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 struct blk_plug plug
;
376 blk_start_plug(&plug
);
377 sync_dirty_inodes(sbi
, FILE_INODE
);
378 blk_finish_plug(&plug
);
380 f2fs_sync_fs(sbi
->sb
, true);
381 stat_inc_bg_cp_count(sbi
->stat_info
);
385 static int issue_flush_thread(void *data
)
387 struct f2fs_sb_info
*sbi
= data
;
388 struct flush_cmd_control
*fcc
= SM_I(sbi
)->cmd_control_info
;
389 wait_queue_head_t
*q
= &fcc
->flush_wait_queue
;
391 if (kthread_should_stop())
394 if (!llist_empty(&fcc
->issue_list
)) {
396 struct flush_cmd
*cmd
, *next
;
399 bio
= f2fs_bio_alloc(0);
401 fcc
->dispatch_list
= llist_del_all(&fcc
->issue_list
);
402 fcc
->dispatch_list
= llist_reverse_order(fcc
->dispatch_list
);
404 bio
->bi_bdev
= sbi
->sb
->s_bdev
;
405 ret
= submit_bio_wait(WRITE_FLUSH
, bio
);
407 llist_for_each_entry_safe(cmd
, next
,
408 fcc
->dispatch_list
, llnode
) {
410 complete(&cmd
->wait
);
413 fcc
->dispatch_list
= NULL
;
416 wait_event_interruptible(*q
,
417 kthread_should_stop() || !llist_empty(&fcc
->issue_list
));
421 int f2fs_issue_flush(struct f2fs_sb_info
*sbi
)
423 struct flush_cmd_control
*fcc
= SM_I(sbi
)->cmd_control_info
;
424 struct flush_cmd cmd
;
426 trace_f2fs_issue_flush(sbi
->sb
, test_opt(sbi
, NOBARRIER
),
427 test_opt(sbi
, FLUSH_MERGE
));
429 if (test_opt(sbi
, NOBARRIER
))
432 if (!test_opt(sbi
, FLUSH_MERGE
)) {
433 struct bio
*bio
= f2fs_bio_alloc(0);
436 bio
->bi_bdev
= sbi
->sb
->s_bdev
;
437 ret
= submit_bio_wait(WRITE_FLUSH
, bio
);
442 init_completion(&cmd
.wait
);
444 llist_add(&cmd
.llnode
, &fcc
->issue_list
);
446 if (!fcc
->dispatch_list
)
447 wake_up(&fcc
->flush_wait_queue
);
449 wait_for_completion(&cmd
.wait
);
454 int create_flush_cmd_control(struct f2fs_sb_info
*sbi
)
456 dev_t dev
= sbi
->sb
->s_bdev
->bd_dev
;
457 struct flush_cmd_control
*fcc
;
460 fcc
= kzalloc(sizeof(struct flush_cmd_control
), GFP_KERNEL
);
463 init_waitqueue_head(&fcc
->flush_wait_queue
);
464 init_llist_head(&fcc
->issue_list
);
465 SM_I(sbi
)->cmd_control_info
= fcc
;
466 fcc
->f2fs_issue_flush
= kthread_run(issue_flush_thread
, sbi
,
467 "f2fs_flush-%u:%u", MAJOR(dev
), MINOR(dev
));
468 if (IS_ERR(fcc
->f2fs_issue_flush
)) {
469 err
= PTR_ERR(fcc
->f2fs_issue_flush
);
471 SM_I(sbi
)->cmd_control_info
= NULL
;
478 void destroy_flush_cmd_control(struct f2fs_sb_info
*sbi
)
480 struct flush_cmd_control
*fcc
= SM_I(sbi
)->cmd_control_info
;
482 if (fcc
&& fcc
->f2fs_issue_flush
)
483 kthread_stop(fcc
->f2fs_issue_flush
);
485 SM_I(sbi
)->cmd_control_info
= NULL
;
488 static void __locate_dirty_segment(struct f2fs_sb_info
*sbi
, unsigned int segno
,
489 enum dirty_type dirty_type
)
491 struct dirty_seglist_info
*dirty_i
= DIRTY_I(sbi
);
493 /* need not be added */
494 if (IS_CURSEG(sbi
, segno
))
497 if (!test_and_set_bit(segno
, dirty_i
->dirty_segmap
[dirty_type
]))
498 dirty_i
->nr_dirty
[dirty_type
]++;
500 if (dirty_type
== DIRTY
) {
501 struct seg_entry
*sentry
= get_seg_entry(sbi
, segno
);
502 enum dirty_type t
= sentry
->type
;
504 if (unlikely(t
>= DIRTY
)) {
508 if (!test_and_set_bit(segno
, dirty_i
->dirty_segmap
[t
]))
509 dirty_i
->nr_dirty
[t
]++;
513 static void __remove_dirty_segment(struct f2fs_sb_info
*sbi
, unsigned int segno
,
514 enum dirty_type dirty_type
)
516 struct dirty_seglist_info
*dirty_i
= DIRTY_I(sbi
);
518 if (test_and_clear_bit(segno
, dirty_i
->dirty_segmap
[dirty_type
]))
519 dirty_i
->nr_dirty
[dirty_type
]--;
521 if (dirty_type
== DIRTY
) {
522 struct seg_entry
*sentry
= get_seg_entry(sbi
, segno
);
523 enum dirty_type t
= sentry
->type
;
525 if (test_and_clear_bit(segno
, dirty_i
->dirty_segmap
[t
]))
526 dirty_i
->nr_dirty
[t
]--;
528 if (get_valid_blocks(sbi
, segno
, sbi
->segs_per_sec
) == 0)
529 clear_bit(GET_SECNO(sbi
, segno
),
530 dirty_i
->victim_secmap
);
535 * Should not occur error such as -ENOMEM.
536 * Adding dirty entry into seglist is not critical operation.
537 * If a given segment is one of current working segments, it won't be added.
539 static void locate_dirty_segment(struct f2fs_sb_info
*sbi
, unsigned int segno
)
541 struct dirty_seglist_info
*dirty_i
= DIRTY_I(sbi
);
542 unsigned short valid_blocks
;
544 if (segno
== NULL_SEGNO
|| IS_CURSEG(sbi
, segno
))
547 mutex_lock(&dirty_i
->seglist_lock
);
549 valid_blocks
= get_valid_blocks(sbi
, segno
, 0);
551 if (valid_blocks
== 0) {
552 __locate_dirty_segment(sbi
, segno
, PRE
);
553 __remove_dirty_segment(sbi
, segno
, DIRTY
);
554 } else if (valid_blocks
< sbi
->blocks_per_seg
) {
555 __locate_dirty_segment(sbi
, segno
, DIRTY
);
557 /* Recovery routine with SSR needs this */
558 __remove_dirty_segment(sbi
, segno
, DIRTY
);
561 mutex_unlock(&dirty_i
->seglist_lock
);
564 static int f2fs_issue_discard(struct f2fs_sb_info
*sbi
,
565 block_t blkstart
, block_t blklen
)
567 sector_t start
= SECTOR_FROM_BLOCK(blkstart
);
568 sector_t len
= SECTOR_FROM_BLOCK(blklen
);
569 struct seg_entry
*se
;
573 for (i
= blkstart
; i
< blkstart
+ blklen
; i
++) {
574 se
= get_seg_entry(sbi
, GET_SEGNO(sbi
, i
));
575 offset
= GET_BLKOFF_FROM_SEG0(sbi
, i
);
577 if (!f2fs_test_and_set_bit(offset
, se
->discard_map
))
580 trace_f2fs_issue_discard(sbi
->sb
, blkstart
, blklen
);
581 return blkdev_issue_discard(sbi
->sb
->s_bdev
, start
, len
, GFP_NOFS
, 0);
584 bool discard_next_dnode(struct f2fs_sb_info
*sbi
, block_t blkaddr
)
586 int err
= -EOPNOTSUPP
;
588 if (test_opt(sbi
, DISCARD
)) {
589 struct seg_entry
*se
= get_seg_entry(sbi
,
590 GET_SEGNO(sbi
, blkaddr
));
591 unsigned int offset
= GET_BLKOFF_FROM_SEG0(sbi
, blkaddr
);
593 if (f2fs_test_bit(offset
, se
->discard_map
))
596 err
= f2fs_issue_discard(sbi
, blkaddr
, 1);
600 update_meta_page(sbi
, NULL
, blkaddr
);
606 static void __add_discard_entry(struct f2fs_sb_info
*sbi
,
607 struct cp_control
*cpc
, struct seg_entry
*se
,
608 unsigned int start
, unsigned int end
)
610 struct list_head
*head
= &SM_I(sbi
)->discard_list
;
611 struct discard_entry
*new, *last
;
613 if (!list_empty(head
)) {
614 last
= list_last_entry(head
, struct discard_entry
, list
);
615 if (START_BLOCK(sbi
, cpc
->trim_start
) + start
==
616 last
->blkaddr
+ last
->len
) {
617 last
->len
+= end
- start
;
622 new = f2fs_kmem_cache_alloc(discard_entry_slab
, GFP_NOFS
);
623 INIT_LIST_HEAD(&new->list
);
624 new->blkaddr
= START_BLOCK(sbi
, cpc
->trim_start
) + start
;
625 new->len
= end
- start
;
626 list_add_tail(&new->list
, head
);
628 SM_I(sbi
)->nr_discards
+= end
- start
;
631 static void add_discard_addrs(struct f2fs_sb_info
*sbi
, struct cp_control
*cpc
)
633 int entries
= SIT_VBLOCK_MAP_SIZE
/ sizeof(unsigned long);
634 int max_blocks
= sbi
->blocks_per_seg
;
635 struct seg_entry
*se
= get_seg_entry(sbi
, cpc
->trim_start
);
636 unsigned long *cur_map
= (unsigned long *)se
->cur_valid_map
;
637 unsigned long *ckpt_map
= (unsigned long *)se
->ckpt_valid_map
;
638 unsigned long *discard_map
= (unsigned long *)se
->discard_map
;
639 unsigned long *dmap
= SIT_I(sbi
)->tmp_map
;
640 unsigned int start
= 0, end
= -1;
641 bool force
= (cpc
->reason
== CP_DISCARD
);
644 if (se
->valid_blocks
== max_blocks
)
648 if (!test_opt(sbi
, DISCARD
) || !se
->valid_blocks
||
649 SM_I(sbi
)->nr_discards
>= SM_I(sbi
)->max_discards
)
653 /* SIT_VBLOCK_MAP_SIZE should be multiple of sizeof(unsigned long) */
654 for (i
= 0; i
< entries
; i
++)
655 dmap
[i
] = force
? ~ckpt_map
[i
] & ~discard_map
[i
] :
656 (cur_map
[i
] ^ ckpt_map
[i
]) & ckpt_map
[i
];
658 while (force
|| SM_I(sbi
)->nr_discards
<= SM_I(sbi
)->max_discards
) {
659 start
= __find_rev_next_bit(dmap
, max_blocks
, end
+ 1);
660 if (start
>= max_blocks
)
663 end
= __find_rev_next_zero_bit(dmap
, max_blocks
, start
+ 1);
664 __add_discard_entry(sbi
, cpc
, se
, start
, end
);
668 void release_discard_addrs(struct f2fs_sb_info
*sbi
)
670 struct list_head
*head
= &(SM_I(sbi
)->discard_list
);
671 struct discard_entry
*entry
, *this;
674 list_for_each_entry_safe(entry
, this, head
, list
) {
675 list_del(&entry
->list
);
676 kmem_cache_free(discard_entry_slab
, entry
);
681 * Should call clear_prefree_segments after checkpoint is done.
