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/swap.h>
23 #include <trace/events/f2fs.h>
25 #define __reverse_ffz(x) __reverse_ffs(~(x))
27 static struct kmem_cache
*discard_entry_slab
;
28 static struct kmem_cache
*sit_entry_set_slab
;
29 static struct kmem_cache
*inmem_entry_slab
;
32 * __reverse_ffs is copied from include/asm-generic/bitops/__ffs.h since
33 * MSB and LSB are reversed in a byte by f2fs_set_bit.
35 static inline unsigned long __reverse_ffs(unsigned long word
)
39 #if BITS_PER_LONG == 64
40 if ((word
& 0xffffffff) == 0) {
45 if ((word
& 0xffff) == 0) {
49 if ((word
& 0xff) == 0) {
53 if ((word
& 0xf0) == 0)
57 if ((word
& 0xc) == 0)
61 if ((word
& 0x2) == 0)
67 * __find_rev_next(_zero)_bit is copied from lib/find_next_bit.c because
68 * f2fs_set_bit makes MSB and LSB reversed in a byte.
71 * f2fs_set_bit(0, bitmap) => 0000 0001
72 * f2fs_set_bit(7, bitmap) => 1000 0000
74 static unsigned long __find_rev_next_bit(const unsigned long *addr
,
75 unsigned long size
, unsigned long offset
)
77 while (!f2fs_test_bit(offset
, (unsigned char *)addr
))
85 const unsigned long *p
= addr
+ BIT_WORD(offset
);
86 unsigned long result
= offset
& ~(BITS_PER_LONG
- 1);
88 unsigned long mask
, submask
;
89 unsigned long quot
, rest
;
95 offset
%= BITS_PER_LONG
;
100 quot
= (offset
>> 3) << 3;
103 submask
= (unsigned char)(0xff << rest
) >> rest
;
107 if (size
< BITS_PER_LONG
)
112 size
-= BITS_PER_LONG
;
113 result
+= BITS_PER_LONG
;
115 while (size
& ~(BITS_PER_LONG
-1)) {
119 result
+= BITS_PER_LONG
;
120 size
-= BITS_PER_LONG
;
126 tmp
&= (~0UL >> (BITS_PER_LONG
- size
));
127 if (tmp
== 0UL) /* Are any bits set? */
128 return result
+ size
; /* Nope. */
130 return result
+ __reverse_ffs(tmp
);
134 static unsigned long __find_rev_next_zero_bit(const unsigned long *addr
,
135 unsigned long size
, unsigned long offset
)
137 while (f2fs_test_bit(offset
, (unsigned char *)addr
))
145 const unsigned long *p
= addr
+ BIT_WORD(offset
);
146 unsigned long result
= offset
& ~(BITS_PER_LONG
- 1);
148 unsigned long mask
, submask
;
149 unsigned long quot
, rest
;
155 offset
%= BITS_PER_LONG
;
160 quot
= (offset
>> 3) << 3;
162 mask
= ~(~0UL << quot
);
163 submask
= (unsigned char)~((unsigned char)(0xff << rest
) >> rest
);
167 if (size
< BITS_PER_LONG
)
172 size
-= BITS_PER_LONG
;
173 result
+= BITS_PER_LONG
;
175 while (size
& ~(BITS_PER_LONG
- 1)) {
179 result
+= BITS_PER_LONG
;
180 size
-= BITS_PER_LONG
;
188 if (tmp
== ~0UL) /* Are any bits zero? */
189 return result
+ size
; /* Nope. */
191 return result
+ __reverse_ffz(tmp
);
195 void register_inmem_page(struct inode
*inode
, struct page
*page
)
197 struct f2fs_inode_info
*fi
= F2FS_I(inode
);
198 struct inmem_pages
*new;
200 f2fs_trace_pid(page
);
202 set_page_private(page
, (unsigned long)ATOMIC_WRITTEN_PAGE
);
203 SetPagePrivate(page
);
205 new = f2fs_kmem_cache_alloc(inmem_entry_slab
, GFP_NOFS
);
207 /* add atomic page indices to the list */
209 INIT_LIST_HEAD(&new->list
);
211 /* increase reference count with clean state */
212 mutex_lock(&fi
->inmem_lock
);
214 list_add_tail(&new->list
, &fi
->inmem_pages
);
215 inc_page_count(F2FS_I_SB(inode
), F2FS_INMEM_PAGES
);
216 mutex_unlock(&fi
->inmem_lock
);
218 trace_f2fs_register_inmem_page(page
, INMEM
);
221 int commit_inmem_pages(struct inode
*inode
, bool abort
)
223 struct f2fs_sb_info
*sbi
= F2FS_I_SB(inode
);
224 struct f2fs_inode_info
*fi
= F2FS_I(inode
);
225 struct inmem_pages
*cur
, *tmp
;
226 bool submit_bio
= false;
227 struct f2fs_io_info fio
= {
230 .rw
= WRITE_SYNC
| REQ_PRIO
,
231 .encrypted_page
= NULL
,
236 * The abort is true only when f2fs_evict_inode is called.
237 * Basically, the f2fs_evict_inode doesn't produce any data writes, so
238 * that we don't need to call f2fs_balance_fs.
239 * Otherwise, f2fs_gc in f2fs_balance_fs can wait forever until this
240 * inode becomes free by iget_locked in f2fs_iget.
243 f2fs_balance_fs(sbi
);
247 mutex_lock(&fi
->inmem_lock
);
248 list_for_each_entry_safe(cur
, tmp
, &fi
->inmem_pages
, list
) {
249 lock_page(cur
->page
);
251 if (cur
->page
->mapping
== inode
->i_mapping
) {
252 set_page_dirty(cur
->page
);
253 f2fs_wait_on_page_writeback(cur
->page
, DATA
);
254 if (clear_page_dirty_for_io(cur
->page
))
255 inode_dec_dirty_pages(inode
);
256 trace_f2fs_commit_inmem_page(cur
->page
, INMEM
);
257 fio
.page
= cur
->page
;
258 err
= do_write_data_page(&fio
);
261 unlock_page(cur
->page
);
266 trace_f2fs_commit_inmem_page(cur
->page
, INMEM_DROP
);
268 set_page_private(cur
->page
, 0);
269 ClearPagePrivate(cur
->page
);
270 f2fs_put_page(cur
->page
, 1);
272 list_del(&cur
->list
);
273 kmem_cache_free(inmem_entry_slab
, cur
);
274 dec_page_count(F2FS_I_SB(inode
), F2FS_INMEM_PAGES
);
276 mutex_unlock(&fi
->inmem_lock
);
281 f2fs_submit_merged_bio(sbi
, DATA
, WRITE
);
287 * This function balances dirty node and dentry pages.
288 * In addition, it controls garbage collection.
290 void f2fs_balance_fs(struct f2fs_sb_info
*sbi
)
293 * We should do GC or end up with checkpoint, if there are so many dirty
294 * dir/node pages without enough free segments.
296 if (has_not_enough_free_secs(sbi
, 0)) {
297 mutex_lock(&sbi
->gc_mutex
);
302 void f2fs_balance_fs_bg(struct f2fs_sb_info
*sbi
)
304 /* try to shrink extent cache when there is no enough memory */
305 if (!available_free_memory(sbi
, EXTENT_CACHE
))
306 f2fs_shrink_extent_tree(sbi
, EXTENT_CACHE_SHRINK_NUMBER
);
308 /* check the # of cached NAT entries */
309 if (!available_free_memory(sbi
, NAT_ENTRIES
))
310 try_to_free_nats(sbi
, NAT_ENTRY_PER_BLOCK
);
312 if (!available_free_memory(sbi
, FREE_NIDS
))
313 try_to_free_nids(sbi
, NAT_ENTRY_PER_BLOCK
* FREE_NID_PAGES
);
315 /* checkpoint is the only way to shrink partial cached entries */
316 if (!available_free_memory(sbi
, NAT_ENTRIES
) ||
317 excess_prefree_segs(sbi
) ||
318 !available_free_memory(sbi
, INO_ENTRIES
))
319 f2fs_sync_fs(sbi
->sb
, true);
322 static int issue_flush_thread(void *data
)
324 struct f2fs_sb_info
*sbi
= data
;
325 struct flush_cmd_control
*fcc
= SM_I(sbi
)->cmd_control_info
;
326 wait_queue_head_t
*q
= &fcc
->flush_wait_queue
;
328 if (kthread_should_stop())
331 if (!llist_empty(&fcc
->issue_list
)) {
333 struct flush_cmd
*cmd
, *next
;
336 bio
= f2fs_bio_alloc(0);
338 fcc
->dispatch_list
= llist_del_all(&fcc
->issue_list
);
339 fcc
->dispatch_list
= llist_reverse_order(fcc
->dispatch_list
);
341 bio
->bi_bdev
= sbi
->sb
->s_bdev
;
342 ret
= submit_bio_wait(WRITE_FLUSH
, bio
);
344 llist_for_each_entry_safe(cmd
, next
,
345 fcc
->dispatch_list
, llnode
) {
347 complete(&cmd
->wait
);
350 fcc
->dispatch_list
= NULL
;
353 wait_event_interruptible(*q
,
354 kthread_should_stop() || !llist_empty(&fcc
->issue_list
));
358 int f2fs_issue_flush(struct f2fs_sb_info
*sbi
)
360 struct flush_cmd_control
*fcc
= SM_I(sbi
)->cmd_control_info
;
361 struct flush_cmd cmd
;
363 trace_f2fs_issue_flush(sbi
->sb
, test_opt(sbi
, NOBARRIER
),
364 test_opt(sbi
, FLUSH_MERGE
));
366 if (test_opt(sbi
, NOBARRIER
))
369 if (!test_opt(sbi
, FLUSH_MERGE
)) {
370 struct bio
*bio
= f2fs_bio_alloc(0);
373 bio
->bi_bdev
= sbi
->sb
->s_bdev
;
374 ret
= submit_bio_wait(WRITE_FLUSH
, bio
);
379 init_completion(&cmd
.wait
);
381 llist_add(&cmd
.llnode
, &fcc
->issue_list
);
383 if (!fcc
->dispatch_list
)
384 wake_up(&fcc
->flush_wait_queue
);
386 wait_for_completion(&cmd
.wait
);
391 int create_flush_cmd_control(struct f2fs_sb_info
*sbi
)
393 dev_t dev
= sbi
->sb
->s_bdev
->bd_dev
;
394 struct flush_cmd_control
*fcc
;
397 fcc
= kzalloc(sizeof(struct flush_cmd_control
), GFP_KERNEL
);
400 init_waitqueue_head(&fcc
->flush_wait_queue
);
401 init_llist_head(&fcc
->issue_list
);
402 SM_I(sbi
)->cmd_control_info
= fcc
;
403 fcc
->f2fs_issue_flush
= kthread_run(issue_flush_thread
, sbi
,
404 "f2fs_flush-%u:%u", MAJOR(dev
), MINOR(dev
));
405 if (IS_ERR(fcc
->f2fs_issue_flush
)) {
406 err
= PTR_ERR(fcc
->f2fs_issue_flush
);
408 SM_I(sbi
)->cmd_control_info
= NULL
;
415 void destroy_flush_cmd_control(struct f2fs_sb_info
*sbi
)
417 struct flush_cmd_control
*fcc
= SM_I(sbi
)->cmd_control_info
;
419 if (fcc
&& fcc
->f2fs_issue_flush
)
420 kthread_stop(fcc
->f2fs_issue_flush
);
422 SM_I(sbi
)->cmd_control_info
= NULL
;
425 static void __locate_dirty_segment(struct f2fs_sb_info
*sbi
, unsigned int segno
,
426 enum dirty_type dirty_type
)
428 struct dirty_seglist_info
*dirty_i
= DIRTY_I(sbi
);
430 /* need not be added */
431 if (IS_CURSEG(sbi
, segno
))
434 if (!test_and_set_bit(segno
, dirty_i
->dirty_segmap
[dirty_type
]))
435 dirty_i
->nr_dirty
[dirty_type
]++;
437 if (dirty_type
== DIRTY
) {
438 struct seg_entry
*sentry
= get_seg_entry(sbi
, segno
);
439 enum dirty_type t
= sentry
->type
;
441 if (unlikely(t
>= DIRTY
)) {
445 if (!test_and_set_bit(segno
, dirty_i
->dirty_segmap
[t
]))
446 dirty_i
->nr_dirty
[t
]++;
450 static void __remove_dirty_segment(struct f2fs_sb_info
*sbi
, unsigned int segno
,
451 enum dirty_type dirty_type
)
453 struct dirty_seglist_info
*dirty_i
= DIRTY_I(sbi
);
455 if (test_and_clear_bit(segno
, dirty_i
->dirty_segmap
[dirty_type
]))
456 dirty_i
->nr_dirty
[dirty_type
]--;
458 if (dirty_type
== DIRTY
) {
459 struct seg_entry
*sentry
= get_seg_entry(sbi
, segno
);
460 enum dirty_type t
= sentry
->type
;
462 if (test_and_clear_bit(segno
, dirty_i
->dirty_segmap
[t
]))
463 dirty_i
->nr_dirty
[t
]--;
465 if (get_valid_blocks(sbi
, segno
, sbi
->segs_per_sec
) == 0)
466 clear_bit(GET_SECNO(sbi
, segno
),
467 dirty_i
->victim_secmap
);
472 * Should not occur error such as -ENOMEM.
