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>
18 #include <linux/timer.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
;
32 static unsigned long __reverse_ulong(unsigned char *str
)
34 unsigned long tmp
= 0;
35 int shift
= 24, idx
= 0;
37 #if BITS_PER_LONG == 64
41 tmp
|= (unsigned long)str
[idx
++] << shift
;
42 shift
-= BITS_PER_BYTE
;
48 * __reverse_ffs is copied from include/asm-generic/bitops/__ffs.h since
49 * MSB and LSB are reversed in a byte by f2fs_set_bit.
51 static inline unsigned long __reverse_ffs(unsigned long word
)
55 #if BITS_PER_LONG == 64
56 if ((word
& 0xffffffff00000000UL
) == 0)
61 if ((word
& 0xffff0000) == 0)
66 if ((word
& 0xff00) == 0)
71 if ((word
& 0xf0) == 0)
76 if ((word
& 0xc) == 0)
81 if ((word
& 0x2) == 0)
87 * __find_rev_next(_zero)_bit is copied from lib/find_next_bit.c because
88 * f2fs_set_bit makes MSB and LSB reversed in a byte.
89 * @size must be integral times of unsigned long.
92 * f2fs_set_bit(0, bitmap) => 1000 0000
93 * f2fs_set_bit(7, bitmap) => 0000 0001
95 static unsigned long __find_rev_next_bit(const unsigned long *addr
,
96 unsigned long size
, unsigned long offset
)
98 const unsigned long *p
= addr
+ BIT_WORD(offset
);
99 unsigned long result
= size
;
105 size
-= (offset
& ~(BITS_PER_LONG
- 1));
106 offset
%= BITS_PER_LONG
;
112 tmp
= __reverse_ulong((unsigned char *)p
);
114 tmp
&= ~0UL >> offset
;
115 if (size
< BITS_PER_LONG
)
116 tmp
&= (~0UL << (BITS_PER_LONG
- size
));
120 if (size
<= BITS_PER_LONG
)
122 size
-= BITS_PER_LONG
;
128 return result
- size
+ __reverse_ffs(tmp
);
131 static unsigned long __find_rev_next_zero_bit(const unsigned long *addr
,
132 unsigned long size
, unsigned long offset
)
134 const unsigned long *p
= addr
+ BIT_WORD(offset
);
135 unsigned long result
= size
;
141 size
-= (offset
& ~(BITS_PER_LONG
- 1));
142 offset
%= BITS_PER_LONG
;
148 tmp
= __reverse_ulong((unsigned char *)p
);
151 tmp
|= ~0UL << (BITS_PER_LONG
- offset
);
152 if (size
< BITS_PER_LONG
)
157 if (size
<= BITS_PER_LONG
)
159 size
-= BITS_PER_LONG
;
165 return result
- size
+ __reverse_ffz(tmp
);
168 void register_inmem_page(struct inode
*inode
, struct page
*page
)
170 struct f2fs_inode_info
*fi
= F2FS_I(inode
);
171 struct inmem_pages
*new;
173 f2fs_trace_pid(page
);
175 set_page_private(page
, (unsigned long)ATOMIC_WRITTEN_PAGE
);
176 SetPagePrivate(page
);
178 new = f2fs_kmem_cache_alloc(inmem_entry_slab
, GFP_NOFS
);
180 /* add atomic page indices to the list */
182 INIT_LIST_HEAD(&new->list
);
184 /* increase reference count with clean state */
185 mutex_lock(&fi
->inmem_lock
);
187 list_add_tail(&new->list
, &fi
->inmem_pages
);
188 inc_page_count(F2FS_I_SB(inode
), F2FS_INMEM_PAGES
);
189 mutex_unlock(&fi
->inmem_lock
);
191 trace_f2fs_register_inmem_page(page
, INMEM
);
194 static int __revoke_inmem_pages(struct inode
*inode
,
195 struct list_head
*head
, bool drop
, bool recover
)
197 struct f2fs_sb_info
*sbi
= F2FS_I_SB(inode
);
198 struct inmem_pages
*cur
, *tmp
;
201 list_for_each_entry_safe(cur
, tmp
, head
, list
) {
202 struct page
*page
= cur
->page
;
205 trace_f2fs_commit_inmem_page(page
, INMEM_DROP
);
210 struct dnode_of_data dn
;
213 trace_f2fs_commit_inmem_page(page
, INMEM_REVOKE
);
215 set_new_dnode(&dn
, inode
, NULL
, NULL
, 0);
216 if (get_dnode_of_data(&dn
, page
->index
, LOOKUP_NODE
)) {
220 get_node_info(sbi
, dn
.nid
, &ni
);
221 f2fs_replace_block(sbi
, &dn
, dn
.data_blkaddr
,
222 cur
->old_addr
, ni
.version
, true, true);
226 /* we don't need to invalidate this in the sccessful status */
228 ClearPageUptodate(page
);
229 set_page_private(page
, 0);
230 ClearPagePrivate(page
);
231 f2fs_put_page(page
, 1);
233 list_del(&cur
->list
);
234 kmem_cache_free(inmem_entry_slab
, cur
);
235 dec_page_count(F2FS_I_SB(inode
), F2FS_INMEM_PAGES
);
240 void drop_inmem_pages(struct inode
*inode
)
242 struct f2fs_inode_info
*fi
= F2FS_I(inode
);
244 clear_inode_flag(inode
, FI_ATOMIC_FILE
);
246 mutex_lock(&fi
->inmem_lock
);
247 __revoke_inmem_pages(inode
, &fi
->inmem_pages
, true, false);
248 mutex_unlock(&fi
->inmem_lock
);
251 static int __commit_inmem_pages(struct inode
*inode
,
252 struct list_head
*revoke_list
)
254 struct f2fs_sb_info
*sbi
= F2FS_I_SB(inode
);
255 struct f2fs_inode_info
*fi
= F2FS_I(inode
);
256 struct inmem_pages
*cur
, *tmp
;
257 struct f2fs_io_info fio
= {
260 .rw
= WRITE_SYNC
| REQ_PRIO
,
261 .encrypted_page
= NULL
,
263 bool submit_bio
= false;
266 list_for_each_entry_safe(cur
, tmp
, &fi
->inmem_pages
, list
) {
267 struct page
*page
= cur
->page
;
270 if (page
->mapping
== inode
->i_mapping
) {
271 trace_f2fs_commit_inmem_page(page
, INMEM
);
273 set_page_dirty(page
);
274 f2fs_wait_on_page_writeback(page
, DATA
, true);
275 if (clear_page_dirty_for_io(page
))
276 inode_dec_dirty_pages(inode
);
279 err
= do_write_data_page(&fio
);
285 /* record old blkaddr for revoking */
286 cur
->old_addr
= fio
.old_blkaddr
;
288 clear_cold_data(page
);
292 list_move_tail(&cur
->list
, revoke_list
);
296 f2fs_submit_merged_bio_cond(sbi
, inode
, NULL
, 0, DATA
, WRITE
);
299 __revoke_inmem_pages(inode
, revoke_list
, false, false);
304 int commit_inmem_pages(struct inode
*inode
)
306 struct f2fs_sb_info
*sbi
= F2FS_I_SB(inode
);
307 struct f2fs_inode_info
*fi
= F2FS_I(inode
);
308 struct list_head revoke_list
;
311 INIT_LIST_HEAD(&revoke_list
);
312 f2fs_balance_fs(sbi
, true);
315 mutex_lock(&fi
->inmem_lock
);
316 err
= __commit_inmem_pages(inode
, &revoke_list
);
320 * try to revoke all committed pages, but still we could fail
321 * due to no memory or other reason, if that happened, EAGAIN
322 * will be returned, which means in such case, transaction is
323 * already not integrity, caller should use journal to do the
324 * recovery or rewrite & commit last transaction. For other
325 * error number, revoking was done by filesystem itself.
327 ret
= __revoke_inmem_pages(inode
, &revoke_list
, false, true);
331 /* drop all uncommitted pages */
332 __revoke_inmem_pages(inode
, &fi
->inmem_pages
, true, false);
334 mutex_unlock(&fi
->inmem_lock
);
341 * This function balances dirty node and dentry pages.
342 * In addition, it controls garbage collection.
344 void f2fs_balance_fs(struct f2fs_sb_info
*sbi
, bool need
)
349 /* balance_fs_bg is able to be pending */
350 if (excess_cached_nats(sbi
))
351 f2fs_balance_fs_bg(sbi
);
354 * We should do GC or end up with checkpoint, if there are so many dirty
355 * dir/node pages without enough free segments.