683 static void set_prefree_as_free_segments(struct f2fs_sb_info
*sbi
)
685 struct dirty_seglist_info
*dirty_i
= DIRTY_I(sbi
);
688 mutex_lock(&dirty_i
->seglist_lock
);
689 for_each_set_bit(segno
, dirty_i
->dirty_segmap
[PRE
], MAIN_SEGS(sbi
))
690 __set_test_and_free(sbi
, segno
);
691 mutex_unlock(&dirty_i
->seglist_lock
);
694 void clear_prefree_segments(struct f2fs_sb_info
*sbi
, struct cp_control
*cpc
)
696 struct list_head
*head
= &(SM_I(sbi
)->discard_list
);
697 struct discard_entry
*entry
, *this;
698 struct dirty_seglist_info
*dirty_i
= DIRTY_I(sbi
);
699 unsigned long *prefree_map
= dirty_i
->dirty_segmap
[PRE
];
700 unsigned int start
= 0, end
= -1;
702 mutex_lock(&dirty_i
->seglist_lock
);
706 start
= find_next_bit(prefree_map
, MAIN_SEGS(sbi
), end
+ 1);
707 if (start
>= MAIN_SEGS(sbi
))
709 end
= find_next_zero_bit(prefree_map
, MAIN_SEGS(sbi
),
712 for (i
= start
; i
< end
; i
++)
713 clear_bit(i
, prefree_map
);
715 dirty_i
->nr_dirty
[PRE
] -= end
- start
;
717 if (!test_opt(sbi
, DISCARD
))
720 f2fs_issue_discard(sbi
, START_BLOCK(sbi
, start
),
721 (end
- start
) << sbi
->log_blocks_per_seg
);
723 mutex_unlock(&dirty_i
->seglist_lock
);
725 /* send small discards */
726 list_for_each_entry_safe(entry
, this, head
, list
) {
727 if (cpc
->reason
== CP_DISCARD
&& entry
->len
< cpc
->trim_minlen
)
729 f2fs_issue_discard(sbi
, entry
->blkaddr
, entry
->len
);
730 cpc
->trimmed
+= entry
->len
;
732 list_del(&entry
->list
);
733 SM_I(sbi
)->nr_discards
-= entry
->len
;
734 kmem_cache_free(discard_entry_slab
, entry
);
738 static bool __mark_sit_entry_dirty(struct f2fs_sb_info
*sbi
, unsigned int segno
)
740 struct sit_info
*sit_i
= SIT_I(sbi
);
742 if (!__test_and_set_bit(segno
, sit_i
->dirty_sentries_bitmap
)) {
743 sit_i
->dirty_sentries
++;
750 static void __set_sit_entry_type(struct f2fs_sb_info
*sbi
, int type
,
751 unsigned int segno
, int modified
)
753 struct seg_entry
*se
= get_seg_entry(sbi
, segno
);
756 __mark_sit_entry_dirty(sbi
, segno
);
759 static void update_sit_entry(struct f2fs_sb_info
*sbi
, block_t blkaddr
, int del
)
761 struct seg_entry
*se
;
762 unsigned int segno
, offset
;
763 long int new_vblocks
;
765 segno
= GET_SEGNO(sbi
, blkaddr
);
767 se
= get_seg_entry(sbi
, segno
);
768 new_vblocks
= se
->valid_blocks
+ del
;
769 offset
= GET_BLKOFF_FROM_SEG0(sbi
, blkaddr
);
771 f2fs_bug_on(sbi
, (new_vblocks
>> (sizeof(unsigned short) << 3) ||
772 (new_vblocks
> sbi
->blocks_per_seg
)));
774 se
->valid_blocks
= new_vblocks
;
775 se
->mtime
= get_mtime(sbi
);
776 SIT_I(sbi
)->max_mtime
= se
->mtime
;
778 /* Update valid block bitmap */
780 if (f2fs_test_and_set_bit(offset
, se
->cur_valid_map
))
782 if (!f2fs_test_and_set_bit(offset
, se
->discard_map
))
785 if (!f2fs_test_and_clear_bit(offset
, se
->cur_valid_map
))
787 if (f2fs_test_and_clear_bit(offset
, se
->discard_map
))
790 if (!f2fs_test_bit(offset
, se
->ckpt_valid_map
))
791 se
->ckpt_valid_blocks
+= del
;
793 __mark_sit_entry_dirty(sbi
, segno
);
795 /* update total number of valid blocks to be written in ckpt area */
796 SIT_I(sbi
)->written_valid_blocks
+= del
;
798 if (sbi
->segs_per_sec
> 1)
799 get_sec_entry(sbi
, segno
)->valid_blocks
+= del
;
802 void refresh_sit_entry(struct f2fs_sb_info
*sbi
, block_t old
, block_t
new)
804 update_sit_entry(sbi
, new, 1);
805 if (GET_SEGNO(sbi
, old
) != NULL_SEGNO
)
806 update_sit_entry(sbi
, old
, -1);
808 locate_dirty_segment(sbi
, GET_SEGNO(sbi
, old
));
809 locate_dirty_segment(sbi
, GET_SEGNO(sbi
, new));
812 void invalidate_blocks(struct f2fs_sb_info
*sbi
, block_t addr
)
814 unsigned int segno
= GET_SEGNO(sbi
, addr
);
815 struct sit_info
*sit_i
= SIT_I(sbi
);
817 f2fs_bug_on(sbi
, addr
== NULL_ADDR
);
818 if (addr
== NEW_ADDR
)
821 /* add it into sit main buffer */
822 mutex_lock(&sit_i
->sentry_lock
);
824 update_sit_entry(sbi
, addr
, -1);
826 /* add it into dirty seglist */
827 locate_dirty_segment(sbi
, segno
);
829 mutex_unlock(&sit_i
->sentry_lock
);
832 bool is_checkpointed_data(struct f2fs_sb_info
*sbi
, block_t blkaddr
)
834 struct sit_info
*sit_i
= SIT_I(sbi
);
835 unsigned int segno
, offset
;
836 struct seg_entry
*se
;
839 if (blkaddr
== NEW_ADDR
|| blkaddr
== NULL_ADDR
)
842 mutex_lock(&sit_i
->sentry_lock
);
844 segno
= GET_SEGNO(sbi
, blkaddr
);
845 se
= get_seg_entry(sbi
, segno
);
846 offset
= GET_BLKOFF_FROM_SEG0(sbi
, blkaddr
);
848 if (f2fs_test_bit(offset
, se
->ckpt_valid_map
))
851 mutex_unlock(&sit_i
->sentry_lock
);
857 * This function should be resided under the curseg_mutex lock
859 static void __add_sum_entry(struct f2fs_sb_info
*sbi
, int type
,
860 struct f2fs_summary
*sum
)
862 struct curseg_info
*curseg
= CURSEG_I(sbi
, type
);
863 void *addr
= curseg
->sum_blk
;
864 addr
+= curseg
->next_blkoff
* sizeof(struct f2fs_summary
);
865 memcpy(addr
, sum
, sizeof(struct f2fs_summary
));
869 * Calculate the number of current summary pages for writing
871 int npages_for_summary_flush(struct f2fs_sb_info
*sbi
, bool for_ra
)
873 int valid_sum_count
= 0;
876 for (i
= CURSEG_HOT_DATA
; i
<= CURSEG_COLD_DATA
; i
++) {
877 if (sbi
->ckpt
->alloc_type
[i
] == SSR
)
878 valid_sum_count
+= sbi
->blocks_per_seg
;
881 valid_sum_count
+= le16_to_cpu(
882 F2FS_CKPT(sbi
)->cur_data_blkoff
[i
]);
884 valid_sum_count
+= curseg_blkoff(sbi
, i
);
888 sum_in_page
= (PAGE_CACHE_SIZE
- 2 * SUM_JOURNAL_SIZE
-
889 SUM_FOOTER_SIZE
) / SUMMARY_SIZE
;
890 if (valid_sum_count
<= sum_in_page
)
892 else if ((valid_sum_count
- sum_in_page
) <=
893 (PAGE_CACHE_SIZE
- SUM_FOOTER_SIZE
) / SUMMARY_SIZE
)
899 * Caller should put this summary page
901 struct page
*get_sum_page(struct f2fs_sb_info
*sbi
, unsigned int segno
)
903 return get_meta_page(sbi
, GET_SUM_BLOCK(sbi
, segno
));
906 void update_meta_page(struct f2fs_sb_info
*sbi
, void *src
, block_t blk_addr
)
908 struct page
*page
= grab_meta_page(sbi
, blk_addr
);
909 void *dst
= page_address(page
);
912 memcpy(dst
, src
, PAGE_CACHE_SIZE
);
914 memset(dst
, 0, PAGE_CACHE_SIZE
);
915 set_page_dirty(page
);
916 f2fs_put_page(page
, 1);
919 static void write_sum_page(struct f2fs_sb_info
*sbi
,
920 struct f2fs_summary_block
*sum_blk
, block_t blk_addr
)
922 update_meta_page(sbi
, (void *)sum_blk
, blk_addr
);
925 static void write_current_sum_page(struct f2fs_sb_info
*sbi
,
926 int type
, block_t blk_addr
)
928 struct curseg_info
*curseg
= CURSEG_I(sbi
, type
);
929 struct page
*page
= grab_meta_page(sbi
, blk_addr
);
930 struct f2fs_summary_block
*src
= curseg
->sum_blk
;
931 struct f2fs_summary_block
*dst
;
933 dst
= (struct f2fs_summary_block
*)page_address(page
);
935 mutex_lock(&curseg
->curseg_mutex
);
937 down_read(&curseg
->journal_rwsem
);
938 memcpy(&dst
->journal
, curseg
->journal
, SUM_JOURNAL_SIZE
);
939 up_read(&curseg
->journal_rwsem
);
941 memcpy(dst
->entries
, src
->entries
, SUM_ENTRY_SIZE
);
942 memcpy(&dst
->footer
, &src
->footer
, SUM_FOOTER_SIZE
);
944 mutex_unlock(&curseg
->curseg_mutex