473 * Adding dirty entry into seglist is not critical operation.
474 * If a given segment is one of current working segments, it won't be added.
476 static void locate_dirty_segment(struct f2fs_sb_info
*sbi
, unsigned int segno
)
478 struct dirty_seglist_info
*dirty_i
= DIRTY_I(sbi
);
479 unsigned short valid_blocks
;
481 if (segno
== NULL_SEGNO
|| IS_CURSEG(sbi
, segno
))
484 mutex_lock(&dirty_i
->seglist_lock
);
486 valid_blocks
= get_valid_blocks(sbi
, segno
, 0);
488 if (valid_blocks
== 0) {
489 __locate_dirty_segment(sbi
, segno
, PRE
);
490 __remove_dirty_segment(sbi
, segno
, DIRTY
);
491 } else if (valid_blocks
< sbi
->blocks_per_seg
) {
492 __locate_dirty_segment(sbi
, segno
, DIRTY
);
494 /* Recovery routine with SSR needs this */
495 __remove_dirty_segment(sbi
, segno
, DIRTY
);
498 mutex_unlock(&dirty_i
->seglist_lock
);
501 static int f2fs_issue_discard(struct f2fs_sb_info
*sbi
,
502 block_t blkstart
, block_t blklen
)
504 sector_t start
= SECTOR_FROM_BLOCK(blkstart
);
505 sector_t len
= SECTOR_FROM_BLOCK(blklen
);
506 struct seg_entry
*se
;
510 for (i
= blkstart
; i
< blkstart
+ blklen
; i
++) {
511 se
= get_seg_entry(sbi
, GET_SEGNO(sbi
, i
));
512 offset
= GET_BLKOFF_FROM_SEG0(sbi
, i
);
514 if (!f2fs_test_and_set_bit(offset
, se
->discard_map
))
517 trace_f2fs_issue_discard(sbi
->sb
, blkstart
, blklen
);
518 return blkdev_issue_discard(sbi
->sb
->s_bdev
, start
, len
, GFP_NOFS
, 0);
521 bool discard_next_dnode(struct f2fs_sb_info
*sbi
, block_t blkaddr
)
525 if (test_opt(sbi
, DISCARD
)) {
526 struct seg_entry
*se
= get_seg_entry(sbi
,
527 GET_SEGNO(sbi
, blkaddr
));
528 unsigned int offset
= GET_BLKOFF_FROM_SEG0(sbi
, blkaddr
);
530 if (f2fs_test_bit(offset
, se
->discard_map
))
533 err
= f2fs_issue_discard(sbi
, blkaddr
, 1);
537 update_meta_page(sbi
, NULL
, blkaddr
);
543 static void __add_discard_entry(struct f2fs_sb_info
*sbi
,
544 struct cp_control
*cpc
, struct seg_entry
*se
,
545 unsigned int start
, unsigned int end
)
547 struct list_head
*head
= &SM_I(sbi
)->discard_list
;
548 struct discard_entry
*new, *last
;
550 if (!list_empty(head
)) {
551 last
= list_last_entry(head
, struct discard_entry
, list
);
552 if (START_BLOCK(sbi
, cpc
->trim_start
) + start
==
553 last
->blkaddr
+ last
->len
) {
554 last
->len
+= end
- start
;
559 new = f2fs_kmem_cache_alloc(discard_entry_slab
, GFP_NOFS
);
560 INIT_LIST_HEAD(&new->list
);
561 new->blkaddr
= START_BLOCK(sbi
, cpc
->trim_start
) + start
;
562 new->len
= end
- start
;
563 list_add_tail(&new->list
, head
);
565 SM_I(sbi
)->nr_discards
+= end
- start
;
568 static void add_discard_addrs(struct f2fs_sb_info
*sbi
, struct cp_control
*cpc
)
570 int entries
= SIT_VBLOCK_MAP_SIZE
/ sizeof(unsigned long);
571 int max_blocks
= sbi
->blocks_per_seg
;
572 struct seg_entry
*se
= get_seg_entry(sbi
, cpc
->trim_start
);
573 unsigned long *cur_map
= (unsigned long *)se
->cur_valid_map
;
574 unsigned long *ckpt_map
= (unsigned long *)se
->ckpt_valid_map
;
575 unsigned long *discard_map
= (unsigned long *)se
->discard_map
;
576 unsigned long *dmap
= SIT_I(sbi
)->tmp_map
;
577 unsigned int start
= 0, end
= -1;
578 bool force
= (cpc
->reason
== CP_DISCARD
);
581 if (se
->valid_blocks
== max_blocks
)
585 if (!test_opt(sbi
, DISCARD
) || !se
->valid_blocks
||
586 SM_I(sbi
)->nr_discards
>= SM_I(sbi
)->max_discards
)
590 /* SIT_VBLOCK_MAP_SIZE should be multiple of sizeof(unsigned long) */
591 for (i
= 0; i
< entries
; i
++)
592 dmap
[i
] = force
? ~ckpt_map
[i
] & ~discard_map
[i
] :
593 (cur_map
[i
] ^ ckpt_map
[i
]) & ckpt_map
[i
];
595 while (force
|| SM_I(sbi
)->nr_discards
<= SM_I(sbi
)->max_discards
) {
596 start
= __find_rev_next_bit(dmap
, max_blocks
, end
+ 1);
597 if (start
>= max_blocks
)
600 end
= __find_rev_next_zero_bit(dmap
, max_blocks
, start
+ 1);
601 __add_discard_entry(sbi
, cpc
, se
, start
, end
);
605 void release_discard_addrs(struct f2fs_sb_info
*sbi
)
607 struct list_head
*head
= &(SM_I(sbi
)->discard_list
);
608 struct discard_entry
*entry
, *this;
611 list_for_each_entry_safe(entry
, this, head
, list
) {
612 list_del(&entry
->list
);
613 kmem_cache_free(discard_entry_slab
, entry
);
618 * Should call clear_prefree_segments after checkpoint is done.
