4 * Copyright (c) 2012 Samsung Electronics Co., Ltd.
5 * http://www.samsung.com/
7 * This program is free software; you can redistribute it and/or modify
8 * it under the terms of the GNU General Public License version 2 as
9 * published by the Free Software Foundation.
12 #include <linux/f2fs_fs.h>
13 #include <linux/bio.h>
14 #include <linux/blkdev.h>
15 #include <linux/prefetch.h>
16 #include <linux/kthread.h>
17 #include <linux/swap.h>
18 #include <linux/timer.h>
24 #include <trace/events/f2fs.h>
26 #define __reverse_ffz(x) __reverse_ffs(~(x))
28 static struct kmem_cache
*discard_entry_slab
;
29 static struct kmem_cache
*sit_entry_set_slab
;
30 static struct kmem_cache
*inmem_entry_slab
;
32 static unsigned long __reverse_ulong(unsigned char *str
)
34 unsigned long tmp
= 0;
35 int shift
= 24, idx
= 0;
37 #if BITS_PER_LONG == 64
41 tmp
|= (unsigned long)str
[idx
++] << shift
;
42 shift
-= BITS_PER_BYTE
;
48 * __reverse_ffs is copied from include/asm-generic/bitops/__ffs.h since
49 * MSB and LSB are reversed in a byte by f2fs_set_bit.
51 static inline unsigned long __reverse_ffs(unsigned long word
)
55 #if BITS_PER_LONG == 64
56 if ((word
& 0xffffffff00000000UL
) == 0)
61 if ((word
& 0xffff0000) == 0)
66 if ((word
& 0xff00) == 0)
71 if ((word
& 0xf0) == 0)
76 if ((word
& 0xc) == 0)
81 if ((word
& 0x2) == 0)
87 * __find_rev_next(_zero)_bit is copied from lib/find_next_bit.c because
88 * f2fs_set_bit makes MSB and LSB reversed in a byte.
89 * @size must be integral times of unsigned long.
92 * f2fs_set_bit(0, bitmap) => 1000 0000
93 * f2fs_set_bit(7, bitmap) => 0000 0001
95 static unsigned long __find_rev_next_bit(const unsigned long *addr
,
96 unsigned long size
, unsigned long offset
)
98 const unsigned long *p
= addr
+ BIT_WORD(offset
);
99 unsigned long result
= size
;
105 size
-= (offset
& ~(BITS_PER_LONG
- 1));
106 offset
%= BITS_PER_LONG
;
112 tmp
= __reverse_ulong((unsigned char *)p
);
114 tmp
&= ~0UL >> offset
;
115 if (size
< BITS_PER_LONG
)
116 tmp
&= (~0UL << (BITS_PER_LONG
- size
));
120 if (size
<= BITS_PER_LONG
)
122 size
-= BITS_PER_LONG
;
128 return result
- size
+ __reverse_ffs(tmp
);
131 static unsigned long __find_rev_next_zero_bit(const unsigned long *addr
,
132 unsigned long size
, unsigned long offset
)
134 const unsigned long *p
= addr
+ BIT_WORD(offset
);
135 unsigned long result
= size
;
141 size
-= (offset
& ~(BITS_PER_LONG
- 1));
142 offset
%= BITS_PER_LONG
;
148 tmp
= __reverse_ulong((unsigned char *)p
);
151 tmp
|= ~0UL << (BITS_PER_LONG
- offset
);
152 if (size
< BITS_PER_LONG
)
157 if (size
<= BITS_PER_LONG
)
159 size
-= BITS_PER_LONG
;
165 return result
- size
+ __reverse_ffz(tmp
);
168 void register_inmem_page(struct inode
*inode
, struct page
*page
)
170 struct f2fs_inode_info
*fi
= F2FS_I(inode
);
171 struct inmem_pages
*new;
173 f2fs_trace_pid(page
);
175 set_page_private(page
, (unsigned long)ATOMIC_WRITTEN_PAGE
);
176 SetPagePrivate(page
);
178 new = f2fs_kmem_cache_alloc(inmem_entry_slab
, GFP_NOFS
);
180 /* add atomic page indices to the list */
182 INIT_LIST_HEAD(&new->list
);
184 /* increase reference count with clean state */
185 mutex_lock(&fi
->inmem_lock
);
187 list_add_tail(&new->list
, &fi
->inmem_pages
);
188 inc_page_count(F2FS_I_SB(inode
), F2FS_INMEM_PAGES
);
189 mutex_unlock(&fi
->inmem_lock
);
191 trace_f2fs_register_inmem_page(page
, INMEM
);
194 static int __revoke_inmem_pages(struct inode
*inode
,
195 struct list_head
*head
, bool drop
, bool recover
)
197 struct f2fs_sb_info
*sbi
= F2FS_I_SB(inode
);
198 struct inmem_pages
*cur
, *tmp
;
201 list_for_each_entry_safe(cur
, tmp
, head
, list
) {
202 struct page
*page
= cur
->page
;
205 trace_f2fs_commit_inmem_page(page
, INMEM_DROP
);
210 struct dnode_of_data dn
;
213 trace_f2fs_commit_inmem_page(page
, INMEM_REVOKE
);
215 set_new_dnode(&dn
, inode
, NULL
, NULL
, 0);
216 if (get_dnode_of_data(&dn
, page
->index
, LOOKUP_NODE
)) {
220 get_node_info(sbi
, dn
.nid
, &ni
);
221 f2fs_replace_block(sbi
, &dn
, dn
.data_blkaddr
,
222 cur
->old_addr
, ni
.version
, true, true);
226 /* we don't need to invalidate this in the sccessful status */
228 ClearPageUptodate(page
);
229 set_page_private(page
, 0);
230 ClearPagePrivate(page
);
231 f2fs_put_page(page
, 1);
233 list_del(&cur
->list
);
234 kmem_cache_free(inmem_entry_slab
, cur
);
235 dec_page_count(F2FS_I_SB(inode
), F2FS_INMEM_PAGES
);
240 void drop_inmem_pages(struct inode
*inode
)
242 struct f2fs_inode_info
*fi
= F2FS_I(inode
);
244 clear_inode_flag(F2FS_I(inode
), FI_ATOMIC_FILE
);
246 mutex_lock(&fi
->inmem_lock
);
247 __revoke_inmem_pages(inode
, &fi
->inmem_pages
, true, false);
248 mutex_unlock(&fi
->inmem_lock
);
251 static int __commit_inmem_pages(struct inode
*inode
,
252 struct list_head
*revoke_list
)
254 struct f2fs_sb_info
*sbi
= F2FS_I_SB(inode
);
255 struct f2fs_inode_info
*fi
= F2FS_I(inode
);
256 struct inmem_pages
*cur
, *tmp
;
257 struct f2fs_io_info fio
= {
261 .op_flags
= WRITE_SYNC
| REQ_PRIO
,
262 .encrypted_page
= NULL
,
264 bool submit_bio
= false;
267 list_for_each_entry_safe(cur
, tmp
, &fi
->inmem_pages
, list
) {
268 struct page
*page
= cur
->page
;
271 if (page
->mapping
== inode
->i_mapping
) {
272 trace_f2fs_commit_inmem_page(page
, INMEM
);
274 set_page_dirty(page
);
275 f2fs_wait_on_page_writeback(page
, DATA
, true);
276 if (clear_page_dirty_for_io(page
))
277 inode_dec_dirty_pages(inode
);
280 err
= do_write_data_page(&fio
);
286 /* record old blkaddr for revoking */
287 cur
->old_addr
= fio
.old_blkaddr
;
289 clear_cold_data(page
);
293 list_move_tail(&cur
->list
, revoke_list
);
297 f2fs_submit_merged_bio_cond(sbi
, inode
, NULL
, 0, DATA
, WRITE
);
300 __revoke_inmem_pages(inode
, revoke_list
, false, false);
305 int commit_inmem_pages(struct inode
*inode
)
307 struct f2fs_sb_info
*sbi
= F2FS_I_SB(inode
);
308 struct f2fs_inode_info
*fi
= F2FS_I(inode
);
309 struct list_head revoke_list
;
312 INIT_LIST_HEAD(&revoke_list
);
313 f2fs_balance_fs(sbi
, true);
316 mutex_lock(&fi
->inmem_lock
);
317 err
= __commit_inmem_pages(inode
, &revoke_list
);
321 * try to revoke all committed pages, but still we could fail
322 * due to no memory or other reason, if that happened, EAGAIN
323 * will be returned, which means in such case, transaction is
324 * already not integrity, caller should use journal to do the
325 * recovery or rewrite & commit last transaction. For other
326 * error number, revoking was done by filesystem itself.
