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 ClearPageUptodate(page
);
227 set_page_private(page
, 0);
228 ClearPageUptodate(page
);
229 f2fs_put_page(page
, 1);
231 list_del(&cur
->list
);
232 kmem_cache_free(inmem_entry_slab
, cur
);
233 dec_page_count(F2FS_I_SB(inode
), F2FS_INMEM_PAGES
);
238 void drop_inmem_pages(struct inode
*inode
)
240 struct f2fs_inode_info
*fi
= F2FS_I(inode
);
242 mutex_lock(&fi
->inmem_lock
);
243 __revoke_inmem_pages(inode
, &fi
->inmem_pages
, true, false);
244 mutex_unlock(&fi
->inmem_lock
);
247 static int __commit_inmem_pages(struct inode
*inode
,
248 struct list_head
*revoke_list
)
250 struct f2fs_sb_info
*sbi
= F2FS_I_SB(inode
);
251 struct f2fs_inode_info
*fi
= F2FS_I(inode
);
252 struct inmem_pages
*cur
, *tmp
;
253 struct f2fs_io_info fio
= {
256 .rw
= WRITE_SYNC
| REQ_PRIO
,
257 .encrypted_page
= NULL
,
259 bool submit_bio
= false;
262 list_for_each_entry_safe(cur
, tmp
, &fi
->inmem_pages
, list
) {
263 struct page
*page
= cur
->page
;
266 if (page
->mapping
== inode
->i_mapping
) {
267 trace_f2fs_commit_inmem_page(page
, INMEM
);
269 set_page_dirty(page
);
270 f2fs_wait_on_page_writeback(page
, DATA
, true);
271 if (clear_page_dirty_for_io(page
))
272 inode_dec_dirty_pages(inode
);
275 err
= do_write_data_page(&fio
);
281 /* record old blkaddr for revoking */
282 cur
->old_addr
= fio
.old_blkaddr
;
284 clear_cold_data(page
);
288 list_move_tail(&cur
->list
, revoke_list
);
292 f2fs_submit_merged_bio_cond(sbi
, inode
, NULL
, 0, DATA
, WRITE
);
295 __revoke_inmem_pages(inode
, revoke_list
, false, false);
300 int commit_inmem_pages(struct inode
*inode
)
302 struct f2fs_sb_info
*sbi
= F2FS_I_SB(inode
);
303 struct f2fs_inode_info
*fi
= F2FS_I(inode
);
304 struct list_head revoke_list
;
307 INIT_LIST_HEAD(&revoke_list
);
308 f2fs_balance_fs(sbi
, true);
311 mutex_lock(&fi
->inmem_lock
);
312 err
= __commit_inmem_pages(inode
, &revoke_list
);
316 * try to revoke all committed pages, but still we could fail
317 * due to no memory or other reason, if that happened, EAGAIN
318 * will be returned, which means in such case, transaction is
319 * already not integrity, caller should use journal to do the
320 * recovery or rewrite & commit last transaction. For other
321 * error number, revoking was done by filesystem itself.
323 ret
= __revoke_inmem_pages(inode
, &revoke_list
, false, true);
327 /* drop all uncommitted pages */
328 __revoke_inmem_pages(inode
, &fi
->inmem_pages
, true, false);
330 mutex_unlock(&fi
->inmem_lock
);
337 * This function balances dirty node and dentry pages.
338 * In addition, it controls garbage collection.
340 void f2fs_balance_fs(struct f2fs_sb_info
*sbi
, bool need
)
345 * We should do GC or end up with checkpoint, if there are so many dirty
346 * dir/node pages without enough free segments.
348 if (has_not_enough_free_secs(sbi
, 0)) {
349 mutex_lock(&sbi
->gc_mutex
);
354 void f2fs_balance_fs_bg(struct f2fs_sb_info
*sbi
)
356 /* try to shrink extent cache when there is no enough memory */
357 if (!available_free_memory(sbi
, EXTENT_CACHE
))
358 f2fs_shrink_extent_tree(sbi
, EXTENT_CACHE_SHRINK_NUMBER
);
360 /* check the # of cached NAT entries */
361 if (!available_free_memory(sbi
, NAT_ENTRIES
))
362 try_to_free_nats(sbi
, NAT_ENTRY_PER_BLOCK
);
364 if (!available_free_memory(sbi
, FREE_NIDS
))
365 try_to_free_nids(sbi
, NAT_ENTRY_PER_BLOCK
* FREE_NID_PAGES
);
367 /* checkpoint is the only way to shrink partial cached entries */
368 if (!available_free_memory(sbi
, NAT_ENTRIES
) ||
369 !available_free_memory(sbi
, INO_ENTRIES
) ||
370 excess_prefree_segs(sbi
) ||
371 excess_dirty_nats(sbi
) ||
372 (is_idle(sbi
) && f2fs_time_over(sbi
, CP_TIME
))) {
373 if (test_opt(sbi
, DATA_FLUSH
))
374 sync_dirty_inodes(sbi
, FILE_INODE
);
375 f2fs_sync_fs(sbi
->sb
, true);
376 stat_inc_bg_cp_count(sbi
->stat_info
);
380 static int issue_flush_thread(void *data
)
382 struct f2fs_sb_info
*sbi
= data
;
383 struct flush_cmd_control
*fcc
= SM_I(sbi
)->cmd_control_info
;
384 wait_queue_head_t
*q
= &fcc
->flush_wait_queue
;
386 if (kthread_should_stop())
389 if (!llist_empty(&fcc
->issue_list
)) {
391 struct flush_cmd
*cmd
, *next
;
394 bio
= f2fs_bio_alloc(0);
396 fcc
->dispatch_list
= llist_del_all(&fcc
->issue_list
);
397 fcc
->dispatch_list
= llist_reverse_order(fcc
->dispatch_list
);
399 bio
->bi_bdev
= sbi
->sb
->s_bdev
;
400 ret
= submit_bio_wait(WRITE_FLUSH
, bio
);
402 llist_for_each_entry_safe(cmd
, next
,
403 fcc
->dispatch_list
, llnode
) {
405 complete(&cmd
->wait
);
408 fcc
->dispatch_list
= NULL
;
411 wait_event_interruptible(*q
,
412 kthread_should_stop() || !llist_empty(&fcc
->issue_list
));
416 int f2fs_issue_flush(struct f2fs_sb_info
*sbi
)
418 struct flush_cmd_control
*fcc
= SM_I(sbi
)->cmd_control_info
;
419 struct flush_cmd cmd
;
421 trace_f2fs_issue_flush(sbi
->sb
, test_opt(sbi
, NOBARRIER
),
422 test_opt(sbi
, FLUSH_MERGE
));
424 if (test_opt(sbi
, NOBARRIER
))
427 if (!test_opt(sbi
, FLUSH_MERGE
)) {
428 struct bio
*bio
= f2fs_bio_alloc(0);
431 bio
->bi_bdev
= sbi
->sb
->s_bdev
;
432 ret
= submit_bio_wait(WRITE_FLUSH
, bio
);
437 init_completion(&cmd
.wait
);
439 llist_add(&cmd
.llnode
, &fcc
->issue_list
);
441 if (!fcc
->dispatch_list
)
442 wake_up(&fcc
->flush_wait_queue
);
444 wait_for_completion(&cmd
.wait
);
449 int create_flush_cmd_control(struct f2fs_sb_info
*sbi
)
451 dev_t dev
= sbi
->sb
->s_bdev
->bd_dev
;
452 struct flush_cmd_control
*fcc
;
455 fcc
= kzalloc(sizeof(struct flush_cmd_control
), GFP_KERNEL
);
458 init_waitqueue_head(&fcc
->flush_wait_queue
);
459 init_llist_head(&fcc
->issue_list
);
460 SM_I(sbi
)->cmd_control_info
= fcc
;
461 fcc
->f2fs_issue_flush
= kthread_run(issue_flush_thread
, sbi
,
462 "f2fs_flush-%u:%u", MAJOR(dev
), MINOR(dev
));
463 if (IS_ERR(fcc
->f2fs_issue_flush
)) {
464 err
= PTR_ERR(fcc
->f2fs_issue_flush
);
466 SM_I(sbi
)->cmd_control_info
= NULL
;
473 void destroy_flush_cmd_control(struct f2fs_sb_info
*sbi
)
475 struct flush_cmd_control
*fcc
= SM_I(sbi
)->cmd_control_info
;
477 if (fcc
&& fcc
->f2fs_issue_flush
)
478 kthread_stop(fcc
->f2fs_issue_flush
);
480 SM_I(sbi
)->cmd_control_info
= NULL
;
483 static void __locate_dirty_segment(struct f2fs_sb_info
*sbi
, unsigned int segno
,
484 enum dirty_type dirty_type
)
486 struct dirty_seglist_info
*dirty_i
= DIRTY_I(sbi
);
488 /* need not be added */
489 if (IS_CURSEG(sbi
, segno
))
492 if (!test_and_set_bit(segno
, dirty_i
->dirty_segmap
[dirty_type
]))
493 dirty_i
->nr_dirty
[dirty_type
]++;
495 if (dirty_type
== DIRTY
) {
496 struct seg_entry
*sentry
= get_seg_entry(sbi
, segno
);
497 enum dirty_type t
= sentry
->type
;
499 if (unlikely(t
>= DIRTY
)) {
503 if (!test_and_set_bit(segno
, dirty_i
->dirty_segmap
[t
]))
504 dirty_i
->nr_dirty
[t
]++;
508 static void __remove_dirty_segment(struct f2fs_sb_info
*sbi
, unsigned int segno
,
509 enum dirty_type dirty_type
)
511 struct dirty_seglist_info
*dirty_i
= DIRTY_I(sbi
);
513 if (test_and_clear_bit(segno
, dirty_i
->dirty_segmap
[dirty_type
]))
514 dirty_i
->nr_dirty
[dirty_type
]--;
516 if (dirty_type
== DIRTY
) {
517 struct seg_entry
*sentry
= get_seg_entry(sbi
, segno
);
518 enum dirty_type t
= sentry
->type
;
520 if (test_and_clear_bit(segno
, dirty_i
->dirty_segmap
[t
]))
521 dirty_i
->nr_dirty
[t
]--;
523 if (get_valid_blocks(sbi
, segno
, sbi
->segs_per_sec
) == 0)
524 clear_bit(GET_SECNO(sbi
, segno
),
525 dirty_i
->victim_secmap
);
530 * Should not occur error such as -ENOMEM.
531 * Adding dirty entry into seglist is not critical operation.
532 * If a given segment is one of current working segments, it won't be added.