620 static void set_prefree_as_free_segments(struct f2fs_sb_info
*sbi
)
622 struct dirty_seglist_info
*dirty_i
= DIRTY_I(sbi
);
625 mutex_lock(&dirty_i
->seglist_lock
);
626 for_each_set_bit(segno
, dirty_i
->dirty_segmap
[PRE
], MAIN_SEGS(sbi
))
627 __set_test_and_free(sbi
, segno
);
628 mutex_unlock(&dirty_i
->seglist_lock
);
631 void clear_prefree_segments(struct f2fs_sb_info
*sbi
, struct cp_control
*cpc
)
633 struct list_head
*head
= &(SM_I(sbi
)->discard_list
);
634 struct discard_entry
*entry
, *this;
635 struct dirty_seglist_info
*dirty_i
= DIRTY_I(sbi
);
636 unsigned long *prefree_map
= dirty_i
->dirty_segmap
[PRE
];
637 unsigned int start
= 0, end
= -1;
639 mutex_lock(&dirty_i
->seglist_lock
);
643 start
= find_next_bit(prefree_map
, MAIN_SEGS(sbi
), end
+ 1);
644 if (start
>= MAIN_SEGS(sbi
))
646 end
= find_next_zero_bit(prefree_map
, MAIN_SEGS(sbi
),
649 for (i
= start
; i
< end
; i
++)
650 clear_bit(i
, prefree_map
);
652 dirty_i
->nr_dirty
[PRE
] -= end
- start
;
654 if (!test_opt(sbi
, DISCARD
))
657 f2fs_issue_discard(sbi
, START_BLOCK(sbi
, start
),
658 (end
- start
) << sbi
->log_blocks_per_seg
);
660 mutex_unlock(&dirty_i
->seglist_lock
);
662 /* send small discards */
663 list_for_each_entry_safe(entry
, this, head
, list
) {
664 if (cpc
->reason
== CP_DISCARD
&& entry
->len
< cpc
->trim_minlen
)
666 f2fs_issue_discard(sbi
, entry
->blkaddr
, entry
->len
);
667 cpc
->trimmed
+= entry
->len
;
669 list_del(&entry
->list
);
670 SM_I(sbi
)->nr_discards
-= entry
->len
;
671 kmem_cache_free(discard_entry_slab
, entry
);
675 static bool __mark_sit_entry_dirty(struct f2fs_sb_info
*sbi
, unsigned int segno
)
677 struct sit_info
*sit_i
= SIT_I(sbi
);
679 if (!__test_and_set_bit(segno
, sit_i
->dirty_sentries_bitmap
)) {
680 sit_i
->dirty_sentries
++;
687 static void __set_sit_entry_type(struct f2fs_sb_info
*sbi
, int type
,
688 unsigned int segno
, int modified
)
690 struct seg_entry
*se
= get_seg_entry(sbi
, segno
);
693 __mark_sit_entry_dirty(sbi
, segno
);
696 static void update_sit_entry(struct f2fs_sb_info
*sbi
, block_t blkaddr
, int del
)
698 struct seg_entry
*se
;
699 unsigned int segno
, offset
;
700 long int new_vblocks
;
702 segno
= GET_SEGNO(sbi
, blkaddr
);
704 se
= get_seg_entry(sbi
, segno
);
705 new_vblocks
= se
->valid_blocks
+ del
;
706 offset
= GET_BLKOFF_FROM_SEG0(sbi
, blkaddr
);
708 f2fs_bug_on(sbi
, (new_vblocks
>> (sizeof(unsigned short) << 3) ||
709 (new_vblocks
> sbi
->blocks_per_seg
)));
711 se
->valid_blocks
= new_vblocks
;
712 se
->mtime
= get_mtime(sbi
);
713 SIT_I(sbi
)->max_mtime
= se
->mtime
;
715 /* Update valid block bitmap */
717 if (f2fs_test_and_set_bit(offset
, se
->cur_valid_map
))
719 if (!f2fs_test_and_set_bit(offset
, se
->discard_map
))
722 if (!f2fs_test_and_clear_bit(offset
, se
->cur_valid_map
))
724 if (f2fs_test_and_clear_bit(offset
, se
->discard_map
))
727 if (!f2fs_test_bit(offset
, se
->ckpt_valid_map
))
728 se
->ckpt_valid_blocks
+= del
;
730 __mark_sit_entry_dirty(sbi
, segno
);
732 /* update total number of valid blocks to be written in ckpt area */
733 SIT_I(sbi
)->written_valid_blocks
+= del
;
735 if (sbi
->segs_per_sec
> 1)
736 get_sec_entry(sbi
, segno
)->valid_blocks
+= del
;
739 void refresh_sit_entry(struct f2fs_sb_info
*sbi
, block_t old
, block_t
new)
741 update_sit_entry(sbi
, new, 1);
742 if (GET_SEGNO(sbi
, old
) != NULL_SEGNO
)
743 update_sit_entry(sbi
, old
, -1);
745 locate_dirty_segment(sbi
, GET_SEGNO(sbi
, old
));
746 locate_dirty_segment(sbi
, GET_SEGNO(sbi
, new));
749 void invalidate_blocks(struct f2fs_sb_info
*sbi
, block_t addr
)
751 unsigned int segno
= GET_SEGNO(sbi
, addr
);
752 struct sit_info
*sit_i
= SIT_I(sbi
);
754 f2fs_bug_on(sbi
, addr
== NULL_ADDR
);
755 if (addr
== NEW_ADDR
)
758 /* add it into sit main buffer */
759 mutex_lock(&sit_i
->sentry_lock
);
761 update_sit_entry(sbi
, addr
, -1);
763 /* add it into dirty seglist */
764 locate_dirty_segment(sbi
, segno
);
766 mutex_unlock(&sit_i
->sentry_lock
);
770 * This function should be resided under the curseg_mutex lock
772 static void __add_sum_entry(struct f2fs_sb_info
*sbi
, int type
,
773 struct f2fs_summary
*sum
)
775 struct curseg_info
*curseg
= CURSEG_I(sbi
, type
);
776 void *addr
= curseg
->sum_blk
;
777 addr
+= curseg
->next_blkoff
* sizeof(struct f2fs_summary
);
778 memcpy(addr
, sum
, sizeof(struct f2fs_summary
));
782 * Calculate the number of current summary pages for writing
784 int npages_for_summary_flush(struct f2fs_sb_info
*sbi
, bool for_ra
)
786 int valid_sum_count
= 0;
789 for (i
= CURSEG_HOT_DATA
; i
<= CURSEG_COLD_DATA
; i
++) {
790 if (sbi
->ckpt
->alloc_type
[i
] == SSR
)
791 valid_sum_count
+= sbi
->blocks_per_seg
;
794 valid_sum_count
+= le16_to_cpu(
795 F2FS_CKPT(sbi
)->cur_data_blkoff
[i
]);
797 valid_sum_count
+= curseg_blkoff(sbi
, i
);
801 sum_in_page
= (PAGE_CACHE_SIZE
- 2 * SUM_JOURNAL_SIZE
-
802 SUM_FOOTER_SIZE
) / SUMMARY_SIZE
;
803 if (valid_sum_count
<= sum_in_page
)
805 else if ((valid_sum_count
- sum_in_page
) <=
806 (PAGE_CACHE_SIZE
- SUM_FOOTER_SIZE
) / SUMMARY_SIZE
)
812 * Caller should put this summary page
814 struct page
*get_sum_page(struct f2fs_sb_info
*sbi
, unsigned int segno
)
816 return get_meta_page(sbi
, GET_SUM_BLOCK(sbi
, segno
));
819 void update_meta_page(struct f2fs_sb_info
*sbi
, void *src
, block_t blk_addr
)
821 struct page
*page
= grab_meta_page(sbi
, blk_addr
);
822 void *dst
= page_address(page
);
825 memcpy(dst
, src
, PAGE_CACHE_SIZE
);
827 memset(dst
, 0, PAGE_CACHE_SIZE
);
828 set_page_dirty(page
);
829 f2fs_put_page(page
, 1);
832 static void write_sum_page(struct f2fs_sb_info
*sbi
,
833 struct f2fs_summary_block
*sum_blk
, block_t blk_addr
)
835 update_meta_page(sbi
, (void *)sum_blk
, blk_addr
);
838 static int is_next_segment_free(struct f2fs_sb_info
*sbi
, int type
)
840 struct curseg_info
*curseg
= CURSEG_I(sbi
, type
);
841 unsigned int segno
= curseg
->segno
+ 1;
842 struct free_segmap_info
*free_i
= FREE_I(sbi
);
844 if (segno
< MAIN_SEGS(sbi
) && segno
% sbi
->segs_per_sec
)
845 return !test_bit(segno
, free_i
->free_segmap
);
850 * Find a new segment from the free segments bitmap to right order
851 * This function should be returned with success, otherwise BUG
853 static void get_new_segment(struct f2fs_sb_info
*sbi
,
854 unsigned int *newseg
, bool new_sec
, int dir
)
856 struct free_segmap_info
*free_i
= FREE_I(sbi
);
857 unsigned int segno
, secno
, zoneno
;
858 unsigned int total_zones
= MAIN_SECS(sbi
) / sbi
->secs_per_zone
;
859 unsigned int hint
= *newseg
/ sbi
->segs_per_sec
;
860 unsigned int old_zoneno
= GET_ZONENO_FROM_SEGNO(sbi
, *newseg
);
861 unsigned int left_start
= hint
;
866 spin_lock(&free_i
->segmap_lock
);
868 if (!new_sec
&& ((*newseg
+ 1) % sbi
->segs_per_sec
)) {
869 segno
= find_next_zero_bit(free_i
->free_segmap
,
870 MAIN_SEGS(sbi
), *newseg
+ 1);
871 if (segno
- *newseg
< sbi
->segs_per_sec
-
872 (*newseg
% sbi
->segs_per_sec
))
876 secno
= find_next_zero_bit(free_i
->free_secmap
, MAIN_SECS(sbi
), hint
);
877 if (secno
>= MAIN_SECS(sbi
)) {
878 if (dir
== ALLOC_RIGHT
) {
879 secno
= find_next_zero_bit(free_i
->free_secmap
,
881 f2fs_bug_on(sbi
, secno
>= MAIN_SECS(sbi
));
884 left_start
= hint
- 1;
890 while (test_bit(left_start
, free_i
->free_secmap
)) {
891 if (left_start
> 0) {
895 left_start
= find_next_zero_bit(free_i
->free_secmap
,
897 f2fs_bug_on(sbi
, left_start
>= MAIN_SECS(sbi
));
903 segno
= secno
* sbi
->segs_per_sec
;
904 zoneno
= secno
/ sbi
->secs_per_zone
;
906 /* give up on finding another zone */
909 if (sbi
->secs_per_zone
== 1)
911 if (zoneno
== old_zoneno
)
913 if (dir
== ALLOC_LEFT
) {
914 if (!go_left
&& zoneno
+ 1 >= total_zones
)
916 if (go_left
&& zoneno
== 0)
919 for (i
= 0; i
< NR_CURSEG_TYPE
; i
++)
920 if (CURSEG_I(sbi
, i
)->zone
== zoneno
)
923 if (i
< NR_CURSEG_TYPE
) {
924 /* zone is in user, try another */
926 hint
= zoneno
* sbi
->secs_per_zone
- 1;
927 else if (zoneno
+ 1 >= total_zones
)
930 hint
= (zoneno
+ 1) * sbi
->secs_per_zone
;
932 goto find_other_zone
;
935 /* set it as dirty segment in free segmap */
936 f2fs_bug_on(sbi
, test_bit(segno
, free_i
->free_segmap
));
937 __set_inuse(sbi
, segno
);
939 spin_unlock(&free_i
->segmap_lock
);
942 static void reset_curseg(struct f2fs_sb_info
*sbi
, int type
, int modified
)
944 struct curseg_info
*curseg
= CURSEG_I(sbi
, type
);
945 struct summary_footer
*sum_footer
;
947 curseg
->segno
= curseg
->next_segno
;
948 curseg
->zone
= GET_ZONENO_FROM_SEGNO(sbi
, curseg
->segno
);
949 curseg
->next_blkoff
= 0;
950 curseg
->next_segno
= NULL_SEGNO
;
952 sum_footer
= &(curseg
->sum_blk
->footer
);
953 memset(sum_footer
, 0, sizeof(struct summary_footer
));
954 if (IS_DATASEG(type
))
955 SET_SUM_TYPE(sum_footer
, SUM_TYPE_DATA
);
956 if (IS_NODESEG(type
))
957 SET_SUM_TYPE(sum_footer
, SUM_TYPE_NODE
);
958 __set_sit_entry_type(sbi
, type
, curseg
->segno
, modified
);
962 * Allocate a current working segment.
963 * This function always allocates a free segment in LFS manner.