357 if (has_not_enough_free_secs(sbi
, 0)) {
358 mutex_lock(&sbi
->gc_mutex
);
363 void f2fs_balance_fs_bg(struct f2fs_sb_info
*sbi
)
365 /* try to shrink extent cache when there is no enough memory */
366 if (!available_free_memory(sbi
, EXTENT_CACHE
))
367 f2fs_shrink_extent_tree(sbi
, EXTENT_CACHE_SHRINK_NUMBER
);
369 /* check the # of cached NAT entries */
370 if (!available_free_memory(sbi
, NAT_ENTRIES
))
371 try_to_free_nats(sbi
, NAT_ENTRY_PER_BLOCK
);
373 if (!available_free_memory(sbi
, FREE_NIDS
))
374 try_to_free_nids(sbi
, MAX_FREE_NIDS
);
376 build_free_nids(sbi
);
378 /* checkpoint is the only way to shrink partial cached entries */
379 if (!available_free_memory(sbi
, NAT_ENTRIES
) ||
380 !available_free_memory(sbi
, INO_ENTRIES
) ||
381 excess_prefree_segs(sbi
) ||
382 excess_dirty_nats(sbi
) ||
383 (is_idle(sbi
) && f2fs_time_over(sbi
, CP_TIME
))) {
384 if (test_opt(sbi
, DATA_FLUSH
))
385 sync_dirty_inodes(sbi
, FILE_INODE
);
386 f2fs_sync_fs(sbi
->sb
, true);
387 stat_inc_bg_cp_count(sbi
->stat_info
);
391 static int issue_flush_thread(void *data
)
393 struct f2fs_sb_info
*sbi
= data
;
394 struct flush_cmd_control
*fcc
= SM_I(sbi
)->cmd_control_info
;
395 wait_queue_head_t
*q
= &fcc
->flush_wait_queue
;
397 if (kthread_should_stop())
400 if (!llist_empty(&fcc
->issue_list
)) {
402 struct flush_cmd
*cmd
, *next
;
405 bio
= f2fs_bio_alloc(0);
407 fcc
->dispatch_list
= llist_del_all(&fcc
->issue_list
);
408 fcc
->dispatch_list
= llist_reverse_order(fcc
->dispatch_list
);
410 bio
->bi_bdev
= sbi
->sb
->s_bdev
;
411 ret
= submit_bio_wait(WRITE_FLUSH
, bio
);
413 llist_for_each_entry_safe(cmd
, next
,
414 fcc
->dispatch_list
, llnode
) {
416 complete(&cmd
->wait
);
419 fcc
->dispatch_list
= NULL
;
422 wait_event_interruptible(*q
,
423 kthread_should_stop() || !llist_empty(&fcc
->issue_list
));
427 int f2fs_issue_flush(struct f2fs_sb_info
*sbi
)
429 struct flush_cmd_control
*fcc
= SM_I(sbi
)->cmd_control_info
;
430 struct flush_cmd cmd
;
432 trace_f2fs_issue_flush(sbi
->sb
, test_opt(sbi
, NOBARRIER
),
433 test_opt(sbi
, FLUSH_MERGE
));
435 if (test_opt(sbi
, NOBARRIER
))
438 if (!test_opt(sbi
, FLUSH_MERGE
) || !atomic_read(&fcc
->submit_flush
)) {
439 struct bio
*bio
= f2fs_bio_alloc(0);
442 atomic_inc(&fcc
->submit_flush
);
443 bio
->bi_bdev
= sbi
->sb
->s_bdev
;
444 ret
= submit_bio_wait(WRITE_FLUSH
, bio
);
445 atomic_dec(&fcc
->submit_flush
);
450 init_completion(&cmd
.wait
);
452 atomic_inc(&fcc
->submit_flush
);
453 llist_add(&cmd
.llnode
, &fcc
->issue_list
);
455 if (!fcc
->dispatch_list
)
456 wake_up(&fcc
->flush_wait_queue
);
458 wait_for_completion(&cmd
.wait
);
459 atomic_dec(&fcc
->submit_flush
);
464 int create_flush_cmd_control(struct f2fs_sb_info
*sbi
)
466 dev_t dev
= sbi
->sb
->s_bdev
->bd_dev
;
467 struct flush_cmd_control
*fcc
;
470 fcc
= kzalloc(sizeof(struct flush_cmd_control
), GFP_KERNEL
);
473 atomic_set(&fcc
->submit_flush
, 0);
474 init_waitqueue_head(&fcc
->flush_wait_queue
);
475 init_llist_head(&fcc
->issue_list
);
476 SM_I(sbi
)->cmd_control_info
= fcc
;
477 fcc
->f2fs_issue_flush
= kthread_run(issue_flush_thread
, sbi
,
478 "f2fs_flush-%u:%u", MAJOR(dev
), MINOR(dev
));
479 if (IS_ERR(fcc
->f2fs_issue_flush
)) {
480 err
= PTR_ERR(fcc
->f2fs_issue_flush
);
482 SM_I(sbi
)->cmd_control_info
= NULL
;
489 void destroy_flush_cmd_control(struct f2fs_sb_info
*sbi
)
491 struct flush_cmd_control
*fcc
= SM_I(sbi
)->cmd_control_info
;
493 if (fcc
&& fcc
->f2fs_issue_flush
)
494 kthread_stop(fcc
->f2fs_issue_flush
);
496 SM_I(sbi
)->cmd_control_info
= NULL
;
499 static void __locate_dirty_segment(struct f2fs_sb_info
*sbi
, unsigned int segno
,
500 enum dirty_type dirty_type
)
502 struct dirty_seglist_info
*dirty_i
= DIRTY_I(sbi
);
504 /* need not be added */
505 if (IS_CURSEG(sbi
, segno
))
508 if (!test_and_set_bit(segno
, dirty_i
->dirty_segmap
[dirty_type
]))
509 dirty_i
->nr_dirty
[dirty_type
]++;
511 if (dirty_type
== DIRTY
) {
512 struct seg_entry
*sentry
= get_seg_entry(sbi
, segno
);
513 enum dirty_type t
= sentry
->type
;
515 if (unlikely(t
>= DIRTY
)) {
519 if (!test_and_set_bit(segno
, dirty_i
->dirty_segmap
[t
]))
520 dirty_i
->nr_dirty
[t
]++;
524 static void __remove_dirty_segment(struct f2fs_sb_info
*sbi
, unsigned int segno
,
525 enum dirty_type dirty_type
)
527 struct dirty_seglist_info
*dirty_i
= DIRTY_I(sbi
);
529 if (test_and_clear_bit(segno
, dirty_i
->dirty_segmap
[dirty_type
]))
530 dirty_i
->nr_dirty
[dirty_type
]--;
532 if (dirty_type
== DIRTY
) {
533 struct seg_entry
*sentry
= get_seg_entry(sbi
, segno
);
534 enum dirty_type t
= sentry
->type
;
536 if (test_and_clear_bit(segno
, dirty_i
->dirty_segmap
[t
]))
537 dirty_i
->nr_dirty
[t
]--;
539 if (get_valid_blocks(sbi
, segno
, sbi
->segs_per_sec
) == 0)
540 clear_bit(GET_SECNO(sbi
, segno
),
541 dirty_i
->victim_secmap
);
546 * Should not occur error such as -ENOMEM.
547 * Adding dirty entry into seglist is not critical operation.
548 * If a given segment is one of current working segments, it won't be added.
550 static void locate_dirty_segment(struct f2fs_sb_info
*sbi
, unsigned int segno
)
552 struct dirty_seglist_info
*dirty_i
= DIRTY_I(sbi
);
553 unsigned short valid_blocks
;
555 if (segno
== NULL_SEGNO
|| IS_CURSEG(sbi
, segno
))
558 mutex_lock(&dirty_i
->seglist_lock
);
560 valid_blocks
= get_valid_blocks(sbi
, segno
, 0);
562 if (valid_blocks
== 0) {
563 __locate_dirty_segment(sbi
, segno
, PRE
);
564 __remove_dirty_segment(sbi
, segno
, DIRTY
);
565 } else if (valid_blocks
< sbi
->blocks_per_seg
) {
566 __locate_dirty_segment(sbi
, segno
, DIRTY
);
568 /* Recovery routine with SSR needs this */
569 __remove_dirty_segment(sbi
, segno
, DIRTY
);
572 mutex_unlock(&dirty_i
->seglist_lock
);
575 static int f2fs_issue_discard(struct f2fs_sb_info
*sbi
,
576 block_t blkstart
, block_t blklen
)
578 sector_t start
= SECTOR_FROM_BLOCK(blkstart
);
579 sector_t len
= SECTOR_FROM_BLOCK(blklen
);
580 struct seg_entry
*se
;
584 for (i
= blkstart
; i
< blkstart
+ blklen
; i
++) {
585 se
= get_seg_entry(sbi
, GET_SEGNO(sbi
, i
));
586 offset
= GET_BLKOFF_FROM_SEG0(sbi
, i
);
588 if (!f2fs_test_and_set_bit(offset
, se
->discard_map
))
591 trace_f2fs_issue_discard(sbi
->sb
, blkstart
, blklen
);
592 return blkdev_issue_discard(sbi
->sb
->s_bdev
, start
, len
, GFP_NOFS
, 0);
595 bool discard_next_dnode(struct f2fs_sb_info
*sbi
, block_t blkaddr
)
597 int err
= -EOPNOTSUPP
;
599 if (test_opt(sbi
, DISCARD
)) {
600 struct seg_entry
*se
= get_seg_entry(sbi
,
601 GET_SEGNO(sbi
, blkaddr
));
602 unsigned int offset
= GET_BLKOFF_FROM_SEG0(sbi
, blkaddr
);
604 if (f2fs_test_bit(offset
, se
->discard_map
))
607 err
= f2fs_issue_discard(sbi
, blkaddr
, 1);
611 update_meta_page(sbi
, NULL
, blkaddr
);
617 static void __add_discard_entry(struct f2fs_sb_info
*sbi
,
618 struct cp_control
*cpc
, struct seg_entry
*se
,
619 unsigned int start
, unsigned int end
)
621 struct list_head
*head
= &SM_I(sbi
)->discard_list
;
622 struct discard_entry
*new, *last
;
624 if (!list_empty(head
)) {
625 last
= list_last_entry(head
, struct discard_entry
, list
);
626 if (START_BLOCK(sbi
, cpc
->trim_start
) + start
==
627 last
->blkaddr
+ last
->len
) {
628 last
->len
+= end
- start
;
633 new = f2fs_kmem_cache_alloc(discard_entry_slab
, GFP_NOFS
);
634 INIT_LIST_HEAD(&new->list
);
635 new->blkaddr
= START_BLOCK(sbi
, cpc
->trim_start
) + start
;
636 new->len
= end
- start
;
637 list_add_tail(&new->list
, head
);
639 SM_I(sbi
)->nr_discards
+= end
- start
;
642 static void add_discard_addrs(struct f2fs_sb_info
*sbi
, struct cp_control
*cpc
)
644 int entries
= SIT_VBLOCK_MAP_SIZE
/ sizeof(unsigned long);
645 int max_blocks
= sbi
->blocks_per_seg
;
646 struct seg_entry
*se
= get_seg_entry(sbi
, cpc
->trim_start
);
647 unsigned long *cur_map
= (unsigned long *)se
->cur_valid_map
;
648 unsigned long *ckpt_map
= (unsigned long *)se
->ckpt_valid_map
;
649 unsigned long *discard_map
= (unsigned long *)se
->discard_map
;
650 unsigned long *dmap
= SIT_I(sbi
)->tmp_map
;
651 unsigned int start
= 0, end
= -1;
652 bool force
= (cpc
->reason
== CP_DISCARD
);
655 if (se
->valid_blocks
== max_blocks
)
659 if (!test_opt(sbi
, DISCARD
) || !se
->valid_blocks
||
660 SM_I(sbi
)->nr_discards
>= SM_I(sbi
)->max_discards
)
664 /* SIT_VBLOCK_MAP_SIZE should be multiple of sizeof(unsigned long) */
665 for (i
= 0; i
< entries
; i
++)
666 dmap
[i
] = force
? ~ckpt_map
[i
] & ~discard_map
[i
] :
667 (cur_map
[i
] ^ ckpt_map
[i
]) & ckpt_map
[i
];
669 while (force
|| SM_I(sbi
)->nr_discards
<= SM_I(sbi
)->max_discards
) {
670 start
= __find_rev_next_bit(dmap
, max_blocks
, end
+ 1);
671 if (start
>= max_blocks
)
674 end
= __find_rev_next_zero_bit(dmap
, max_blocks
, start
+ 1);
675 __add_discard_entry(sbi
, cpc
, se
, start
, end
);
679 void release_discard_addrs(struct f2fs_sb_info
*sbi
)
681 struct list_head
*head
= &(SM_I(sbi
)->discard_list
);
682 struct discard_entry
*entry
, *this;
685 list_for_each_entry_safe(entry
, this, head
, list
) {
686 list_del(&entry
->list
);
687 kmem_cache_free(discard_entry_slab
, entry
);
692 * Should call clear_prefree_segments after checkpoint is done.
