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/kthread.h>
17 #include <linux/vmalloc.h>
18 #include <linux/swap.h>
24 #include <trace/events/f2fs.h>
26 #define __reverse_ffz(x) __reverse_ffs(~(x))
28 static struct kmem_cache
*discard_entry_slab
;
29 static struct kmem_cache
*sit_entry_set_slab
;
30 static struct kmem_cache
*inmem_entry_slab
;
33 * __reverse_ffs is copied from include/asm-generic/bitops/__ffs.h since
34 * MSB and LSB are reversed in a byte by f2fs_set_bit.
36 static inline unsigned long __reverse_ffs(unsigned long word
)
40 #if BITS_PER_LONG == 64
41 if ((word
& 0xffffffff) == 0) {
46 if ((word
& 0xffff) == 0) {
50 if ((word
& 0xff) == 0) {
54 if ((word
& 0xf0) == 0)
58 if ((word
& 0xc) == 0)
62 if ((word
& 0x2) == 0)
68 * __find_rev_next(_zero)_bit is copied from lib/find_next_bit.c because
69 * f2fs_set_bit makes MSB and LSB reversed in a byte.
72 * f2fs_set_bit(0, bitmap) => 0000 0001
73 * f2fs_set_bit(7, bitmap) => 1000 0000
75 static unsigned long __find_rev_next_bit(const unsigned long *addr
,
76 unsigned long size
, unsigned long offset
)
78 const unsigned long *p
= addr
+ BIT_WORD(offset
);
79 unsigned long result
= offset
& ~(BITS_PER_LONG
- 1);
81 unsigned long mask
, submask
;
82 unsigned long quot
, rest
;
88 offset
%= BITS_PER_LONG
;
93 quot
= (offset
>> 3) << 3;
96 submask
= (unsigned char)(0xff << rest
) >> rest
;
100 if (size
< BITS_PER_LONG
)
105 size
-= BITS_PER_LONG
;
106 result
+= BITS_PER_LONG
;
108 while (size
& ~(BITS_PER_LONG
-1)) {
112 result
+= BITS_PER_LONG
;
113 size
-= BITS_PER_LONG
;
119 tmp
&= (~0UL >> (BITS_PER_LONG
- size
));
120 if (tmp
== 0UL) /* Are any bits set? */
121 return result
+ size
; /* Nope. */
123 return result
+ __reverse_ffs(tmp
);
126 static unsigned long __find_rev_next_zero_bit(const unsigned long *addr
,
127 unsigned long size
, unsigned long offset
)
129 const unsigned long *p
= addr
+ BIT_WORD(offset
);
130 unsigned long result
= offset
& ~(BITS_PER_LONG
- 1);
132 unsigned long mask
, submask
;
133 unsigned long quot
, rest
;
139 offset
%= BITS_PER_LONG
;
144 quot
= (offset
>> 3) << 3;
146 mask
= ~(~0UL << quot
);
147 submask
= (unsigned char)~((unsigned char)(0xff << rest
) >> rest
);
151 if (size
< BITS_PER_LONG
)
156 size
-= BITS_PER_LONG
;
157 result
+= BITS_PER_LONG
;
159 while (size
& ~(BITS_PER_LONG
- 1)) {
163 result
+= BITS_PER_LONG
;
164 size
-= BITS_PER_LONG
;
172 if (tmp
== ~0UL) /* Are any bits zero? */
173 return result
+ size
; /* Nope. */
175 return result
+ __reverse_ffz(tmp
);
178 void register_inmem_page(struct inode
*inode
, struct page
*page
)
180 struct f2fs_inode_info
*fi
= F2FS_I(inode
);
181 struct inmem_pages
*new;
184 SetPagePrivate(page
);
185 f2fs_trace_pid(page
);
187 new = f2fs_kmem_cache_alloc(inmem_entry_slab
, GFP_NOFS
);
189 /* add atomic page indices to the list */
191 INIT_LIST_HEAD(&new->list
);
193 /* increase reference count with clean state */
194 mutex_lock(&fi
->inmem_lock
);
195 err
= radix_tree_insert(&fi
->inmem_root
, page
->index
, new);
196 if (err
== -EEXIST
) {
197 mutex_unlock(&fi
->inmem_lock
);
198 kmem_cache_free(inmem_entry_slab
, new);
201 mutex_unlock(&fi
->inmem_lock
);
205 list_add_tail(&new->list
, &fi
->inmem_pages
);
206 inc_page_count(F2FS_I_SB(inode
), F2FS_INMEM_PAGES
);
207 mutex_unlock(&fi
->inmem_lock
);
210 void commit_inmem_pages(struct inode
*inode
, bool abort
)
212 struct f2fs_sb_info
*sbi
= F2FS_I_SB(inode
);
213 struct f2fs_inode_info
*fi
= F2FS_I(inode
);
214 struct inmem_pages
*cur
, *tmp
;
215 bool submit_bio
= false;
216 struct f2fs_io_info fio
= {
218 .rw
= WRITE_SYNC
| REQ_PRIO
,
222 * The abort is true only when f2fs_evict_inode is called.
223 * Basically, the f2fs_evict_inode doesn't produce any data writes, so
224 * that we don't need to call f2fs_balance_fs.
225 * Otherwise, f2fs_gc in f2fs_balance_fs can wait forever until this
226 * inode becomes free by iget_locked in f2fs_iget.
229 f2fs_balance_fs(sbi
);
233 mutex_lock(&fi
->inmem_lock
);
234 list_for_each_entry_safe(cur
, tmp
, &fi
->inmem_pages
, list
) {
236 lock_page(cur
->page
);
237 if (cur
->page
->mapping
== inode
->i_mapping
) {
238 f2fs_wait_on_page_writeback(cur
->page
, DATA
);
239 if (clear_page_dirty_for_io(cur
->page
))
240 inode_dec_dirty_pages(inode
);
241 do_write_data_page(cur
->page
, &fio
);
244 f2fs_put_page(cur
->page
, 1);
248 radix_tree_delete(&fi
->inmem_root
, cur
->page
->index
);
249 list_del(&cur
->list
);
250 kmem_cache_free(inmem_entry_slab
, cur
);
251 dec_page_count(F2FS_I_SB(inode
), F2FS_INMEM_PAGES
);
253 mutex_unlock(&fi
->inmem_lock
);
258 f2fs_submit_merged_bio(sbi
, DATA
, WRITE
);
263 * This function balances dirty node and dentry pages.
264 * In addition, it controls garbage collection.
266 void f2fs_balance_fs(struct f2fs_sb_info
*sbi
)
269 * We should do GC or end up with checkpoint, if there are so many dirty
270 * dir/node pages without enough free segments.
272 if (has_not_enough_free_secs(sbi
, 0)) {
273 mutex_lock(&sbi
->gc_mutex
);
278 void f2fs_balance_fs_bg(struct f2fs_sb_info
*sbi
)
280 /* check the # of cached NAT entries and prefree segments */
281 if (try_to_free_nats(sbi
, NAT_ENTRY_PER_BLOCK
) ||
282 excess_prefree_segs(sbi
) ||
283 !available_free_memory(sbi
, INO_ENTRIES
))
284 f2fs_sync_fs(sbi
->sb
, true);
287 static int issue_flush_thread(void *data
)
289 struct f2fs_sb_info
*sbi
= data
;
290 struct flush_cmd_control
*fcc
= SM_I(sbi
)->cmd_control_info
;
291 wait_queue_head_t
*q
= &fcc
->flush_wait_queue
;
293 if (kthread_should_stop())
296 if (!llist_empty(&fcc
->issue_list
)) {
297 struct bio
*bio
= bio_alloc(GFP_NOIO
, 0);
298 struct flush_cmd
*cmd
, *next
;
301 fcc
->dispatch_list
= llist_del_all(&fcc
->issue_list
);
302 fcc
->dispatch_list
= llist_reverse_order(fcc
->dispatch_list
);
304 bio
->bi_bdev
= sbi
->sb
->s_bdev
;
305 ret
= submit_bio_wait(WRITE_FLUSH
, bio
);
307 llist_for_each_entry_safe(cmd
, next
,
308 fcc
->dispatch_list
, llnode
) {
310 complete(&cmd
->wait
);
313 fcc
->dispatch_list
= NULL
;
316 wait_event_interruptible(*q
,
317 kthread_should_stop() || !llist_empty(&fcc
->issue_list
));
321 int f2fs_issue_flush(struct f2fs_sb_info
*sbi
)
323 struct flush_cmd_control
*fcc
= SM_I(sbi
)->cmd_control_info
;
324 struct flush_cmd cmd
;
326 trace_f2fs_issue_flush(sbi
->sb
, test_opt(sbi
, NOBARRIER
),
327 test_opt(sbi
, FLUSH_MERGE
));
329 if (test_opt(sbi
, NOBARRIER
))
332 if (!test_opt(sbi
, FLUSH_MERGE
))
333 return blkdev_issue_flush(sbi
->sb
->s_bdev
, GFP_KERNEL
, NULL
);
335 init_completion(&cmd
.wait
);
337 llist_add(&cmd
.llnode
, &fcc
->issue_list
);
339 if (!fcc
->dispatch_list
)
340 wake_up(&fcc
->flush_wait_queue
);
342 wait_for_completion(&cmd
.wait
);
347 int create_flush_cmd_control(struct f2fs_sb_info
*sbi
)
349 dev_t dev
= sbi
->sb
->s_bdev
->bd_dev
;
350 struct flush_cmd_control
*fcc
;
353 fcc
= kzalloc(sizeof(struct flush_cmd_control
), GFP_KERNEL
);
356 init_waitqueue_head(&fcc
->flush_wait_queue
);
357 init_llist_head(&fcc
->issue_list
);
358 SM_I(sbi
)->cmd_control_info
= fcc
;
359 fcc
->f2fs_issue_flush
= kthread_run(issue_flush_thread
, sbi
,
360 "f2fs_flush-%u:%u", MAJOR(dev
), MINOR(dev
));
361 if (IS_ERR(fcc
->f2fs_issue_flush
)) {
362 err
= PTR_ERR(fcc
->f2fs_issue_flush
);
364 SM_I(sbi
)->cmd_control_info
= NULL
;
371 void destroy_flush_cmd_control(struct f2fs_sb_info
*sbi
)
373 struct flush_cmd_control
*fcc
= SM_I(sbi
)->cmd_control_info
;
375 if (fcc
&& fcc
->f2fs_issue_flush
)
376 kthread_stop(fcc
->f2fs_issue_flush
);
378 SM_I(sbi
)->cmd_control_info
= NULL
;
381 static void __locate_dirty_segment(struct f2fs_sb_info
*sbi
, unsigned int segno
,
382 enum dirty_type dirty_type
)
384 struct dirty_seglist_info
*dirty_i
= DIRTY_I(sbi
);
386 /* need not be added */
387 if (IS_CURSEG(sbi
, segno
))
390 if (!test_and_set_bit(segno
, dirty_i
->dirty_segmap
[dirty_type
]))
391 dirty_i
->nr_dirty
[dirty_type
]++;
393 if (dirty_type
== DIRTY
) {
394 struct seg_entry
*sentry
= get_seg_entry(sbi
, segno
);
395 enum dirty_type t
= sentry
->type
;
397 if (unlikely(t
>= DIRTY
)) {
401 if (!test_and_set_bit(segno
, dirty_i
->dirty_segmap
[t
]))
402 dirty_i
->nr_dirty
[t
]++;
406 static void __remove_dirty_segment(struct f2fs_sb_info
*sbi
, unsigned int segno
,
407 enum dirty_type dirty_type
)
409 struct dirty_seglist_info
*dirty_i
= DIRTY_I(sbi
);
411 if (test_and_clear_bit(segno
, dirty_i
->dirty_segmap
[dirty_type
]))
412 dirty_i
->nr_dirty
[dirty_type
]--;
414 if (dirty_type
== DIRTY
) {
415 struct seg_entry
*sentry
= get_seg_entry(sbi
, segno
);
416 enum dirty_type t
= sentry
->type
;
418 if (test_and_clear_bit(segno
, dirty_i
->dirty_segmap
[t
]))
419 dirty_i
->nr_dirty
[t
]--;
421 if (get_valid_blocks(sbi
, segno
, sbi
->segs_per_sec
) == 0)
422 clear_bit(GET_SECNO(sbi
, segno
),
423 dirty_i
->victim_secmap
);
428 * Should not occur error such as -ENOMEM.
