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 * We should do GC or end up with checkpoint, if there are so many dirty
350 * dir/node pages without enough free segments.
352 if (has_not_enough_free_secs(sbi
, 0)) {
353 mutex_lock(&sbi
->gc_mutex
);
358 void f2fs_balance_fs_bg(struct f2fs_sb_info
*sbi
)
360 /* try to shrink extent cache when there is no enough memory */
361 if (!available_free_memory(sbi
, EXTENT_CACHE
))
362 f2fs_shrink_extent_tree(sbi
, EXTENT_CACHE_SHRINK_NUMBER
);
364 /* check the # of cached NAT entries */
365 if (!available_free_memory(sbi
, NAT_ENTRIES
))
366 try_to_free_nats(sbi
, NAT_ENTRY_PER_BLOCK
);
368 if (!available_free_memory(sbi
, FREE_NIDS
))
369 try_to_free_nids(sbi
, NAT_ENTRY_PER_BLOCK
* FREE_NID_PAGES
);
371 /* checkpoint is the only way to shrink partial cached entries */
372 if (!available_free_memory(sbi
, NAT_ENTRIES
) ||
373 !available_free_memory(sbi
, INO_ENTRIES
) ||
374 excess_prefree_segs(sbi
) ||
375 excess_dirty_nats(sbi
) ||
376 (is_idle(sbi
) && f2fs_time_over(sbi
, CP_TIME
))) {
377 if (test_opt(sbi
, DATA_FLUSH
)) {
378 struct blk_plug plug
;
380 blk_start_plug(&plug
);
381 sync_dirty_inodes(sbi
, FILE_INODE
);
382 blk_finish_plug(&plug
);
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;
711 mutex_lock(&dirty_i
->seglist_lock
);
715 start
= find_next_bit(prefree_map
, MAIN_SEGS(sbi
), end
+ 1);
716 if (start
>= MAIN_SEGS(sbi
))
718 end
= find_next_zero_bit(prefree_map
, MAIN_SEGS(sbi
),
721 for (i
= start
; i
< end
; i
++)
722 clear_bit(i
, prefree_map
);
724 dirty_i
->nr_dirty
[PRE
] -= end
- start
;
726 if (!test_opt(sbi
, DISCARD
))
729 f2fs_issue_discard(sbi
, START_BLOCK(sbi
, start
),
730 (end
- start
) << sbi
->log_blocks_per_seg
);
732 mutex_unlock(&dirty_i
->seglist_lock
);
734 /* send small discards */
735 list_for_each_entry_safe(entry
, this, head
, list
) {
736 if (cpc
->reason
== CP_DISCARD
&& entry
->len
< cpc
->trim_minlen
)
738 f2fs_issue_discard(sbi
, entry
->blkaddr
, entry
->len
);
739 cpc
->trimmed
+= entry
->len
;
741 list_del(&entry
->list
);
742 SM_I(sbi
)->nr_discards
-= entry
->len
;
743 kmem_cache_free(discard_entry_slab
, entry
);
747 static bool __mark_sit_entry_dirty(struct f2fs_sb_info
*sbi
, unsigned int segno
)
749 struct sit_info
*sit_i
= SIT_I(sbi
);
751 if (!__test_and_set_bit(segno
, sit_i
->dirty_sentries_bitmap
)) {
752 sit_i
->dirty_sentries
++;
759 static void __set_sit_entry_type(struct f2fs_sb_info
*sbi
, int type
,
760 unsigned int segno
, int modified
)
762 struct seg_entry
*se
= get_seg_entry(sbi
, segno
);
765 __mark_sit_entry_dirty(sbi
, segno
);
768 static void update_sit_entry(struct f2fs_sb_info
*sbi
, block_t blkaddr
, int del
)
770 struct seg_entry
*se
;
771 unsigned int segno
, offset
;
772 long int new_vblocks
;
774 segno
= GET_SEGNO(sbi
, blkaddr
);
776 se
= get_seg_entry(sbi
, segno
);
777 new_vblocks
= se
->valid_blocks
+ del
;
778 offset
= GET_BLKOFF_FROM_SEG0(sbi
, blkaddr
);
780 f2fs_bug_on(sbi
, (new_vblocks
>> (sizeof(unsigned short) << 3) ||
781 (new_vblocks
> sbi
->blocks_per_seg
)));
783 se
->valid_blocks
= new_vblocks
;
784 se
->mtime
= get_mtime(sbi
);
785 SIT_I(sbi
)->max_mtime
= se
->mtime
;
787 /* Update valid block bitmap */
789 if (f2fs_test_and_set_bit(offset
, se
->cur_valid_map
))
791 if (!f2fs_test_and_set_bit(offset
, se
->discard_map
))
794 if (!f2fs_test_and_clear_bit(offset
, se
->cur_valid_map
))
796 if (f2fs_test_and_clear_bit(offset
, se
->discard_map
))
799 if (!f2fs_test_bit(offset
, se
->ckpt_valid_map
))
800 se
->ckpt_valid_blocks
+= del
;
802 __mark_sit_entry_dirty(sbi
, segno
);
804 /* update total number of valid blocks to be written in ckpt area */
805 SIT_I(sbi
)->written_valid_blocks
+= del
;
807 if (sbi
->segs_per_sec
> 1)
808 get_sec_entry(sbi
, segno
)->valid_blocks
+= del
;
811 void refresh_sit_entry(struct f2fs_sb_info
*sbi
, block_t old
, block_t
new)
813 update_sit_entry(sbi
, new, 1);
814 if (GET_SEGNO(sbi
, old
) != NULL_SEGNO
)
815 update_sit_entry(sbi
, old
, -1);
817 locate_dirty_segment(sbi
, GET_SEGNO(sbi
, old
));
818 locate_dirty_segment(sbi
, GET_SEGNO(sbi
, new));
821 void invalidate_blocks(struct f2fs_sb_info
*sbi
, block_t addr
)
823 unsigned int segno
= GET_SEGNO(sbi
, addr
);
824 struct sit_info
*sit_i
= SIT_I(sbi
);
826 f2fs_bug_on(sbi
, addr
== NULL_ADDR
);
827 if (addr
== NEW_ADDR
)
830 /* add it into sit main buffer */
831 mutex_lock(&sit_i
->sentry_lock
);
833 update_sit_entry(sbi
, addr
, -1);
835 /* add it into dirty seglist */
836 locate_dirty_segment(sbi
, segno
);
838 mutex_unlock(&sit_i
->sentry_lock
);
841 bool is_checkpointed_data(struct f2fs_sb_info
*sbi
, block_t blkaddr
)
843 struct sit_info
*sit_i
= SIT_I(sbi
);
844 unsigned int segno
, offset
;
845 struct seg_entry
*se
;
848 if (blkaddr
== NEW_ADDR
|| blkaddr
== NULL_ADDR
)
851 mutex_lock(&sit_i
->sentry_lock
);
853 segno
= GET_SEGNO(sbi
, blkaddr
);
854 se
= get_seg_entry(sbi
, segno
);
855 offset
= GET_BLKOFF_FROM_SEG0(sbi
, blkaddr
);
857 if (f2fs_test_bit(offset
, se
->ckpt_valid_map
))
860 mutex_unlock(&sit_i
->sentry_lock
);
866 * This function should be resided under the curseg_mutex lock
868 static void __add_sum_entry(struct f2fs_sb_info
*sbi
, int type
,
869 struct f2fs_summary
*sum
)
871 struct curseg_info
*curseg
= CURSEG_I(sbi
, type
);
872 void *addr
= curseg
->sum_blk
;
873 addr
+= curseg
->next_blkoff
* sizeof(struct f2fs_summary
);
874 memcpy(addr
, sum
, sizeof(struct f2fs_summary
));
878 * Calculate the number of current summary pages for writing
880 int npages_for_summary_flush(struct f2fs_sb_info
*sbi
, bool for_ra
)
882 int valid_sum_count
= 0;
885 for (i
= CURSEG_HOT_DATA
; i
<= CURSEG_COLD_DATA
; i
++) {
886 if (sbi
->ckpt
->alloc_type
[i
] == SSR
)
887 valid_sum_count
+= sbi
->blocks_per_seg
;
890 valid_sum_count
+= le16_to_cpu(
891 F2FS_CKPT(sbi
)->cur_data_blkoff
[i
]);
893 valid_sum_count
+= curseg_blkoff(sbi
, i
);
897 sum_in_page
= (PAGE_SIZE
- 2 * SUM_JOURNAL_SIZE
-
898 SUM_FOOTER_SIZE
) / SUMMARY_SIZE
;
899 if (valid_sum_count
<= sum_in_page
)
901 else if ((valid_sum_count
- sum_in_page
) <=
902 (PAGE_SIZE
- SUM_FOOTER_SIZE
) / SUMMARY_SIZE
)
908 * Caller should put this summary page
910 struct page
*get_sum_page(struct f2fs_sb_info
*sbi
, unsigned int segno
)
912 return get_meta_page(sbi
, GET_SUM_BLOCK(sbi
, segno
));
915 void update_meta_page(struct f2fs_sb_info
*sbi
, void *src
, block_t blk_addr
)
917 struct page
*page
= grab_meta_page(sbi
, blk_addr
);
918 void *dst
= page_address(page
);
