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
, NAT_ENTRY_PER_BLOCK
* FREE_NID_PAGES
);
376 /* checkpoint is the only way to shrink partial cached entries */
377 if (!available_free_memory(sbi
, NAT_ENTRIES
) ||
378 !available_free_memory(sbi
, INO_ENTRIES
) ||
379 excess_prefree_segs(sbi
) ||
380 excess_dirty_nats(sbi
) ||
381 (is_idle(sbi
) && f2fs_time_over(sbi
, CP_TIME
))) {
382 if (test_opt(sbi
, DATA_FLUSH
))
383 sync_dirty_inodes(sbi
, FILE_INODE
);
384 f2fs_sync_fs(sbi
->sb
, true);
385 stat_inc_bg_cp_count(sbi
->stat_info
);
389 static int issue_flush_thread(void *data
)
391 struct f2fs_sb_info
*sbi
= data
;
392 struct flush_cmd_control
*fcc
= SM_I(sbi
)->cmd_control_info
;
393 wait_queue_head_t
*q
= &fcc
->flush_wait_queue
;
395 if (kthread_should_stop())
398 if (!llist_empty(&fcc
->issue_list
)) {
400 struct flush_cmd
*cmd
, *next
;
403 bio
= f2fs_bio_alloc(0);
405 fcc
->dispatch_list
= llist_del_all(&fcc
->issue_list
);
406 fcc
->dispatch_list
= llist_reverse_order(fcc
->dispatch_list
);
408 bio
->bi_bdev
= sbi
->sb
->s_bdev
;
409 ret
= submit_bio_wait(WRITE_FLUSH
, bio
);
411 llist_for_each_entry_safe(cmd
, next
,
412 fcc
->dispatch_list
, llnode
) {
414 complete(&cmd
->wait
);
417 fcc
->dispatch_list
= NULL
;
420 wait_event_interruptible(*q
,
421 kthread_should_stop() || !llist_empty(&fcc
->issue_list
));
425 int f2fs_issue_flush(struct f2fs_sb_info
*sbi
)
427 struct flush_cmd_control
*fcc
= SM_I(sbi
)->cmd_control_info
;
428 struct flush_cmd cmd
;
430 trace_f2fs_issue_flush(sbi
->sb
, test_opt(sbi
, NOBARRIER
),
431 test_opt(sbi
, FLUSH_MERGE
));
433 if (test_opt(sbi
, NOBARRIER
))
436 if (!test_opt(sbi
, FLUSH_MERGE
) || !atomic_read(&fcc
->submit_flush
)) {
437 struct bio
*bio
= f2fs_bio_alloc(0);
440 atomic_inc(&fcc
->submit_flush
);
441 bio
->bi_bdev
= sbi
->sb
->s_bdev
;
442 ret
= submit_bio_wait(WRITE_FLUSH
, bio
);
443 atomic_dec(&fcc
->submit_flush
);
448 init_completion(&cmd
.wait
);
450 atomic_inc(&fcc
->submit_flush
);
451 llist_add(&cmd
.llnode
, &fcc
->issue_list
);
453 if (!fcc
->dispatch_list
)
454 wake_up(&fcc
->flush_wait_queue
);
456 wait_for_completion(&cmd
.wait
);
457 atomic_dec(&fcc
->submit_flush
);
462 int create_flush_cmd_control(struct f2fs_sb_info
*sbi
)
464 dev_t dev
= sbi
->sb
->s_bdev
->bd_dev
;
465 struct flush_cmd_control
*fcc
;
468 fcc
= kzalloc(sizeof(struct flush_cmd_control
), GFP_KERNEL
);
471 atomic_set(&fcc
->submit_flush
, 0);
472 init_waitqueue_head(&fcc
->flush_wait_queue
);
473 init_llist_head(&fcc
->issue_list
);
474 SM_I(sbi
)->cmd_control_info
= fcc
;
475 fcc
->f2fs_issue_flush
= kthread_run(issue_flush_thread
, sbi
,
476 "f2fs_flush-%u:%u", MAJOR(dev
), MINOR(dev
));
477 if (IS_ERR(fcc
->f2fs_issue_flush
)) {
478 err
= PTR_ERR(fcc
->f2fs_issue_flush
);
480 SM_I(sbi
)->cmd_control_info
= NULL
;
487 void destroy_flush_cmd_control(struct f2fs_sb_info
*sbi
)
489 struct flush_cmd_control
*fcc
= SM_I(sbi
)->cmd_control_info
;
491 if (fcc
&& fcc
->f2fs_issue_flush
)
492 kthread_stop(fcc
->f2fs_issue_flush
);
494 SM_I(sbi
)->cmd_control_info
= NULL
;
497 static void __locate_dirty_segment(struct f2fs_sb_info
*sbi
, unsigned int segno
,
498 enum dirty_type dirty_type
)
500 struct dirty_seglist_info
*dirty_i
= DIRTY_I(sbi
);
502 /* need not be added */
503 if (IS_CURSEG(sbi
, segno
))
506 if (!test_and_set_bit(segno
, dirty_i
->dirty_segmap
[dirty_type
]))
507 dirty_i
->nr_dirty
[dirty_type
]++;
509 if (dirty_type
== DIRTY
) {
510 struct seg_entry
*sentry
= get_seg_entry(sbi
, segno
);
511 enum dirty_type t
= sentry
->type
;
513 if (unlikely(t
>= DIRTY
)) {
517 if (!test_and_set_bit(segno
, dirty_i
->dirty_segmap
[t
]))
518 dirty_i
->nr_dirty
[t
]++;
522 static void __remove_dirty_segment(struct f2fs_sb_info
*sbi
, unsigned int segno
,
523 enum dirty_type dirty_type
)
525 struct dirty_seglist_info
*dirty_i
= DIRTY_I(sbi
);
527 if (test_and_clear_bit(segno
, dirty_i
->dirty_segmap
[dirty_type
]))
528 dirty_i
->nr_dirty
[dirty_type
]--;
530 if (dirty_type
== DIRTY
) {
531 struct seg_entry
*sentry
= get_seg_entry(sbi
, segno
);
532 enum dirty_type t
= sentry
->type
;
534 if (test_and_clear_bit(segno
, dirty_i
->dirty_segmap
[t
]))
535 dirty_i
->nr_dirty
[t
]--;
537 if (get_valid_blocks(sbi
, segno
, sbi
->segs_per_sec
) == 0)
538 clear_bit(GET_SECNO(sbi
, segno
),
539 dirty_i
->victim_secmap
);
544 * Should not occur error such as -ENOMEM.
545 * Adding dirty entry into seglist is not critical operation.
546 * If a given segment is one of current working segments, it won't be added.
548 static void locate_dirty_segment(struct f2fs_sb_info
*sbi
, unsigned int segno
)
550 struct dirty_seglist_info
*dirty_i
= DIRTY_I(sbi
);
551 unsigned short valid_blocks
;
553 if (segno
== NULL_SEGNO
|| IS_CURSEG(sbi
, segno
))
556 mutex_lock(&dirty_i
->seglist_lock
);
558 valid_blocks
= get_valid_blocks(sbi
, segno
, 0);
560 if (valid_blocks
== 0) {
561 __locate_dirty_segment(sbi
, segno
, PRE
);
562 __remove_dirty_segment(sbi
, segno
, DIRTY
);
563 } else if (valid_blocks
< sbi
->blocks_per_seg
) {
564 __locate_dirty_segment(sbi
, segno
, DIRTY
);
566 /* Recovery routine with SSR needs this */
567 __remove_dirty_segment(sbi
, segno
, DIRTY
);
570 mutex_unlock(&dirty_i
->seglist_lock
);
573 static int f2fs_issue_discard(struct f2fs_sb_info
*sbi
,
574 block_t blkstart
, block_t blklen
)
576 sector_t start
= SECTOR_FROM_BLOCK(blkstart
);
577 sector_t len
= SECTOR_FROM_BLOCK(blklen
);
578 struct seg_entry
*se
;
582 for (i
= blkstart
; i
< blkstart
+ blklen
; i
++) {
583 se
= get_seg_entry(sbi
, GET_SEGNO(sbi
, i
));
584 offset
= GET_BLKOFF_FROM_SEG0(sbi
, i
);
586 if (!f2fs_test_and_set_bit(offset
, se
->discard_map
))
589 trace_f2fs_issue_discard(sbi
->sb
, blkstart
, blklen
);
590 return blkdev_issue_discard(sbi
->sb
->s_bdev
, start
, len
, GFP_NOFS
, 0);
593 bool discard_next_dnode(struct f2fs_sb_info
*sbi
, block_t blkaddr
)
595 int err
= -EOPNOTSUPP
;
597 if (test_opt(sbi
, DISCARD
)) {
598 struct seg_entry
*se
= get_seg_entry(sbi
,
599 GET_SEGNO(sbi
, blkaddr
));
600 unsigned int offset
= GET_BLKOFF_FROM_SEG0(sbi
, blkaddr
);
602 if (f2fs_test_bit(offset
, se
->discard_map
))
605 err
= f2fs_issue_discard(sbi
, blkaddr
, 1);
609 update_meta_page(sbi
, NULL
, blkaddr
);
615 static void __add_discard_entry(struct f2fs_sb_info
*sbi
,
616 struct cp_control
*cpc
, struct seg_entry
*se
,
617 unsigned int start
, unsigned int end
)
619 struct list_head
*head
= &SM_I(sbi
)->discard_list
;
620 struct discard_entry
*new, *last
;
622 if (!list_empty(head
)) {
623 last
= list_last_entry(head
, struct discard_entry
, list
);
624 if (START_BLOCK(sbi
, cpc
->trim_start
) + start
==
625 last
->blkaddr
+ last
->len
) {
626 last
->len
+= end
- start
;
631 new = f2fs_kmem_cache_alloc(discard_entry_slab
, GFP_NOFS
);
632 INIT_LIST_HEAD(&new->list
);
633 new->blkaddr
= START_BLOCK(sbi
, cpc
->trim_start
) + start
;
634 new->len
= end
- start
;
635 list_add_tail(&new->list
, head
);
637 SM_I(sbi
)->nr_discards
+= end
- start
;
640 static void add_discard_addrs(struct f2fs_sb_info
*sbi
, struct cp_control
*cpc
)
642 int entries
= SIT_VBLOCK_MAP_SIZE
/ sizeof(unsigned long);
643 int max_blocks
= sbi
->blocks_per_seg
;
644 struct seg_entry
*se
= get_seg_entry(sbi
, cpc
->trim_start
);
645 unsigned long *cur_map
= (unsigned long *)se
->cur_valid_map
;
646 unsigned long *ckpt_map
= (unsigned long *)se
->ckpt_valid_map
;
647 unsigned long *discard_map
= (unsigned long *)se
->discard_map
;
648 unsigned long *dmap
= SIT_I(sbi
)->tmp_map
;
649 unsigned int start
= 0, end
= -1;
650 bool force
= (cpc
->reason
== CP_DISCARD
);
653 if (se
->valid_blocks
== max_blocks
)
657 if (!test_opt(sbi
, DISCARD
) || !se
->valid_blocks
||
658 SM_I(sbi
)->nr_discards
>= SM_I(sbi
)->max_discards
)
662 /* SIT_VBLOCK_MAP_SIZE should be multiple of sizeof(unsigned long) */
663 for (i
= 0; i
< entries
; i
++)
664 dmap
[i
] = force
? ~ckpt_map
[i
] & ~discard_map
[i
] :
665 (cur_map
[i
] ^ ckpt_map
[i
]) & ckpt_map
[i
];
667 while (force
|| SM_I(sbi
)->nr_discards
<= SM_I(sbi
)->max_discards
) {
668 start
= __find_rev_next_bit(dmap
, max_blocks
, end
+ 1);
669 if (start
>= max_blocks
)
672 end
= __find_rev_next_zero_bit(dmap
, max_blocks
, start
+ 1);
673 __add_discard_entry(sbi
, cpc
, se
, start
, end
);
677 void release_discard_addrs(struct f2fs_sb_info
*sbi
)
679 struct list_head
*head
= &(SM_I(sbi
)->discard_list
);
680 struct discard_entry
*entry
, *this;
683 list_for_each_entry_safe(entry
, this, head
, list
) {
684 list_del(&entry
->list
);
685 kmem_cache_free(discard_entry_slab
, entry
);
690 * Should call clear_prefree_segments after checkpoint is done.