965 static void new_curseg(struct f2fs_sb_info
*sbi
, int type
, bool new_sec
)
967 struct curseg_info
*curseg
= CURSEG_I(sbi
, type
);
968 unsigned int segno
= curseg
->segno
;
969 int dir
= ALLOC_LEFT
;
971 write_sum_page(sbi
, curseg
->sum_blk
,
972 GET_SUM_BLOCK(sbi
, segno
));
973 if (type
== CURSEG_WARM_DATA
|| type
== CURSEG_COLD_DATA
)
976 if (test_opt(sbi
, NOHEAP
))
979 get_new_segment(sbi
, &segno
, new_sec
, dir
);
980 curseg
->next_segno
= segno
;
981 reset_curseg(sbi
, type
, 1);
982 curseg
->alloc_type
= LFS
;
985 static void __next_free_blkoff(struct f2fs_sb_info
*sbi
,
986 struct curseg_info
*seg
, block_t start
)
988 struct seg_entry
*se
= get_seg_entry(sbi
, seg
->segno
);
989 int entries
= SIT_VBLOCK_MAP_SIZE
/ sizeof(unsigned long);
990 unsigned long *target_map
= SIT_I(sbi
)->tmp_map
;
991 unsigned long *ckpt_map
= (unsigned long *)se
->ckpt_valid_map
;
992 unsigned long *cur_map
= (unsigned long *)se
->cur_valid_map
;
995 for (i
= 0; i
< entries
; i
++)
996 target_map
[i
] = ckpt_map
[i
] | cur_map
[i
];
998 pos
= __find_rev_next_zero_bit(target_map
, sbi
->blocks_per_seg
, start
);
1000 seg
->next_blkoff
= pos
;
1004 * If a segment is written by LFS manner, next block offset is just obtained
1005 * by increasing the current block offset. However, if a segment is written by
1006 * SSR manner, next block offset obtained by calling __next_free_blkoff
1008 static void __refresh_next_blkoff(struct f2fs_sb_info
*sbi
,
1009 struct curseg_info
*seg
)
1011 if (seg
->alloc_type
== SSR
)
1012 __next_free_blkoff(sbi
, seg
, seg
->next_blkoff
+ 1);
1018 * This function always allocates a used segment(from dirty seglist) by SSR
1019 * manner, so it should recover the existing segment information of valid blocks
1021 static void change_curseg(struct f2fs_sb_info
*sbi
, int type
, bool reuse
)
1023 struct dirty_seglist_info
*dirty_i
= DIRTY_I(sbi
);
1024 struct curseg_info
*curseg
= CURSEG_I(sbi
, type
);
1025 unsigned int new_segno
= curseg
->next_segno
;
1026 struct f2fs_summary_block
*sum_node
;
1027 struct page
*sum_page
;
1029 write_sum_page(sbi
, curseg
->sum_blk
,
1030 GET_SUM_BLOCK(sbi
, curseg
->segno
));
1031 __set_test_and_inuse(sbi
, new_segno
);
1033 mutex_lock(&dirty_i
->seglist_lock
);
1034 __remove_dirty_segment(sbi
, new_segno
, PRE
);
1035 __remove_dirty_segment(sbi
, new_segno
, DIRTY
);
1036 mutex_unlock(&dirty_i
->seglist_lock
);
1038 reset_curseg(sbi
, type
, 1);
1039 curseg
->alloc_type
= SSR
;
1040 __next_free_blkoff(sbi
, curseg
, 0);
1043 sum_page
= get_sum_page(sbi
, new_segno
);
1044 sum_node
= (struct f2fs_summary_block
*)page_address(sum_page
);
1045 memcpy(curseg
->sum_blk
, sum_node
, SUM_ENTRY_SIZE
);
1046 f2fs_put_page(sum_page
, 1);
1050 static int get_ssr_segment(struct f2fs_sb_info
*sbi
, int type
)
1052 struct curseg_info
*curseg
= CURSEG_I(sbi
, type
);
1053 const struct victim_selection
*v_ops
= DIRTY_I(sbi
)->v_ops
;
1055 if (IS_NODESEG(type
) || !has_not_enough_free_secs(sbi
, 0))
1056 return v_ops
->get_victim(sbi
,
1057 &(curseg
)->next_segno
, BG_GC
, type
, SSR
);
1059 /* For data segments, let's do SSR more intensively */
1060 for (; type
>= CURSEG_HOT_DATA
; type
--)
1061 if (v_ops
->get_victim(sbi
, &(curseg
)->next_segno
,
1068 * flush out current segment and replace it with new segment
1069 * This function should be returned with success, otherwise BUG
1071 static void allocate_segment_by_default(struct f2fs_sb_info
*sbi
,
1072 int type
, bool force
)
1074 struct curseg_info
*curseg
= CURSEG_I(sbi
, type
);
1077 new_curseg(sbi
, type
, true);
1078 else if (type
== CURSEG_WARM_NODE
)
1079 new_curseg(sbi
, type
, false);
1080 else if (curseg
->alloc_type
== LFS
&& is_next_segment_free(sbi
, type
))
1081 new_curseg(sbi
, type
, false);
1082 else if (need_SSR(sbi
) && get_ssr_segment(sbi
, type
))
1083 change_curseg(sbi
, type
, true);
1085 new_curseg(sbi
, type
, false);
1087 stat_inc_seg_type(sbi
, curseg
);
1090 static void __allocate_new_segments(struct f2fs_sb_info
*sbi
, int type
)
1092 struct curseg_info
*curseg
= CURSEG_I(sbi
, type
);
1093 unsigned int old_segno
;
1095 old_segno
= curseg
->segno
;
1096 SIT_I(sbi
)->s_ops
->allocate_segment(sbi
, type
, true);
1097 locate_dirty_segment(sbi
, old_segno
);
1100 void allocate_new_segments(struct f2fs_sb_info
*sbi
)
1104 for (i
= CURSEG_HOT_DATA
; i
<= CURSEG_COLD_DATA
; i
++)
1105 __allocate_new_segments(sbi
, i
);
1108 static const struct segment_allocation default_salloc_ops
= {
1109 .allocate_segment
= allocate_segment_by_default
,
1112 int f2fs_trim_fs(struct f2fs_sb_info
*sbi
, struct fstrim_range
*range
)
1114 __u64 start
= F2FS_BYTES_TO_BLK(range
->start
);
1115 __u64 end
= start
+ F2FS_BYTES_TO_BLK(range
->len
) - 1;
1116 unsigned int start_segno
, end_segno
;
1117 struct cp_control cpc
;
1119 if (start
>= MAX_BLKADDR(sbi
) || range
->len
< sbi
->blocksize
)
1123 if (end
<= MAIN_BLKADDR(sbi
))
1126 /* start/end segment number in main_area */
1127 start_segno
= (start
<= MAIN_BLKADDR(sbi
)) ? 0 : GET_SEGNO(sbi
, start
);
1128 end_segno
= (end
>= MAX_BLKADDR(sbi
)) ? MAIN_SEGS(sbi
) - 1 :
1129 GET_SEGNO(sbi
, end
);
1130 cpc
.reason
= CP_DISCARD
;
1131 cpc
.trim_minlen
= max_t(__u64
, 1, F2FS_BYTES_TO_BLK(range
->minlen
));
1133 /* do checkpoint to issue discard commands safely */
1134 for (; start_segno
<= end_segno
; start_segno
= cpc
.trim_end
+ 1) {
1135 cpc
.trim_start
= start_segno
;
1137 if (sbi
->discard_blks
== 0)
1139 else if (sbi
->discard_blks
< BATCHED_TRIM_BLOCKS(sbi
))
1140 cpc
.trim_end
= end_segno
;
1142 cpc
.trim_end
= min_t(unsigned int,
1143 rounddown(start_segno
+
1144 BATCHED_TRIM_SEGMENTS(sbi
),
1145 sbi
->segs_per_sec
) - 1, end_segno
);
1147 mutex_lock(&sbi
->gc_mutex
);
1148 write_checkpoint(sbi
, &cpc
);
1149 mutex_unlock(&sbi
->gc_mutex
);
1152 range
->len
= F2FS_BLK_TO_BYTES(cpc
.trimmed
);
1156 static bool __has_curseg_space(struct f2fs_sb_info
*sbi
, int type
)
1158 struct curseg_info
*curseg
= CURSEG_I(sbi
, type
);
1159 if (curseg
->next_blkoff
< sbi
->blocks_per_seg
)
1164 static int __get_segment_type_2(struct page
*page
, enum page_type p_type
)
1167 return CURSEG_HOT_DATA
;
1169 return CURSEG_HOT_NODE
;
1172 static int __get_segment_type_4(struct page
*page
, enum page_type p_type
)
1174 if (p_type
== DATA
) {
1175 struct inode
*inode
= page
->mapping
->host
;
1177 if (S_ISDIR(inode
->i_mode
))
1178 return CURSEG_HOT_DATA
;
1180 return CURSEG_COLD_DATA
;
1182 if (IS_DNODE(page
) && is_cold_node(page
))
1183 return CURSEG_WARM_NODE
;
1185 return CURSEG_COLD_NODE
;
1189 static int __get_segment_type_6(struct page
*page
, enum page_type p_type
)
1191 if (p_type
== DATA
) {
1192 struct inode
*inode
= page
->mapping
->host
;
1194 if (S_ISDIR(inode
->i_mode
))
1195 return CURSEG_HOT_DATA
;
1196 else if (is_cold_data(page
) || file_is_cold(inode
))
1197 return CURSEG_COLD_DATA
;
1199 return CURSEG_WARM_DATA
;
1202 return is_cold_node(page
) ? CURSEG_WARM_NODE
:
1205 return CURSEG_COLD_NODE
;
1209 static int __get_segment_type(struct page
*page
, enum page_type p_type
)
1211 switch (F2FS_P_SB(page
)->active_logs
) {
1213 return __get_segment_type_2(page
, p_type
);
1215 return __get_segment_type_4(page
, p_type
);
1217 /* NR_CURSEG_TYPE(6) logs by default */
1218 f2fs_bug_on(F2FS_P_SB(page
),
1219 F2FS_P_SB(page
)->active_logs
!= NR_CURSEG_TYPE
);
1220 return __get_segment_type_6(page
, p_type
);
1223 void allocate_data_block(struct f2fs_sb_info
*sbi
, struct page
*page
,
1224 block_t old_blkaddr
, block_t
*new_blkaddr
,
1225 struct f2fs_summary
*sum
, int type
)
1227 struct sit_info
*sit_i
= SIT_I(sbi
);
1228 struct curseg_info
*curseg
;
1229 bool direct_io
= (type
== CURSEG_DIRECT_IO
);
1231 type
= direct_io
? CURSEG_WARM_DATA
: type
;
1233 curseg
= CURSEG_I(sbi
, type
);
1235 mutex_lock(&curseg
->curseg_mutex
);
1236 mutex_lock(&sit_i
->sentry_lock
);
1238 /* direct_io'ed data is aligned to the segment for better performance */
1239 if (direct_io
&& curseg
->next_blkoff
&&
1240 !has_not_enough_free_secs(sbi
, 0))
1241 __allocate_new_segments(sbi
, type
);
1243 *new_blkaddr
= NEXT_FREE_BLKADDR(sbi
, curseg
);
1246 * __add_sum_entry should be resided under the curseg_mutex
1247 * because, this function updates a summary entry in the
1248 * current summary block.
1250 __add_sum_entry(sbi
, type
, sum
);
1252 __refresh_next_blkoff(sbi
, curseg
);
1254 stat_inc_block_count(sbi
, curseg
);
1256 if (!__has_curseg_space(sbi
, type
))
1257 sit_i
->s_ops
->allocate_segment(sbi
, type
, false);
1259 * SIT information should be updated before segment allocation,
1260 * since SSR needs latest valid block information.