429 * Adding dirty entry into seglist is not critical operation.
430 * If a given segment is one of current working segments, it won't be added.
432 static void locate_dirty_segment(struct f2fs_sb_info
*sbi
, unsigned int segno
)
434 struct dirty_seglist_info
*dirty_i
= DIRTY_I(sbi
);
435 unsigned short valid_blocks
;
437 if (segno
== NULL_SEGNO
|| IS_CURSEG(sbi
, segno
))
440 mutex_lock(&dirty_i
->seglist_lock
);
442 valid_blocks
= get_valid_blocks(sbi
, segno
, 0);
444 if (valid_blocks
== 0) {
445 __locate_dirty_segment(sbi
, segno
, PRE
);
446 __remove_dirty_segment(sbi
, segno
, DIRTY
);
447 } else if (valid_blocks
< sbi
->blocks_per_seg
) {
448 __locate_dirty_segment(sbi
, segno
, DIRTY
);
450 /* Recovery routine with SSR needs this */
451 __remove_dirty_segment(sbi
, segno
, DIRTY
);
454 mutex_unlock(&dirty_i
->seglist_lock
);
457 static int f2fs_issue_discard(struct f2fs_sb_info
*sbi
,
458 block_t blkstart
, block_t blklen
)
460 sector_t start
= SECTOR_FROM_BLOCK(blkstart
);
461 sector_t len
= SECTOR_FROM_BLOCK(blklen
);
462 trace_f2fs_issue_discard(sbi
->sb
, blkstart
, blklen
);
463 return blkdev_issue_discard(sbi
->sb
->s_bdev
, start
, len
, GFP_NOFS
, 0);
466 void discard_next_dnode(struct f2fs_sb_info
*sbi
, block_t blkaddr
)
468 if (f2fs_issue_discard(sbi
, blkaddr
, 1)) {
469 struct page
*page
= grab_meta_page(sbi
, blkaddr
);
470 /* zero-filled page */
471 set_page_dirty(page
);
472 f2fs_put_page(page
, 1);
476 static void __add_discard_entry(struct f2fs_sb_info
*sbi
,
477 struct cp_control
*cpc
, unsigned int start
, unsigned int end
)
479 struct list_head
*head
= &SM_I(sbi
)->discard_list
;
480 struct discard_entry
*new, *last
;
482 if (!list_empty(head
)) {
483 last
= list_last_entry(head
, struct discard_entry
, list
);
484 if (START_BLOCK(sbi
, cpc
->trim_start
) + start
==
485 last
->blkaddr
+ last
->len
) {
486 last
->len
+= end
- start
;
491 new = f2fs_kmem_cache_alloc(discard_entry_slab
, GFP_NOFS
);
492 INIT_LIST_HEAD(&new->list
);
493 new->blkaddr
= START_BLOCK(sbi
, cpc
->trim_start
) + start
;
494 new->len
= end
- start
;
495 list_add_tail(&new->list
, head
);
497 SM_I(sbi
)->nr_discards
+= end
- start
;
498 cpc
->trimmed
+= end
- start
;
501 static void add_discard_addrs(struct f2fs_sb_info
*sbi
, struct cp_control
*cpc
)
503 int entries
= SIT_VBLOCK_MAP_SIZE
/ sizeof(unsigned long);
504 int max_blocks
= sbi
->blocks_per_seg
;
505 struct seg_entry
*se
= get_seg_entry(sbi
, cpc
->trim_start
);
506 unsigned long *cur_map
= (unsigned long *)se
->cur_valid_map
;
507 unsigned long *ckpt_map
= (unsigned long *)se
->ckpt_valid_map
;
508 unsigned long dmap
[entries
];
509 unsigned int start
= 0, end
= -1;
510 bool force
= (cpc
->reason
== CP_DISCARD
);
513 if (!force
&& (!test_opt(sbi
, DISCARD
) ||
514 SM_I(sbi
)->nr_discards
>= SM_I(sbi
)->max_discards
))
517 if (force
&& !se
->valid_blocks
) {
518 struct dirty_seglist_info
*dirty_i
= DIRTY_I(sbi
);
520 * if this segment is registered in the prefree list, then
521 * we should skip adding a discard candidate, and let the
522 * checkpoint do that later.
524 mutex_lock(&dirty_i
->seglist_lock
);
525 if (test_bit(cpc
->trim_start
, dirty_i
->dirty_segmap
[PRE
])) {
526 mutex_unlock(&dirty_i
->seglist_lock
);
527 cpc
->trimmed
+= sbi
->blocks_per_seg
;
530 mutex_unlock(&dirty_i
->seglist_lock
);
532 __add_discard_entry(sbi
, cpc
, 0, sbi
->blocks_per_seg
);
536 /* zero block will be discarded through the prefree list */
537 if (!se
->valid_blocks
|| se
->valid_blocks
== max_blocks
)
540 /* SIT_VBLOCK_MAP_SIZE should be multiple of sizeof(unsigned long) */
541 for (i
= 0; i
< entries
; i
++)
542 dmap
[i
] = force
? ~ckpt_map
[i
] :
543 (cur_map
[i
] ^ ckpt_map
[i
]) & ckpt_map
[i
];
545 while (force
|| SM_I(sbi
)->nr_discards
<= SM_I(sbi
)->max_discards
) {
546 start
= __find_rev_next_bit(dmap
, max_blocks
, end
+ 1);
547 if (start
>= max_blocks
)
550 end
= __find_rev_next_zero_bit(dmap
, max_blocks
, start
+ 1);
552 if (end
- start
< cpc
->trim_minlen
)
555 __add_discard_entry(sbi
, cpc
, start
, end
);
559 void release_discard_addrs(struct f2fs_sb_info
*sbi
)
561 struct list_head
*head
= &(SM_I(sbi
)->discard_list
);
562 struct discard_entry
*entry
, *this;
565 list_for_each_entry_safe(entry
, this, head
, list
) {
566 list_del(&entry
->list
);
567 kmem_cache_free(discard_entry_slab
, entry
);
572 * Should call clear_prefree_segments after checkpoint is done.