921 memcpy(dst
, src
, PAGE_SIZE
);
923 memset(dst
, 0, PAGE_SIZE
);
924 set_page_dirty(page
);
925 f2fs_put_page(page
, 1);
928 static void write_sum_page(struct f2fs_sb_info
*sbi
,
929 struct f2fs_summary_block
*sum_blk
, block_t blk_addr
)
931 update_meta_page(sbi
, (void *)sum_blk
, blk_addr
);
934 static void write_current_sum_page(struct f2fs_sb_info
*sbi
,
935 int type
, block_t blk_addr
)
937 struct curseg_info
*curseg
= CURSEG_I(sbi
, type
);
938 struct page
*page
= grab_meta_page(sbi
, blk_addr
);
939 struct f2fs_summary_block
*src
= curseg
->sum_blk
;
940 struct f2fs_summary_block
*dst
;
942 dst
= (struct f2fs_summary_block
*)page_address(page
);
944 mutex_lock(&curseg
->curseg_mutex
);
946 down_read(&curseg
->journal_rwsem
);
947 memcpy(&dst
->journal
, curseg
->journal
, SUM_JOURNAL_SIZE
);
948 up_read(&curseg
->journal_rwsem
);
950 memcpy(dst
->entries
, src
->entries
, SUM_ENTRY_SIZE
);
951 memcpy(&dst
->footer
, &src
->footer
, SUM_FOOTER_SIZE
);
953 mutex_unlock(&curseg
->curseg_mutex
);
955 set_page_dirty(page
);
956 f2fs_put_page(page
, 1);
959 static int is_next_segment_free(struct f2fs_sb_info
*sbi
, int type
)
961 struct curseg_info
*curseg
= CURSEG_I(sbi
, type
);
962 unsigned int segno
= curseg
->segno
+ 1;
963 struct free_segmap_info
*free_i
= FREE_I(sbi
);
965 if (segno
< MAIN_SEGS(sbi
) && segno
% sbi
->segs_per_sec
)
966 return !test_bit(segno
, free_i
->free_segmap
);
971 * Find a new segment from the free segments bitmap to right order
972 * This function should be returned with success, otherwise BUG
974 static void get_new_segment(struct f2fs_sb_info
*sbi
,
975 unsigned int *newseg
, bool new_sec
, int dir
)
977 struct free_segmap_info
*free_i
= FREE_I(sbi
);
978 unsigned int segno
, secno
, zoneno
;
979 unsigned int total_zones
= MAIN_SECS(sbi
) / sbi
->secs_per_zone
;
980 unsigned int hint
= *newseg
/ sbi
->segs_per_sec
;
981 unsigned int old_zoneno
= GET_ZONENO_FROM_SEGNO(sbi
, *newseg
);
982 unsigned int left_start
= hint
;
987 spin_lock(&free_i
->segmap_lock
);
989 if (!new_sec
&& ((*newseg
+ 1) % sbi
->segs_per_sec
)) {
990 segno
= find_next_zero_bit(free_i
->free_segmap
,
991 (hint
+ 1) * sbi
->segs_per_sec
, *newseg
+ 1);
992 if (segno
< (hint
+ 1) * sbi
->segs_per_sec
)
996 secno
= find_next_zero_bit(free_i
->free_secmap
, MAIN_SECS(sbi
), hint
);
997 if (secno
>= MAIN_SECS(sbi
)) {
998 if (dir
== ALLOC_RIGHT
) {
999 secno
= find_next_zero_bit(free_i
->free_secmap
,
1001 f2fs_bug_on(sbi
, secno
>= MAIN_SECS(sbi
));
1004 left_start
= hint
- 1;
1010 while (test_bit(left_start
, free_i
->free_secmap
)) {
1011 if (left_start
> 0) {
1015 left_start
= find_next_zero_bit(free_i
->free_secmap
,
1017 f2fs_bug_on(sbi
, left_start
>= MAIN_SECS(sbi
));
1023 segno
= secno
* sbi
->segs_per_sec
;
1024 zoneno
= secno
/ sbi
->secs_per_zone
;
1026 /* give up on finding another zone */
1029 if (sbi
->secs_per_zone
== 1)
1031 if (zoneno
== old_zoneno
)
1033 if (dir
== ALLOC_LEFT
) {
1034 if (!go_left
&& zoneno
+ 1 >= total_zones
)
1036 if (go_left
&& zoneno
== 0)
1039 for (i
= 0; i
< NR_CURSEG_TYPE
; i
++)
1040 if (CURSEG_I(sbi
, i
)->zone
== zoneno
)
1043 if (i
< NR_CURSEG_TYPE
) {
1044 /* zone is in user, try another */
1046 hint
= zoneno
* sbi
->secs_per_zone
- 1;
1047 else if (zoneno
+ 1 >= total_zones
)
1050 hint
= (zoneno
+ 1) * sbi
->secs_per_zone
;
1052 goto find_other_zone
;
1055 /* set it as dirty segment in free segmap */
1056 f2fs_bug_on(sbi
, test_bit(segno
, free_i
->free_segmap
));
1057 __set_inuse(sbi
, segno
);
1059 spin_unlock(&free_i
->segmap_lock
);
1062 static void reset_curseg(struct f2fs_sb_info
*sbi
, int type
, int modified
)
1064 struct curseg_info
*curseg
= CURSEG_I(sbi
, type
);
1065 struct summary_footer
*sum_footer
;
1067 curseg
->segno
= curseg
->next_segno
;
1068 curseg
->zone
= GET_ZONENO_FROM_SEGNO(sbi
, curseg
->segno
);
1069 curseg
->next_blkoff
= 0;
1070 curseg
->next_segno
= NULL_SEGNO
;
1072 sum_footer
= &(curseg
->sum_blk
->footer
);
1073 memset(sum_footer
, 0, sizeof(struct summary_footer
));
1074 if (IS_DATASEG(type
))
1075 SET_SUM_TYPE(sum_footer
, SUM_TYPE_DATA
);
1076 if (IS_NODESEG(type
))
1077 SET_SUM_TYPE(sum_footer
, SUM_TYPE_NODE
);
1078 __set_sit_entry_type(sbi
, type
, curseg
->segno
, modified
);
1082 * Allocate a current working segment.
1083 * This function always allocates a free segment in LFS manner.
1085 static void new_curseg(struct f2fs_sb_info
*sbi
, int type
, bool new_sec
)
1087 struct curseg_info
*curseg
= CURSEG_I(sbi
, type
);
1088 unsigned int segno
= curseg
->segno
;
1089 int dir
= ALLOC_LEFT
;
1091 write_sum_page(sbi
, curseg
->sum_blk
,
1092 GET_SUM_BLOCK(sbi
, segno
));
1093 if (type
== CURSEG_WARM_DATA
|| type
== CURSEG_COLD_DATA
)
1096 if (test_opt(sbi
, NOHEAP
))
1099 get_new_segment(sbi
, &segno
, new_sec
, dir
);
1100 curseg
->next_segno
= segno
;
1101 reset_curseg(sbi
, type
, 1);
1102 curseg
->alloc_type
= LFS
;
1105 static void __next_free_blkoff(struct f2fs_sb_info
*sbi
,
1106 struct curseg_info
*seg
, block_t start
)
1108 struct seg_entry
*se
= get_seg_entry(sbi
, seg
->segno
);
1109 int entries
= SIT_VBLOCK_MAP_SIZE
/ sizeof(unsigned long);
1110 unsigned long *target_map
= SIT_I(sbi
)->tmp_map
;
1111 unsigned long *ckpt_map
= (unsigned long *)se
->ckpt_valid_map
;
1112 unsigned long *cur_map
= (unsigned long *)se
->cur_valid_map
;
1115 for (i
= 0; i
< entries
; i
++)
1116 target_map
[i
] = ckpt_map
[i
] | cur_map
[i
];
1118 pos
= __find_rev_next_zero_bit(target_map
, sbi
->blocks_per_seg
, start
);
1120 seg
->next_blkoff
= pos
;
1124 * If a segment is written by LFS manner, next block offset is just obtained
1125 * by increasing the current block offset. However, if a segment is written by
1126 * SSR manner, next block offset obtained by calling __next_free_blkoff
1128 static void __refresh_next_blkoff(struct f2fs_sb_info
*sbi
,
1129 struct curseg_info
*seg
)
1131 if (seg
->alloc_type
== SSR
)
1132 __next_free_blkoff(sbi
, seg
, seg
->next_blkoff
+ 1);
1138 * This function always allocates a used segment(from dirty seglist) by SSR
1139 * manner, so it should recover the existing segment information of valid blocks
1141 static void change_curseg(struct f2fs_sb_info
*sbi
, int type
, bool reuse
)
1143 struct dirty_seglist_info
*dirty_i
= DIRTY_I(sbi
);
1144 struct curseg_info
*curseg
= CURSEG_I(sbi
, type
);
1145 unsigned int new_segno
= curseg
->next_segno
;
1146 struct f2fs_summary_block
*sum_node
;
1147 struct page
*sum_page
;
1149 write_sum_page(sbi
, curseg
->sum_blk
,