);
946 set_page_dirty(page
);
947 f2fs_put_page(page
, 1);
950 static int is_next_segment_free(struct f2fs_sb_info
*sbi
, int type
)
952 struct curseg_info
*curseg
= CURSEG_I(sbi
, type
);
953 unsigned int segno
= curseg
->segno
+ 1;
954 struct free_segmap_info
*free_i
= FREE_I(sbi
);
956 if (segno
< MAIN_SEGS(sbi
) && segno
% sbi
->segs_per_sec
)
957 return !test_bit(segno
, free_i
->free_segmap
);
962 * Find a new segment from the free segments bitmap to right order
963 * This function should be returned with success, otherwise BUG
965 static void get_new_segment(struct f2fs_sb_info
*sbi
,
966 unsigned int *newseg
, bool new_sec
, int dir
)
968 struct free_segmap_info
*free_i
= FREE_I(sbi
);
969 unsigned int segno
, secno
, zoneno
;
970 unsigned int total_zones
= MAIN_SECS(sbi
) / sbi
->secs_per_zone
;
971 unsigned int hint
= *newseg
/ sbi
->segs_per_sec
;
972 unsigned int old_zoneno
= GET_ZONENO_FROM_SEGNO(sbi
, *newseg
);
973 unsigned int left_start
= hint
;
978 spin_lock(&free_i
->segmap_lock
);
980 if (!new_sec
&& ((*newseg
+ 1) % sbi
->segs_per_sec
)) {
981 segno
= find_next_zero_bit(free_i
->free_segmap
,
982 (hint
+ 1) * sbi
->segs_per_sec
, *newseg
+ 1);
983 if (segno
< (hint
+ 1) * sbi
->segs_per_sec
)
987 secno
= find_next_zero_bit(free_i
->free_secmap
, MAIN_SECS(sbi
), hint
);
988 if (secno
>= MAIN_SECS(sbi
)) {
989 if (dir
== ALLOC_RIGHT
) {
990 secno
= find_next_zero_bit(free_i
->free_secmap
,
992 f2fs_bug_on(sbi
, secno
>= MAIN_SECS(sbi
));
995 left_start
= hint
- 1;
1001 while (test_bit(left_start
, free_i
->free_secmap
)) {
1002 if (left_start
> 0) {
1006 left_start
= find_next_zero_bit(free_i
->free_secmap
,
1008 f2fs_bug_on(sbi
, left_start
>= MAIN_SECS(sbi
));
1014 segno
= secno
* sbi
->segs_per_sec
;
1015 zoneno
= secno
/ sbi
->secs_per_zone
;
1017 /* give up on finding another zone */
1020 if (sbi
->secs_per_zone
== 1)
1022 if (zoneno
== old_zoneno
)
1024 if (dir
== ALLOC_LEFT
) {
1025 if (!go_left
&& zoneno
+ 1 >= total_zones
)
1027 if (go_left
&& zoneno
== 0)
1030 for (i
= 0; i
< NR_CURSEG_TYPE
; i
++)
1031 if (CURSEG_I(sbi
, i
)->zone
== zoneno
)
1034 if (i
< NR_CURSEG_TYPE
) {
1035 /* zone is in user, try another */
1037 hint
= zoneno
* sbi
->secs_per_zone
- 1;
1038 else if (zoneno
+ 1 >= total_zones
)
1041 hint
= (zoneno
+ 1) * sbi
->secs_per_zone
;
1043 goto find_other_zone
;
1046 /* set it as dirty segment in free segmap */
1047 f2fs_bug_on(sbi
, test_bit(segno
, free_i
->free_segmap
));
1048 __set_inuse(sbi
, segno
);
1050 spin_unlock(&free_i
->segmap_lock
);
1053 static void reset_curseg(struct f2fs_sb_info
*sbi
, int type
, int modified
)
1055 struct curseg_info
*curseg
= CURSEG_I(sbi
, type
);
1056 struct summary_footer
*sum_footer
;
1058 curseg
->segno
= curseg
->next_segno
;
1059 curseg
->zone
= GET_ZONENO_FROM_SEGNO(sbi
, curseg
->segno
);
1060 curseg
->next_blkoff
= 0;
1061 curseg
->next_segno
= NULL_SEGNO
;
1063 sum_footer
= &(curseg
->sum_blk
->footer
);
1064 memset(sum_footer
, 0, sizeof(struct summary_footer
));
1065 if (IS_DATASEG(type
))
1066 SET_SUM_TYPE(sum_footer
, SUM_TYPE_DATA
);
1067 if (IS_NODESEG(type
))
1068 SET_SUM_TYPE(sum_footer
, SUM_TYPE_NODE
);
1069 __set_sit_entry_type(sbi
, type
, curseg
->segno
, modified
);
1073 * Allocate a current working segment.
1074 * This function always allocates a free segment in LFS manner.
1076 static void new_curseg(struct f2fs_sb_info
*sbi
, int type
, bool new_sec
)
1078 struct curseg_info
*curseg
= CURSEG_I(sbi
, type
);
1079 unsigned int segno
= curseg
->segno
;
1080 int dir
= ALLOC_LEFT
;
1082 write_sum_page(sbi
, curseg
->sum_blk
,
1083 GET_SUM_BLOCK(sbi
, segno
));
1084 if (type
== CURSEG_WARM_DATA
|| type
== CURSEG_COLD_DATA
)
1087 if (test_opt(sbi
, NOHEAP
))
1090 get_new_segment(sbi
, &segno
, new_sec
, dir
);
1091 curseg
->next_segno
= segno
;
1092 reset_curseg(sbi
, type
, 1);
1093 curseg
->alloc_type
= LFS
;
1096 static void __next_free_blkoff(struct f2fs_sb_info
*sbi
,
1097 struct curseg_info
*seg
, block_t start
)
1099 struct seg_entry
*se
= get_seg_entry(sbi
, seg
->segno
);
1100 int entries
= SIT_VBLOCK_MAP_SIZE
/ sizeof(unsigned long);
1101 unsigned long *target_map
= SIT_I(sbi
)->tmp_map
;
1102 unsigned long *ckpt_map
= (unsigned long *)se
->ckpt_valid_map
;
1103 unsigned long *cur_map
= (unsigned long *)se
->cur_valid_map
;
1106 for (i
= 0; i
< entries
; i
++)
1107 target_map
[i
] = ckpt_map
[i
] | cur_map
[i
];
1109 pos
= __find_rev_next_zero_bit(target_map
, sbi
->blocks_per_seg
, start
);
1111 seg
->next_blkoff
= pos
;
1115 * If a segment is written by LFS manner, next block offset is just obtained
1116 * by increasing the current block offset. However, if a segment is written by
1117 * SSR manner, next block offset obtained by calling __next_free_blkoff
1119 static void __refresh_next_blkoff(struct f2fs_sb_info
*sbi
,
1120 struct curseg_info
*seg
)
1122 if (seg
->alloc_type
== SSR
)
1123 __next_free_blkoff(sbi
, seg
, seg
->next_blkoff
+ 1);
1129 * This function always allocates a used segment(from dirty seglist) by SSR
1130 * manner, so it should recover the existing segment information of valid blocks
1132 static void change_curseg(struct f2fs_sb_info
*sbi
, int type
, bool reuse
)
1134 struct dirty_seglist_info
*dirty_i
= DIRTY_I(sbi
);
1135 struct curseg_info
*curseg
= CURSEG_I(sbi
, type
);
1136 unsigned int new_segno
= curseg
->next_segno
;
1137 struct f2fs_summary_block
*sum_node
;
1138 struct page
*sum_page
;
1140 write_sum_page(sbi
, curseg
->sum_blk
,
1141 GET_SUM_BLOCK(sbi
, curseg
->segno
));
1142 __set_test_and_inuse(sbi
, new_segno
);
1144 mutex_lock(&dirty_i
->seglist_lock
);
1145 __remove_dirty_segment(sbi
, new_segno
, PRE
);
1146 __remove_dirty_segment(sbi
, new_segno
, DIRTY
);
1147 mutex_unlock(&dirty_i
->seglist_lock
);
1149 reset_curseg(sbi
, type
, 1);
1150 curseg
->alloc_type
= SSR
;
1151 __next_free_blkoff(sbi
, curseg
, 0);
1154 sum_page
= get_sum_page(sbi
, new_segno
);
1155 sum_node
= (struct f2fs_summary_block
*)page_address(sum_page
);
1156 memcpy(curseg
->sum_blk
, sum_node
, SUM_ENTRY_SIZE
);
1157 f2fs_put_page(sum_page
, 1);
1161 static int get_ssr_segment(struct f2fs_sb_info
*sbi
, int type
)
1163 struct curseg_info
*curseg
= CURSEG_I(sbi
, type
);
1164 const struct victim_selection
*v_ops
= DIRTY_I(sbi
)->v_ops
;
1166 if (IS_NODESEG(type
) || !has_not_enough_free_secs(sbi
, 0))
1167 return v_ops
->get_victim(sbi
,
1168 &(curseg
)->next_segno
, BG_GC
, type
, SSR
);
1170 /* For data segments, let's do SSR more intensively */
1171 for (; type
>= CURSEG_HOT_DATA
; type
--)
1172 if (v_ops
->get_victim(sbi
, &(curseg
)->next_segno
,
1179 * flush out current segment and replace it with new segment
1180 * This function should be returned with success, otherwise BUG
1182 static void allocate_segment_by_default(struct f2fs_sb_info
*sbi
,
1183 int type
, bool force
)
1185 struct curseg_info
*curseg
= CURSEG_I(sbi
, type
);
1188 new_curseg(sbi
, type
, true);