534 static void locate_dirty_segment(struct f2fs_sb_info
*sbi
, unsigned int segno
)
536 struct dirty_seglist_info
*dirty_i
= DIRTY_I(sbi
);
537 unsigned short valid_blocks
;
539 if (segno
== NULL_SEGNO
|| IS_CURSEG(sbi
, segno
))
542 mutex_lock(&dirty_i
->seglist_lock
);
544 valid_blocks
= get_valid_blocks(sbi
, segno
, 0);
546 if (valid_blocks
== 0) {
547 __locate_dirty_segment(sbi
, segno
, PRE
);
548 __remove_dirty_segment(sbi
, segno
, DIRTY
);
549 } else if (valid_blocks
< sbi
->blocks_per_seg
) {
550 __locate_dirty_segment(sbi
, segno
, DIRTY
);
552 /* Recovery routine with SSR needs this */
553 __remove_dirty_segment(sbi
, segno
, DIRTY
);
556 mutex_unlock(&dirty_i
->seglist_lock
);
559 static int f2fs_issue_discard(struct f2fs_sb_info
*sbi
,
560 block_t blkstart
, block_t blklen
)
562 sector_t start
= SECTOR_FROM_BLOCK(blkstart
);
563 sector_t len
= SECTOR_FROM_BLOCK(blklen
);
564 struct seg_entry
*se
;
568 for (i
= blkstart
; i
< blkstart
+ blklen
; i
++) {
569 se
= get_seg_entry(sbi
, GET_SEGNO(sbi
, i
));
570 offset
= GET_BLKOFF_FROM_SEG0(sbi
, i
);
572 if (!f2fs_test_and_set_bit(offset
, se
->discard_map
))
575 trace_f2fs_issue_discard(sbi
->sb
, blkstart
, blklen
);
576 return blkdev_issue_discard(sbi
->sb
->s_bdev
, start
, len
, GFP_NOFS
, 0);
579 bool discard_next_dnode(struct f2fs_sb_info
*sbi
, block_t blkaddr
)
581 int err
= -EOPNOTSUPP
;
583 if (test_opt(sbi
, DISCARD
)) {
584 struct seg_entry
*se
= get_seg_entry(sbi
,
585 GET_SEGNO(sbi
, blkaddr
));
586 unsigned int offset
= GET_BLKOFF_FROM_SEG0(sbi
, blkaddr
);
588 if (f2fs_test_bit(offset
, se
->discard_map
))
591 err
= f2fs_issue_discard(sbi
, blkaddr
, 1);
595 update_meta_page(sbi
, NULL
, blkaddr
);
601 static void __add_discard_entry(struct f2fs_sb_info
*sbi
,
602 struct cp_control
*cpc
, struct seg_entry
*se
,
603 unsigned int start
, unsigned int end
)
605 struct list_head
*head
= &SM_I(sbi
)->discard_list
;
606 struct discard_entry
*new, *last
;
608 if (!list_empty(head
)) {
609 last
= list_last_entry(head
, struct discard_entry
, list
);
610 if (START_BLOCK(sbi
, cpc
->trim_start
) + start
==
611 last
->blkaddr
+ last
->len
) {
612 last
->len
+= end
- start
;
617 new = f2fs_kmem_cache_alloc(discard_entry_slab
, GFP_NOFS
);
618 INIT_LIST_HEAD(&new->list
);
619 new->blkaddr
= START_BLOCK(sbi
, cpc
->trim_start
) + start
;
620 new->len
= end
- start
;
621 list_add_tail(&new->list
, head
);
623 SM_I(sbi
)->nr_discards
+= end
- start
;
626 static void add_discard_addrs(struct f2fs_sb_info
*sbi
, struct cp_control
*cpc
)
628 int entries
= SIT_VBLOCK_MAP_SIZE
/ sizeof(unsigned long);
629 int max_blocks
= sbi
->blocks_per_seg
;
630 struct seg_entry
*se
= get_seg_entry(sbi
, cpc
->trim_start
);
631 unsigned long *cur_map
= (unsigned long *)se
->cur_valid_map
;
632 unsigned long *ckpt_map
= (unsigned long *)se
->ckpt_valid_map
;
633 unsigned long *discard_map
= (unsigned long *)se
->discard_map
;
634 unsigned long *dmap
= SIT_I(sbi
)->tmp_map
;
635 unsigned int start
= 0, end
= -1;
636 bool force
= (cpc
->reason
== CP_DISCARD
);
639 if (se
->valid_blocks
== max_blocks
)
643 if (!test_opt(sbi
, DISCARD
) || !se
->valid_blocks
||
644 SM_I(sbi
)->nr_discards
>= SM_I(sbi
)->max_discards
)
648 /* SIT_VBLOCK_MAP_SIZE should be multiple of sizeof(unsigned long) */
649 for (i
= 0; i
< entries
; i
++)
650 dmap
[i
] = force
? ~ckpt_map
[i
] & ~discard_map
[i
] :
651 (cur_map
[i
] ^ ckpt_map
[i
]) & ckpt_map
[i
];
653 while (force
|| SM_I(sbi
)->nr_discards
<= SM_I(sbi
)->max_discards
) {
654 start
= __find_rev_next_bit(dmap
, max_blocks
, end
+ 1);
655 if (start
>= max_blocks
)
658 end
= __find_rev_next_zero_bit(dmap
, max_blocks
, start
+ 1);
659 __add_discard_entry(sbi
, cpc
, se
, start
, end
);
663 void release_discard_addrs(struct f2fs_sb_info
*sbi
)
665 struct list_head
*head
= &(SM_I(sbi
)->discard_list
);
666 struct discard_entry
*entry
, *this;
669 list_for_each_entry_safe(entry
, this, head
, list
) {
670 list_del(&entry
->list
);
671 kmem_cache_free(discard_entry_slab
, entry
);
676 * Should call clear_prefree_segments after checkpoint is done.
678 static void set_prefree_as_free_segments(struct f2fs_sb_info
*sbi
)
680 struct dirty_seglist_info
*dirty_i
= DIRTY_I(sbi
);
683 mutex_lock(&dirty_i
->seglist_lock
);
684 for_each_set_bit(segno
, dirty_i
->dirty_segmap
[PRE
], MAIN_SEGS(sbi
))
685 __set_test_and_free(sbi
, segno
);
686 mutex_unlock(&dirty_i
->seglist_lock
);
689 void clear_prefree_segments(struct f2fs_sb_info
*sbi
, struct cp_control
*cpc
)
691 struct list_head
*head
= &(SM_I(sbi
)->discard_list
);
692 struct discard_entry
*entry
, *this;
693 struct dirty_seglist_info
*dirty_i
= DIRTY_I(sbi
);
694 unsigned long *prefree_map
= dirty_i
->dirty_segmap
[PRE
];
695 unsigned int start
= 0, end
= -1;
697 mutex_lock(&dirty_i
->seglist_lock
);
701 start
= find_next_bit(prefree_map
, MAIN_SEGS(sbi
), end
+ 1);
702 if (start
>= MAIN_SEGS(sbi
))
704 end
= find_next_zero_bit(prefree_map
, MAIN_SEGS(sbi
),
707 for (i
= start
; i
< end
; i
++)
708 clear_bit(i
, prefree_map
);
710 dirty_i
->nr_dirty
[PRE
] -= end
- start
;
712 if (!test_opt(sbi
, DISCARD
))
715 f2fs_issue_discard(sbi
, START_BLOCK(sbi
, start
),
716 (end
- start
) << sbi
->log_blocks_per_seg
);
718 mutex_unlock(&dirty_i
->seglist_lock
);
720 /* send small discards */
721 list_for_each_entry_safe(entry
, this, head
, list
) {
722 if (cpc
->reason
== CP_DISCARD
&& entry
->len
< cpc
->trim_minlen
)
724 f2fs_issue_discard(sbi
, entry
->blkaddr
, entry
->len
);
725 cpc
->trimmed
+= entry
->len
;
727 list_del(&entry
->list
);
728 SM_I(sbi
)->nr_discards
-= entry
->len
;
729 kmem_cache_free(discard_entry_slab
, entry
);
733 static bool __mark_sit_entry_dirty(struct f2fs_sb_info
*sbi
, unsigned int segno
)
735 struct sit_info
*sit_i
= SIT_I(sbi
);
737 if (!__test_and_set_bit(segno
, sit_i
->dirty_sentries_bitmap
)) {
738 sit_i
->dirty_sentries
++;
745 static void __set_sit_entry_type(struct f2fs_sb_info
*sbi
, int type
,
746 unsigned int segno
, int modified
)
748 struct seg_entry
*se
= get_seg_entry(sbi
, segno
);
751 __mark_sit_entry_dirty(sbi
, segno
);
754 static void update_sit_entry(struct f2fs_sb_info
*sbi
, block_t blkaddr
, int del
)
756 struct seg_entry
*se
;
757 unsigned int segno
, offset
;
758 long int new_vblocks
;
760 segno
= GET_SEGNO(sbi
, blkaddr
);
762 se
= get_seg_entry(sbi
, segno
);
763 new_vblocks
= se
->valid_blocks
+ del
;
764 offset
= GET_BLKOFF_FROM_SEG0(sbi
, blkaddr
);
766 f2fs_bug_on(sbi
, (new_vblocks
>> (sizeof(unsigned short) << 3) ||
767 (new_vblocks
> sbi
->blocks_per_seg
)));
769 se
->valid_blocks
= new_vblocks
;
770 se
->mtime
= get_mtime(sbi
);
771 SIT_I(sbi
)->max_mtime
= se
->mtime
;
773 /* Update valid block bitmap */
775 if (f2fs_test_and_set_bit(offset
, se
->cur_valid_map
))