574 static void set_prefree_as_free_segments(struct f2fs_sb_info
*sbi
)
576 struct dirty_seglist_info
*dirty_i
= DIRTY_I(sbi
);
579 mutex_lock(&dirty_i
->seglist_lock
);
580 for_each_set_bit(segno
, dirty_i
->dirty_segmap
[PRE
], MAIN_SEGS(sbi
))
581 __set_test_and_free(sbi
, segno
);
582 mutex_unlock(&dirty_i
->seglist_lock
);
585 void clear_prefree_segments(struct f2fs_sb_info
*sbi
)
587 struct list_head
*head
= &(SM_I(sbi
)->discard_list
);
588 struct discard_entry
*entry
, *this;
589 struct dirty_seglist_info
*dirty_i
= DIRTY_I(sbi
);
590 unsigned long *prefree_map
= dirty_i
->dirty_segmap
[PRE
];
591 unsigned int start
= 0, end
= -1;
593 mutex_lock(&dirty_i
->seglist_lock
);
597 start
= find_next_bit(prefree_map
, MAIN_SEGS(sbi
), end
+ 1);
598 if (start
>= MAIN_SEGS(sbi
))
600 end
= find_next_zero_bit(prefree_map
, MAIN_SEGS(sbi
),
603 for (i
= start
; i
< end
; i
++)
604 clear_bit(i
, prefree_map
);
606 dirty_i
->nr_dirty
[PRE
] -= end
- start
;
608 if (!test_opt(sbi
, DISCARD
))
611 f2fs_issue_discard(sbi
, START_BLOCK(sbi
, start
),
612 (end
- start
) << sbi
->log_blocks_per_seg
);
614 mutex_unlock(&dirty_i
->seglist_lock
);
616 /* send small discards */
617 list_for_each_entry_safe(entry
, this, head
, list
) {
618 f2fs_issue_discard(sbi
, entry
->blkaddr
, entry
->len
);
619 list_del(&entry
->list
);
620 SM_I(sbi
)->nr_discards
-= entry
->len
;
621 kmem_cache_free(discard_entry_slab
, entry
);
625 static bool __mark_sit_entry_dirty(struct f2fs_sb_info
*sbi
, unsigned int segno
)
627 struct sit_info
*sit_i
= SIT_I(sbi
);
629 if (!__test_and_set_bit(segno
, sit_i
->dirty_sentries_bitmap
)) {
630 sit_i
->dirty_sentries
++;
637 static void __set_sit_entry_type(struct f2fs_sb_info
*sbi
, int type
,
638 unsigned int segno
, int modified
)
640 struct seg_entry
*se
= get_seg_entry(sbi
, segno
);
643 __mark_sit_entry_dirty(sbi
, segno
);
646 static void update_sit_entry(struct f2fs_sb_info
*sbi
, block_t blkaddr
, int del
)
648 struct seg_entry
*se
;
649 unsigned int segno
, offset
;
650 long int new_vblocks
;
652 segno
= GET_SEGNO(sbi
, blkaddr
);
654 se
= get_seg_entry(sbi
, segno
);
655 new_vblocks
= se
->valid_blocks
+ del
;
656 offset
= GET_BLKOFF_FROM_SEG0(sbi
, blkaddr
);
658 f2fs_bug_on(sbi
, (new_vblocks
>> (sizeof(unsigned short) << 3) ||
659 (new_vblocks
> sbi
->blocks_per_seg
)));
661 se
->valid_blocks
= new_vblocks
;
662 se
->mtime
= get_mtime(sbi
);
663 SIT_I(sbi
)->max_mtime
= se
->mtime
;
665 /* Update valid block bitmap */
667 if (f2fs_test_and_set_bit(offset
, se
->cur_valid_map
))
670 if (!f2fs_test_and_clear_bit(offset
, se
->cur_valid_map
))
673 if (!f2fs_test_bit(offset
, se
->ckpt_valid_map
))
674 se
->ckpt_valid_blocks
+= del
;
676 __mark_sit_entry_dirty(sbi
, segno
);
678 /* update total number of valid blocks to be written in ckpt area */
679 SIT_I(sbi
)->written_valid_blocks
+= del
;
681 if (sbi
->segs_per_sec
> 1)
682 get_sec_entry(sbi
, segno
)->valid_blocks
+= del
;
685 void refresh_sit_entry(struct f2fs_sb_info
*sbi
, block_t old
, block_t
new)
687 update_sit_entry(sbi
, new, 1);
688 if (GET_SEGNO(sbi
, old
) != NULL_SEGNO
)
689 update_sit_entry(sbi
, old
, -1);
691 locate_dirty_segment(sbi
, GET_SEGNO(sbi
, old
));
692 locate_dirty_segment(sbi
, GET_SEGNO(sbi
, new));
695 void invalidate_blocks(struct f2fs_sb_info
*sbi
, block_t addr
)
697 unsigned int segno
= GET_SEGNO(sbi
, addr
);
698 struct sit_info
*sit_i
= SIT_I(sbi
);
700 f2fs_bug_on(sbi
, addr
== NULL_ADDR
);
701 if (addr
== NEW_ADDR
)
704 /* add it into sit main buffer */
705 mutex_lock(&sit_i
->sentry_lock
);
707 update_sit_entry(sbi
, addr
, -1);
709 /* add it into dirty seglist */
710 locate_dirty_segment(sbi
, segno
);
712 mutex_unlock(&sit_i
->sentry_lock
);
716 * This function should be resided under the curseg_mutex lock
718 static void __add_sum_entry(struct f2fs_sb_info
*sbi
, int type
,
719 struct f2fs_summary
*sum
)
721 struct curseg_info
*curseg
= CURSEG_I(sbi
, type
);
722 void *addr
= curseg
->sum_blk
;
723 addr
+= curseg
->next_blkoff
* sizeof(struct f2fs_summary
);
724 memcpy(addr
, sum
, sizeof(struct f2fs_summary
));
728 * Calculate the number of current summary pages for writing
730 int npages_for_summary_flush(struct f2fs_sb_info
*sbi
, bool for_ra
)
732 int valid_sum_count
= 0;
735 for (i
= CURSEG_HOT_DATA
; i
<= CURSEG_COLD_DATA
; i
++) {
736 if (sbi
->ckpt
->alloc_type
[i
] == SSR
)
737 valid_sum_count
+= sbi
->blocks_per_seg
;
740 valid_sum_count
+= le16_to_cpu(
741 F2FS_CKPT(sbi
)->cur_data_blkoff
[i
]);
743 valid_sum_count
+= curseg_blkoff(sbi
, i
);
747 sum_in_page
= (PAGE_CACHE_SIZE
- 2 * SUM_JOURNAL_SIZE
-
748 SUM_FOOTER_SIZE
) / SUMMARY_SIZE
;
749 if (valid_sum_count
<= sum_in_page
)
751 else if ((valid_sum_count
- sum_in_page
) <=
752 (PAGE_CACHE_SIZE
- SUM_FOOTER_SIZE
) / SUMMARY_SIZE
)
758 * Caller should put this summary page
760 struct page
*get_sum_page(struct f2fs_sb_info
*sbi
, unsigned int segno
)
762 return get_meta_page(sbi
, GET_SUM_BLOCK(sbi
, segno
));
765 static void write_sum_page(struct f2fs_sb_info
*sbi
,
766 struct f2fs_summary_block
*sum_blk
, block_t blk_addr
)
768 struct page
*page
= grab_meta_page(sbi
, blk_addr
);
769 void *kaddr
= page_address(page
);
770 memcpy(kaddr
, sum_blk
, PAGE_CACHE_SIZE
);
771 set_page_dirty(page
);
772 f2fs_put_page(page
, 1);
775 static int is_next_segment_free(struct f2fs_sb_info
*sbi
, int type
)
777 struct curseg_info
*curseg
= CURSEG_I(sbi
, type
);
778 unsigned int segno
= curseg
->segno
+ 1;
779 struct free_segmap_info
*free_i
= FREE_I(sbi
);
781 if (segno
< MAIN_SEGS(sbi
) && segno
% sbi
->segs_per_sec
)
782 return !test_bit(segno
, free_i
->free_segmap
);
787 * Find a new segment from the free segments bitmap to right order
788 * This function should be returned with success, otherwise BUG
790 static void get_new_segment(struct f2fs_sb_info
*sbi
,
791 unsigned int *newseg
, bool new_sec
, int dir
)
793 struct free_segmap_info
*free_i
= FREE_I(sbi
);
794 unsigned int segno
, secno
, zoneno
;
795 unsigned int total_zones
= MAIN_SECS(sbi
) / sbi
->secs_per_zone
;
796 unsigned int hint
= *newseg
/ sbi
->segs_per_sec
;
797 unsigned int old_zoneno
= GET_ZONENO_FROM_SEGNO(sbi
, *newseg
);
798 unsigned int left_start
= hint
;
803 write_lock(&free_i
->segmap_lock
);
805 if (!new_sec
&& ((*newseg
+ 1) % sbi
->segs_per_sec
)) {
806 segno
= find_next_zero_bit(free_i
->free_segmap
,
807 MAIN_SEGS(sbi
), *newseg
+ 1);
808 if (segno
- *newseg
< sbi
->segs_per_sec
-
809 (*newseg
% sbi
->segs_per_sec
))
813 secno
= find_next_zero_bit(free_i
->free_secmap
, MAIN_SECS(sbi
), hint
);
814 if (secno
>= MAIN_SECS(sbi
)) {
815 if (dir
== ALLOC_RIGHT
) {
816 secno
= find_next_zero_bit(free_i
->free_secmap
,
818 f2fs_bug_on(sbi
, secno
>= MAIN_SECS(sbi
));
821 left_start
= hint
- 1;
827 while (test_bit(left_start
, free_i
->free_secmap
)) {
828 if (left_start
> 0) {
832 left_start
= find_next_zero_bit(free_i
->free_secmap
,
834 f2fs_bug_on(sbi
, left_start
>= MAIN_SECS(sbi
));
840 segno
= secno
* sbi
->segs_per_sec
;
841 zoneno
= secno
/ sbi
->secs_per_zone
;
843 /* give up on finding another zone */
846 if (sbi
->secs_per_zone
== 1)
848 if (zoneno
== old_zoneno
)
850 if (dir
== ALLOC_LEFT
) {
851 if (!go_left
&& zoneno
+ 1 >= total_zones
)
853 if (go_left
&& zoneno
== 0)
856 for (i
= 0; i
< NR_CURSEG_TYPE
; i
++)
857 if (CURSEG_I(sbi
, i
)->zone
== zoneno
)
860 if (i
< NR_CURSEG_TYPE
) {
861 /* zone is in user, try another */
863 hint
= zoneno
* sbi
->secs_per_zone
- 1;
864 else if (zoneno
+ 1 >= total_zones
)
867 hint
= (zoneno
+ 1) * sbi
->secs_per_zone
;
869 goto find_other_zone
;
872 /* set it as dirty segment in free segmap */
873 f2fs_bug_on(sbi
, test_bit(segno
, free_i
->free_segmap
));
874 __set_inuse(sbi
, segno
);
876 write_unlock(&free_i
->segmap_lock
);
879 static void reset_curseg(struct f2fs_sb_info
*sbi
, int type
, int modified
)
881 struct curseg_info
*curseg
= CURSEG_I(sbi
, type
);
882 struct summary_footer
*sum_footer
;
884 curseg
->segno
= curseg
->next_segno
;
885 curseg
->zone
= GET_ZONENO_FROM_SEGNO(sbi
, curseg
->segno
);
886 curseg
->next_blkoff
= 0;
887 curseg
->next_segno
= NULL_SEGNO
;
889 sum_footer
= &(curseg
->sum_blk
->footer
);
890 memset(sum_footer
, 0, sizeof(struct summary_footer
));
891 if (IS_DATASEG(type
))
892 SET_SUM_TYPE(sum_footer
, SUM_TYPE_DATA
);
893 if (IS_NODESEG(type
))
894 SET_SUM_TYPE(sum_footer
, SUM_TYPE_NODE
);
895 __set_sit_entry_type(sbi
, type
, curseg
->segno
, modified
);
899 * Allocate a current working segment.
900 * This function always allocates a free segment in LFS manner.