1262 refresh_sit_entry(sbi
, old_blkaddr
, *new_blkaddr
);
1264 mutex_unlock(&sit_i
->sentry_lock
);
1266 if (page
&& IS_NODESEG(type
))
1267 fill_node_footer_blkaddr(page
, NEXT_FREE_BLKADDR(sbi
, curseg
));
1269 mutex_unlock(&curseg
->curseg_mutex
);
1272 static void do_write_page(struct f2fs_summary
*sum
, struct f2fs_io_info
*fio
)
1274 int type
= __get_segment_type(fio
->page
, fio
->type
);
1276 allocate_data_block(fio
->sbi
, fio
->page
, fio
->blk_addr
,
1277 &fio
->blk_addr
, sum
, type
);
1279 /* writeout dirty page into bdev */
1280 f2fs_submit_page_mbio(fio
);
1283 void write_meta_page(struct f2fs_sb_info
*sbi
, struct page
*page
)
1285 struct f2fs_io_info fio
= {
1288 .rw
= WRITE_SYNC
| REQ_META
| REQ_PRIO
,
1289 .blk_addr
= page
->index
,
1291 .encrypted_page
= NULL
,
1294 set_page_writeback(page
);
1295 f2fs_submit_page_mbio(&fio
);
1298 void write_node_page(unsigned int nid
, struct f2fs_io_info
*fio
)
1300 struct f2fs_summary sum
;
1302 set_summary(&sum
, nid
, 0, 0);
1303 do_write_page(&sum
, fio
);
1306 void write_data_page(struct dnode_of_data
*dn
, struct f2fs_io_info
*fio
)
1308 struct f2fs_sb_info
*sbi
= fio
->sbi
;
1309 struct f2fs_summary sum
;
1310 struct node_info ni
;
1312 f2fs_bug_on(sbi
, dn
->data_blkaddr
== NULL_ADDR
);
1313 get_node_info(sbi
, dn
->nid
, &ni
);
1314 set_summary(&sum
, dn
->nid
, dn
->ofs_in_node
, ni
.version
);
1315 do_write_page(&sum
, fio
);
1316 dn
->data_blkaddr
= fio
->blk_addr
;
1319 void rewrite_data_page(struct f2fs_io_info
*fio
)
1321 stat_inc_inplace_blocks(fio
->sbi
);
1322 f2fs_submit_page_mbio(fio
);
1325 static void __f2fs_replace_block(struct f2fs_sb_info
*sbi
,
1326 struct f2fs_summary
*sum
,
1327 block_t old_blkaddr
, block_t new_blkaddr
,
1328 bool recover_curseg
)
1330 struct sit_info
*sit_i
= SIT_I(sbi
);
1331 struct curseg_info
*curseg
;
1332 unsigned int segno
, old_cursegno
;
1333 struct seg_entry
*se
;
1335 unsigned short old_blkoff
;
1337 segno
= GET_SEGNO(sbi
, new_blkaddr
);
1338 se
= get_seg_entry(sbi
, segno
);
1341 if (!recover_curseg
) {
1342 /* for recovery flow */
1343 if (se
->valid_blocks
== 0 && !IS_CURSEG(sbi
, segno
)) {
1344 if (old_blkaddr
== NULL_ADDR
)
1345 type
= CURSEG_COLD_DATA
;
1347 type
= CURSEG_WARM_DATA
;
1350 if (!IS_CURSEG(sbi
, segno
))
1351 type
= CURSEG_WARM_DATA
;
1354 curseg
= CURSEG_I(sbi
, type
);
1356 mutex_lock(&curseg
->curseg_mutex
);
1357 mutex_lock(&sit_i
->sentry_lock
);
1359 old_cursegno
= curseg
->segno
;
1360 old_blkoff
= curseg
->next_blkoff
;
1362 /* change the current segment */
1363 if (segno
!= curseg
->segno
) {
1364 curseg
->next_segno
= segno
;
1365 change_curseg(sbi
, type
, true);
1368 curseg
->next_blkoff
= GET_BLKOFF_FROM_SEG0(sbi
, new_blkaddr
);
1369 __add_sum_entry(sbi
, type
, sum
);
1371 refresh_sit_entry(sbi
, old_blkaddr
, new_blkaddr
);
1372 locate_dirty_segment(sbi
, old_cursegno
);
1374 if (recover_curseg
) {
1375 if (old_cursegno
!= curseg
->segno
) {
1376 curseg
->next_segno
= old_cursegno
;
1377 change_curseg(sbi
, type
, true);
1379 curseg
->next_blkoff
= old_blkoff
;
1382 mutex_unlock(&sit_i
->sentry_lock
);
1383 mutex_unlock(&curseg
->curseg_mutex
);
1386 void f2fs_replace_block(struct f2fs_sb_info
*sbi
, struct dnode_of_data
*dn
,
1387 block_t old_addr
, block_t new_addr
,
1388 unsigned char version
, bool recover_curseg
)
1390 struct f2fs_summary sum
;
1392 set_summary(&sum
, dn
->nid
, dn
->ofs_in_node
, version
);
1394 __f2fs_replace_block(sbi
, &sum
, old_addr
, new_addr
, recover_curseg
);
1396 dn
->data_blkaddr
= new_addr
;
1397 set_data_blkaddr(dn
);
1398 f2fs_update_extent_cache(dn
);
1401 static inline bool is_merged_page(struct f2fs_sb_info
*sbi
,
1402 struct page
*page
, enum page_type type
)
1404 enum page_type btype
= PAGE_TYPE_OF_BIO(type
);
1405 struct f2fs_bio_info
*io
= &sbi
->write_io
[btype
];
1406 struct bio_vec
*bvec
;
1407 struct page
*target
;
1410 down_read(&io
->io_rwsem
);
1412 up_read(&io
->io_rwsem
);
1416 bio_for_each_segment_all(bvec
, io
->bio
, i
) {
1418 if (bvec
->bv_page
->mapping
) {
1419 target
= bvec
->bv_page
;
1421 struct f2fs_crypto_ctx
*ctx
;
1423 /* encrypted page */
1424 ctx
= (struct f2fs_crypto_ctx
*)page_private(
1426 target
= ctx
->w
.control_page
;
1429 if (page
== target
) {
1430 up_read(&io
->io_rwsem
);
1435 up_read(&io
->io_rwsem
);
1439 void f2fs_wait_on_page_writeback(struct page
*page
,
1440 enum page_type type
)
1442 if (PageWriteback(page
)) {
1443 struct f2fs_sb_info
*sbi
= F2FS_P_SB(page
);
1445 if (is_merged_page(sbi
, page
, type
))
1446 f2fs_submit_merged_bio(sbi
, type
, WRITE
);
1447 wait_on_page_writeback(page
);
1451 static int read_compacted_summaries(struct f2fs_sb_info
*sbi
)
1453 struct f2fs_checkpoint
*ckpt
= F2FS_CKPT(sbi
);
1454 struct curseg_info
*seg_i
;
1455 unsigned char *kaddr
;
1460 start
= start_sum_block(sbi
);
1462 page
= get_meta_page(sbi
, start
++);
1463 kaddr
= (unsigned char *)page_address(page
);
1465 /* Step 1: restore nat cache */
1466 seg_i
= CURSEG_I(sbi
, CURSEG_HOT_DATA
);
1467 memcpy(&seg_i
->sum_blk
->n_nats
, kaddr
, SUM_JOURNAL_SIZE
);
1469 /* Step 2: restore sit cache */
1470 seg_i
= CURSEG_I(sbi
, CURSEG_COLD_DATA
);
1471 memcpy(&seg_i
->sum_blk
->n_sits
, kaddr
+ SUM_JOURNAL_SIZE
,
1473 offset
= 2 * SUM_JOURNAL_SIZE
;
1475 /* Step 3: restore summary entries */
1476 for (i
= CURSEG_HOT_DATA
; i
<= CURSEG_COLD_DATA
; i
++) {
1477 unsigned short blk_off
;
1480 seg_i
= CURSEG_I(sbi
, i
);
1481 segno
= le32_to_cpu(ckpt
->cur_data_segno
[i
]);
1482 blk_off
= le16_to_cpu(ckpt
->cur_data_blkoff
[i
]);
1483 seg_i
->next_segno
= segno
;
1484 reset_curseg(sbi
, i
, 0);
1485 seg_i
->alloc_type
= ckpt
->alloc_type
[i
];
1486 seg_i
->next_blkoff
= blk_off
;
1488 if (seg_i
->alloc_type
== SSR
)
1489 blk_off
= sbi
->blocks_per_seg
;
1491 for (j
= 0; j
< blk_off
; j
++) {
1492 struct f2fs_summary
*s
;
1493 s
= (struct f2fs_summary
*)(kaddr
+ offset
);
1494 seg_i
->sum_blk
->entries
[j
] = *s
;
1495 offset
+= SUMMARY_SIZE
;
1496 if (offset
+ SUMMARY_SIZE
<= PAGE_CACHE_SIZE
-
1500 f2fs_put_page(page
, 1);
1503 page
= get_meta_page(sbi
, start
++);
1504 kaddr
= (unsigned char *)page_address(page
);
1508 f2fs_put_page(page
, 1);
1512 static int read_normal_summaries(struct f2fs_sb_info
*sbi
, int type
)
1514 struct f2fs_checkpoint
*ckpt
= F2FS_CKPT(sbi
);
1515 struct f2fs_summary_block
*sum
;
1516 struct curseg_info
*curseg
;
1518 unsigned short blk_off
;
1519 unsigned int segno
= 0;
1520 block_t blk_addr
= 0;
1522 /* get segment number and block addr */
1523 if (IS_DATASEG(type
)) {
1524 segno
= le32_to_cpu(ckpt
->cur_data_segno
[type
]);
1525 blk_off
= le16_to_cpu(ckpt
->cur_data_blkoff
[type
-
1527 if (__exist_node_summaries(sbi
))
1528 blk_addr
= sum_blk_addr(sbi
, NR_CURSEG_TYPE
, type
);
1530 blk_addr
= sum_blk_addr(sbi
, NR_CURSEG_DATA_TYPE
, type
);
1532 segno
= le32_to_cpu(ckpt
->cur_node_segno
[type
-
1534 blk_off
= le16_to_cpu(ckpt
->cur_node_blkoff
[type
-
1536 if (__exist_node_summaries(sbi
))
1537 blk_addr
= sum_blk_addr(sbi
, NR_CURSEG_NODE_TYPE
,
1538 type
- CURSEG_HOT_NODE
);
1540 blk_addr
= GET_SUM_BLOCK(sbi
, segno
);
1543 new = get_meta_page(sbi
, blk_addr
);
1544 sum
= (struct f2fs_summary_block
*)page_address(new);
1546 if (IS_NODESEG(type
)) {
1547 if (__exist_node_summaries(sbi
)) {
1548 struct f2fs_summary
*ns
= &sum
->entries
[0];
1550 for (i
= 0; i
< sbi
->blocks_per_seg
; i
++, ns
++) {
1552 ns
->ofs_in_node
= 0;
1557 err
= restore_node_summary(sbi
, segno
, sum
);
1559 f2fs_put_page(new, 1);
1565 /* set uncompleted segment to curseg */
1566 curseg
= CURSEG_I(sbi
, type
);
1567 mutex_lock(&curseg
->curseg_mutex
);
1568 memcpy(curseg
->sum_blk
, sum
, PAGE_CACHE_SIZE
);
1569 curseg
->next_segno
= segno
;
1570 reset_curseg(sbi
, type