777 if (!f2fs_test_and_set_bit(offset
, se
->discard_map
))
780 if (!f2fs_test_and_clear_bit(offset
, se
->cur_valid_map
))
782 if (f2fs_test_and_clear_bit(offset
, se
->discard_map
))
785 if (!f2fs_test_bit(offset
, se
->ckpt_valid_map
))
786 se
->ckpt_valid_blocks
+= del
;
788 __mark_sit_entry_dirty(sbi
, segno
);
790 /* update total number of valid blocks to be written in ckpt area */
791 SIT_I(sbi
)->written_valid_blocks
+= del
;
793 if (sbi
->segs_per_sec
> 1)
794 get_sec_entry(sbi
, segno
)->valid_blocks
+= del
;
797 void refresh_sit_entry(struct f2fs_sb_info
*sbi
, block_t old
, block_t
new)
799 update_sit_entry(sbi
, new, 1);
800 if (GET_SEGNO(sbi
, old
) != NULL_SEGNO
)
801 update_sit_entry(sbi
, old
, -1);
803 locate_dirty_segment(sbi
, GET_SEGNO(sbi
, old
));
804 locate_dirty_segment(sbi
, GET_SEGNO(sbi
, new));
807 void invalidate_blocks(struct f2fs_sb_info
*sbi
, block_t addr
)
809 unsigned int segno
= GET_SEGNO(sbi
, addr
);
810 struct sit_info
*sit_i
= SIT_I(sbi
);
812 f2fs_bug_on(sbi
, addr
== NULL_ADDR
);
813 if (addr
== NEW_ADDR
)
816 /* add it into sit main buffer */
817 mutex_lock(&sit_i
->sentry_lock
);
819 update_sit_entry(sbi
, addr
, -1);
821 /* add it into dirty seglist */
822 locate_dirty_segment(sbi
, segno
);
824 mutex_unlock(&sit_i
->sentry_lock
);
827 bool is_checkpointed_data(struct f2fs_sb_info
*sbi
, block_t blkaddr
)
829 struct sit_info
*sit_i
= SIT_I(sbi
);
830 unsigned int segno
, offset
;
831 struct seg_entry
*se
;
834 if (blkaddr
== NEW_ADDR
|| blkaddr
== NULL_ADDR
)
837 mutex_lock(&sit_i
->sentry_lock
);
839 segno
= GET_SEGNO(sbi
, blkaddr
);
840 se
= get_seg_entry(sbi
, segno
);
841 offset
= GET_BLKOFF_FROM_SEG0(sbi
, blkaddr
);
843 if (f2fs_test_bit(offset
, se
->ckpt_valid_map
))
846 mutex_unlock(&sit_i
->sentry_lock
);
852 * This function should be resided under the curseg_mutex lock
854 static void __add_sum_entry(struct f2fs_sb_info
*sbi
, int type
,
855 struct f2fs_summary
*sum
)
857 struct curseg_info
*curseg
= CURSEG_I(sbi
, type
);
858 void *addr
= curseg
->sum_blk
;
859 addr
+= curseg
->next_blkoff
* sizeof(struct f2fs_summary
);
860 memcpy(addr
, sum
, sizeof(struct f2fs_summary
));
864 * Calculate the number of current summary pages for writing
866 int npages_for_summary_flush(struct f2fs_sb_info
*sbi
, bool for_ra
)
868 int valid_sum_count
= 0;
871 for (i
= CURSEG_HOT_DATA
; i
<= CURSEG_COLD_DATA
; i
++) {
872 if (sbi
->ckpt
->alloc_type
[i
] == SSR
)
873 valid_sum_count
+= sbi
->blocks_per_seg
;
876 valid_sum_count
+= le16_to_cpu(
877 F2FS_CKPT(sbi
)->cur_data_blkoff
[i
]);
879 valid_sum_count
+= curseg_blkoff(sbi
, i
);
883 sum_in_page
= (PAGE_CACHE_SIZE
- 2 * SUM_JOURNAL_SIZE
-
884 SUM_FOOTER_SIZE
) / SUMMARY_SIZE
;
885 if (valid_sum_count
<= sum_in_page
)
887 else if ((valid_sum_count
- sum_in_page
) <=
888 (PAGE_CACHE_SIZE
- SUM_FOOTER_SIZE
) / SUMMARY_SIZE
)
894 * Caller should put this summary page
896 struct page
*get_sum_page(struct f2fs_sb_info
*sbi
, unsigned int segno
)
898 return get_meta_page(sbi
, GET_SUM_BLOCK(sbi
, segno
));
901 void update_meta_page(struct f2fs_sb_info
*sbi
, void *src
, block_t blk_addr
)
903 struct page
*page
= grab_meta_page(sbi
, blk_addr
);
904 void *dst
= page_address(page
);
907 memcpy(dst
, src
, PAGE_CACHE_SIZE
);
909 memset(dst
, 0, PAGE_CACHE_SIZE
);
910 set_page_dirty(page
);
911 f2fs_put_page(page
, 1);
914 static void write_sum_page(struct f2fs_sb_info
*sbi
,
915 struct f2fs_summary_block
*sum_blk
, block_t blk_addr
)
917 update_meta_page(sbi
, (void *)sum_blk
, blk_addr
);
920 static int is_next_segment_free(struct f2fs_sb_info
*sbi
, int type
)
922 struct curseg_info
*curseg
= CURSEG_I(sbi
, type
);
923 unsigned int segno
= curseg
->segno
+ 1;
924 struct free_segmap_info
*free_i
= FREE_I(sbi
);
926 if (segno
< MAIN_SEGS(sbi
) && segno
% sbi
->segs_per_sec
)
927 return !test_bit(segno
, free_i
->free_segmap
);
932 * Find a new segment from the free segments bitmap to right order
933 * This function should be returned with success, otherwise BUG
935 static void get_new_segment(struct f2fs_sb_info
*sbi
,
936 unsigned int *newseg
, bool new_sec
, int dir
)
938 struct free_segmap_info
*free_i
= FREE_I(sbi
);
939 unsigned int segno
, secno
, zoneno
;
940 unsigned int total_zones
= MAIN_SECS(sbi
) / sbi
->secs_per_zone
;
941 unsigned int hint
= *newseg
/ sbi
->segs_per_sec
;
942 unsigned int old_zoneno
= GET_ZONENO_FROM_SEGNO(sbi
, *newseg
);
943 unsigned int left_start
= hint
;
948 spin_lock(&free_i
->segmap_lock
);
950 if (!new_sec
&& ((*newseg
+ 1) % sbi
->segs_per_sec
)) {
951 segno
= find_next_zero_bit(free_i
->free_segmap
,
952 (hint
+ 1) * sbi
->segs_per_sec
, *newseg
+ 1);
953 if (segno
< (hint
+ 1) * sbi
->segs_per_sec
)
957 secno
= find_next_zero_bit(free_i
->free_secmap
, MAIN_SECS(sbi
), hint
);
958 if (secno
>= MAIN_SECS(sbi
)) {
959 if (dir
== ALLOC_RIGHT
) {
960 secno
= find_next_zero_bit(free_i
->free_secmap
,
962 f2fs_bug_on(sbi
, secno
>= MAIN_SECS(sbi
));
965 left_start
= hint
- 1;
971 while (test_bit(left_start
, free_i
->free_secmap
)) {
972 if (left_start
> 0) {
976 left_start
= find_next_zero_bit(free_i
->free_secmap
,
978 f2fs_bug_on(sbi
, left_start
>= MAIN_SECS(sbi
));
984 segno
= secno
* sbi
->segs_per_sec
;
985 zoneno
= secno
/ sbi
->secs_per_zone
;
987 /* give up on finding another zone */
990 if (sbi
->secs_per_zone
== 1)
992 if (zoneno
== old_zoneno
)
994 if (dir
== ALLOC_LEFT
) {
995 if (!go_left
&& zoneno
+ 1 >= total_zones
)
997 if (go_left
&& zoneno
== 0)
1000 for (i
= 0; i
< NR_CURSEG_TYPE
; i
++)
1001 if (CURSEG_I(sbi
, i
)->zone
== zoneno
)
1004 if (i
< NR_CURSEG_TYPE
) {
1005 /* zone is in user, try another */
1007 hint
= zoneno
* sbi
->secs_per_zone
- 1;
1008 else if (zoneno
+ 1 >= total_zones
)
1011 hint
= (zoneno
+ 1) * sbi
->secs_per_zone
;
1013 goto find_other_zone
;
1016 /* set it as dirty segment in free segmap */
1017 f2fs_bug_on(sbi
, test_bit(segno
, free_i
->free_segmap
));
1018 __set_inuse(sbi
, segno
);
1020 spin_unlock(&free_i
->segmap_lock
);
1023 static void reset_curseg(struct f2fs_sb_info
*sbi
, int type
, int modified
)
1025 struct curseg_info
*curseg
= CURSEG_I(sbi
, type
);
1026 struct summary_footer
*sum_footer
;
1028 curseg
->segno
= curseg
->next_segno
;
1029 curseg
->zone
= GET_ZONENO_FROM_SEGNO(sbi
, curseg
->segno
);
1030 curseg
->next_blkoff
= 0;
1031 curseg
->next_segno
= NULL_SEGNO
;
1033 sum_footer
= &(curseg
->sum_blk
->footer
);
1034 memset(sum_footer
, 0, sizeof(struct summary_footer
));
1035 if (IS_DATASEG(type
))
1036 SET_SUM_TYPE(sum_footer
, SUM_TYPE_DATA
);
1037 if (IS_NODESEG(type
))
1038 SET_SUM_TYPE(sum_footer
, SUM_TYPE_NODE
);
1039 __set_sit_entry_type(sbi
, type
, curseg
->segno
, modified
);
1043 * Allocate a current working segment.
1044 * This function always allocates a free segment in LFS manner.