902 static void new_curseg(struct f2fs_sb_info
*sbi
, int type
, bool new_sec
)
904 struct curseg_info
*curseg
= CURSEG_I(sbi
, type
);
905 unsigned int segno
= curseg
->segno
;
906 int dir
= ALLOC_LEFT
;
908 write_sum_page(sbi
, curseg
->sum_blk
,
909 GET_SUM_BLOCK(sbi
, segno
));
910 if (type
== CURSEG_WARM_DATA
|| type
== CURSEG_COLD_DATA
)
913 if (test_opt(sbi
, NOHEAP
))
916 get_new_segment(sbi
, &segno
, new_sec
, dir
);
917 curseg
->next_segno
= segno
;
918 reset_curseg(sbi
, type
, 1);
919 curseg
->alloc_type
= LFS
;
922 static void __next_free_blkoff(struct f2fs_sb_info
*sbi
,
923 struct curseg_info
*seg
, block_t start
)
925 struct seg_entry
*se
= get_seg_entry(sbi
, seg
->segno
);
926 int entries
= SIT_VBLOCK_MAP_SIZE
/ sizeof(unsigned long);
927 unsigned long target_map
[entries
];
928 unsigned long *ckpt_map
= (unsigned long *)se
->ckpt_valid_map
;
929 unsigned long *cur_map
= (unsigned long *)se
->cur_valid_map
;
932 for (i
= 0; i
< entries
; i
++)
933 target_map
[i
] = ckpt_map
[i
] | cur_map
[i
];
935 pos
= __find_rev_next_zero_bit(target_map
, sbi
->blocks_per_seg
, start
);
937 seg
->next_blkoff
= pos
;
941 * If a segment is written by LFS manner, next block offset is just obtained
942 * by increasing the current block offset. However, if a segment is written by
943 * SSR manner, next block offset obtained by calling __next_free_blkoff
945 static void __refresh_next_blkoff(struct f2fs_sb_info
*sbi
,
946 struct curseg_info
*seg
)
948 if (seg
->alloc_type
== SSR
)
949 __next_free_blkoff(sbi
, seg
, seg
->next_blkoff
+ 1);
955 * This function always allocates a used segment(from dirty seglist) by SSR
956 * manner, so it should recover the existing segment information of valid blocks
958 static void change_curseg(struct f2fs_sb_info
*sbi
, int type
, bool reuse
)
960 struct dirty_seglist_info
*dirty_i
= DIRTY_I(sbi
);
961 struct curseg_info
*curseg
= CURSEG_I(sbi
, type
);
962 unsigned int new_segno
= curseg
->next_segno
;
963 struct f2fs_summary_block
*sum_node
;
964 struct page
*sum_page
;
966 write_sum_page(sbi
, curseg
->sum_blk
,
967 GET_SUM_BLOCK(sbi
, curseg
->segno
));
968 __set_test_and_inuse(sbi
, new_segno
);
970 mutex_lock(&dirty_i
->seglist_lock
);
971 __remove_dirty_segment(sbi
, new_segno
, PRE
);
972 __remove_dirty_segment(sbi
, new_segno
, DIRTY
);
973 mutex_unlock(&dirty_i
->seglist_lock
);
975 reset_curseg(sbi
, type
, 1);
976 curseg
->alloc_type
= SSR
;
977 __next_free_blkoff(sbi
, curseg
, 0);
980 sum_page
= get_sum_page(sbi
, new_segno
);
981 sum_node
= (struct f2fs_summary_block
*)page_address(sum_page
);
982 memcpy(curseg
->sum_blk
, sum_node
, SUM_ENTRY_SIZE
);
983 f2fs_put_page(sum_page
, 1);
987 static int get_ssr_segment(struct f2fs_sb_info
*sbi
, int type
)
989 struct curseg_info
*curseg
= CURSEG_I(sbi
, type
);
990 const struct victim_selection
*v_ops
= DIRTY_I(sbi
)->v_ops
;
992 if (IS_NODESEG(type
) || !has_not_enough_free_secs(sbi
, 0))
993 return v_ops
->get_victim(sbi
,
994 &(curseg
)->next_segno
, BG_GC
, type
, SSR
);
996 /* For data segments, let's do SSR more intensively */
997 for (; type
>= CURSEG_HOT_DATA
; type
--)
998 if (v_ops
->get_victim(sbi
, &(curseg
)->next_segno
,
1005 * flush out current segment and replace it with new segment
1006 * This function should be returned with success, otherwise BUG
1008 static void allocate_segment_by_default(struct f2fs_sb_info
*sbi
,
1009 int type
, bool force
)
1011 struct curseg_info
*curseg
= CURSEG_I(sbi
, type
);
1014 new_curseg(sbi
, type
, true);
1015 else if (type
== CURSEG_WARM_NODE
)
1016 new_curseg(sbi
, type
, false);
1017 else if (curseg
->alloc_type
== LFS
&& is_next_segment_free(sbi
, type
))
1018 new_curseg(sbi
, type
, false);
1019 else if (need_SSR(sbi
) && get_ssr_segment(sbi
, type
))
1020 change_curseg(sbi
, type
, true);
1022 new_curseg(sbi
, type
, false);
1024 stat_inc_seg_type(sbi
, curseg
);
1027 static void __allocate_new_segments(struct f2fs_sb_info
*sbi
, int type
)
1029 struct curseg_info
*curseg
= CURSEG_I(sbi
, type
);
1030 unsigned int old_segno
;
1032 old_segno
= curseg
->segno
;
1033 SIT_I(sbi
)->s_ops
->allocate_segment(sbi
, type
, true);
1034 locate_dirty_segment(sbi
, old_segno
);
1037 void allocate_new_segments(struct f2fs_sb_info
*sbi
)
1041 for (i
= CURSEG_HOT_DATA
; i
<= CURSEG_COLD_DATA
; i
++)
1042 __allocate_new_segments(sbi
, i
);
1045 static const struct segment_allocation default_salloc_ops
= {
1046 .allocate_segment
= allocate_segment_by_default
,
1049 int f2fs_trim_fs(struct f2fs_sb_info
*sbi
, struct fstrim_range
*range
)
1051 __u64 start
= range
->start
>> sbi
->log_blocksize
;
1052 __u64 end
= start
+ (range
->len
>> sbi
->log_blocksize
) - 1;
1053 unsigned int start_segno
, end_segno
;
1054 struct cp_control cpc
;
1056 if (range
->minlen
> SEGMENT_SIZE(sbi
) || start
>= MAX_BLKADDR(sbi
) ||
1057 range
->len
< sbi
->blocksize
)
1061 if (end
<= MAIN_BLKADDR(sbi
))
1064 /* start/end segment number in main_area */
1065 start_segno
= (start
<= MAIN_BLKADDR(sbi
)) ? 0 : GET_SEGNO(sbi
, start
);
1066 end_segno
= (end
>= MAX_BLKADDR(sbi
)) ? MAIN_SEGS(sbi
) - 1 :
1067 GET_SEGNO(sbi
, end
);
1068 cpc
.reason
= CP_DISCARD
;
1069 cpc
.trim_minlen
= range
->minlen
>> sbi
->log_blocksize
;
1071 /* do checkpoint to issue discard commands safely */
1072 for (; start_segno
<= end_segno
; start_segno
= cpc
.trim_end
+ 1) {
1073 cpc
.trim_start
= start_segno
;
1074 cpc
.trim_end
= min_t(unsigned int, rounddown(start_segno
+
1075 BATCHED_TRIM_SEGMENTS(sbi
),
1076 sbi
->segs_per_sec
) - 1, end_segno
);
1078 mutex_lock(&sbi
->gc_mutex
);
1079 write_checkpoint(sbi
, &cpc
);
1080 mutex_unlock(&sbi
->gc_mutex
);
1083 range
->len
= cpc
.trimmed
<< sbi
->log_blocksize
;
1087 static bool __has_curseg_space(struct f2fs_sb_info
*sbi
, int type
)
1089 struct curseg_info
*curseg
= CURSEG_I(sbi
, type
);
1090 if (curseg
->next_blkoff
< sbi
->blocks_per_seg
)
1095 static int __get_segment_type_2(struct page
*page
, enum page_type p_type
)
1098 return CURSEG_HOT_DATA
;
1100 return CURSEG_HOT_NODE
;
1103 static int __get_segment_type_4(struct page
*page
, enum page_type p_type
)
1105 if (p_type
== DATA
) {
1106 struct inode
*inode
= page
->mapping
->host
;
1108 if (S_ISDIR(inode
->i_mode
))
1109 return CURSEG_HOT_DATA
;
1111 return CURSEG_COLD_DATA
;
1113 if (IS_DNODE(page
) && is_cold_node(page
))
1114 return CURSEG_WARM_NODE
;
1116 return CURSEG_COLD_NODE
;
1120 static int __get_segment_type_6(struct page
*page
, enum page_type p_type
)
1122 if (p_type
== DATA
) {
1123 struct inode
*inode
= page
->mapping
->host
;
1125 if (S_ISDIR(inode
->i_mode
))
1126 return CURSEG_HOT_DATA
;
1127 else if (is_cold_data(page
) || file_is_cold(inode
))
1128 return CURSEG_COLD_DATA
;
1130 return CURSEG_WARM_DATA
;
1133 return is_cold_node(page
) ? CURSEG_WARM_NODE
:
1136 return CURSEG_COLD_NODE
;
1140 static int __get_segment_type(struct page
*page
, enum page_type p_type
)
1142 switch (F2FS_P_SB(page
)->active_logs
) {
1144 return __get_segment_type_2(page
, p_type
);
1146 return __get_segment_type_4(page
, p_type
);
1148 /* NR_CURSEG_TYPE(6) logs by default */
1149 f2fs_bug_on(F2FS_P_SB(page
),
1150 F2FS_P_SB(page
)->active_logs
!= NR_CURSEG_TYPE
);
1151 return __get_segment_type_6(page
, p_type
);
1154 void allocate_data_block(struct f2fs_sb_info
*sbi
, struct page
*page
,
1155 block_t old_blkaddr
, block_t
*new_blkaddr
,
1156 struct f2fs_summary
*sum
, int type
)
1158 struct sit_info
*sit_i
= SIT_I(sbi
);
1159 struct curseg_info
*curseg
;
1160 bool direct_io
= (type
== CURSEG_DIRECT_IO
);
1162 type
= direct_io
? CURSEG_WARM_DATA
: type
;
1164 curseg
= CURSEG_I(sbi
, type
);
1166 mutex_lock(&curseg
->curseg_mutex
);
1168 /* direct_io'ed data is aligned to the segment for better performance */
1169 if (direct_io
&& curseg
->next_blkoff
)
1170 __allocate_new_segments(sbi
, type
);
1172 *new_blkaddr
= NEXT_FREE_BLKADDR(sbi
, curseg
);
1175 * __add_sum_entry should be resided under the curseg_mutex
1176 * because, this function updates a summary entry in the
1177 * current summary block.
1179 __add_sum_entry(sbi
, type
, sum
);
1181 mutex_lock(&sit_i
->sentry_lock
);
1182 __refresh_next_blkoff(sbi
, curseg
);
1184 stat_inc_block_count(sbi
, curseg
);
1186 if (!__has_curseg_space(sbi
, type
))
1187 sit_i
->s_ops
->allocate_segment(sbi
, type
, false);
1189 * SIT information should be updated before segment allocation,
1190 * since SSR needs latest valid block information.