694 static void set_prefree_as_free_segments(struct f2fs_sb_info
*sbi
)
696 struct dirty_seglist_info
*dirty_i
= DIRTY_I(sbi
);
699 mutex_lock(&dirty_i
->seglist_lock
);
700 for_each_set_bit(segno
, dirty_i
->dirty_segmap
[PRE
], MAIN_SEGS(sbi
))
701 __set_test_and_free(sbi
, segno
);
702 mutex_unlock(&dirty_i
->seglist_lock
);
705 void clear_prefree_segments(struct f2fs_sb_info
*sbi
, struct cp_control
*cpc
)
707 struct list_head
*head
= &(SM_I(sbi
)->discard_list
);
708 struct discard_entry
*entry
, *this;
709 struct dirty_seglist_info
*dirty_i
= DIRTY_I(sbi
);
710 unsigned long *prefree_map
= dirty_i
->dirty_segmap
[PRE
];
711 unsigned int start
= 0, end
= -1;
712 unsigned int secno
, start_segno
;
714 mutex_lock(&dirty_i
->seglist_lock
);
718 start
= find_next_bit(prefree_map
, MAIN_SEGS(sbi
), end
+ 1);
719 if (start
>= MAIN_SEGS(sbi
))
721 end
= find_next_zero_bit(prefree_map
, MAIN_SEGS(sbi
),
724 for (i
= start
; i
< end
; i
++)
725 clear_bit(i
, prefree_map
);
727 dirty_i
->nr_dirty
[PRE
] -= end
- start
;
729 if (!test_opt(sbi
, DISCARD
))
732 if (!test_opt(sbi
, LFS
) || sbi
->segs_per_sec
== 1) {
733 f2fs_issue_discard(sbi
, START_BLOCK(sbi
, start
),
734 (end
- start
) << sbi
->log_blocks_per_seg
);
738 secno
= GET_SECNO(sbi
, start
);
739 start_segno
= secno
* sbi
->segs_per_sec
;
740 if (!IS_CURSEC(sbi
, secno
) &&
741 !get_valid_blocks(sbi
, start
, sbi
->segs_per_sec
))
742 f2fs_issue_discard(sbi
, START_BLOCK(sbi
, start_segno
),
743 sbi
->segs_per_sec
<< sbi
->log_blocks_per_seg
);
745 start
= start_segno
+ sbi
->segs_per_sec
;
749 mutex_unlock(&dirty_i
->seglist_lock
);
751 /* send small discards */
752 list_for_each_entry_safe(entry
, this, head
, list
) {
753 if (cpc
->reason
== CP_DISCARD
&& entry
->len
< cpc
->trim_minlen
)
755 f2fs_issue_discard(sbi
, entry
->blkaddr
, entry
->len
);
756 cpc
->trimmed
+= entry
->len
;
758 list_del(&entry
->list
);
759 SM_I(sbi
)->nr_discards
-= entry
->len
;
760 kmem_cache_free(discard_entry_slab
, entry
);
764 static bool __mark_sit_entry_dirty(struct f2fs_sb_info
*sbi
, unsigned int segno
)
766 struct sit_info
*sit_i
= SIT_I(sbi
);
768 if (!__test_and_set_bit(segno
, sit_i
->dirty_sentries_bitmap
)) {
769 sit_i
->dirty_sentries
++;
776 static void __set_sit_entry_type(struct f2fs_sb_info
*sbi
, int type
,
777 unsigned int segno
, int modified
)
779 struct seg_entry
*se
= get_seg_entry(sbi
, segno
);
782 __mark_sit_entry_dirty(sbi
, segno
);
785 static void update_sit_entry(struct f2fs_sb_info
*sbi
, block_t blkaddr
, int del
)
787 struct seg_entry
*se
;
788 unsigned int segno
, offset
;
789 long int new_vblocks
;
791 segno
= GET_SEGNO(sbi
, blkaddr
);
793 se
= get_seg_entry(sbi
, segno
);
794 new_vblocks
= se
->valid_blocks
+ del
;
795 offset
= GET_BLKOFF_FROM_SEG0(sbi
, blkaddr
);
797 f2fs_bug_on(sbi
, (new_vblocks
>> (sizeof(unsigned short) << 3) ||
798 (new_vblocks
> sbi
->blocks_per_seg
)));
800 se
->valid_blocks
= new_vblocks
;
801 se
->mtime
= get_mtime(sbi
);
802 SIT_I(sbi
)->max_mtime
= se
->mtime
;
804 /* Update valid block bitmap */
806 if (f2fs_test_and_set_bit(offset
, se
->cur_valid_map
))
808 if (!f2fs_test_and_set_bit(offset
, se
->discard_map
))
811 if (!f2fs_test_and_clear_bit(offset
, se
->cur_valid_map
))
813 if (f2fs_test_and_clear_bit(offset
, se
->discard_map
))
816 if (!f2fs_test_bit(offset
, se
->ckpt_valid_map
))
817 se
->ckpt_valid_blocks
+= del
;
819 __mark_sit_entry_dirty(sbi
, segno
);
821 /* update total number of valid blocks to be written in ckpt area */
822 SIT_I(sbi
)->written_valid_blocks
+= del
;
824 if (sbi
->segs_per_sec
> 1)
825 get_sec_entry(sbi
, segno
)->valid_blocks
+= del
;
828 void refresh_sit_entry(struct f2fs_sb_info
*sbi
, block_t old
, block_t
new)
830 update_sit_entry(sbi
, new, 1);
831 if (GET_SEGNO(sbi
, old
) != NULL_SEGNO
)
832 update_sit_entry(sbi
, old
, -1);
834 locate_dirty_segment(sbi
, GET_SEGNO(sbi
, old
));
835 locate_dirty_segment(sbi
, GET_SEGNO(sbi
, new));
838 void invalidate_blocks(struct f2fs_sb_info
*sbi
, block_t addr
)
840 unsigned int segno
= GET_SEGNO(sbi
, addr
);
841 struct sit_info
*sit_i
= SIT_I(sbi
);
843 f2fs_bug_on(sbi
, addr
== NULL_ADDR
);
844 if (addr
== NEW_ADDR
)
847 /* add it into sit main buffer */
848 mutex_lock(&sit_i
->sentry_lock
);
850 update_sit_entry(sbi
, addr
, -1);
852 /* add it into dirty seglist */
853 locate_dirty_segment(sbi
, segno
);
855 mutex_unlock(&sit_i
->sentry_lock
);
858 bool is_checkpointed_data(struct f2fs_sb_info
*sbi
, block_t blkaddr
)
860 struct sit_info
*sit_i
= SIT_I(sbi
);
861 unsigned int segno
, offset
;
862 struct seg_entry
*se
;
865 if (blkaddr
== NEW_ADDR
|| blkaddr
== NULL_ADDR
)
868 mutex_lock(&sit_i
->sentry_lock
);
870 segno
= GET_SEGNO(sbi
, blkaddr
);
871 se
= get_seg_entry(sbi
, segno
);
872 offset
= GET_BLKOFF_FROM_SEG0(sbi
, blkaddr
);
874 if (f2fs_test_bit(offset
, se
->ckpt_valid_map
))
877 mutex_unlock(&sit_i
->sentry_lock
);
883 * This function should be resided under the curseg_mutex lock
885 static void __add_sum_entry(struct f2fs_sb_info
*sbi
, int type
,
886 struct f2fs_summary
*sum
)
888 struct curseg_info
*curseg
= CURSEG_I(sbi
, type
);
889 void *addr
= curseg
->sum_blk
;
890 addr
+= curseg
->next_blkoff
* sizeof(struct f2fs_summary
);
891 memcpy(addr
, sum
, sizeof(struct f2fs_summary
));
895 * Calculate the number of current summary pages for writing
897 int npages_for_summary_flush(struct f2fs_sb_info
*sbi
, bool for_ra
)
899 int valid_sum_count
= 0;
902 for (i
= CURSEG_HOT_DATA
; i
<= CURSEG_COLD_DATA
; i
++) {
903 if (sbi
->ckpt
->alloc_type
[i
] == SSR
)
904 valid_sum_count
+= sbi
->blocks_per_seg
;
907 valid_sum_count
+= le16_to_cpu(
908 F2FS_CKPT(sbi
)->cur_data_blkoff
[i
]);
910 valid_sum_count
+= curseg_blkoff(sbi
, i
);
914 sum_in_page
= (PAGE_SIZE
- 2 * SUM_JOURNAL_SIZE
-
915 SUM_FOOTER_SIZE
) / SUMMARY_SIZE
;
916 if (valid_sum_count
<= sum_in_page
)
918 else if ((valid_sum_count
- sum_in_page
) <=
919 (PAGE_SIZE
- SUM_FOOTER_SIZE
) / SUMMARY_SIZE
)
925 * Caller should put this summary page
927 struct page
*get_sum_page(struct f2fs_sb_info
*sbi
, unsigned int segno
)
929 return get_meta_page(sbi
, GET_SUM_BLOCK(sbi
, segno
));
932 void update_meta_page(struct f2fs_sb_info
*sbi
, void *src
, block_t blk_addr
)
934 struct page
*page
= grab_meta_page(sbi
, blk_addr
);
935 void *dst
= page_address(page
);
938 memcpy(dst
, src
, PAGE_SIZE
);
940 memset(dst
, 0, PAGE_SIZE
);
941 set_page_dirty(page
);
942 f2fs_put_page(page
, 1);
945 static void write_sum_page(struct f2fs_sb_info
*sbi
,
946 struct f2fs_summary_block
*sum_blk
, block_t blk_addr
)
948 update_meta_page(sbi
, (void *)sum_blk
, blk_addr
);
951 static void write_current_sum_page(struct f2fs_sb_info
*sbi
,
952 int type
, block_t blk_addr
)
954 struct curseg_info
*curseg
= CURSEG_I(sbi
, type
);
955 struct page
*page
= grab_meta_page(sbi
, blk_addr
);
956 struct f2fs_summary_block
*src
= curseg
->sum_blk
;
957 struct f2fs_summary_block
*dst
;
959 dst
= (struct f2fs_summary_block
*)page_address(page
);
961 mutex_lock(&curseg
->curseg_mutex
);
963 down_read(&curseg
->journal_rwsem
);
964 memcpy(&dst
->journal
, curseg
->journal
, SUM_JOURNAL_SIZE
);
965 up_read(&curseg
->journal_rwsem
);
967 memcpy(dst
->entries
, src
->entries
, SUM_ENTRY_SIZE
);
968 memcpy(&dst
->footer
, &src
->footer
, SUM_FOOTER_SIZE
);
970 mutex_unlock(&curseg
->curseg_mutex
);
972 set_page_dirty(page
);
973 f2fs_put_page(page
, 1);
976 static int is_next_segment_free(struct f2fs_sb_info
*sbi
, int type
)
978 struct curseg_info
*curseg
= CURSEG_I(sbi
, type
);
979 unsigned int segno
= curseg
->segno
+ 1;
980 struct free_segmap_info
*free_i
= FREE_I(sbi
);
982 if (segno
< MAIN_SEGS(sbi
) && segno
% sbi
->segs_per_sec
)
983 return !test_bit(segno
, free_i
->free_segmap
);
988 * Find a new segment from the free segments bitmap to right order
989 * This function should be returned with success, otherwise BUG
991 static void get_new_segment(struct f2fs_sb_info
*sbi
,
992 unsigned int *newseg
, bool new_sec
, int dir
)
994 struct free_segmap_info
*free_i
= FREE_I(sbi
);
995 unsigned int segno
, secno
, zoneno
;
996 unsigned int total_zones
= MAIN_SECS(sbi
) / sbi
->secs_per_zone
;
997 unsigned int hint
= *newseg
/ sbi
->segs_per_sec
;
998 unsigned int old_zoneno
= GET_ZONENO_FROM_SEGNO(sbi
, *newseg
);
999 unsigned int left_start
= hint
;
1004 spin_lock(&free_i
->segmap_lock
);
1006 if (!new_sec
&& ((*newseg
+ 1) % sbi
->segs_per_sec
)) {
1007 segno
= find_next_zero_bit(free_i
->free_segmap
,
1008 (hint
+ 1) * sbi
->segs_per_sec
, *newseg
+ 1);
1009 if (segno
< (hint
+ 1) * sbi
->segs_per_sec
)
1013 secno
= find_next_zero_bit(free_i
->free_secmap
, MAIN_SECS(sbi
), hint
);
1014 if (secno
>= MAIN_SECS(sbi
)) {
1015 if (dir
== ALLOC_RIGHT
) {
1016 secno
= find_next_zero_bit(free_i
->free_secmap
,
1018 f2fs_bug_on(sbi
, secno
>= MAIN_SECS(sbi
));
1021 left_start
= hint
- 1;
1027 while (test_bit(left_start
, free_i
->free_secmap
)) {
1028 if (left_start
> 0) {
1032 left_start
= find_next_zero_bit(free_i
->free_secmap
,
1034 f2fs_bug_on(sbi
, left_start
>= MAIN_SECS(sbi
));
1040 segno
= secno
* sbi
->segs_per_sec
;
1041 zoneno
= secno
/ sbi
->secs_per_zone
;
1043 /* give up on finding another zone */
1046 if (sbi
->secs_per_zone
== 1)
1048 if (zoneno
== old_zoneno
)
1050 if (dir
== ALLOC_LEFT
) {
1051 if (!go_left
&& zoneno
+ 1 >= total_zones
)
1053 if (go_left
&& zoneno
== 0)
1056 for (i
= 0; i
< NR_CURSEG_TYPE
; i
++)
1057 if (CURSEG_I(sbi
, i
)->zone
== zoneno
)
1060 if (i
< NR_CURSEG_TYPE
) {
1061 /* zone is in user, try another */
1063 hint
= zoneno
* sbi
->secs_per_zone
- 1;
1064 else if (zoneno
+ 1 >= total_zones
)
1067 hint
= (zoneno
+ 1) * sbi
->secs_per_zone
;
1069 goto find_other_zone
;
1072 /* set it as dirty segment in free segmap */
1073 f2fs_bug_on(sbi
, test_bit(segno
, free_i
->free_segmap
));
1074 __set_inuse(sbi
, segno
);
1076 spin_unlock(&free_i
->segmap_lock
);
1079 static void reset_curseg(struct f2fs_sb_info
*sbi
, int type
, int modified
)
1081 struct curseg_info
*curseg
= CURSEG_I(sbi
, type
);
1082 struct summary_footer
*sum_footer
;
1084 curseg
->segno
= curseg
->next_segno
;
1085 curseg
->zone
= GET_ZONENO_FROM_SEGNO(sbi
, curseg
->segno
);
1086 curseg
->next_blkoff
= 0;
1087 curseg
->next_segno
= NULL_SEGNO
;
1089 sum_footer
= &(curseg
->sum_blk
->footer
);
1090 memset(sum_footer
, 0, sizeof(struct summary_footer
));
1091 if (IS_DATASEG(type
))
1092 SET_SUM_TYPE(sum_footer
, SUM_TYPE_DATA
);
1093 if (IS_NODESEG(type
))
1094 SET_SUM_TYPE(sum_footer
, SUM_TYPE_NODE
);
1095 __set_sit_entry_type(sbi
, type
, curseg
->segno
, modified
);
1099 * Allocate a current working segment.