1150 GET_SUM_BLOCK(sbi
, curseg
->segno
));
1151 __set_test_and_inuse(sbi
, new_segno
);
1153 mutex_lock(&dirty_i
->seglist_lock
);
1154 __remove_dirty_segment(sbi
, new_segno
, PRE
);
1155 __remove_dirty_segment(sbi
, new_segno
, DIRTY
);
1156 mutex_unlock(&dirty_i
->seglist_lock
);
1158 reset_curseg(sbi
, type
, 1);
1159 curseg
->alloc_type
= SSR
;
1160 __next_free_blkoff(sbi
, curseg
, 0);
1163 sum_page
= get_sum_page(sbi
, new_segno
);
1164 sum_node
= (struct f2fs_summary_block
*)page_address(sum_page
);
1165 memcpy(curseg
->sum_blk
, sum_node
, SUM_ENTRY_SIZE
);
1166 f2fs_put_page(sum_page
, 1);
1170 static int get_ssr_segment(struct f2fs_sb_info
*sbi
, int type
)
1172 struct curseg_info
*curseg
= CURSEG_I(sbi
, type
);
1173 const struct victim_selection
*v_ops
= DIRTY_I(sbi
)->v_ops
;
1175 if (IS_NODESEG(type
) || !has_not_enough_free_secs(sbi
, 0))
1176 return v_ops
->get_victim(sbi
,
1177 &(curseg
)->next_segno
, BG_GC
, type
, SSR
);
1179 /* For data segments, let's do SSR more intensively */
1180 for (; type
>= CURSEG_HOT_DATA
; type
--)
1181 if (v_ops
->get_victim(sbi
, &(curseg
)->next_segno
,
1188 * flush out current segment and replace it with new segment
1189 * This function should be returned with success, otherwise BUG
1191 static void allocate_segment_by_default(struct f2fs_sb_info
*sbi
,
1192 int type
, bool force
)
1194 struct curseg_info
*curseg
= CURSEG_I(sbi
, type
);
1197 new_curseg(sbi
, type
, true);
1198 else if (type
== CURSEG_WARM_NODE
)
1199 new_curseg(sbi
, type
, false);
1200 else if (curseg
->alloc_type
== LFS
&& is_next_segment_free(sbi
, type
))
1201 new_curseg(sbi
, type
, false);
1202 else if (need_SSR(sbi
) && get_ssr_segment(sbi
, type
))
1203 change_curseg(sbi
, type
, true);
1205 new_curseg(sbi
, type
, false);
1207 stat_inc_seg_type(sbi
, curseg
);
1210 static void __allocate_new_segments(struct f2fs_sb_info
*sbi
, int type
)
1212 struct curseg_info
*curseg
= CURSEG_I(sbi
, type
);
1213 unsigned int old_segno
;
1215 old_segno
= curseg
->segno
;
1216 SIT_I(sbi
)->s_ops
->allocate_segment(sbi
, type
, true);
1217 locate_dirty_segment(sbi
, old_segno
);
1220 void allocate_new_segments(struct f2fs_sb_info
*sbi
)
1224 for (i
= CURSEG_HOT_DATA
; i
<= CURSEG_COLD_DATA
; i
++)
1225 __allocate_new_segments(sbi
, i
);
1228 static const struct segment_allocation default_salloc_ops
= {
1229 .allocate_segment
= allocate_segment_by_default
,
1232 int f2fs_trim_fs(struct f2fs_sb_info
*sbi
, struct fstrim_range
*range
)
1234 __u64 start
= F2FS_BYTES_TO_BLK(range
->start
);
1235 __u64 end
= start
+ F2FS_BYTES_TO_BLK(range
->len
) - 1;
1236 unsigned int start_segno
, end_segno
;
1237 struct cp_control cpc
;
1240 if (start
>= MAX_BLKADDR(sbi
) || range
->len
< sbi
->blocksize
)
1244 if (end
<= MAIN_BLKADDR(sbi
))
1247 /* start/end segment number in main_area */
1248 start_segno
= (start
<= MAIN_BLKADDR(sbi
)) ? 0 : GET_SEGNO(sbi
, start
);
1249 end_segno
= (end
>= MAX_BLKADDR(sbi
)) ? MAIN_SEGS(sbi
) - 1 :
1250 GET_SEGNO(sbi
, end
);
1251 cpc
.reason
= CP_DISCARD
;
1252 cpc
.trim_minlen
= max_t(__u64
, 1, F2FS_BYTES_TO_BLK(range
->minlen
));
1254 /* do checkpoint to issue discard commands safely */
1255 for (; start_segno
<= end_segno
; start_segno
= cpc
.trim_end
+ 1) {
1256 cpc
.trim_start
= start_segno
;
1258 if (sbi
->discard_blks
== 0)
1260 else if (sbi
->discard_blks
< BATCHED_TRIM_BLOCKS(sbi
))
1261 cpc
.trim_end
= end_segno
;
1263 cpc
.trim_end
= min_t(unsigned int,
1264 rounddown(start_segno
+
1265 BATCHED_TRIM_SEGMENTS(sbi
),
1266 sbi
->segs_per_sec
) - 1, end_segno
);
1268 mutex_lock(&sbi
->gc_mutex
);
1269 err
= write_checkpoint(sbi
, &cpc
);
1270 mutex_unlock(&sbi
->gc_mutex
);
1273 range
->len
= F2FS_BLK_TO_BYTES(cpc
.trimmed
);
1277 static bool __has_curseg_space(struct f2fs_sb_info
*sbi
, int type
)
1279 struct curseg_info
*curseg
= CURSEG_I(sbi
, type
);
1280 if (curseg
->next_blkoff
< sbi
->blocks_per_seg
)
1285 static int __get_segment_type_2(struct page
*page
, enum page_type p_type
)
1288 return CURSEG_HOT_DATA
;
1290 return CURSEG_HOT_NODE
;
1293 static int __get_segment_type_4(struct page
*page
, enum page_type p_type
)
1295 if (p_type
== DATA
) {
1296 struct inode
*inode
= page
->mapping
->host
;
1298 if (S_ISDIR(inode
->i_mode
))
1299 return CURSEG_HOT_DATA
;
1301 return CURSEG_COLD_DATA
;
1303 if (IS_DNODE(page
) && is_cold_node(page
))
1304 return CURSEG_WARM_NODE
;
1306 return CURSEG_COLD_NODE
;
1310 static int __get_segment_type_6(struct page
*page
, enum page_type p_type
)
1312 if (p_type
== DATA
) {
1313 struct inode
*inode
= page
->mapping
->host
;
1315 if (S_ISDIR(inode
->i_mode
))
1316 return CURSEG_HOT_DATA
;
1317 else if (is_cold_data(page
) || file_is_cold(inode
))
1318 return CURSEG_COLD_DATA
;
1320 return CURSEG_WARM_DATA
;
1323 return is_cold_node(page
) ? CURSEG_WARM_NODE
:
1326 return CURSEG_COLD_NODE
;
1330 static int __get_segment_type(struct page
*page
, enum page_type p_type
)
1332 switch (F2FS_P_SB(page
)->active_logs
) {
1334 return __get_segment_type_2(page
, p_type
);
1336 return __get_segment_type_4(page
, p_type
);
1338 /* NR_CURSEG_TYPE(6) logs by default */
1339 f2fs_bug_on(F2FS_P_SB(page
),
1340 F2FS_P_SB(page
)->active_logs
!= NR_CURSEG_TYPE
);
1341 return __get_segment_type_6(page
, p_type
);
1344 void allocate_data_block(struct f2fs_sb_info
*sbi
, struct page
*page
,
1345 block_t old_blkaddr
, block_t
*new_blkaddr
,
1346 struct f2fs_summary
*sum
, int type
)
1348 struct sit_info
*sit_i
= SIT_I(sbi
);
1349 struct curseg_info
*curseg
;
1350 bool direct_io
= (type
== CURSEG_DIRECT_IO
);
1352 type
= direct_io
? CURSEG_WARM_DATA
: type
;
1354 curseg
= CURSEG_I(sbi
, type
);
1356 mutex_lock(&curseg
->curseg_mutex
);
1357 mutex_lock(&sit_i
->sentry_lock
);
1359 /* direct_io'ed data is aligned to the segment for better performance */
1360 if (direct_io
&& curseg
->next_blkoff
&&
1361 !has_not_enough_free_secs(sbi
, 0))
1362 __allocate_new_segments(sbi
, type
);
1364 *new_blkaddr
= NEXT_FREE_BLKADDR(sbi
, curseg
);
1367 * __add_sum_entry should be resided under the curseg_mutex
1368 * because, this function updates a summary entry in the
1369 * current summary block.
1371 __add_sum_entry(sbi
, type
, sum
);
1373 __refresh_next_blkoff(sbi
, curseg
);
1375 stat_inc_block_count(sbi
, curseg
);
1377 if (!__has_curseg_space(sbi
, type
))
1378 sit_i
->s_ops
->allocate_segment(sbi
, type
, false);
1380 * SIT information should be updated before segment allocation,
1381 * since SSR needs latest valid block information.