328 ret
= __revoke_inmem_pages(inode
, &revoke_list
, false, true);
332 /* drop all uncommitted pages */
333 __revoke_inmem_pages(inode
, &fi
->inmem_pages
, true, false);
335 mutex_unlock(&fi
->inmem_lock
);
342 * This function balances dirty node and dentry pages.
343 * In addition, it controls garbage collection.
345 void f2fs_balance_fs(struct f2fs_sb_info
*sbi
, bool need
)
350 * We should do GC or end up with checkpoint, if there are so many dirty
351 * dir/node pages without enough free segments.
353 if (has_not_enough_free_secs(sbi
, 0)) {
354 mutex_lock(&sbi
->gc_mutex
);
359 void f2fs_balance_fs_bg(struct f2fs_sb_info
*sbi
)
361 /* try to shrink extent cache when there is no enough memory */
362 if (!available_free_memory(sbi
, EXTENT_CACHE
))
363 f2fs_shrink_extent_tree(sbi
, EXTENT_CACHE_SHRINK_NUMBER
);
365 /* check the # of cached NAT entries */
366 if (!available_free_memory(sbi
, NAT_ENTRIES
))
367 try_to_free_nats(sbi
, NAT_ENTRY_PER_BLOCK
);
369 if (!available_free_memory(sbi
, FREE_NIDS
))
370 try_to_free_nids(sbi
, NAT_ENTRY_PER_BLOCK
* FREE_NID_PAGES
);
372 /* checkpoint is the only way to shrink partial cached entries */
373 if (!available_free_memory(sbi
, NAT_ENTRIES
) ||
374 !available_free_memory(sbi
, INO_ENTRIES
) ||
375 excess_prefree_segs(sbi
) ||
376 excess_dirty_nats(sbi
) ||
377 (is_idle(sbi
) && f2fs_time_over(sbi
, CP_TIME
))) {
378 if (test_opt(sbi
, DATA_FLUSH
)) {
379 struct blk_plug plug
;
381 blk_start_plug(&plug
);
382 sync_dirty_inodes(sbi
, FILE_INODE
);
383 blk_finish_plug(&plug
);
385 f2fs_sync_fs(sbi
->sb
, true);
386 stat_inc_bg_cp_count(sbi
->stat_info
);
390 static int issue_flush_thread(void *data
)
392 struct f2fs_sb_info
*sbi
= data
;
393 struct flush_cmd_control
*fcc
= SM_I(sbi
)->cmd_control_info
;
394 wait_queue_head_t
*q
= &fcc
->flush_wait_queue
;
396 if (kthread_should_stop())
399 if (!llist_empty(&fcc
->issue_list
)) {
401 struct flush_cmd
*cmd
, *next
;
404 bio
= f2fs_bio_alloc(0);
406 fcc
->dispatch_list
= llist_del_all(&fcc
->issue_list
);
407 fcc
->dispatch_list
= llist_reverse_order(fcc
->dispatch_list
);
409 bio
->bi_bdev
= sbi
->sb
->s_bdev
;
410 bio_set_op_attrs(bio
, REQ_OP_WRITE
, WRITE_FLUSH
);
411 ret
= submit_bio_wait(bio
);
413 llist_for_each_entry_safe(cmd
, next
,
414 fcc
->dispatch_list
, llnode
) {
416 complete(&cmd
->wait
);
419 fcc
->dispatch_list
= NULL
;
422 wait_event_interruptible(*q
,
423 kthread_should_stop() || !llist_empty(&fcc
->issue_list
));
427 int f2fs_issue_flush(struct f2fs_sb_info
*sbi
)
429 struct flush_cmd_control
*fcc
= SM_I(sbi
)->cmd_control_info
;
430 struct flush_cmd cmd
;
432 trace_f2fs_issue_flush(sbi
->sb
, test_opt(sbi
, NOBARRIER
),
433 test_opt(sbi
, FLUSH_MERGE
));
435 if (test_opt(sbi
, NOBARRIER
))
438 if (!test_opt(sbi
, FLUSH_MERGE
)) {
439 struct bio
*bio
= f2fs_bio_alloc(0);
442 bio
->bi_bdev
= sbi
->sb
->s_bdev
;
443 bio_set_op_attrs(bio
, REQ_OP_WRITE
, WRITE_FLUSH
);
444 ret
= submit_bio_wait(bio
);
449 init_completion(&cmd
.wait
);
451 llist_add(&cmd
.llnode
, &fcc
->issue_list
);
453 if (!fcc
->dispatch_list
)
454 wake_up(&fcc
->flush_wait_queue
);
456 wait_for_completion(&cmd
.wait
);
461 int create_flush_cmd_control(struct f2fs_sb_info
*sbi
)
463 dev_t dev
= sbi
->sb
->s_bdev
->bd_dev
;
464 struct flush_cmd_control
*fcc
;
467 fcc
= kzalloc(sizeof(struct flush_cmd_control
), GFP_KERNEL
);
470 init_waitqueue_head(&fcc
->flush_wait_queue
);
471 init_llist_head(&fcc
->issue_list
);
472 SM_I(sbi
)->cmd_control_info
= fcc
;
473 fcc
->f2fs_issue_flush
= kthread_run(issue_flush_thread
, sbi
,
474 "f2fs_flush-%u:%u", MAJOR(dev
), MINOR(dev
));
475 if (IS_ERR(fcc
->f2fs_issue_flush
)) {
476 err
= PTR_ERR(fcc
->f2fs_issue_flush
);
478 SM_I(sbi
)->cmd_control_info
= NULL
;
485 void destroy_flush_cmd_control(struct f2fs_sb_info
*sbi
)
487 struct flush_cmd_control
*fcc
= SM_I(sbi
)->cmd_control_info
;
489 if (fcc
&& fcc
->f2fs_issue_flush
)
490 kthread_stop(fcc
->f2fs_issue_flush
);
492 SM_I(sbi
)->cmd_control_info
= NULL
;
495 static void __locate_dirty_segment(struct f2fs_sb_info
*sbi
, unsigned int segno
,
496 enum dirty_type dirty_type
)
498 struct dirty_seglist_info
*dirty_i
= DIRTY_I(sbi
);
500 /* need not be added */
501 if (IS_CURSEG(sbi
, segno
))
504 if (!test_and_set_bit(segno
, dirty_i
->dirty_segmap
[dirty_type
]))
505 dirty_i
->nr_dirty
[dirty_type
]++;
507 if (dirty_type
== DIRTY
) {
508 struct seg_entry
*sentry
= get_seg_entry(sbi
, segno
);
509 enum dirty_type t
= sentry
->type
;
511 if (unlikely(t
>= DIRTY
)) {
515 if (!test_and_set_bit(segno
, dirty_i
->dirty_segmap
[t
]))
516 dirty_i
->nr_dirty
[t
]++;
520 static void __remove_dirty_segment(struct f2fs_sb_info
*sbi
, unsigned int segno
,
521 enum dirty_type dirty_type
)
523 struct dirty_seglist_info
*dirty_i
= DIRTY_I(sbi
);
525 if (test_and_clear_bit(segno
, dirty_i
->dirty_segmap
[dirty_type
]))
526 dirty_i
->nr_dirty
[dirty_type
]--;
528 if (dirty_type
== DIRTY
) {
529 struct seg_entry
*sentry
= get_seg_entry(sbi
, segno
);
530 enum dirty_type t
= sentry
->type
;
532 if (test_and_clear_bit(segno
, dirty_i
->dirty_segmap
[t
]))
533 dirty_i
->nr_dirty
[t
]--;
535 if (get_valid_blocks(sbi
, segno
, sbi
->segs_per_sec
) == 0)
536 clear_bit(GET_SECNO(sbi
, segno
),
537 dirty_i
->victim_secmap
);
542 * Should not occur error such as -ENOMEM.