1189 else if (type
== CURSEG_WARM_NODE
)
1190 new_curseg(sbi
, type
, false);
1191 else if (curseg
->alloc_type
== LFS
&& is_next_segment_free(sbi
, type
))
1192 new_curseg(sbi
, type
, false);
1193 else if (need_SSR(sbi
) && get_ssr_segment(sbi
, type
))
1194 change_curseg(sbi
, type
, true);
1196 new_curseg(sbi
, type
, false);
1198 stat_inc_seg_type(sbi
, curseg
);
1201 static void __allocate_new_segments(struct f2fs_sb_info
*sbi
, int type
)
1203 struct curseg_info
*curseg
= CURSEG_I(sbi
, type
);
1204 unsigned int old_segno
;
1206 old_segno
= curseg
->segno
;
1207 SIT_I(sbi
)->s_ops
->allocate_segment(sbi
, type
, true);
1208 locate_dirty_segment(sbi
, old_segno
);
1211 void allocate_new_segments(struct f2fs_sb_info
*sbi
)
1215 for (i
= CURSEG_HOT_DATA
; i
<= CURSEG_COLD_DATA
; i
++)
1216 __allocate_new_segments(sbi
, i
);
1219 static const struct segment_allocation default_salloc_ops
= {
1220 .allocate_segment
= allocate_segment_by_default
,
1223 int f2fs_trim_fs(struct f2fs_sb_info
*sbi
, struct fstrim_range
*range
)
1225 __u64 start
= F2FS_BYTES_TO_BLK(range
->start
);
1226 __u64 end
= start
+ F2FS_BYTES_TO_BLK(range
->len
) - 1;
1227 unsigned int start_segno
, end_segno
;
1228 struct cp_control cpc
;
1231 if (start
>= MAX_BLKADDR(sbi
) || range
->len
< sbi
->blocksize
)
1235 if (end
<= MAIN_BLKADDR(sbi
))
1238 /* start/end segment number in main_area */
1239 start_segno
= (start
<= MAIN_BLKADDR(sbi
)) ? 0 : GET_SEGNO(sbi
, start
);
1240 end_segno
= (end
>= MAX_BLKADDR(sbi
)) ? MAIN_SEGS(sbi
) - 1 :
1241 GET_SEGNO(sbi
, end
);
1242 cpc
.reason
= CP_DISCARD
;
1243 cpc
.trim_minlen
= max_t(__u64
, 1, F2FS_BYTES_TO_BLK(range
->minlen
));
1245 /* do checkpoint to issue discard commands safely */
1246 for (; start_segno
<= end_segno
; start_segno
= cpc
.trim_end
+ 1) {
1247 cpc
.trim_start
= start_segno
;
1249 if (sbi
->discard_blks
== 0)
1251 else if (sbi
->discard_blks
< BATCHED_TRIM_BLOCKS(sbi
))
1252 cpc
.trim_end
= end_segno
;
1254 cpc
.trim_end
= min_t(unsigned int,
1255 rounddown(start_segno
+
1256 BATCHED_TRIM_SEGMENTS(sbi
),
1257 sbi
->segs_per_sec
) - 1, end_segno
);
1259 mutex_lock(&sbi
->gc_mutex
);
1260 err
= write_checkpoint(sbi
, &cpc
);
1261 mutex_unlock(&sbi
->gc_mutex
);
1264 range
->len
= F2FS_BLK_TO_BYTES(cpc
.trimmed
);
1268 static bool __has_curseg_space(struct f2fs_sb_info
*sbi
, int type
)
1270 struct curseg_info
*curseg
= CURSEG_I(sbi
, type
);
1271 if (curseg
->next_blkoff
< sbi
->blocks_per_seg
)
1276 static int __get_segment_type_2(struct page
*page
, enum page_type p_type
)
1279 return CURSEG_HOT_DATA
;
1281 return CURSEG_HOT_NODE
;
1284 static int __get_segment_type_4(struct page
*page
, enum page_type p_type
)
1286 if (p_type
== DATA
) {
1287 struct inode
*inode
= page
->mapping
->host
;
1289 if (S_ISDIR(inode
->i_mode
))
1290 return CURSEG_HOT_DATA
;
1292 return CURSEG_COLD_DATA
;
1294 if (IS_DNODE(page
) && is_cold_node(page
))
1295 return CURSEG_WARM_NODE
;
1297 return CURSEG_COLD_NODE
;
1301 static int __get_segment_type_6(struct page
*page
, enum page_type p_type
)
1303 if (p_type
== DATA
) {
1304 struct inode
*inode
= page
->mapping
->host
;
1306 if (S_ISDIR(inode
->i_mode
))
1307 return CURSEG_HOT_DATA
;
1308 else if (is_cold_data(page
) || file_is_cold(inode
))
1309 return CURSEG_COLD_DATA
;
1311 return CURSEG_WARM_DATA
;
1314 return is_cold_node(page
) ? CURSEG_WARM_NODE
:
1317 return CURSEG_COLD_NODE
;
1321 static int __get_segment_type(struct page
*page
, enum page_type p_type
)
1323 switch (F2FS_P_SB(page
)->active_logs
) {
1325 return __get_segment_type_2(page
, p_type
);
1327 return __get_segment_type_4(page
, p_type
);
1329 /* NR_CURSEG_TYPE(6) logs by default */
1330 f2fs_bug_on(F2FS_P_SB(page
),
1331 F2FS_P_SB(page
)->active_logs
!= NR_CURSEG_TYPE
);
1332 return __get_segment_type_6(page
, p_type
);
1335 void allocate_data_block(struct f2fs_sb_info
*sbi
, struct page
*page
,
1336 block_t old_blkaddr
, block_t
*new_blkaddr
,
1337 struct f2fs_summary
*sum
, int type
)
1339 struct sit_info
*sit_i
= SIT_I(sbi
);
1340 struct curseg_info
*curseg
;
1341 bool direct_io
= (type
== CURSEG_DIRECT_IO
);
1343 type
= direct_io
? CURSEG_WARM_DATA
: type
;
1345 curseg
= CURSEG_I(sbi
, type
);
1347 mutex_lock(&curseg
->curseg_mutex
);
1348 mutex_lock(&sit_i
->sentry_lock
);
1350 /* direct_io'ed data is aligned to the segment for better performance */
1351 if (direct_io
&& curseg
->next_blkoff
&&
1352 !has_not_enough_free_secs(sbi
, 0))
1353 __allocate_new_segments(sbi
, type
);
1355 *new_blkaddr
= NEXT_FREE_BLKADDR(sbi
, curseg
);
1358 * __add_sum_entry should be resided under the curseg_mutex
1359 * because, this function updates a summary entry in the
1360 * current summary block.
1362 __add_sum_entry(sbi
, type
, sum
);
1364 __refresh_next_blkoff(sbi
, curseg
);
1366 stat_inc_block_count(sbi
, curseg
);
1368 if (!__has_curseg_space(sbi
, type
))
1369 sit_i
->s_ops
->allocate_segment(sbi
, type
, false);
1371 * SIT information should be updated before segment allocation,
1372 * since SSR needs latest valid block information.
1374 refresh_sit_entry(sbi
, old_blkaddr
, *new_blkaddr
);
1376 mutex_unlock(&sit_i
->sentry_lock
);
1378 if (page
&& IS_NODESEG(type
))
1379 fill_node_footer_blkaddr(page
, NEXT_FREE_BLKADDR(sbi
, curseg
));
1381 mutex_unlock(&curseg
->curseg_mutex
);
1384 static void do_write_page(struct f2fs_summary
*sum
, struct f2fs_io_info
*fio
)
1386 int type
= __get_segment_type(fio
->page
, fio
->type
);
1388 allocate_data_block(fio
->sbi
, fio
->page
, fio
->old_blkaddr
,
1389 &fio
->new_blkaddr
, sum
, type
);
1391 /* writeout dirty page into bdev */
1392 f2fs_submit_page_mbio(fio
);
1395 void write_meta_page(struct f2fs_sb_info
*sbi
, struct page
*page
)
1397 struct f2fs_io_info fio
= {
1400 .rw
= WRITE_SYNC
| REQ_META
| REQ_PRIO
,
1401 .old_blkaddr
= page
->index
,
1402 .new_blkaddr
= page
->index
,
1404 .encrypted_page
= NULL
,
1407 if (unlikely(page
->index
>= MAIN_BLKADDR(sbi
)))
1408 fio
.rw
&= ~REQ_META
;
1410 set_page_writeback(page
);
1411 f2fs_submit_page_mbio(&fio
);
1414 void write_node_page(unsigned int nid
, struct f2fs_io_info
*fio
)
1416 struct f2fs_summary sum
;
1418 set_summary(&sum
, nid
, 0, 0);
1419 do_write_page(&sum
, fio
);
1422 void write_data_page(struct dnode_of_data
*dn
, struct f2fs_io_info
*fio
)
1424 struct f2fs_sb_info
*sbi
= fio
->sbi
;
1425 struct f2fs_summary sum
;
1426 struct node_info ni
;
1428 f2fs_bug_on(sbi
, dn
->data_blkaddr
== NULL_ADDR
);
1429 get_node_info(sbi
, dn
->nid
, &ni
);
1430 set_summary(&sum
, dn
->nid
, dn
->ofs_in_node
, ni
.version
);
1431 do_write_page(&sum
, fio
);
1432 f2fs_update_data_blkaddr(dn
, fio
->new_blkaddr
);
1435 void rewrite_data_page(struct f2fs_io_info
*fio
)
1437 fio
->new_blkaddr
= fio
->old_blkaddr
;
1438 stat_inc_inplace_blocks(fio
->sbi
);
1439 f2fs_submit_page_mbio(fio
);
1442 void __f2fs_replace_block(struct f2fs_sb_info
*sbi
, struct f2fs_summary
*sum
,
1443 block_t old_blkaddr