692 static void set_prefree_as_free_segments(struct f2fs_sb_info
*sbi
)
694 struct dirty_seglist_info
*dirty_i
= DIRTY_I(sbi
);
697 mutex_lock(&dirty_i
->seglist_lock
);
698 for_each_set_bit(segno
, dirty_i
->dirty_segmap
[PRE
], MAIN_SEGS(sbi
))
699 __set_test_and_free(sbi
, segno
);
700 mutex_unlock(&dirty_i
->seglist_lock
);
703 void clear_prefree_segments(struct f2fs_sb_info
*sbi
, struct cp_control
*cpc
)
705 struct list_head
*head
= &(SM_I(sbi
)->discard_list
);
706 struct discard_entry
*entry
, *this;
707 struct dirty_seglist_info
*dirty_i
= DIRTY_I(sbi
);
708 unsigned long *prefree_map
= dirty_i
->dirty_segmap
[PRE
];
709 unsigned int start
= 0, end
= -1;
710 unsigned int secno
, start_segno
;
712 mutex_lock(&dirty_i
->seglist_lock
);
716 start
= find_next_bit(prefree_map
, MAIN_SEGS(sbi
), end
+ 1);
717 if (start
>= MAIN_SEGS(sbi
))
719 end
= find_next_zero_bit(prefree_map
, MAIN_SEGS(sbi
),
722 for (i
= start
; i
< end
; i
++)
723 clear_bit(i
, prefree_map
);
725 dirty_i
->nr_dirty
[PRE
] -= end
- start
;
727 if (!test_opt(sbi
, DISCARD
))
730 if (!test_opt(sbi
, LFS
) || sbi
->segs_per_sec
== 1) {
731 f2fs_issue_discard(sbi
, START_BLOCK(sbi
, start
),
732 (end
- start
) << sbi
->log_blocks_per_seg
);
736 secno
= GET_SECNO(sbi
, start
);
737 start_segno
= secno
* sbi
->segs_per_sec
;
738 if (!IS_CURSEC(sbi
, secno
) &&
739 !get_valid_blocks(sbi
, start
, sbi
->segs_per_sec
))
740 f2fs_issue_discard(sbi
, START_BLOCK(sbi
, start_segno
),
741 sbi
->segs_per_sec
<< sbi
->log_blocks_per_seg
);
743 start
= start_segno
+ sbi
->segs_per_sec
;
747 mutex_unlock(&dirty_i
->seglist_lock
);
749 /* send small discards */
750 list_for_each_entry_safe(entry
, this, head
, list
) {
751 if (cpc
->reason
== CP_DISCARD
&& entry
->len
< cpc
->trim_minlen
)
753 f2fs_issue_discard(sbi
, entry
->blkaddr
, entry
->len
);
754 cpc
->trimmed
+= entry
->len
;
756 list_del(&entry
->list
);
757 SM_I(sbi
)->nr_discards
-= entry
->len
;
758 kmem_cache_free(discard_entry_slab
, entry
);
762 static bool __mark_sit_entry_dirty(struct f2fs_sb_info
*sbi
, unsigned int segno
)
764 struct sit_info
*sit_i
= SIT_I(sbi
);
766 if (!__test_and_set_bit(segno
, sit_i
->dirty_sentries_bitmap
)) {
767 sit_i
->dirty_sentries
++;
774 static void __set_sit_entry_type(struct f2fs_sb_info
*sbi
, int type
,
775 unsigned int segno
, int modified
)
777 struct seg_entry
*se
= get_seg_entry(sbi
, segno
);
780 __mark_sit_entry_dirty(sbi
, segno
);
783 static void update_sit_entry(struct f2fs_sb_info
*sbi
, block_t blkaddr
, int del
)
785 struct seg_entry
*se
;
786 unsigned int segno
, offset
;
787 long int new_vblocks
;
789 segno
= GET_SEGNO(sbi
, blkaddr
);
791 se
= get_seg_entry(sbi
, segno
);
792 new_vblocks
= se
->valid_blocks
+ del
;
793 offset
= GET_BLKOFF_FROM_SEG0(sbi
, blkaddr
);
795 f2fs_bug_on(sbi
, (new_vblocks
>> (sizeof(unsigned short) << 3) ||
796 (new_vblocks
> sbi
->blocks_per_seg
)));
798 se
->valid_blocks
= new_vblocks
;
799 se
->mtime
= get_mtime(sbi
);
800 SIT_I(sbi
)->max_mtime
= se
->mtime
;
802 /* Update valid block bitmap */
804 if (f2fs_test_and_set_bit(offset
, se
->cur_valid_map
))
806 if (!f2fs_test_and_set_bit(offset
, se
->discard_map
))
809 if (!f2fs_test_and_clear_bit(offset
, se
->cur_valid_map
))
811 if (f2fs_test_and_clear_bit(offset
, se
->discard_map
))
814 if (!f2fs_test_bit(offset
, se
->ckpt_valid_map
))
815 se
->ckpt_valid_blocks
+= del
;
817 __mark_sit_entry_dirty(sbi
, segno
);
819 /* update total number of valid blocks to be written in ckpt area */
820 SIT_I(sbi
)->written_valid_blocks
+= del
;
822 if (sbi
->segs_per_sec
> 1)
823 get_sec_entry(sbi
, segno
)->valid_blocks
+= del
;
826 void refresh_sit_entry(struct f2fs_sb_info
*sbi
, block_t old
, block_t
new)
828 update_sit_entry(sbi
, new, 1);
829 if (GET_SEGNO(sbi
, old
) != NULL_SEGNO
)
830 update_sit_entry(sbi
, old
, -1);
832 locate_dirty_segment(sbi
, GET_SEGNO(sbi
, old
));
833 locate_dirty_segment(sbi
, GET_SEGNO(sbi
, new));
836 void invalidate_blocks(struct f2fs_sb_info
*sbi
, block_t addr
)
838 unsigned int segno
= GET_SEGNO(sbi
, addr
);
839 struct sit_info
*sit_i
= SIT_I(sbi
);
841 f2fs_bug_on(sbi
, addr
== NULL_ADDR
);
842 if (addr
== NEW_ADDR
)
845 /* add it into sit main buffer */
846 mutex_lock(&sit_i
->sentry_lock
);
848 update_sit_entry(sbi
, addr
, -1);
850 /* add it into dirty seglist */
851 locate_dirty_segment(sbi
, segno
);
853 mutex_unlock(&sit_i
->sentry_lock
);
856 bool is_checkpointed_data(struct f2fs_sb_info
*sbi
, block_t blkaddr
)
858 struct sit_info
*sit_i
= SIT_I(sbi
);
859 unsigned int segno
, offset
;
860 struct seg_entry
*se
;
863 if (blkaddr
== NEW_ADDR
|| blkaddr
== NULL_ADDR
)
866 mutex_lock(&sit_i
->sentry_lock
);
868 segno
= GET_SEGNO(sbi
, blkaddr
);
869 se
= get_seg_entry(sbi
, segno
);
870 offset
= GET_BLKOFF_FROM_SEG0(sbi
, blkaddr
);
872 if (f2fs_test_bit(offset
, se
->ckpt_valid_map
))
875 mutex_unlock(&sit_i
->sentry_lock
);
881 * This function should be resided under the curseg_mutex lock
883 static void __add_sum_entry(struct f2fs_sb_info
*sbi
, int type
,
884 struct f2fs_summary
*sum
)
886 struct curseg_info
*curseg
= CURSEG_I(sbi
, type
);
887 void *addr
= curseg
->sum_blk
;
888 addr
+= curseg
->next_blkoff
* sizeof(struct f2fs_summary
);
889 memcpy(addr
, sum
, sizeof(struct f2fs_summary
));
893 * Calculate the number of current summary pages for writing
895 int npages_for_summary_flush(struct f2fs_sb_info
*sbi
, bool for_ra
)
897 int valid_sum_count
= 0;
900 for (i
= CURSEG_HOT_DATA
; i
<= CURSEG_COLD_DATA
; i
++) {
901 if (sbi
->ckpt
->alloc_type
[i
] == SSR
)
902 valid_sum_count
+= sbi
->blocks_per_seg
;
905 valid_sum_count
+= le16_to_cpu(
906 F2FS_CKPT(sbi
)->cur_data_blkoff
[i
]);
908 valid_sum_count
+= curseg_blkoff(sbi
, i
);
912 sum_in_page
= (PAGE_SIZE
- 2 * SUM_JOURNAL_SIZE
-
913 SUM_FOOTER_SIZE
) / SUMMARY_SIZE
;
914 if (valid_sum_count
<= sum_in_page
)
916 else if ((valid_sum_count
- sum_in_page
) <=
917 (PAGE_SIZE
- SUM_FOOTER_SIZE
) / SUMMARY_SIZE
)
923 * Caller should put this summary page
925 struct page
*get_sum_page(struct f2fs_sb_info
*sbi
, unsigned int segno
)
927 return get_meta_page(sbi
, GET_SUM_BLOCK(sbi
, segno
));
930 void update_meta_page(struct f2fs_sb_info
*sbi
, void *src
, block_t blk_addr
)
932 struct page
*page
= grab_meta_page(sbi
, blk_addr
);
933 void *dst
= page_address(page
);
936 memcpy(dst
, src
, PAGE_SIZE
);
938 memset(dst
, 0, PAGE_SIZE
);
939 set_page_dirty(page
);
940 f2fs_put_page(page
, 1);