1383 refresh_sit_entry(sbi
, old_blkaddr
, *new_blkaddr
);
1385 mutex_unlock(&sit_i
->sentry_lock
);
1387 if (page
&& IS_NODESEG(type
))
1388 fill_node_footer_blkaddr(page
, NEXT_FREE_BLKADDR(sbi
, curseg
));
1390 mutex_unlock(&curseg
->curseg_mutex
);
1393 static void do_write_page(struct f2fs_summary
*sum
, struct f2fs_io_info
*fio
)
1395 int type
= __get_segment_type(fio
->page
, fio
->type
);
1397 allocate_data_block(fio
->sbi
, fio
->page
, fio
->old_blkaddr
,
1398 &fio
->new_blkaddr
, sum
, type
);
1400 /* writeout dirty page into bdev */
1401 f2fs_submit_page_mbio(fio
);
1404 void write_meta_page(struct f2fs_sb_info
*sbi
, struct page
*page
)
1406 struct f2fs_io_info fio
= {
1409 .rw
= WRITE_SYNC
| REQ_META
| REQ_PRIO
,
1410 .old_blkaddr
= page
->index
,
1411 .new_blkaddr
= page
->index
,
1413 .encrypted_page
= NULL
,
1416 if (unlikely(page
->index
>= MAIN_BLKADDR(sbi
)))
1417 fio
.rw
&= ~REQ_META
;
1419 set_page_writeback(page
);
1420 f2fs_submit_page_mbio(&fio
);
1423 void write_node_page(unsigned int nid
, struct f2fs_io_info
*fio
)
1425 struct f2fs_summary sum
;
1427 set_summary(&sum
, nid
, 0, 0);
1428 do_write_page(&sum
, fio
);
1431 void write_data_page(struct dnode_of_data
*dn
, struct f2fs_io_info
*fio
)
1433 struct f2fs_sb_info
*sbi
= fio
->sbi
;
1434 struct f2fs_summary sum
;
1435 struct node_info ni
;
1437 f2fs_bug_on(sbi
, dn
->data_blkaddr
== NULL_ADDR
);
1438 get_node_info(sbi
, dn
->nid
, &ni
);
1439 set_summary(&sum
, dn
->nid
, dn
->ofs_in_node
, ni
.version
);
1440 do_write_page(&sum
, fio
);
1441 f2fs_update_data_blkaddr(dn
, fio
->new_blkaddr
);
1444 void rewrite_data_page(struct f2fs_io_info
*fio
)
1446 fio
->new_blkaddr
= fio
->old_blkaddr
;
1447 stat_inc_inplace_blocks(fio
->sbi
);
1448 f2fs_submit_page_mbio(fio
);
1451 void __f2fs_replace_block(struct f2fs_sb_info
*sbi
, struct f2fs_summary
*sum
,
1452 block_t old_blkaddr
, block_t new_blkaddr
,
1453 bool recover_curseg
, bool recover_newaddr
)
1455 struct sit_info
*sit_i
= SIT_I(sbi
);
1456 struct curseg_info
*curseg
;
1457 unsigned int segno
, old_cursegno
;
1458 struct seg_entry
*se
;
1460 unsigned short old_blkoff
;
1462 segno
= GET_SEGNO(sbi
, new_blkaddr
);
1463 se
= get_seg_entry(sbi
, segno
);
1466 if (!recover_curseg
) {
1467 /* for recovery flow */
1468 if (se
->valid_blocks
== 0 && !IS_CURSEG(sbi
, segno
)) {
1469 if (old_blkaddr
== NULL_ADDR
)
1470 type
= CURSEG_COLD_DATA
;
1472 type
= CURSEG_WARM_DATA
;
1475 if (!IS_CURSEG(sbi
, segno
))
1476 type
= CURSEG_WARM_DATA
;
1479 curseg
= CURSEG_I(sbi
, type
);
1481 mutex_lock(&curseg
->curseg_mutex
);
1482 mutex_lock(&sit_i
->sentry_lock
);
1484 old_cursegno
= curseg
->segno
;
1485 old_blkoff
= curseg
->next_blkoff
;
1487 /* change the current segment */
1488 if (segno
!= curseg
->segno
) {
1489 curseg
->next_segno
= segno
;
1490 change_curseg(sbi
, type
, true);
1493 curseg
->next_blkoff
= GET_BLKOFF_FROM_SEG0(sbi
, new_blkaddr
);
1494 __add_sum_entry(sbi
, type
, sum
);
1496 if (!recover_curseg
|| recover_newaddr
)
1497 update_sit_entry(sbi
, new_blkaddr
, 1);
1498 if (GET_SEGNO(sbi
, old_blkaddr
) != NULL_SEGNO
)
1499 update_sit_entry(sbi
, old_blkaddr
, -1);
1501 locate_dirty_segment(sbi
, GET_SEGNO(sbi
, old_blkaddr
));
1502 locate_dirty_segment(sbi
, GET_SEGNO(sbi
, new_blkaddr
));
1504 locate_dirty_segment(sbi
, old_cursegno
);
1506 if (recover_curseg
) {
1507 if (old_cursegno
!= curseg
->segno
) {
1508 curseg
->next_segno
= old_cursegno
;
1509 change_curseg(sbi
, type
, true);
1511 curseg
->next_blkoff
= old_blkoff
;
1514 mutex_unlock(&sit_i
->sentry_lock
);
1515 mutex_unlock(&curseg
->curseg_mutex
);
1518 void f2fs_replace_block(struct f2fs_sb_info
*sbi
, struct dnode_of_data
*dn
,
1519 block_t old_addr
, block_t new_addr
,
1520 unsigned char version
, bool recover_curseg
,
1521 bool recover_newaddr
)
1523 struct f2fs_summary sum
;
1525 set_summary(&sum
, dn
->nid
, dn
->ofs_in_node
, version
);
1527 __f2fs_replace_block(sbi
, &sum
, old_addr
, new_addr
,
1528 recover_curseg
, recover_newaddr
);
1530 f2fs_update_data_blkaddr(dn
, new_addr
);
1533 void f2fs_wait_on_page_writeback(struct page
*page
,
1534 enum page_type type
, bool ordered
)
1536 if (PageWriteback(page
)) {
1537 struct f2fs_sb_info
*sbi
= F2FS_P_SB(page
);
1539 f2fs_submit_merged_bio_cond(sbi
, NULL
, page
, 0, type
, WRITE
);
1541 wait_on_page_writeback(page
);
1543 wait_for_stable_page(page
);
1547 void f2fs_wait_on_encrypted_page_writeback(struct f2fs_sb_info
*sbi
,
1552 if (blkaddr
== NEW_ADDR
)
1555 f2fs_bug_on(sbi
, blkaddr
== NULL_ADDR
);
1557 cpage
= find_lock_page(META_MAPPING(sbi
), blkaddr
);
1559 f2fs_wait_on_page_writeback(cpage
, DATA
, true);
1560 f2fs_put_page(cpage
, 1);
1564 static int read_compacted_summaries(struct f2fs_sb_info
*sbi
)
1566 struct f2fs_checkpoint
*ckpt
= F2FS_CKPT(sbi
);
1567 struct curseg_info
*seg_i
;
1568 unsigned char *kaddr
;
1573 start
= start_sum_block(sbi
);
1575 page
= get_meta_page(sbi
, start
++);
1576 kaddr
= (unsigned char *)page_address(page
);
1578 /* Step 1: restore nat cache */
1579 seg_i
= CURSEG_I(sbi
, CURSEG_HOT_DATA
);
1580 memcpy(seg_i
->journal
, kaddr
, SUM_JOURNAL_SIZE
);
1582 /* Step 2: restore sit cache */
1583 seg_i
= CURSEG_I(sbi
, CURSEG_COLD_DATA
);
1584 memcpy(seg_i
->journal
, kaddr
+ SUM_JOURNAL_SIZE
, SUM_JOURNAL_SIZE
);
1585 offset
= 2 * SUM_JOURNAL_SIZE
;
1587 /* Step 3: restore summary entries */
1588 for (i
= CURSEG_HOT_DATA
; i
<= CURSEG_COLD_DATA
; i
++) {
1589 unsigned short blk_off
;
1592 seg_i
= CURSEG_I(sbi
, i
);
1593 segno
= le32_to_cpu(ckpt
->cur_data_segno
[i
]);
1594 blk_off
= le16_to_cpu(ckpt
->cur_data_blkoff
[i
]);
1595 seg_i
->next_segno
= segno
;
1596 reset_curseg(sbi
, i
, 0);
1597 seg_i
->alloc_type
= ckpt
->alloc_type
[i
];
1598 seg_i
->next_blkoff
= blk_off
;
1600 if (seg_i
->alloc_type
== SSR
)
1601 blk_off
= sbi
->blocks_per_seg
;
1603 for (j
= 0; j
< blk_off
; j
++) {
1604 struct f2fs_summary
*s
;
1605 s
= (struct f2fs_summary
*)(kaddr
+ offset
);
1606 seg_i
->sum_blk
->entries
[j
] = *s
;
1607 offset
+= SUMMARY_SIZE
;
1608 if (offset
+ SUMMARY_SIZE
<= PAGE_SIZE
-
1612 f2fs_put_page(page
, 1);
1615 page
= get_meta_page(sbi
, start
++);
1616 kaddr