1046 static void new_curseg(struct f2fs_sb_info
*sbi
, int type
, bool new_sec
)
1048 struct curseg_info
*curseg
= CURSEG_I(sbi
, type
);
1049 unsigned int segno
= curseg
->segno
;
1050 int dir
= ALLOC_LEFT
;
1052 write_sum_page(sbi
, curseg
->sum_blk
,
1053 GET_SUM_BLOCK(sbi
, segno
));
1054 if (type
== CURSEG_WARM_DATA
|| type
== CURSEG_COLD_DATA
)
1057 if (test_opt(sbi
, NOHEAP
))
1060 get_new_segment(sbi
, &segno
, new_sec
, dir
);
1061 curseg
->next_segno
= segno
;
1062 reset_curseg(sbi
, type
, 1);
1063 curseg
->alloc_type
= LFS
;
1066 static void __next_free_blkoff(struct f2fs_sb_info
*sbi
,
1067 struct curseg_info
*seg
, block_t start
)
1069 struct seg_entry
*se
= get_seg_entry(sbi
, seg
->segno
);
1070 int entries
= SIT_VBLOCK_MAP_SIZE
/ sizeof(unsigned long);
1071 unsigned long *target_map
= SIT_I(sbi
)->tmp_map
;
1072 unsigned long *ckpt_map
= (unsigned long *)se
->ckpt_valid_map
;
1073 unsigned long *cur_map
= (unsigned long *)se
->cur_valid_map
;
1076 for (i
= 0; i
< entries
; i
++)
1077 target_map
[i
] = ckpt_map
[i
] | cur_map
[i
];
1079 pos
= __find_rev_next_zero_bit(target_map
, sbi
->blocks_per_seg
, start
);
1081 seg
->next_blkoff
= pos
;
1085 * If a segment is written by LFS manner, next block offset is just obtained
1086 * by increasing the current block offset. However, if a segment is written by
1087 * SSR manner, next block offset obtained by calling __next_free_blkoff
1089 static void __refresh_next_blkoff(struct f2fs_sb_info
*sbi
,
1090 struct curseg_info
*seg
)
1092 if (seg
->alloc_type
== SSR
)
1093 __next_free_blkoff(sbi
, seg
, seg
->next_blkoff
+ 1);
1099 * This function always allocates a used segment(from dirty seglist) by SSR
1100 * manner, so it should recover the existing segment information of valid blocks
1102 static void change_curseg(struct f2fs_sb_info
*sbi
, int type
, bool reuse
)
1104 struct dirty_seglist_info
*dirty_i
= DIRTY_I(sbi
);
1105 struct curseg_info
*curseg
= CURSEG_I(sbi
, type
);
1106 unsigned int new_segno
= curseg
->next_segno
;
1107 struct f2fs_summary_block
*sum_node
;
1108 struct page
*sum_page
;
1110 write_sum_page(sbi
, curseg
->sum_blk
,
1111 GET_SUM_BLOCK(sbi
, curseg
->segno
));
1112 __set_test_and_inuse(sbi
, new_segno
);
1114 mutex_lock(&dirty_i
->seglist_lock
);
1115 __remove_dirty_segment(sbi
, new_segno
, PRE
);
1116 __remove_dirty_segment(sbi
, new_segno
, DIRTY
);
1117 mutex_unlock(&dirty_i
->seglist_lock
);
1119 reset_curseg(sbi
, type
, 1);
1120 curseg
->alloc_type
= SSR
;
1121 __next_free_blkoff(sbi
, curseg
, 0);
1124 sum_page
= get_sum_page(sbi
, new_segno
);
1125 sum_node
= (struct f2fs_summary_block
*)page_address(sum_page
);
1126 memcpy(curseg
->sum_blk
, sum_node
, SUM_ENTRY_SIZE
);
1127 f2fs_put_page(sum_page
, 1);
1131 static int get_ssr_segment(struct f2fs_sb_info
*sbi
, int type
)
1133 struct curseg_info
*curseg
= CURSEG_I(sbi
, type
);
1134 const struct victim_selection
*v_ops
= DIRTY_I(sbi
)->v_ops
;
1136 if (IS_NODESEG(type
) || !has_not_enough_free_secs(sbi
, 0))
1137 return v_ops
->get_victim(sbi
,
1138 &(curseg
)->next_segno
, BG_GC
, type
, SSR
);
1140 /* For data segments, let's do SSR more intensively */
1141 for (; type
>= CURSEG_HOT_DATA
; type
--)
1142 if (v_ops
->get_victim(sbi
, &(curseg
)->next_segno
,
1149 * flush out current segment and replace it with new segment
1150 * This function should be returned with success, otherwise BUG
1152 static void allocate_segment_by_default(struct f2fs_sb_info
*sbi
,
1153 int type
, bool force
)
1155 struct curseg_info
*curseg
= CURSEG_I(sbi
, type
);
1158 new_curseg(sbi
, type
, true);
1159 else if (type
== CURSEG_WARM_NODE
)
1160 new_curseg(sbi
, type
, false);
1161 else if (curseg
->alloc_type
== LFS
&& is_next_segment_free(sbi
, type
))
1162 new_curseg(sbi
, type
, false);
1163 else if (need_SSR(sbi
) && get_ssr_segment(sbi
, type
))
1164 change_curseg(sbi
, type
, true);
1166 new_curseg(sbi
, type
, false);
1168 stat_inc_seg_type(sbi
, curseg
);
1171 static void __allocate_new_segments(struct f2fs_sb_info
*sbi
, int type
)
1173 struct curseg_info
*curseg
= CURSEG_I(sbi
, type
);
1174 unsigned int old_segno
;
1176 old_segno
= curseg
->segno
;
1177 SIT_I(sbi
)->s_ops
->allocate_segment(sbi
, type
, true);
1178 locate_dirty_segment(sbi
, old_segno
);
1181 void allocate_new_segments(struct f2fs_sb_info
*sbi
)
1185 for (i
= CURSEG_HOT_DATA
; i
<= CURSEG_COLD_DATA
; i
++)
1186 __allocate_new_segments(sbi
, i
);
1189 static const struct segment_allocation default_salloc_ops
= {
1190 .allocate_segment
= allocate_segment_by_default
,
1193 int f2fs_trim_fs(struct f2fs_sb_info
*sbi
, struct fstrim_range
*range
)
1195 __u64 start
= F2FS_BYTES_TO_BLK(range
->start
);
1196 __u64 end
= start
+ F2FS_BYTES_TO_BLK(range
->len
) - 1;
1197 unsigned int start_segno
, end_segno
;
1198 struct cp_control cpc
;
1201 if (start
>= MAX_BLKADDR(sbi
) || range
->len
< sbi
->blocksize
)
1205 if (end
<= MAIN_BLKADDR(sbi
))
1208 /* start/end segment number in main_area */
1209 start_segno
= (start
<= MAIN_BLKADDR(sbi
)) ? 0 : GET_SEGNO(sbi
, start
);
1210 end_segno
= (end
>= MAX_BLKADDR(sbi
)) ? MAIN_SEGS(sbi
) - 1 :
1211 GET_SEGNO(sbi
, end
);
1212 cpc
.reason
= CP_DISCARD
;
1213 cpc
.trim_minlen
= max_t(__u64
, 1, F2FS_BYTES_TO_BLK(range
->minlen
));
1215 /* do checkpoint to issue discard commands safely */
1216 for (; start_segno
<= end_segno
; start_segno
= cpc
.trim_end
+ 1) {
1217 cpc
.trim_start
= start_segno
;
1219 if (sbi
->discard_blks
== 0)
1221 else if (sbi
->discard_blks
< BATCHED_TRIM_BLOCKS(sbi
))
1222 cpc
.trim_end
= end_segno
;
1224 cpc
.trim_end
= min_t(unsigned int,
1225 rounddown(start_segno
+
1226 BATCHED_TRIM_SEGMENTS(sbi
),
1227 sbi
->segs_per_sec
) - 1, end_segno
);
1229 mutex_lock(&sbi
->gc_mutex
);
1230 err
= write_checkpoint(sbi
, &cpc
);
1231 mutex_unlock(&sbi
->gc_mutex
);
1234 range
->len
= F2FS_BLK_TO_BYTES(cpc
.trimmed
);
1238 static bool __has_curseg_space(struct f2fs_sb_info
*sbi
, int type
)
1240 struct curseg_info
*curseg
= CURSEG_I(sbi
, type
);
1241 if (curseg
->next_blkoff
< sbi
->blocks_per_seg
)
1246 static int __get_segment_type_2(struct page
*page
, enum page_type p_type
)
1249 return CURSEG_HOT_DATA
;
1251 return CURSEG_HOT_NODE
;
1254 static int __get_segment_type_4(struct page
*page
, enum page_type p_type
)
1256 if (p_type
== DATA
) {
1257 struct inode
*inode
= page
->mapping
->host
;
1259 if (S_ISDIR(inode
->i_mode
))
1260 return CURSEG_HOT_DATA
;
1262 return CURSEG_COLD_DATA
;
1264 if (IS_DNODE(page
) && is_cold_node(page
))
1265 return CURSEG_WARM_NODE
;
1267 return CURSEG_COLD_NODE
;
1271 static int __get_segment_type_6(struct page
*page
, enum page_type p_type
)
1273 if (p_type
== DATA
) {
1274 struct inode
*inode
= page
->mapping
->host
;
1276 if (S_ISDIR(inode
->i_mode
))
1277 return CURSEG_HOT_DATA
;
1278 else if (is_cold_data(page
) || file_is_cold(inode
))
1279 return CURSEG_COLD_DATA
;
1281 return CURSEG_WARM_DATA
;
1284 return is_cold_node(page
) ? CURSEG_WARM_NODE
:
1287 return CURSEG_COLD_NODE
;
1291 static int __get_segment_type(struct page
*page
, enum page_type p_type
)
1293 switch (F2FS_P_SB(page
)->active_logs
) {
1295 return __get_segment_type_2(page
, p_type
);
1297 return __get_segment_type_4(page
, p_type
);
1299 /* NR_CURSEG_TYPE(6) logs by default */
1300 f2fs_bug_on(F2FS_P_SB(page
),
1301 F2FS_P_SB(page
)->active_logs
!= NR_CURSEG_TYPE
);
1302 return __get_segment_type_6(page
, p_type
);
1305 void allocate_data_block(struct f2fs_sb_info
*sbi
, struct page
*page
,
1306 block_t old_blkaddr
, block_t
*new_blkaddr
,
1307 struct f2fs_summary
*sum
, int type
)
1309 struct sit_info
*sit_i
= SIT_I(sbi
);
1310 struct curseg_info
*curseg
;
1311 bool direct_io
= (type
== CURSEG_DIRECT_IO
);
1313 type
= direct_io
? CURSEG_WARM_DATA
: type
;
1315 curseg
= CURSEG_I(sbi
, type
);
1317 mutex_lock(&curseg
->curseg_mutex
);
1318 mutex_lock(&sit_i
->sentry_lock
);
1320 /* direct_io'ed data is aligned to the segment for better performance */
1321 if (direct_io
&& curseg
->next_blkoff
&&
1322 !has_not_enough_free_secs(sbi
, 0))
1323 __allocate_new_segments(sbi
, type
);
1325 *new_blkaddr
= NEXT_FREE_BLKADDR(sbi
, curseg
);
1328 * __add_sum_entry should be resided under the curseg_mutex
1329 * because, this function updates a summary entry in the
1330 * current summary block.
1332 __add_sum_entry(sbi
, type
, sum
);
1334 __refresh_next_blkoff(sbi
, curseg
);
1336 stat_inc_block_count(sbi
, curseg
);
1338 if (!__has_curseg_space(sbi
, type
))
1339 sit_i
->s_ops
->allocate_segment(sbi
, type
, false);
1341 * SIT information should be updated before segment allocation,
1342 * since SSR needs latest valid block information.