543 * Adding dirty entry into seglist is not critical operation.
544 * If a given segment is one of current working segments, it won't be added.
546 static void locate_dirty_segment(struct f2fs_sb_info
*sbi
, unsigned int segno
)
548 struct dirty_seglist_info
*dirty_i
= DIRTY_I(sbi
);
549 unsigned short valid_blocks
;
551 if (segno
== NULL_SEGNO
|| IS_CURSEG(sbi
, segno
))
554 mutex_lock(&dirty_i
->seglist_lock
);
556 valid_blocks
= get_valid_blocks(sbi
, segno
, 0);
558 if (valid_blocks
== 0) {
559 __locate_dirty_segment(sbi
, segno
, PRE
);
560 __remove_dirty_segment(sbi
, segno
, DIRTY
);
561 } else if (valid_blocks
< sbi
->blocks_per_seg
) {
562 __locate_dirty_segment(sbi
, segno
, DIRTY
);
564 /* Recovery routine with SSR needs this */
565 __remove_dirty_segment(sbi
, segno
, DIRTY
);
568 mutex_unlock(&dirty_i
->seglist_lock
);
571 static int f2fs_issue_discard(struct f2fs_sb_info
*sbi
,
572 block_t blkstart
, block_t blklen
)
574 sector_t start
= SECTOR_FROM_BLOCK(blkstart
);
575 sector_t len
= SECTOR_FROM_BLOCK(blklen
);
576 struct seg_entry
*se
;
580 for (i
= blkstart
; i
< blkstart
+ blklen
; i
++) {
581 se
= get_seg_entry(sbi
, GET_SEGNO(sbi
, i
));
582 offset
= GET_BLKOFF_FROM_SEG0(sbi
, i
);
584 if (!f2fs_test_and_set_bit(offset
, se
->discard_map
))
587 trace_f2fs_issue_discard(sbi
->sb
, blkstart
, blklen
);
588 return blkdev_issue_discard(sbi
->sb
->s_bdev
, start
, len
, GFP_NOFS
, 0);
591 bool discard_next_dnode(struct f2fs_sb_info
*sbi
, block_t blkaddr
)
593 int err
= -EOPNOTSUPP
;
595 if (test_opt(sbi
, DISCARD
)) {
596 struct seg_entry
*se
= get_seg_entry(sbi
,
597 GET_SEGNO(sbi
, blkaddr
));
598 unsigned int offset
= GET_BLKOFF_FROM_SEG0(sbi
, blkaddr
);
600 if (f2fs_test_bit(offset
, se
->discard_map
))
603 err
= f2fs_issue_discard(sbi
, blkaddr
, 1);
607 update_meta_page(sbi
, NULL
, blkaddr
);
613 static void __add_discard_entry(struct f2fs_sb_info
*sbi
,
614 struct cp_control
*cpc
, struct seg_entry
*se
,
615 unsigned int start
, unsigned int end
)
617 struct list_head
*head
= &SM_I(sbi
)->discard_list
;
618 struct discard_entry
*new, *last
;
620 if (!list_empty(head
)) {
621 last
= list_last_entry(head
, struct discard_entry
, list
);
622 if (START_BLOCK(sbi
, cpc
->trim_start
) + start
==
623 last
->blkaddr
+ last
->len
) {
624 last
->len
+= end
- start
;
629 new = f2fs_kmem_cache_alloc(discard_entry_slab
, GFP_NOFS
);
630 INIT_LIST_HEAD(&new->list
);
631 new->blkaddr
= START_BLOCK(sbi
, cpc
->trim_start
) + start
;
632 new->len
= end
- start
;
633 list_add_tail(&new->list
, head
);
635 SM_I(sbi
)->nr_discards
+= end
- start
;
638 static void add_discard_addrs(struct f2fs_sb_info
*sbi
, struct cp_control
*cpc
)
640 int entries
= SIT_VBLOCK_MAP_SIZE
/ sizeof(unsigned long);
641 int max_blocks
= sbi
->blocks_per_seg
;
642 struct seg_entry
*se
= get_seg_entry(sbi
, cpc
->trim_start
);
643 unsigned long *cur_map
= (unsigned long *)se
->cur_valid_map
;
644 unsigned long *ckpt_map
= (unsigned long *)se
->ckpt_valid_map
;
645 unsigned long *discard_map
= (unsigned long *)se
->discard_map
;
646 unsigned long *dmap
= SIT_I(sbi
)->tmp_map
;
647 unsigned int start
= 0, end
= -1;
648 bool force
= (cpc
->reason
== CP_DISCARD
);
651 if (se
->valid_blocks
== max_blocks
)
655 if (!test_opt(sbi
, DISCARD
) || !se
->valid_blocks
||
656 SM_I(sbi
)->nr_discards
>= SM_I(sbi
)->max_discards
)
660 /* SIT_VBLOCK_MAP_SIZE should be multiple of sizeof(unsigned long) */
661 for (i
= 0; i
< entries
; i
++)
662 dmap
[i
] = force
? ~ckpt_map
[i
] & ~discard_map
[i
] :
663 (cur_map
[i
] ^ ckpt_map
[i
]) & ckpt_map
[i
];
665 while (force
|| SM_I(sbi
)->nr_discards
<= SM_I(sbi
)->max_discards
) {
666 start
= __find_rev_next_bit(dmap
, max_blocks
, end
+ 1);
667 if (start
>= max_blocks
)
670 end
= __find_rev_next_zero_bit(dmap
, max_blocks
, start
+ 1);
671 __add_discard_entry(sbi
, cpc
, se
, start
, end
);
675 void release_discard_addrs(struct f2fs_sb_info
*sbi
)
677 struct list_head
*head
= &(SM_I(sbi
)->discard_list
);
678 struct discard_entry
*entry
, *this;
681 list_for_each_entry_safe(entry
, this, head
, list
) {
682 list_del(&entry
->list
);
683 kmem_cache_free(discard_entry_slab
, entry
);
688 * Should call clear_prefree_segments after checkpoint is done.