= (unsigned char *)page_address(page
);
1620 f2fs_put_page(page
, 1);
1624 static int read_normal_summaries(struct f2fs_sb_info
*sbi
, int type
)
1626 struct f2fs_checkpoint
*ckpt
= F2FS_CKPT(sbi
);
1627 struct f2fs_summary_block
*sum
;
1628 struct curseg_info
*curseg
;
1630 unsigned short blk_off
;
1631 unsigned int segno
= 0;
1632 block_t blk_addr
= 0;
1634 /* get segment number and block addr */
1635 if (IS_DATASEG(type
)) {
1636 segno
= le32_to_cpu(ckpt
->cur_data_segno
[type
]);
1637 blk_off
= le16_to_cpu(ckpt
->cur_data_blkoff
[type
-
1639 if (__exist_node_summaries(sbi
))
1640 blk_addr
= sum_blk_addr(sbi
, NR_CURSEG_TYPE
, type
);
1642 blk_addr
= sum_blk_addr(sbi
, NR_CURSEG_DATA_TYPE
, type
);
1644 segno
= le32_to_cpu(ckpt
->cur_node_segno
[type
-
1646 blk_off
= le16_to_cpu(ckpt
->cur_node_blkoff
[type
-
1648 if (__exist_node_summaries(sbi
))
1649 blk_addr
= sum_blk_addr(sbi
, NR_CURSEG_NODE_TYPE
,
1650 type
- CURSEG_HOT_NODE
);
1652 blk_addr
= GET_SUM_BLOCK(sbi
, segno
);
1655 new = get_meta_page(sbi
, blk_addr
);
1656 sum
= (struct f2fs_summary_block
*)page_address(new);
1658 if (IS_NODESEG(type
)) {
1659 if (__exist_node_summaries(sbi
)) {
1660 struct f2fs_summary
*ns
= &sum
->entries
[0];
1662 for (i
= 0; i
< sbi
->blocks_per_seg
; i
++, ns
++) {
1664 ns
->ofs_in_node
= 0;
1669 err
= restore_node_summary(sbi
, segno
, sum
);
1671 f2fs_put_page(new, 1);
1677 /* set uncompleted segment to curseg */
1678 curseg
= CURSEG_I(sbi
, type
);
1679 mutex_lock(&curseg
->curseg_mutex
);
1681 /* update journal info */
1682 down_write(&curseg
->journal_rwsem
);
1683 memcpy(curseg
->journal
, &sum
->journal
, SUM_JOURNAL_SIZE
);
1684 up_write(&curseg
->journal_rwsem
);
1686 memcpy(curseg
->sum_blk
->entries
, sum
->entries
, SUM_ENTRY_SIZE
);
1687 memcpy(&curseg
->sum_blk
->footer
, &sum
->footer
, SUM_FOOTER_SIZE
);
1688 curseg
->next_segno
= segno
;
1689 reset_curseg(sbi
, type
, 0);
1690 curseg
->alloc_type
= ckpt
->alloc_type
[type
];
1691 curseg
->next_blkoff
= blk_off
;
1692 mutex_unlock(&curseg
->curseg_mutex
);
1693 f2fs_put_page(new, 1);
1697 static int restore_curseg_summaries(struct f2fs_sb_info
*sbi
)
1699 int type
= CURSEG_HOT_DATA
;
1702 if (is_set_ckpt_flags(F2FS_CKPT(sbi
), CP_COMPACT_SUM_FLAG
)) {
1703 int npages
= npages_for_summary_flush(sbi
, true);
1706 ra_meta_pages(sbi
, start_sum_block(sbi
), npages
,
1709 /* restore for compacted data summary */
1710 if (read_compacted_summaries(sbi
))
1712 type
= CURSEG_HOT_NODE
;
1715 if (__exist_node_summaries(sbi
))
1716 ra_meta_pages(sbi
, sum_blk_addr(sbi
, NR_CURSEG_TYPE
, type
),
1717 NR_CURSEG_TYPE
- type
, META_CP
, true);
1719 for (; type
<= CURSEG_COLD_NODE
; type
++) {
1720 err
= read_normal_summaries(sbi
, type
);
1728 static void write_compacted_summaries(struct f2fs_sb_info
*sbi
, block_t blkaddr
)
1731 unsigned char *kaddr
;
1732 struct f2fs_summary
*summary
;
1733 struct curseg_info
*seg_i
;
1734 int written_size
= 0;
1737 page
= grab_meta_page(sbi
, blkaddr
++);
1738 kaddr
= (unsigned char *)page_address(page
);
1740 /* Step 1: write nat cache */
1741 seg_i
= CURSEG_I(sbi
, CURSEG_HOT_DATA
);
1742 memcpy(kaddr
, seg_i
->journal
, SUM_JOURNAL_SIZE
);
1743 written_size
+= SUM_JOURNAL_SIZE
;
1745 /* Step 2: write sit cache */
1746 seg_i
= CURSEG_I(sbi
, CURSEG_COLD_DATA
);
1747 memcpy(kaddr
+ written_size
, seg_i
->journal
, SUM_JOURNAL_SIZE
);
1748 written_size
+= SUM_JOURNAL_SIZE
;
1750 /* Step 3: write summary entries */
1751 for (i
= CURSEG_HOT_DATA
; i
<= CURSEG_COLD_DATA
; i
++) {
1752 unsigned short blkoff
;
1753 seg_i
= CURSEG_I(sbi
, i
);
1754 if (sbi
->ckpt
->alloc_type
[i
] == SSR
)
1755 blkoff
= sbi
->blocks_per_seg
;
1757 blkoff
= curseg_blkoff(sbi
, i
);
1759 for (j
= 0; j
< blkoff
; j
++) {
1761 page
= grab_meta_page(sbi
, blkaddr
++);
1762 kaddr
= (unsigned char *)page_address(page
);
1765 summary
= (struct f2fs_summary
*)(kaddr
+ written_size
);
1766 *summary
= seg_i
->sum_blk
->entries
[j
];
1767 written_size
+= SUMMARY_SIZE
;
1769 if (written_size
+ SUMMARY_SIZE
<= PAGE_SIZE
-
1773 set_page_dirty(page
);
1774 f2fs_put_page(page
, 1);
1779 set_page_dirty(page
);
1780 f2fs_put_page(page
, 1);
1784 static void write_normal_summaries(struct f2fs_sb_info
*sbi
,
1785 block_t blkaddr
, int type
)
1788 if (IS_DATASEG(type
))
1789 end
= type
+ NR_CURSEG_DATA_TYPE
;
1791 end
= type
+ NR_CURSEG_NODE_TYPE
;
1793 for (i
= type
; i
< end
; i
++)
1794 write_current_sum_page(sbi
, i
, blkaddr
+ (i
- type
));
1797 void write_data_summaries(struct f2fs_sb_info
*sbi
, block_t start_blk
)
1799 if (is_set_ckpt_flags(F2FS_CKPT(sbi
), CP_COMPACT_SUM_FLAG
))
1800 write_compacted_summaries(sbi
, start_blk
);
1802 write_normal_summaries(sbi
, start_blk
, CURSEG_HOT_DATA
);
1805 void write_node_summaries(struct f2fs_sb_info
*sbi
, block_t start_blk
)
1807 write_normal_summaries(sbi
, start_blk
, CURSEG_HOT_NODE
);
1810 int lookup_journal_in_cursum(struct f2fs_journal
*journal
, int type
,
1811 unsigned int val
, int alloc
)
1815 if (type
== NAT_JOURNAL
) {
1816 for (i
= 0; i
< nats_in_cursum(journal
); i
++) {
1817 if (le32_to_cpu(nid_in_journal(journal
, i
)) == val
)
1820 if (alloc
&& __has_cursum_space(journal
, 1, NAT_JOURNAL
))
1821 return update_nats_in_cursum(journal
, 1);
1822 } else if (type
== SIT_JOURNAL
) {
1823 for (i
= 0; i
< sits_in_cursum(journal
); i
++)
1824 if (le32_to_cpu(segno_in_journal(journal
, i
)) == val
)
1826 if (alloc
&& __has_cursum_space(journal
, 1, SIT_JOURNAL
))
1827 return update_sits_in_cursum(journal
, 1);
1832 static struct page
*get_current_sit_page(struct f2fs_sb_info
*sbi
,
1835 return get_meta_page(sbi
, current_sit_addr(sbi
, segno
));
1838 static struct page
*get_next_sit_page(struct f2fs_sb_info
*sbi
,
1841 struct sit_info
*sit_i
= SIT_I(sbi
);
1842 struct page
*src_page
, *dst_page
;
1843 pgoff_t src_off
, dst_off
;
1844 void *src_addr
, *dst_addr
;
1846 src_off
= current_sit_addr(sbi
, start
);
1847 dst_off
= next_sit_addr(sbi
, src_off
);
1849 /* get current sit block page without lock */
1850 src_page
= get_meta_page(sbi
, src_off
);
1851 dst_page
= grab_meta_page(sbi
, dst_off
);
1852 f2fs_bug_on(sbi
, PageDirty(src_page
));
1854 src_addr