1344 refresh_sit_entry(sbi
, old_blkaddr
, *new_blkaddr
);
1346 mutex_unlock(&sit_i
->sentry_lock
);
1348 if (page
&& IS_NODESEG(type
))
1349 fill_node_footer_blkaddr(page
, NEXT_FREE_BLKADDR(sbi
, curseg
));
1351 mutex_unlock(&curseg
->curseg_mutex
);
1354 static void do_write_page(struct f2fs_summary
*sum
, struct f2fs_io_info
*fio
)
1356 int type
= __get_segment_type(fio
->page
, fio
->type
);
1358 allocate_data_block(fio
->sbi
, fio
->page
, fio
->blk_addr
,
1359 &fio
->blk_addr
, sum
, type
);
1361 /* writeout dirty page into bdev */
1362 f2fs_submit_page_mbio(fio
);
1365 void write_meta_page(struct f2fs_sb_info
*sbi
, struct page
*page
)
1367 struct f2fs_io_info fio
= {
1370 .rw
= WRITE_SYNC
| REQ_META
| REQ_PRIO
,
1371 .blk_addr
= page
->index
,
1373 .encrypted_page
= NULL
,
1376 if (unlikely(page
->index
>= MAIN_BLKADDR(sbi
)))
1377 fio
.rw
&= ~REQ_META
;
1379 set_page_writeback(page
);
1380 f2fs_submit_page_mbio(&fio
);
1383 void write_node_page(unsigned int nid
, struct f2fs_io_info
*fio
)
1385 struct f2fs_summary sum
;
1387 set_summary(&sum
, nid
, 0, 0);
1388 do_write_page(&sum
, fio
);
1391 void write_data_page(struct dnode_of_data
*dn
, struct f2fs_io_info
*fio
)
1393 struct f2fs_sb_info
*sbi
= fio
->sbi
;
1394 struct f2fs_summary sum
;
1395 struct node_info ni
;
1397 f2fs_bug_on(sbi
, dn
->data_blkaddr
== NULL_ADDR
);
1398 get_node_info(sbi
, dn
->nid
, &ni
);
1399 set_summary(&sum
, dn
->nid
, dn
->ofs_in_node
, ni
.version
);
1400 do_write_page(&sum
, fio
);
1401 dn
->data_blkaddr
= fio
->blk_addr
;
1404 void rewrite_data_page(struct f2fs_io_info
*fio
)
1406 stat_inc_inplace_blocks(fio
->sbi
);
1407 f2fs_submit_page_mbio(fio
);
1410 static void __f2fs_replace_block(struct f2fs_sb_info
*sbi
,
1411 struct f2fs_summary
*sum
,
1412 block_t old_blkaddr
, block_t new_blkaddr
,
1413 bool recover_curseg
, bool recover_newaddr
)
1415 struct sit_info
*sit_i
= SIT_I(sbi
);
1416 struct curseg_info
*curseg
;
1417 unsigned int segno
, old_cursegno
;
1418 struct seg_entry
*se
;
1420 unsigned short old_blkoff
;
1422 segno
= GET_SEGNO(sbi
, new_blkaddr
);
1423 se
= get_seg_entry(sbi
, segno
);
1426 if (!recover_curseg
) {
1427 /* for recovery flow */
1428 if (se
->valid_blocks
== 0 && !IS_CURSEG(sbi
, segno
)) {
1429 if (old_blkaddr
== NULL_ADDR
)
1430 type
= CURSEG_COLD_DATA
;
1432 type
= CURSEG_WARM_DATA
;
1435 if (!IS_CURSEG(sbi
, segno
))
1436 type
= CURSEG_WARM_DATA
;
1439 curseg
= CURSEG_I(sbi
, type
);
1441 mutex_lock(&curseg
->curseg_mutex
);
1442 mutex_lock(&sit_i
->sentry_lock
);
1444 old_cursegno
= curseg
->segno
;
1445 old_blkoff
= curseg
->next_blkoff
;
1447 /* change the current segment */
1448 if (segno
!= curseg
->segno
) {
1449 curseg
->next_segno
= segno
;
1450 change_curseg(sbi
, type
, true);
1453 curseg
->next_blkoff
= GET_BLKOFF_FROM_SEG0(sbi
, new_blkaddr
);
1454 __add_sum_entry(sbi
, type
, sum
);
1456 if (!recover_curseg
|| recover_newaddr
)
1457 update_sit_entry(sbi
, new_blkaddr
, 1);
1458 if (GET_SEGNO(sbi
, old_blkaddr
) != NULL_SEGNO
)
1459 update_sit_entry(sbi
, old_blkaddr
, -1);
1461 locate_dirty_segment(sbi
, GET_SEGNO(sbi
, old_blkaddr
));
1462 locate_dirty_segment(sbi
, GET_SEGNO(sbi
, new_blkaddr
));
1464 locate_dirty_segment(sbi
, old_cursegno
);
1466 if (recover_curseg
) {
1467 if (old_cursegno
!= curseg
->segno
) {
1468 curseg
->next_segno
= old_cursegno
;
1469 change_curseg(sbi
, type
, true);
1471 curseg
->next_blkoff
= old_blkoff
;
1474 mutex_unlock(&sit_i
->sentry_lock
);
1475 mutex_unlock(&curseg
->curseg_mutex
);
1478 void f2fs_replace_block(struct f2fs_sb_info
*sbi
, struct dnode_of_data
*dn
,
1479 block_t old_addr
, block_t new_addr
,
1480 unsigned char version
, bool recover_curseg
,
1481 bool recover_newaddr
)
1483 struct f2fs_summary sum
;
1485 set_summary(&sum
, dn
->nid
, dn
->ofs_in_node
, version
);
1487 __f2fs_replace_block(sbi
, &sum
, old_addr
, new_addr
,
1488 recover_curseg
, recover_newaddr
);
1490 dn
->data_blkaddr
= new_addr
;
1491 set_data_blkaddr(dn
);
1492 f2fs_update_extent_cache(dn
);
1495 void f2fs_wait_on_page_writeback(struct page
*page
,
1496 enum page_type type
, bool ordered
)
1498 if (PageWriteback(page
)) {
1499 struct f2fs_sb_info
*sbi
= F2FS_P_SB(page
);
1501 f2fs_submit_merged_bio_cond(sbi
, NULL
, page
, 0, type
, WRITE
);
1503 wait_on_page_writeback(page
);
1505 wait_for_stable_page(page
);
1509 void f2fs_wait_on_encrypted_page_writeback(struct f2fs_sb_info
*sbi
,
1514 if (blkaddr
== NEW_ADDR
)
1517 f2fs_bug_on(sbi
, blkaddr
== NULL_ADDR
);
1519 cpage
= find_lock_page(META_MAPPING(sbi
), blkaddr
);
1521 f2fs_wait_on_page_writeback(cpage
, DATA
, true);
1522 f2fs_put_page(cpage
, 1);
1526 static int read_compacted_summaries(struct f2fs_sb_info
*sbi
)
1528 struct f2fs_checkpoint
*ckpt
= F2FS_CKPT(sbi
);
1529 struct curseg_info
*seg_i
;
1530 unsigned char *kaddr
;
1535 start
= start_sum_block(sbi
);
1537 page
= get_meta_page(sbi
, start
++);
1538 kaddr
= (unsigned char *)page_address(page
);
1540 /* Step 1: restore nat cache */
1541 seg_i
= CURSEG_I(sbi
, CURSEG_HOT_DATA
);
1542 memcpy(&seg_i
->sum_blk
->journal
.n_nats
, kaddr
, SUM_JOURNAL_SIZE
);
1544 /* Step 2: restore sit cache */
1545 seg_i
= CURSEG_I(sbi
, CURSEG_COLD_DATA
);
1546 memcpy(&seg_i
->sum_blk
->journal
.n_sits
, kaddr
+ SUM_JOURNAL_SIZE
,
1548 offset
= 2 * SUM_JOURNAL_SIZE
;
1550 /* Step 3: restore summary entries */
1551 for (i
= CURSEG_HOT_DATA
; i
<= CURSEG_COLD_DATA
; i
++) {
1552 unsigned short blk_off
;
1555 seg_i
= CURSEG_I(sbi
, i
);
1556 segno
= le32_to_cpu(ckpt
->cur_data_segno
[i
]);
1557 blk_off
= le16_to_cpu(ckpt
->cur_data_blkoff
[i
]);
1558 seg_i
->next_segno
= segno
;
1559 reset_curseg(sbi
, i
, 0);
1560 seg_i
->alloc_type
= ckpt
->alloc_type
[i
];
1561 seg_i
->next_blkoff
= blk_off
;
1563 if (seg_i
->alloc_type
== SSR
)
1564 blk_off
= sbi
->blocks_per_seg
;
1566 for (j
= 0; j
< blk_off
; j
++) {
1567 struct f2fs_summary
*s
;
1568 s
= (struct f2fs_summary
*)(kaddr
+ offset
);
1569 seg_i
->sum_blk
->entries
[j
] = *s
;
1570 offset
+= SUMMARY_SIZE
;
1571 if (offset
+ SUMMARY_SIZE
<= PAGE_CACHE_SIZE
-
1575 f2fs_put_page(page
, 1);
1578 page
= get_meta_page(sbi
, start
++);
1579 kaddr
= (unsigned char *)page_address(page
);
1583 f2fs_put_page(page
, 1);
1587 static int read_normal_summaries(struct f2fs_sb_info
*sbi
, int type
)
1589 struct f2fs_checkpoint
*ckpt
= F2FS_CKPT(sbi
);
1590 struct f2fs_summary_block
*sum
;
1591 struct curseg_info
*curseg
;
1593 unsigned short blk_off
;
1594 unsigned int segno
= 0;
1595 block_t blk_addr
= 0;
1597 /* get segment number and block addr */
1598 if (IS_DATASEG(type
)) {
1599 segno
= le32_to_cpu(ckpt
->cur_data_segno
[type
]);
1600 blk_off
= le16_to_cpu(ckpt
->cur_data_blkoff
[type
-
1602 if (__exist_node_summaries(sbi
))
1603 blk_addr
= sum_blk_addr(sbi
, NR_CURSEG_TYPE
, type
);
1605 blk_addr
= sum_blk_addr(sbi
, NR_CURSEG_DATA_TYPE
, type
);
1607 segno
= le32_to_cpu(ckpt
->cur_node_segno
[type
-
1609 blk_off
= le16_to_cpu(ckpt
->cur_node_blkoff
[type
-
1611 if (__exist_node_summaries(sbi
))
1612 blk_addr
= sum_blk_addr(sbi
, NR_CURSEG_NODE_TYPE
,
1613 type
- CURSEG_HOT_NODE
);
1615 blk_addr
= GET_SUM_BLOCK(sbi
, segno
);
1618 new = get_meta_page(sbi
, blk_addr
);
1619 sum
= (struct f2fs_summary_block
*)page_address(new);
1621 if (IS_NODESEG(type
)) {
1622 if (__exist_node_summaries(sbi
)) {
1623 struct f2fs_summary
*ns
= &sum
->entries
[0];
1625 for (i
= 0; i
< sbi
->blocks_per_seg
; i
++, ns
++) {
1627 ns
->ofs_in_node
= 0;
1632 err
= restore_node_summary(sbi
, segno
, sum
);
1634 f2fs_put_page(new, 1);
1640 /* set uncompleted segment to curseg */
1641 curseg
= CURSEG_I(sbi
, type
);
1642 mutex_lock(&curseg
->curseg_mutex
);
1643 memcpy(curseg
->sum_blk
, sum
, PAGE_CACHE_SIZE
);
1644 curseg
->next_segno
= segno
;
1645 reset_curseg(sbi
, type