1192 refresh_sit_entry(sbi
, old_blkaddr
, *new_blkaddr
);
1194 mutex_unlock(&sit_i
->sentry_lock
);
1196 if (page
&& IS_NODESEG(type
))
1197 fill_node_footer_blkaddr(page
, NEXT_FREE_BLKADDR(sbi
, curseg
));
1199 mutex_unlock(&curseg
->curseg_mutex
);
1202 static void do_write_page(struct f2fs_sb_info
*sbi
, struct page
*page
,
1203 struct f2fs_summary
*sum
,
1204 struct f2fs_io_info
*fio
)
1206 int type
= __get_segment_type(page
, fio
->type
);
1208 allocate_data_block(sbi
, page
, fio
->blk_addr
, &fio
->blk_addr
, sum
, type
);
1210 /* writeout dirty page into bdev */
1211 f2fs_submit_page_mbio(sbi
, page
, fio
);
1214 void write_meta_page(struct f2fs_sb_info
*sbi
, struct page
*page
)
1216 struct f2fs_io_info fio
= {
1218 .rw
= WRITE_SYNC
| REQ_META
| REQ_PRIO
,
1219 .blk_addr
= page
->index
,
1222 set_page_writeback(page
);
1223 f2fs_submit_page_mbio(sbi
, page
, &fio
);
1226 void write_node_page(struct f2fs_sb_info
*sbi
, struct page
*page
,
1227 unsigned int nid
, struct f2fs_io_info
*fio
)
1229 struct f2fs_summary sum
;
1230 set_summary(&sum
, nid
, 0, 0);
1231 do_write_page(sbi
, page
, &sum
, fio
);
1234 void write_data_page(struct page
*page
, struct dnode_of_data
*dn
,
1235 struct f2fs_io_info
*fio
)
1237 struct f2fs_sb_info
*sbi
= F2FS_I_SB(dn
->inode
);
1238 struct f2fs_summary sum
;
1239 struct node_info ni
;
1241 f2fs_bug_on(sbi
, dn
->data_blkaddr
== NULL_ADDR
);
1242 get_node_info(sbi
, dn
->nid
, &ni
);
1243 set_summary(&sum
, dn
->nid
, dn
->ofs_in_node
, ni
.version
);
1244 do_write_page(sbi
, page
, &sum
, fio
);
1245 dn
->data_blkaddr
= fio
->blk_addr
;
1248 void rewrite_data_page(struct page
*page
, struct f2fs_io_info
*fio
)
1250 stat_inc_inplace_blocks(F2FS_P_SB(page
));
1251 f2fs_submit_page_mbio(F2FS_P_SB(page
), page
, fio
);
1254 void recover_data_page(struct f2fs_sb_info
*sbi
,
1255 struct page
*page
, struct f2fs_summary
*sum
,
1256 block_t old_blkaddr
, block_t new_blkaddr
)
1258 struct sit_info
*sit_i
= SIT_I(sbi
);
1259 struct curseg_info
*curseg
;
1260 unsigned int segno
, old_cursegno
;
1261 struct seg_entry
*se
;
1264 segno
= GET_SEGNO(sbi
, new_blkaddr
);
1265 se
= get_seg_entry(sbi
, segno
);
1268 if (se
->valid_blocks
== 0 && !IS_CURSEG(sbi
, segno
)) {
1269 if (old_blkaddr
== NULL_ADDR
)
1270 type
= CURSEG_COLD_DATA
;
1272 type
= CURSEG_WARM_DATA
;
1274 curseg
= CURSEG_I(sbi
, type
);
1276 mutex_lock(&curseg
->curseg_mutex
);
1277 mutex_lock(&sit_i
->sentry_lock
);
1279 old_cursegno
= curseg
->segno
;
1281 /* change the current segment */
1282 if (segno
!= curseg
->segno
) {
1283 curseg
->next_segno
= segno
;
1284 change_curseg(sbi
, type
, true);
1287 curseg
->next_blkoff
= GET_BLKOFF_FROM_SEG0(sbi
, new_blkaddr
);
1288 __add_sum_entry(sbi
, type
, sum
);
1290 refresh_sit_entry(sbi
, old_blkaddr
, new_blkaddr
);
1291 locate_dirty_segment(sbi
, old_cursegno
);
1293 mutex_unlock(&sit_i
->sentry_lock
);
1294 mutex_unlock(&curseg
->curseg_mutex
);
1297 static inline bool is_merged_page(struct f2fs_sb_info
*sbi
,
1298 struct page
*page
, enum page_type type
)
1300 enum page_type btype
= PAGE_TYPE_OF_BIO(type
);
1301 struct f2fs_bio_info
*io
= &sbi
->write_io
[btype
];
1302 struct bio_vec
*bvec
;
1305 down_read(&io
->io_rwsem
);
1309 bio_for_each_segment_all(bvec
, io
->bio
, i
) {
1310 if (page
== bvec
->bv_page
) {
1311 up_read(&io
->io_rwsem
);
1317 up_read(&io
->io_rwsem
);
1321 void f2fs_wait_on_page_writeback(struct page
*page
,
1322 enum page_type type
)
1324 if (PageWriteback(page
)) {
1325 struct f2fs_sb_info
*sbi
= F2FS_P_SB(page
);
1327 if (is_merged_page(sbi
, page
, type
))
1328 f2fs_submit_merged_bio(sbi
, type
, WRITE
);
1329 wait_on_page_writeback(page
);
1333 static int read_compacted_summaries(struct f2fs_sb_info
*sbi
)
1335 struct f2fs_checkpoint
*ckpt
= F2FS_CKPT(sbi
);
1336 struct curseg_info
*seg_i
;
1337 unsigned char *kaddr
;
1342 start
= start_sum_block(sbi
);
1344 page
= get_meta_page(sbi
, start
++);
1345 kaddr
= (unsigned char *)page_address(page
);
1347 /* Step 1: restore nat cache */
1348 seg_i
= CURSEG_I(sbi
, CURSEG_HOT_DATA
);
1349 memcpy(&seg_i
->sum_blk
->n_nats
, kaddr
, SUM_JOURNAL_SIZE
);
1351 /* Step 2: restore sit cache */
1352 seg_i
= CURSEG_I(sbi
, CURSEG_COLD_DATA
);
1353 memcpy(&seg_i
->sum_blk
->n_sits
, kaddr
+ SUM_JOURNAL_SIZE
,
1355 offset
= 2 * SUM_JOURNAL_SIZE
;
1357 /* Step 3: restore summary entries */
1358 for (i
= CURSEG_HOT_DATA
; i
<= CURSEG_COLD_DATA
; i
++) {
1359 unsigned short blk_off
;
1362 seg_i
= CURSEG_I(sbi
, i
);
1363 segno
= le32_to_cpu(ckpt
->cur_data_segno
[i
]);
1364 blk_off
= le16_to_cpu(ckpt
->cur_data_blkoff
[i
]);
1365 seg_i
->next_segno
= segno
;
1366 reset_curseg(sbi
, i
, 0);
1367 seg_i
->alloc_type
= ckpt
->alloc_type
[i
];
1368 seg_i
->next_blkoff
= blk_off
;
1370 if (seg_i
->alloc_type
== SSR
)
1371 blk_off
= sbi
->blocks_per_seg
;
1373 for (j
= 0; j
< blk_off
; j
++) {
1374 struct f2fs_summary
*s
;
1375 s
= (struct f2fs_summary
*)(kaddr
+ offset
);
1376 seg_i
->sum_blk
->entries
[j
] = *s
;
1377 offset
+= SUMMARY_SIZE
;
1378 if (offset
+ SUMMARY_SIZE
<= PAGE_CACHE_SIZE
-
1382 f2fs_put_page(page
, 1);
1385 page
= get_meta_page(sbi
, start
++);
1386 kaddr
= (unsigned char *)page_address(page
);
1390 f2fs_put_page(page
, 1);
1394 static int read_normal_summaries(struct f2fs_sb_info
*sbi
, int type
)
1396 struct f2fs_checkpoint
*ckpt
= F2FS_CKPT(sbi
);
1397 struct f2fs_summary_block
*sum
;
1398 struct curseg_info
*curseg
;
1400 unsigned short blk_off
;
1401 unsigned int segno
= 0;
1402 block_t blk_addr
= 0;
1404 /* get segment number and block addr */
1405 if (IS_DATASEG(type
)) {
1406 segno
= le32_to_cpu(ckpt
->cur_data_segno
[type
]);
1407 blk_off
= le16_to_cpu(ckpt
->cur_data_blkoff
[type
-
1409 if (__exist_node_summaries(sbi
))
1410 blk_addr
= sum_blk_addr(sbi
, NR_CURSEG_TYPE
, type
);
1412 blk_addr
= sum_blk_addr(sbi
, NR_CURSEG_DATA_TYPE
, type
);
1414 segno
= le32_to_cpu(ckpt
->cur_node_segno
[type
-
1416 blk_off
= le16_to_cpu(ckpt
->cur_node_blkoff
[type
-
1418 if (__exist_node_summaries(sbi
))
1419 blk_addr
= sum_blk_addr(sbi
, NR_CURSEG_NODE_TYPE
,
1420 type
- CURSEG_HOT_NODE
);
1422 blk_addr
= GET_SUM_BLOCK(sbi
, segno
);
1425 new = get_meta_page(sbi
, blk_addr
);
1426 sum
= (struct f2fs_summary_block
*)page_address(new);
1428 if (IS_NODESEG(type
)) {
1429 if (__exist_node_summaries(sbi
)) {
1430 struct f2fs_summary
*ns
= &sum
->entries
[0];
1432 for (i
= 0; i
< sbi
->blocks_per_seg
; i
++, ns
++) {
1434 ns
->ofs_in_node
= 0;
1439 err
= restore_node_summary(sbi
, segno
, sum
);
1441 f2fs_put_page(new, 1);
1447 /* set uncompleted segment to curseg */
1448 curseg
= CURSEG_I(sbi
, type
);
1449 mutex_lock(&curseg
->curseg_mutex
);
1450 memcpy(curseg
->sum_blk
, sum
, PAGE_CACHE_SIZE
);
1451 curseg
->next_segno
= segno
;
1452 reset_curseg(sbi
, type
, 0);
1453 curseg
->alloc_type
= ckpt
->alloc_type
[type
];
1454 curseg
->next_blkoff
= blk_off
;
1455 mutex_unlock(&curseg
->curseg_mutex
);
1456 f2fs_put_page(new, 1);
1460 static int restore_curseg_summaries(struct f2fs_sb_info
*sbi
)
1462 int type
= CURSEG_HOT_DATA
;
1465 if (is_set_ckpt_flags(F2FS_CKPT(sbi
), CP_COMPACT_SUM_FLAG
)) {
1466 int npages
= npages_for_summary_flush(sbi
, true);
1469 ra_meta_pages(sbi
, start_sum_block(sbi
), npages
,
