!= NULL_SEGNO
)
590 change_curseg(sbi
, type
, false);
591 else if (type
== CURSEG_WARM_NODE
)
592 new_curseg(sbi
, type
, false);
593 else if (need_SSR(sbi
) && get_ssr_segment(sbi
, type
))
594 change_curseg(sbi
, type
, true);
596 new_curseg(sbi
, type
, false);
598 sbi
->segment_count
[curseg
->alloc_type
]++;
601 void allocate_new_segments(struct f2fs_sb_info
*sbi
)
603 struct curseg_info
*curseg
;
604 unsigned int old_curseg
;
607 for (i
= CURSEG_HOT_DATA
; i
<= CURSEG_COLD_DATA
; i
++) {
608 curseg
= CURSEG_I(sbi
, i
);
609 old_curseg
= curseg
->segno
;
610 SIT_I(sbi
)->s_ops
->allocate_segment(sbi
, i
, true);
611 locate_dirty_segment(sbi
, old_curseg
);
615 static const struct segment_allocation default_salloc_ops
= {
616 .allocate_segment
= allocate_segment_by_default
,
619 static void f2fs_end_io_write(struct bio
*bio
, int err
)
621 const int uptodate
= test_bit(BIO_UPTODATE
, &bio
->bi_flags
);
622 struct bio_vec
*bvec
= bio
->bi_io_vec
+ bio
->bi_vcnt
- 1;
623 struct bio_private
*p
= bio
->bi_private
;
626 struct page
*page
= bvec
->bv_page
;
628 if (--bvec
>= bio
->bi_io_vec
)
629 prefetchw(&bvec
->bv_page
->flags
);
633 set_bit(AS_EIO
, &page
->mapping
->flags
);
634 set_ckpt_flags(p
->sbi
->ckpt
, CP_ERROR_FLAG
);
636 end_page_writeback(page
);
637 dec_page_count(p
->sbi
, F2FS_WRITEBACK
);
638 } while (bvec
>= bio
->bi_io_vec
);
646 struct bio
*f2fs_bio_alloc(struct block_device
*bdev
, int npages
)
649 struct bio_private
*priv
;
651 priv
= kmalloc(sizeof(struct bio_private
), GFP_NOFS
);
657 /* No failure on bio allocation */
658 bio
= bio_alloc(GFP_NOIO
, npages
);
660 bio
->bi_private
= priv
;
664 static void do_submit_bio(struct f2fs_sb_info
*sbi
,
665 enum page_type type
, bool sync
)
667 int rw
= sync
? WRITE_SYNC
: WRITE
;
668 enum page_type btype
= type
> META
? META
: type
;
670 if (type
>= META_FLUSH
)
671 rw
= WRITE_FLUSH_FUA
;
673 if (sbi
->bio
[btype
]) {
674 struct bio_private
*p
= sbi
->bio
[btype
]->bi_private
;
676 sbi
->bio
[btype
]->bi_end_io
= f2fs_end_io_write
;
677 if (type
== META_FLUSH
) {
678 DECLARE_COMPLETION_ONSTACK(wait
);
681 submit_bio(rw
, sbi
->bio
[btype
]);
682 wait_for_completion(&wait
);
685 submit_bio(rw
, sbi
->bio
[btype
]);
687 sbi
->bio
[btype
] = NULL
;
691 void f2fs_submit_bio(struct f2fs_sb_info
*sbi
, enum page_type type
, bool sync
)
693 down_write(&sbi
->bio_sem
);
694 do_submit_bio(sbi
, type
, sync
);
695 up_write(&sbi
->bio_sem
);
698 static void submit_write_page(struct f2fs_sb_info
*sbi
, struct page
*page
,
699 block_t blk_addr
, enum page_type type
)
701 struct block_device
*bdev
= sbi
->sb
->s_bdev
;
703 verify_block_addr(sbi
, blk_addr
);
705 down_write(&sbi
->bio_sem
);
707 inc_page_count(sbi
, F2FS_WRITEBACK
);
709 if (sbi
->bio
[type
] && sbi
->last_block_in_bio
[type
] != blk_addr
- 1)
710 do_submit_bio(sbi
, type
, false);
712 if (sbi
->bio
[type
] == NULL
) {
713 sbi
->bio
[type
] = f2fs_bio_alloc(bdev
, bio_get_nr_vecs(bdev
));
714 sbi
->bio
[type
]->bi_sector
= SECTOR_FROM_BLOCK(sbi
, blk_addr
);
716 * The end_io will be assigned at the sumbission phase.