, 0);
1646 curseg
->alloc_type
= ckpt
->alloc_type
[type
];
1647 curseg
->next_blkoff
= blk_off
;
1648 mutex_unlock(&curseg
->curseg_mutex
);
1649 f2fs_put_page(new, 1);
1653 static int restore_curseg_summaries(struct f2fs_sb_info
*sbi
)
1655 int type
= CURSEG_HOT_DATA
;
1658 if (is_set_ckpt_flags(F2FS_CKPT(sbi
), CP_COMPACT_SUM_FLAG
)) {
1659 int npages
= npages_for_summary_flush(sbi
, true);
1662 ra_meta_pages(sbi
, start_sum_block(sbi
), npages
,
1665 /* restore for compacted data summary */
1666 if (read_compacted_summaries(sbi
))
1668 type
= CURSEG_HOT_NODE
;
1671 if (__exist_node_summaries(sbi
))
1672 ra_meta_pages(sbi
, sum_blk_addr(sbi
, NR_CURSEG_TYPE
, type
),
1673 NR_CURSEG_TYPE
- type
, META_CP
, true);
1675 for (; type
<= CURSEG_COLD_NODE
; type
++) {
1676 err
= read_normal_summaries(sbi
, type
);
1684 static void write_compacted_summaries(struct f2fs_sb_info
*sbi
, block_t blkaddr
)
1687 unsigned char *kaddr
;
1688 struct f2fs_summary
*summary
;
1689 struct curseg_info
*seg_i
;
1690 int written_size
= 0;
1693 page
= grab_meta_page(sbi
, blkaddr
++);
1694 kaddr
= (unsigned char *)page_address(page
);
1696 /* Step 1: write nat cache */
1697 seg_i
= CURSEG_I(sbi
, CURSEG_HOT_DATA
);
1698 memcpy(kaddr
, &seg_i
->sum_blk
->journal
.n_nats
, SUM_JOURNAL_SIZE
);
1699 written_size
+= SUM_JOURNAL_SIZE
;
1701 /* Step 2: write sit cache */
1702 seg_i
= CURSEG_I(sbi
, CURSEG_COLD_DATA
);
1703 memcpy(kaddr
+ written_size
, &seg_i
->sum_blk
->journal
.n_sits
,
1705 written_size
+= SUM_JOURNAL_SIZE
;
1707 /* Step 3: write summary entries */
1708 for (i
= CURSEG_HOT_DATA
; i
<= CURSEG_COLD_DATA
; i
++) {
1709 unsigned short blkoff
;
1710 seg_i
= CURSEG_I(sbi
, i
);
1711 if (sbi
->ckpt
->alloc_type
[i
] == SSR
)
1712 blkoff
= sbi
->blocks_per_seg
;
1714 blkoff
= curseg_blkoff(sbi
, i
);
1716 for (j
= 0; j
< blkoff
; j
++) {
1718 page
= grab_meta_page(sbi
, blkaddr
++);
1719 kaddr
= (unsigned char *)page_address(page
);
1722 summary
= (struct f2fs_summary
*)(kaddr
+ written_size
);
1723 *summary
= seg_i
->sum_blk
->entries
[j
];
1724 written_size
+= SUMMARY_SIZE
;
1726 if (written_size
+ SUMMARY_SIZE
<= PAGE_CACHE_SIZE
-
1730 set_page_dirty(page
);
1731 f2fs_put_page(page
, 1);
1736 set_page_dirty(page
);
1737 f2fs_put_page(page
, 1);
1741 static void write_normal_summaries(struct f2fs_sb_info
*sbi
,
1742 block_t blkaddr
, int type
)
1745 if (IS_DATASEG(type
))
1746 end
= type
+ NR_CURSEG_DATA_TYPE
;
1748 end
= type
+ NR_CURSEG_NODE_TYPE
;
1750 for (i
= type
; i
< end
; i
++) {
1751 struct curseg_info
*sum
= CURSEG_I(sbi
, i
);
1752 mutex_lock(&sum
->curseg_mutex
);
1753 write_sum_page(sbi
, sum
->sum_blk
, blkaddr
+ (i
- type
));
1754 mutex_unlock(&sum
->curseg_mutex
);
1758 void write_data_summaries(struct f2fs_sb_info
*sbi
, block_t start_blk
)
1760 if (is_set_ckpt_flags(F2FS_CKPT(sbi
), CP_COMPACT_SUM_FLAG
))
1761 write_compacted_summaries(sbi
, start_blk
);
1763 write_normal_summaries(sbi
, start_blk
, CURSEG_HOT_DATA
);
1766 void write_node_summaries(struct f2fs_sb_info
*sbi
, block_t start_blk
)
1768 write_normal_summaries(sbi
, start_blk
, CURSEG_HOT_NODE
);
1771 int lookup_journal_in_cursum(struct f2fs_journal
*journal
, int type
,
1772 unsigned int val
, int alloc
)
1776 if (type
== NAT_JOURNAL
) {
1777 for (i
= 0; i
< nats_in_cursum(journal
); i
++) {
1778 if (le32_to_cpu(nid_in_journal(journal
, i
)) == val
)
1781 if (alloc
&& __has_cursum_space(journal
, 1, NAT_JOURNAL
))
1782 return update_nats_in_cursum(journal
, 1);
1783 } else if (type
== SIT_JOURNAL
) {
1784 for (i
= 0; i
< sits_in_cursum(journal
); i
++)
1785 if (le32_to_cpu(segno_in_journal(journal
, i
)) == val
)
1787 if (alloc
&& __has_cursum_space(journal
, 1, SIT_JOURNAL
))
1788 return update_sits_in_cursum(journal
, 1);
1793 static struct page
*get_current_sit_page(struct f2fs_sb_info
*sbi
,
1796 return get_meta_page(sbi
, current_sit_addr(sbi
, segno
));
1799 static struct page
*get_next_sit_page(struct f2fs_sb_info
*sbi
,
1802 struct sit_info
*sit_i
= SIT_I(sbi
);
1803 struct page
*src_page
, *dst_page
;
1804 pgoff_t src_off
, dst_off
;
1805 void *src_addr
, *dst_addr
;
1807 src_off
= current_sit_addr(sbi
, start
);
1808 dst_off
= next_sit_addr(sbi
, src_off
);
1810 /* get current sit block page without lock */
1811 src_page
= get_meta_page(sbi
, src_off
);
1812 dst_page
= grab_meta_page(sbi
, dst_off
);
1813 f2fs_bug_on(sbi
, PageDirty(src_page
));
1815 src_addr
= page_address(src_page
);
1816 dst_addr
= page_address(dst_page
);
1817 memcpy(dst_addr
, src_addr
, PAGE_CACHE_SIZE
);
1819 set_page_dirty(dst_page
);
1820 f2fs_put_page(src_page
, 1);
1822 set_to_next_sit(sit_i
, start
);
1827 static struct sit_entry_set
*grab_sit_entry_set(void)
1829 struct sit_entry_set
*ses
=
1830 f2fs_kmem_cache_alloc(sit_entry_set_slab
, GFP_NOFS
);
1833 INIT_LIST_HEAD(&ses
->set_list
);
1837 static void release_sit_entry_set(struct sit_entry_set
*ses
)
1839 list_del(&ses
->set_list
);
1840 kmem_cache_free(sit_entry_set_slab
, ses
);
1843 static void adjust_sit_entry_set(struct sit_entry_set
*ses
,
1844 struct list_head
*head
)
1846 struct sit_entry_set
*next
= ses
;
1848 if (list_is_last(&ses
->set_list
, head
))
1851 list_for_each_entry_continue(next
, head
, set_list
)
1852 if (ses
->entry_cnt
<= next
->entry_cnt
)
1855 list_move_tail(&ses
->set_list
, &next
->set_list
);
1858 static void add_sit_entry(unsigned int segno
, struct list_head
*head
)
1860 struct sit_entry_set
*ses
;
1861 unsigned int start_segno
= START_SEGNO(segno
);
1863 list_for_each_entry(ses
, head
, set_list
) {
1864 if (ses
->start_segno
== start_segno
) {
1866 adjust_sit_entry_set(ses
, head
);
1871 ses
= grab_sit_entry_set();
1873 ses
->start_segno
= start_segno
;
1875 list_add(&ses
->set_list
, head
);
1878 static void add_sits_in_set(struct f2fs_sb_info
*sbi
)
1880 struct f2fs_sm_info
*sm_info
= SM_I(sbi
);
1881 struct list_head
*set_list
= &sm_info
->sit_entry_set
;
1882 unsigned long *bitmap
= SIT_I(sbi
)->dirty_sentries_bitmap
;
1885 for_each_set_bit(segno
, bitmap
, MAIN_SEGS(sbi
))
1886 add_sit_entry(segno
, set_list
);
1889 static void remove_sits_in_journal(struct f2fs_sb_info
*sbi
)
1891 struct curseg_info
*curseg
= CURSEG_I(sbi
, CURSEG_COLD_DATA
);
1892 struct f2fs_journal
*journal
= &curseg
->sum_blk
->journal
;
1895 for (i
= 0; i
< sits_in_cursum(journal
); i
++) {
1899 segno
= le32_to_cpu(segno_in_journal(journal
, i
));
1900 dirtied
= __mark_sit_entry_dirty(sbi
, segno
);
1903 add_sit_entry(segno
, &SM_I(sbi
)->sit_entry_set
);
1905 update_sits_in_cursum(journal
, -i
);
1909 * CP calls this function, which flushes SIT entries including sit_journal,
1910 * and moves prefree segs to free segs.
1912 void flush_sit_entries(struct f2fs_sb_info
*sbi
, struct cp_control
*cpc
)
1914 struct sit_info
*sit_i
= SIT_I(sbi
);
1915 unsigned long *bitmap
= sit_i
->dirty_sentries_bitmap
;
1916 struct curseg_info
*curseg
= CURSEG_I(sbi
, CURSEG_COLD_DATA
);
1917 struct f2fs_journal
*journal
= &curseg
->sum_blk
->journal
;
1918 struct sit_entry_set
*ses
, *tmp
;
1919 struct list_head
*head
= &SM_I(sbi
)->sit_entry_set
;
1920 bool to_journal
= true;
1921 struct seg_entry
*se
;
1923 mutex_lock(&curseg
->curseg_mutex
);
1924 mutex_lock(&sit_i
->sentry_lock
);
1926 if (!sit_i
->dirty_sentries
)
1930 * add and account sit entries of dirty bitmap in sit entry
1933 add_sits_in_set(sbi
);
1936 * if there are no enough space in journal to store dirty sit
1937 * entries, remove all entries from journal and add and account
1938 * them in sit entry set.
1940 if (!__has_cursum_space(journal
, sit_i
->dirty_sentries
, SIT_JOURNAL
))
1941 remove_sits_in_journal(sbi
);
1944 * there are two steps to flush sit entries:
1945 * #1, flush sit entries to journal in current cold data summary block.
1946 * #2, flush sit entries to sit page.