943 static void write_sum_page(struct f2fs_sb_info
*sbi
,
944 struct f2fs_summary_block
*sum_blk
, block_t blk_addr
)
946 update_meta_page(sbi
, (void *)sum_blk
, blk_addr
);
949 static void write_current_sum_page(struct f2fs_sb_info
*sbi
,
950 int type
, block_t blk_addr
)
952 struct curseg_info
*curseg
= CURSEG_I(sbi
, type
);
953 struct page
*page
= grab_meta_page(sbi
, blk_addr
);
954 struct f2fs_summary_block
*src
= curseg
->sum_blk
;
955 struct f2fs_summary_block
*dst
;
957 dst
= (struct f2fs_summary_block
*)page_address(page
);
959 mutex_lock(&curseg
->curseg_mutex
);
961 down_read(&curseg
->journal_rwsem
);
962 memcpy(&dst
->journal
, curseg
->journal
, SUM_JOURNAL_SIZE
);
963 up_read(&curseg
->journal_rwsem
);
965 memcpy(dst
->entries
, src
->entries
, SUM_ENTRY_SIZE
);
966 memcpy(&dst
->footer
, &src
->footer
, SUM_FOOTER_SIZE
);
968 mutex_unlock(&curseg
->curseg_mutex
);
970 set_page_dirty(page
);
971 f2fs_put_page(page
, 1);
974 static int is_next_segment_free(struct f2fs_sb_info
*sbi
, int type
)
976 struct curseg_info
*curseg
= CURSEG_I(sbi
, type
);
977 unsigned int segno
= curseg
->segno
+ 1;
978 struct free_segmap_info
*free_i
= FREE_I(sbi
);
980 if (segno
< MAIN_SEGS(sbi
) && segno
% sbi
->segs_per_sec
)
981 return !test_bit(segno
, free_i
->free_segmap
);
986 * Find a new segment from the free segments bitmap to right order
987 * This function should be returned with success, otherwise BUG
989 static void get_new_segment(struct f2fs_sb_info
*sbi
,
990 unsigned int *newseg
, bool new_sec
, int dir
)
992 struct free_segmap_info
*free_i
= FREE_I(sbi
);
993 unsigned int segno
, secno
, zoneno
;
994 unsigned int total_zones
= MAIN_SECS(sbi
) / sbi
->secs_per_zone
;
995 unsigned int hint
= *newseg
/ sbi
->segs_per_sec
;
996 unsigned int old_zoneno
= GET_ZONENO_FROM_SEGNO(sbi
, *newseg
);
997 unsigned int left_start
= hint
;
1002 spin_lock(&free_i
->segmap_lock
);
1004 if (!new_sec
&& ((*newseg
+ 1) % sbi
->segs_per_sec
)) {
1005 segno
= find_next_zero_bit(free_i
->free_segmap
,
1006 (hint
+ 1) * sbi
->segs_per_sec
, *newseg
+ 1);
1007 if (segno
< (hint
+ 1) * sbi
->segs_per_sec
)
1011 secno
= find_next_zero_bit(free_i
->free_secmap
, MAIN_SECS(sbi
), hint
);
1012 if (secno
>= MAIN_SECS(sbi
)) {
1013 if (dir
== ALLOC_RIGHT
) {
1014 secno
= find_next_zero_bit(free_i
->free_secmap
,
1016 f2fs_bug_on(sbi
, secno
>= MAIN_SECS(sbi
));
1019 left_start
= hint
- 1;
1025 while (test_bit(left_start
, free_i
->free_secmap
)) {
1026 if (left_start
> 0) {
1030 left_start
= find_next_zero_bit(free_i
->free_secmap
,
1032 f2fs_bug_on(sbi
, left_start
>= MAIN_SECS(sbi
));
1038 segno
= secno
* sbi
->segs_per_sec
;
1039 zoneno
= secno
/ sbi
->secs_per_zone
;
1041 /* give up on finding another zone */
1044 if (sbi
->secs_per_zone
== 1)
1046 if (zoneno
== old_zoneno
)
1048 if (dir
== ALLOC_LEFT
) {
1049 if (!go_left
&& zoneno
+ 1 >= total_zones
)
1051 if (go_left
&& zoneno
== 0)
1054 for (i
= 0; i
< NR_CURSEG_TYPE
; i
++)
1055 if (CURSEG_I(sbi
, i
)->zone
== zoneno
)
1058 if (i
< NR_CURSEG_TYPE
) {
1059 /* zone is in user, try another */
1061 hint
= zoneno
* sbi
->secs_per_zone
- 1;
1062 else if (zoneno
+ 1 >= total_zones
)
1065 hint
= (zoneno
+ 1) * sbi
->secs_per_zone
;
1067 goto find_other_zone
;
1070 /* set it as dirty segment in free segmap */
1071 f2fs_bug_on(sbi
, test_bit(segno
, free_i
->free_segmap
));
1072 __set_inuse(sbi
, segno
);
1074 spin_unlock(&free_i
->segmap_lock
);
1077 static void reset_curseg(struct f2fs_sb_info
*sbi
, int type
, int modified
)
1079 struct curseg_info
*curseg
= CURSEG_I(sbi
, type
);
1080 struct summary_footer
*sum_footer
;
1082 curseg
->segno
= curseg
->next_segno
;
1083 curseg
->zone
= GET_ZONENO_FROM_SEGNO(sbi
, curseg
->segno
);
1084 curseg
->next_blkoff
= 0;
1085 curseg
->next_segno
= NULL_SEGNO
;
1087 sum_footer
= &(curseg
->sum_blk
->footer
);
1088 memset(sum_footer
, 0, sizeof(struct summary_footer
));
1089 if (IS_DATASEG(type
))
1090 SET_SUM_TYPE(sum_footer
, SUM_TYPE_DATA
);
1091 if (IS_NODESEG(type
))
1092 SET_SUM_TYPE(sum_footer
, SUM_TYPE_NODE
);
1093 __set_sit_entry_type(sbi
, type
, curseg
->segno
, modified
);
1097 * Allocate a current working segment.
1098 * This function always allocates a free segment in LFS manner.
1100 static void new_curseg(struct f2fs_sb_info
*sbi
, int type
, bool new_sec
)
1102 struct curseg_info
*curseg
= CURSEG_I(sbi
, type
);
1103 unsigned int segno
= curseg
->segno
;
1104 int dir
= ALLOC_LEFT
;
1106 write_sum_page(sbi
, curseg
->sum_blk
,
1107 GET_SUM_BLOCK(sbi
, segno
));
1108 if (type
== CURSEG_WARM_DATA
|| type
== CURSEG_COLD_DATA
)
1111 if (test_opt(sbi
, NOHEAP
))
1114 get_new_segment(sbi
, &segno
, new_sec
, dir
);
1115 curseg
->next_segno
= segno
;
1116 reset_curseg(sbi
, type
, 1);
1117 curseg
->alloc_type
= LFS
;
1120 static void __next_free_blkoff(struct f2fs_sb_info
*sbi
,
1121 struct curseg_info
*seg
, block_t start
)
1123 struct seg_entry
*se
= get_seg_entry(sbi
, seg
->segno
);
1124 int entries
= SIT_VBLOCK_MAP_SIZE
/ sizeof(unsigned long);
1125 unsigned long *target_map
= SIT_I(sbi
)->tmp_map
;
1126 unsigned long *ckpt_map
= (unsigned long *)se
->ckpt_valid_map
;
1127 unsigned long *cur_map
= (unsigned long *)se
->cur_valid_map
;
1130 for (i
= 0; i
< entries
; i
++)
1131 target_map
[i
] = ckpt_map
[i
] | cur_map
[i
];
1133 pos
= __find_rev_next_zero_bit(target_map
, sbi
->blocks_per_seg
, start
);
1135 seg
->next_blkoff
= pos
;
1139 * If a segment is written by LFS manner, next block offset is just obtained
1140 * by increasing the current block offset. However, if a segment is written by
1141 * SSR manner, next block offset obtained by calling __next_free_blkoff
1143 static void __refresh_next_blkoff(struct f2fs_sb_info
*sbi
,
1144 struct curseg_info
*seg
)
1146 if (seg
->alloc_type
== SSR
)
1147 __next_free_blkoff(sbi
, seg
, seg
->next_blkoff
+ 1);
1153 * This function always allocates a used segment(from dirty seglist) by SSR
1154 * manner, so it should recover the existing segment information of valid blocks
1156 static void change_curseg(struct f2fs_sb_info
*sbi
, int type
, bool reuse
)
1158 struct dirty_seglist_info
*dirty_i
= DIRTY_I(sbi
);
1159 struct curseg_info
*curseg
= CURSEG_I(sbi
, type
);
1160 unsigned int new_segno
= curseg
->next_segno
;
1161 struct f2fs_summary_block
*sum_node
;
1162 struct page
*sum_page
;
1164 write_sum_page(sbi
, curseg
->sum_blk
,
1165 GET_SUM_BLOCK(sbi
, curseg
->segno
));
1166 __set_test_and_inuse(sbi