1100 * This function always allocates a free segment in LFS manner.
1102 static void new_curseg(struct f2fs_sb_info
*sbi
, int type
, bool new_sec
)
1104 struct curseg_info
*curseg
= CURSEG_I(sbi
, type
);
1105 unsigned int segno
= curseg
->segno
;
1106 int dir
= ALLOC_LEFT
;
1108 write_sum_page(sbi
, curseg
->sum_blk
,
1109 GET_SUM_BLOCK(sbi
, segno
));
1110 if (type
== CURSEG_WARM_DATA
|| type
== CURSEG_COLD_DATA
)
1113 if (test_opt(sbi
, NOHEAP
))
1116 get_new_segment(sbi
, &segno
, new_sec
, dir
);
1117 curseg
->next_segno
= segno
;
1118 reset_curseg(sbi
, type
, 1);
1119 curseg
->alloc_type
= LFS
;
1122 static void __next_free_blkoff(struct f2fs_sb_info
*sbi
,
1123 struct curseg_info
*seg
, block_t start
)
1125 struct seg_entry
*se
= get_seg_entry(sbi
, seg
->segno
);
1126 int entries
= SIT_VBLOCK_MAP_SIZE
/ sizeof(unsigned long);
1127 unsigned long *target_map
= SIT_I(sbi
)->tmp_map
;
1128 unsigned long *ckpt_map
= (unsigned long *)se
->ckpt_valid_map
;
1129 unsigned long *cur_map
= (unsigned long *)se
->cur_valid_map
;
1132 for (i
= 0; i
< entries
; i
++)
1133 target_map
[i
] = ckpt_map
[i
] | cur_map
[i
];
1135 pos
= __find_rev_next_zero_bit(target_map
, sbi
->blocks_per_seg
, start
);
1137 seg
->next_blkoff
= pos
;
1141 * If a segment is written by LFS manner, next block offset is just obtained
1142 * by increasing the current block offset. However, if a segment is written by
1143 * SSR manner, next block offset obtained by calling __next_free_blkoff
1145 static void __refresh_next_blkoff(struct f2fs_sb_info
*sbi
,
1146 struct curseg_info
*seg
)
1148 if (seg
->alloc_type
== SSR
)
1149 __next_free_blkoff(sbi
, seg
, seg
->next_blkoff
+ 1);
1155 * This function always allocates a used segment(from dirty seglist) by SSR
1156 * manner, so it should recover the existing segment information of valid blocks
1158 static void change_curseg(struct f2fs_sb_info
*sbi
, int type
, bool reuse
)
1160 struct dirty_seglist_info
*dirty_i
= DIRTY_I(sbi
);
1161 struct curseg_info
*curseg
= CURSEG_I(sbi
, type
);
1162 unsigned int new_segno
= curseg
->next_segno
;
1163 struct f2fs_summary_block
*sum_node
;
1164 struct page
*sum_page
;
1166 write_sum_page(sbi
, curseg
->sum_blk
,
1167 GET_SUM_BLOCK(sbi
, curseg
->segno
));
1168 __set_test_and_inuse(sbi
, new_segno
);
1170 mutex_lock(&dirty_i
->seglist_lock
);
1171 __remove_dirty_segment(sbi
, new_segno
, PRE
);
1172 __remove_dirty_segment(sbi
, new_segno
, DIRTY
);
1173 mutex_unlock(&dirty_i
->seglist_lock
);
1175 reset_curseg(sbi
, type
, 1);
1176 curseg
->alloc_type
= SSR
;
1177 __next_free_blkoff(sbi
, curseg
, 0);
1180 sum_page
= get_sum_page(sbi
, new_segno
);
1181 sum_node
= (struct f2fs_summary_block
*)page_address(sum_page
);
1182 memcpy(curseg
->sum_blk
, sum_node
, SUM_ENTRY_SIZE
);
1183 f2fs_put_page(sum_page
, 1);
1187 static int get_ssr_segment(struct f2fs_sb_info
*sbi
, int type
)
1189 struct curseg_info
*curseg
= CURSEG_I(sbi
, type
);
1190 const struct victim_selection
*v_ops
= DIRTY_I(sbi
)->v_ops
;
1192 if (IS_NODESEG(type
) || !has_not_enough_free_secs(sbi
, 0))
1193 return v_ops
->get_victim(sbi
,
1194 &(curseg
)->next_segno
, BG_GC
, type
, SSR
);
1196 /* For data segments, let's do SSR more intensively */
1197 for (; type
>= CURSEG_HOT_DATA
; type
--)
1198 if (v_ops
->get_victim(sbi
, &(curseg
)->next_segno
,
1205 * flush out current segment and replace it with new segment
1206 * This function should be returned with success, otherwise BUG
1208 static void allocate_segment_by_default(struct f2fs_sb_info
*sbi
,
1209 int type
, bool force
)
1211 struct curseg_info
*curseg
= CURSEG_I(sbi
, type
);
1214 new_curseg(sbi
, type
, true);
1215 else if (type
== CURSEG_WARM_NODE
)
1216 new_curseg(sbi
, type
, false);
1217 else if (curseg
->alloc_type
== LFS
&& is_next_segment_free(sbi
, type
))
1218 new_curseg(sbi
, type
, false);
1219 else if (need_SSR(sbi
) && get_ssr_segment(sbi
, type
))
1220 change_curseg(sbi
, type
, true);
1222 new_curseg(sbi
, type
, false);
1224 stat_inc_seg_type(sbi
, curseg
);
1227 static void __allocate_new_segments(struct f2fs_sb_info
*sbi
, int type
)
1229 struct curseg_info
*curseg
= CURSEG_I(sbi
, type
);
1230 unsigned int old_segno
;
1232 old_segno
= curseg
->segno
;
1233 SIT_I(sbi
)->s_ops
->allocate_segment(sbi
, type
, true);
1234 locate_dirty_segment(sbi
, old_segno
);
1237 void allocate_new_segments(struct f2fs_sb_info
*sbi
)
1241 if (test_opt(sbi
, LFS
))
1244 for (i
= CURSEG_HOT_DATA
; i
<= CURSEG_COLD_DATA
; i
++)
1245 __allocate_new_segments(sbi
, i
);
1248 static const struct segment_allocation default_salloc_ops
= {
1249 .allocate_segment
= allocate_segment_by_default
,
1252 int f2fs_trim_fs(struct f2fs_sb_info
*sbi
, struct fstrim_range
*range
)
1254 __u64 start
= F2FS_BYTES_TO_BLK(range
->start
);
1255 __u64 end
= start
+ F2FS_BYTES_TO_BLK(range
->len
) - 1;
1256 unsigned int start_segno
, end_segno
;
1257 struct cp_control cpc
;
1260 if (start
>= MAX_BLKADDR(sbi
) || range
->len
< sbi
->blocksize
)
1264 if (end
<= MAIN_BLKADDR(sbi
))
1267 /* start/end segment number in main_area */
1268 start_segno
= (start
<= MAIN_BLKADDR(sbi
)) ? 0 : GET_SEGNO(sbi
, start
);
1269 end_segno
= (end
>= MAX_BLKADDR(sbi
)) ? MAIN_SEGS(sbi
) - 1 :
1270 GET_SEGNO(sbi
, end
);
1271 cpc
.reason
= CP_DISCARD
;
1272 cpc
.trim_minlen
= max_t(__u64
, 1, F2FS_BYTES_TO_BLK(range
->minlen
));
1274 /* do checkpoint to issue discard commands safely */
1275 for (; start_segno
<= end_segno
; start_segno
= cpc
.trim_end
+ 1) {
1276 cpc
.trim_start
= start_segno
;
1278 if (sbi
->discard_blks
== 0)
1280 else if (sbi
->discard_blks
< BATCHED_TRIM_BLOCKS(sbi
))
1281 cpc
.trim_end
= end_segno
;
1283 cpc
.trim_end
= min_t(unsigned int,
1284 rounddown(start_segno
+
1285 BATCHED_TRIM_SEGMENTS(sbi
),
1286 sbi
->segs_per_sec
) - 1, end_segno
);
1288 mutex_lock(&sbi
->gc_mutex
);
1289 err
= write_checkpoint(sbi
, &cpc
);
1290 mutex_unlock(&sbi
->gc_mutex
);
1293 range
->len
= F2FS_BLK_TO_BYTES(cpc
.trimmed
);
1297 static bool __has_curseg_space(struct f2fs_sb_info
*sbi
, int type
)
1299 struct curseg_info
*curseg
= CURSEG_I(sbi
, type
);
1300 if (curseg
->next_blkoff
< sbi
->blocks_per_seg
)
1305 static int __get_segment_type_2(struct page
*page
, enum page_type p_type
)
1308 return CURSEG_HOT_DATA
;
1310 return CURSEG_HOT_NODE
;
1313 static int __get_segment_type_4(struct page
*page
, enum page_type p_type
)
1315 if (p_type
== DATA
) {
1316 struct inode
*inode
= page
->mapping
->host
;
1318 if (S_ISDIR(inode
->i_mode
))
1319 return CURSEG_HOT_DATA
;
1321 return CURSEG_COLD_DATA
;
1323 if (IS_DNODE(page
) && is_cold_node(page
))
1324 return CURSEG_WARM_NODE
;
1326 return CURSEG_COLD_NODE
;
1330 static int __get_segment_type_6(struct page
*page
, enum page_type p_type
)
1332 if (p_type
== DATA
) {
1333 struct inode
*inode
= page
->mapping
->host
;
1335 if (S_ISDIR(inode
->i_mode
))
1336 return CURSEG_HOT_DATA
;
1337 else if (is_cold_data(page
) || file_is_cold(inode
))
1338 return CURSEG_COLD_DATA
;
1340 return CURSEG_WARM_DATA
;
1343 return is_cold_node(page
) ? CURSEG_WARM_NODE
:
1346 return CURSEG_COLD_NODE
;
1350 static int __get_segment_type(struct page
*page
, enum page_type p_type
)
1352 switch (F2FS_P_SB(page
)->active_logs
) {
1354 return __get_segment_type_2(page
, p_type
);
1356 return __get_segment_type_4(page
, p_type
);
1358 /* NR_CURSEG_TYPE(6) logs by default */
1359 f2fs_bug_on(F2FS_P_SB(page
),
1360 F2FS_P_SB(page
)->active_logs
!= NR_CURSEG_TYPE
);
1361 return __get_segment_type_6(page
, p_type
);
1364 void allocate_data_block(struct f2fs_sb_info
*sbi
, struct page
*page
,
1365 block_t old_blkaddr
, block_t
*new_blkaddr
,
1366 struct f2fs_summary
*sum
, int type
)
1368 struct sit_info
*sit_i
= SIT_I(sbi
);
1369 struct curseg_info
*curseg
;
1370 bool direct_io
= (type
== CURSEG_DIRECT_IO
);
1372 type
= direct_io
? CURSEG_WARM_DATA
: type
;
1374 curseg
= CURSEG_I(sbi
, type
);
1376 mutex_lock(&curseg
->curseg_mutex
);
1377 mutex_lock(&sit_i
->sentry_lock
);
1379 /* direct_io'ed data is aligned to the segment for better performance */
1380 if (direct_io
&& curseg
->next_blkoff
&&
1381 !has_not_enough_free_secs(sbi
, 0))
1382 __allocate_new_segments(sbi
, type
);
1384 *new_blkaddr
= NEXT_FREE_BLKADDR(sbi
, curseg
);
1387 * __add_sum_entry should be resided under the curseg_mutex
1388 * because, this function updates a summary entry in the
1389 * current summary block.