, 0);
1571 curseg
->alloc_type
= ckpt
->alloc_type
[type
];
1572 curseg
->next_blkoff
= blk_off
;
1573 mutex_unlock(&curseg
->curseg_mutex
);
1574 f2fs_put_page(new, 1);
1578 static int restore_curseg_summaries(struct f2fs_sb_info
*sbi
)
1580 int type
= CURSEG_HOT_DATA
;
1583 if (is_set_ckpt_flags(F2FS_CKPT(sbi
), CP_COMPACT_SUM_FLAG
)) {
1584 int npages
= npages_for_summary_flush(sbi
, true);
1587 ra_meta_pages(sbi
, start_sum_block(sbi
), npages
,
1590 /* restore for compacted data summary */
1591 if (read_compacted_summaries(sbi
))
1593 type
= CURSEG_HOT_NODE
;
1596 if (__exist_node_summaries(sbi
))
1597 ra_meta_pages(sbi
, sum_blk_addr(sbi
, NR_CURSEG_TYPE
, type
),
1598 NR_CURSEG_TYPE
- type
, META_CP
);
1600 for (; type
<= CURSEG_COLD_NODE
; type
++) {
1601 err
= read_normal_summaries(sbi
, type
);
1609 static void write_compacted_summaries(struct f2fs_sb_info
*sbi
, block_t blkaddr
)
1612 unsigned char *kaddr
;
1613 struct f2fs_summary
*summary
;
1614 struct curseg_info
*seg_i
;
1615 int written_size
= 0;
1618 page
= grab_meta_page(sbi
, blkaddr
++);
1619 kaddr
= (unsigned char *)page_address(page
);
1621 /* Step 1: write nat cache */
1622 seg_i
= CURSEG_I(sbi
, CURSEG_HOT_DATA
);
1623 memcpy(kaddr
, &seg_i
->sum_blk
->n_nats
, SUM_JOURNAL_SIZE
);
1624 written_size
+= SUM_JOURNAL_SIZE
;
1626 /* Step 2: write sit cache */
1627 seg_i
= CURSEG_I(sbi
, CURSEG_COLD_DATA
);
1628 memcpy(kaddr
+ written_size
, &seg_i
->sum_blk
->n_sits
,
1630 written_size
+= SUM_JOURNAL_SIZE
;
1632 /* Step 3: write summary entries */
1633 for (i
= CURSEG_HOT_DATA
; i
<= CURSEG_COLD_DATA
; i
++) {
1634 unsigned short blkoff
;
1635 seg_i
= CURSEG_I(sbi
, i
);
1636 if (sbi
->ckpt
->alloc_type
[i
] == SSR
)
1637 blkoff
= sbi
->blocks_per_seg
;
1639 blkoff
= curseg_blkoff(sbi
, i
);
1641 for (j
= 0; j
< blkoff
; j
++) {
1643 page
= grab_meta_page(sbi
, blkaddr
++);
1644 kaddr
= (unsigned char *)page_address(page
);
1647 summary
= (struct f2fs_summary
*)(kaddr
+ written_size
);
1648 *summary
= seg_i
->sum_blk
->entries
[j
];
1649 written_size
+= SUMMARY_SIZE
;
1651 if (written_size
+ SUMMARY_SIZE
<= PAGE_CACHE_SIZE
-
1655 set_page_dirty(page
);
1656 f2fs_put_page(page
, 1);
1661 set_page_dirty(page
);
1662 f2fs_put_page(page
, 1);
1666 static void write_normal_summaries(struct f2fs_sb_info
*sbi
,
1667 block_t blkaddr
, int type
)
1670 if (IS_DATASEG(type
))
1671 end
= type
+ NR_CURSEG_DATA_TYPE
;
1673 end
= type
+ NR_CURSEG_NODE_TYPE
;
1675 for (i
= type
; i
< end
; i
++) {
1676 struct curseg_info
*sum
= CURSEG_I(sbi
, i
);
1677 mutex_lock(&sum
->curseg_mutex
);
1678 write_sum_page(sbi
, sum
->sum_blk
, blkaddr
+ (i
- type
));
1679 mutex_unlock(&sum
->curseg_mutex
);
1683 void write_data_summaries(struct f2fs_sb_info
*sbi
, block_t start_blk
)
1685 if (is_set_ckpt_flags(F2FS_CKPT(sbi
), CP_COMPACT_SUM_FLAG
))
1686 write_compacted_summaries(sbi
, start_blk
);
1688 write_normal_summaries(sbi
, start_blk
, CURSEG_HOT_DATA
);
1691 void write_node_summaries(struct f2fs_sb_info
*sbi
, block_t start_blk
)
1693 write_normal_summaries(sbi
, start_blk
, CURSEG_HOT_NODE
);
1696 int lookup_journal_in_cursum(struct f2fs_summary_block
*sum
, int type
,
1697 unsigned int val
, int alloc
)
1701 if (type
== NAT_JOURNAL
) {
1702 for (i
= 0; i
< nats_in_cursum(sum
); i
++) {
1703 if (le32_to_cpu(nid_in_journal(sum
, i
)) == val
)
1706 if (alloc
&& nats_in_cursum(sum
) < NAT_JOURNAL_ENTRIES
)
1707 return update_nats_in_cursum(sum
, 1);
1708 } else if (type
== SIT_JOURNAL
) {
1709 for (i
= 0; i
< sits_in_cursum(sum
); i
++)
1710 if (le32_to_cpu(segno_in_journal(sum
, i
)) == val
)
1712 if (alloc
&& sits_in_cursum(sum
) < SIT_JOURNAL_ENTRIES
)
1713 return update_sits_in_cursum(sum
, 1);
1718 static struct page
*get_current_sit_page(struct f2fs_sb_info
*sbi
,
1721 return get_meta_page(sbi
, current_sit_addr(sbi
, segno
));
1724 static struct page
*get_next_sit_page(struct f2fs_sb_info
*sbi
,
1727 struct sit_info
*sit_i
= SIT_I(sbi
);
1728 struct page
*src_page
, *dst_page
;
1729 pgoff_t src_off
, dst_off
;
1730 void *src_addr
, *dst_addr
;
1732 src_off
= current_sit_addr(sbi
, start
);
1733 dst_off
= next_sit_addr(sbi
, src_off
);
1735 /* get current sit block page without lock */
1736 src_page
= get_meta_page(sbi
, src_off
);
1737 dst_page
= grab_meta_page(sbi
, dst_off
);
1738 f2fs_bug_on(sbi
, PageDirty(src_page
));
1740 src_addr
= page_address(src_page
);
1741 dst_addr
= page_address(dst_page
);
1742 memcpy(dst_addr
, src_addr
, PAGE_CACHE_SIZE
);
1744 set_page_dirty(dst_page
);
1745 f2fs_put_page(src_page
, 1);
1747 set_to_next_sit(sit_i
, start
);
1752 static struct sit_entry_set
*grab_sit_entry_set(void)
1754 struct sit_entry_set
*ses
=
1755 f2fs_kmem_cache_alloc(sit_entry_set_slab
, GFP_NOFS
);
1758 INIT_LIST_HEAD(&ses
->set_list
);
1762 static void release_sit_entry_set(struct sit_entry_set
*ses
)
1764 list_del(&ses
->set_list
);
1765 kmem_cache_free(sit_entry_set_slab
, ses
);
1768 static void adjust_sit_entry_set(struct sit_entry_set
*ses
,
1769 struct list_head
*head
)
1771 struct sit_entry_set
*next
= ses
;
1773 if (list_is_last(&ses
->set_list
, head
))
1776 list_for_each_entry_continue(next
, head
, set_list
)
1777 if (ses
->entry_cnt
<= next
->entry_cnt
)
1780 list_move_tail(&ses
->set_list
, &next
->set_list
);
1783 static void add_sit_entry(unsigned int segno
, struct list_head
*head
)
1785 struct sit_entry_set
*ses
;
1786 unsigned int start_segno
= START_SEGNO(segno
);
1788 list_for_each_entry(ses
, head
, set_list
) {
1789 if (ses
->start_segno
== start_segno
) {
1791 adjust_sit_entry_set(ses
, head
);
1796 ses
= grab_sit_entry_set();
1798 ses
->start_segno
= start_segno
;
1800 list_add(&ses
->set_list
, head
);
1803 static void add_sits_in_set(struct f2fs_sb_info
*sbi
)
1805 struct f2fs_sm_info
*sm_info
= SM_I(sbi
);
1806 struct list_head
*set_list
= &sm_info
->sit_entry_set
;
1807 unsigned long *bitmap
= SIT_I(sbi
)->dirty_sentries_bitmap
;
1810 for_each_set_bit(segno
, bitmap
, MAIN_SEGS(sbi
))
1811 add_sit_entry(segno
, set_list
);
1814 static void remove_sits_in_journal(struct f2fs_sb_info
*sbi
)
1816 struct curseg_info
*curseg
= CURSEG_I(sbi
, CURSEG_COLD_DATA
);
1817 struct f2fs_summary_block
*sum
= curseg
->sum_blk
;
1820 for (i
= sits_in_cursum(sum
) - 1; i
>= 0; i
--) {
1824 segno
= le32_to_cpu(segno_in_journal(sum
, i
));
1825 dirtied
= __mark_sit_entry_dirty(sbi
, segno
);
1828 add_sit_entry(segno
, &SM_I(sbi
)->sit_entry_set
);
1830 update_sits_in_cursum(sum
, -sits_in_cursum(sum
));
1834 * CP calls this function, which flushes SIT entries including sit_journal,
1835 * and moves prefree segs to free segs.
1837 void flush_sit_entries(struct f2fs_sb_info
*sbi
, struct cp_control
*cpc
)
1839 struct sit_info
*sit_i
= SIT_I(sbi
);
1840 unsigned long *bitmap
= sit_i
->dirty_sentries_bitmap
;
1841 struct curseg_info
*curseg
= CURSEG_I(sbi
, CURSEG_COLD_DATA
);
1842 struct f2fs_summary_block
*sum
= curseg
->sum_blk
;
1843 struct sit_entry_set
*ses
, *tmp
;
1844 struct list_head
*head
= &SM_I(sbi
)->sit_entry_set
;
1845 bool to_journal
= true;
1846 struct seg_entry
*se
;
1848 mutex_lock(&curseg
->curseg_mutex
);
1849 mutex_lock(&sit_i
->sentry_lock
);
1851 if (!sit_i
->dirty_sentries
)
1855 * add and account sit entries of dirty bitmap in sit entry
1858 add_sits_in_set(sbi
);
1861 * if there are no enough space in journal to store dirty sit
1862 * entries, remove all entries from journal and add and account
1863 * them in sit entry set.
1865 if (!__has_cursum_space(sum
, sit_i
->dirty_sentries
, SIT_JOURNAL
))
1866 remove_sits_in_journal(sbi
);
1869 * there are two steps to flush sit entries:
1870 * #1, flush sit entries to journal in current cold data summary block.
1871 * #2, flush sit entries to sit page.