717 * Until then, let bio_add_page() merge consecutive IOs as much
722 if (bio_add_page(sbi
->bio
[type
], page
, PAGE_CACHE_SIZE
, 0) <
724 do_submit_bio(sbi
, type
, false);
728 sbi
->last_block_in_bio
[type
] = blk_addr
;
730 up_write(&sbi
->bio_sem
);
733 static bool __has_curseg_space(struct f2fs_sb_info
*sbi
, int type
)
735 struct curseg_info
*curseg
= CURSEG_I(sbi
, type
);
736 if (curseg
->next_blkoff
< sbi
->blocks_per_seg
)
741 static int __get_segment_type_2(struct page
*page
, enum page_type p_type
)
744 return CURSEG_HOT_DATA
;
746 return CURSEG_HOT_NODE
;
749 static int __get_segment_type_4(struct page
*page
, enum page_type p_type
)
751 if (p_type
== DATA
) {
752 struct inode
*inode
= page
->mapping
->host
;
754 if (S_ISDIR(inode
->i_mode
))
755 return CURSEG_HOT_DATA
;
757 return CURSEG_COLD_DATA
;
759 if (IS_DNODE(page
) && !is_cold_node(page
))
760 return CURSEG_HOT_NODE
;
762 return CURSEG_COLD_NODE
;
766 static int __get_segment_type_6(struct page
*page
, enum page_type p_type
)
768 if (p_type
== DATA
) {
769 struct inode
*inode
= page
->mapping
->host
;
771 if (S_ISDIR(inode
->i_mode
))
772 return CURSEG_HOT_DATA
;
773 else if (is_cold_data(page
) || is_cold_file(inode
))
774 return CURSEG_COLD_DATA
;
776 return CURSEG_WARM_DATA
;
779 return is_cold_node(page
) ? CURSEG_WARM_NODE
:
782 return CURSEG_COLD_NODE
;
786 static int __get_segment_type(struct page
*page
, enum page_type p_type
)
788 struct f2fs_sb_info
*sbi
= F2FS_SB(page
->mapping
->host
->i_sb
);
789 switch (sbi
->active_logs
) {
791 return __get_segment_type_2(page
, p_type
);
793 return __get_segment_type_4(page
, p_type
);
795 /* NR_CURSEG_TYPE(6) logs by default */
796 BUG_ON(sbi
->active_logs
!= NR_CURSEG_TYPE
);
797 return __get_segment_type_6(page
, p_type
);
800 static void do_write_page(struct f2fs_sb_info
*sbi
, struct page
*page
,
801 block_t old_blkaddr
, block_t
*new_blkaddr
,
802 struct f2fs_summary
*sum
, enum page_type p_type
)
804 struct sit_info
*sit_i
= SIT_I(sbi
);
805 struct curseg_info
*curseg
;
806 unsigned int old_cursegno
;
809 type
= __get_segment_type(page
, p_type
);
810 curseg
= CURSEG_I(sbi
, type
);
812 mutex_lock(&curseg
->curseg_mutex
);
814 *new_blkaddr
= NEXT_FREE_BLKADDR(sbi
, curseg
);
815 old_cursegno
= curseg
->segno
;
818 * __add_sum_entry should be resided under the curseg_mutex
819 * because, this function updates a summary entry in the
820 * current summary block.
822 __add_sum_entry(sbi
, type
, sum
, curseg
->next_blkoff
);
824 mutex_lock(&sit_i
->sentry_lock
);
825 __refresh_next_blkoff(sbi
, curseg
);
826 sbi
->block_count
[curseg
->alloc_type
]++;
829 * SIT information should be updated before segment allocation,
830 * since SSR needs latest valid block information.
832 refresh_sit_entry(sbi
, old_blkaddr
, *new_blkaddr
);
834 if (!__has_curseg_space(sbi
, type
))
835 sit_i
->s_ops
->allocate_segment(sbi
, type
, false);
837 locate_dirty_segment(sbi
, old_cursegno
);
838 locate_dirty_segment(sbi
, GET_SEGNO(sbi
, old_blkaddr
));
839 mutex_unlock(&sit_i
->sentry_lock
);
842 fill_node_footer_blkaddr(page
, NEXT_FREE_BLKADDR(sbi
, curseg
));
844 /* writeout dirty page into bdev */
845 submit_write_page(sbi
, page
, *new_blkaddr
, p_type
);
847 mutex_unlock(&curseg
->curseg_mutex
);
850 int write_meta_page(struct f2fs_sb_info
*sbi
, struct page
*page
,
851 struct writeback_control
*wbc
)
853 if (wbc
->for_reclaim
)
854 return AOP_WRITEPAGE_ACTIVATE
;
856 set_page_writeback(page
);
857 submit_write_page(sbi
, page
, page
->index
, META
);
861 void write_node_page(struct f2fs_sb_info
*sbi
, struct page
*page
,
862 unsigned int nid
, block_t old_blkaddr
, block_t
*new_blkaddr
)
864 struct f2fs_summary sum
;
865 set_summary(&sum
, nid
, 0, 0);
866 do_write_page(sbi
, page
, old_blkaddr
, new_blkaddr
, &sum
, NODE
);
869 void write_data_page(struct inode
*inode
, struct page
*page
,
870 struct dnode_of_data
*dn
, block_t old_blkaddr
,
871 block_t
*new_blkaddr
)
873 struct f2fs_sb_info
*sbi
= F2FS_SB(inode
->i_sb
);
874 struct f2fs_summary sum
;
877 BUG_ON(old_blkaddr
== NULL_ADDR
);
878 get_node_info(sbi
, dn
->nid
, &ni
);
879 set_summary(&sum