, new_segno
);
1168 mutex_lock(&dirty_i
->seglist_lock
);
1169 __remove_dirty_segment(sbi
, new_segno
, PRE
);
1170 __remove_dirty_segment(sbi
, new_segno
, DIRTY
);
1171 mutex_unlock(&dirty_i
->seglist_lock
);
1173 reset_curseg(sbi
, type
, 1);
1174 curseg
->alloc_type
= SSR
;
1175 __next_free_blkoff(sbi
, curseg
, 0);
1178 sum_page
= get_sum_page(sbi
, new_segno
);
1179 sum_node
= (struct f2fs_summary_block
*)page_address(sum_page
);
1180 memcpy(curseg
->sum_blk
, sum_node
, SUM_ENTRY_SIZE
);
1181 f2fs_put_page(sum_page
, 1);
1185 static int get_ssr_segment(struct f2fs_sb_info
*sbi
, int type
)
1187 struct curseg_info
*curseg
= CURSEG_I(sbi
, type
);
1188 const struct victim_selection
*v_ops
= DIRTY_I(sbi
)->v_ops
;
1190 if (IS_NODESEG(type
) || !has_not_enough_free_secs(sbi
, 0))
1191 return v_ops
->get_victim(sbi
,
1192 &(curseg
)->next_segno
, BG_GC
, type
, SSR
);
1194 /* For data segments, let's do SSR more intensively */
1195 for (; type
>= CURSEG_HOT_DATA
; type
--)
1196 if (v_ops
->get_victim(sbi
, &(curseg
)->next_segno
,
1203 * flush out current segment and replace it with new segment
1204 * This function should be returned with success, otherwise BUG
1206 static void allocate_segment_by_default(struct f2fs_sb_info
*sbi
,
1207 int type
, bool force
)
1209 struct curseg_info
*curseg
= CURSEG_I(sbi
, type
);
1212 new_curseg(sbi
, type
, true);
1213 else if (type
== CURSEG_WARM_NODE
)
1214 new_curseg(sbi
, type
, false);
1215 else if (curseg
->alloc_type
== LFS
&& is_next_segment_free(sbi
, type
))
1216 new_curseg(sbi
, type
, false);
1217 else if (need_SSR(sbi
) && get_ssr_segment(sbi
, type
))
1218 change_curseg(sbi
, type
, true);
1220 new_curseg(sbi
, type
, false);
1222 stat_inc_seg_type(sbi
, curseg
);
1225 static void __allocate_new_segments(struct f2fs_sb_info
*sbi
, int type
)
1227 struct curseg_info
*curseg
= CURSEG_I(sbi
, type
);
1228 unsigned int old_segno
;
1230 old_segno
= curseg
->segno
;
1231 SIT_I(sbi
)->s_ops
->allocate_segment(sbi
, type
, true);
1232 locate_dirty_segment(sbi
, old_segno
);
1235 void allocate_new_segments(struct f2fs_sb_info
*sbi
)
1239 if (test_opt(sbi
, LFS
))
1242 for (i
= CURSEG_HOT_DATA
; i
<= CURSEG_COLD_DATA
; i
++)
1243 __allocate_new_segments(sbi
, i
);
1246 static const struct segment_allocation default_salloc_ops
= {
1247 .allocate_segment
= allocate_segment_by_default
,
1250 int f2fs_trim_fs(struct f2fs_sb_info
*sbi
, struct fstrim_range
*range
)
1252 __u64 start
= F2FS_BYTES_TO_BLK(range
->start
);
1253 __u64 end
= start
+ F2FS_BYTES_TO_BLK(range
->len
) - 1;
1254 unsigned int start_segno
, end_segno
;
1255 struct cp_control cpc
;
1258 if (start
>= MAX_BLKADDR(sbi
) || range
->len
< sbi
->blocksize
)
1262 if (end
<= MAIN_BLKADDR(sbi
))
1265 /* start/end segment number in main_area */
1266 start_segno
= (start
<= MAIN_BLKADDR(sbi
)) ? 0 : GET_SEGNO(sbi
, start
);
1267 end_segno
= (end
>= MAX_BLKADDR(sbi
)) ? MAIN_SEGS(sbi
) - 1 :
1268 GET_SEGNO(sbi
, end
);
1269 cpc
.reason
= CP_DISCARD
;
1270 cpc
.trim_minlen
= max_t(__u64
, 1, F2FS_BYTES_TO_BLK(range
->minlen
));
1272 /* do checkpoint to issue discard commands safely */
1273 for (; start_segno
<= end_segno
; start_segno
= cpc
.trim_end
+ 1) {
1274 cpc
.trim_start
= start_segno
;
1276 if (sbi
->discard_blks
== 0)
1278 else if (sbi
->discard_blks
< BATCHED_TRIM_BLOCKS(sbi
))
1279 cpc
.trim_end
= end_segno
;
1281 cpc
.trim_end
= min_t(unsigned int,
1282 rounddown(start_segno
+
1283 BATCHED_TRIM_SEGMENTS(sbi
),
1284 sbi
->segs_per_sec
) - 1, end_segno
);
1286 mutex_lock(&sbi
->gc_mutex
);
1287 err
= write_checkpoint(sbi
, &cpc
);
1288 mutex_unlock(&sbi
->gc_mutex
);
1291 range
->len
= F2FS_BLK_TO_BYTES(cpc
.trimmed
);
1295 static bool __has_curseg_space(struct f2fs_sb_info
*sbi
, int type
)
1297 struct curseg_info
*curseg
= CURSEG_I(sbi
, type
);
1298 if (curseg
->next_blkoff
< sbi
->blocks_per_seg
)
1303 static int __get_segment_type_2(struct page
*page
, enum page_type p_type
)
1306 return CURSEG_HOT_DATA
;
1308 return CURSEG_HOT_NODE
;
1311 static int __get_segment_type_4(struct page
*page
, enum page_type p_type
)
1313 if (p_type
== DATA
) {
1314 struct inode
*inode
= page
->mapping
->host
;
1316 if (S_ISDIR(inode
->i_mode
))
1317 return CURSEG_HOT_DATA
;
1319 return CURSEG_COLD_DATA
;
1321 if (IS_DNODE(page
) && is_cold_node(page
))
1322 return CURSEG_WARM_NODE
;
1324 return CURSEG_COLD_NODE
;
1328 static int __get_segment_type_6(struct page
*page
, enum page_type p_type
)
1330 if (p_type
== DATA
) {
1331 struct inode
*inode
= page
->mapping
->host
;
1333 if (S_ISDIR(inode
->i_mode
))
1334 return CURSEG_HOT_DATA
;
1335 else if (is_cold_data(page
) || file_is_cold(inode
))
1336 return CURSEG_COLD_DATA
;
1338 return CURSEG_WARM_DATA
;
1341 return is_cold_node(page
) ? CURSEG_WARM_NODE
:
1344 return CURSEG_COLD_NODE
;
1348 static int __get_segment_type(struct page
*page
, enum page_type p_type
)
1350 switch (F2FS_P_SB(page
)->active_logs
) {
1352 return __get_segment_type_2(page
, p_type
);
1354 return __get_segment_type_4(page
, p_type
);
1356 /* NR_CURSEG_TYPE(6) logs by default */
1357 f2fs_bug_on(F2FS_P_SB(page
),
1358 F2FS_P_SB(page
)->active_logs
!= NR_CURSEG_TYPE
);
1359 return __get_segment_type_6(page
, p_type
);
1362 void allocate_data_block(struct f2fs_sb_info
*sbi
, struct page
*page
,
1363 block_t old_blkaddr
, block_t
*new_blkaddr
,
1364 struct f2fs_summary
*sum
, int type
)
1366 struct sit_info
*sit_i
= SIT_I(sbi
);
1367 struct curseg_info
*curseg
;
1368 bool direct_io
= (type
== CURSEG_DIRECT_IO
);
1370 type
= direct_io
? CURSEG_WARM_DATA
: type
;
1372 curseg
= CURSEG_I(sbi
, type
);
1374 mutex_lock(&curseg
->curseg_mutex
);
1375 mutex_lock(&sit_i
->sentry_lock
);
1377 /* direct_io'ed data is aligned to the segment for better performance */
1378 if (direct_io
&& curseg
->next_blkoff
&&
1379 !has_not_enough_free_secs(sbi
, 0))
1380 __allocate_new_segments(sbi
, type
);
1382 *new_blkaddr
= NEXT_FREE_BLKADDR(sbi
, curseg
);
1385 * __add_sum_entry should be resided under the curseg_mutex
1386 * because, this function updates a summary entry in the
1387 * current summary block.
1389 __add_sum_entry(sbi
, type
, sum
);
1391 __refresh_next_blkoff(sbi
, curseg
);
1393 stat_inc_block_count(sbi
, curseg
);
1395 if (!__has_curseg_space(sbi
, type
))
1396 sit_i
->s_ops
->allocate_segment(sbi
, type
, false);
1398 * SIT information should be updated before segment allocation,
1399 * since SSR needs latest valid block information.