1873 list_for_each_entry_safe(ses
, tmp
, head
, set_list
) {
1874 struct page
*page
= NULL
;
1875 struct f2fs_sit_block
*raw_sit
= NULL
;
1876 unsigned int start_segno
= ses
->start_segno
;
1877 unsigned int end
= min(start_segno
+ SIT_ENTRY_PER_BLOCK
,
1878 (unsigned long)MAIN_SEGS(sbi
));
1879 unsigned int segno
= start_segno
;
1882 !__has_cursum_space(sum
, ses
->entry_cnt
, SIT_JOURNAL
))
1886 page
= get_next_sit_page(sbi
, start_segno
);
1887 raw_sit
= page_address(page
);
1890 /* flush dirty sit entries in region of current sit set */
1891 for_each_set_bit_from(segno
, bitmap
, end
) {
1892 int offset
, sit_offset
;
1894 se
= get_seg_entry(sbi
, segno
);
1896 /* add discard candidates */
1897 if (cpc
->reason
!= CP_DISCARD
) {
1898 cpc
->trim_start
= segno
;
1899 add_discard_addrs(sbi
, cpc
);
1903 offset
= lookup_journal_in_cursum(sum
,
1904 SIT_JOURNAL
, segno
, 1);
1905 f2fs_bug_on(sbi
, offset
< 0);
1906 segno_in_journal(sum
, offset
) =
1908 seg_info_to_raw_sit(se
,
1909 &sit_in_journal(sum
, offset
));
1911 sit_offset
= SIT_ENTRY_OFFSET(sit_i
, segno
);
1912 seg_info_to_raw_sit(se
,
1913 &raw_sit
->entries
[sit_offset
]);
1916 __clear_bit(segno
, bitmap
);
1917 sit_i
->dirty_sentries
--;
1922 f2fs_put_page(page
, 1);
1924 f2fs_bug_on(sbi
, ses
->entry_cnt
);
1925 release_sit_entry_set(ses
);
1928 f2fs_bug_on(sbi
, !list_empty(head
));
1929 f2fs_bug_on(sbi
, sit_i
->dirty_sentries
);
1931 if (cpc
->reason
== CP_DISCARD
) {
1932 for (; cpc
->trim_start
<= cpc
->trim_end
; cpc
->trim_start
++)
1933 add_discard_addrs(sbi
, cpc
);
1935 mutex_unlock(&sit_i
->sentry_lock
);
1936 mutex_unlock(&curseg
->curseg_mutex
);
1938 set_prefree_as_free_segments(sbi
);
1941 static int build_sit_info(struct f2fs_sb_info
*sbi
)
1943 struct f2fs_super_block
*raw_super
= F2FS_RAW_SUPER(sbi
);
1944 struct f2fs_checkpoint
*ckpt
= F2FS_CKPT(sbi
);
1945 struct sit_info
*sit_i
;
1946 unsigned int sit_segs
, start
;
1947 char *src_bitmap
, *dst_bitmap
;
1948 unsigned int bitmap_size
;
1950 /* allocate memory for SIT information */
1951 sit_i
= kzalloc(sizeof(struct sit_info
), GFP_KERNEL
);
1955 SM_I(sbi
)->sit_info
= sit_i
;
1957 sit_i
->sentries
= f2fs_kvzalloc(MAIN_SEGS(sbi
) *
1958 sizeof(struct seg_entry
), GFP_KERNEL
);
1959 if (!sit_i
->sentries
)
1962 bitmap_size
= f2fs_bitmap_size(MAIN_SEGS(sbi
));
1963 sit_i
->dirty_sentries_bitmap
= f2fs_kvzalloc(bitmap_size
, GFP_KERNEL
);
1964 if (!sit_i
->dirty_sentries_bitmap
)
1967 for (start
= 0; start
< MAIN_SEGS(sbi
); start
++) {
1968 sit_i
->sentries
[start
].cur_valid_map
1969 = kzalloc(SIT_VBLOCK_MAP_SIZE
, GFP_KERNEL
);
1970 sit_i
->sentries
[start
].ckpt_valid_map
1971 = kzalloc(SIT_VBLOCK_MAP_SIZE
, GFP_KERNEL
);
1972 sit_i
->sentries
[start
].discard_map
1973 = kzalloc(SIT_VBLOCK_MAP_SIZE
, GFP_KERNEL
);
1974 if (!sit_i
->sentries
[start
].cur_valid_map
||
1975 !sit_i
->sentries
[start
].ckpt_valid_map
||
1976 !sit_i
->sentries
[start
].discard_map
)
1980 sit_i
->tmp_map
= kzalloc(SIT_VBLOCK_MAP_SIZE
, GFP_KERNEL
);
1981 if (!sit_i
->tmp_map
)
1984 if (sbi
->segs_per_sec
> 1) {
1985 sit_i
->sec_entries
= f2fs_kvzalloc(MAIN_SECS(sbi
) *
1986 sizeof(struct sec_entry
), GFP_KERNEL
);
1987 if (!sit_i
->sec_entries
)
1991 /* get information related with SIT */
1992 sit_segs
= le32_to_cpu(raw_super
->segment_count_sit
) >> 1;
1994 /* setup SIT bitmap from ckeckpoint pack */
1995 bitmap_size
= __bitmap_size(sbi
, SIT_BITMAP
);
1996 src_bitmap
= __bitmap_ptr(sbi
, SIT_BITMAP
);
1998 dst_bitmap
= kmemdup(src_bitmap
, bitmap_size
, GFP_KERNEL
);
2002 /* init SIT information */
2003 sit_i
->s_ops
= &default_salloc_ops
;
2005 sit_i
->sit_base_addr
= le32_to_cpu(raw_super
->sit_blkaddr
);
2006 sit_i
->sit_blocks
= sit_segs
<< sbi
->log_blocks_per_seg
;
2007 sit_i
->written_valid_blocks
= le64_to_cpu(ckpt
->valid_block_count
);
2008 sit_i
->sit_bitmap
= dst_bitmap
;
2009 sit_i
->bitmap_size
= bitmap_size
;
2010 sit_i
->dirty_sentries
= 0;
2011 sit_i
->sents_per_block
= SIT_ENTRY_PER_BLOCK
;
2012 sit_i
->elapsed_time
= le64_to_cpu(sbi
->ckpt
->elapsed_time
);
2013 sit_i
->mounted_time
= CURRENT_TIME_SEC
.tv_sec
;
2014 mutex_init(&sit_i
->sentry_lock
);
2018 static int build_free_segmap(struct f2fs_sb_info
*sbi
)
2020 struct free_segmap_info
*free_i
;
2021 unsigned int bitmap_size
, sec_bitmap_size
;
2023 /* allocate memory for free segmap information */
2024 free_i
= kzalloc(sizeof(struct free_segmap_info
), GFP_KERNEL
);
2028 SM_I(sbi
)->free_info
= free_i
;
2030 bitmap_size
= f2fs_bitmap_size(MAIN_SEGS(sbi
));
2031 free_i
->free_segmap
= f2fs_kvmalloc(bitmap_size
, GFP_KERNEL
);
2032 if (!free_i
->free_segmap
)
2035 sec_bitmap_size
= f2fs_bitmap_size(MAIN_SECS(sbi
));
2036 free_i
->free_secmap
= f2fs_kvmalloc(sec_bitmap_size
, GFP_KERNEL
);
2037 if (!free_i
->free_secmap
)
2040 /* set all segments as dirty temporarily */
2041 memset(free_i
->free_segmap
, 0xff, bitmap_size
);
2042 memset(free_i
->free_secmap
, 0xff, sec_bitmap_size
);
2044 /* init free segmap information */
2045 free_i
->start_segno
= GET_SEGNO_FROM_SEG0(sbi
, MAIN_BLKADDR(sbi
));
2046 free_i
->free_segments
= 0;
2047 free_i
->free_sections
= 0;
2048 spin_lock_init(&free_i
->segmap_lock
);
2052 static int build_curseg(struct f2fs_sb_info
*sbi
)
2054 struct curseg_info
*array
;
2057 array
= kcalloc(NR_CURSEG_TYPE
, sizeof(*array
), GFP_KERNEL
);
2061 SM_I(sbi
)->curseg_array
= array
;
2063 for (i
= 0; i
< NR_CURSEG_TYPE
; i
++) {
2064 mutex_init(&array
[i
].curseg_mutex
);
2065 array
[i
].sum_blk
= kzalloc(PAGE_CACHE_SIZE
, GFP_KERNEL
);
2066 if (!array
[i
].sum_blk
)
2068 array
[i
].segno
= NULL_SEGNO
;
2069 array
[i
].next_blkoff
= 0;
2071 return restore_curseg_summaries(sbi
);
2074 static void build_sit_entries(struct f2fs_sb_info
*sbi
)
2076 struct sit_info
*sit_i
= SIT_I(sbi
);
2077 struct curseg_info
*curseg
= CURSEG_I(sbi
, CURSEG_COLD_DATA
);
2078 struct f2fs_summary_block
*sum
= curseg
->sum_blk
;
2079 int sit_blk_cnt
= SIT_BLK_CNT(sbi
);
2080 unsigned int i
, start
, end
;
2081 unsigned int readed
, start_blk
= 0;
2082 int nrpages
= MAX_BIO_BLOCKS(sbi
);
2085 readed
= ra_meta_pages(sbi
, start_blk
, nrpages
, META_SIT
);
2087 start
= start_blk
* sit_i
->sents_per_block
;
2088 end
= (start_blk
+ readed
) * sit_i
->sents_per_block
;
2090 for (; start
< end
&& start
< MAIN_SEGS(sbi
); start
++) {
2091 struct seg_entry
*se
= &sit_i
->sentries
[start
];
2092 struct f2fs_sit_block
*sit_blk
;
2093 struct f2fs_sit_entry sit
;
2096 mutex_lock(&curseg
->curseg_mutex
);
2097 for (i
= 0; i
< sits_in_cursum(sum
); i
++) {
2098 if (le32_to_cpu(segno_in_journal(sum
, i
))
2100 sit
= sit_in_journal(sum
, i
);
2101 mutex_unlock(&curseg
->curseg_mutex
);
2105 mutex_unlock(&curseg
->curseg_mutex
);
2107 page
= get_current_sit_page(sbi
, start
);
2108 sit_blk
= (struct f2fs_sit_block
*)page_address(page
);
2109 sit
= sit_blk
->entries
[SIT_ENTRY_OFFSET(sit_i
, start
)];
2110 f2fs_put_page(page
, 1);
2112 check_block_count(sbi
, start
, &sit
);
2113 seg_info_from_raw_sit(se
, &sit
);
2115 /* build discard map only one time */
2116 memcpy(se
->discard_map
, se
->cur_valid_map
, SIT_VBLOCK_MAP_SIZE
);
2117 sbi
->discard_blks
+= sbi
->blocks_per_seg
- se
->valid_blocks
;
2119 if (sbi
->segs_per_sec
> 1) {
2120 struct sec_entry
*e
= get_sec_entry(sbi
, start
);
2121 e
->valid_blocks
+= se
->valid_blocks
;
2124 start_blk
+= readed
;
2125 } while (start_blk
< sit_blk_cnt
);
2128 static void init_free_segmap(struct f2fs_sb_info