, dn
->nid
, dn
->ofs_in_node
, ni
.version
);
881 do_write_page(sbi
, page
, old_blkaddr
,
882 new_blkaddr
, &sum
, DATA
);
885 void rewrite_data_page(struct f2fs_sb_info
*sbi
, struct page
*page
,
886 block_t old_blk_addr
)
888 submit_write_page(sbi
, page
, old_blk_addr
, DATA
);
891 void recover_data_page(struct f2fs_sb_info
*sbi
,
892 struct page
*page
, struct f2fs_summary
*sum
,
893 block_t old_blkaddr
, block_t new_blkaddr
)
895 struct sit_info
*sit_i
= SIT_I(sbi
);
896 struct curseg_info
*curseg
;
897 unsigned int segno
, old_cursegno
;
898 struct seg_entry
*se
;
901 segno
= GET_SEGNO(sbi
, new_blkaddr
);
902 se
= get_seg_entry(sbi
, segno
);
905 if (se
->valid_blocks
== 0 && !IS_CURSEG(sbi
, segno
)) {
906 if (old_blkaddr
== NULL_ADDR
)
907 type
= CURSEG_COLD_DATA
;
909 type
= CURSEG_WARM_DATA
;
911 curseg
= CURSEG_I(sbi
, type
);
913 mutex_lock(&curseg
->curseg_mutex
);
914 mutex_lock(&sit_i
->sentry_lock
);
916 old_cursegno
= curseg
->segno
;
918 /* change the current segment */
919 if (segno
!= curseg
->segno
) {
920 curseg
->next_segno
= segno
;
921 change_curseg(sbi
, type
, true);
924 curseg
->next_blkoff
= GET_SEGOFF_FROM_SEG0(sbi
, new_blkaddr
) &
925 (sbi
->blocks_per_seg
- 1);
926 __add_sum_entry(sbi
, type
, sum
, curseg
->next_blkoff
);
928 refresh_sit_entry(sbi
, old_blkaddr
, new_blkaddr
);
930 locate_dirty_segment(sbi
, old_cursegno
);
931 locate_dirty_segment(sbi
, GET_SEGNO(sbi
, old_blkaddr
));
933 mutex_unlock(&sit_i
->sentry_lock
);
934 mutex_unlock(&curseg
->curseg_mutex
);
937 void rewrite_node_page(struct f2fs_sb_info
*sbi
,
938 struct page
*page
, struct f2fs_summary
*sum
,
939 block_t old_blkaddr
, block_t new_blkaddr
)
941 struct sit_info
*sit_i
= SIT_I(sbi
);
942 int type
= CURSEG_WARM_NODE
;
943 struct curseg_info
*curseg
;
944 unsigned int segno
, old_cursegno
;
945 block_t next_blkaddr
= next_blkaddr_of_node(page
);
946 unsigned int next_segno
= GET_SEGNO(sbi
, next_blkaddr
);
948 curseg
= CURSEG_I(sbi
, type
);
950 mutex_lock(&curseg
->curseg_mutex
);
951 mutex_lock(&sit_i
->sentry_lock
);
953 segno
= GET_SEGNO(sbi
, new_blkaddr
);
954 old_cursegno
= curseg
->segno
;
956 /* change the current segment */
957 if (segno
!= curseg
->segno
) {
958 curseg
->next_segno
= segno
;
959 change_curseg(sbi
, type
, true);
961 curseg
->next_blkoff
= GET_SEGOFF_FROM_SEG0(sbi
, new_blkaddr
) &
962 (sbi
->blocks_per_seg
- 1);
963 __add_sum_entry(sbi
, type
, sum
, curseg
->next_blkoff
);
965 /* change the current log to the next block addr in advance */
966 if (next_segno
!= segno
) {
967 curseg
->next_segno
= next_segno
;
968 change_curseg(sbi
, type
, true);
970 curseg
->next_blkoff
= GET_SEGOFF_FROM_SEG0(sbi
, next_blkaddr
) &
971 (sbi
->blocks_per_seg
- 1);
973 /* rewrite node page */
974 set_page_writeback(page
);
975 submit_write_page(sbi
, page
, new_blkaddr
, NODE
);
976 f2fs_submit_bio(sbi
, NODE
, true);
977 refresh_sit_entry(sbi
, old_blkaddr
, new_blkaddr
);
979 locate_dirty_segment(sbi
, old_cursegno
);
980 locate_dirty_segment(sbi
, GET_SEGNO(sbi
, old_blkaddr
));
982 mutex_unlock(&sit_i
->sentry_lock
);
983 mutex_unlock(&curseg
->curseg_mutex
);
986 static int read_compacted_summaries(struct f2fs_sb_info
*sbi
)
988 struct f2fs_checkpoint
*ckpt
= F2FS_CKPT(sbi
);
989 struct curseg_info
*seg_i
;
990 unsigned char *kaddr
;
995 start
= start_sum_block(sbi
);
997 page
= get_meta_page(sbi
, start
++);
998 kaddr
= (unsigned char *)page_address(page
);
1000 /* Step 1: restore nat cache */
1001 seg_i
= CURSEG_I(sbi
, CURSEG_HOT_DATA
);
1002 memcpy(&seg_i
->sum_blk
->n_nats
, kaddr
, SUM_JOURNAL_SIZE
);
1004 /* Step 2: restore sit cache */
1005 seg_i
= CURSEG_I(sbi
, CURSEG_COLD_DATA
);
1006 memcpy(&seg_i
->sum_blk
->n_sits
, kaddr
+ SUM_JOURNAL_SIZE
,
1008 offset
= 2 * SUM_JOURNAL_SIZE
;
1010 /* Step 3: restore summary entries */
1011 for (i
= CURSEG_HOT_DATA
; i
<= CURSEG_COLD_DATA
; i
++) {
1012 unsigned short blk_off
;
1015 seg_i
= CURSEG_I(sbi
, i
);
1016 segno
= le32_to_cpu(ckpt
->cur_data_segno
[i
]);
1017 blk_off
= le16_to_cpu(ckpt
->cur_data_blkoff
[i
]);
1018 seg_i
->next_segno
= segno
;
1019 reset_curseg(sbi
, i
, 0);
1020 seg_i