1401 refresh_sit_entry(sbi
, old_blkaddr
, *new_blkaddr
);
1403 mutex_unlock(&sit_i
->sentry_lock
);
1405 if (page
&& IS_NODESEG(type
))
1406 fill_node_footer_blkaddr(page
, NEXT_FREE_BLKADDR(sbi
, curseg
));
1408 mutex_unlock(&curseg
->curseg_mutex
);
1411 static void do_write_page(struct f2fs_summary
*sum
, struct f2fs_io_info
*fio
)
1413 int type
= __get_segment_type(fio
->page
, fio
->type
);
1415 if (fio
->type
== NODE
|| fio
->type
== DATA
)
1416 mutex_lock(&fio
->sbi
->wio_mutex
[fio
->type
]);
1418 allocate_data_block(fio
->sbi
, fio
->page
, fio
->old_blkaddr
,
1419 &fio
->new_blkaddr
, sum
, type
);
1421 /* writeout dirty page into bdev */
1422 f2fs_submit_page_mbio(fio
);
1424 if (fio
->type
== NODE
|| fio
->type
== DATA
)
1425 mutex_unlock(&fio
->sbi
->wio_mutex
[fio
->type
]);
1428 void write_meta_page(struct f2fs_sb_info
*sbi
, struct page
*page
)
1430 struct f2fs_io_info fio
= {
1433 .rw
= WRITE_SYNC
| REQ_META
| REQ_PRIO
,
1434 .old_blkaddr
= page
->index
,
1435 .new_blkaddr
= page
->index
,
1437 .encrypted_page
= NULL
,
1440 if (unlikely(page
->index
>= MAIN_BLKADDR(sbi
)))
1441 fio
.rw
&= ~REQ_META
;
1443 set_page_writeback(page
);
1444 f2fs_submit_page_mbio(&fio
);
1447 void write_node_page(unsigned int nid
, struct f2fs_io_info
*fio
)
1449 struct f2fs_summary sum
;
1451 set_summary(&sum
, nid
, 0, 0);
1452 do_write_page(&sum
, fio
);
1455 void write_data_page(struct dnode_of_data
*dn
, struct f2fs_io_info
*fio
)
1457 struct f2fs_sb_info
*sbi
= fio
->sbi
;
1458 struct f2fs_summary sum
;
1459 struct node_info ni
;
1461 f2fs_bug_on(sbi
, dn
->data_blkaddr
== NULL_ADDR
);
1462 get_node_info(sbi
, dn
->nid
, &ni
);
1463 set_summary(&sum
, dn
->nid
, dn
->ofs_in_node
, ni
.version
);
1464 do_write_page(&sum
, fio
);
1465 f2fs_update_data_blkaddr(dn
, fio
->new_blkaddr
);
1468 void rewrite_data_page(struct f2fs_io_info
*fio
)
1470 fio
->new_blkaddr
= fio
->old_blkaddr
;
1471 stat_inc_inplace_blocks(fio
->sbi
);
1472 f2fs_submit_page_mbio(fio
);
1475 void __f2fs_replace_block(struct f2fs_sb_info
*sbi
, struct f2fs_summary
*sum
,
1476 block_t old_blkaddr
, block_t new_blkaddr
,
1477 bool recover_curseg
, bool recover_newaddr
)
1479 struct sit_info
*sit_i
= SIT_I(sbi
);
1480 struct curseg_info
*curseg
;
1481 unsigned int segno
, old_cursegno
;
1482 struct seg_entry
*se
;
1484 unsigned short old_blkoff
;
1486 segno
= GET_SEGNO(sbi
, new_blkaddr
);
1487 se
= get_seg_entry(sbi
, segno
);
1490 if (!recover_curseg
) {
1491 /* for recovery flow */
1492 if (se
->valid_blocks
== 0 && !IS_CURSEG(sbi
, segno
)) {
1493 if (old_blkaddr
== NULL_ADDR
)
1494 type
= CURSEG_COLD_DATA
;
1496 type
= CURSEG_WARM_DATA
;
1499 if (!IS_CURSEG(sbi
, segno
))
1500 type
= CURSEG_WARM_DATA
;
1503 curseg
= CURSEG_I(sbi
, type
);
1505 mutex_lock(&curseg
->curseg_mutex
);
1506 mutex_lock(&sit_i
->sentry_lock
);
1508 old_cursegno
= curseg
->segno
;
1509 old_blkoff
= curseg
->next_blkoff
;
1511 /* change the current segment */
1512 if (segno
!= curseg
->segno
) {
1513 curseg
->next_segno
= segno
;
1514 change_curseg(sbi
, type
, true);
1517 curseg
->next_blkoff
= GET_BLKOFF_FROM_SEG0(sbi
, new_blkaddr
);
1518 __add_sum_entry(sbi
, type
, sum
);
1520 if (!recover_curseg
|| recover_newaddr
)
1521 update_sit_entry(sbi
, new_blkaddr
, 1);
1522 if (GET_SEGNO(sbi
, old_blkaddr
) != NULL_SEGNO
)
1523 update_sit_entry(sbi
, old_blkaddr
, -1);
1525 locate_dirty_segment(sbi
, GET_SEGNO(sbi
, old_blkaddr
));
1526 locate_dirty_segment(sbi
, GET_SEGNO(sbi
, new_blkaddr
));
1528 locate_dirty_segment(sbi
, old_cursegno
);
1530 if (recover_curseg
) {
1531 if (old_cursegno
!= curseg
->segno
) {
1532 curseg
->next_segno
= old_cursegno
;
1533 change_curseg(sbi
, type
, true);
1535 curseg
->next_blkoff
= old_blkoff
;
1538 mutex_unlock(&sit_i
->sentry_lock
);
1539 mutex_unlock(&curseg
->curseg_mutex
);
1542 void f2fs_replace_block(struct f2fs_sb_info
*sbi
, struct dnode_of_data
*dn
,
1543 block_t old_addr
, block_t new_addr
,
1544 unsigned char version
, bool recover_curseg
,
1545 bool recover_newaddr
)
1547 struct f2fs_summary sum
;
1549 set_summary(&sum
, dn
->nid
, dn
->ofs_in_node
, version
);
1551 __f2fs_replace_block(sbi
, &sum
, old_addr
, new_addr
,
1552 recover_curseg
, recover_newaddr
);
1554 f2fs_update_data_blkaddr(dn
, new_addr
);
1557 void f2fs_wait_on_page_writeback(struct page
*page
,
1558 enum page_type type
, bool ordered
)
1560 if (PageWriteback(page
)) {
1561 struct f2fs_sb_info
*sbi
= F2FS_P_SB(page
);
1563 f2fs_submit_merged_bio_cond(sbi
, NULL
, page
, 0, type
, WRITE
);
1565 wait_on_page_writeback(page
);
1567 wait_for_stable_page(page
);
1571 void f2fs_wait_on_encrypted_page_writeback(struct f2fs_sb_info
*sbi
,
1576 if (blkaddr
== NEW_ADDR
)
1579 f2fs_bug_on(sbi
, blkaddr
== NULL_ADDR
);
1581 cpage
= find_lock_page(META_MAPPING(sbi
), blkaddr
);
1583 f2fs_wait_on_page_writeback(cpage
, DATA
, true);
1584 f2fs_put_page(cpage
, 1);
1588 static int read_compacted_summaries(struct f2fs_sb_info
*sbi
)
1590 struct f2fs_checkpoint
*ckpt
= F2FS_CKPT(sbi
);
1591 struct curseg_info
*seg_i
;
1592 unsigned char *kaddr
;
1597 start
= start_sum_block(sbi
);
1599 page
= get_meta_page(sbi
, start
++);
1600 kaddr
= (unsigned char *)page_address(page
);
1602 /* Step 1: restore nat cache */
1603 seg_i
= CURSEG_I(sbi
, CURSEG_HOT_DATA
);
1604 memcpy(seg_i
->journal
, kaddr
, SUM_JOURNAL_SIZE
);
1606 /* Step 2: restore sit cache */
1607 seg_i
= CURSEG_I(sbi
, CURSEG_COLD_DATA
);
1608 memcpy(seg_i
->journal
, kaddr
+ SUM_JOURNAL_SIZE
, SUM_JOURNAL_SIZE
);
1609 offset
= 2 * SUM_JOURNAL_SIZE
;
1611 /* Step 3: restore summary entries */
1612 for (i
= CURSEG_HOT_DATA
; i
<= CURSEG_COLD_DATA
; i
++) {
1613 unsigned short blk_off
;
1616 seg_i
= CURSEG_I(sbi
, i
);
1617 segno
= le32_to_cpu(ckpt
->cur_data_segno
[i
]);
1618 blk_off
= le16_to_cpu(ckpt
->cur_data_blkoff
[i
]);
1619 seg_i
->next_segno
= segno
;
1620 reset_curseg(sbi
, i
, 0);
1621 seg_i
->alloc_type
= ckpt
->alloc_type
[i
];
1622 seg_i
->next_blkoff
= blk_off
;
1624 if (seg_i
->alloc_type
== SSR
)
1625 blk_off
= sbi
->blocks_per_seg
;
1627 for (j
= 0; j
< blk_off
; j
++) {
1628 struct f2fs_summary
*s
;
1629 s
= (struct f2fs_summary
*)(kaddr
+ offset
);
1630 seg_i
->sum_blk
->entries
[j
] = *s
;
1631 offset
+= SUMMARY_SIZE
;
1632 if (offset
+ SUMMARY_SIZE
<= PAGE_SIZE
-
1636 f2fs_put_page(page
, 1);
1639 page
= get_meta_page(sbi
, start
++);
1640 kaddr
= (unsigned char *)page_address(page
);
1644 f2fs_put_page(page
, 1);
1648 static int read_normal_summaries(struct f2fs_sb_info
*sbi
, int type
)
1650 struct f2fs_checkpoint
*ckpt
= F2FS_CKPT(sbi
);
1651 struct f2fs_summary_block
*sum
;
1652 struct curseg_info
*curseg
;
1654 unsigned short blk_off
;
1655 unsigned int segno
= 0;
1656 block_t blk_addr
= 0;
1658 /* get segment number and block addr */
1659 if (IS_DATASEG(type
)) {
1660 segno
= le32_to_cpu(ckpt
->cur_data_segno
[type
]);
1661 blk_off
= le16_to_cpu(ckpt
->cur_data_blkoff
[type
-
1663 if (__exist_node_summaries(sbi
))
1664 blk_addr
= sum_blk_addr(sbi
, NR_CURSEG_TYPE
, type
);
1666 blk_addr
= sum_blk_addr(sbi
, NR_CURSEG_DATA_TYPE
, type
);
1668 segno
= le32_to_cpu(ckpt
->cur_node_segno
[type
-
1670 blk_off
= le16_to_cpu(ckpt
->cur_node_blkoff
[type
-
1672 if (__exist_node_summaries(sbi
))
1673 blk_addr
= sum_blk_addr(sbi
, NR_CURSEG_NODE_TYPE
,
1674 type
- CURSEG_HOT_NODE
);
1676 blk_addr
= GET_SUM_BLOCK(sbi
, segno
);
1679 new = get_meta_page(sbi
, blk_addr
);
1680 sum
= (struct f2fs_summary_block
*)page_address(new);
1682 if (IS_NODESEG(type
)) {
1683 if (__exist_node_summaries(sbi
)) {
1684 struct f2fs_summary
*ns
= &sum
->entries
[0];
1686 for (i
= 0; i
< sbi
->blocks_per_seg
; i
++, ns
++) {
1688 ns
->ofs_in_node
= 0;
1693 err
= restore_node_summary(sbi
, segno
, sum
);
1695 f2fs_put_page(new, 1);
1701 /* set uncompleted segment to curseg */
1702 curseg
= CURSEG_I(sbi
, type
);
1703 mutex_lock(&curseg
->curseg_mutex
);
1705 /* update journal info */
1706 down_write(&curseg
->journal_rwsem
);