1391 __add_sum_entry(sbi
, type
, sum
);
1393 __refresh_next_blkoff(sbi
, curseg
);
1395 stat_inc_block_count(sbi
, curseg
);
1397 if (!__has_curseg_space(sbi
, type
))
1398 sit_i
->s_ops
->allocate_segment(sbi
, type
, false);
1400 * SIT information should be updated before segment allocation,
1401 * since SSR needs latest valid block information.
1403 refresh_sit_entry(sbi
, old_blkaddr
, *new_blkaddr
);
1405 mutex_unlock(&sit_i
->sentry_lock
);
1407 if (page
&& IS_NODESEG(type
))
1408 fill_node_footer_blkaddr(page
, NEXT_FREE_BLKADDR(sbi
, curseg
));
1410 mutex_unlock(&curseg
->curseg_mutex
);
1413 static void do_write_page(struct f2fs_summary
*sum
, struct f2fs_io_info
*fio
)
1415 int type
= __get_segment_type(fio
->page
, fio
->type
);
1417 if (fio
->type
== NODE
|| fio
->type
== DATA
)
1418 mutex_lock(&fio
->sbi
->wio_mutex
[fio
->type
]);
1420 allocate_data_block(fio
->sbi
, fio
->page
, fio
->old_blkaddr
,
1421 &fio
->new_blkaddr
, sum
, type
);
1423 /* writeout dirty page into bdev */
1424 f2fs_submit_page_mbio(fio
);
1426 if (fio
->type
== NODE
|| fio
->type
== DATA
)
1427 mutex_unlock(&fio
->sbi
->wio_mutex
[fio
->type
]);
1430 void write_meta_page(struct f2fs_sb_info
*sbi
, struct page
*page
)
1432 struct f2fs_io_info fio
= {
1435 .rw
= WRITE_SYNC
| REQ_META
| REQ_PRIO
,
1436 .old_blkaddr
= page
->index
,
1437 .new_blkaddr
= page
->index
,
1439 .encrypted_page
= NULL
,
1442 if (unlikely(page
->index
>= MAIN_BLKADDR(sbi
)))
1443 fio
.rw
&= ~REQ_META
;
1445 set_page_writeback(page
);
1446 f2fs_submit_page_mbio(&fio
);
1449 void write_node_page(unsigned int nid
, struct f2fs_io_info
*fio
)
1451 struct f2fs_summary sum
;
1453 set_summary(&sum
, nid
, 0, 0);
1454 do_write_page(&sum
, fio
);
1457 void write_data_page(struct dnode_of_data
*dn
, struct f2fs_io_info
*fio
)
1459 struct f2fs_sb_info
*sbi
= fio
->sbi
;
1460 struct f2fs_summary sum
;
1461 struct node_info ni
;
1463 f2fs_bug_on(sbi
, dn
->data_blkaddr
== NULL_ADDR
);
1464 get_node_info(sbi
, dn
->nid
, &ni
);
1465 set_summary(&sum
, dn
->nid
, dn
->ofs_in_node
, ni
.version
);
1466 do_write_page(&sum
, fio
);
1467 f2fs_update_data_blkaddr(dn
, fio
->new_blkaddr
);
1470 void rewrite_data_page(struct f2fs_io_info
*fio
)
1472 fio
->new_blkaddr
= fio
->old_blkaddr
;
1473 stat_inc_inplace_blocks(fio
->sbi
);
1474 f2fs_submit_page_mbio(fio
);
1477 void __f2fs_replace_block(struct f2fs_sb_info
*sbi
, struct f2fs_summary
*sum
,
1478 block_t old_blkaddr
, block_t new_blkaddr
,
1479 bool recover_curseg
, bool recover_newaddr
)
1481 struct sit_info
*sit_i
= SIT_I(sbi
);
1482 struct curseg_info
*curseg
;
1483 unsigned int segno
, old_cursegno
;
1484 struct seg_entry
*se
;
1486 unsigned short old_blkoff
;
1488 segno
= GET_SEGNO(sbi
, new_blkaddr
);
1489 se
= get_seg_entry(sbi
, segno
);
1492 if (!recover_curseg
) {
1493 /* for recovery flow */
1494 if (se
->valid_blocks
== 0 && !IS_CURSEG(sbi
, segno
)) {
1495 if (old_blkaddr
== NULL_ADDR
)
1496 type
= CURSEG_COLD_DATA
;
1498 type
= CURSEG_WARM_DATA
;
1501 if (!IS_CURSEG(sbi
, segno
))
1502 type
= CURSEG_WARM_DATA
;
1505 curseg
= CURSEG_I(sbi
, type
);
1507 mutex_lock(&curseg
->curseg_mutex
);
1508 mutex_lock(&sit_i
->sentry_lock
);
1510 old_cursegno
= curseg
->segno
;
1511 old_blkoff
= curseg
->next_blkoff
;
1513 /* change the current segment */
1514 if (segno
!= curseg
->segno
) {
1515 curseg
->next_segno
= segno
;
1516 change_curseg(sbi
, type
, true);
1519 curseg
->next_blkoff
= GET_BLKOFF_FROM_SEG0(sbi
, new_blkaddr
);
1520 __add_sum_entry(sbi
, type
, sum
);
1522 if (!recover_curseg
|| recover_newaddr
)
1523 update_sit_entry(sbi
, new_blkaddr
, 1);
1524 if (GET_SEGNO(sbi
, old_blkaddr
) != NULL_SEGNO
)
1525 update_sit_entry(sbi
, old_blkaddr
, -1);
1527 locate_dirty_segment(sbi
, GET_SEGNO(sbi
, old_blkaddr
));
1528 locate_dirty_segment(sbi
, GET_SEGNO(sbi
, new_blkaddr
));
1530 locate_dirty_segment(sbi
, old_cursegno
);
1532 if (recover_curseg
) {
1533 if (old_cursegno
!= curseg
->segno
) {
1534 curseg
->next_segno
= old_cursegno
;
1535 change_curseg(sbi
, type
, true);
1537 curseg
->next_blkoff
= old_blkoff
;
1540 mutex_unlock(&sit_i
->sentry_lock
);
1541 mutex_unlock(&curseg
->curseg_mutex
);
1544 void f2fs_replace_block(struct f2fs_sb_info
*sbi
, struct dnode_of_data
*dn
,
1545 block_t old_addr
, block_t new_addr
,
1546 unsigned char version
, bool recover_curseg
,
1547 bool recover_newaddr
)
1549 struct f2fs_summary sum
;
1551 set_summary(&sum
, dn
->nid
, dn
->ofs_in_node
, version
);
1553 __f2fs_replace_block(sbi
, &sum
, old_addr
, new_addr
,
1554 recover_curseg
, recover_newaddr
);
1556 f2fs_update_data_blkaddr(dn
, new_addr
);
1559 void f2fs_wait_on_page_writeback(struct page
*page
,
1560 enum page_type type
, bool ordered
)
1562 if (PageWriteback(page
)) {
1563 struct f2fs_sb_info
*sbi
= F2FS_P_SB(page
);
1565 f2fs_submit_merged_bio_cond(sbi
, NULL
, page
, 0, type
, WRITE
);
1567 wait_on_page_writeback(page
);
1569 wait_for_stable_page(page
);
1573 void f2fs_wait_on_encrypted_page_writeback(struct f2fs_sb_info
*sbi
,
1578 if (blkaddr
== NEW_ADDR
)
1581 f2fs_bug_on(sbi
, blkaddr
== NULL_ADDR
);
1583 cpage
= find_lock_page(META_MAPPING(sbi
), blkaddr
);
1585 f2fs_wait_on_page_writeback(cpage
, DATA
, true);
1586 f2fs_put_page(cpage
, 1);
1590 static int read_compacted_summaries(struct f2fs_sb_info
*sbi
)
1592 struct f2fs_checkpoint
*ckpt
= F2FS_CKPT(sbi
);
1593 struct curseg_info
*seg_i
;
1594 unsigned char *kaddr
;
1599 start
= start_sum_block(sbi
);
1601 page
= get_meta_page(sbi
, start
++);
1602 kaddr
= (unsigned char *)page_address(page
);
1604 /* Step 1: restore nat cache */
1605 seg_i
= CURSEG_I(sbi
, CURSEG_HOT_DATA
);
1606 memcpy(seg_i
->journal
, kaddr
, SUM_JOURNAL_SIZE
);
1608 /* Step 2: restore sit cache */
1609 seg_i
= CURSEG_I(sbi
, CURSEG_COLD_DATA
);
1610 memcpy(seg_i
->journal
, kaddr
+ SUM_JOURNAL_SIZE
, SUM_JOURNAL_SIZE
);
1611 offset
= 2 * SUM_JOURNAL_SIZE
;
1613 /* Step 3: restore summary entries */
1614 for (i
= CURSEG_HOT_DATA
; i
<= CURSEG_COLD_DATA
; i
++) {
1615 unsigned short blk_off
;
1618 seg_i
= CURSEG_I(sbi
, i
);
1619 segno
= le32_to_cpu(ckpt
->cur_data_segno
[i
]);
1620 blk_off
= le16_to_cpu(ckpt
->cur_data_blkoff
[i
]);
1621 seg_i
->next_segno
= segno
;
1622 reset_curseg(sbi
, i
, 0);
1623 seg_i
->alloc_type
= ckpt
->alloc_type
[i
];
1624 seg_i
->next_blkoff
= blk_off
;
1626 if (seg_i
->alloc_type
== SSR
)
1627 blk_off
= sbi
->blocks_per_seg
;
1629 for (j
= 0; j
< blk_off
; j
++) {
1630 struct f2fs_summary
*s
;
1631 s
= (struct f2fs_summary
*)(kaddr
+ offset
);
1632 seg_i
->sum_blk
->entries
[j
] = *s
;
1633 offset
+= SUMMARY_SIZE
;
1634 if (offset
+ SUMMARY_SIZE
<= PAGE_SIZE
-
1638 f2fs_put_page(page
, 1);
1641 page
= get_meta_page(sbi
, start
++);
1642 kaddr
= (unsigned char *)page_address(page
);
1646 f2fs_put_page(page
, 1);
1650 static int read_normal_summaries(struct f2fs_sb_info
*sbi
, int type
)
1652 struct f2fs_checkpoint
*ckpt
= F2FS_CKPT(sbi
);
1653 struct f2fs_summary_block
*sum
;
1654 struct curseg_info
*curseg
;
1656 unsigned short blk_off
;
1657 unsigned int segno
= 0;
1658 block_t blk_addr
= 0;
1660 /* get segment number and block addr */
1661 if (IS_DATASEG(type
)) {
1662 segno
= le32_to_cpu(ckpt
->cur_data_segno
[type
]);
1663 blk_off
= le16_to_cpu(ckpt
->cur_data_blkoff
[type
-
1665 if (__exist_node_summaries(sbi
))
1666 blk_addr
= sum_blk_addr(sbi
, NR_CURSEG_TYPE
, type
);
1668 blk_addr
= sum_blk_addr(sbi
, NR_CURSEG_DATA_TYPE
, type
);
1670 segno
= le32_to_cpu(ckpt
->cur_node_segno
[type
-
1672 blk_off
= le16_to_cpu(ckpt
->cur_node_blkoff
[type
-
1674 if (__exist_node_summaries(sbi
))
1675 blk_addr