->alloc_type
= ckpt
->alloc_type
[i
];
1021 seg_i
->next_blkoff
= blk_off
;
1023 if (seg_i
->alloc_type
== SSR
)
1024 blk_off
= sbi
->blocks_per_seg
;
1026 for (j
= 0; j
< blk_off
; j
++) {
1027 struct f2fs_summary
*s
;
1028 s
= (struct f2fs_summary
*)(kaddr
+ offset
);
1029 seg_i
->sum_blk
->entries
[j
] = *s
;
1030 offset
+= SUMMARY_SIZE
;
1031 if (offset
+ SUMMARY_SIZE
<= PAGE_CACHE_SIZE
-
1035 f2fs_put_page(page
, 1);
1038 page
= get_meta_page(sbi
, start
++);
1039 kaddr
= (unsigned char *)page_address(page
);
1043 f2fs_put_page(page
, 1);
1047 static int read_normal_summaries(struct f2fs_sb_info
*sbi
, int type
)
1049 struct f2fs_checkpoint
*ckpt
= F2FS_CKPT(sbi
);
1050 struct f2fs_summary_block
*sum
;
1051 struct curseg_info
*curseg
;
1053 unsigned short blk_off
;
1054 unsigned int segno
= 0;
1055 block_t blk_addr
= 0;
1057 /* get segment number and block addr */
1058 if (IS_DATASEG(type
)) {
1059 segno
= le32_to_cpu(ckpt
->cur_data_segno
[type
]);
1060 blk_off
= le16_to_cpu(ckpt
->cur_data_blkoff
[type
-
1062 if (is_set_ckpt_flags(ckpt
, CP_UMOUNT_FLAG
))
1063 blk_addr
= sum_blk_addr(sbi
, NR_CURSEG_TYPE
, type
);
1065 blk_addr
= sum_blk_addr(sbi
, NR_CURSEG_DATA_TYPE
, type
);
1067 segno
= le32_to_cpu(ckpt
->cur_node_segno
[type
-
1069 blk_off
= le16_to_cpu(ckpt
->cur_node_blkoff
[type
-
1071 if (is_set_ckpt_flags(ckpt
, CP_UMOUNT_FLAG
))
1072 blk_addr
= sum_blk_addr(sbi
, NR_CURSEG_NODE_TYPE
,
1073 type
- CURSEG_HOT_NODE
);
1075 blk_addr
= GET_SUM_BLOCK(sbi
, segno
);
1078 new = get_meta_page(sbi
, blk_addr
);
1079 sum
= (struct f2fs_summary_block
*)page_address(new);
1081 if (IS_NODESEG(type
)) {
1082 if (is_set_ckpt_flags(ckpt
, CP_UMOUNT_FLAG
)) {
1083 struct f2fs_summary
*ns
= &sum
->entries
[0];
1085 for (i
= 0; i
< sbi
->blocks_per_seg
; i
++, ns
++) {
1087 ns
->ofs_in_node
= 0;
1090 if (restore_node_summary(sbi
, segno
, sum
)) {
1091 f2fs_put_page(new, 1);
1097 /* set uncompleted segment to curseg */
1098 curseg
= CURSEG_I(sbi
, type
);
1099 mutex_lock(&curseg
->curseg_mutex
);
1100 memcpy(curseg
->sum_blk
, sum
, PAGE_CACHE_SIZE
);
1101 curseg
->next_segno
= segno
;
1102 reset_curseg(sbi
, type
, 0);
1103 curseg
->alloc_type
= ckpt
->alloc_type
[type
];
1104 curseg
->next_blkoff
= blk_off
;
1105 mutex_unlock(&curseg
->curseg_mutex
);
1106 f2fs_put_page(new, 1);
1110 static int restore_curseg_summaries(struct f2fs_sb_info
*sbi
)
1112 int type
= CURSEG_HOT_DATA
;
1114 if (is_set_ckpt_flags(F2FS_CKPT(sbi
), CP_COMPACT_SUM_FLAG
)) {
1115 /* restore for compacted data summary */
1116 if (read_compacted_summaries(sbi
))
1118 type
= CURSEG_HOT_NODE
;
1121 for (; type
<= CURSEG_COLD_NODE
; type
++)
1122 if (read_normal_summaries(sbi
, type
))
1127 static void write_compacted_summaries(struct f2fs_sb_info
*sbi
, block_t blkaddr
)
1130 unsigned char *kaddr
;
1131 struct f2fs_summary
*summary
;
1132 struct curseg_info
*seg_i
;
1133 int written_size
= 0;
1136 page
= grab_meta_page(sbi
, blkaddr
++);
1137 kaddr
= (unsigned char *)page_address(page
);
1139 /* Step 1: write nat cache */
1140 seg_i
= CURSEG_I(sbi
, CURSEG_HOT_DATA
);
1141 memcpy(kaddr
, &seg_i
->sum_blk
->n_nats
, SUM_JOURNAL_SIZE
);
1142 written_size
+= SUM_JOURNAL_SIZE
;
1144 /* Step 2: write sit cache */
1145 seg_i
= CURSEG_I(sbi
, CURSEG_COLD_DATA
);
1146 memcpy(kaddr
+ written_size
, &seg_i
->sum_blk
->n_sits
,
1148 written_size
+= SUM_JOURNAL_SIZE
;
1150 set_page_dirty(page
);
1152 /* Step 3: write summary entries */
1153 for (i
= CURSEG_HOT_DATA
; i
<= CURSEG_COLD_DATA
; i
++) {
1154 unsigned short blkoff
;
1155 seg_i
= CURSEG_I(sbi
, i
);
1156 if (sbi
->ckpt
->alloc_type
[i
] == SSR
)
1157 blkoff
= sbi
->blocks_per_seg
;
1159 blkoff
= curseg_blkoff(sbi
, i
);
1161 for (j
= 0; j
< blkoff
; j
++) {
1163 page
= grab_meta_page(sbi
, blkaddr
++);
1164 kaddr
= (unsigned char *)page_address(page
);
1167 summary
= (struct f2fs_summary
*)(kaddr
+ written_size
);
1168 *summary
= seg_i
->sum_blk
->entries
[j
];
1169 written_size
+= SUMMARY_SIZE
;
1170 set_page_dirty(page
);
1172 if (written_size
+ SUMMARY_SIZE
<= PAGE_CACHE_SIZE
-
1176 f2fs_put_page(page
, 1);
1181 f2fs_put_page(page
, 1);