= page_address(src_page
);
1855 dst_addr
= page_address(dst_page
);
1856 memcpy(dst_addr
, src_addr
, PAGE_SIZE
);
1858 set_page_dirty(dst_page
);
1859 f2fs_put_page(src_page
, 1);
1861 set_to_next_sit(sit_i
, start
);
1866 static struct sit_entry_set
*grab_sit_entry_set(void)
1868 struct sit_entry_set
*ses
=
1869 f2fs_kmem_cache_alloc(sit_entry_set_slab
, GFP_NOFS
);
1872 INIT_LIST_HEAD(&ses
->set_list
);
1876 static void release_sit_entry_set(struct sit_entry_set
*ses
)
1878 list_del(&ses
->set_list
);
1879 kmem_cache_free(sit_entry_set_slab
, ses
);
1882 static void adjust_sit_entry_set(struct sit_entry_set
*ses
,
1883 struct list_head
*head
)
1885 struct sit_entry_set
*next
= ses
;
1887 if (list_is_last(&ses
->set_list
, head
))
1890 list_for_each_entry_continue(next
, head
, set_list
)
1891 if (ses
->entry_cnt
<= next
->entry_cnt
)
1894 list_move_tail(&ses
->set_list
, &next
->set_list
);
1897 static void add_sit_entry(unsigned int segno
, struct list_head
*head
)
1899 struct sit_entry_set
*ses
;
1900 unsigned int start_segno
= START_SEGNO(segno
);
1902 list_for_each_entry(ses
, head
, set_list
) {
1903 if (ses
->start_segno
== start_segno
) {
1905 adjust_sit_entry_set(ses
, head
);
1910 ses
= grab_sit_entry_set();
1912 ses
->start_segno
= start_segno
;
1914 list_add(&ses
->set_list
, head
);
1917 static void add_sits_in_set(struct f2fs_sb_info
*sbi
)
1919 struct f2fs_sm_info
*sm_info
= SM_I(sbi
);
1920 struct list_head
*set_list
= &sm_info
->sit_entry_set
;
1921 unsigned long *bitmap
= SIT_I(sbi
)->dirty_sentries_bitmap
;
1924 for_each_set_bit(segno
, bitmap
, MAIN_SEGS(sbi
))
1925 add_sit_entry(segno
, set_list
);
1928 static void remove_sits_in_journal(struct f2fs_sb_info
*sbi
)
1930 struct curseg_info
*curseg
= CURSEG_I(sbi
, CURSEG_COLD_DATA
);
1931 struct f2fs_journal
*journal
= curseg
->journal
;
1934 down_write(&curseg
->journal_rwsem
);
1935 for (i
= 0; i
< sits_in_cursum(journal
); i
++) {
1939 segno
= le32_to_cpu(segno_in_journal(journal
, i
));
1940 dirtied
= __mark_sit_entry_dirty(sbi
, segno
);
1943 add_sit_entry(segno
, &SM_I(sbi
)->sit_entry_set
);
1945 update_sits_in_cursum(journal
, -i
);
1946 up_write(&curseg
->journal_rwsem
);
1950 * CP calls this function, which flushes SIT entries including sit_journal,
1951 * and moves prefree segs to free segs.
1953 void flush_sit_entries(struct f2fs_sb_info
*sbi
, struct cp_control
*cpc
)
1955 struct sit_info
*sit_i
= SIT_I(sbi
);
1956 unsigned long *bitmap
= sit_i
->dirty_sentries_bitmap
;
1957 struct curseg_info
*curseg
= CURSEG_I(sbi
, CURSEG_COLD_DATA
);
1958 struct f2fs_journal
*journal
= curseg
->journal
;
1959 struct sit_entry_set
*ses
, *tmp
;
1960 struct list_head
*head
= &SM_I(sbi
)->sit_entry_set
;
1961 bool to_journal
= true;
1962 struct seg_entry
*se
;
1964 mutex_lock(&sit_i
->sentry_lock
);
1966 if (!sit_i
->dirty_sentries
)
1970 * add and account sit entries of dirty bitmap in sit entry
1973 add_sits_in_set(sbi
);
1976 * if there are no enough space in journal to store dirty sit
1977 * entries, remove all entries from journal and add and account
1978 * them in sit entry set.
1980 if (!__has_cursum_space(journal
, sit_i
->dirty_sentries
, SIT_JOURNAL
))
1981 remove_sits_in_journal(sbi
);
1984 * there are two steps to flush sit entries:
1985 * #1, flush sit entries to journal in current cold data summary block.
1986 * #2, flush sit entries to sit page.
1988 list_for_each_entry_safe(ses
, tmp
, head
, set_list
) {
1989 struct page
*page
= NULL
;
1990 struct f2fs_sit_block
*raw_sit
= NULL
;
1991 unsigned int start_segno
= ses
->start_segno
;
1992 unsigned int end
= min(start_segno
+ SIT_ENTRY_PER_BLOCK
,
1993 (unsigned long)MAIN_SEGS(sbi
));
1994 unsigned int segno
= start_segno
;
1997 !__has_cursum_space(journal
, ses
->entry_cnt
, SIT_JOURNAL
))
2001 down_write(&curseg
->journal_rwsem
);
2003 page
= get_next_sit_page(sbi
, start_segno
);
2004 raw_sit
= page_address(page
);
2007 /* flush dirty sit entries in region of current sit set */
2008 for_each_set_bit_from(segno
, bitmap
, end
) {
2009 int offset
, sit_offset
;
2011 se
= get_seg_entry(sbi
, segno
);
2013 /* add discard candidates */
2014 if (cpc
->reason
!= CP_DISCARD
) {
2015 cpc
->trim_start
= segno
;
2016 add_discard_addrs(sbi
, cpc
);
2020 offset
= lookup_journal_in_cursum(journal
,
2021 SIT_JOURNAL
, segno
, 1);
2022 f2fs_bug_on(sbi
, offset
< 0);
2023 segno_in_journal(journal
, offset
) =
2025 seg_info_to_raw_sit(se
,
2026 &sit_in_journal(journal
, offset
));
2028 sit_offset
= SIT_ENTRY_OFFSET(sit_i
, segno
);
2029 seg_info_to_raw_sit(se
,
2030 &raw_sit
->entries
[sit_offset
]);
2033 __clear_bit(segno
, bitmap
);
2034 sit_i
->dirty_sentries
--;
2039 up_write(&curseg
->journal_rwsem
);
2041 f2fs_put_page(page
, 1);
2043 f2fs_bug_on(sbi
, ses
->entry_cnt
);
2044 release_sit_entry_set(ses
);
2047 f2fs_bug_on(sbi
, !list_empty(head
));
2048 f2fs_bug_on(sbi
, sit_i
->dirty_sentries
);
2050 if (cpc
->reason
== CP_DISCARD
) {
2051 for (; cpc
->trim_start
<= cpc
->trim_end
; cpc
->trim_start
++)
2052 add_discard_addrs(sbi
, cpc
);
2054 mutex_unlock(&sit_i
->sentry_lock
);
2056 set_prefree_as_free_segments(sbi
);
2059 static int build_sit_info(struct f2fs_sb_info
*sbi
)
2061 struct f2fs_super_block
*raw_super
= F2FS_RAW_SUPER(sbi
);
2062 struct f2fs_checkpoint
*ckpt
= F2FS_CKPT(sbi
);
2063 struct sit_info
*sit_i
;
2064 unsigned int sit_segs
, start
;
2065 char *src_bitmap
, *dst_bitmap
;
2066 unsigned int bitmap_size
;
2068 /* allocate memory for SIT information */
2069 sit_i
= kzalloc(sizeof(struct sit_info
), GFP_KERNEL
);
2073 SM_I(sbi
)->sit_info
= sit_i
;
2075 sit_i
->sentries
= f2fs_kvzalloc(MAIN_SEGS(sbi
) *
2076 sizeof(struct seg_entry
), GFP_KERNEL
);
2077 if (!sit_i
->sentries
)
2080 bitmap_size
= f2fs_bitmap_size(MAIN_SEGS(sbi
));
2081 sit_i
->dirty_sentries_bitmap
= f2fs_kvzalloc(bitmap_size
, GFP_KERNEL
);
2082 if (!sit_i
->dirty_sentries_bitmap
)
2085 for (start
= 0; start
< MAIN_SEGS(sbi
); start
++) {
2086 sit_i
->sentries
[start
].cur_valid_map
2087 = kzalloc(SIT_VBLOCK_MAP_SIZE
, GFP_KERNEL
);
2088 sit_i
->sentries
[start
].ckpt_valid_map
2089 = kzalloc(SIT_VBLOCK_MAP_SIZE
, GFP_KERNEL
);
2090 sit_i
->sentries