= sum_blk_addr(sbi
, NR_CURSEG_NODE_TYPE
,
1676 type
- CURSEG_HOT_NODE
);
1678 blk_addr
= GET_SUM_BLOCK(sbi
, segno
);
1681 new = get_meta_page(sbi
, blk_addr
);
1682 sum
= (struct f2fs_summary_block
*)page_address(new);
1684 if (IS_NODESEG(type
)) {
1685 if (__exist_node_summaries(sbi
)) {
1686 struct f2fs_summary
*ns
= &sum
->entries
[0];
1688 for (i
= 0; i
< sbi
->blocks_per_seg
; i
++, ns
++) {
1690 ns
->ofs_in_node
= 0;
1695 err
= restore_node_summary(sbi
, segno
, sum
);
1697 f2fs_put_page(new, 1);
1703 /* set uncompleted segment to curseg */
1704 curseg
= CURSEG_I(sbi
, type
);
1705 mutex_lock(&curseg
->curseg_mutex
);
1707 /* update journal info */
1708 down_write(&curseg
->journal_rwsem
);
1709 memcpy(curseg
->journal
, &sum
->journal
, SUM_JOURNAL_SIZE
);
1710 up_write(&curseg
->journal_rwsem
);
1712 memcpy(curseg
->sum_blk
->entries
, sum
->entries
, SUM_ENTRY_SIZE
);
1713 memcpy(&curseg
->sum_blk
->footer
, &sum
->footer
, SUM_FOOTER_SIZE
);
1714 curseg
->next_segno
= segno
;
1715 reset_curseg(sbi
, type
, 0);
1716 curseg
->alloc_type
= ckpt
->alloc_type
[type
];
1717 curseg
->next_blkoff
= blk_off
;
1718 mutex_unlock(&curseg
->curseg_mutex
);
1719 f2fs_put_page(new, 1);
1723 static int restore_curseg_summaries(struct f2fs_sb_info
*sbi
)
1725 int type
= CURSEG_HOT_DATA
;
1728 if (is_set_ckpt_flags(F2FS_CKPT(sbi
), CP_COMPACT_SUM_FLAG
)) {
1729 int npages
= npages_for_summary_flush(sbi
, true);
1732 ra_meta_pages(sbi
, start_sum_block(sbi
), npages
,
1735 /* restore for compacted data summary */
1736 if (read_compacted_summaries(sbi
))
1738 type
= CURSEG_HOT_NODE
;
1741 if (__exist_node_summaries(sbi
))
1742 ra_meta_pages(sbi
, sum_blk_addr(sbi
, NR_CURSEG_TYPE
, type
),
1743 NR_CURSEG_TYPE
- type
, META_CP
, true);
1745 for (; type
<= CURSEG_COLD_NODE
; type
++) {
1746 err
= read_normal_summaries(sbi
, type
);
1754 static void write_compacted_summaries(struct f2fs_sb_info
*sbi
, block_t blkaddr
)
1757 unsigned char *kaddr
;
1758 struct f2fs_summary
*summary
;
1759 struct curseg_info
*seg_i
;
1760 int written_size
= 0;
1763 page
= grab_meta_page(sbi
, blkaddr
++);
1764 kaddr
= (unsigned char *)page_address(page
);
1766 /* Step 1: write nat cache */
1767 seg_i
= CURSEG_I(sbi
, CURSEG_HOT_DATA
);
1768 memcpy(kaddr
, seg_i
->journal
, SUM_JOURNAL_SIZE
);
1769 written_size
+= SUM_JOURNAL_SIZE
;
1771 /* Step 2: write sit cache */
1772 seg_i
= CURSEG_I(sbi
, CURSEG_COLD_DATA
);
1773 memcpy(kaddr
+ written_size
, seg_i
->journal
, SUM_JOURNAL_SIZE
);
1774 written_size
+= SUM_JOURNAL_SIZE
;
1776 /* Step 3: write summary entries */
1777 for (i
= CURSEG_HOT_DATA
; i
<= CURSEG_COLD_DATA
; i
++) {
1778 unsigned short blkoff
;
1779 seg_i
= CURSEG_I(sbi
, i
);
1780 if (sbi
->ckpt
->alloc_type
[i
] == SSR
)
1781 blkoff
= sbi
->blocks_per_seg
;
1783 blkoff
= curseg_blkoff(sbi
, i
);
1785 for (j
= 0; j
< blkoff
; j
++) {
1787 page
= grab_meta_page(sbi
, blkaddr
++);
1788 kaddr
= (unsigned char *)page_address(page
);
1791 summary
= (struct f2fs_summary
*)(kaddr
+ written_size
);
1792 *summary
= seg_i
->sum_blk
->entries
[j
];
1793 written_size
+= SUMMARY_SIZE
;
1795 if (written_size
+ SUMMARY_SIZE
<= PAGE_SIZE
-
1799 set_page_dirty(page
);
1800 f2fs_put_page(page
, 1);
1805 set_page_dirty(page
);
1806 f2fs_put_page(page
, 1);
1810 static void write_normal_summaries(struct f2fs_sb_info
*sbi
,
1811 block_t blkaddr
, int type
)
1814 if (IS_DATASEG(type
))
1815 end
= type
+ NR_CURSEG_DATA_TYPE
;
1817 end
= type
+ NR_CURSEG_NODE_TYPE
;
1819 for (i
= type
; i
< end
; i
++)
1820 write_current_sum_page(sbi
, i
, blkaddr
+ (i
- type
));
1823 void write_data_summaries(struct f2fs_sb_info
*sbi
, block_t start_blk
)
1825 if (is_set_ckpt_flags(F2FS_CKPT(sbi
), CP_COMPACT_SUM_FLAG
))
1826 write_compacted_summaries(sbi
, start_blk
);
1828 write_normal_summaries(sbi
, start_blk
, CURSEG_HOT_DATA
);
1831 void write_node_summaries(struct f2fs_sb_info
*sbi
, block_t start_blk
)
1833 write_normal_summaries(sbi
, start_blk
, CURSEG_HOT_NODE
);
1836 int lookup_journal_in_cursum(struct f2fs_journal
*journal
, int type
,
1837 unsigned int val
, int alloc
)
1841 if (type
== NAT_JOURNAL
) {
1842 for (i
= 0; i
< nats_in_cursum(journal
); i
++) {
1843 if (le32_to_cpu(nid_in_journal(journal
, i
)) == val
)
1846 if (alloc
&& __has_cursum_space(journal
, 1, NAT_JOURNAL
))
1847 return update_nats_in_cursum(journal
, 1);
1848 } else if (type
== SIT_JOURNAL
) {
1849 for (i
= 0; i
< sits_in_cursum(journal
); i
++)
1850 if (le32_to_cpu(segno_in_journal(journal
, i
)) == val
)
1852 if (alloc
&& __has_cursum_space(journal
, 1, SIT_JOURNAL
))
1853 return update_sits_in_cursum(journal
, 1);
1858 static struct page
*get_current_sit_page(struct f2fs_sb_info
*sbi
,
1861 return get_meta_page(sbi
, current_sit_addr(sbi
, segno
));
1864 static struct page
*get_next_sit_page(struct f2fs_sb_info
*sbi
,
1867 struct sit_info
*sit_i
= SIT_I(sbi
);
1868 struct page
*src_page
, *dst_page
;
1869 pgoff_t src_off
, dst_off
;
1870 void *src_addr
, *dst_addr
;
1872 src_off
= current_sit_addr(sbi
, start
);
1873 dst_off
= next_sit_addr(sbi
, src_off
);
1875 /* get current sit block page without lock */
1876 src_page
= get_meta_page(sbi
, src_off
);
1877 dst_page
= grab_meta_page(sbi
, dst_off
);
1878 f2fs_bug_on(sbi
, PageDirty(src_page
));
1880 src_addr
= page_address(src_page
);
1881 dst_addr
= page_address(dst_page
);
1882 memcpy(dst_addr
, src_addr
, PAGE_SIZE
);
1884 set_page_dirty(dst_page
);
1885 f2fs_put_page(src_page
, 1);
1887 set_to_next_sit(sit_i
, start
);
1892 static struct sit_entry_set
*grab_sit_entry_set(void)
1894 struct sit_entry_set
*ses
=
1895 f2fs_kmem_cache_alloc(sit_entry_set_slab
, GFP_NOFS
);
1898 INIT_LIST_HEAD(&ses
->set_list
);
1902 static void release_sit_entry_set(struct sit_entry_set
*ses
)
1904 list_del(&ses
->set_list
);
1905 kmem_cache_free(sit_entry_set_slab
, ses
);
1908 static void adjust_sit_entry_set(struct sit_entry_set
*ses
,
1909 struct list_head
*head
)
1911 struct sit_entry_set
*next
= ses
;
1913 if (list_is_last(&ses
->set_list
, head
))
1916 list_for_each_entry_continue(next
, head
, set_list
)
1917 if (ses
->entry_cnt
<= next
->entry_cnt
)
1920 list_move_tail(&ses
->set_list
, &next
->set_list
);
1923 static void add_sit_entry(unsigned int segno
, struct list_head
*head
)
1925 struct sit_entry_set
*ses
;
1926 unsigned int start_segno
= START_SEGNO(segno
);
1928 list_for_each_entry(ses
, head
, set_list
) {
1929 if (ses
->start_segno
== start_segno
) {
1931 adjust_sit_entry_set(ses
, head
);
1936 ses
= grab_sit_entry_set();
1938 ses
->start_segno
= start_segno
;
1940 list_add(&ses
->set_list
, head
);
1943 static void add_sits_in_set(struct f2fs_sb_info
*sbi
)
1945 struct f2fs_sm_info
*sm_info
= SM_I(sbi
);
1946 struct list_head
*set_list
= &sm_info
->sit_entry_set
;
1947 unsigned long *bitmap
= SIT_I(sbi
)->dirty_sentries_bitmap
;
1950 for_each_set_bit(segno
, bitmap
, MAIN_SEGS(sbi
))
1951 add_sit_entry(segno
, set_list
);
1954 static void remove_sits_in_journal(struct f2fs_sb_info
*sbi
)
1956 struct curseg_info
*curseg
= CURSEG_I(sbi
, CURSEG_COLD_DATA
);
1957 struct f2fs_journal
*journal
= curseg
->journal
;
1960 down_write(&curseg
->journal_rwsem
);
1961 for (i
= 0; i
< sits_in_cursum(journal
); i
++) {
1965 segno
= le32_to_cpu(segno_in_journal(journal
, i
));
1966 dirtied
= __mark_sit_entry_dirty(sbi
, segno
);
1969 add_sit_entry(segno
, &SM_I(sbi
)->sit_entry_set
);
1971 update_sits_in_cursum(journal
, -i
);
1972 up_write(&curseg
->journal_rwsem
);
1976 * CP calls this function, which flushes SIT entries including sit_journal,
1977 * and moves prefree segs to free segs.