1707 memcpy(curseg
->journal
, &sum
->journal
, SUM_JOURNAL_SIZE
);
1708 up_write(&curseg
->journal_rwsem
);
1710 memcpy(curseg
->sum_blk
->entries
, sum
->entries
, SUM_ENTRY_SIZE
);
1711 memcpy(&curseg
->sum_blk
->footer
, &sum
->footer
, SUM_FOOTER_SIZE
);
1712 curseg
->next_segno
= segno
;
1713 reset_curseg(sbi
, type
, 0);
1714 curseg
->alloc_type
= ckpt
->alloc_type
[type
];
1715 curseg
->next_blkoff
= blk_off
;
1716 mutex_unlock(&curseg
->curseg_mutex
);
1717 f2fs_put_page(new, 1);
1721 static int restore_curseg_summaries(struct f2fs_sb_info
*sbi
)
1723 int type
= CURSEG_HOT_DATA
;
1726 if (is_set_ckpt_flags(F2FS_CKPT(sbi
), CP_COMPACT_SUM_FLAG
)) {
1727 int npages
= npages_for_summary_flush(sbi
, true);
1730 ra_meta_pages(sbi
, start_sum_block(sbi
), npages
,
1733 /* restore for compacted data summary */
1734 if (read_compacted_summaries(sbi
))
1736 type
= CURSEG_HOT_NODE
;
1739 if (__exist_node_summaries(sbi
))
1740 ra_meta_pages(sbi
, sum_blk_addr(sbi
, NR_CURSEG_TYPE
, type
),
1741 NR_CURSEG_TYPE
- type
, META_CP
, true);
1743 for (; type
<= CURSEG_COLD_NODE
; type
++) {
1744 err
= read_normal_summaries(sbi
, type
);
1752 static void write_compacted_summaries(struct f2fs_sb_info
*sbi
, block_t blkaddr
)
1755 unsigned char *kaddr
;
1756 struct f2fs_summary
*summary
;
1757 struct curseg_info
*seg_i
;
1758 int written_size
= 0;
1761 page
= grab_meta_page(sbi
, blkaddr
++);
1762 kaddr
= (unsigned char *)page_address(page
);
1764 /* Step 1: write nat cache */
1765 seg_i
= CURSEG_I(sbi
, CURSEG_HOT_DATA
);
1766 memcpy(kaddr
, seg_i
->journal
, SUM_JOURNAL_SIZE
);
1767 written_size
+= SUM_JOURNAL_SIZE
;
1769 /* Step 2: write sit cache */
1770 seg_i
= CURSEG_I(sbi
, CURSEG_COLD_DATA
);
1771 memcpy(kaddr
+ written_size
, seg_i
->journal
, SUM_JOURNAL_SIZE
);
1772 written_size
+= SUM_JOURNAL_SIZE
;
1774 /* Step 3: write summary entries */
1775 for (i
= CURSEG_HOT_DATA
; i
<= CURSEG_COLD_DATA
; i
++) {
1776 unsigned short blkoff
;
1777 seg_i
= CURSEG_I(sbi
, i
);
1778 if (sbi
->ckpt
->alloc_type
[i
] == SSR
)
1779 blkoff
= sbi
->blocks_per_seg
;
1781 blkoff
= curseg_blkoff(sbi
, i
);
1783 for (j
= 0; j
< blkoff
; j
++) {
1785 page
= grab_meta_page(sbi
, blkaddr
++);
1786 kaddr
= (unsigned char *)page_address(page
);
1789 summary
= (struct f2fs_summary
*)(kaddr
+ written_size
);
1790 *summary
= seg_i
->sum_blk
->entries
[j
];
1791 written_size
+= SUMMARY_SIZE
;
1793 if (written_size
+ SUMMARY_SIZE
<= PAGE_SIZE
-
1797 set_page_dirty(page
);
1798 f2fs_put_page(page
, 1);
1803 set_page_dirty(page
);
1804 f2fs_put_page(page
, 1);
1808 static void write_normal_summaries(struct f2fs_sb_info
*sbi
,
1809 block_t blkaddr
, int type
)
1812 if (IS_DATASEG(type
))
1813 end
= type
+ NR_CURSEG_DATA_TYPE
;
1815 end
= type
+ NR_CURSEG_NODE_TYPE
;
1817 for (i
= type
; i
< end
; i
++)
1818 write_current_sum_page(sbi
, i
, blkaddr
+ (i
- type
));
1821 void write_data_summaries(struct f2fs_sb_info
*sbi
, block_t start_blk
)
1823 if (is_set_ckpt_flags(F2FS_CKPT(sbi
), CP_COMPACT_SUM_FLAG
))
1824 write_compacted_summaries(sbi
, start_blk
);
1826 write_normal_summaries(sbi
, start_blk
, CURSEG_HOT_DATA
);
1829 void write_node_summaries(struct f2fs_sb_info
*sbi
, block_t start_blk
)
1831 write_normal_summaries(sbi
, start_blk
, CURSEG_HOT_NODE
);
1834 int lookup_journal_in_cursum(struct f2fs_journal
*journal
, int type
,
1835 unsigned int val
, int alloc
)
1839 if (type
== NAT_JOURNAL
) {
1840 for (i
= 0; i
< nats_in_cursum(journal
); i
++) {
1841 if (le32_to_cpu(nid_in_journal(journal
, i
)) == val
)
1844 if (alloc
&& __has_cursum_space(journal
, 1, NAT_JOURNAL
))
1845 return update_nats_in_cursum(journal
, 1);
1846 } else if (type
== SIT_JOURNAL
) {
1847 for (i
= 0; i
< sits_in_cursum(journal
); i
++)
1848 if (le32_to_cpu(segno_in_journal(journal
, i
)) == val
)
1850 if (alloc
&& __has_cursum_space(journal
, 1, SIT_JOURNAL
))
1851 return update_sits_in_cursum(journal
, 1);
1856 static struct page
*get_current_sit_page(struct f2fs_sb_info
*sbi
,
1859 return get_meta_page(sbi
, current_sit_addr(sbi
, segno
));
1862 static struct page
*get_next_sit_page(struct f2fs_sb_info
*sbi
,
1865 struct sit_info
*sit_i
= SIT_I(sbi
);
1866 struct page
*src_page
, *dst_page
;
1867 pgoff_t src_off
, dst_off
;
1868 void *src_addr
, *dst_addr
;
1870 src_off
= current_sit_addr(sbi
, start
);
1871 dst_off
= next_sit_addr(sbi
, src_off
);
1873 /* get current sit block page without lock */
1874 src_page
= get_meta_page(sbi
, src_off
);
1875 dst_page
= grab_meta_page(sbi
, dst_off
);
1876 f2fs_bug_on(sbi
, PageDirty(src_page
));
1878 src_addr
= page_address(src_page
);
1879 dst_addr
= page_address(dst_page
);
1880 memcpy(dst_addr
, src_addr
, PAGE_SIZE
);
1882 set_page_dirty(dst_page
);
1883 f2fs_put_page(src_page
, 1);
1885 set_to_next_sit(sit_i
, start
);
1890 static struct sit_entry_set
*grab_sit_entry_set(void)
1892 struct sit_entry_set
*ses
=
1893 f2fs_kmem_cache_alloc(sit_entry_set_slab
, GFP_NOFS
);
1896 INIT_LIST_HEAD(&ses
->set_list
);
1900 static void release_sit_entry_set(struct sit_entry_set
*ses
)
1902 list_del(&ses
->set_list
);
1903 kmem_cache_free(sit_entry_set_slab
, ses
);
1906 static void adjust_sit_entry_set(struct sit_entry_set
*ses
,
1907 struct list_head
*head
)
1909 struct sit_entry_set
*next
= ses
;
1911 if (list_is_last(&ses
->set_list
, head
))
1914 list_for_each_entry_continue(next
, head
, set_list
)
1915 if (ses
->entry_cnt
<= next
->entry_cnt
)
1918 list_move_tail(&ses
->set_list
, &next
->set_list
);
1921 static void add_sit_entry(unsigned int segno
, struct list_head
*head
)
1923 struct sit_entry_set
*ses
;
1924 unsigned int start_segno
= START_SEGNO(segno
);
1926 list_for_each_entry(ses
, head
, set_list
) {
1927 if (ses
->start_segno
== start_segno
) {
1929 adjust_sit_entry_set(ses
, head
);
1934 ses
= grab_sit_entry_set();
1936 ses
->start_segno
= start_segno
;
1938 list_add(&ses
->set_list
, head
);
1941 static void add_sits_in_set(struct f2fs_sb_info
*sbi
)
1943 struct f2fs_sm_info
*sm_info
= SM_I(sbi
);
1944 struct list_head
*set_list
= &sm_info
->sit_entry_set
;
1945 unsigned long *bitmap
= SIT_I(sbi
)->dirty_sentries_bitmap
;
1948 for_each_set_bit(segno
, bitmap
, MAIN_SEGS(sbi
))
1949 add_sit_entry(segno
, set_list
);
1952 static void remove_sits_in_journal(struct f2fs_sb_info
*sbi
)
1954 struct curseg_info
*curseg
= CURSEG_I(sbi
, CURSEG_COLD_DATA
);
1955 struct f2fs_journal
*journal
= curseg
->journal
;
1958 down_write(&curseg
->journal_rwsem
);
1959 for (i
= 0; i
< sits_in_cursum(journal
); i
++) {
1963 segno
= le32_to_cpu(segno_in_journal(journal
, i
));
1964 dirtied
= __mark_sit_entry_dirty(sbi
, segno
);
1967 add_sit_entry(segno
, &SM_I(sbi
)->sit_entry_set
);
1969 update_sits_in_cursum(journal
, -i
);
1970 up_write(&curseg
->journal_rwsem
);
1974 * CP calls this function, which flushes SIT entries including sit_journal,
1975 * and moves prefree segs to free segs.
1977 void flush_sit_entries(struct f2fs_sb_info
*sbi
, struct cp_control
*cpc
)
1979 struct sit_info
*sit_i
= SIT_I(sbi
);
1980 unsigned long *bitmap
= sit_i
->dirty_sentries_bitmap
;
1981 struct curseg_info
*curseg
= CURSEG_I(sbi
, CURSEG_COLD_DATA
);
1982 struct f2fs_journal
*journal
= curseg
->journal
;
1983 struct sit_entry_set
*ses
, *tmp
;
1984 struct list_head
*head
= &SM_I(sbi
)->sit_entry_set
;
1985 bool to_journal
= true;
1986 struct seg_entry
*se
;
1988 mutex_lock(&sit_i
->sentry_lock
);
1990 if (!sit_i
->dirty_sentries
)
1994 * add and account sit entries of dirty bitmap in sit entry
1997 add_sits_in_set(sbi
);
2000 * if there are no enough space in journal to store dirty sit
2001 * entries, remove all entries from journal and add and account
2002 * them in sit entry set.
2004 if (!__has_cursum_space(journal
, sit_i
->dirty_sentries
, SIT_JOURNAL
))
2005 remove_sits_in_journal(sbi
);
2008 * there are two steps to flush sit entries:
2009 * #1, flush sit entries to journal in current cold data summary block.
2010 * #2, flush sit entries to sit page.