690 static void set_prefree_as_free_segments(struct f2fs_sb_info
*sbi
)
692 struct dirty_seglist_info
*dirty_i
= DIRTY_I(sbi
);
695 mutex_lock(&dirty_i
->seglist_lock
);
696 for_each_set_bit(segno
, dirty_i
->dirty_segmap
[PRE
], MAIN_SEGS(sbi
))
697 __set_test_and_free(sbi
, segno
);
698 mutex_unlock(&dirty_i
->seglist_lock
);
701 void clear_prefree_segments(struct f2fs_sb_info
*sbi
, struct cp_control
*cpc
)
703 struct list_head
*head
= &(SM_I(sbi
)->discard_list
);
704 struct discard_entry
*entry
, *this;
705 struct dirty_seglist_info
*dirty_i
= DIRTY_I(sbi
);
706 unsigned long *prefree_map
= dirty_i
->dirty_segmap
[PRE
];
707 unsigned int start
= 0, end
= -1;
709 mutex_lock(&dirty_i
->seglist_lock
);
713 start
= find_next_bit(prefree_map
, MAIN_SEGS(sbi
), end
+ 1);
714 if (start
>= MAIN_SEGS(sbi
))
716 end
= find_next_zero_bit(prefree_map
, MAIN_SEGS(sbi
),
719 for (i
= start
; i
< end
; i
++)
720 clear_bit(i
, prefree_map
);
722 dirty_i
->nr_dirty
[PRE
] -= end
- start
;
724 if (!test_opt(sbi
, DISCARD
))
727 f2fs_issue_discard(sbi
, START_BLOCK(sbi
, start
),
728 (end
- start
) << sbi
->log_blocks_per_seg
);
730 mutex_unlock(&dirty_i
->seglist_lock
);
732 /* send small discards */
733 list_for_each_entry_safe(entry
, this, head
, list
) {
734 if (cpc
->reason
== CP_DISCARD
&& entry
->len
< cpc
->trim_minlen
)
736 f2fs_issue_discard(sbi
, entry
->blkaddr
, entry
->len
);
737 cpc
->trimmed
+= entry
->len
;
739 list_del(&entry
->list
);
740 SM_I(sbi
)->nr_discards
-= entry
->len
;
741 kmem_cache_free(discard_entry_slab
, entry
);
745 static bool __mark_sit_entry_dirty(struct f2fs_sb_info
*sbi
, unsigned int segno
)
747 struct sit_info
*sit_i
= SIT_I(sbi
);
749 if (!__test_and_set_bit(segno
, sit_i
->dirty_sentries_bitmap
)) {
750 sit_i
->dirty_sentries
++;
757 static void __set_sit_entry_type(struct f2fs_sb_info
*sbi
, int type
,
758 unsigned int segno
, int modified
)
760 struct seg_entry
*se
= get_seg_entry(sbi
, segno
);
763 __mark_sit_entry_dirty(sbi
, segno
);
766 static void update_sit_entry(struct f2fs_sb_info
*sbi
, block_t blkaddr
, int del
)
768 struct seg_entry
*se
;
769 unsigned int segno
, offset
;
770 long int new_vblocks
;
772 segno
= GET_SEGNO(sbi
, blkaddr
);
774 se
= get_seg_entry(sbi
, segno
);
775 new_vblocks
= se
->valid_blocks
+ del
;
776 offset
= GET_BLKOFF_FROM_SEG0(sbi
, blkaddr
);
778 f2fs_bug_on(sbi
, (new_vblocks
>> (sizeof(unsigned short) << 3) ||
779 (new_vblocks
> sbi
->blocks_per_seg
)));
781 se
->valid_blocks
= new_vblocks
;
782 se
->mtime
= get_mtime(sbi
);
783 SIT_I(sbi
)->max_mtime
= se
->mtime
;
785 /* Update valid block bitmap */
787 if (f2fs_test_and_set_bit(offset
, se
->cur_valid_map
))
789 if (!f2fs_test_and_set_bit(offset
, se
->discard_map
))
792 if (!f2fs_test_and_clear_bit(offset
, se
->cur_valid_map
))
794 if (f2fs_test_and_clear_bit(offset
, se
->discard_map
))
797 if (!f2fs_test_bit(offset
, se
->ckpt_valid_map
))
798 se
->ckpt_valid_blocks
+= del
;
800 __mark_sit_entry_dirty(sbi
, segno
);
802 /* update total number of valid blocks to be written in ckpt area */
803 SIT_I(sbi
)->written_valid_blocks
+= del
;
805 if (sbi
->segs_per_sec
> 1)
806 get_sec_entry(sbi
, segno
)->valid_blocks
+= del
;
809 void refresh_sit_entry(struct f2fs_sb_info
*sbi
, block_t old
, block_t
new)
811 update_sit_entry(sbi
, new, 1);
812 if (GET_SEGNO(sbi
, old
) != NULL_SEGNO
)
813 update_sit_entry(sbi
, old
, -1);
815 locate_dirty_segment(sbi
, GET_SEGNO(sbi
, old
));
816 locate_dirty_segment(sbi
, GET_SEGNO(sbi
, new));
819 void invalidate_blocks(struct f2fs_sb_info
*sbi
, block_t addr
)
821 unsigned int segno
= GET_SEGNO(sbi
, addr
);
822 struct sit_info
*sit_i
= SIT_I(sbi
);
824 f2fs_bug_on(sbi
, addr
== NULL_ADDR
);
825 if (addr
== NEW_ADDR
)
828 /* add it into sit main buffer */
829 mutex_lock(&sit_i
->sentry_lock
);
831 update_sit_entry(sbi
, addr
, -1);
833 /* add it into dirty seglist */
834 locate_dirty_segment(sbi
, segno
);
836 mutex_unlock(&sit_i
->sentry_lock
);
839 bool is_checkpointed_data(struct f2fs_sb_info
*sbi
, block_t blkaddr
)
841 struct sit_info
*sit_i
= SIT_I(sbi
);
842 unsigned int segno
, offset
;
843 struct seg_entry
*se
;
846 if (blkaddr
== NEW_ADDR
|| blkaddr
== NULL_ADDR
)
849 mutex_lock(&sit_i
->sentry_lock
);
851 segno
= GET_SEGNO(sbi
, blkaddr
);
852 se
= get_seg_entry(sbi
, segno
);
853 offset
= GET_BLKOFF_FROM_SEG0(sbi
, blkaddr
);
855 if (f2fs_test_bit(offset
, se
->ckpt_valid_map
))
858 mutex_unlock(&sit_i
->sentry_lock
);
864 * This function should be resided under the curseg_mutex lock
866 static void __add_sum_entry(struct f2fs_sb_info
*sbi
, int type
,
867 struct f2fs_summary
*sum
)
869 struct curseg_info
*curseg
= CURSEG_I(sbi
, type
);
870 void *addr
= curseg
->sum_blk
;
871 addr
+= curseg
->next_blkoff
* sizeof(struct f2fs_summary
);
872 memcpy(addr
, sum
, sizeof(struct f2fs_summary
));
876 * Calculate the number of current summary pages for writing
878 int npages_for_summary_flush(struct f2fs_sb_info
*sbi
, bool for_ra
)
880 int valid_sum_count
= 0;
883 for (i
= CURSEG_HOT_DATA
; i
<= CURSEG_COLD_DATA
; i
++) {
884 if (sbi
->ckpt
->alloc_type
[i
] == SSR
)
885 valid_sum_count
+= sbi
->blocks_per_seg
;
888 valid_sum_count
+= le16_to_cpu(
889 F2FS_CKPT(sbi
)->cur_data_blkoff
[i
]);
891 valid_sum_count
+= curseg_blkoff(sbi
, i
);
895 sum_in_page
= (PAGE_SIZE
- 2 * SUM_JOURNAL_SIZE
-
896 SUM_FOOTER_SIZE
) / SUMMARY_SIZE
;
897 if (valid_sum_count
<= sum_in_page
)
899 else if ((valid_sum_count
- sum_in_page
) <=
900 (PAGE_SIZE
- SUM_FOOTER_SIZE
) / SUMMARY_SIZE
)
906 * Caller should put this summary page
908 struct page
*get_sum_page(struct f2fs_sb_info
*sbi
, unsigned int segno
)
910 return get_meta_page(sbi
, GET_SUM_BLOCK(sbi
, segno
));
913 void update_meta_page(struct f2fs_sb_info
*sbi
, void *src
, block_t blk_addr
)
915 struct page
*page
= grab_meta_page(sbi
, blk_addr
);
916 void *dst
= page_address(page
);
919 memcpy(dst
, src
, PAGE_SIZE
);
921 memset(dst
, 0, PAGE_SIZE
);
922 set_page_dirty(page
);
923 f2fs_put_page(page
, 1);
926 static void write_sum_page(struct f2fs_sb_info
*sbi
,
927 struct f2fs_summary_block
*sum_blk
, block_t blk_addr
)
929 update_meta_page(sbi
, (void *)sum_blk