1184 static void write_normal_summaries(struct f2fs_sb_info
*sbi
,
1185 block_t blkaddr
, int type
)
1188 if (IS_DATASEG(type
))
1189 end
= type
+ NR_CURSEG_DATA_TYPE
;
1191 end
= type
+ NR_CURSEG_NODE_TYPE
;
1193 for (i
= type
; i
< end
; i
++) {
1194 struct curseg_info
*sum
= CURSEG_I(sbi
, i
);
1195 mutex_lock(&sum
->curseg_mutex
);
1196 write_sum_page(sbi
, sum
->sum_blk
, blkaddr
+ (i
- type
));
1197 mutex_unlock(&sum
->curseg_mutex
);
1201 void write_data_summaries(struct f2fs_sb_info
*sbi
, block_t start_blk
)
1203 if (is_set_ckpt_flags(F2FS_CKPT(sbi
), CP_COMPACT_SUM_FLAG
))
1204 write_compacted_summaries(sbi
, start_blk
);
1206 write_normal_summaries(sbi
, start_blk
, CURSEG_HOT_DATA
);
1209 void write_node_summaries(struct f2fs_sb_info
*sbi
, block_t start_blk
)
1211 if (is_set_ckpt_flags(F2FS_CKPT(sbi
), CP_UMOUNT_FLAG
))
1212 write_normal_summaries(sbi
, start_blk
, CURSEG_HOT_NODE
);
1216 int lookup_journal_in_cursum(struct f2fs_summary_block
*sum
, int type
,
1217 unsigned int val
, int alloc
)
1221 if (type
== NAT_JOURNAL
) {
1222 for (i
= 0; i
< nats_in_cursum(sum
); i
++) {
1223 if (le32_to_cpu(nid_in_journal(sum
, i
)) == val
)
1226 if (alloc
&& nats_in_cursum(sum
) < NAT_JOURNAL_ENTRIES
)
1227 return update_nats_in_cursum(sum
, 1);
1228 } else if (type
== SIT_JOURNAL
) {
1229 for (i
= 0; i
< sits_in_cursum(sum
); i
++)
1230 if (le32_to_cpu(segno_in_journal(sum
, i
)) == val
)
1232 if (alloc
&& sits_in_cursum(sum
) < SIT_JOURNAL_ENTRIES
)
1233 return update_sits_in_cursum(sum
, 1);
1238 static struct page
*get_current_sit_page(struct f2fs_sb_info
*sbi
,
1241 struct sit_info
*sit_i
= SIT_I(sbi
);
1242 unsigned int offset
= SIT_BLOCK_OFFSET(sit_i
, segno
);
1243 block_t blk_addr
= sit_i
->sit_base_addr
+ offset
;
1245 check_seg_range(sbi
, segno
);
1247 /* calculate sit block address */
1248 if (f2fs_test_bit(offset
, sit_i
->sit_bitmap
))
1249 blk_addr
+= sit_i
->sit_blocks
;
1251 return get_meta_page(sbi
, blk_addr
);
1254 static struct page
*get_next_sit_page(struct f2fs_sb_info
*sbi
,
1257 struct sit_info
*sit_i
= SIT_I(sbi
);
1258 struct page
*src_page
, *dst_page
;
1259 pgoff_t src_off
, dst_off
;
1260 void *src_addr
, *dst_addr
;
1262 src_off
= current_sit_addr(sbi
, start
);
1263 dst_off
= next_sit_addr(sbi
, src_off
);
1265 /* get current sit block page without lock */
1266 src_page
= get_meta_page(sbi
, src_off
);
1267 dst_page
= grab_meta_page(sbi
, dst_off
);
1268 BUG_ON(PageDirty(src_page
));
1270 src_addr
= page_address(src_page
);
1271 dst_addr
= page_address(dst_page
);
1272 memcpy(dst_addr
, src_addr
, PAGE_CACHE_SIZE
);
1274 set_page_dirty(dst_page
);
1275 f2fs_put_page(src_page
, 1);
1277 set_to_next_sit(sit_i
, start
);
1282 static bool flush_sits_in_journal(struct f2fs_sb_info
*sbi
)
1284 struct curseg_info
*curseg
= CURSEG_I(sbi
, CURSEG_COLD_DATA
);
1285 struct f2fs_summary_block
*sum
= curseg
->sum_blk
;
1289 * If the journal area in the current summary is full of sit entries,
1290 * all the sit entries will be flushed. Otherwise the sit entries
1291 * are not able to replace with newly hot sit entries.
1293 if (sits_in_cursum(sum
) >= SIT_JOURNAL_ENTRIES
) {
1294 for (i
= sits_in_cursum(sum
) - 1; i
>= 0; i
--) {
1296 segno
= le32_to_cpu(segno_in_journal(sum
, i
));
1297 __mark_sit_entry_dirty(sbi
, segno
);
1299 update_sits_in_cursum(sum
, -sits_in_cursum(sum
));
1306 * CP calls this function, which flushes SIT entries including sit_journal,
1307 * and moves prefree segs to free segs.
1309 void flush_sit_entries(struct f2fs_sb_info
*sbi
)
1311 struct sit_info
*sit_i
= SIT_I(sbi
);
1312 unsigned long *bitmap
= sit_i
->dirty_sentries_bitmap
;
1313 struct curseg_info
*curseg
= CURSEG_I(sbi
, CURSEG_COLD_DATA
);
1314 struct f2fs_summary_block
*sum
= curseg
->sum_blk
;
1315 unsigned long nsegs
= TOTAL_SEGS(sbi
);
1316 struct page
*page
= NULL
;
1317 struct f2fs_sit_block
*raw_sit
= NULL
;
1318 unsigned int start
= 0, end
= 0;
1319 unsigned int segno
= -1;
1322 mutex_lock(&curseg
->curseg_mutex
);
1323 mutex_lock(&sit_i
->sentry_lock
);
1326 * "flushed" indicates whether sit entries in journal are flushed
1327 * to the SIT area or not.