[start
].discard_map
2091 = kzalloc(SIT_VBLOCK_MAP_SIZE
, GFP_KERNEL
);
2092 if (!sit_i
->sentries
[start
].cur_valid_map
||
2093 !sit_i
->sentries
[start
].ckpt_valid_map
||
2094 !sit_i
->sentries
[start
].discard_map
)
2098 sit_i
->tmp_map
= kzalloc(SIT_VBLOCK_MAP_SIZE
, GFP_KERNEL
);
2099 if (!sit_i
->tmp_map
)
2102 if (sbi
->segs_per_sec
> 1) {
2103 sit_i
->sec_entries
= f2fs_kvzalloc(MAIN_SECS(sbi
) *
2104 sizeof(struct sec_entry
), GFP_KERNEL
);
2105 if (!sit_i
->sec_entries
)
2109 /* get information related with SIT */
2110 sit_segs
= le32_to_cpu(raw_super
->segment_count_sit
) >> 1;
2112 /* setup SIT bitmap from ckeckpoint pack */
2113 bitmap_size
= __bitmap_size(sbi
, SIT_BITMAP
);
2114 src_bitmap
= __bitmap_ptr(sbi
, SIT_BITMAP
);
2116 dst_bitmap
= kmemdup(src_bitmap
, bitmap_size
, GFP_KERNEL
);
2120 /* init SIT information */
2121 sit_i
->s_ops
= &default_salloc_ops
;
2123 sit_i
->sit_base_addr
= le32_to_cpu(raw_super
->sit_blkaddr
);
2124 sit_i
->sit_blocks
= sit_segs
<< sbi
->log_blocks_per_seg
;
2125 sit_i
->written_valid_blocks
= le64_to_cpu(ckpt
->valid_block_count
);
2126 sit_i
->sit_bitmap
= dst_bitmap
;
2127 sit_i
->bitmap_size
= bitmap_size
;
2128 sit_i
->dirty_sentries
= 0;
2129 sit_i
->sents_per_block
= SIT_ENTRY_PER_BLOCK
;
2130 sit_i
->elapsed_time
= le64_to_cpu(sbi
->ckpt
->elapsed_time
);
2131 sit_i
->mounted_time
= CURRENT_TIME_SEC
.tv_sec
;
2132 mutex_init(&sit_i
->sentry_lock
);
2136 static int build_free_segmap(struct f2fs_sb_info
*sbi
)
2138 struct free_segmap_info
*free_i
;
2139 unsigned int bitmap_size
, sec_bitmap_size
;
2141 /* allocate memory for free segmap information */
2142 free_i
= kzalloc(sizeof(struct free_segmap_info
), GFP_KERNEL
);
2146 SM_I(sbi
)->free_info
= free_i
;
2148 bitmap_size
= f2fs_bitmap_size(MAIN_SEGS(sbi
));
2149 free_i
->free_segmap
= f2fs_kvmalloc(bitmap_size
, GFP_KERNEL
);
2150 if (!free_i
->free_segmap
)
2153 sec_bitmap_size
= f2fs_bitmap_size(MAIN_SECS(sbi
));
2154 free_i
->free_secmap
= f2fs_kvmalloc(sec_bitmap_size
, GFP_KERNEL
);
2155 if (!free_i
->free_secmap
)
2158 /* set all segments as dirty temporarily */
2159 memset(free_i
->free_segmap
, 0xff, bitmap_size
);
2160 memset(free_i
->free_secmap
, 0xff, sec_bitmap_size
);
2162 /* init free segmap information */
2163 free_i
->start_segno
= GET_SEGNO_FROM_SEG0(sbi
, MAIN_BLKADDR(sbi
));
2164 free_i
->free_segments
= 0;
2165 free_i
->free_sections
= 0;
2166 spin_lock_init(&free_i
->segmap_lock
);
2170 static int build_curseg(struct f2fs_sb_info
*sbi
)
2172 struct curseg_info
*array
;
2175 array
= kcalloc(NR_CURSEG_TYPE
, sizeof(*array
), GFP_KERNEL
);
2179 SM_I(sbi
)->curseg_array
= array
;
2181 for (i
= 0; i
< NR_CURSEG_TYPE
; i
++) {
2182 mutex_init(&array
[i
].curseg_mutex
);
2183 array
[i
].sum_blk
= kzalloc(PAGE_SIZE
, GFP_KERNEL
);
2184 if (!array
[i
].sum_blk
)
2186 init_rwsem(&array
[i
].journal_rwsem
);
2187 array
[i
].journal
= kzalloc(sizeof(struct f2fs_journal
),
2189 if (!array
[i
].journal
)
2191 array
[i
].segno
= NULL_SEGNO
;
2192 array
[i
].next_blkoff
= 0;
2194 return restore_curseg_summaries(sbi
);
2197 static void build_sit_entries(struct f2fs_sb_info
*sbi
)
2199 struct sit_info
*sit_i
= SIT_I(sbi
);
2200 struct curseg_info
*curseg
= CURSEG_I(sbi
, CURSEG_COLD_DATA
);
2201 struct f2fs_journal
*journal
= curseg
->journal
;
2202 int sit_blk_cnt
= SIT_BLK_CNT(sbi
);
2203 unsigned int i
, start
, end
;
2204 unsigned int readed
, start_blk
= 0;
2205 int nrpages
= MAX_BIO_BLOCKS(sbi
) * 8;
2208 readed
= ra_meta_pages(sbi
, start_blk
, nrpages
, META_SIT
, true);
2210 start
= start_blk
* sit_i
->sents_per_block
;
2211 end
= (start_blk
+ readed
) * sit_i
->sents_per_block
;
2213 for (; start
< end
&& start
< MAIN_SEGS(sbi
); start
++) {
2214 struct seg_entry
*se
= &sit_i
->sentries
[start
];
2215 struct f2fs_sit_block
*sit_blk
;
2216 struct f2fs_sit_entry sit
;
2219 down_read(&curseg
->journal_rwsem
);
2220 for (i
= 0; i
< sits_in_cursum(journal
); i
++) {
2221 if (le32_to_cpu(segno_in_journal(journal
, i
))
2223 sit
= sit_in_journal(journal
, i
);
2224 up_read(&curseg
->journal_rwsem
);
2228 up_read(&curseg
->journal_rwsem
);
2230 page
= get_current_sit_page(sbi
, start
);
2231 sit_blk
= (struct f2fs_sit_block
*)page_address(page
);
2232 sit
= sit_blk
->entries
[SIT_ENTRY_OFFSET(sit_i
, start
)];
2233 f2fs_put_page(page
, 1);
2235 check_block_count(sbi
, start
, &sit
);
2236 seg_info_from_raw_sit(se
, &sit
);
2238 /* build discard map only one time */
2239 memcpy(se
->discard_map
, se
->cur_valid_map
, SIT_VBLOCK_MAP_SIZE
);
2240 sbi
->discard_blks
+= sbi
->blocks_per_seg
- se
->valid_blocks
;
2242 if (sbi
->segs_per_sec
> 1) {
2243 struct sec_entry
*e
= get_sec_entry(sbi
, start
);
2244 e
->valid_blocks
+= se
->valid_blocks
;
2247 start_blk
+= readed
;
2248 } while (start_blk
< sit_blk_cnt
);
2251 static void init_free_segmap(struct f2fs_sb_info
*sbi
)
2256 for (start
= 0; start
< MAIN_SEGS(sbi
); start
++) {
2257 struct seg_entry
*sentry
= get_seg_entry(sbi
, start
);
2258 if (!sentry
->valid_blocks
)
2259 __set_free(sbi
, start
);
2262 /* set use the current segments */
2263 for (type
= CURSEG_HOT_DATA
; type
<= CURSEG_COLD_NODE
; type
++) {
2264 struct curseg_info
*curseg_t
= CURSEG_I(sbi
, type
);
2265 __set_test_and_inuse(sbi
, curseg_t
->segno
);
2269 static void init_dirty_segmap(struct f2fs_sb_info
*sbi
)
2271 struct dirty_seglist_info
*dirty_i
= DIRTY_I(sbi
);
2272 struct free_segmap_info
*free_i
= FREE_I(sbi
);
2273 unsigned int segno
= 0, offset
= 0;
2274 unsigned short valid_blocks
;
2277 /* find dirty segment based on free segmap */
2278 segno
= find_next_inuse(free_i
, MAIN_SEGS(sbi
), offset
);
2279 if (segno
>= MAIN_SEGS(sbi
))
2282 valid_blocks
= get_valid_blocks(sbi
, segno
, 0);
2283 if (valid_blocks
== sbi
->blocks_per_seg
|| !valid_blocks
)
2285 if (valid_blocks
> sbi
->blocks_per_seg
) {
2286 f2fs_bug_on(sbi
, 1);
2289 mutex_lock(&dirty_i
->seglist_lock
);
2290 __locate_dirty_segment(sbi
, segno
, DIRTY
);
2291 mutex_unlock(&dirty_i
->seglist_lock
);
2295 static int init_victim_secmap(struct f2fs_sb_info
*sbi
)
2297 struct dirty_seglist_info
*dirty_i
= DIRTY_I(sbi
);
2298 unsigned int bitmap_size