, blk_addr
);
932 static void write_current_sum_page(struct f2fs_sb_info
*sbi
,
933 int type
, block_t blk_addr
)
935 struct curseg_info
*curseg
= CURSEG_I(sbi
, type
);
936 struct page
*page
= grab_meta_page(sbi
, blk_addr
);
937 struct f2fs_summary_block
*src
= curseg
->sum_blk
;
938 struct f2fs_summary_block
*dst
;
940 dst
= (struct f2fs_summary_block
*)page_address(page
);
942 mutex_lock(&curseg
->curseg_mutex
);
944 down_read(&curseg
->journal_rwsem
);
945 memcpy(&dst
->journal
, curseg
->journal
, SUM_JOURNAL_SIZE
);
946 up_read(&curseg
->journal_rwsem
);
948 memcpy(dst
->entries
, src
->entries
, SUM_ENTRY_SIZE
);
949 memcpy(&dst
->footer
, &src
->footer
, SUM_FOOTER_SIZE
);
951 mutex_unlock(&curseg
->curseg_mutex
);
953 set_page_dirty(page
);
954 f2fs_put_page(page
, 1);
957 static int is_next_segment_free(struct f2fs_sb_info
*sbi
, int type
)
959 struct curseg_info
*curseg
= CURSEG_I(sbi
, type
);
960 unsigned int segno
= curseg
->segno
+ 1;
961 struct free_segmap_info
*free_i
= FREE_I(sbi
);
963 if (segno
< MAIN_SEGS(sbi
) && segno
% sbi
->segs_per_sec
)
964 return !test_bit(segno
, free_i
->free_segmap
);
969 * Find a new segment from the free segments bitmap to right order
970 * This function should be returned with success, otherwise BUG
972 static void get_new_segment(struct f2fs_sb_info
*sbi
,
973 unsigned int *newseg
, bool new_sec
, int dir
)
975 struct free_segmap_info
*free_i
= FREE_I(sbi
);
976 unsigned int segno
, secno
, zoneno
;
977 unsigned int total_zones
= MAIN_SECS(sbi
) / sbi
->secs_per_zone
;
978 unsigned int hint
= *newseg
/ sbi
->segs_per_sec
;
979 unsigned int old_zoneno
= GET_ZONENO_FROM_SEGNO(sbi
, *newseg
);
980 unsigned int left_start
= hint
;
985 spin_lock(&free_i
->segmap_lock
);
987 if (!new_sec
&& ((*newseg
+ 1) % sbi
->segs_per_sec
)) {
988 segno
= find_next_zero_bit(free_i
->free_segmap
,
989 (hint
+ 1) * sbi
->segs_per_sec
, *newseg
+ 1);
990 if (segno
< (hint
+ 1) * sbi
->segs_per_sec
)
994 secno
= find_next_zero_bit(free_i
->free_secmap
, MAIN_SECS(sbi
), hint
);
995 if (secno
>= MAIN_SECS(sbi
)) {
996 if (dir
== ALLOC_RIGHT
) {
997 secno
= find_next_zero_bit(free_i
->free_secmap
,
999 f2fs_bug_on(sbi
, secno
>= MAIN_SECS(sbi
));
1002 left_start
= hint
- 1;
1008 while (test_bit(left_start
, free_i
->free_secmap
)) {
1009 if (left_start
> 0) {
1013 left_start
= find_next_zero_bit(free_i
->free_secmap
,
1015 f2fs_bug_on(sbi
, left_start
>= MAIN_SECS(sbi
));
1021 segno
= secno
* sbi
->segs_per_sec
;
1022 zoneno
= secno
/ sbi
->secs_per_zone
;
1024 /* give up on finding another zone */
1027 if (sbi
->secs_per_zone
== 1)
1029 if (zoneno
== old_zoneno
)
1031 if (dir
== ALLOC_LEFT
) {
1032 if (!go_left
&& zoneno
+ 1 >= total_zones
)
1034 if (go_left
&& zoneno
== 0)
1037 for (i
= 0; i
< NR_CURSEG_TYPE
; i
++)
1038 if (CURSEG_I(sbi
, i
)->zone
== zoneno
)
1041 if (i
< NR_CURSEG_TYPE
) {
1042 /* zone is in user, try another */
1044 hint
= zoneno
* sbi
->secs_per_zone
- 1;
1045 else if (zoneno
+ 1 >= total_zones
)
1048 hint
= (zoneno
+ 1) * sbi
->secs_per_zone
;
1050 goto find_other_zone
;
1053 /* set it as dirty segment in free segmap */
1054 f2fs_bug_on(sbi
, test_bit(segno
, free_i
->free_segmap
));
1055 __set_inuse(sbi
, segno
);
1057 spin_unlock(&free_i
->segmap_lock
);
1060 static void reset_curseg(struct f2fs_sb_info
*sbi
, int type
, int modified
)
1062 struct curseg_info
*curseg
= CURSEG_I(sbi
, type
);
1063 struct summary_footer
*sum_footer
;
1065 curseg
->segno
= curseg
->next_segno
;
1066 curseg
->zone
= GET_ZONENO_FROM_SEGNO(sbi
, curseg
->segno
);
1067 curseg
->next_blkoff
= 0;
1068 curseg
->next_segno
= NULL_SEGNO
;
1070 sum_footer
= &(curseg
->sum_blk
->footer
);
1071 memset(sum_footer
, 0, sizeof(struct summary_footer
));
1072 if (IS_DATASEG(type
))
1073 SET_SUM_TYPE(sum_footer
, SUM_TYPE_DATA
);
1074 if (IS_NODESEG(type
))
1075 SET_SUM_TYPE(sum_footer
, SUM_TYPE_NODE
);
1076 __set_sit_entry_type(sbi
, type
, curseg
->segno
, modified
);
1080 * Allocate a current working segment.
1081 * This function always allocates a free segment in LFS manner.
1083 static void new_curseg(struct f2fs_sb_info
*sbi
, int type
, bool new_sec
)
1085 struct curseg_info
*curseg
= CURSEG_I(sbi
, type
);
1086 unsigned int segno
= curseg
->segno
;
1087 int dir
= ALLOC_LEFT
;
1089 write_sum_page(sbi
, curseg
->sum_blk
,
1090 GET_SUM_BLOCK(sbi
, segno
));
1091 if (type
== CURSEG_WARM_DATA
|| type
== CURSEG_COLD_DATA
)
1094 if (test_opt(sbi
, NOHEAP
))
1097 get_new_segment(sbi
, &segno
, new_sec
, dir
);
1098 curseg
->next_segno
= segno
;
1099 reset_curseg(sbi
, type
, 1);
1100 curseg
->alloc_type
= LFS
;
1103 static void __next_free_blkoff(struct f2fs_sb_info
*sbi
,
1104 struct curseg_info
*seg
, block_t start
)
1106 struct seg_entry
*se
= get_seg_entry(sbi
, seg
->segno
);
1107 int entries
= SIT_VBLOCK_MAP_SIZE
/ sizeof(unsigned long);
1108 unsigned long *target_map
= SIT_I(sbi
)->tmp_map
;
1109 unsigned long *ckpt_map
= (unsigned long *)se
->ckpt_valid_map
;
1110 unsigned long *cur_map
= (unsigned long *)se
->cur_valid_map
;
1113 for (i
= 0; i
< entries
; i
++)
1114 target_map
[i
] = ckpt_map
[i
] | cur_map
[i
];
1116 pos
= __find_rev_next_zero_bit(target_map
, sbi
->blocks_per_seg
, start
);
1118 seg
->next_blkoff
= pos
;
1122 * If a segment is written by LFS manner, next block offset is just obtained
1123 * by increasing the current block offset. However, if a segment is written by
1124 * SSR manner, next block offset obtained by calling __next_free_blkoff
1126 static void __refresh_next_blkoff(struct f2fs_sb_info
*sbi
,
1127 struct curseg_info
*seg
)
1129 if (seg
->alloc_type
== SSR
)
1130 __next_free_blkoff(sbi
, seg
, seg
->next_blkoff
+ 1);
1136 * This function always allocates a used segment(from dirty seglist) by SSR
1137 * manner, so it should recover the existing segment information of valid blocks
1139 static void change_curseg(struct f2fs_sb_info
*sbi
, int type
, bool reuse
)
1141 struct dirty_seglist_info
*dirty_i
= DIRTY_I(sbi
);
1142 struct curseg_info
*curseg
= CURSEG_I(sbi
, type
);
1143 unsigned int new_segno
= curseg
->next_segno
;
1144 struct f2fs_summary_block
*sum_node
;
1145 struct page
*sum_page
;
1147 write_sum_page(sbi