2012 list_for_each_entry_safe(ses
, tmp
, head
, set_list
) {
2013 struct page
*page
= NULL
;
2014 struct f2fs_sit_block
*raw_sit
= NULL
;
2015 unsigned int start_segno
= ses
->start_segno
;
2016 unsigned int end
= min(start_segno
+ SIT_ENTRY_PER_BLOCK
,
2017 (unsigned long)MAIN_SEGS(sbi
));
2018 unsigned int segno
= start_segno
;
2021 !__has_cursum_space(journal
, ses
->entry_cnt
, SIT_JOURNAL
))
2025 down_write(&curseg
->journal_rwsem
);
2027 page
= get_next_sit_page(sbi
, start_segno
);
2028 raw_sit
= page_address(page
);
2031 /* flush dirty sit entries in region of current sit set */
2032 for_each_set_bit_from(segno
, bitmap
, end
) {
2033 int offset
, sit_offset
;
2035 se
= get_seg_entry(sbi
, segno
);
2037 /* add discard candidates */
2038 if (cpc
->reason
!= CP_DISCARD
) {
2039 cpc
->trim_start
= segno
;
2040 add_discard_addrs(sbi
, cpc
);
2044 offset
= lookup_journal_in_cursum(journal
,
2045 SIT_JOURNAL
, segno
, 1);
2046 f2fs_bug_on(sbi
, offset
< 0);
2047 segno_in_journal(journal
, offset
) =
2049 seg_info_to_raw_sit(se
,
2050 &sit_in_journal(journal
, offset
));
2052 sit_offset
= SIT_ENTRY_OFFSET(sit_i
, segno
);
2053 seg_info_to_raw_sit(se
,
2054 &raw_sit
->entries
[sit_offset
]);
2057 __clear_bit(segno
, bitmap
);
2058 sit_i
->dirty_sentries
--;
2063 up_write(&curseg
->journal_rwsem
);
2065 f2fs_put_page(page
, 1);
2067 f2fs_bug_on(sbi
, ses
->entry_cnt
);
2068 release_sit_entry_set(ses
);
2071 f2fs_bug_on(sbi
, !list_empty(head
));
2072 f2fs_bug_on(sbi
, sit_i
->dirty_sentries
);
2074 if (cpc
->reason
== CP_DISCARD
) {
2075 for (; cpc
->trim_start
<= cpc
->trim_end
; cpc
->trim_start
++)
2076 add_discard_addrs(sbi
, cpc
);
2078 mutex_unlock(&sit_i
->sentry_lock
);
2080 set_prefree_as_free_segments(sbi
);
2083 static int build_sit_info(struct f2fs_sb_info
*sbi
)
2085 struct f2fs_super_block
*raw_super
= F2FS_RAW_SUPER(sbi
);
2086 struct f2fs_checkpoint
*ckpt
= F2FS_CKPT(sbi
);
2087 struct sit_info
*sit_i
;
2088 unsigned int sit_segs
, start
;
2089 char *src_bitmap
, *dst_bitmap
;
2090 unsigned int bitmap_size
;
2092 /* allocate memory for SIT information */
2093 sit_i
= kzalloc(sizeof(struct sit_info
), GFP_KERNEL
);
2097 SM_I(sbi
)->sit_info
= sit_i
;
2099 sit_i
->sentries
= f2fs_kvzalloc(MAIN_SEGS(sbi
) *
2100 sizeof(struct seg_entry
), GFP_KERNEL
);
2101 if (!sit_i
->sentries
)
2104 bitmap_size
= f2fs_bitmap_size(MAIN_SEGS(sbi
));
2105 sit_i
->dirty_sentries_bitmap
= f2fs_kvzalloc(bitmap_size
, GFP_KERNEL
);
2106 if (!sit_i
->dirty_sentries_bitmap
)
2109 for (start
= 0; start
< MAIN_SEGS(sbi
); start
++) {
2110 sit_i
->sentries
[start
].cur_valid_map
2111 = kzalloc(SIT_VBLOCK_MAP_SIZE
, GFP_KERNEL
);
2112 sit_i
->sentries
[start
].ckpt_valid_map
2113 = kzalloc(SIT_VBLOCK_MAP_SIZE
, GFP_KERNEL
);
2114 sit_i
->sentries
[start
].discard_map
2115 = kzalloc(SIT_VBLOCK_MAP_SIZE
, GFP_KERNEL
);
2116 if (!sit_i
->sentries
[start
].cur_valid_map
||
2117 !sit_i
->sentries
[start
].ckpt_valid_map
||
2118 !sit_i
->sentries
[start
].discard_map
)
2122 sit_i
->tmp_map
= kzalloc(SIT_VBLOCK_MAP_SIZE
, GFP_KERNEL
);
2123 if (!sit_i
->tmp_map
)
2126 if (sbi
->segs_per_sec
> 1) {
2127 sit_i
->sec_entries
= f2fs_kvzalloc(MAIN_SECS(sbi
) *
2128 sizeof(struct sec_entry
), GFP_KERNEL
);
2129 if (!sit_i
->sec_entries
)
2133 /* get information related with SIT */
2134 sit_segs
= le32_to_cpu(raw_super
->segment_count_sit
) >> 1;
2136 /* setup SIT bitmap from ckeckpoint pack */
2137 bitmap_size
= __bitmap_size(sbi
, SIT_BITMAP
);
2138 src_bitmap
= __bitmap_ptr(sbi
, SIT_BITMAP
);
2140 dst_bitmap
= kmemdup(src_bitmap
, bitmap_size
, GFP_KERNEL
);
2144 /* init SIT information */
2145 sit_i
->s_ops
= &default_salloc_ops
;
2147 sit_i
->sit_base_addr
= le32_to_cpu(raw_super
->sit_blkaddr
);
2148 sit_i
->sit_blocks
= sit_segs
<< sbi
->log_blocks_per_seg
;
2149 sit_i
->written_valid_blocks
= le64_to_cpu(ckpt
->valid_block_count
);
2150 sit_i
->sit_bitmap
= dst_bitmap
;
2151 sit_i
->bitmap_size
= bitmap_size
;
2152 sit_i
->dirty_sentries
= 0;
2153 sit_i
->sents_per_block
= SIT_ENTRY_PER_BLOCK
;
2154 sit_i
->elapsed_time
= le64_to_cpu(sbi
->ckpt
->elapsed_time
);
2155 sit_i
->mounted_time
= CURRENT_TIME_SEC
.tv_sec
;
2156 mutex_init(&sit_i
->sentry_lock
);
2160 static int build_free_segmap(struct f2fs_sb_info
*sbi
)
2162 struct free_segmap_info
*free_i
;
2163 unsigned int bitmap_size
, sec_bitmap_size
;
2165 /* allocate memory for free segmap information */
2166 free_i
= kzalloc(sizeof(struct free_segmap_info
), GFP_KERNEL
);
2170 SM_I(sbi
)->free_info
= free_i
;
2172 bitmap_size
= f2fs_bitmap_size(MAIN_SEGS(sbi
));
2173 free_i
->free_segmap
= f2fs_kvmalloc(bitmap_size
, GFP_KERNEL
);
2174 if (!free_i
->free_segmap
)
2177 sec_bitmap_size
= f2fs_bitmap_size(MAIN_SECS(sbi
));
2178 free_i
->free_secmap
= f2fs_kvmalloc(sec_bitmap_size
, GFP_KERNEL
);
2179 if (!free_i
->free_secmap
)
2182 /* set all segments as dirty temporarily */
2183 memset(free_i
->free_segmap
, 0xff, bitmap_size
);
2184 memset(free_i
->free_secmap
, 0xff, sec_bitmap_size
);
2186 /* init free segmap information */
2187 free_i
->start_segno
= GET_SEGNO_FROM_SEG0(sbi
, MAIN_BLKADDR(sbi
));
2188 free_i
->free_segments
= 0;
2189 free_i
->free_sections
= 0;
2190 spin_lock_init(&free_i
->segmap_lock
);
2194 static int build_curseg(struct f2fs_sb_info
*sbi
)
2196 struct curseg_info
*array
;
2199 array
= kcalloc(NR_CURSEG_TYPE
, sizeof(*array
), GFP_KERNEL
);
2203 SM_I(sbi
)->curseg_array
= array
;
2205 for (i
= 0; i
< NR_CURSEG_TYPE
; i
++) {
2206 mutex_init(&array
[i
].curseg_mutex
);
2207 array
[i
].sum_blk
= kzalloc(PAGE_SIZE
, GFP_KERNEL
);
2208 if (!array
[i
].sum_blk
)
2210 init_rwsem(&array
[i
].journal_rwsem
);
2211 array
[i
].journal
= kzalloc(sizeof(struct f2fs_journal
),
2213 if (!array
[i
].journal
)
2215 array
[i
].segno
= NULL_SEGNO
;
2216 array
[i
].next_blkoff
= 0;
2218 return restore_curseg_summaries(sbi
);
2221 static void build_sit_entries(struct f2fs_sb_info
*sbi
)
2223 struct sit_info
*sit_i
= SIT_I(sbi
);
2224 struct curseg_info
*curseg
= CURSEG_I(sbi
, CURSEG_COLD_DATA
);
2225 struct f2fs_journal
*journal
= curseg
->journal
;
2226 int sit_blk_cnt
= SIT_BLK_CNT(sbi
);
2227 unsigned int i
, start
, end
;
2228 unsigned int readed
, start_blk
= 0;
2229 int nrpages
= MAX_BIO_BLOCKS(sbi
) * 8;
2232 readed
= ra_meta_pages(sbi
, start_blk
, nrpages
, META_SIT
, true);
2234 start
= start_blk
* sit_i
->sents_per_block
;
2235 end
= (start_blk
+ readed
) * sit_i
->sents_per_block
;
2237 for (; start
< end
&& start
< MAIN_SEGS(sbi
); start
++) {
2238 struct seg_entry
*se
= &sit_i
->sentries
[start
];
2239 struct f2fs_sit_block
*sit_blk
;
2240 struct f2fs_sit_entry sit
;
2243 down_read(&curseg
->journal_rwsem
);
2244 for (i
= 0; i
< sits_in_cursum(journal
); i
++) {
2245 if (le32_to_cpu(segno_in_journal(journal
, i
))
2247 sit
= sit_in_journal(journal
, i
);
2248 up_read(&curseg
->journal_rwsem
);
2252 up_read(&curseg
->journal_rwsem
);
2254 page
= get_current_sit_page(sbi
, start
);
2255 sit_blk
= (struct f2fs_sit_block
*)page_address(page
);
2256 sit
= sit_blk
->entries
[SIT_ENTRY_OFFSET(sit_i
, start
)];
2257 f2fs_put_page(page
, 1);
2259 check_block_count(sbi
, start
, &sit
);
2260 seg_info_from_raw_sit(se
, &sit
);
2262 /* build discard map only one time */
2263 memcpy(se
->discard_map
, se
->cur_valid_map
, SIT_VBLOCK_MAP_SIZE
);
2264 sbi
->discard_blks
+= sbi
->blocks_per_seg
- se
->valid_blocks
;
2266 if (sbi
->segs_per_sec
> 1) {
2267 struct sec_entry
*e
= get_sec_entry(sbi
, start
);
2268 e
->valid_blocks
+= se
->valid_blocks
;
2271 start_blk
+= readed
;
2272 } while (start_blk