= f2fs_bitmap_size(MAIN_SECS(sbi
));
2300 dirty_i
->victim_secmap
= f2fs_kvzalloc(bitmap_size
, GFP_KERNEL
);
2301 if (!dirty_i
->victim_secmap
)
2306 static int build_dirty_segmap(struct f2fs_sb_info
*sbi
)
2308 struct dirty_seglist_info
*dirty_i
;
2309 unsigned int bitmap_size
, i
;
2311 /* allocate memory for dirty segments list information */
2312 dirty_i
= kzalloc(sizeof(struct dirty_seglist_info
), GFP_KERNEL
);
2316 SM_I(sbi
)->dirty_info
= dirty_i
;
2317 mutex_init(&dirty_i
->seglist_lock
);
2319 bitmap_size
= f2fs_bitmap_size(MAIN_SEGS(sbi
));
2321 for (i
= 0; i
< NR_DIRTY_TYPE
; i
++) {
2322 dirty_i
->dirty_segmap
[i
] = f2fs_kvzalloc(bitmap_size
, GFP_KERNEL
);
2323 if (!dirty_i
->dirty_segmap
[i
])
2327 init_dirty_segmap(sbi
);
2328 return init_victim_secmap(sbi
);
2332 * Update min, max modified time for cost-benefit GC algorithm
2334 static void init_min_max_mtime(struct f2fs_sb_info
*sbi
)
2336 struct sit_info
*sit_i
= SIT_I(sbi
);
2339 mutex_lock(&sit_i
->sentry_lock
);
2341 sit_i
->min_mtime
= LLONG_MAX
;
2343 for (segno
= 0; segno
< MAIN_SEGS(sbi
); segno
+= sbi
->segs_per_sec
) {
2345 unsigned long long mtime
= 0;
2347 for (i
= 0; i
< sbi
->segs_per_sec
; i
++)
2348 mtime
+= get_seg_entry(sbi
, segno
+ i
)->mtime
;
2350 mtime
= div_u64(mtime
, sbi
->segs_per_sec
);
2352 if (sit_i
->min_mtime
> mtime
)
2353 sit_i
->min_mtime
= mtime
;
2355 sit_i
->max_mtime
= get_mtime(sbi
);
2356 mutex_unlock(&sit_i
->sentry_lock
);
2359 int build_segment_manager(struct f2fs_sb_info
*sbi
)
2361 struct f2fs_super_block
*raw_super
= F2FS_RAW_SUPER(sbi
);
2362 struct f2fs_checkpoint
*ckpt
= F2FS_CKPT(sbi
);
2363 struct f2fs_sm_info
*sm_info
;
2366 sm_info
= kzalloc(sizeof(struct f2fs_sm_info
), GFP_KERNEL
);
2371 sbi
->sm_info
= sm_info
;
2372 sm_info
->seg0_blkaddr
= le32_to_cpu(raw_super
->segment0_blkaddr
);
2373 sm_info
->main_blkaddr
= le32_to_cpu(raw_super
->main_blkaddr
);
2374 sm_info
->segment_count
= le32_to_cpu(raw_super
->segment_count
);
2375 sm_info
->reserved_segments
= le32_to_cpu(ckpt
->rsvd_segment_count
);
2376 sm_info
->ovp_segments
= le32_to_cpu(ckpt
->overprov_segment_count
);
2377 sm_info
->main_segments
= le32_to_cpu(raw_super
->segment_count_main
);
2378 sm_info
->ssa_blkaddr
= le32_to_cpu(raw_super
->ssa_blkaddr
);
2379 sm_info
->rec_prefree_segments
= sm_info
->main_segments
*
2380 DEF_RECLAIM_PREFREE_SEGMENTS
/ 100;
2381 sm_info
->ipu_policy
= 1 << F2FS_IPU_FSYNC
;
2382 sm_info
->min_ipu_util
= DEF_MIN_IPU_UTIL
;
2383 sm_info
->min_fsync_blocks
= DEF_MIN_FSYNC_BLOCKS
;
2385 INIT_LIST_HEAD(&sm_info
->discard_list
);
2386 sm_info
->nr_discards
= 0;
2387 sm_info
->max_discards
= 0;
2389 sm_info
->trim_sections
= DEF_BATCHED_TRIM_SECTIONS
;
2391 INIT_LIST_HEAD(&sm_info
->sit_entry_set
);
2393 if (test_opt(sbi
, FLUSH_MERGE
) && !f2fs_readonly(sbi
->sb
)) {
2394 err
= create_flush_cmd_control(sbi
);
2399 err
= build_sit_info(sbi
);
2402 err
= build_free_segmap(sbi
);
2405 err
= build_curseg(sbi
);
2409 /* reinit free segmap based on SIT */
2410 build_sit_entries(sbi
);
2412 init_free_segmap(sbi
);
2413 err
= build_dirty_segmap(sbi
);
2417 init_min_max_mtime(sbi
);
2421 static void discard_dirty_segmap(struct f2fs_sb_info
*sbi
,
2422 enum dirty_type dirty_type
)
2424 struct dirty_seglist_info
*dirty_i
= DIRTY_I(sbi
);
2426 mutex_lock(&dirty_i
->seglist_lock
);
2427 kvfree(dirty_i
->dirty_segmap
[dirty_type
]);
2428 dirty_i
->nr_dirty
[dirty_type
] = 0;
2429 mutex_unlock(&dirty_i
->seglist_lock
);
2432 static void destroy_victim_secmap(struct f2fs_sb_info
*sbi
)
2434 struct dirty_seglist_info
*dirty_i
= DIRTY_I(sbi
);
2435 kvfree(dirty_i
->victim_secmap
);
2438 static void destroy_dirty_segmap(struct f2fs_sb_info
*sbi
)
2440 struct dirty_seglist_info
*dirty_i
= DIRTY_I(sbi
);
2446 /* discard pre-free/dirty segments list */
2447 for (i
= 0; i
< NR_DIRTY_TYPE
; i
++)
2448 discard_dirty_segmap(sbi
, i
);
2450 destroy_victim_secmap(sbi
);
2451 SM_I(sbi
)->dirty_info
= NULL
;
2455 static void destroy_curseg(struct f2fs_sb_info
*sbi
)
2457 struct curseg_info
*array
= SM_I(sbi
)->curseg_array
;
2462 SM_I(sbi
)->curseg_array
= NULL
;
2463 for (i
= 0; i
< NR_CURSEG_TYPE
; i
++) {
2464 kfree(array
[i
].sum_blk
);
2465 kfree(array
[i
].journal
);
2470 static void destroy_free_segmap(struct f2fs_sb_info
*sbi
)
2472 struct free_segmap_info
*free_i
= SM_I(sbi
)->free_info
;
2475 SM_I(sbi
)->free_info
= NULL
;
2476 kvfree(free_i
->free_segmap
);
2477 kvfree(free_i
->free_secmap
);
2481 static void destroy_sit_info(struct f2fs_sb_info
*sbi
)
2483 struct sit_info
*sit_i
= SIT_I(sbi
);
2489 if (sit_i
->sentries
) {
2490 for (start
= 0; start
< MAIN_SEGS(sbi
); start
++) {
2491 kfree(sit_i
->sentries
[start
].cur_valid_map
);
2492 kfree(sit_i
->sentries
[start
].ckpt_valid_map
);
2493 kfree(sit_i
->sentries
[start
].discard_map
);
2496 kfree(sit_i
->tmp_map
);
2498 kvfree(sit_i
->sentries
);
2499 kvfree(sit_i
->sec_entries
);
2500 kvfree(sit_i
->dirty_sentries_bitmap
);
2502 SM_I(sbi
)->sit_info
= NULL
;
2503 kfree(sit_i
->sit_bitmap
);
2507 void destroy_segment_manager(struct f2fs_sb_info
*sbi
)
2509 struct f2fs_sm_info
*sm_info
= SM_I(sbi
);
2513 destroy_flush_cmd_control(sbi
);
2514 destroy_dirty_segmap(sbi
);
2515 destroy_curseg(sbi
);
2516 destroy_free_segmap(sbi
);
2517 destroy_sit_info(sbi
);
2518 sbi
->sm_info
= NULL
;
2522 int __init
create_segment_manager_caches(void)
2524 discard_entry_slab
= f2fs_kmem_cache_create("discard_entry",
2525 sizeof(struct discard_entry
));
2526 if (!discard_entry_slab
)
2529 sit_entry_set_slab
= f2fs_kmem_cache_create("sit_entry_set",
2530 sizeof(struct sit_entry_set
));
2531 if (!sit_entry_set_slab
)
2532 goto destory_discard_entry
;
2534 inmem_entry_slab
= f2fs_kmem_cache_create("inmem_page_entry",
2535 sizeof(struct inmem_pages
));
2536 if (!inmem_entry_slab
)
2537 goto destroy_sit_entry_set
;
2540 destroy_sit_entry_set
:
2541 kmem_cache_destroy(sit_entry_set_slab
);
2542 destory_discard_entry
:
2543 kmem_cache_destroy(discard_entry_slab
);
2548 void destroy_segment_manager_caches(void)
2550 kmem_cache_destroy(sit_entry_set_slab
);
2551 kmem_cache_destroy(discard_entry_slab
);
2552 kmem_cache_destroy(inmem_entry_slab
);