, block_t new_blkaddr
,
1444 bool recover_curseg
, bool recover_newaddr
)
1446 struct sit_info
*sit_i
= SIT_I(sbi
);
1447 struct curseg_info
*curseg
;
1448 unsigned int segno
, old_cursegno
;
1449 struct seg_entry
*se
;
1451 unsigned short old_blkoff
;
1453 segno
= GET_SEGNO(sbi
, new_blkaddr
);
1454 se
= get_seg_entry(sbi
, segno
);
1457 if (!recover_curseg
) {
1458 /* for recovery flow */
1459 if (se
->valid_blocks
== 0 && !IS_CURSEG(sbi
, segno
)) {
1460 if (old_blkaddr
== NULL_ADDR
)
1461 type
= CURSEG_COLD_DATA
;
1463 type
= CURSEG_WARM_DATA
;
1466 if (!IS_CURSEG(sbi
, segno
))
1467 type
= CURSEG_WARM_DATA
;
1470 curseg
= CURSEG_I(sbi
, type
);
1472 mutex_lock(&curseg
->curseg_mutex
);
1473 mutex_lock(&sit_i
->sentry_lock
);
1475 old_cursegno
= curseg
->segno
;
1476 old_blkoff
= curseg
->next_blkoff
;
1478 /* change the current segment */
1479 if (segno
!= curseg
->segno
) {
1480 curseg
->next_segno
= segno
;
1481 change_curseg(sbi
, type
, true);
1484 curseg
->next_blkoff
= GET_BLKOFF_FROM_SEG0(sbi
, new_blkaddr
);
1485 __add_sum_entry(sbi
, type
, sum
);
1487 if (!recover_curseg
|| recover_newaddr
)
1488 update_sit_entry(sbi
, new_blkaddr
, 1);
1489 if (GET_SEGNO(sbi
, old_blkaddr
) != NULL_SEGNO
)
1490 update_sit_entry(sbi
, old_blkaddr
, -1);
1492 locate_dirty_segment(sbi
, GET_SEGNO(sbi
, old_blkaddr
));
1493 locate_dirty_segment(sbi
, GET_SEGNO(sbi
, new_blkaddr
));
1495 locate_dirty_segment(sbi
, old_cursegno
);
1497 if (recover_curseg
) {
1498 if (old_cursegno
!= curseg
->segno
) {
1499 curseg
->next_segno
= old_cursegno
;
1500 change_curseg(sbi
, type
, true);
1502 curseg
->next_blkoff
= old_blkoff
;
1505 mutex_unlock(&sit_i
->sentry_lock
);
1506 mutex_unlock(&curseg
->curseg_mutex
);
1509 void f2fs_replace_block(struct f2fs_sb_info
*sbi
, struct dnode_of_data
*dn
,
1510 block_t old_addr
, block_t new_addr
,
1511 unsigned char version
, bool recover_curseg
,
1512 bool recover_newaddr
)
1514 struct f2fs_summary sum
;
1516 set_summary(&sum
, dn
->nid
, dn
->ofs_in_node
, version
);
1518 __f2fs_replace_block(sbi
, &sum
, old_addr
, new_addr
,
1519 recover_curseg
, recover_newaddr
);
1521 f2fs_update_data_blkaddr(dn
, new_addr
);
1524 void f2fs_wait_on_page_writeback(struct page
*page
,
1525 enum page_type type
, bool ordered
)
1527 if (PageWriteback(page
)) {
1528 struct f2fs_sb_info
*sbi
= F2FS_P_SB(page
);
1530 f2fs_submit_merged_bio_cond(sbi
, NULL
, page
, 0, type
, WRITE
);
1532 wait_on_page_writeback(page
);
1534 wait_for_stable_page(page
);
1538 void f2fs_wait_on_encrypted_page_writeback(struct f2fs_sb_info
*sbi
,
1543 if (blkaddr
== NEW_ADDR
)
1546 f2fs_bug_on(sbi
, blkaddr
== NULL_ADDR
);
1548 cpage
= find_lock_page(META_MAPPING(sbi
), blkaddr
);
1550 f2fs_wait_on_page_writeback(cpage
, DATA
, true);
1551 f2fs_put_page(cpage
, 1);
1555 static int read_compacted_summaries(struct f2fs_sb_info
*sbi
)
1557 struct f2fs_checkpoint
*ckpt
= F2FS_CKPT(sbi
);
1558 struct curseg_info
*seg_i
;
1559 unsigned char *kaddr
;
1564 start
= start_sum_block(sbi
);
1566 page
= get_meta_page(sbi
, start
++);
1567 kaddr
= (unsigned char *)page_address(page
);
1569 /* Step 1: restore nat cache */
1570 seg_i
= CURSEG_I(sbi
, CURSEG_HOT_DATA
);
1571 memcpy(seg_i
->journal
, kaddr
, SUM_JOURNAL_SIZE
);
1573 /* Step 2: restore sit cache */
1574 seg_i
= CURSEG_I(sbi
, CURSEG_COLD_DATA
);
1575 memcpy(seg_i
->journal
, kaddr
+ SUM_JOURNAL_SIZE
, SUM_JOURNAL_SIZE
);
1576 offset
= 2 * SUM_JOURNAL_SIZE
;
1578 /* Step 3: restore summary entries */
1579 for (i
= CURSEG_HOT_DATA
; i
<= CURSEG_COLD_DATA
; i
++) {
1580 unsigned short blk_off
;
1583 seg_i
= CURSEG_I(sbi
, i
);
1584 segno
= le32_to_cpu(ckpt
->cur_data_segno
[i
]);
1585 blk_off
= le16_to_cpu(ckpt
->cur_data_blkoff
[i
]);
1586 seg_i
->next_segno
= segno
;
1587 reset_curseg(sbi
, i
, 0);
1588 seg_i
->alloc_type
= ckpt
->alloc_type
[i
];
1589 seg_i
->next_blkoff
= blk_off
;
1591 if (seg_i
->alloc_type
== SSR
)
1592 blk_off
= sbi
->blocks_per_seg
;
1594 for (j
= 0; j
< blk_off
; j
++) {
1595 struct f2fs_summary
*s
;
1596 s
= (struct f2fs_summary
*)(kaddr
+ offset
);
1597 seg_i
->sum_blk
->entries
[j
] = *s
;
1598 offset
+= SUMMARY_SIZE
;
1599 if (offset
+ SUMMARY_SIZE
<= PAGE_CACHE_SIZE
-
1603 f2fs_put_page(page
, 1);
1606 page
= get_meta_page(sbi
, start
++);
1607 kaddr
= (unsigned char *)page_address(page
);
1611 f2fs_put_page(page
, 1);
1615 static int read_normal_summaries(struct f2fs_sb_info
*sbi
, int type
)
1617 struct f2fs_checkpoint
*ckpt
= F2FS_CKPT(sbi
);
1618 struct f2fs_summary_block
*sum
;
1619 struct curseg_info
*curseg
;
1621 unsigned short blk_off
;
1622 unsigned int segno
= 0;
1623 block_t blk_addr
= 0;
1625 /* get segment number and block addr */
1626 if (IS_DATASEG(type
)) {
1627 segno
= le32_to_cpu(ckpt
->cur_data_segno
[type
]);
1628 blk_off
= le16_to_cpu(ckpt
->cur_data_blkoff
[type
-
1630 if (__exist_node_summaries(sbi
))
1631 blk_addr
= sum_blk_addr(sbi
, NR_CURSEG_TYPE
, type
);
1633 blk_addr
= sum_blk_addr(sbi
, NR_CURSEG_DATA_TYPE
, type
);
1635 segno
= le32_to_cpu(ckpt
->cur_node_segno
[type
-
1637 blk_off
= le16_to_cpu(ckpt
->cur_node_blkoff
[type
-
1639 if (__exist_node_summaries(sbi
))
1640 blk_addr
= sum_blk_addr(sbi
, NR_CURSEG_NODE_TYPE
,
1641 type
- CURSEG_HOT_NODE
);
1643 blk_addr
= GET_SUM_BLOCK(sbi
, segno
);
1646 new = get_meta_page(sbi
, blk_addr
);
1647 sum
= (struct f2fs_summary_block
*)page_address(new);
1649 if (IS_NODESEG(type
)) {
1650 if (__exist_node_summaries(sbi
)) {
1651 struct f2fs_summary
*ns
= &sum
->entries
[0];
1653 for (i
= 0; i
< sbi
->blocks_per_seg
; i
++, ns
++) {
1655 ns
->ofs_in_node
= 0;
1660 err
= restore_node_summary(sbi
, segno
, sum
);
1662 f2fs_put_page(new, 1);
1668 /* set uncompleted segment to curseg */
1669 curseg
= CURSEG_I(sbi
, type
);
1670 mutex_lock(&curseg
->curseg_mutex
);
1672 /* update journal info */
1673 down_write(&curseg
->journal_rwsem
);
1674 memcpy(curseg
->journal
, &sum
->journal
, SUM_JOURNAL_SIZE
);
1675 up_write(&curseg
->journal_rwsem
);
1677 memcpy(curseg
->sum_blk
->entries
, sum
->entries
, SUM_ENTRY_SIZE
);
1678 memcpy(&curseg
->sum_blk
->footer
, &sum
->footer
, SUM_FOOTER_SIZE
);
1679 curseg
->next_segno
= segno
;
1680 reset_curseg(sbi
, type
, 0);
1681 curseg
->alloc_type
= ckpt
->alloc_type
[type
];
1682 curseg
->next_blkoff
= blk_off
;
1683 mutex_unlock(&curseg
->curseg_mutex
);
1684 f2fs_put_page(new, 1);
1688 static int restore_curseg_summaries(struct f2fs_sb_info
*sbi
)
1690 int type
= CURSEG_HOT_DATA
;
1693 if (is_set_ckpt_flags(F2FS_CKPT(sbi
), CP_COMPACT_SUM_FLAG
)) {
1694 int npages
= npages_for_summary_flush(sbi
, true);
1697 ra_meta_pages(sbi
, start_sum_block(sbi
), npages
,
1700 /* restore for compacted data summary */
1701 if (read_compacted_summaries(sbi
))
1703 type
= CURSEG_HOT_NODE
;
1706 if (__exist_node_summaries(sbi
))
1707 ra_meta_pages(sbi
, sum_blk_addr(sbi
, NR_CURSEG_TYPE
, type
),
1708 NR_CURSEG_TYPE
- type
, META_CP
, true);