, curseg
->sum_blk
,
1148 GET_SUM_BLOCK(sbi
, curseg
->segno
));
1149 __set_test_and_inuse(sbi
, new_segno
);
1151 mutex_lock(&dirty_i
->seglist_lock
);
1152 __remove_dirty_segment(sbi
, new_segno
, PRE
);
1153 __remove_dirty_segment(sbi
, new_segno
, DIRTY
);
1154 mutex_unlock(&dirty_i
->seglist_lock
);
1156 reset_curseg(sbi
, type
, 1);
1157 curseg
->alloc_type
= SSR
;
1158 __next_free_blkoff(sbi
, curseg
, 0);
1161 sum_page
= get_sum_page(sbi
, new_segno
);
1162 sum_node
= (struct f2fs_summary_block
*)page_address(sum_page
);
1163 memcpy(curseg
->sum_blk
, sum_node
, SUM_ENTRY_SIZE
);
1164 f2fs_put_page(sum_page
, 1);
1168 static int get_ssr_segment(struct f2fs_sb_info
*sbi
, int type
)
1170 struct curseg_info
*curseg
= CURSEG_I(sbi
, type
);
1171 const struct victim_selection
*v_ops
= DIRTY_I(sbi
)->v_ops
;
1173 if (IS_NODESEG(type
) || !has_not_enough_free_secs(sbi
, 0))
1174 return v_ops
->get_victim(sbi
,
1175 &(curseg
)->next_segno
, BG_GC
, type
, SSR
);
1177 /* For data segments, let's do SSR more intensively */
1178 for (; type
>= CURSEG_HOT_DATA
; type
--)
1179 if (v_ops
->get_victim(sbi
, &(curseg
)->next_segno
,
1186 * flush out current segment and replace it with new segment
1187 * This function should be returned with success, otherwise BUG
1189 static void allocate_segment_by_default(struct f2fs_sb_info
*sbi
,
1190 int type
, bool force
)
1192 struct curseg_info
*curseg
= CURSEG_I(sbi
, type
);
1195 new_curseg(sbi
, type
, true);
1196 else if (type
== CURSEG_WARM_NODE
)
1197 new_curseg(sbi
, type
, false);
1198 else if (curseg
->alloc_type
== LFS
&& is_next_segment_free(sbi
, type
))
1199 new_curseg(sbi
, type
, false);
1200 else if (need_SSR(sbi
) && get_ssr_segment(sbi
, type
))
1201 change_curseg(sbi
, type
, true);
1203 new_curseg(sbi
, type
, false);
1205 stat_inc_seg_type(sbi
, curseg
);
1208 static void __allocate_new_segments(struct f2fs_sb_info
*sbi
, int type
)
1210 struct curseg_info
*curseg
= CURSEG_I(sbi
, type
);
1211 unsigned int old_segno
;
1213 old_segno
= curseg
->segno
;
1214 SIT_I(sbi
)->s_ops
->allocate_segment(sbi
, type
, true);
1215 locate_dirty_segment(sbi
, old_segno
);
1218 void allocate_new_segments(struct f2fs_sb_info
*sbi
)
1222 for (i
= CURSEG_HOT_DATA
; i
<= CURSEG_COLD_DATA
; i
++)
1223 __allocate_new_segments(sbi
, i
);
1226 static const struct segment_allocation default_salloc_ops
= {
1227 .allocate_segment
= allocate_segment_by_default
,
1230 int f2fs_trim_fs(struct f2fs_sb_info
*sbi
, struct fstrim_range
*range
)
1232 __u64 start
= F2FS_BYTES_TO_BLK(range
->start
);
1233 __u64 end
= start
+ F2FS_BYTES_TO_BLK(range
->len
) - 1;
1234 unsigned int start_segno
, end_segno
;
1235 struct cp_control cpc
;
1238 if (start
>= MAX_BLKADDR(sbi
) || range
->len
< sbi
->blocksize
)
1242 if (end
<= MAIN_BLKADDR(sbi
))
1245 /* start/end segment number in main_area */
1246 start_segno
= (start
<= MAIN_BLKADDR(sbi
)) ? 0 : GET_SEGNO(sbi
, start
);
1247 end_segno
= (end
>= MAX_BLKADDR(sbi
)) ? MAIN_SEGS(sbi
) - 1 :
1248 GET_SEGNO(sbi
, end
);
1249 cpc
.reason
= CP_DISCARD
;
1250 cpc
.trim_minlen
= max_t(__u64
, 1, F2FS_BYTES_TO_BLK(range
->minlen
));
1252 /* do checkpoint to issue discard commands safely */
1253 for (; start_segno
<= end_segno
; start_segno
= cpc
.trim_end
+ 1) {
1254 cpc
.trim_start
= start_segno
;
1256 if (sbi
->discard_blks
== 0)
1258 else if (sbi
->discard_blks
< BATCHED_TRIM_BLOCKS(sbi
))
1259 cpc
.trim_end
= end_segno
;
1261 cpc
.trim_end
= min_t(unsigned int,
1262 rounddown(start_segno
+
1263 BATCHED_TRIM_SEGMENTS(sbi
),
1264 sbi
->segs_per_sec
) - 1, end_segno
);
1266 mutex_lock(&sbi
->gc_mutex
);
1267 err
= write_checkpoint(sbi
, &cpc
);
1268 mutex_unlock(&sbi
->gc_mutex
);
1271 range
->len
= F2FS_BLK_TO_BYTES(cpc
.trimmed
);
1275 static bool __has_curseg_space(struct f2fs_sb_info
*sbi
, int type
)
1277 struct curseg_info
*curseg
= CURSEG_I(sbi
, type
);
1278 if (curseg
->next_blkoff
< sbi
->blocks_per_seg
)
1283 static int __get_segment_type_2(struct page
*page
, enum page_type p_type
)
1286 return CURSEG_HOT_DATA
;
1288 return CURSEG_HOT_NODE
;
1291 static int __get_segment_type_4(struct page
*page
, enum page_type p_type
)
1293 if (p_type
== DATA
) {
1294 struct inode
*inode
= page
->mapping
->host
;
1296 if (S_ISDIR(inode
->i_mode
))
1297 return CURSEG_HOT_DATA
;
1299 return CURSEG_COLD_DATA
;
1301 if (IS_DNODE(page
) && is_cold_node(page
))
1302 return CURSEG_WARM_NODE
;
1304 return CURSEG_COLD_NODE
;
1308 static int __get_segment_type_6(struct page
*page
, enum page_type p_type
)
1310 if (p_type
== DATA
) {
1311 struct inode
*inode
= page
->mapping
->host
;
1313 if (S_ISDIR(inode
->i_mode
))
1314 return CURSEG_HOT_DATA
;
1315 else if (is_cold_data(page
) || file_is_cold(inode
))
1316 return CURSEG_COLD_DATA
;
1318 return CURSEG_WARM_DATA
;
1321 return is_cold_node(page
) ? CURSEG_WARM_NODE
:
1324 return CURSEG_COLD_NODE
;
1328 static int __get_segment_type(struct page
*page
, enum page_type p_type
)
1330 switch (F2FS_P_SB(page
)->active_logs
) {
1332 return __get_segment_type_2(page
, p_type
);
1334 return __get_segment_type_4(page
, p_type
);
1336 /* NR_CURSEG_TYPE(6) logs by default */
1337 f2fs_bug_on(F2FS_P_SB(page
),
1338 F2FS_P_SB(page
)->active_logs
!= NR_CURSEG_TYPE
);
1339 return __get_segment_type_6(page
, p_type
);
1342 void allocate_data_block(struct f2fs_sb_info
*sbi
, struct page
*page
,
1343 block_t old_blkaddr
, block_t
*new_blkaddr
,
1344 struct f2fs_summary
*sum
, int type
)
1346 struct sit_info
*sit_i
= SIT_I(sbi
);
1347 struct curseg_info
*curseg
;
1348 bool direct_io
= (type
== CURSEG_DIRECT_IO
);
1350 type
= direct_io
? CURSEG_WARM_DATA
: type
;
1352 curseg
= CURSEG_I(sbi
, type
);
1354 mutex_lock(&curseg
->curseg_mutex
);
1355 mutex_lock(&sit_i
->sentry_lock
);
1357 /* direct_io'ed data is aligned to the segment for better performance */
1358 if (direct_io
&& curseg
->next_blkoff
&&
1359 !has_not_enough_free_secs(sbi
, 0))
1360 __allocate_new_segments(sbi
, type
);
1362 *new_blkaddr
= NEXT_FREE_BLKADDR(sbi
, curseg
);
1365 * __add_sum_entry should be resided under the curseg_mutex
1366 * because, this function updates a summary entry in the
1367 * current summary block.