1472 /* restore for compacted data summary */
1473 if (read_compacted_summaries(sbi
))
1475 type
= CURSEG_HOT_NODE
;
1478 if (__exist_node_summaries(sbi
))
1479 ra_meta_pages(sbi
, sum_blk_addr(sbi
, NR_CURSEG_TYPE
, type
),
1480 NR_CURSEG_TYPE
- type
, META_CP
);
1482 for (; type
<= CURSEG_COLD_NODE
; type
++) {
1483 err
= read_normal_summaries(sbi
, type
);
1491 static void write_compacted_summaries(struct f2fs_sb_info
*sbi
, block_t blkaddr
)
1494 unsigned char *kaddr
;
1495 struct f2fs_summary
*summary
;
1496 struct curseg_info
*seg_i
;
1497 int written_size
= 0;
1500 page
= grab_meta_page(sbi
, blkaddr
++);
1501 kaddr
= (unsigned char *)page_address(page
);
1503 /* Step 1: write nat cache */
1504 seg_i
= CURSEG_I(sbi
, CURSEG_HOT_DATA
);
1505 memcpy(kaddr
, &seg_i
->sum_blk
->n_nats
, SUM_JOURNAL_SIZE
);
1506 written_size
+= SUM_JOURNAL_SIZE
;
1508 /* Step 2: write sit cache */
1509 seg_i
= CURSEG_I(sbi
, CURSEG_COLD_DATA
);
1510 memcpy(kaddr
+ written_size
, &seg_i
->sum_blk
->n_sits
,
1512 written_size
+= SUM_JOURNAL_SIZE
;
1514 /* Step 3: write summary entries */
1515 for (i
= CURSEG_HOT_DATA
; i
<= CURSEG_COLD_DATA
; i
++) {
1516 unsigned short blkoff
;
1517 seg_i
= CURSEG_I(sbi
, i
);
1518 if (sbi
->ckpt
->alloc_type
[i
] == SSR
)
1519 blkoff
= sbi
->blocks_per_seg
;
1521 blkoff
= curseg_blkoff(sbi
, i
);
1523 for (j
= 0; j
< blkoff
; j
++) {
1525 page
= grab_meta_page(sbi
, blkaddr
++);
1526 kaddr
= (unsigned char *)page_address(page
);
1529 summary
= (struct f2fs_summary
*)(kaddr
+ written_size
);
1530 *summary
= seg_i
->sum_blk
->entries
[j
];
1531 written_size
+= SUMMARY_SIZE
;
1533 if (written_size
+ SUMMARY_SIZE
<= PAGE_CACHE_SIZE
-
1537 set_page_dirty(page
);
1538 f2fs_put_page(page
, 1);
1543 set_page_dirty(page
);
1544 f2fs_put_page(page
, 1);
1548 static void write_normal_summaries(struct f2fs_sb_info
*sbi
,
1549 block_t blkaddr
, int type
)
1552 if (IS_DATASEG(type
))
1553 end
= type
+ NR_CURSEG_DATA_TYPE
;
1555 end
= type
+ NR_CURSEG_NODE_TYPE
;
1557 for (i
= type
; i
< end
; i
++) {
1558 struct curseg_info
*sum
= CURSEG_I(sbi
, i
);
1559 mutex_lock(&sum
->curseg_mutex
);
1560 write_sum_page(sbi
, sum
->sum_blk
, blkaddr
+ (i
- type
));
1561 mutex_unlock(&sum
->curseg_mutex
);
1565 void write_data_summaries(struct f2fs_sb_info
*sbi
, block_t start_blk
)
1567 if (is_set_ckpt_flags(F2FS_CKPT(sbi
), CP_COMPACT_SUM_FLAG
))
1568 write_compacted_summaries(sbi
, start_blk
);
1570 write_normal_summaries(sbi
, start_blk
, CURSEG_HOT_DATA
);
1573 void write_node_summaries(struct f2fs_sb_info
*sbi
, block_t start_blk
)
1575 write_normal_summaries(sbi
, start_blk
, CURSEG_HOT_NODE
);
1578 int lookup_journal_in_cursum(struct f2fs_summary_block
*sum
, int type
,
1579 unsigned int val
, int alloc
)
1583 if (type
== NAT_JOURNAL
) {
1584 for (i
= 0; i
< nats_in_cursum(sum
); i
++) {
1585 if (le32_to_cpu(nid_in_journal(sum
, i
)) == val
)
1588 if (alloc
&& nats_in_cursum(sum
) < NAT_JOURNAL_ENTRIES
)
1589 return update_nats_in_cursum(sum
, 1);
1590 } else if (type
== SIT_JOURNAL
) {
1591 for (i
= 0; i
< sits_in_cursum(sum
); i
++)
1592 if (le32_to_cpu(segno_in_journal(sum
, i
)) == val
)
1594 if (alloc
&& sits_in_cursum(sum
) < SIT_JOURNAL_ENTRIES
)
1595 return update_sits_in_cursum(sum
, 1);
1600 static struct page
*get_current_sit_page(struct f2fs_sb_info
*sbi
,
1603 return get_meta_page(sbi
, current_sit_addr(sbi
, segno
));
1606 static struct page
*get_next_sit_page(struct f2fs_sb_info
*sbi
,
1609 struct sit_info
*sit_i
= SIT_I(sbi
);
1610 struct page
*src_page
, *dst_page
;
1611 pgoff_t src_off
, dst_off
;
1612 void *src_addr
, *dst_addr
;
1614 src_off
= current_sit_addr(sbi
, start
);
1615 dst_off
= next_sit_addr(sbi
, src_off
);
1617 /* get current sit block page without lock */
1618 src_page
= get_meta_page(sbi
, src_off
);
1619 dst_page
= grab_meta_page(sbi
, dst_off
);
1620 f2fs_bug_on(sbi
, PageDirty(src_page
));
1622 src_addr
= page_address(src_page
);
1623 dst_addr
= page_address(dst_page
);
1624 memcpy(dst_addr
, src_addr
, PAGE_CACHE_SIZE
);
1626 set_page_dirty(dst_page
);
1627 f2fs_put_page(src_page
, 1);
1629 set_to_next_sit(sit_i
, start
);
1634 static struct sit_entry_set
*grab_sit_entry_set(void)
1636 struct sit_entry_set
*ses
=
1637 f2fs_kmem_cache_alloc(sit_entry_set_slab
, GFP_ATOMIC
);
1640 INIT_LIST_HEAD(&ses
->set_list
);
1644 static void release_sit_entry_set(struct sit_entry_set
*ses
)
1646 list_del(&ses
->set_list
);
1647 kmem_cache_free(sit_entry_set_slab
, ses
);
1650 static void adjust_sit_entry_set(struct sit_entry_set
*ses
,
1651 struct list_head
*head
)
1653 struct sit_entry_set
*next
= ses
;
1655 if (list_is_last(&ses
->set_list
, head
))
1658 list_for_each_entry_continue(next
, head
, set_list
)
1659 if (ses
->entry_cnt
<= next
->entry_cnt
)
1662 list_move_tail(&ses
->set_list
, &next
->set_list
);
1665 static void add_sit_entry(unsigned int segno
, struct list_head
*head
)
1667 struct sit_entry_set
*ses
;
1668 unsigned int start_segno
= START_SEGNO(segno
);
1670 list_for_each_entry(ses
, head
, set_list
) {
1671 if (ses
->start_segno
== start_segno
) {
1673 adjust_sit_entry_set(ses
, head
);
1678 ses
= grab_sit_entry_set();
1680 ses
->start_segno
= start_segno
;
1682 list_add(&ses
->set_list
, head
);
1685 static void add_sits_in_set(struct f2fs_sb_info
*sbi
)
1687 struct f2fs_sm_info
*sm_info
= SM_I(sbi
);
1688 struct list_head
*set_list
= &sm_info
->sit_entry_set
;
1689 unsigned long *bitmap
= SIT_I(sbi
)->dirty_sentries_bitmap
;
1692 for_each_set_bit(segno
, bitmap
, MAIN_SEGS(sbi
))
1693 add_sit_entry(segno
, set_list
);
1696 static void remove_sits_in_journal(struct f2fs_sb_info
*sbi
)
1698 struct curseg_info
*curseg
= CURSEG_I(sbi
, CURSEG_COLD_DATA
);
1699 struct f2fs_summary_block
*sum
= curseg
->sum_blk
;
1702 for (i
= sits_in_cursum(sum
) - 1; i
>= 0; i
--) {
1706 segno
= le32_to_cpu(segno_in_journal(sum
, i
));
1707 dirtied
= __mark_sit_entry_dirty(sbi
, segno
);
1710 add_sit_entry(segno
, &SM_I(sbi
)->sit_entry_set
);
1712 update_sits_in_cursum(sum
, -sits_in_cursum(sum
));
1716 * CP calls this function, which flushes SIT entries including sit_journal,
1717 * and moves prefree segs to free segs.
1719 void flush_sit_entries(struct f2fs_sb_info
*sbi
, struct cp_control
*cpc
)
1721 struct sit_info
*sit_i
= SIT_I(sbi
);
1722 unsigned long *bitmap
= sit_i
->dirty_sentries_bitmap
;
1723 struct curseg_info
*curseg
= CURSEG_I(sbi
, CURSEG_COLD_DATA
);
1724 struct f2fs_summary_block
*sum
= curseg
->sum_blk
;
1725 struct sit_entry_set
*ses
, *tmp
;
1726 struct list_head
*head
= &SM_I(sbi
)->sit_entry_set
;
1727 bool to_journal
= true;
1728 struct seg_entry
*se
;
1730 mutex_lock(&curseg
->curseg_mutex
);
1731 mutex_lock(&sit_i
->sentry_lock
);
1734 * add and account sit entries of dirty bitmap in sit entry
1737 add_sits_in_set(sbi
);
1740 * if there are no enough space in journal to store dirty sit
1741 * entries, remove all entries from journal and add and account
1742 * them in sit entry set.
1744 if (!__has_cursum_space(sum
, sit_i
->dirty_sentries
, SIT_JOURNAL
))
1745 remove_sits_in_journal(sbi
);
1747 if (!sit_i
->dirty_sentries
)
1751 * there are two steps to flush sit entries:
1752 * #1, flush sit entries to journal in current cold data summary block.
1753 * #2, flush sit entries to sit page.