*sbi
)
2133 for (start
= 0; start
< MAIN_SEGS(sbi
); start
++) {
2134 struct seg_entry
*sentry
= get_seg_entry(sbi
, start
);
2135 if (!sentry
->valid_blocks
)
2136 __set_free(sbi
, start
);
2139 /* set use the current segments */
2140 for (type
= CURSEG_HOT_DATA
; type
<= CURSEG_COLD_NODE
; type
++) {
2141 struct curseg_info
*curseg_t
= CURSEG_I(sbi
, type
);
2142 __set_test_and_inuse(sbi
, curseg_t
->segno
);
2146 static void init_dirty_segmap(struct f2fs_sb_info
*sbi
)
2148 struct dirty_seglist_info
*dirty_i
= DIRTY_I(sbi
);
2149 struct free_segmap_info
*free_i
= FREE_I(sbi
);
2150 unsigned int segno
= 0, offset
= 0;
2151 unsigned short valid_blocks
;
2154 /* find dirty segment based on free segmap */
2155 segno
= find_next_inuse(free_i
, MAIN_SEGS(sbi
), offset
);
2156 if (segno
>= MAIN_SEGS(sbi
))
2159 valid_blocks
= get_valid_blocks(sbi
, segno
, 0);
2160 if (valid_blocks
== sbi
->blocks_per_seg
|| !valid_blocks
)
2162 if (valid_blocks
> sbi
->blocks_per_seg
) {
2163 f2fs_bug_on(sbi
, 1);
2166 mutex_lock(&dirty_i
->seglist_lock
);
2167 __locate_dirty_segment(sbi
, segno
, DIRTY
);
2168 mutex_unlock(&dirty_i
->seglist_lock
);
2172 static int init_victim_secmap(struct f2fs_sb_info
*sbi
)
2174 struct dirty_seglist_info
*dirty_i
= DIRTY_I(sbi
);
2175 unsigned int bitmap_size
= f2fs_bitmap_size(MAIN_SECS(sbi
));
2177 dirty_i
->victim_secmap
= f2fs_kvzalloc(bitmap_size
, GFP_KERNEL
);
2178 if (!dirty_i
->victim_secmap
)
2183 static int build_dirty_segmap(struct f2fs_sb_info
*sbi
)
2185 struct dirty_seglist_info
*dirty_i
;
2186 unsigned int bitmap_size
, i
;
2188 /* allocate memory for dirty segments list information */
2189 dirty_i
= kzalloc(sizeof(struct dirty_seglist_info
), GFP_KERNEL
);
2193 SM_I(sbi
)->dirty_info
= dirty_i
;
2194 mutex_init(&dirty_i
->seglist_lock
);
2196 bitmap_size
= f2fs_bitmap_size(MAIN_SEGS(sbi
));
2198 for (i
= 0; i
< NR_DIRTY_TYPE
; i
++) {
2199 dirty_i
->dirty_segmap
[i
] = f2fs_kvzalloc(bitmap_size
, GFP_KERNEL
);
2200 if (!dirty_i
->dirty_segmap
[i
])
2204 init_dirty_segmap(sbi
);
2205 return init_victim_secmap(sbi
);
2209 * Update min, max modified time for cost-benefit GC algorithm
2211 static void init_min_max_mtime(struct f2fs_sb_info
*sbi
)
2213 struct sit_info
*sit_i
= SIT_I(sbi
);
2216 mutex_lock(&sit_i
->sentry_lock
);
2218 sit_i
->min_mtime
= LLONG_MAX
;
2220 for (segno
= 0; segno
< MAIN_SEGS(sbi
); segno
+= sbi
->segs_per_sec
) {
2222 unsigned long long mtime
= 0;
2224 for (i
= 0; i
< sbi
->segs_per_sec
; i
++)
2225 mtime
+= get_seg_entry(sbi
, segno
+ i
)->mtime
;
2227 mtime
= div_u64(mtime
, sbi
->segs_per_sec
);
2229 if (sit_i
->min_mtime
> mtime
)
2230 sit_i
->min_mtime
= mtime
;
2232 sit_i
->max_mtime
= get_mtime(sbi
);
2233 mutex_unlock(&sit_i
->sentry_lock
);
2236 int build_segment_manager(struct f2fs_sb_info
*sbi
)
2238 struct f2fs_super_block
*raw_super
= F2FS_RAW_SUPER(sbi
);
2239 struct f2fs_checkpoint
*ckpt
= F2FS_CKPT(sbi
);
2240 struct f2fs_sm_info
*sm_info
;
2243 sm_info
= kzalloc(sizeof(struct f2fs_sm_info
), GFP_KERNEL
);
2248 sbi
->sm_info
= sm_info
;
2249 sm_info
->seg0_blkaddr
= le32_to_cpu(raw_super
->segment0_blkaddr
);
2250 sm_info
->main_blkaddr
= le32_to_cpu(raw_super
->main_blkaddr
);
2251 sm_info
->segment_count
= le32_to_cpu(raw_super
->segment_count
);
2252 sm_info
->reserved_segments
= le32_to_cpu(ckpt
->rsvd_segment_count
);
2253 sm_info
->ovp_segments
= le32_to_cpu(ckpt
->overprov_segment_count
);
2254 sm_info
->main_segments
= le32_to_cpu(raw_super
->segment_count_main
);
2255 sm_info
->ssa_blkaddr
= le32_to_cpu(raw_super
->ssa_blkaddr
);
2256 sm_info
->rec_prefree_segments
= sm_info
->main_segments
*
2257 DEF_RECLAIM_PREFREE_SEGMENTS
/ 100;
2258 sm_info
->ipu_policy
= 1 << F2FS_IPU_FSYNC
;
2259 sm_info
->min_ipu_util
= DEF_MIN_IPU_UTIL
;
2260 sm_info
->min_fsync_blocks
= DEF_MIN_FSYNC_BLOCKS
;
2262 INIT_LIST_HEAD(&sm_info
->discard_list
);
2263 sm_info
->nr_discards
= 0;
2264 sm_info
->max_discards
= 0;
2266 sm_info
->trim_sections
= DEF_BATCHED_TRIM_SECTIONS
;
2268 INIT_LIST_HEAD(&sm_info
->sit_entry_set
);
2270 if (test_opt(sbi
, FLUSH_MERGE
) && !f2fs_readonly(sbi
->sb
)) {
2271 err
= create_flush_cmd_control(sbi
);
2276 err
= build_sit_info(sbi
);
2279 err
= build_free_segmap(sbi
);
2282 err
= build_curseg(sbi
);
2286 /* reinit free segmap based on SIT */
2287 build_sit_entries(sbi
);
2289 init_free_segmap(sbi
);
2290 err
= build_dirty_segmap(sbi
);
2294 init_min_max_mtime(sbi
);
2298 static void discard_dirty_segmap(struct f2fs_sb_info
*sbi
,
2299 enum dirty_type dirty_type
)
2301 struct dirty_seglist_info
*dirty_i
= DIRTY_I(sbi
);
2303 mutex_lock(&dirty_i
->seglist_lock
);
2304 kvfree(dirty_i
->dirty_segmap
[dirty_type
]);
2305 dirty_i
->nr_dirty
[dirty_type
] = 0;
2306 mutex_unlock(&dirty_i
->seglist_lock
);
2309 static void destroy_victim_secmap(struct f2fs_sb_info
*sbi
)
2311 struct dirty_seglist_info
*dirty_i
= DIRTY_I(sbi
);
2312 kvfree(dirty_i
->victim_secmap
);
2315 static void destroy_dirty_segmap(struct f2fs_sb_info
*sbi
)
2317 struct dirty_seglist_info
*dirty_i
= DIRTY_I(sbi
);
2323 /* discard pre-free/dirty segments list */
2324 for (i
= 0; i
< NR_DIRTY_TYPE
; i
++)
2325 discard_dirty_segmap(sbi
, i
);
2327 destroy_victim_secmap(sbi
);
2328 SM_I(sbi
)->dirty_info
= NULL
;
2332 static void destroy_curseg(struct f2fs_sb_info
*sbi
)
2334 struct curseg_info
*array
= SM_I(sbi
)->curseg_array
;
2339 SM_I(sbi
)->curseg_array
= NULL
;
2340 for (i
= 0; i
< NR_CURSEG_TYPE
; i
++)
2341 kfree(array
[i
].sum_blk
);
2345 static void destroy_free_segmap(struct f2fs_sb_info
*sbi
)
2347 struct free_segmap_info
*free_i
= SM_I(sbi
)->free_info
;
2350 SM_I(sbi
)->free_info
= NULL
;
2351 kvfree(free_i
->free_segmap
);
2352 kvfree(free_i
->free_secmap
);
2356 static void destroy_sit_info(struct f2fs_sb_info
*sbi
)
2358 struct sit_info
*sit_i
= SIT_I(sbi
);
2364 if (sit_i
->sentries
) {
2365 for (start
= 0; start
< MAIN_SEGS(sbi
); start
++) {
2366 kfree(sit_i
->sentries
[start
].cur_valid_map
);
2367 kfree(sit_i
->sentries
[start
].ckpt_valid_map
);
2368 kfree(sit_i
->sentries
[start
].discard_map
);
2371 kfree(sit_i
->tmp_map
);
2373 kvfree(sit_i
->sentries
);
2374 kvfree(sit_i
->sec_entries
);
2375 kvfree(sit_i
->dirty_sentries_bitmap
);
2377 SM_I(sbi
)->sit_info
= NULL
;
2378 kfree(sit_i
->sit_bitmap
);
2382 void destroy_segment_manager(struct f2fs_sb_info
*sbi
)
2384 struct f2fs_sm_info
*sm_info
= SM_I(sbi
);
2388 destroy_flush_cmd_control(sbi
);
2389 destroy_dirty_segmap(sbi
);
2390 destroy_curseg(sbi
);
2391 destroy_free_segmap(sbi
);
2392 destroy_sit_info(sbi
);
2393 sbi
->sm_info
= NULL
;
2397 int __init
create_segment_manager_caches(void)
2399 discard_entry_slab
= f2fs_kmem_cache_create("discard_entry",
2400 sizeof(struct discard_entry
));
2401 if (!discard_entry_slab
)
2404 sit_entry_set_slab
= f2fs_kmem_cache_create("sit_entry_set",
2405 sizeof(struct sit_entry_set
));
2406 if (!sit_entry_set_slab
)
2407 goto destory_discard_entry
;
2409 inmem_entry_slab
= f2fs_kmem_cache_create("inmem_page_entry",
2410 sizeof(struct inmem_pages
));
2411 if (!inmem_entry_slab
)
2412 goto destroy_sit_entry_set
;
2415 destroy_sit_entry_set
:
2416 kmem_cache_destroy(sit_entry_set_slab
);
2417 destory_discard_entry
:
2418 kmem_cache_destroy(discard_entry_slab
);
2423 void destroy_segment_manager_caches(void)
2425 kmem_cache_destroy(sit_entry_set_slab
);
2426 kmem_cache_destroy(discard_entry_slab
);
2427 kmem_cache_destroy(inmem_entry_slab
);