1948 list_for_each_entry_safe(ses
, tmp
, head
, set_list
) {
1949 struct page
*page
= NULL
;
1950 struct f2fs_sit_block
*raw_sit
= NULL
;
1951 unsigned int start_segno
= ses
->start_segno
;
1952 unsigned int end
= min(start_segno
+ SIT_ENTRY_PER_BLOCK
,
1953 (unsigned long)MAIN_SEGS(sbi
));
1954 unsigned int segno
= start_segno
;
1957 !__has_cursum_space(journal
, ses
->entry_cnt
, SIT_JOURNAL
))
1961 page
= get_next_sit_page(sbi
, start_segno
);
1962 raw_sit
= page_address(page
);
1965 /* flush dirty sit entries in region of current sit set */
1966 for_each_set_bit_from(segno
, bitmap
, end
) {
1967 int offset
, sit_offset
;
1969 se
= get_seg_entry(sbi
, segno
);
1971 /* add discard candidates */
1972 if (cpc
->reason
!= CP_DISCARD
) {
1973 cpc
->trim_start
= segno
;
1974 add_discard_addrs(sbi
, cpc
);
1978 offset
= lookup_journal_in_cursum(journal
,
1979 SIT_JOURNAL
, segno
, 1);
1980 f2fs_bug_on(sbi
, offset
< 0);
1981 segno_in_journal(journal
, offset
) =
1983 seg_info_to_raw_sit(se
,
1984 &sit_in_journal(journal
, offset
));
1986 sit_offset
= SIT_ENTRY_OFFSET(sit_i
, segno
);
1987 seg_info_to_raw_sit(se
,
1988 &raw_sit
->entries
[sit_offset
]);
1991 __clear_bit(segno
, bitmap
);
1992 sit_i
->dirty_sentries
--;
1997 f2fs_put_page(page
, 1);
1999 f2fs_bug_on(sbi
, ses
->entry_cnt
);
2000 release_sit_entry_set(ses
);
2003 f2fs_bug_on(sbi
, !list_empty(head
));
2004 f2fs_bug_on(sbi
, sit_i
->dirty_sentries
);
2006 if (cpc
->reason
== CP_DISCARD
) {
2007 for (; cpc
->trim_start
<= cpc
->trim_end
; cpc
->trim_start
++)
2008 add_discard_addrs(sbi
, cpc
);
2010 mutex_unlock(&sit_i
->sentry_lock
);
2011 mutex_unlock(&curseg
->curseg_mutex
);
2013 set_prefree_as_free_segments(sbi
);
2016 static int build_sit_info(struct f2fs_sb_info
*sbi
)
2018 struct f2fs_super_block
*raw_super
= F2FS_RAW_SUPER(sbi
);
2019 struct f2fs_checkpoint
*ckpt
= F2FS_CKPT(sbi
);
2020 struct sit_info
*sit_i
;
2021 unsigned int sit_segs
, start
;
2022 char *src_bitmap
, *dst_bitmap
;
2023 unsigned int bitmap_size
;
2025 /* allocate memory for SIT information */
2026 sit_i
= kzalloc(sizeof(struct sit_info
), GFP_KERNEL
);
2030 SM_I(sbi
)->sit_info
= sit_i
;
2032 sit_i
->sentries
= f2fs_kvzalloc(MAIN_SEGS(sbi
) *
2033 sizeof(struct seg_entry
), GFP_KERNEL
);
2034 if (!sit_i
->sentries
)
2037 bitmap_size
= f2fs_bitmap_size(MAIN_SEGS(sbi
));
2038 sit_i
->dirty_sentries_bitmap
= f2fs_kvzalloc(bitmap_size
, GFP_KERNEL
);
2039 if (!sit_i
->dirty_sentries_bitmap
)
2042 for (start
= 0; start
< MAIN_SEGS(sbi
); start
++) {
2043 sit_i
->sentries
[start
].cur_valid_map
2044 = kzalloc(SIT_VBLOCK_MAP_SIZE
, GFP_KERNEL
);
2045 sit_i
->sentries
[start
].ckpt_valid_map
2046 = kzalloc(SIT_VBLOCK_MAP_SIZE
, GFP_KERNEL
);
2047 sit_i
->sentries
[start
].discard_map
2048 = kzalloc(SIT_VBLOCK_MAP_SIZE
, GFP_KERNEL
);
2049 if (!sit_i
->sentries
[start
].cur_valid_map
||
2050 !sit_i
->sentries
[start
].ckpt_valid_map
||
2051 !sit_i
->sentries
[start
].discard_map
)
2055 sit_i
->tmp_map
= kzalloc(SIT_VBLOCK_MAP_SIZE
, GFP_KERNEL
);
2056 if (!sit_i
->tmp_map
)
2059 if (sbi
->segs_per_sec
> 1) {
2060 sit_i
->sec_entries
= f2fs_kvzalloc(MAIN_SECS(sbi
) *
2061 sizeof(struct sec_entry
), GFP_KERNEL
);
2062 if (!sit_i
->sec_entries
)
2066 /* get information related with SIT */
2067 sit_segs
= le32_to_cpu(raw_super
->segment_count_sit
) >> 1;
2069 /* setup SIT bitmap from ckeckpoint pack */
2070 bitmap_size
= __bitmap_size(sbi
, SIT_BITMAP
);
2071 src_bitmap
= __bitmap_ptr(sbi
, SIT_BITMAP
);
2073 dst_bitmap
= kmemdup(src_bitmap
, bitmap_size
, GFP_KERNEL
);
2077 /* init SIT information */
2078 sit_i
->s_ops
= &default_salloc_ops
;
2080 sit_i
->sit_base_addr
= le32_to_cpu(raw_super
->sit_blkaddr
);
2081 sit_i
->sit_blocks
= sit_segs
<< sbi
->log_blocks_per_seg
;
2082 sit_i
->written_valid_blocks
= le64_to_cpu(ckpt
->valid_block_count
);
2083 sit_i
->sit_bitmap
= dst_bitmap
;
2084 sit_i
->bitmap_size
= bitmap_size
;
2085 sit_i
->dirty_sentries
= 0;
2086 sit_i
->sents_per_block
= SIT_ENTRY_PER_BLOCK
;
2087 sit_i
->elapsed_time
= le64_to_cpu(sbi
->ckpt
->elapsed_time
);
2088 sit_i
->mounted_time
= CURRENT_TIME_SEC
.tv_sec
;
2089 mutex_init(&sit_i
->sentry_lock
);
2093 static int build_free_segmap(struct f2fs_sb_info
*sbi
)
2095 struct free_segmap_info
*free_i
;
2096 unsigned int bitmap_size
, sec_bitmap_size
;
2098 /* allocate memory for free segmap information */
2099 free_i
= kzalloc(sizeof(struct free_segmap_info
), GFP_KERNEL
);
2103 SM_I(sbi
)->free_info
= free_i
;
2105 bitmap_size
= f2fs_bitmap_size(MAIN_SEGS(sbi
));
2106 free_i
->free_segmap
= f2fs_kvmalloc(bitmap_size
, GFP_KERNEL
);
2107 if (!free_i
->free_segmap
)
2110 sec_bitmap_size
= f2fs_bitmap_size(MAIN_SECS(sbi
));
2111 free_i
->free_secmap
= f2fs_kvmalloc(sec_bitmap_size
, GFP_KERNEL
);
2112 if (!free_i
->free_secmap
)
2115 /* set all segments as dirty temporarily */
2116 memset(free_i
->free_segmap
, 0xff, bitmap_size
);
2117 memset(free_i
->free_secmap
, 0xff, sec_bitmap_size
);
2119 /* init free segmap information */
2120 free_i
->start_segno
= GET_SEGNO_FROM_SEG0(sbi
, MAIN_BLKADDR(sbi
));
2121 free_i
->free_segments
= 0;
2122 free_i
->free_sections
= 0;
2123 spin_lock_init(&free_i
->segmap_lock
);
2127 static int build_curseg(struct f2fs_sb_info
*sbi
)
2129 struct curseg_info
*array
;
2132 array
= kcalloc(NR_CURSEG_TYPE
, sizeof(*array
), GFP_KERNEL
);
2136 SM_I(sbi
)->curseg_array
= array
;
2138 for (i
= 0; i
< NR_CURSEG_TYPE
; i
++) {
2139 mutex_init(&array
[i
].curseg_mutex
);
2140 array
[i
].sum_blk
= kzalloc(PAGE_CACHE_SIZE
, GFP_KERNEL
);
2141 if (!array
[i
].sum_blk
)
2143 array
[i
].segno
= NULL_SEGNO
;
2144 array
[i
].next_blkoff
= 0;
2146 return restore_curseg_summaries(sbi
);
2149 static void build_sit_entries(struct f2fs_sb_info
*sbi
)
2151 struct sit_info
*sit_i
= SIT_I(sbi
);
2152 struct curseg_info
*curseg
= CURSEG_I(sbi
, CURSEG_COLD_DATA
);
2153 struct f2fs_journal
*journal
= &curseg
->sum_blk
->journal
;
2154 int sit_blk_cnt
= SIT_BLK_CNT(sbi
);
2155 unsigned int i
, start
, end
;
2156 unsigned int readed
, start_blk
= 0;
2157 int nrpages
= MAX_BIO_BLOCKS(sbi
);
2160 readed
= ra_meta_pages(sbi
, start_blk
, nrpages
, META_SIT
, true);
2162 start
= start_blk
* sit_i
->sents_per_block
;
2163 end
= (start_blk
+ readed
) * sit_i
->sents_per_block
;
2165 for (; start
< end
&& start
< MAIN_SEGS(sbi
); start
++) {
2166 struct seg_entry
*se
= &sit_i
->sentries
[start
];
2167 struct f2fs_sit_block
*sit_blk
;
2168 struct f2fs_sit_entry sit
;
2171 mutex_lock(&curseg
->curseg_mutex
);
2172 for (i
= 0; i
< sits_in_cursum(journal
); i
++) {
2173 if (le32_to_cpu(segno_in_journal(journal
, i
))
2175 sit
= sit_in_journal(journal
, i
);
2176 mutex_unlock(&curseg
->curseg_mutex
);
2180 mutex_unlock(&curseg
->curseg_mutex
);
2182 page
= get_current_sit_page(sbi
, start
);
2183 sit_blk
= (struct f2fs_sit_block
*)page_address(page
);
2184 sit
= sit_blk
->entries
[SIT_ENTRY_OFFSET(sit_i
, start
)];
2185 f2fs_put_page(page
, 1);
2187 check_block_count(sbi
, start
, &sit
);
2188 seg_info_from_raw_sit(se
, &sit
);
2190 /* build discard map only one time */
2191 memcpy(se
->discard_map
, se
->cur_valid_map
, SIT_VBLOCK_MAP_SIZE
);
2192 sbi
->discard_blks
+= sbi
->blocks_per_seg
- se
->valid_blocks
;
2194 if (sbi
->segs_per_sec
> 1) {
2195 struct sec_entry
*e
= get_sec_entry(sbi
, start
);
2196 e
->valid_blocks
+= se
->valid_blocks
;
2199 start_blk
+= readed
;
2200 } while (start_blk
< sit_blk_cnt
);
2203 static void init_free_segmap(struct f2fs_sb_info