1710 for (; type
<= CURSEG_COLD_NODE
; type
++) {
1711 err
= read_normal_summaries(sbi
, type
);
1719 static void write_compacted_summaries(struct f2fs_sb_info
*sbi
, block_t blkaddr
)
1722 unsigned char *kaddr
;
1723 struct f2fs_summary
*summary
;
1724 struct curseg_info
*seg_i
;
1725 int written_size
= 0;
1728 page
= grab_meta_page(sbi
, blkaddr
++);
1729 kaddr
= (unsigned char *)page_address(page
);
1731 /* Step 1: write nat cache */
1732 seg_i
= CURSEG_I(sbi
, CURSEG_HOT_DATA
);
1733 memcpy(kaddr
, seg_i
->journal
, SUM_JOURNAL_SIZE
);
1734 written_size
+= SUM_JOURNAL_SIZE
;
1736 /* Step 2: write sit cache */
1737 seg_i
= CURSEG_I(sbi
, CURSEG_COLD_DATA
);
1738 memcpy(kaddr
+ written_size
, seg_i
->journal
, SUM_JOURNAL_SIZE
);
1739 written_size
+= SUM_JOURNAL_SIZE
;
1741 /* Step 3: write summary entries */
1742 for (i
= CURSEG_HOT_DATA
; i
<= CURSEG_COLD_DATA
; i
++) {
1743 unsigned short blkoff
;
1744 seg_i
= CURSEG_I(sbi
, i
);
1745 if (sbi
->ckpt
->alloc_type
[i
] == SSR
)
1746 blkoff
= sbi
->blocks_per_seg
;
1748 blkoff
= curseg_blkoff(sbi
, i
);
1750 for (j
= 0; j
< blkoff
; j
++) {
1752 page
= grab_meta_page(sbi
, blkaddr
++);
1753 kaddr
= (unsigned char *)page_address(page
);
1756 summary
= (struct f2fs_summary
*)(kaddr
+ written_size
);
1757 *summary
= seg_i
->sum_blk
->entries
[j
];
1758 written_size
+= SUMMARY_SIZE
;
1760 if (written_size
+ SUMMARY_SIZE
<= PAGE_CACHE_SIZE
-
1764 set_page_dirty(page
);
1765 f2fs_put_page(page
, 1);
1770 set_page_dirty(page
);
1771 f2fs_put_page(page
, 1);
1775 static void write_normal_summaries(struct f2fs_sb_info
*sbi
,
1776 block_t blkaddr
, int type
)
1779 if (IS_DATASEG(type
))
1780 end
= type
+ NR_CURSEG_DATA_TYPE
;
1782 end
= type
+ NR_CURSEG_NODE_TYPE
;
1784 for (i
= type
; i
< end
; i
++)
1785 write_current_sum_page(sbi
, i
, blkaddr
+ (i
- type
));
1788 void write_data_summaries(struct f2fs_sb_info
*sbi
, block_t start_blk
)
1790 if (is_set_ckpt_flags(F2FS_CKPT(sbi
), CP_COMPACT_SUM_FLAG
))
1791 write_compacted_summaries(sbi
, start_blk
);
1793 write_normal_summaries(sbi
, start_blk
, CURSEG_HOT_DATA
);
1796 void write_node_summaries(struct f2fs_sb_info
*sbi
, block_t start_blk
)
1798 write_normal_summaries(sbi
, start_blk
, CURSEG_HOT_NODE
);
1801 int lookup_journal_in_cursum(struct f2fs_journal
*journal
, int type
,
1802 unsigned int val
, int alloc
)
1806 if (type
== NAT_JOURNAL
) {
1807 for (i
= 0; i
< nats_in_cursum(journal
); i
++) {
1808 if (le32_to_cpu(nid_in_journal(journal
, i
)) == val
)
1811 if (alloc
&& __has_cursum_space(journal
, 1, NAT_JOURNAL
))
1812 return update_nats_in_cursum(journal
, 1);
1813 } else if (type
== SIT_JOURNAL
) {
1814 for (i
= 0; i
< sits_in_cursum(journal
); i
++)
1815 if (le32_to_cpu(segno_in_journal(journal
, i
)) == val
)
1817 if (alloc
&& __has_cursum_space(journal
, 1, SIT_JOURNAL
))
1818 return update_sits_in_cursum(journal
, 1);
1823 static struct page
*get_current_sit_page(struct f2fs_sb_info
*sbi
,
1826 return get_meta_page(sbi
, current_sit_addr(sbi
, segno
));
1829 static struct page
*get_next_sit_page(struct f2fs_sb_info
*sbi
,
1832 struct sit_info
*sit_i
= SIT_I(sbi
);
1833 struct page
*src_page
, *dst_page
;
1834 pgoff_t src_off
, dst_off
;
1835 void *src_addr
, *dst_addr
;
1837 src_off
= current_sit_addr(sbi
, start
);
1838 dst_off
= next_sit_addr(sbi
, src_off
);
1840 /* get current sit block page without lock */
1841 src_page
= get_meta_page(sbi
, src_off
);
1842 dst_page
= grab_meta_page(sbi
, dst_off
);
1843 f2fs_bug_on(sbi
, PageDirty(src_page
));
1845 src_addr
= page_address(src_page
);
1846 dst_addr
= page_address(dst_page
);
1847 memcpy(dst_addr
, src_addr
, PAGE_CACHE_SIZE
);
1849 set_page_dirty(dst_page
);
1850 f2fs_put_page(src_page
, 1);
1852 set_to_next_sit(sit_i
, start
);
1857 static struct sit_entry_set
*grab_sit_entry_set(void)
1859 struct sit_entry_set
*ses
=
1860 f2fs_kmem_cache_alloc(sit_entry_set_slab
, GFP_NOFS
);
1863 INIT_LIST_HEAD(&ses
->set_list
);
1867 static void release_sit_entry_set(struct sit_entry_set
*ses
)
1869 list_del(&ses
->set_list
);
1870 kmem_cache_free(sit_entry_set_slab
, ses
);
1873 static void adjust_sit_entry_set(struct sit_entry_set
*ses
,
1874 struct list_head
*head
)
1876 struct sit_entry_set
*next
= ses
;
1878 if (list_is_last(&ses
->set_list
, head
))
1881 list_for_each_entry_continue(next
, head
, set_list
)
1882 if (ses
->entry_cnt
<= next
->entry_cnt
)
1885 list_move_tail(&ses
->set_list
, &next
->set_list
);
1888 static void add_sit_entry(unsigned int segno
, struct list_head
*head
)
1890 struct sit_entry_set
*ses
;
1891 unsigned int start_segno
= START_SEGNO(segno
);
1893 list_for_each_entry(ses
, head
, set_list
) {
1894 if (ses
->start_segno
== start_segno
) {
1896 adjust_sit_entry_set(ses
, head
);
1901 ses
= grab_sit_entry_set();
1903 ses
->start_segno
= start_segno
;
1905 list_add(&ses
->set_list
, head
);
1908 static void add_sits_in_set(struct f2fs_sb_info
*sbi
)
1910 struct f2fs_sm_info
*sm_info
= SM_I(sbi
);
1911 struct list_head
*set_list
= &sm_info
->sit_entry_set
;
1912 unsigned long *bitmap
= SIT_I(sbi
)->dirty_sentries_bitmap
;
1915 for_each_set_bit(segno
, bitmap
, MAIN_SEGS(sbi
))
1916 add_sit_entry(segno
, set_list
);
1919 static void remove_sits_in_journal(struct f2fs_sb_info
*sbi
)
1921 struct curseg_info
*curseg
= CURSEG_I(sbi
, CURSEG_COLD_DATA
);
1922 struct f2fs_journal
*journal
= curseg
->journal
;
1925 down_write(&curseg
->journal_rwsem
);
1926 for (i
= 0; i
< sits_in_cursum(journal
); i
++) {
1930 segno
= le32_to_cpu(segno_in_journal(journal
, i
));
1931 dirtied
= __mark_sit_entry_dirty(sbi
, segno
);
1934 add_sit_entry(segno
, &SM_I(sbi
)->sit_entry_set
);
1936 update_sits_in_cursum(journal
, -i
);
1937 up_write(&curseg
->journal_rwsem
);
1941 * CP calls this function, which flushes SIT entries including sit_journal,
1942 * and moves prefree segs to free segs.
1944 void flush_sit_entries(struct f2fs_sb_info
*sbi
, struct cp_control
*cpc
)
1946 struct sit_info
*sit_i
= SIT_I(sbi
);
1947 unsigned long *bitmap
= sit_i
->dirty_sentries_bitmap
;
1948 struct curseg_info
*curseg
= CURSEG_I(sbi
, CURSEG_COLD_DATA
);
1949 struct f2fs_journal
*journal
= curseg
->journal
;
1950 struct sit_entry_set
*ses
, *tmp
;
1951 struct list_head
*head
= &SM_I(sbi
)->sit_entry_set
;
1952 bool to_journal
= true;
1953 struct seg_entry
*se
;
1955 mutex_lock(&sit_i
->sentry_lock
);
1957 if (!sit_i
->dirty_sentries
)
1961 * add and account sit entries of dirty bitmap in sit entry
1964 add_sits_in_set(sbi
);
1967 * if there are no enough space in journal to store dirty sit
1968 * entries, remove all entries from journal and add and account
1969 * them in sit entry set.
1971 if (!__has_cursum_space(journal
, sit_i
->dirty_sentries
, SIT_JOURNAL
))
1972 remove_sits_in_journal(sbi
);
1975 * there are two steps to flush sit entries:
1976 * #1, flush sit entries to journal in current cold data summary block.
1977 * #2, flush sit entries to sit page.