1369 __add_sum_entry(sbi
, type
, sum
);
1371 __refresh_next_blkoff(sbi
, curseg
);
1373 stat_inc_block_count(sbi
, curseg
);
1375 if (!__has_curseg_space(sbi
, type
))
1376 sit_i
->s_ops
->allocate_segment(sbi
, type
, false);
1378 * SIT information should be updated before segment allocation,
1379 * since SSR needs latest valid block information.
1381 refresh_sit_entry(sbi
, old_blkaddr
, *new_blkaddr
);
1383 mutex_unlock(&sit_i
->sentry_lock
);
1385 if (page
&& IS_NODESEG(type
))
1386 fill_node_footer_blkaddr(page
, NEXT_FREE_BLKADDR(sbi
, curseg
));
1388 mutex_unlock(&curseg
->curseg_mutex
);
1391 static void do_write_page(struct f2fs_summary
*sum
, struct f2fs_io_info
*fio
)
1393 int type
= __get_segment_type(fio
->page
, fio
->type
);
1395 allocate_data_block(fio
->sbi
, fio
->page
, fio
->old_blkaddr
,
1396 &fio
->new_blkaddr
, sum
, type
);
1398 /* writeout dirty page into bdev */
1399 f2fs_submit_page_mbio(fio
);
1402 void write_meta_page(struct f2fs_sb_info
*sbi
, struct page
*page
)
1404 struct f2fs_io_info fio
= {
1408 .op_flags
= WRITE_SYNC
| REQ_META
| REQ_PRIO
,
1409 .old_blkaddr
= page
->index
,
1410 .new_blkaddr
= page
->index
,
1412 .encrypted_page
= NULL
,
1415 if (unlikely(page
->index
>= MAIN_BLKADDR(sbi
)))
1416 fio
.op_flags
&= ~REQ_META
;
1418 set_page_writeback(page
);
1419 f2fs_submit_page_mbio(&fio
);
1422 void write_node_page(unsigned int nid
, struct f2fs_io_info
*fio
)
1424 struct f2fs_summary sum
;
1426 set_summary(&sum
, nid
, 0, 0);
1427 do_write_page(&sum
, fio
);
1430 void write_data_page(struct dnode_of_data
*dn
, struct f2fs_io_info
*fio
)
1432 struct f2fs_sb_info
*sbi
= fio
->sbi
;
1433 struct f2fs_summary sum
;
1434 struct node_info ni
;
1436 f2fs_bug_on(sbi
, dn
->data_blkaddr
== NULL_ADDR
);
1437 get_node_info(sbi
, dn
->nid
, &ni
);
1438 set_summary(&sum
, dn
->nid
, dn
->ofs_in_node
, ni
.version
);
1439 do_write_page(&sum
, fio
);
1440 f2fs_update_data_blkaddr(dn
, fio
->new_blkaddr
);
1443 void rewrite_data_page(struct f2fs_io_info
*fio
)
1445 fio
->new_blkaddr
= fio
->old_blkaddr
;
1446 stat_inc_inplace_blocks(fio
->sbi
);
1447 f2fs_submit_page_mbio(fio
);
1450 void __f2fs_replace_block(struct f2fs_sb_info
*sbi
, struct f2fs_summary
*sum
,
1451 block_t old_blkaddr
, block_t new_blkaddr
,
1452 bool recover_curseg
, bool recover_newaddr
)
1454 struct sit_info
*sit_i
= SIT_I(sbi
);
1455 struct curseg_info
*curseg
;
1456 unsigned int segno
, old_cursegno
;
1457 struct seg_entry
*se
;
1459 unsigned short old_blkoff
;
1461 segno
= GET_SEGNO(sbi
, new_blkaddr
);
1462 se
= get_seg_entry(sbi
, segno
);
1465 if (!recover_curseg
) {
1466 /* for recovery flow */
1467 if (se
->valid_blocks
== 0 && !IS_CURSEG(sbi
, segno
)) {
1468 if (old_blkaddr
== NULL_ADDR
)
1469 type
= CURSEG_COLD_DATA
;
1471 type
= CURSEG_WARM_DATA
;
1474 if (!IS_CURSEG(sbi
, segno
))
1475 type
= CURSEG_WARM_DATA
;
1478 curseg
= CURSEG_I(sbi
, type
);
1480 mutex_lock(&curseg
->curseg_mutex
);
1481 mutex_lock(&sit_i
->sentry_lock
);
1483 old_cursegno
= curseg
->segno
;
1484 old_blkoff
= curseg
->next_blkoff
;
1486 /* change the current segment */
1487 if (segno
!= curseg
->segno
) {
1488 curseg
->next_segno
= segno
;
1489 change_curseg(sbi
, type
, true);
1492 curseg
->next_blkoff
= GET_BLKOFF_FROM_SEG0(sbi
, new_blkaddr
);
1493 __add_sum_entry(sbi
, type
, sum
);
1495 if (!recover_curseg
|| recover_newaddr
)
1496 update_sit_entry(sbi
, new_blkaddr
, 1);
1497 if (GET_SEGNO(sbi
, old_blkaddr
) != NULL_SEGNO
)
1498 update_sit_entry(sbi
, old_blkaddr
, -1);
1500 locate_dirty_segment(sbi
, GET_SEGNO(sbi
, old_blkaddr
));
1501 locate_dirty_segment(sbi
, GET_SEGNO(sbi
, new_blkaddr
));
1503 locate_dirty_segment(sbi
, old_cursegno
);
1505 if (recover_curseg
) {
1506 if (old_cursegno
!= curseg
->segno
) {
1507 curseg
->next_segno
= old_cursegno
;
1508 change_curseg(sbi
, type
, true);
1510 curseg
->next_blkoff
= old_blkoff
;
1513 mutex_unlock(&sit_i
->sentry_lock
);
1514 mutex_unlock(&curseg
->curseg_mutex
);
1517 void f2fs_replace_block(struct f2fs_sb_info
*sbi
, struct dnode_of_data
*dn
,
1518 block_t old_addr
, block_t new_addr
,
1519 unsigned char version
, bool recover_curseg
,
1520 bool recover_newaddr
)
1522 struct f2fs_summary sum
;
1524 set_summary(&sum
, dn
->nid
, dn
->ofs_in_node
, version
);
1526 __f2fs_replace_block(sbi
, &sum
, old_addr
, new_addr
,
1527 recover_curseg
, recover_newaddr
);
1529 f2fs_update_data_blkaddr(dn
, new_addr
);
1532 void f2fs_wait_on_page_writeback(struct page
*page
,
1533 enum page_type type
, bool ordered
)
1535 if (PageWriteback(page
)) {
1536 struct f2fs_sb_info
*sbi
= F2FS_P_SB(page
);
1538 f2fs_submit_merged_bio_cond(sbi
, NULL
, page
, 0, type
, WRITE
);
1540 wait_on_page_writeback(page
);
1542 wait_for_stable_page(page
);
1546 void f2fs_wait_on_encrypted_page_writeback(struct f2fs_sb_info
*sbi
,
1551 if (blkaddr
== NEW_ADDR
)
1554 f2fs_bug_on(sbi
, blkaddr
== NULL_ADDR
);
1556 cpage
= find_lock_page(META_MAPPING(sbi
), blkaddr
);
1558 f2fs_wait_on_page_writeback(cpage
, DATA
, true);
1559 f2fs_put_page(cpage
, 1);
1563 static int read_compacted_summaries(struct f2fs_sb_info
*sbi
)
1565 struct f2fs_checkpoint
*ckpt
= F2FS_CKPT(sbi
);
1566 struct curseg_info
*seg_i
;
1567 unsigned char *kaddr
;
1572 start
= start_sum_block(sbi
);
1574 page
= get_meta_page(sbi
, start
++);
1575 kaddr
= (unsigned char *)page_address(page
);
1577 /* Step 1: restore nat cache */
1578 seg_i
= CURSEG_I(sbi
, CURSEG_HOT_DATA
);
1579 memcpy(seg_i