1755 list_for_each_entry_safe(ses
, tmp
, head
, set_list
) {
1756 struct page
*page
= NULL
;
1757 struct f2fs_sit_block
*raw_sit
= NULL
;
1758 unsigned int start_segno
= ses
->start_segno
;
1759 unsigned int end
= min(start_segno
+ SIT_ENTRY_PER_BLOCK
,
1760 (unsigned long)MAIN_SEGS(sbi
));
1761 unsigned int segno
= start_segno
;
1764 !__has_cursum_space(sum
, ses
->entry_cnt
, SIT_JOURNAL
))
1768 page
= get_next_sit_page(sbi
, start_segno
);
1769 raw_sit
= page_address(page
);
1772 /* flush dirty sit entries in region of current sit set */
1773 for_each_set_bit_from(segno
, bitmap
, end
) {
1774 int offset
, sit_offset
;
1776 se
= get_seg_entry(sbi
, segno
);
1778 /* add discard candidates */
1779 if (cpc
->reason
!= CP_DISCARD
) {
1780 cpc
->trim_start
= segno
;
1781 add_discard_addrs(sbi
, cpc
);
1785 offset
= lookup_journal_in_cursum(sum
,
1786 SIT_JOURNAL
, segno
, 1);
1787 f2fs_bug_on(sbi
, offset
< 0);
1788 segno_in_journal(sum
, offset
) =
1790 seg_info_to_raw_sit(se
,
1791 &sit_in_journal(sum
, offset
));
1793 sit_offset
= SIT_ENTRY_OFFSET(sit_i
, segno
);
1794 seg_info_to_raw_sit(se
,
1795 &raw_sit
->entries
[sit_offset
]);
1798 __clear_bit(segno
, bitmap
);
1799 sit_i
->dirty_sentries
--;
1804 f2fs_put_page(page
, 1);
1806 f2fs_bug_on(sbi
, ses
->entry_cnt
);
1807 release_sit_entry_set(ses
);
1810 f2fs_bug_on(sbi
, !list_empty(head
));
1811 f2fs_bug_on(sbi
, sit_i
->dirty_sentries
);
1813 if (cpc
->reason
== CP_DISCARD
) {
1814 for (; cpc
->trim_start
<= cpc
->trim_end
; cpc
->trim_start
++)
1815 add_discard_addrs(sbi
, cpc
);
1817 mutex_unlock(&sit_i
->sentry_lock
);
1818 mutex_unlock(&curseg
->curseg_mutex
);
1820 set_prefree_as_free_segments(sbi
);
1823 static int build_sit_info(struct f2fs_sb_info
*sbi
)
1825 struct f2fs_super_block
*raw_super
= F2FS_RAW_SUPER(sbi
);
1826 struct f2fs_checkpoint
*ckpt
= F2FS_CKPT(sbi
);
1827 struct sit_info
*sit_i
;
1828 unsigned int sit_segs
, start
;
1829 char *src_bitmap
, *dst_bitmap
;
1830 unsigned int bitmap_size
;
1832 /* allocate memory for SIT information */
1833 sit_i
= kzalloc(sizeof(struct sit_info
), GFP_KERNEL
);
1837 SM_I(sbi
)->sit_info
= sit_i
;
1839 sit_i
->sentries
= vzalloc(MAIN_SEGS(sbi
) * sizeof(struct seg_entry
));
1840 if (!sit_i
->sentries
)
1843 bitmap_size
= f2fs_bitmap_size(MAIN_SEGS(sbi
));
1844 sit_i
->dirty_sentries_bitmap
= kzalloc(bitmap_size
, GFP_KERNEL
);
1845 if (!sit_i
->dirty_sentries_bitmap
)
1848 for (start
= 0; start
< MAIN_SEGS(sbi
); start
++) {
1849 sit_i
->sentries
[start
].cur_valid_map
1850 = kzalloc(SIT_VBLOCK_MAP_SIZE
, GFP_KERNEL
);
1851 sit_i
->sentries
[start
].ckpt_valid_map
1852 = kzalloc(SIT_VBLOCK_MAP_SIZE
, GFP_KERNEL
);
1853 if (!sit_i
->sentries
[start
].cur_valid_map
1854 || !sit_i
->sentries
[start
].ckpt_valid_map
)
1858 if (sbi
->segs_per_sec
> 1) {
1859 sit_i
->sec_entries
= vzalloc(MAIN_SECS(sbi
) *
1860 sizeof(struct sec_entry
));
1861 if (!sit_i
->sec_entries
)
1865 /* get information related with SIT */
1866 sit_segs
= le32_to_cpu(raw_super
->segment_count_sit
) >> 1;
1868 /* setup SIT bitmap from ckeckpoint pack */
1869 bitmap_size
= __bitmap_size(sbi
, SIT_BITMAP
);
1870 src_bitmap
= __bitmap_ptr(sbi
, SIT_BITMAP
);
1872 dst_bitmap
= kmemdup(src_bitmap
, bitmap_size
, GFP_KERNEL
);
1876 /* init SIT information */
1877 sit_i
->s_ops
= &default_salloc_ops
;
1879 sit_i
->sit_base_addr
= le32_to_cpu(raw_super
->sit_blkaddr
);
1880 sit_i
->sit_blocks
= sit_segs
<< sbi
->log_blocks_per_seg
;
1881 sit_i
->written_valid_blocks
= le64_to_cpu(ckpt
->valid_block_count
);
1882 sit_i
->sit_bitmap
= dst_bitmap
;
1883 sit_i
->bitmap_size
= bitmap_size
;
1884 sit_i
->dirty_sentries
= 0;
1885 sit_i
->sents_per_block
= SIT_ENTRY_PER_BLOCK
;
1886 sit_i
->elapsed_time
= le64_to_cpu(sbi
->ckpt
->elapsed_time
);
1887 sit_i
->mounted_time
= CURRENT_TIME_SEC
.tv_sec
;
1888 mutex_init(&sit_i
->sentry_lock
);
1892 static int build_free_segmap(struct f2fs_sb_info
*sbi
)
1894 struct free_segmap_info
*free_i
;
1895 unsigned int bitmap_size
, sec_bitmap_size
;
1897 /* allocate memory for free segmap information */
1898 free_i
= kzalloc(sizeof(struct free_segmap_info
), GFP_KERNEL
);
1902 SM_I(sbi
)->free_info
= free_i
;
1904 bitmap_size
= f2fs_bitmap_size(MAIN_SEGS(sbi
));
1905 free_i
->free_segmap
= kmalloc(bitmap_size
, GFP_KERNEL
);
1906 if (!free_i
->free_segmap
)
1909 sec_bitmap_size
= f2fs_bitmap_size(MAIN_SECS(sbi
));
1910 free_i
->free_secmap
= kmalloc(sec_bitmap_size
, GFP_KERNEL
);
1911 if (!free_i
->free_secmap
)
1914 /* set all segments as dirty temporarily */
1915 memset(free_i
->free_segmap
, 0xff, bitmap_size
);
1916 memset(free_i
->free_secmap
, 0xff, sec_bitmap_size
);
1918 /* init free segmap information */
1919 free_i
->start_segno
= GET_SEGNO_FROM_SEG0(sbi
, MAIN_BLKADDR(sbi
));
1920 free_i
->free_segments
= 0;
1921 free_i
->free_sections
= 0;
1922 rwlock_init(&free_i
->segmap_lock
);
1926 static int build_curseg(struct f2fs_sb_info
*sbi
)
1928 struct curseg_info
*array
;
1931 array
= kcalloc(NR_CURSEG_TYPE
, sizeof(*array
), GFP_KERNEL
);
1935 SM_I(sbi
)->curseg_array
= array
;
1937 for (i
= 0; i
< NR_CURSEG_TYPE
; i
++) {
1938 mutex_init(&array
[i
].curseg_mutex
);
1939 array
[i
].sum_blk
= kzalloc(PAGE_CACHE_SIZE
, GFP_KERNEL
);
1940 if (!array
[i
].sum_blk
)
1942 array
[i
].segno
= NULL_SEGNO
;
1943 array
[i
].next_blkoff
= 0;
1945 return restore_curseg_summaries(sbi
);
1948 static void build_sit_entries(struct f2fs_sb_info
*sbi
)
1950 struct sit_info
*sit_i
= SIT_I(sbi
);
1951 struct curseg_info
*curseg
= CURSEG_I(sbi
, CURSEG_COLD_DATA
);
1952 struct f2fs_summary_block
*sum
= curseg
->sum_blk
;
1953 int sit_blk_cnt
= SIT_BLK_CNT(sbi
);
1954 unsigned int i
, start
, end
;
1955 unsigned int readed
, start_blk
= 0;
1956 int nrpages
= MAX_BIO_BLOCKS(sbi
);
1959 readed
= ra_meta_pages(sbi
, start_blk
, nrpages
, META_SIT
);
1961 start
= start_blk
* sit_i
->sents_per_block
;
1962 end
= (start_blk
+ readed
) * sit_i
->sents_per_block
;
1964 for (; start
< end
&& start
< MAIN_SEGS(sbi
); start
++) {
1965 struct seg_entry
*se
= &sit_i
->sentries
[start
];
1966 struct f2fs_sit_block
*sit_blk
;
1967 struct f2fs_sit_entry sit
;
1970 mutex_lock(&curseg
->curseg_mutex
);
1971 for (i
= 0; i
< sits_in_cursum(sum
); i
++) {
1972 if (le32_to_cpu(segno_in_journal(sum
, i
))
1974 sit
= sit_in_journal(sum
, i
);
1975 mutex_unlock(&curseg
->curseg_mutex
);
1979 mutex_unlock(&curseg
->curseg_mutex
);
1981 page
= get_current_sit_page(sbi
, start
);
1982 sit_blk
= (struct f2fs_sit_block
*)page_address(page
);
1983 sit
= sit_blk
->entries
[SIT_ENTRY_OFFSET(sit_i
, start
)];
1984 f2fs_put_page(page
, 1);
1986 check_block_count(sbi
, start
, &sit
);
1987 seg_info_from_raw_sit(se
, &sit
);
1988 if (sbi
->segs_per_sec
> 1) {
1989 struct sec_entry
*e
= get_sec_entry(sbi
, start
);
1990 e
->valid_blocks
+= se
->valid_blocks
;
1993 start_blk
+= readed
;
1994 } while (start_blk
< sit_blk_cnt
);
1997 static void init_free_segmap(struct f2fs_sb_info
*sbi
)
2002 for (start
= 0; start
< MAIN_SEGS(sbi
); start
++) {
2003 struct seg_entry
*sentry
= get_seg_entry(sbi
, start
);
2004 if (!sentry
->valid_blocks
)
2005 __set_free(sbi
, start
);