1979 list_for_each_entry_safe(ses
, tmp
, head
, set_list
) {
1980 struct page
*page
= NULL
;
1981 struct f2fs_sit_block
*raw_sit
= NULL
;
1982 unsigned int start_segno
= ses
->start_segno
;
1983 unsigned int end
= min(start_segno
+ SIT_ENTRY_PER_BLOCK
,
1984 (unsigned long)MAIN_SEGS(sbi
));
1985 unsigned int segno
= start_segno
;
1988 !__has_cursum_space(journal
, ses
->entry_cnt
, SIT_JOURNAL
))
1992 down_write(&curseg
->journal_rwsem
);
1994 page
= get_next_sit_page(sbi
, start_segno
);
1995 raw_sit
= page_address(page
);
1998 /* flush dirty sit entries in region of current sit set */
1999 for_each_set_bit_from(segno
, bitmap
, end
) {
2000 int offset
, sit_offset
;
2002 se
= get_seg_entry(sbi
, segno
);
2004 /* add discard candidates */
2005 if (cpc
->reason
!= CP_DISCARD
) {
2006 cpc
->trim_start
= segno
;
2007 add_discard_addrs(sbi
, cpc
);
2011 offset
= lookup_journal_in_cursum(journal
,
2012 SIT_JOURNAL
, segno
, 1);
2013 f2fs_bug_on(sbi
, offset
< 0);
2014 segno_in_journal(journal
, offset
) =
2016 seg_info_to_raw_sit(se
,
2017 &sit_in_journal(journal
, offset
));
2019 sit_offset
= SIT_ENTRY_OFFSET(sit_i
, segno
);
2020 seg_info_to_raw_sit(se
,
2021 &raw_sit
->entries
[sit_offset
]);
2024 __clear_bit(segno
, bitmap
);
2025 sit_i
->dirty_sentries
--;
2030 up_write(&curseg
->journal_rwsem
);
2032 f2fs_put_page(page
, 1);
2034 f2fs_bug_on(sbi
, ses
->entry_cnt
);
2035 release_sit_entry_set(ses
);
2038 f2fs_bug_on(sbi
, !list_empty(head
));
2039 f2fs_bug_on(sbi
, sit_i
->dirty_sentries
);
2041 if (cpc
->reason
== CP_DISCARD
) {
2042 for (; cpc
->trim_start
<= cpc
->trim_end
; cpc
->trim_start
++)
2043 add_discard_addrs(sbi
, cpc
);
2045 mutex_unlock(&sit_i
->sentry_lock
);
2047 set_prefree_as_free_segments(sbi
);
2050 static int build_sit_info(struct f2fs_sb_info
*sbi
)
2052 struct f2fs_super_block
*raw_super
= F2FS_RAW_SUPER(sbi
);
2053 struct f2fs_checkpoint
*ckpt
= F2FS_CKPT(sbi
);
2054 struct sit_info
*sit_i
;
2055 unsigned int sit_segs
, start
;
2056 char *src_bitmap
, *dst_bitmap
;
2057 unsigned int bitmap_size
;
2059 /* allocate memory for SIT information */
2060 sit_i
= kzalloc(sizeof(struct sit_info
), GFP_KERNEL
);
2064 SM_I(sbi
)->sit_info
= sit_i
;
2066 sit_i
->sentries
= f2fs_kvzalloc(MAIN_SEGS(sbi
) *
2067 sizeof(struct seg_entry
), GFP_KERNEL
);
2068 if (!sit_i
->sentries
)
2071 bitmap_size
= f2fs_bitmap_size(MAIN_SEGS(sbi
));
2072 sit_i
->dirty_sentries_bitmap
= f2fs_kvzalloc(bitmap_size
, GFP_KERNEL
);
2073 if (!sit_i
->dirty_sentries_bitmap
)
2076 for (start
= 0; start
< MAIN_SEGS(sbi
); start
++) {
2077 sit_i
->sentries
[start
].cur_valid_map
2078 = kzalloc(SIT_VBLOCK_MAP_SIZE
, GFP_KERNEL
);
2079 sit_i
->sentries
[start
].ckpt_valid_map
2080 = kzalloc(SIT_VBLOCK_MAP_SIZE
, GFP_KERNEL
);
2081 sit_i
->sentries
[start
].discard_map
2082 = kzalloc(SIT_VBLOCK_MAP_SIZE
, GFP_KERNEL
);
2083 if (!sit_i
->sentries
[start
].cur_valid_map
||
2084 !sit_i
->sentries
[start
].ckpt_valid_map
||
2085 !sit_i
->sentries
[start
].discard_map
)
2089 sit_i
->tmp_map
= kzalloc(SIT_VBLOCK_MAP_SIZE
, GFP_KERNEL
);
2090 if (!sit_i
->tmp_map
)
2093 if (sbi
->segs_per_sec
> 1) {
2094 sit_i
->sec_entries
= f2fs_kvzalloc(MAIN_SECS(sbi
) *
2095 sizeof(struct sec_entry
), GFP_KERNEL
);
2096 if (!sit_i
->sec_entries
)
2100 /* get information related with SIT */
2101 sit_segs
= le32_to_cpu(raw_super
->segment_count_sit
) >> 1;
2103 /* setup SIT bitmap from ckeckpoint pack */
2104 bitmap_size
= __bitmap_size(sbi
, SIT_BITMAP
);
2105 src_bitmap
= __bitmap_ptr(sbi
, SIT_BITMAP
);
2107 dst_bitmap
= kmemdup(src_bitmap
, bitmap_size
, GFP_KERNEL
);
2111 /* init SIT information */
2112 sit_i
->s_ops
= &default_salloc_ops
;
2114 sit_i
->sit_base_addr
= le32_to_cpu(raw_super
->sit_blkaddr
);
2115 sit_i
->sit_blocks
= sit_segs
<< sbi
->log_blocks_per_seg
;
2116 sit_i
->written_valid_blocks
= le64_to_cpu(ckpt
->valid_block_count
);
2117 sit_i
->sit_bitmap
= dst_bitmap
;
2118 sit_i
->bitmap_size
= bitmap_size
;
2119 sit_i
->dirty_sentries
= 0;
2120 sit_i
->sents_per_block
= SIT_ENTRY_PER_BLOCK
;
2121 sit_i
->elapsed_time
= le64_to_cpu(sbi
->ckpt
->elapsed_time
);
2122 sit_i
->mounted_time
= CURRENT_TIME_SEC
.tv_sec
;
2123 mutex_init(&sit_i
->sentry_lock
);
2127 static int build_free_segmap(struct f2fs_sb_info
*sbi
)
2129 struct free_segmap_info
*free_i
;
2130 unsigned int bitmap_size
, sec_bitmap_size
;
2132 /* allocate memory for free segmap information */
2133 free_i
= kzalloc(sizeof(struct free_segmap_info
), GFP_KERNEL
);
2137 SM_I(sbi
)->free_info
= free_i
;
2139 bitmap_size
= f2fs_bitmap_size(MAIN_SEGS(sbi
));
2140 free_i
->free_segmap
= f2fs_kvmalloc(bitmap_size
, GFP_KERNEL
);
2141 if (!free_i
->free_segmap
)
2144 sec_bitmap_size
= f2fs_bitmap_size(MAIN_SECS(sbi
));
2145 free_i
->free_secmap
= f2fs_kvmalloc(sec_bitmap_size
, GFP_KERNEL
);
2146 if (!free_i
->free_secmap
)
2149 /* set all segments as dirty temporarily */
2150 memset(free_i
->free_segmap
, 0xff, bitmap_size
);
2151 memset(free_i
->free_secmap
, 0xff, sec_bitmap_size
);
2153 /* init free segmap information */
2154 free_i
->start_segno
= GET_SEGNO_FROM_SEG0(sbi
, MAIN_BLKADDR(sbi
));
2155 free_i
->free_segments
= 0;
2156 free_i
->free_sections
= 0;
2157 spin_lock_init(&free_i
->segmap_lock
);
2161 static int build_curseg(struct f2fs_sb_info
*sbi
)
2163 struct curseg_info
*array
;
2166 array
= kcalloc(NR_CURSEG_TYPE
, sizeof(*array
), GFP_KERNEL
);
2170 SM_I(sbi
)->curseg_array
= array
;
2172 for (i
= 0; i
< NR_CURSEG_TYPE
; i
++) {
2173 mutex_init(&array
[i
].curseg_mutex
);
2174 array
[i
].sum_blk
= kzalloc(PAGE_CACHE_SIZE
, GFP_KERNEL
);
2175 if (!array
[i
].sum_blk
)
2177 init_rwsem(&array
[i
].journal_rwsem
);
2178 array
[i
].journal
= kzalloc(sizeof(struct f2fs_journal
),
2180 if (!array
[i
].journal
)
2182 array
[i
].segno
= NULL_SEGNO
;
2183 array
[i
].next_blkoff
= 0;
2185 return restore_curseg_summaries(sbi
);
2188 static void build_sit_entries(struct f2fs_sb_info
*sbi
)
2190 struct sit_info
*sit_i
= SIT_I(sbi
);
2191 struct curseg_info
*curseg
= CURSEG_I(sbi
, CURSEG_COLD_DATA
);
2192 struct f2fs_journal
*journal
= curseg
->journal
;
2193 int sit_blk_cnt
= SIT_BLK_CNT(sbi
);
2194 unsigned int i
, start
, end
;
2195 unsigned int readed
, start_blk
= 0;
2196 int nrpages
= MAX_BIO_BLOCKS(sbi
) * 8;
2199 readed
= ra_meta_pages(sbi
, start_blk
, nrpages
, META_SIT
, true);
2201 start
= start_blk
* sit_i
->sents_per_block
;
2202 end
= (start_blk
+ readed
) * sit_i
->sents_per_block
;
2204 for (; start
< end
&& start
< MAIN_SEGS(sbi
); start
++) {
2205 struct seg_entry
*se
= &sit_i
->sentries
[start
];
2206 struct f2fs_sit_block
*sit_blk
;
2207 struct f2fs_sit_entry sit
;
2210 down_read(&curseg
->journal_rwsem
);
2211 for (i
= 0; i
< sits_in_cursum(journal
); i
++) {
2212 if (le32_to_cpu(segno_in_journal(journal
, i
))
2214 sit
= sit_in_journal(journal
, i
);
2215 up_read(&curseg
->journal_rwsem
);
2219 up_read(&curseg
->journal_rwsem
);
2221 page
= get_current_sit_page(sbi
, start
);
2222 sit_blk
= (struct f2fs_sit_block
*)page_address(page
);
2223 sit
= sit_blk
->entries
[SIT_ENTRY_OFFSET(sit_i
, start
)];
2224 f2fs_put_page(page
, 1);
2226 check_block_count(sbi
, start
, &sit
);
2227 seg_info_from_raw_sit(se
, &sit
);
2229 /* build discard map only one time */
2230 memcpy(se
->discard_map
, se
->cur_valid_map
, SIT_VBLOCK_MAP_SIZE
);
2231 sbi
->discard_blks
+= sbi
->blocks_per_seg
- se
->valid_blocks
;
2233 if (sbi
->segs_per_sec
> 1) {
2234 struct sec_entry
*e
= get_sec_entry(sbi
, start
);
2235 e
->valid_blocks
+= se
->valid_blocks
;
2238 start_blk
+= readed
;