*sbi
)
2208 for (start
= 0; start
< MAIN_SEGS(sbi
); start
++) {
2209 struct seg_entry
*sentry
= get_seg_entry(sbi
, start
);
2210 if (!sentry
->valid_blocks
)
2211 __set_free(sbi
, start
);
2214 /* set use the current segments */
2215 for (type
= CURSEG_HOT_DATA
; type
<= CURSEG_COLD_NODE
; type
++) {
2216 struct curseg_info
*curseg_t
= CURSEG_I(sbi
, type
);
2217 __set_test_and_inuse(sbi
, curseg_t
->segno
);
2221 static void init_dirty_segmap(struct f2fs_sb_info
*sbi
)
2223 struct dirty_seglist_info
*dirty_i
= DIRTY_I(sbi
);
2224 struct free_segmap_info
*free_i
= FREE_I(sbi
);
2225 unsigned int segno
= 0, offset
= 0;
2226 unsigned short valid_blocks
;
2229 /* find dirty segment based on free segmap */
2230 segno
= find_next_inuse(free_i
, MAIN_SEGS(sbi
), offset
);
2231 if (segno
>= MAIN_SEGS(sbi
))
2234 valid_blocks
= get_valid_blocks(sbi
, segno
, 0);
2235 if (valid_blocks
== sbi
->blocks_per_seg
|| !valid_blocks
)
2237 if (valid_blocks
> sbi
->blocks_per_seg
) {
2238 f2fs_bug_on(sbi
, 1);
2241 mutex_lock(&dirty_i
->seglist_lock
);
2242 __locate_dirty_segment(sbi
, segno
, DIRTY
);
2243 mutex_unlock(&dirty_i
->seglist_lock
);
2247 static int init_victim_secmap(struct f2fs_sb_info
*sbi
)
2249 struct dirty_seglist_info
*dirty_i
= DIRTY_I(sbi
);
2250 unsigned int bitmap_size
= f2fs_bitmap_size(MAIN_SECS(sbi
));
2252 dirty_i
->victim_secmap
= f2fs_kvzalloc(bitmap_size
, GFP_KERNEL
);
2253 if (!dirty_i
->victim_secmap
)
2258 static int build_dirty_segmap(struct f2fs_sb_info
*sbi
)
2260 struct dirty_seglist_info
*dirty_i
;
2261 unsigned int bitmap_size
, i
;
2263 /* allocate memory for dirty segments list information */
2264 dirty_i
= kzalloc(sizeof(struct dirty_seglist_info
), GFP_KERNEL
);
2268 SM_I(sbi
)->dirty_info
= dirty_i
;
2269 mutex_init(&dirty_i
->seglist_lock
);
2271 bitmap_size
= f2fs_bitmap_size(MAIN_SEGS(sbi
));
2273 for (i
= 0; i
< NR_DIRTY_TYPE
; i
++) {
2274 dirty_i
->dirty_segmap
[i
] = f2fs_kvzalloc(bitmap_size
, GFP_KERNEL
);
2275 if (!dirty_i
->dirty_segmap
[i
])
2279 init_dirty_segmap(sbi
);
2280 return init_victim_secmap(sbi
);
2284 * Update min, max modified time for cost-benefit GC algorithm
2286 static void init_min_max_mtime(struct f2fs_sb_info
*sbi
)
2288 struct sit_info
*sit_i
= SIT_I(sbi
);
2291 mutex_lock(&sit_i
->sentry_lock
);
2293 sit_i
->min_mtime
= LLONG_MAX
;
2295 for (segno
= 0; segno
< MAIN_SEGS(sbi
); segno
+= sbi
->segs_per_sec
) {
2297 unsigned long long mtime
= 0;
2299 for (i
= 0; i
< sbi
->segs_per_sec
; i
++)
2300 mtime
+= get_seg_entry(sbi
, segno
+ i
)->mtime
;
2302 mtime
= div_u64(mtime
, sbi
->segs_per_sec
);
2304 if (sit_i
->min_mtime
> mtime
)
2305 sit_i
->min_mtime
= mtime
;
2307 sit_i
->max_mtime
= get_mtime(sbi
);
2308 mutex_unlock(&sit_i
->sentry_lock
);
2311 int build_segment_manager(struct f2fs_sb_info
*sbi
)
2313 struct f2fs_super_block
*raw_super
= F2FS_RAW_SUPER(sbi
);
2314 struct f2fs_checkpoint
*ckpt
= F2FS_CKPT(sbi
);
2315 struct f2fs_sm_info
*sm_info
;
2318 sm_info
= kzalloc(sizeof(struct f2fs_sm_info
), GFP_KERNEL
);
2323 sbi
->sm_info
= sm_info
;
2324 sm_info
->seg0_blkaddr
= le32_to_cpu(raw_super
->segment0_blkaddr
);
2325 sm_info
->main_blkaddr
= le32_to_cpu(raw_super
->main_blkaddr
);
2326 sm_info
->segment_count
= le32_to_cpu(raw_super
->segment_count
);
2327 sm_info
->reserved_segments
= le32_to_cpu(ckpt
->rsvd_segment_count
);
2328 sm_info
->ovp_segments
= le32_to_cpu(ckpt
->overprov_segment_count
);
2329 sm_info
->main_segments
= le32_to_cpu(raw_super
->segment_count_main
);
2330 sm_info
->ssa_blkaddr
= le32_to_cpu(raw_super
->ssa_blkaddr
);
2331 sm_info
->rec_prefree_segments
= sm_info
->main_segments
*
2332 DEF_RECLAIM_PREFREE_SEGMENTS
/ 100;
2333 sm_info
->ipu_policy
= 1 << F2FS_IPU_FSYNC
;
2334 sm_info
->min_ipu_util
= DEF_MIN_IPU_UTIL
;
2335 sm_info
->min_fsync_blocks
= DEF_MIN_FSYNC_BLOCKS
;
2337 INIT_LIST_HEAD(&sm_info
->discard_list
);
2338 sm_info
->nr_discards
= 0;
2339 sm_info
->max_discards
= 0;
2341 sm_info
->trim_sections
= DEF_BATCHED_TRIM_SECTIONS
;
2343 INIT_LIST_HEAD(&sm_info
->sit_entry_set
);
2345 if (test_opt(sbi
, FLUSH_MERGE
) && !f2fs_readonly(sbi
->sb
)) {
2346 err
= create_flush_cmd_control(sbi
);
2351 err
= build_sit_info(sbi
);
2354 err
= build_free_segmap(sbi
);
2357 err
= build_curseg(sbi
);
2361 /* reinit free segmap based on SIT */
2362 build_sit_entries(sbi
);
2364 init_free_segmap(sbi
);
2365 err
= build_dirty_segmap(sbi
);
2369 init_min_max_mtime(sbi
);
2373 static void discard_dirty_segmap(struct f2fs_sb_info
*sbi
,
2374 enum dirty_type dirty_type
)
2376 struct dirty_seglist_info
*dirty_i
= DIRTY_I(sbi
);
2378 mutex_lock(&dirty_i
->seglist_lock
);
2379 kvfree(dirty_i
->dirty_segmap
[dirty_type
]);
2380 dirty_i
->nr_dirty
[dirty_type
] = 0;
2381 mutex_unlock(&dirty_i
->seglist_lock
);
2384 static void destroy_victim_secmap(struct f2fs_sb_info
*sbi
)
2386 struct dirty_seglist_info
*dirty_i
= DIRTY_I(sbi
);
2387 kvfree(dirty_i
->victim_secmap
);
2390 static void destroy_dirty_segmap(struct f2fs_sb_info
*sbi
)
2392 struct dirty_seglist_info
*dirty_i
= DIRTY_I(sbi
);
2398 /* discard pre-free/dirty segments list */
2399 for (i
= 0; i
< NR_DIRTY_TYPE
; i
++)
2400 discard_dirty_segmap(sbi
, i
);
2402 destroy_victim_secmap(sbi
);
2403 SM_I(sbi
)->dirty_info
= NULL
;
2407 static void destroy_curseg(struct f2fs_sb_info
*sbi
)
2409 struct curseg_info
*array
= SM_I(sbi
)->curseg_array
;
2414 SM_I(sbi
)->curseg_array
= NULL
;
2415 for (i
= 0; i
< NR_CURSEG_TYPE
; i
++)
2416 kfree(array
[i
].sum_blk
);
2420 static void destroy_free_segmap(struct f2fs_sb_info
*sbi
)
2422 struct free_segmap_info
*free_i
= SM_I(sbi
)->free_info
;
2425 SM_I(sbi
)->free_info
= NULL
;
2426 kvfree(free_i
->free_segmap
);
2427 kvfree(free_i
->free_secmap
);
2431 static void destroy_sit_info(struct f2fs_sb_info
*sbi
)
2433 struct sit_info
*sit_i
= SIT_I(sbi
);
2439 if (sit_i
->sentries
) {
2440 for (start
= 0; start
< MAIN_SEGS(sbi
); start
++) {
2441 kfree(sit_i
->sentries
[start
].cur_valid_map
);
2442 kfree(sit_i
->sentries
[start
].ckpt_valid_map
);
2443 kfree(sit_i
->sentries
[start
].discard_map
);
2446 kfree(sit_i
->tmp_map
);
2448 kvfree(sit_i
->sentries
);
2449 kvfree(sit_i
->sec_entries
);
2450 kvfree(sit_i
->dirty_sentries_bitmap
);
2452 SM_I(sbi
)->sit_info
= NULL
;
2453 kfree(sit_i
->sit_bitmap
);
2457 void destroy_segment_manager(struct f2fs_sb_info
*sbi
)
2459 struct f2fs_sm_info
*sm_info
= SM_I(sbi
);
2463 destroy_flush_cmd_control(sbi
);
2464 destroy_dirty_segmap(sbi
);
2465 destroy_curseg(sbi
);
2466 destroy_free_segmap(sbi
);
2467 destroy_sit_info(sbi
);
2468 sbi
->sm_info
= NULL
;
2472 int __init
create_segment_manager_caches(void)
2474 discard_entry_slab
= f2fs_kmem_cache_create("discard_entry",
2475 sizeof(struct discard_entry
));
2476 if (!discard_entry_slab
)
2479 sit_entry_set_slab
= f2fs_kmem_cache_create("sit_entry_set",
2480 sizeof(struct sit_entry_set
));
2481 if (!sit_entry_set_slab
)
2482 goto destory_discard_entry
;
2484 inmem_entry_slab
= f2fs_kmem_cache_create("inmem_page_entry",
2485 sizeof(struct inmem_pages
));
2486 if (!inmem_entry_slab
)
2487 goto destroy_sit_entry_set
;
2490 destroy_sit_entry_set
:
2491 kmem_cache_destroy(sit_entry_set_slab
);
2492 destory_discard_entry
:
2493 kmem_cache_destroy(discard_entry_slab
);
2498 void destroy_segment_manager_caches(void)
2500 kmem_cache_destroy(sit_entry_set_slab
);
2501 kmem_cache_destroy(discard_entry_slab
);
2502 kmem_cache_destroy(inmem_entry_slab
);