2008 /* set use the current segments */
2009 for (type
= CURSEG_HOT_DATA
; type
<= CURSEG_COLD_NODE
; type
++) {
2010 struct curseg_info
*curseg_t
= CURSEG_I(sbi
, type
);
2011 __set_test_and_inuse(sbi
, curseg_t
->segno
);
2015 static void init_dirty_segmap(struct f2fs_sb_info
*sbi
)
2017 struct dirty_seglist_info
*dirty_i
= DIRTY_I(sbi
);
2018 struct free_segmap_info
*free_i
= FREE_I(sbi
);
2019 unsigned int segno
= 0, offset
= 0;
2020 unsigned short valid_blocks
;
2023 /* find dirty segment based on free segmap */
2024 segno
= find_next_inuse(free_i
, MAIN_SEGS(sbi
), offset
);
2025 if (segno
>= MAIN_SEGS(sbi
))
2028 valid_blocks
= get_valid_blocks(sbi
, segno
, 0);
2029 if (valid_blocks
== sbi
->blocks_per_seg
|| !valid_blocks
)
2031 if (valid_blocks
> sbi
->blocks_per_seg
) {
2032 f2fs_bug_on(sbi
, 1);
2035 mutex_lock(&dirty_i
->seglist_lock
);
2036 __locate_dirty_segment(sbi
, segno
, DIRTY
);
2037 mutex_unlock(&dirty_i
->seglist_lock
);
2041 static int init_victim_secmap(struct f2fs_sb_info
*sbi
)
2043 struct dirty_seglist_info
*dirty_i
= DIRTY_I(sbi
);
2044 unsigned int bitmap_size
= f2fs_bitmap_size(MAIN_SECS(sbi
));
2046 dirty_i
->victim_secmap
= kzalloc(bitmap_size
, GFP_KERNEL
);
2047 if (!dirty_i
->victim_secmap
)
2052 static int build_dirty_segmap(struct f2fs_sb_info
*sbi
)
2054 struct dirty_seglist_info
*dirty_i
;
2055 unsigned int bitmap_size
, i
;
2057 /* allocate memory for dirty segments list information */
2058 dirty_i
= kzalloc(sizeof(struct dirty_seglist_info
), GFP_KERNEL
);
2062 SM_I(sbi
)->dirty_info
= dirty_i
;
2063 mutex_init(&dirty_i
->seglist_lock
);
2065 bitmap_size
= f2fs_bitmap_size(MAIN_SEGS(sbi
));
2067 for (i
= 0; i
< NR_DIRTY_TYPE
; i
++) {
2068 dirty_i
->dirty_segmap
[i
] = kzalloc(bitmap_size
, GFP_KERNEL
);
2069 if (!dirty_i
->dirty_segmap
[i
])
2073 init_dirty_segmap(sbi
);
2074 return init_victim_secmap(sbi
);
2078 * Update min, max modified time for cost-benefit GC algorithm
2080 static void init_min_max_mtime(struct f2fs_sb_info
*sbi
)
2082 struct sit_info
*sit_i
= SIT_I(sbi
);
2085 mutex_lock(&sit_i
->sentry_lock
);
2087 sit_i
->min_mtime
= LLONG_MAX
;
2089 for (segno
= 0; segno
< MAIN_SEGS(sbi
); segno
+= sbi
->segs_per_sec
) {
2091 unsigned long long mtime
= 0;
2093 for (i
= 0; i
< sbi
->segs_per_sec
; i
++)
2094 mtime
+= get_seg_entry(sbi
, segno
+ i
)->mtime
;
2096 mtime
= div_u64(mtime
, sbi
->segs_per_sec
);
2098 if (sit_i
->min_mtime
> mtime
)
2099 sit_i
->min_mtime
= mtime
;
2101 sit_i
->max_mtime
= get_mtime(sbi
);
2102 mutex_unlock(&sit_i
->sentry_lock
);
2105 int build_segment_manager(struct f2fs_sb_info
*sbi
)
2107 struct f2fs_super_block
*raw_super
= F2FS_RAW_SUPER(sbi
);
2108 struct f2fs_checkpoint
*ckpt
= F2FS_CKPT(sbi
);
2109 struct f2fs_sm_info
*sm_info
;
2112 sm_info
= kzalloc(sizeof(struct f2fs_sm_info
), GFP_KERNEL
);
2117 sbi
->sm_info
= sm_info
;
2118 sm_info
->seg0_blkaddr
= le32_to_cpu(raw_super
->segment0_blkaddr
);
2119 sm_info
->main_blkaddr
= le32_to_cpu(raw_super
->main_blkaddr
);
2120 sm_info
->segment_count
= le32_to_cpu(raw_super
->segment_count
);
2121 sm_info
->reserved_segments
= le32_to_cpu(ckpt
->rsvd_segment_count
);
2122 sm_info
->ovp_segments
= le32_to_cpu(ckpt
->overprov_segment_count
);
2123 sm_info
->main_segments
= le32_to_cpu(raw_super
->segment_count_main
);
2124 sm_info
->ssa_blkaddr
= le32_to_cpu(raw_super
->ssa_blkaddr
);
2125 sm_info
->rec_prefree_segments
= sm_info
->main_segments
*
2126 DEF_RECLAIM_PREFREE_SEGMENTS
/ 100;
2127 sm_info
->ipu_policy
= 1 << F2FS_IPU_FSYNC
;
2128 sm_info
->min_ipu_util
= DEF_MIN_IPU_UTIL
;
2129 sm_info
->min_fsync_blocks
= DEF_MIN_FSYNC_BLOCKS
;
2131 INIT_LIST_HEAD(&sm_info
->discard_list
);
2132 sm_info
->nr_discards
= 0;
2133 sm_info
->max_discards
= 0;
2135 sm_info
->trim_sections
= DEF_BATCHED_TRIM_SECTIONS
;
2137 INIT_LIST_HEAD(&sm_info
->sit_entry_set
);
2139 if (test_opt(sbi
, FLUSH_MERGE
) && !f2fs_readonly(sbi
->sb
)) {
2140 err
= create_flush_cmd_control(sbi
);
2145 err
= build_sit_info(sbi
);
2148 err
= build_free_segmap(sbi
);
2151 err
= build_curseg(sbi
);
2155 /* reinit free segmap based on SIT */
2156 build_sit_entries(sbi
);
2158 init_free_segmap(sbi
);
2159 err
= build_dirty_segmap(sbi
);
2163 init_min_max_mtime(sbi
);
2167 static void discard_dirty_segmap(struct f2fs_sb_info
*sbi
,
2168 enum dirty_type dirty_type
)
2170 struct dirty_seglist_info
*dirty_i
= DIRTY_I(sbi
);
2172 mutex_lock(&dirty_i
->seglist_lock
);
2173 kfree(dirty_i
->dirty_segmap
[dirty_type
]);
2174 dirty_i
->nr_dirty
[dirty_type
] = 0;
2175 mutex_unlock(&dirty_i
->seglist_lock
);
2178 static void destroy_victim_secmap(struct f2fs_sb_info
*sbi
)
2180 struct dirty_seglist_info
*dirty_i
= DIRTY_I(sbi
);
2181 kfree(dirty_i
->victim_secmap
);
2184 static void destroy_dirty_segmap(struct f2fs_sb_info
*sbi
)
2186 struct dirty_seglist_info
*dirty_i
= DIRTY_I(sbi
);
2192 /* discard pre-free/dirty segments list */
2193 for (i
= 0; i
< NR_DIRTY_TYPE
; i
++)
2194 discard_dirty_segmap(sbi
, i
);
2196 destroy_victim_secmap(sbi
);
2197 SM_I(sbi
)->dirty_info
= NULL
;
2201 static void destroy_curseg(struct f2fs_sb_info
*sbi
)
2203 struct curseg_info
*array
= SM_I(sbi
)->curseg_array
;
2208 SM_I(sbi
)->curseg_array
= NULL
;
2209 for (i
= 0; i
< NR_CURSEG_TYPE
; i
++)
2210 kfree(array
[i
].sum_blk
);
2214 static void destroy_free_segmap(struct f2fs_sb_info
*sbi
)
2216 struct free_segmap_info
*free_i
= SM_I(sbi
)->free_info
;
2219 SM_I(sbi
)->free_info
= NULL
;
2220 kfree(free_i
->free_segmap
);
2221 kfree(free_i
->free_secmap
);
2225 static void destroy_sit_info(struct f2fs_sb_info
*sbi
)
2227 struct sit_info
*sit_i
= SIT_I(sbi
);
2233 if (sit_i
->sentries
) {
2234 for (start
= 0; start
< MAIN_SEGS(sbi
); start
++) {
2235 kfree(sit_i
->sentries
[start
].cur_valid_map
);
2236 kfree(sit_i
->sentries
[start
].ckpt_valid_map
);
2239 vfree(sit_i
->sentries
);
2240 vfree(sit_i
->sec_entries
);
2241 kfree(sit_i
->dirty_sentries_bitmap
);
2243 SM_I(sbi
)->sit_info
= NULL
;
2244 kfree(sit_i
->sit_bitmap
);
2248 void destroy_segment_manager(struct f2fs_sb_info
*sbi
)
2250 struct f2fs_sm_info
*sm_info
= SM_I(sbi
);
2254 destroy_flush_cmd_control(sbi
);
2255 destroy_dirty_segmap(sbi
);
2256 destroy_curseg(sbi
);
2257 destroy_free_segmap(sbi
);
2258 destroy_sit_info(sbi
);
2259 sbi
->sm_info
= NULL
;
2263 int __init
create_segment_manager_caches(void)
2265 discard_entry_slab
= f2fs_kmem_cache_create("discard_entry",
2266 sizeof(struct discard_entry
));
2267 if (!discard_entry_slab
)
2270 sit_entry_set_slab
= f2fs_kmem_cache_create("sit_entry_set",
2271 sizeof(struct sit_entry_set
));
2272 if (!sit_entry_set_slab
)
2273 goto destory_discard_entry
;
2275 inmem_entry_slab
= f2fs_kmem_cache_create("inmem_page_entry",
2276 sizeof(struct inmem_pages
));
2277 if (!inmem_entry_slab
)
2278 goto destroy_sit_entry_set
;
2281 destroy_sit_entry_set
:
2282 kmem_cache_destroy(sit_entry_set_slab
);
2283 destory_discard_entry
:
2284 kmem_cache_destroy(discard_entry_slab
);
2289 void destroy_segment_manager_caches(void)
2291 kmem_cache_destroy(sit_entry_set_slab
);
2292 kmem_cache_destroy(discard_entry_slab
);
2293 kmem_cache_destroy(inmem_entry_slab
);