2239 } while (start_blk
< sit_blk_cnt
);
2242 static void init_free_segmap(struct f2fs_sb_info
*sbi
)
2247 for (start
= 0; start
< MAIN_SEGS(sbi
); start
++) {
2248 struct seg_entry
*sentry
= get_seg_entry(sbi
, start
);
2249 if (!sentry
->valid_blocks
)
2250 __set_free(sbi
, start
);
2253 /* set use the current segments */
2254 for (type
= CURSEG_HOT_DATA
; type
<= CURSEG_COLD_NODE
; type
++) {
2255 struct curseg_info
*curseg_t
= CURSEG_I(sbi
, type
);
2256 __set_test_and_inuse(sbi
, curseg_t
->segno
);
2260 static void init_dirty_segmap(struct f2fs_sb_info
*sbi
)
2262 struct dirty_seglist_info
*dirty_i
= DIRTY_I(sbi
);
2263 struct free_segmap_info
*free_i
= FREE_I(sbi
);
2264 unsigned int segno
= 0, offset
= 0;
2265 unsigned short valid_blocks
;
2268 /* find dirty segment based on free segmap */
2269 segno
= find_next_inuse(free_i
, MAIN_SEGS(sbi
), offset
);
2270 if (segno
>= MAIN_SEGS(sbi
))
2273 valid_blocks
= get_valid_blocks(sbi
, segno
, 0);
2274 if (valid_blocks
== sbi
->blocks_per_seg
|| !valid_blocks
)
2276 if (valid_blocks
> sbi
->blocks_per_seg
) {
2277 f2fs_bug_on(sbi
, 1);
2280 mutex_lock(&dirty_i
->seglist_lock
);
2281 __locate_dirty_segment(sbi
, segno
, DIRTY
);
2282 mutex_unlock(&dirty_i
->seglist_lock
);
2286 static int init_victim_secmap(struct f2fs_sb_info
*sbi
)
2288 struct dirty_seglist_info
*dirty_i
= DIRTY_I(sbi
);
2289 unsigned int bitmap_size
= f2fs_bitmap_size(MAIN_SECS(sbi
));
2291 dirty_i
->victim_secmap
= f2fs_kvzalloc(bitmap_size
, GFP_KERNEL
);
2292 if (!dirty_i
->victim_secmap
)
2297 static int build_dirty_segmap(struct f2fs_sb_info
*sbi
)
2299 struct dirty_seglist_info
*dirty_i
;
2300 unsigned int bitmap_size
, i
;
2302 /* allocate memory for dirty segments list information */
2303 dirty_i
= kzalloc(sizeof(struct dirty_seglist_info
), GFP_KERNEL
);
2307 SM_I(sbi
)->dirty_info
= dirty_i
;
2308 mutex_init(&dirty_i
->seglist_lock
);
2310 bitmap_size
= f2fs_bitmap_size(MAIN_SEGS(sbi
));
2312 for (i
= 0; i
< NR_DIRTY_TYPE
; i
++) {
2313 dirty_i
->dirty_segmap
[i
] = f2fs_kvzalloc(bitmap_size
, GFP_KERNEL
);
2314 if (!dirty_i
->dirty_segmap
[i
])
2318 init_dirty_segmap(sbi
);
2319 return init_victim_secmap(sbi
);
2323 * Update min, max modified time for cost-benefit GC algorithm
2325 static void init_min_max_mtime(struct f2fs_sb_info
*sbi
)
2327 struct sit_info
*sit_i
= SIT_I(sbi
);
2330 mutex_lock(&sit_i
->sentry_lock
);
2332 sit_i
->min_mtime
= LLONG_MAX
;
2334 for (segno
= 0; segno
< MAIN_SEGS(sbi
); segno
+= sbi
->segs_per_sec
) {
2336 unsigned long long mtime
= 0;
2338 for (i
= 0; i
< sbi
->segs_per_sec
; i
++)
2339 mtime
+= get_seg_entry(sbi
, segno
+ i
)->mtime
;
2341 mtime
= div_u64(mtime
, sbi
->segs_per_sec
);
2343 if (sit_i
->min_mtime
> mtime
)
2344 sit_i
->min_mtime
= mtime
;
2346 sit_i
->max_mtime
= get_mtime(sbi
);
2347 mutex_unlock(&sit_i
->sentry_lock
);
2350 int build_segment_manager(struct f2fs_sb_info
*sbi
)
2352 struct f2fs_super_block
*raw_super
= F2FS_RAW_SUPER(sbi
);
2353 struct f2fs_checkpoint
*ckpt
= F2FS_CKPT(sbi
);
2354 struct f2fs_sm_info
*sm_info
;
2357 sm_info
= kzalloc(sizeof(struct f2fs_sm_info
), GFP_KERNEL
);
2362 sbi
->sm_info
= sm_info
;
2363 sm_info
->seg0_blkaddr
= le32_to_cpu(raw_super
->segment0_blkaddr
);
2364 sm_info
->main_blkaddr
= le32_to_cpu(raw_super
->main_blkaddr
);
2365 sm_info
->segment_count
= le32_to_cpu(raw_super
->segment_count
);
2366 sm_info
->reserved_segments
= le32_to_cpu(ckpt
->rsvd_segment_count
);
2367 sm_info
->ovp_segments
= le32_to_cpu(ckpt
->overprov_segment_count
);
2368 sm_info
->main_segments
= le32_to_cpu(raw_super
->segment_count_main
);
2369 sm_info
->ssa_blkaddr
= le32_to_cpu(raw_super
->ssa_blkaddr
);
2370 sm_info
->rec_prefree_segments
= sm_info
->main_segments
*
2371 DEF_RECLAIM_PREFREE_SEGMENTS
/ 100;
2372 sm_info
->ipu_policy
= 1 << F2FS_IPU_FSYNC
;
2373 sm_info
->min_ipu_util
= DEF_MIN_IPU_UTIL
;
2374 sm_info
->min_fsync_blocks
= DEF_MIN_FSYNC_BLOCKS
;
2376 INIT_LIST_HEAD(&sm_info
->discard_list
);
2377 sm_info
->nr_discards
= 0;
2378 sm_info
->max_discards
= 0;
2380 sm_info
->trim_sections
= DEF_BATCHED_TRIM_SECTIONS
;
2382 INIT_LIST_HEAD(&sm_info
->sit_entry_set
);
2384 if (test_opt(sbi
, FLUSH_MERGE
) && !f2fs_readonly(sbi
->sb
)) {
2385 err
= create_flush_cmd_control(sbi
);
2390 err
= build_sit_info(sbi
);
2393 err
= build_free_segmap(sbi
);
2396 err
= build_curseg(sbi
);
2400 /* reinit free segmap based on SIT */
2401 build_sit_entries(sbi
);
2403 init_free_segmap(sbi
);
2404 err
= build_dirty_segmap(sbi
);
2408 init_min_max_mtime(sbi
);
2412 static void discard_dirty_segmap(struct f2fs_sb_info
*sbi
,
2413 enum dirty_type dirty_type
)
2415 struct dirty_seglist_info
*dirty_i
= DIRTY_I(sbi
);
2417 mutex_lock(&dirty_i
->seglist_lock
);
2418 kvfree(dirty_i
->dirty_segmap
[dirty_type
]);
2419 dirty_i
->nr_dirty
[dirty_type
] = 0;
2420 mutex_unlock(&dirty_i
->seglist_lock
);
2423 static void destroy_victim_secmap(struct f2fs_sb_info
*sbi
)
2425 struct dirty_seglist_info
*dirty_i
= DIRTY_I(sbi
);
2426 kvfree(dirty_i
->victim_secmap
);
2429 static void destroy_dirty_segmap(struct f2fs_sb_info
*sbi
)
2431 struct dirty_seglist_info
*dirty_i
= DIRTY_I(sbi
);
2437 /* discard pre-free/dirty segments list */
2438 for (i
= 0; i
< NR_DIRTY_TYPE
; i
++)
2439 discard_dirty_segmap(sbi
, i
);
2441 destroy_victim_secmap(sbi
);
2442 SM_I(sbi
)->dirty_info
= NULL
;
2446 static void destroy_curseg(struct f2fs_sb_info
*sbi
)
2448 struct curseg_info
*array
= SM_I(sbi
)->curseg_array
;
2453 SM_I(sbi
)->curseg_array
= NULL
;
2454 for (i
= 0; i
< NR_CURSEG_TYPE
; i
++) {
2455 kfree(array
[i
].sum_blk
);
2456 kfree(array
[i
].journal
);
2461 static void destroy_free_segmap(struct f2fs_sb_info
*sbi
)
2463 struct free_segmap_info
*free_i
= SM_I(sbi
)->free_info
;
2466 SM_I(sbi
)->free_info
= NULL
;
2467 kvfree(free_i
->free_segmap
);
2468 kvfree(free_i
->free_secmap
);
2472 static void destroy_sit_info(struct f2fs_sb_info
*sbi
)
2474 struct sit_info
*sit_i
= SIT_I(sbi
);
2480 if (sit_i
->sentries
) {
2481 for (start
= 0; start
< MAIN_SEGS(sbi
); start
++) {
2482 kfree(sit_i
->sentries
[start
].cur_valid_map
);
2483 kfree(sit_i
->sentries
[start
].ckpt_valid_map
);
2484 kfree(sit_i
->sentries
[start
].discard_map
);
2487 kfree(sit_i
->tmp_map
);
2489 kvfree(sit_i
->sentries
);
2490 kvfree(sit_i
->sec_entries
);
2491 kvfree(sit_i
->dirty_sentries_bitmap
);
2493 SM_I(sbi
)->sit_info
= NULL
;
2494 kfree(sit_i
->sit_bitmap
);
2498 void destroy_segment_manager(struct f2fs_sb_info
*sbi
)
2500 struct f2fs_sm_info
*sm_info
= SM_I(sbi
);
2504 destroy_flush_cmd_control(sbi
);
2505 destroy_dirty_segmap(sbi
);
2506 destroy_curseg(sbi
);
2507 destroy_free_segmap(sbi
);
2508 destroy_sit_info(sbi
);
2509 sbi
->sm_info
= NULL
;
2513 int __init
create_segment_manager_caches(void)
2515 discard_entry_slab
= f2fs_kmem_cache_create("discard_entry",
2516 sizeof(struct discard_entry
));
2517 if (!discard_entry_slab
)
2520 sit_entry_set_slab
= f2fs_kmem_cache_create("sit_entry_set",
2521 sizeof(struct sit_entry_set
));
2522 if (!sit_entry_set_slab
)
2523 goto destory_discard_entry
;
2525 inmem_entry_slab
= f2fs_kmem_cache_create("inmem_page_entry",
2526 sizeof(struct inmem_pages
));
2527 if (!inmem_entry_slab
)
2528 goto destroy_sit_entry_set
;
2531 destroy_sit_entry_set
:
2532 kmem_cache_destroy(sit_entry_set_slab
);
2533 destory_discard_entry
:
2534 kmem_cache_destroy(discard_entry_slab
);
2539 void destroy_segment_manager_caches(void)
2541 kmem_cache_destroy(sit_entry_set_slab
);
2542 kmem_cache_destroy(discard_entry_slab
);
2543 kmem_cache_destroy(inmem_entry_slab
);