< sit_blk_cnt
);
2275 static void init_free_segmap(struct f2fs_sb_info
*sbi
)
2280 for (start
= 0; start
< MAIN_SEGS(sbi
); start
++) {
2281 struct seg_entry
*sentry
= get_seg_entry(sbi
, start
);
2282 if (!sentry
->valid_blocks
)
2283 __set_free(sbi
, start
);
2286 /* set use the current segments */
2287 for (type
= CURSEG_HOT_DATA
; type
<= CURSEG_COLD_NODE
; type
++) {
2288 struct curseg_info
*curseg_t
= CURSEG_I(sbi
, type
);
2289 __set_test_and_inuse(sbi
, curseg_t
->segno
);
2293 static void init_dirty_segmap(struct f2fs_sb_info
*sbi
)
2295 struct dirty_seglist_info
*dirty_i
= DIRTY_I(sbi
);
2296 struct free_segmap_info
*free_i
= FREE_I(sbi
);
2297 unsigned int segno
= 0, offset
= 0;
2298 unsigned short valid_blocks
;
2301 /* find dirty segment based on free segmap */
2302 segno
= find_next_inuse(free_i
, MAIN_SEGS(sbi
), offset
);
2303 if (segno
>= MAIN_SEGS(sbi
))
2306 valid_blocks
= get_valid_blocks(sbi
, segno
, 0);
2307 if (valid_blocks
== sbi
->blocks_per_seg
|| !valid_blocks
)
2309 if (valid_blocks
> sbi
->blocks_per_seg
) {
2310 f2fs_bug_on(sbi
, 1);
2313 mutex_lock(&dirty_i
->seglist_lock
);
2314 __locate_dirty_segment(sbi
, segno
, DIRTY
);
2315 mutex_unlock(&dirty_i
->seglist_lock
);
2319 static int init_victim_secmap(struct f2fs_sb_info
*sbi
)
2321 struct dirty_seglist_info
*dirty_i
= DIRTY_I(sbi
);
2322 unsigned int bitmap_size
= f2fs_bitmap_size(MAIN_SECS(sbi
));
2324 dirty_i
->victim_secmap
= f2fs_kvzalloc(bitmap_size
, GFP_KERNEL
);
2325 if (!dirty_i
->victim_secmap
)
2330 static int build_dirty_segmap(struct f2fs_sb_info
*sbi
)
2332 struct dirty_seglist_info
*dirty_i
;
2333 unsigned int bitmap_size
, i
;
2335 /* allocate memory for dirty segments list information */
2336 dirty_i
= kzalloc(sizeof(struct dirty_seglist_info
), GFP_KERNEL
);
2340 SM_I(sbi
)->dirty_info
= dirty_i
;
2341 mutex_init(&dirty_i
->seglist_lock
);
2343 bitmap_size
= f2fs_bitmap_size(MAIN_SEGS(sbi
));
2345 for (i
= 0; i
< NR_DIRTY_TYPE
; i
++) {
2346 dirty_i
->dirty_segmap
[i
] = f2fs_kvzalloc(bitmap_size
, GFP_KERNEL
);
2347 if (!dirty_i
->dirty_segmap
[i
])
2351 init_dirty_segmap(sbi
);
2352 return init_victim_secmap(sbi
);
2356 * Update min, max modified time for cost-benefit GC algorithm
2358 static void init_min_max_mtime(struct f2fs_sb_info
*sbi
)
2360 struct sit_info
*sit_i
= SIT_I(sbi
);
2363 mutex_lock(&sit_i
->sentry_lock
);
2365 sit_i
->min_mtime
= LLONG_MAX
;
2367 for (segno
= 0; segno
< MAIN_SEGS(sbi
); segno
+= sbi
->segs_per_sec
) {
2369 unsigned long long mtime
= 0;
2371 for (i
= 0; i
< sbi
->segs_per_sec
; i
++)
2372 mtime
+= get_seg_entry(sbi
, segno
+ i
)->mtime
;
2374 mtime
= div_u64(mtime
, sbi
->segs_per_sec
);
2376 if (sit_i
->min_mtime
> mtime
)
2377 sit_i
->min_mtime
= mtime
;
2379 sit_i
->max_mtime
= get_mtime(sbi
);
2380 mutex_unlock(&sit_i
->sentry_lock
);
2383 int build_segment_manager(struct f2fs_sb_info
*sbi
)
2385 struct f2fs_super_block
*raw_super
= F2FS_RAW_SUPER(sbi
);
2386 struct f2fs_checkpoint
*ckpt
= F2FS_CKPT(sbi
);
2387 struct f2fs_sm_info
*sm_info
;
2390 sm_info
= kzalloc(sizeof(struct f2fs_sm_info
), GFP_KERNEL
);
2395 sbi
->sm_info
= sm_info
;
2396 sm_info
->seg0_blkaddr
= le32_to_cpu(raw_super
->segment0_blkaddr
);
2397 sm_info
->main_blkaddr
= le32_to_cpu(raw_super
->main_blkaddr
);
2398 sm_info
->segment_count
= le32_to_cpu(raw_super
->segment_count
);
2399 sm_info
->reserved_segments
= le32_to_cpu(ckpt
->rsvd_segment_count
);
2400 sm_info
->ovp_segments
= le32_to_cpu(ckpt
->overprov_segment_count
);
2401 sm_info
->main_segments
= le32_to_cpu(raw_super
->segment_count_main
);
2402 sm_info
->ssa_blkaddr
= le32_to_cpu(raw_super
->ssa_blkaddr
);
2403 sm_info
->rec_prefree_segments
= sm_info
->main_segments
*
2404 DEF_RECLAIM_PREFREE_SEGMENTS
/ 100;
2405 if (!test_opt(sbi
, LFS
))
2406 sm_info
->ipu_policy
= 1 << F2FS_IPU_FSYNC
;
2407 sm_info
->min_ipu_util
= DEF_MIN_IPU_UTIL
;
2408 sm_info
->min_fsync_blocks
= DEF_MIN_FSYNC_BLOCKS
;
2410 INIT_LIST_HEAD(&sm_info
->discard_list
);
2411 sm_info
->nr_discards
= 0;
2412 sm_info
->max_discards
= 0;
2414 sm_info
->trim_sections
= DEF_BATCHED_TRIM_SECTIONS
;
2416 INIT_LIST_HEAD(&sm_info
->sit_entry_set
);
2418 if (test_opt(sbi
, FLUSH_MERGE
) && !f2fs_readonly(sbi
->sb
)) {
2419 err
= create_flush_cmd_control(sbi
);
2424 err
= build_sit_info(sbi
);
2427 err
= build_free_segmap(sbi
);
2430 err
= build_curseg(sbi
);
2434 /* reinit free segmap based on SIT */
2435 build_sit_entries(sbi
);
2437 init_free_segmap(sbi
);
2438 err
= build_dirty_segmap(sbi
);
2442 init_min_max_mtime(sbi
);
2446 static void discard_dirty_segmap(struct f2fs_sb_info
*sbi
,
2447 enum dirty_type dirty_type
)
2449 struct dirty_seglist_info
*dirty_i
= DIRTY_I(sbi
);
2451 mutex_lock(&dirty_i
->seglist_lock
);
2452 kvfree(dirty_i
->dirty_segmap
[dirty_type
]);
2453 dirty_i
->nr_dirty
[dirty_type
] = 0;
2454 mutex_unlock(&dirty_i
->seglist_lock
);
2457 static void destroy_victim_secmap(struct f2fs_sb_info
*sbi
)
2459 struct dirty_seglist_info
*dirty_i
= DIRTY_I(sbi
);
2460 kvfree(dirty_i
->victim_secmap
);
2463 static void destroy_dirty_segmap(struct f2fs_sb_info
*sbi
)
2465 struct dirty_seglist_info
*dirty_i
= DIRTY_I(sbi
);
2471 /* discard pre-free/dirty segments list */
2472 for (i
= 0; i
< NR_DIRTY_TYPE
; i
++)
2473 discard_dirty_segmap(sbi
, i
);
2475 destroy_victim_secmap(sbi
);
2476 SM_I(sbi
)->dirty_info
= NULL
;
2480 static void destroy_curseg(struct f2fs_sb_info
*sbi
)
2482 struct curseg_info
*array
= SM_I(sbi
)->curseg_array
;
2487 SM_I(sbi
)->curseg_array
= NULL
;
2488 for (i
= 0; i
< NR_CURSEG_TYPE
; i
++) {
2489 kfree(array
[i
].sum_blk
);
2490 kfree(array
[i
].journal
);
2495 static void destroy_free_segmap(struct f2fs_sb_info
*sbi
)
2497 struct free_segmap_info
*free_i
= SM_I(sbi
)->free_info
;
2500 SM_I(sbi
)->free_info
= NULL
;
2501 kvfree(free_i
->free_segmap
);
2502 kvfree(free_i
->free_secmap
);
2506 static void destroy_sit_info(struct f2fs_sb_info
*sbi
)
2508 struct sit_info
*sit_i
= SIT_I(sbi
);
2514 if (sit_i
->sentries
) {
2515 for (start
= 0; start
< MAIN_SEGS(sbi
); start
++) {
2516 kfree(sit_i
->sentries
[start
].cur_valid_map
);
2517 kfree(sit_i
->sentries
[start
].ckpt_valid_map
);
2518 kfree(sit_i
->sentries
[start
].discard_map
);
2521 kfree(sit_i
->tmp_map
);
2523 kvfree(sit_i
->sentries
);
2524 kvfree(sit_i
->sec_entries
);
2525 kvfree(sit_i
->dirty_sentries_bitmap
);
2527 SM_I(sbi
)->sit_info
= NULL
;
2528 kfree(sit_i
->sit_bitmap
);
2532 void destroy_segment_manager(struct f2fs_sb_info
*sbi
)
2534 struct f2fs_sm_info
*sm_info
= SM_I(sbi
);
2538 destroy_flush_cmd_control(sbi
);
2539 destroy_dirty_segmap(sbi
);
2540 destroy_curseg(sbi
);
2541 destroy_free_segmap(sbi
);
2542 destroy_sit_info(sbi
);
2543 sbi
->sm_info
= NULL
;
2547 int __init
create_segment_manager_caches(void)
2549 discard_entry_slab
= f2fs_kmem_cache_create("discard_entry",
2550 sizeof(struct discard_entry
));
2551 if (!discard_entry_slab
)
2554 sit_entry_set_slab
= f2fs_kmem_cache_create("sit_entry_set",
2555 sizeof(struct sit_entry_set
));
2556 if (!sit_entry_set_slab
)
2557 goto destory_discard_entry
;
2559 inmem_entry_slab
= f2fs_kmem_cache_create("inmem_page_entry",
2560 sizeof(struct inmem_pages
));
2561 if (!inmem_entry_slab
)
2562 goto destroy_sit_entry_set
;
2565 destroy_sit_entry_set
:
2566 kmem_cache_destroy(sit_entry_set_slab
);
2567 destory_discard_entry
:
2568 kmem_cache_destroy(discard_entry_slab
);
2573 void destroy_segment_manager_caches(void)
2575 kmem_cache_destroy(sit_entry_set_slab
);
2576 kmem_cache_destroy(discard_entry_slab
);
2577 kmem_cache_destroy(inmem_entry_slab
);