1329 flushed
= flush_sits_in_journal(sbi
);
1331 while ((segno
= find_next_bit(bitmap
, nsegs
, segno
+ 1)) < nsegs
) {
1332 struct seg_entry
*se
= get_seg_entry(sbi
, segno
);
1333 int sit_offset
, offset
;
1335 sit_offset
= SIT_ENTRY_OFFSET(sit_i
, segno
);
1340 offset
= lookup_journal_in_cursum(sum
, SIT_JOURNAL
, segno
, 1);
1342 segno_in_journal(sum
, offset
) = cpu_to_le32(segno
);
1343 seg_info_to_raw_sit(se
, &sit_in_journal(sum
, offset
));
1347 if (!page
|| (start
> segno
) || (segno
> end
)) {
1349 f2fs_put_page(page
, 1);
1353 start
= START_SEGNO(sit_i
, segno
);
1354 end
= start
+ SIT_ENTRY_PER_BLOCK
- 1;
1356 /* read sit block that will be updated */
1357 page
= get_next_sit_page(sbi
, start
);
1358 raw_sit
= page_address(page
);
1361 /* udpate entry in SIT block */
1362 seg_info_to_raw_sit(se
, &raw_sit
->entries
[sit_offset
]);
1364 __clear_bit(segno
, bitmap
);
1365 sit_i
->dirty_sentries
--;
1367 mutex_unlock(&sit_i
->sentry_lock
);
1368 mutex_unlock(&curseg
->curseg_mutex
);
1370 /* writeout last modified SIT block */
1371 f2fs_put_page(page
, 1);
1373 set_prefree_as_free_segments(sbi
);
1376 static int build_sit_info(struct f2fs_sb_info
*sbi
)
1378 struct f2fs_super_block
*raw_super
= F2FS_RAW_SUPER(sbi
);
1379 struct f2fs_checkpoint
*ckpt
= F2FS_CKPT(sbi
);
1380 struct sit_info
*sit_i
;
1381 unsigned int sit_segs
, start
;
1382 char *src_bitmap
, *dst_bitmap
;
1383 unsigned int bitmap_size
;
1385 /* allocate memory for SIT information */
1386 sit_i
= kzalloc(sizeof(struct sit_info
), GFP_KERNEL
);
1390 SM_I(sbi
)->sit_info
= sit_i
;
1392 sit_i
->sentries
= vzalloc(TOTAL_SEGS(sbi
) * sizeof(struct seg_entry
));
1393 if (!sit_i
->sentries
)
1396 bitmap_size
= f2fs_bitmap_size(TOTAL_SEGS(sbi
));
1397 sit_i
->dirty_sentries_bitmap
= kzalloc(bitmap_size
, GFP_KERNEL
);
1398 if (!sit_i
->dirty_sentries_bitmap
)
1401 for (start
= 0; start
< TOTAL_SEGS(sbi
); start
++) {
1402 sit_i
->sentries
[start
].cur_valid_map
1403 = kzalloc(SIT_VBLOCK_MAP_SIZE
, GFP_KERNEL
);
1404 sit_i
->sentries
[start
].ckpt_valid_map
1405 = kzalloc(SIT_VBLOCK_MAP_SIZE
, GFP_KERNEL
);
1406 if (!sit_i
->sentries
[start
].cur_valid_map
1407 || !sit_i
->sentries
[start
].ckpt_valid_map
)
1411 if (sbi
->segs_per_sec
> 1) {
1412 sit_i
->sec_entries
= vzalloc(sbi
->total_sections
*
1413 sizeof(struct sec_entry
));
1414 if (!sit_i
->sec_entries
)
1418 /* get information related with SIT */
1419 sit_segs
= le32_to_cpu(raw_super
->segment_count_sit
) >> 1;
1421 /* setup SIT bitmap from ckeckpoint pack */
1422 bitmap_size
= __bitmap_size(sbi
, SIT_BITMAP
);
1423 src_bitmap
= __bitmap_ptr(sbi
, SIT_BITMAP
);
1425 dst_bitmap
= kzalloc(bitmap_size
, GFP_KERNEL
);
1428 memcpy(dst_bitmap
, src_bitmap
, bitmap_size
);
1430 /* init SIT information */
1431 sit_i
->s_ops
= &default_salloc_ops
;
1433 sit_i
->sit_base_addr
= le32_to_cpu(raw_super
->sit_blkaddr
);
1434 sit_i
->sit_blocks
= sit_segs
<< sbi
->log_blocks_per_seg
;
1435 sit_i
->written_valid_blocks
= le64_to_cpu(ckpt
->valid_block_count
);
1436 sit_i
->sit_bitmap
= dst_bitmap
;
1437 sit_i
->bitmap_size
= bitmap_size
;
1438 sit_i
->dirty_sentries
= 0;
1439 sit_i
->sents_per_block
= SIT_ENTRY_PER_BLOCK
;
1440 sit_i
->elapsed_time
= le64_to_cpu(sbi
->ckpt
->elapsed_time
);
1441 sit_i
->mounted_time
= CURRENT_TIME_SEC
.tv_sec
;
1442 mutex_init(&sit_i
->sentry_lock
);
1446 static int build_free_segmap(struct f2fs_sb_info
*sbi
)
1448 struct f2fs_sm_info
*sm_info
= SM_I(sbi
);
1449 struct free_segmap_info
*free_i
;
1450 unsigned int bitmap_size
, sec_bitmap_size
;
1452 /* allocate memory for free segmap information */
1453 free_i
= kzalloc(sizeof(struct free_segmap_info
), GFP_KERNEL
);
1457 SM_I(sbi
)->free_info
= free_i
;
1459 bitmap_size
= f2fs_bitmap_size(TOTAL_SEGS(sbi
));
1460 free_i
->free_segmap
= kmalloc(bitmap_size
, GFP_KERNEL
);
1461 if (!free_i
->free_segmap
)
1464 sec_bitmap_size
= f2fs_bitmap_size(sbi
->total_sections
);
1465 free_i
->free_secmap
= kmalloc(sec_bitmap_size
, GFP_KERNEL
);
1466 if (!free_i
->free_secmap
)
1469 /* set all segments as dirty temporarily */
1470 memset(free_i
->free_segmap
, 0xff, bitmap_size
);
1471 memset(free_i