1979 void flush_sit_entries(struct f2fs_sb_info
*sbi
, struct cp_control
*cpc
)
1981 struct sit_info
*sit_i
= SIT_I(sbi
);
1982 unsigned long *bitmap
= sit_i
->dirty_sentries_bitmap
;
1983 struct curseg_info
*curseg
= CURSEG_I(sbi
, CURSEG_COLD_DATA
);
1984 struct f2fs_journal
*journal
= curseg
->journal
;
1985 struct sit_entry_set
*ses
, *tmp
;
1986 struct list_head
*head
= &SM_I(sbi
)->sit_entry_set
;
1987 bool to_journal
= true;
1988 struct seg_entry
*se
;
1990 mutex_lock(&sit_i
->sentry_lock
);
1992 if (!sit_i
->dirty_sentries
)
1996 * add and account sit entries of dirty bitmap in sit entry
1999 add_sits_in_set(sbi
);
2002 * if there are no enough space in journal to store dirty sit
2003 * entries, remove all entries from journal and add and account
2004 * them in sit entry set.
2006 if (!__has_cursum_space(journal
, sit_i
->dirty_sentries
, SIT_JOURNAL
))
2007 remove_sits_in_journal(sbi
);
2010 * there are two steps to flush sit entries:
2011 * #1, flush sit entries to journal in current cold data summary block.
2012 * #2, flush sit entries to sit page.
2014 list_for_each_entry_safe(ses
, tmp
, head
, set_list
) {
2015 struct page
*page
= NULL
;
2016 struct f2fs_sit_block
*raw_sit
= NULL
;
2017 unsigned int start_segno
= ses
->start_segno
;
2018 unsigned int end
= min(start_segno
+ SIT_ENTRY_PER_BLOCK
,
2019 (unsigned long)MAIN_SEGS(sbi
));
2020 unsigned int segno
= start_segno
;
2023 !__has_cursum_space(journal
, ses
->entry_cnt
, SIT_JOURNAL
))
2027 down_write(&curseg
->journal_rwsem
);
2029 page
= get_next_sit_page(sbi
, start_segno
);
2030 raw_sit
= page_address(page
);
2033 /* flush dirty sit entries in region of current sit set */
2034 for_each_set_bit_from(segno
, bitmap
, end
) {
2035 int offset
, sit_offset
;
2037 se
= get_seg_entry(sbi
, segno
);
2039 /* add discard candidates */
2040 if (cpc
->reason
!= CP_DISCARD
) {
2041 cpc
->trim_start
= segno
;
2042 add_discard_addrs(sbi
, cpc
);
2046 offset
= lookup_journal_in_cursum(journal
,
2047 SIT_JOURNAL
, segno
, 1);
2048 f2fs_bug_on(sbi
, offset
< 0);
2049 segno_in_journal(journal
, offset
) =
2051 seg_info_to_raw_sit(se
,
2052 &sit_in_journal(journal
, offset
));
2054 sit_offset
= SIT_ENTRY_OFFSET(sit_i
, segno
);
2055 seg_info_to_raw_sit(se
,
2056 &raw_sit
->entries
[sit_offset
]);
2059 __clear_bit(segno
, bitmap
);
2060 sit_i
->dirty_sentries
--;
2065 up_write(&curseg
->journal_rwsem
);
2067 f2fs_put_page(page
, 1);
2069 f2fs_bug_on(sbi
, ses
->entry_cnt
);
2070 release_sit_entry_set(ses
);
2073 f2fs_bug_on(sbi
, !list_empty(head
));
2074 f2fs_bug_on(sbi
, sit_i
->dirty_sentries
);
2076 if (cpc
->reason
== CP_DISCARD
) {
2077 for (; cpc
->trim_start
<= cpc
->trim_end
; cpc
->trim_start
++)
2078 add_discard_addrs(sbi
, cpc
);
2080 mutex_unlock(&sit_i
->sentry_lock
);
2082 set_prefree_as_free_segments(sbi
);
2085 static int build_sit_info(struct f2fs_sb_info
*sbi
)
2087 struct f2fs_super_block
*raw_super
= F2FS_RAW_SUPER(sbi
);
2088 struct f2fs_checkpoint
*ckpt
= F2FS_CKPT(sbi
);
2089 struct sit_info
*sit_i
;
2090 unsigned int sit_segs
, start
;
2091 char *src_bitmap
, *dst_bitmap
;
2092 unsigned int bitmap_size
;
2094 /* allocate memory for SIT information */
2095 sit_i
= kzalloc(sizeof(struct sit_info
), GFP_KERNEL
);
2099 SM_I(sbi
)->sit_info
= sit_i
;
2101 sit_i
->sentries
= f2fs_kvzalloc(MAIN_SEGS(sbi
) *
2102 sizeof(struct seg_entry
), GFP_KERNEL
);
2103 if (!sit_i
->sentries
)
2106 bitmap_size
= f2fs_bitmap_size(MAIN_SEGS(sbi
));
2107 sit_i
->dirty_sentries_bitmap
= f2fs_kvzalloc(bitmap_size
, GFP_KERNEL
);
2108 if (!sit_i
->dirty_sentries_bitmap
)
2111 for (start
= 0; start
< MAIN_SEGS(sbi
); start
++) {
2112 sit_i
->sentries
[start
].cur_valid_map
2113 = kzalloc(SIT_VBLOCK_MAP_SIZE
, GFP_KERNEL
);
2114 sit_i
->sentries
[start
].ckpt_valid_map
2115 = kzalloc(SIT_VBLOCK_MAP_SIZE
, GFP_KERNEL
);
2116 sit_i
->sentries
[start
].discard_map
2117 = kzalloc(SIT_VBLOCK_MAP_SIZE
, GFP_KERNEL
);
2118 if (!sit_i
->sentries
[start
].cur_valid_map
||
2119 !sit_i
->sentries
[start
].ckpt_valid_map
||
2120 !sit_i
->sentries
[start
].discard_map
)
2124 sit_i
->tmp_map
= kzalloc(SIT_VBLOCK_MAP_SIZE
, GFP_KERNEL
);
2125 if (!sit_i
->tmp_map
)
2128 if (sbi
->segs_per_sec
> 1) {
2129 sit_i
->sec_entries
= f2fs_kvzalloc(MAIN_SECS(sbi
) *
2130 sizeof(struct sec_entry
), GFP_KERNEL
);
2131 if (!sit_i
->sec_entries
)
2135 /* get information related with SIT */
2136 sit_segs
= le32_to_cpu(raw_super
->segment_count_sit
) >> 1;
2138 /* setup SIT bitmap from ckeckpoint pack */
2139 bitmap_size
= __bitmap_size(sbi
, SIT_BITMAP
);
2140 src_bitmap
= __bitmap_ptr(sbi
, SIT_BITMAP
);
2142 dst_bitmap
= kmemdup(src_bitmap
, bitmap_size
, GFP_KERNEL
);
2146 /* init SIT information */
2147 sit_i
->s_ops
= &default_salloc_ops
;
2149 sit_i
->sit_base_addr
= le32_to_cpu(raw_super
->sit_blkaddr
);
2150 sit_i
->sit_blocks
= sit_segs
<< sbi
->log_blocks_per_seg
;
2151 sit_i
->written_valid_blocks
= le64_to_cpu(ckpt
->valid_block_count
);
2152 sit_i
->sit_bitmap
= dst_bitmap
;
2153 sit_i
->bitmap_size
= bitmap_size
;
2154 sit_i
->dirty_sentries
= 0;
2155 sit_i
->sents_per_block
= SIT_ENTRY_PER_BLOCK
;
2156 sit_i
->elapsed_time
= le64_to_cpu(sbi
->ckpt
->elapsed_time
);
2157 sit_i
->mounted_time
= CURRENT_TIME_SEC
.tv_sec
;
2158 mutex_init(&sit_i
->sentry_lock
);
2162 static int build_free_segmap(struct f2fs_sb_info
*sbi
)
2164 struct free_segmap_info
*free_i
;
2165 unsigned int bitmap_size
, sec_bitmap_size
;
2167 /* allocate memory for free segmap information */
2168 free_i
= kzalloc(sizeof(struct free_segmap_info
), GFP_KERNEL
);
2172 SM_I(sbi
)->free_info
= free_i
;
2174 bitmap_size
= f2fs_bitmap_size(MAIN_SEGS(sbi
));
2175 free_i
->free_segmap
= f2fs_kvmalloc(bitmap_size
, GFP_KERNEL
);
2176 if (!free_i
->free_segmap
)
2179 sec_bitmap_size
= f2fs_bitmap_size(MAIN_SECS(sbi
));
2180 free_i
->free_secmap
= f2fs_kvmalloc(sec_bitmap_size
, GFP_KERNEL
);
2181 if (!free_i
->free_secmap
)
2184 /* set all segments as dirty temporarily */
2185 memset(free_i
->free_segmap
, 0xff, bitmap_size
);
2186 memset(free_i
->free_secmap
, 0xff, sec_bitmap_size
);
2188 /* init free segmap information */
2189 free_i
->start_segno
= GET_SEGNO_FROM_SEG0(sbi
, MAIN_BLKADDR(sbi
));
2190 free_i
->free_segments
= 0;
2191 free_i
->free_sections
= 0;
2192 spin_lock_init(&free_i
->segmap_lock
);
2196 static int build_curseg(struct f2fs_sb_info
*sbi
)
2198 struct curseg_info
*array
;
2201 array
= kcalloc(NR_CURSEG_TYPE
, sizeof(*array
), GFP_KERNEL
);
2205 SM_I(sbi
)->curseg_array
= array
;
2207 for (i
= 0; i
< NR_CURSEG_TYPE
; i
++) {
2208 mutex_init(&array
[i
].curseg_mutex
);
2209 array
[i
].sum_blk
= kzalloc(PAGE_SIZE
, GFP_KERNEL
);
2210 if (!array
[i
].sum_blk
)
2212 init_rwsem(&array
[i
].journal_rwsem
);
2213 array
[i
].journal
= kzalloc(sizeof(struct f2fs_journal
),
2215 if (!array
[i
].journal
)
2217 array
[i
].segno
= NULL_SEGNO
;
2218 array
[i
].next_blkoff
= 0;
2220 return restore_curseg_summaries(sbi
);
2223 static void build_sit_entries(struct f2fs_sb_info
*sbi
)
2225 struct sit_info
*sit_i
= SIT_I(sbi
);
2226 struct curseg_info
*curseg
= CURSEG_I(sbi
, CURSEG_COLD_DATA
);
2227 struct f2fs_journal
*journal
= curseg
->journal
;
2228 int sit_blk_cnt
= SIT_BLK_CNT(sbi
);
2229 unsigned int i
, start
, end
;
2230 unsigned int readed
, start_blk
= 0;
2231 int nrpages
= MAX_BIO_BLOCKS(sbi
) * 8;
2234 readed
= ra_meta_pages(sbi
, start_blk
, nrpages
, META_SIT
, true);
2236 start
= start_blk
* sit_i
->sents_per_block
;
2237 end
= (start_blk
+ readed
) * sit_i
->sents_per_block
;
2239 for (; start
< end
&& start
< MAIN_SEGS(sbi
); start
++) {
2240 struct seg_entry
*se
= &sit_i
->sentries
[start
];
2241 struct f2fs_sit_block
*sit_blk
;
2242 struct f2fs_sit_entry sit
;
2245 down_read(&curseg
->journal_rwsem
);
2246 for (i
= 0; i
< sits_in_cursum(journal
); i
++) {
2247 if (le32_to_cpu(segno_in_journal(journal
, i
))
2249 sit
= sit_in_journal(journal
, i
);
2250 up_read(&curseg
->journal_rwsem
);
2254 up_read(&curseg
->journal_rwsem
);
2256 page
= get_current_sit_page(sbi
, start
);
2257 sit_blk
= (struct f2fs_sit_block
*)page_address(page
);
2258 sit
= sit_blk