->free_secmap
, 0xff, sec_bitmap_size
);
1473 /* init free segmap information */
1474 free_i
->start_segno
=
1475 (unsigned int) GET_SEGNO_FROM_SEG0(sbi
, sm_info
->main_blkaddr
);
1476 free_i
->free_segments
= 0;
1477 free_i
->free_sections
= 0;
1478 rwlock_init(&free_i
->segmap_lock
);
1482 static int build_curseg(struct f2fs_sb_info
*sbi
)
1484 struct curseg_info
*array
;
1487 array
= kzalloc(sizeof(*array
) * NR_CURSEG_TYPE
, GFP_KERNEL
);
1491 SM_I(sbi
)->curseg_array
= array
;
1493 for (i
= 0; i
< NR_CURSEG_TYPE
; i
++) {
1494 mutex_init(&array
[i
].curseg_mutex
);
1495 array
[i
].sum_blk
= kzalloc(PAGE_CACHE_SIZE
, GFP_KERNEL
);
1496 if (!array
[i
].sum_blk
)
1498 array
[i
].segno
= NULL_SEGNO
;
1499 array
[i
].next_blkoff
= 0;
1501 return restore_curseg_summaries(sbi
);
1504 static void build_sit_entries(struct f2fs_sb_info
*sbi
)
1506 struct sit_info
*sit_i
= SIT_I(sbi
);
1507 struct curseg_info
*curseg
= CURSEG_I(sbi
, CURSEG_COLD_DATA
);
1508 struct f2fs_summary_block
*sum
= curseg
->sum_blk
;
1511 for (start
= 0; start
< TOTAL_SEGS(sbi
); start
++) {
1512 struct seg_entry
*se
= &sit_i
->sentries
[start
];
1513 struct f2fs_sit_block
*sit_blk
;
1514 struct f2fs_sit_entry sit
;
1518 mutex_lock(&curseg
->curseg_mutex
);
1519 for (i
= 0; i
< sits_in_cursum(sum
); i
++) {
1520 if (le32_to_cpu(segno_in_journal(sum
, i
)) == start
) {
1521 sit
= sit_in_journal(sum
, i
);
1522 mutex_unlock(&curseg
->curseg_mutex
);
1526 mutex_unlock(&curseg
->curseg_mutex
);
1527 page
= get_current_sit_page(sbi
, start
);
1528 sit_blk
= (struct f2fs_sit_block
*)page_address(page
);
1529 sit
= sit_blk
->entries
[SIT_ENTRY_OFFSET(sit_i
, start
)];
1530 f2fs_put_page(page
, 1);
1532 check_block_count(sbi
, start
, &sit
);
1533 seg_info_from_raw_sit(se
, &sit
);
1534 if (sbi
->segs_per_sec
> 1) {
1535 struct sec_entry
*e
= get_sec_entry(sbi
, start
);
1536 e
->valid_blocks
+= se
->valid_blocks
;
1541 static void init_free_segmap(struct f2fs_sb_info
*sbi
)
1546 for (start
= 0; start
< TOTAL_SEGS(sbi
); start
++) {
1547 struct seg_entry
*sentry
= get_seg_entry(sbi
, start
);
1548 if (!sentry
->valid_blocks
)
1549 __set_free(sbi
, start
);
1552 /* set use the current segments */
1553 for (type
= CURSEG_HOT_DATA
; type
<= CURSEG_COLD_NODE
; type
++) {
1554 struct curseg_info
*curseg_t
= CURSEG_I(sbi
, type
);
1555 __set_test_and_inuse(sbi
, curseg_t
->segno
);
1559 static void init_dirty_segmap(struct f2fs_sb_info
*sbi
)
1561 struct dirty_seglist_info
*dirty_i
= DIRTY_I(sbi
);
1562 struct free_segmap_info
*free_i
= FREE_I(sbi
);
1563 unsigned int segno
= 0, offset
= 0;
1564 unsigned short valid_blocks
;
1566 while (segno
< TOTAL_SEGS(sbi
)) {
1567 /* find dirty segment based on free segmap */
1568 segno
= find_next_inuse(free_i
, TOTAL_SEGS(sbi
), offset
);
1569 if (segno
>= TOTAL_SEGS(sbi
))
1572 valid_blocks
= get_valid_blocks(sbi
, segno
, 0);
1573 if (valid_blocks
>= sbi
->blocks_per_seg
|| !valid_blocks
)
1575 mutex_lock(&dirty_i
->seglist_lock
);
1576 __locate_dirty_segment(sbi
, segno
, DIRTY
);
1577 mutex_unlock(&dirty_i
->seglist_lock
);
1581 static int init_victim_segmap(struct f2fs_sb_info
*sbi
)
1583 struct dirty_seglist_info
*dirty_i
= DIRTY_I(sbi
);
1584 unsigned int bitmap_size
= f2fs_bitmap_size(TOTAL_SEGS(sbi
));
1586 dirty_i
->victim_segmap
[FG_GC
] = kzalloc(bitmap_size
, GFP_KERNEL
);
1587 dirty_i
->victim_segmap
[BG_GC
] = kzalloc(bitmap_size
, GFP_KERNEL
);
1588 if (!dirty_i
->victim_segmap
[FG_GC
] || !dirty_i
->victim_segmap
[BG_GC
])
1593 static int build_dirty_segmap(struct f2fs_sb_info
*sbi
)
1595 struct dirty_seglist_info
*dirty_i
;
1596 unsigned int bitmap_size
, i
;
1598 /* allocate memory for dirty segments list information */
1599 dirty_i
= kzalloc(sizeof(struct dirty_seglist_info
), GFP_KERNEL
);
1603 SM_I(sbi
)->dirty_info
= dirty_i
;
1604 mutex_init(&dirty_i
->seglist_lock
);
1606 bitmap_size
= f2fs_bitmap_size(TOTAL_SEGS(sbi
));
1608 for (i
= 0; i
< NR_DIRTY_TYPE
; i
++) {
1609 dirty_i
->dirty_segmap
[i
] = kzalloc(bitmap_size
, GFP_KERNEL
);
1610 dirty_i
->nr_dirty
[i
] = 0;
1611 if (!dirty_i
->dirty_segmap
[i
])
1615 init_dirty_segmap(sbi
);
1616 return init_victim_segmap(sbi
);
1620 * Update min, max modified time for cost-benefit GC algorithm
1622 static void init_min_max_mtime(struct f2fs_sb_info
*sbi
)
1624 struct sit_info
*sit_i
= SIT_I(sbi
);
1627 mutex_lock(&sit_i
->sentry_lock
);
1629 sit_i
->min_mtime
= LLONG_MAX
;
1631 for (segno
= 0; segno
< TOTAL_SEGS(sbi
); segno
+= sbi
->segs_per_sec
) {
1633 unsigned long long mtime
= 0;
1635 for (i
= 0; i
< sbi
->segs_per_sec
; i
++)
1636 mtime
+= get_seg_entry(sbi
, segno
+ i
)->mtime
;
1638 mtime
= div_u64(mtime
, sbi
->segs_per_sec
);
1640 if (sit_i
->min_mtime
> mtime
)
1641 sit_i
->min_mtime
= mtime
;
1643 sit_i
->max_mtime
= get_mtime(sbi
);
1644 mutex_unlock(&sit_i
->sentry_lock
);
1647 int build_segment_manager(struct f2fs_sb_info
*sbi
)
1649 struct f2fs_super_block
*raw_super
= F2FS_RAW_SUPER(sbi
);
1650 struct f2fs_checkpoint
*ckpt
= F2FS_CKPT(sbi
);
1651 struct f2fs_sm_info
*sm_info
;
1654 sm_info
= kzalloc(sizeof(struct f2fs_sm_info
), GFP_KERNEL
);
1659 sbi
->sm_info
= sm_info
;
1660 INIT_LIST_HEAD(&sm_info
->wblist_head
);
1661 spin_lock_init(&sm_info
->wblist_lock
);
1662 sm_info
->seg0_blkaddr
= le32_to_cpu(raw_super
->segment0_blkaddr
);
1663 sm_info
->main_blkaddr
= le32_to_cpu(raw_super
->main_blkaddr
);
1664 sm_info
->segment_count
= le32_to_cpu(raw_super
->segment_count
);
1665 sm_info
->reserved_segments
= le32_to_cpu(ckpt
->rsvd_segment_count
);
1666 sm_info
->ovp_segments
= le32_to_cpu(ckpt
->overprov_segment_count
);
1667 sm_info
->main_segments
= le32_to_cpu(raw_super
->segment_count_main
);
1668 sm_info
->ssa_blkaddr
= le32_to_cpu(raw_super
->ssa_blkaddr
);
1670 err
= build_sit_info(sbi
);
1673 err
= build_free_segmap(sbi
);
1676 err
= build_curseg(sbi
);
1680 /* reinit free segmap based on SIT */
1681 build_sit_entries(sbi
);
1683 init_free_segmap(sbi
);
1684 err
= build_dirty_segmap(sbi
);
1688 init_min_max_mtime(sbi
);
1692 static void discard_dirty_segmap(struct f2fs_sb_info
*sbi
,
1693 enum dirty_type dirty_type
)
1695 struct dirty_seglist_info
*dirty_i
= DIRTY_I(sbi
);
1697 mutex_lock(&dirty_i
->seglist_lock
);
1698 kfree(dirty_i
->dirty_segmap
[dirty_type
]);
1699 dirty_i
->nr_dirty
[dirty_type
] = 0;
1700 mutex_unlock(&dirty_i
->seglist_lock
);
1703 void reset_victim_segmap(struct f2fs_sb_info
*sbi
)
1705 unsigned int bitmap_size
= f2fs_bitmap_size(TOTAL_SEGS(sbi
));
1706 memset(DIRTY_I(sbi
)->victim_segmap
[FG_GC
], 0, bitmap_size
);
1709 static void destroy_victim_segmap(struct f2fs_sb_info
*sbi
)
1711 struct dirty_seglist_info
*dirty_i
= DIRTY_I(sbi
);
1713 kfree(dirty_i
->victim_segmap
[FG_GC
]);
1714 kfree(dirty_i
->victim_segmap
[BG_GC
]);
1717 static void destroy_dirty_segmap(struct f2fs_sb_info
*sbi
)
1719 struct dirty_seglist_info
*dirty_i
= DIRTY_I(sbi
);
1725 /* discard pre-free/dirty segments list */
1726 for (i
= 0; i
< NR_DIRTY_TYPE
; i
++)
1727 discard_dirty_segmap(sbi
, i
);
1729 destroy_victim_segmap(sbi
);
1730 SM_I(sbi
)->dirty_info
= NULL
;
1734 static void destroy_curseg(struct f2fs_sb_info
*sbi
)
1736 struct curseg_info
*array
= SM_I(sbi
)->curseg_array
;
1741 SM_I(sbi
)->curseg_array
= NULL
;
1742 for (i
= 0; i
< NR_CURSEG_TYPE
; i
++)
1743 kfree(array
[i
].sum_blk
);
1747 static void destroy_free_segmap(struct f2fs_sb_info
*sbi
)
1749 struct free_segmap_info
*free_i
= SM_I(sbi
)->free_info
;
1752 SM_I(sbi
)->free_info
= NULL
;
1753 kfree(free_i
->free_segmap
);
1754 kfree(free_i
->free_secmap
);
1758 static void destroy_sit_info(struct f2fs_sb_info
*sbi
)
1760 struct sit_info
*sit_i
= SIT_I(sbi
);
1766 if (sit_i
->sentries
) {
1767 for (start
= 0; start
< TOTAL_SEGS(sbi
); start
++) {
1768 kfree(sit_i
->sentries
[start
].cur_valid_map
);
1769 kfree(sit_i
->sentries
[start
].ckpt_valid_map
);
1772 vfree(sit_i
->sentries
);
1773 vfree(sit_i
->sec_entries
);
1774 kfree(sit_i
->dirty_sentries_bitmap
);
1776 SM_I(sbi
)->sit_info
= NULL
;
1777 kfree(sit_i
->sit_bitmap
);
1781 void destroy_segment_manager(struct f2fs_sb_info
*sbi
)
1783 struct f2fs_sm_info
*sm_info
= SM_I(sbi
);
1784 destroy_dirty_segmap(sbi
);
1785 destroy_curseg(sbi
);
1786 destroy_free_segmap(sbi
);
1787 destroy_sit_info(sbi
);
1788 sbi
->sm_info
= NULL
;