->entries
[SIT_ENTRY_OFFSET(sit_i
, start
)];
2259 f2fs_put_page(page
, 1);
2261 check_block_count(sbi
, start
, &sit
);
2262 seg_info_from_raw_sit(se
, &sit
);
2264 /* build discard map only one time */
2265 memcpy(se
->discard_map
, se
->cur_valid_map
, SIT_VBLOCK_MAP_SIZE
);
2266 sbi
->discard_blks
+= sbi
->blocks_per_seg
- se
->valid_blocks
;
2268 if (sbi
->segs_per_sec
> 1) {
2269 struct sec_entry
*e
= get_sec_entry(sbi
, start
);
2270 e
->valid_blocks
+= se
->valid_blocks
;
2273 start_blk
+= readed
;
2274 } while (start_blk
< sit_blk_cnt
);
2277 static void init_free_segmap(struct f2fs_sb_info
*sbi
)
2282 for (start
= 0; start
< MAIN_SEGS(sbi
); start
++) {
2283 struct seg_entry
*sentry
= get_seg_entry(sbi
, start
);
2284 if (!sentry
->valid_blocks
)
2285 __set_free(sbi
, start
);
2288 /* set use the current segments */
2289 for (type
= CURSEG_HOT_DATA
; type
<= CURSEG_COLD_NODE
; type
++) {
2290 struct curseg_info
*curseg_t
= CURSEG_I(sbi
, type
);
2291 __set_test_and_inuse(sbi
, curseg_t
->segno
);
2295 static void init_dirty_segmap(struct f2fs_sb_info
*sbi
)
2297 struct dirty_seglist_info
*dirty_i
= DIRTY_I(sbi
);
2298 struct free_segmap_info
*free_i
= FREE_I(sbi
);
2299 unsigned int segno
= 0, offset
= 0;
2300 unsigned short valid_blocks
;
2303 /* find dirty segment based on free segmap */
2304 segno
= find_next_inuse(free_i
, MAIN_SEGS(sbi
), offset
);
2305 if (segno
>= MAIN_SEGS(sbi
))
2308 valid_blocks
= get_valid_blocks(sbi
, segno
, 0);
2309 if (valid_blocks
== sbi
->blocks_per_seg
|| !valid_blocks
)
2311 if (valid_blocks
> sbi
->blocks_per_seg
) {
2312 f2fs_bug_on(sbi
, 1);
2315 mutex_lock(&dirty_i
->seglist_lock
);
2316 __locate_dirty_segment(sbi
, segno
, DIRTY
);
2317 mutex_unlock(&dirty_i
->seglist_lock
);
2321 static int init_victim_secmap(struct f2fs_sb_info
*sbi
)
2323 struct dirty_seglist_info
*dirty_i
= DIRTY_I(sbi
);
2324 unsigned int bitmap_size
= f2fs_bitmap_size(MAIN_SECS(sbi
));
2326 dirty_i
->victim_secmap
= f2fs_kvzalloc(bitmap_size
, GFP_KERNEL
);
2327 if (!dirty_i
->victim_secmap
)
2332 static int build_dirty_segmap(struct f2fs_sb_info
*sbi
)
2334 struct dirty_seglist_info
*dirty_i
;
2335 unsigned int bitmap_size
, i
;
2337 /* allocate memory for dirty segments list information */
2338 dirty_i
= kzalloc(sizeof(struct dirty_seglist_info
), GFP_KERNEL
);
2342 SM_I(sbi
)->dirty_info
= dirty_i
;
2343 mutex_init(&dirty_i
->seglist_lock
);
2345 bitmap_size
= f2fs_bitmap_size(MAIN_SEGS(sbi
));
2347 for (i
= 0; i
< NR_DIRTY_TYPE
; i
++) {
2348 dirty_i
->dirty_segmap
[i
] = f2fs_kvzalloc(bitmap_size
, GFP_KERNEL
);
2349 if (!dirty_i
->dirty_segmap
[i
])
2353 init_dirty_segmap(sbi
);
2354 return init_victim_secmap(sbi
);
2358 * Update min, max modified time for cost-benefit GC algorithm
2360 static void init_min_max_mtime(struct f2fs_sb_info
*sbi
)
2362 struct sit_info
*sit_i
= SIT_I(sbi
);
2365 mutex_lock(&sit_i
->sentry_lock
);
2367 sit_i
->min_mtime
= LLONG_MAX
;
2369 for (segno
= 0; segno
< MAIN_SEGS(sbi
); segno
+= sbi
->segs_per_sec
) {
2371 unsigned long long mtime
= 0;
2373 for (i
= 0; i
< sbi
->segs_per_sec
; i
++)
2374 mtime
+= get_seg_entry(sbi
, segno
+ i
)->mtime
;
2376 mtime
= div_u64(mtime
, sbi
->segs_per_sec
);
2378 if (sit_i
->min_mtime
> mtime
)
2379 sit_i
->min_mtime
= mtime
;
2381 sit_i
->max_mtime
= get_mtime(sbi
);
2382 mutex_unlock(&sit_i
->sentry_lock
);
2385 int build_segment_manager(struct f2fs_sb_info
*sbi
)
2387 struct f2fs_super_block
*raw_super
= F2FS_RAW_SUPER(sbi
);
2388 struct f2fs_checkpoint
*ckpt
= F2FS_CKPT(sbi
);
2389 struct f2fs_sm_info
*sm_info
;
2392 sm_info
= kzalloc(sizeof(struct f2fs_sm_info
), GFP_KERNEL
);
2397 sbi
->sm_info
= sm_info
;
2398 sm_info
->seg0_blkaddr
= le32_to_cpu(raw_super
->segment0_blkaddr
);
2399 sm_info
->main_blkaddr
= le32_to_cpu(raw_super
->main_blkaddr
);
2400 sm_info
->segment_count
= le32_to_cpu(raw_super
->segment_count
);
2401 sm_info
->reserved_segments
= le32_to_cpu(ckpt
->rsvd_segment_count
);
2402 sm_info
->ovp_segments
= le32_to_cpu(ckpt
->overprov_segment_count
);
2403 sm_info
->main_segments
= le32_to_cpu(raw_super
->segment_count_main
);
2404 sm_info
->ssa_blkaddr
= le32_to_cpu(raw_super
->ssa_blkaddr
);
2405 sm_info
->rec_prefree_segments
= sm_info
->main_segments
*
2406 DEF_RECLAIM_PREFREE_SEGMENTS
/ 100;
2407 if (!test_opt(sbi
, LFS
))
2408 sm_info
->ipu_policy
= 1 << F2FS_IPU_FSYNC
;
2409 sm_info
->min_ipu_util
= DEF_MIN_IPU_UTIL
;
2410 sm_info
->min_fsync_blocks
= DEF_MIN_FSYNC_BLOCKS
;
2412 INIT_LIST_HEAD(&sm_info
->discard_list
);
2413 sm_info
->nr_discards
= 0;
2414 sm_info
->max_discards
= 0;
2416 sm_info
->trim_sections
= DEF_BATCHED_TRIM_SECTIONS
;
2418 INIT_LIST_HEAD(&sm_info
->sit_entry_set
);
2420 if (test_opt(sbi
, FLUSH_MERGE
) && !f2fs_readonly(sbi
->sb
)) {
2421 err
= create_flush_cmd_control(sbi
);
2426 err
= build_sit_info(sbi
);
2429 err
= build_free_segmap(sbi
);
2432 err
= build_curseg(sbi
);
2436 /* reinit free segmap based on SIT */
2437 build_sit_entries(sbi
);
2439 init_free_segmap(sbi
);
2440 err
= build_dirty_segmap(sbi
);
2444 init_min_max_mtime(sbi
);
2448 static void discard_dirty_segmap(struct f2fs_sb_info
*sbi
,
2449 enum dirty_type dirty_type
)
2451 struct dirty_seglist_info
*dirty_i
= DIRTY_I(sbi
);
2453 mutex_lock(&dirty_i
->seglist_lock
);
2454 kvfree(dirty_i
->dirty_segmap
[dirty_type
]);
2455 dirty_i
->nr_dirty
[dirty_type
] = 0;
2456 mutex_unlock(&dirty_i
->seglist_lock
);
2459 static void destroy_victim_secmap(struct f2fs_sb_info
*sbi
)
2461 struct dirty_seglist_info
*dirty_i
= DIRTY_I(sbi
);
2462 kvfree(dirty_i
->victim_secmap
);
2465 static void destroy_dirty_segmap(struct f2fs_sb_info
*sbi
)
2467 struct dirty_seglist_info
*dirty_i
= DIRTY_I(sbi
);
2473 /* discard pre-free/dirty segments list */
2474 for (i
= 0; i
< NR_DIRTY_TYPE
; i
++)
2475 discard_dirty_segmap(sbi
, i
);
2477 destroy_victim_secmap(sbi
);
2478 SM_I(sbi
)->dirty_info
= NULL
;
2482 static void destroy_curseg(struct f2fs_sb_info
*sbi
)
2484 struct curseg_info
*array
= SM_I(sbi
)->curseg_array
;
2489 SM_I(sbi
)->curseg_array
= NULL
;
2490 for (i
= 0; i
< NR_CURSEG_TYPE
; i
++) {
2491 kfree(array
[i
].sum_blk
);
2492 kfree(array
[i
].journal
);
2497 static void destroy_free_segmap(struct f2fs_sb_info
*sbi
)
2499 struct free_segmap_info
*free_i
= SM_I(sbi
)->free_info
;
2502 SM_I(sbi
)->free_info
= NULL
;
2503 kvfree(free_i
->free_segmap
);
2504 kvfree(free_i
->free_secmap
);
2508 static void destroy_sit_info(struct f2fs_sb_info
*sbi
)
2510 struct sit_info
*sit_i
= SIT_I(sbi
);
2516 if (sit_i
->sentries
) {
2517 for (start
= 0; start
< MAIN_SEGS(sbi
); start
++) {
2518 kfree(sit_i
->sentries
[start
].cur_valid_map
);
2519 kfree(sit_i
->sentries
[start
].ckpt_valid_map
);
2520 kfree(sit_i
->sentries
[start
].discard_map
);
2523 kfree(sit_i
->tmp_map
);
2525 kvfree(sit_i
->sentries
);
2526 kvfree(sit_i
->sec_entries
);
2527 kvfree(sit_i
->dirty_sentries_bitmap
);
2529 SM_I(sbi
)->sit_info
= NULL
;
2530 kfree(sit_i
->sit_bitmap
);
2534 void destroy_segment_manager(struct f2fs_sb_info
*sbi
)
2536 struct f2fs_sm_info
*sm_info
= SM_I(sbi
);
2540 destroy_flush_cmd_control(sbi
);
2541 destroy_dirty_segmap(sbi
);
2542 destroy_curseg(sbi
);
2543 destroy_free_segmap(sbi
);
2544 destroy_sit_info(sbi
);
2545 sbi
->sm_info
= NULL
;
2549 int __init
create_segment_manager_caches(void)
2551 discard_entry_slab
= f2fs_kmem_cache_create("discard_entry",
2552 sizeof(struct discard_entry
));
2553 if (!discard_entry_slab
)
2556 sit_entry_set_slab
= f2fs_kmem_cache_create("sit_entry_set",
2557 sizeof(struct sit_entry_set
));
2558 if (!sit_entry_set_slab
)
2559 goto destory_discard_entry
;
2561 inmem_entry_slab
= f2fs_kmem_cache_create("inmem_page_entry",
2562 sizeof(struct inmem_pages
));
2563 if (!inmem_entry_slab
)
2564 goto destroy_sit_entry_set
;
2567 destroy_sit_entry_set
:
2568 kmem_cache_destroy(sit_entry_set_slab
);
2569 destory_discard_entry
:
2570 kmem_cache_destroy(discard_entry_slab
);
2575 void destroy_segment_manager_caches(void)
2577 kmem_cache_destroy(sit_entry_set_slab
);
2578 kmem_cache_destroy(discard_entry_slab
);
2579 kmem_cache_destroy(inmem_entry_slab
);