->journal
, kaddr
, SUM_JOURNAL_SIZE
);
1581 /* Step 2: restore sit cache */
1582 seg_i
= CURSEG_I(sbi
, CURSEG_COLD_DATA
);
1583 memcpy(seg_i
->journal
, kaddr
+ SUM_JOURNAL_SIZE
, SUM_JOURNAL_SIZE
);
1584 offset
= 2 * SUM_JOURNAL_SIZE
;
1586 /* Step 3: restore summary entries */
1587 for (i
= CURSEG_HOT_DATA
; i
<= CURSEG_COLD_DATA
; i
++) {
1588 unsigned short blk_off
;
1591 seg_i
= CURSEG_I(sbi
, i
);
1592 segno
= le32_to_cpu(ckpt
->cur_data_segno
[i
]);
1593 blk_off
= le16_to_cpu(ckpt
->cur_data_blkoff
[i
]);
1594 seg_i
->next_segno
= segno
;
1595 reset_curseg(sbi
, i
, 0);
1596 seg_i
->alloc_type
= ckpt
->alloc_type
[i
];
1597 seg_i
->next_blkoff
= blk_off
;
1599 if (seg_i
->alloc_type
== SSR
)
1600 blk_off
= sbi
->blocks_per_seg
;
1602 for (j
= 0; j
< blk_off
; j
++) {
1603 struct f2fs_summary
*s
;
1604 s
= (struct f2fs_summary
*)(kaddr
+ offset
);
1605 seg_i
->sum_blk
->entries
[j
] = *s
;
1606 offset
+= SUMMARY_SIZE
;
1607 if (offset
+ SUMMARY_SIZE
<= PAGE_SIZE
-
1611 f2fs_put_page(page
, 1);
1614 page
= get_meta_page(sbi
, start
++);
1615 kaddr
= (unsigned char *)page_address(page
);
1619 f2fs_put_page(page
, 1);
1623 static int read_normal_summaries(struct f2fs_sb_info
*sbi
, int type
)
1625 struct f2fs_checkpoint
*ckpt
= F2FS_CKPT(sbi
);
1626 struct f2fs_summary_block
*sum
;
1627 struct curseg_info
*curseg
;
1629 unsigned short blk_off
;
1630 unsigned int segno
= 0;
1631 block_t blk_addr
= 0;
1633 /* get segment number and block addr */
1634 if (IS_DATASEG(type
)) {
1635 segno
= le32_to_cpu(ckpt
->cur_data_segno
[type
]);
1636 blk_off
= le16_to_cpu(ckpt
->cur_data_blkoff
[type
-
1638 if (__exist_node_summaries(sbi
))
1639 blk_addr
= sum_blk_addr(sbi
, NR_CURSEG_TYPE
, type
);
1641 blk_addr
= sum_blk_addr(sbi
, NR_CURSEG_DATA_TYPE
, type
);
1643 segno
= le32_to_cpu(ckpt
->cur_node_segno
[type
-
1645 blk_off
= le16_to_cpu(ckpt
->cur_node_blkoff
[type
-
1647 if (__exist_node_summaries(sbi
))
1648 blk_addr
= sum_blk_addr(sbi
, NR_CURSEG_NODE_TYPE
,
1649 type
- CURSEG_HOT_NODE
);
1651 blk_addr
= GET_SUM_BLOCK(sbi
, segno
);
1654 new = get_meta_page(sbi
, blk_addr
);
1655 sum
= (struct f2fs_summary_block
*)page_address(new);
1657 if (IS_NODESEG(type
)) {
1658 if (__exist_node_summaries(sbi
)) {
1659 struct f2fs_summary
*ns
= &sum
->entries
[0];
1661 for (i
= 0; i
< sbi
->blocks_per_seg
; i
++, ns
++) {
1663 ns
->ofs_in_node
= 0;
1668 err
= restore_node_summary(sbi
, segno
, sum
);
1670 f2fs_put_page(new, 1);
1676 /* set uncompleted segment to curseg */
1677 curseg
= CURSEG_I(sbi
, type
);
1678 mutex_lock(&curseg
->curseg_mutex
);
1680 /* update journal info */
1681 down_write(&curseg
->journal_rwsem
);
1682 memcpy(curseg
->journal
, &sum
->journal
, SUM_JOURNAL_SIZE
);
1683 up_write(&curseg
->journal_rwsem
);
1685 memcpy(curseg
->sum_blk
->entries
, sum
->entries
, SUM_ENTRY_SIZE
);
1686 memcpy(&curseg
->sum_blk
->footer
, &sum
->footer
, SUM_FOOTER_SIZE
);
1687 curseg
->next_segno
= segno
;
1688 reset_curseg(sbi
, type
, 0);
1689 curseg
->alloc_type
= ckpt
->alloc_type
[type
];
1690 curseg
->next_blkoff
= blk_off
;
1691 mutex_unlock(&curseg
->curseg_mutex
);
1692 f2fs_put_page(new, 1);
1696 static int restore_curseg_summaries(struct f2fs_sb_info
*sbi
)
1698 int type
= CURSEG_HOT_DATA
;
1701 if (is_set_ckpt_flags(F2FS_CKPT(sbi
), CP_COMPACT_SUM_FLAG
)) {
1702 int npages
= npages_for_summary_flush(sbi
, true);
1705 ra_meta_pages(sbi
, start_sum_block(sbi
), npages
,
1708 /* restore for compacted data summary */
1709 if (read_compacted_summaries(sbi
))
1711 type
= CURSEG_HOT_NODE
;
1714 if (__exist_node_summaries(sbi
))
1715 ra_meta_pages(sbi
, sum_blk_addr(sbi
, NR_CURSEG_TYPE
, type
),
1716 NR_CURSEG_TYPE
- type
, META_CP
, true);
1718 for (; type
<= CURSEG_COLD_NODE
; type
++) {
1719 err
= read_normal_summaries(sbi
, type
);
1727 static void write_compacted_summaries(struct f2fs_sb_info
*sbi
, block_t blkaddr
)
1730 unsigned char *kaddr
;
1731 struct f2fs_summary
*summary
;
1732 struct curseg_info
*seg_i
;
1733 int written_size
= 0;
1736 page
= grab_meta_page(sbi
, blkaddr
++);
1737 kaddr
= (unsigned char *)page_address(page
);
1739 /* Step 1: write nat cache */
1740 seg_i
= CURSEG_I(sbi
, CURSEG_HOT_DATA
);
1741 memcpy(kaddr
, seg_i
->journal
, SUM_JOURNAL_SIZE
);
1742 written_size
+= SUM_JOURNAL_SIZE
;
1744 /* Step 2: write sit cache */
1745 seg_i
= CURSEG_I(sbi
, CURSEG_COLD_DATA
);
1746 memcpy(kaddr
+ written_size
, seg_i
->journal
, SUM_JOURNAL_SIZE
);
1747 written_size
+= SUM_JOURNAL_SIZE
;
1749 /* Step 3: write summary entries */
1750 for (i
= CURSEG_HOT_DATA
; i
<= CURSEG_COLD_DATA
; i
++) {
1751 unsigned short blkoff
;
1752 seg_i
= CURSEG_I(sbi
, i
);
1753 if (sbi
->ckpt
->alloc_type
[i
] == SSR
)
1754 blkoff
= sbi
->blocks_per_seg
;
1756 blkoff
= curseg_blkoff(sbi
, i
);
1758 for (j
= 0; j
< blkoff
; j
++) {
1760 page
= grab_meta_page(sbi
, blkaddr
++);
1761 kaddr
= (unsigned char *)page_address(page
);
1764 summary
= (struct f2fs_summary
*)(kaddr
+ written_size
);
1765 *summary
= seg_i
->sum_blk
->entries
[j
];
1766 written_size
+= SUMMARY_SIZE
;
1768 if (written_size
+ SUMMARY_SIZE
<= PAGE_SIZE
-
1772 set_page_dirty(page
);
1773 f2fs_put_page(page
, 1);
1778 set_page_dirty(page
);
1779 f2fs_put_page(page
, 1);
1783 static void write_normal_summaries(struct f2fs_sb_info
*sbi
,
1784 block_t blkaddr
, int type
)
1787 if (IS_DATASEG(type
))
1788 end
= type
+ NR_CURSEG_DATA_TYPE
;
1790 end
= type
+ NR_CURSEG_NODE_TYPE
;
1792 for (i
= type
; i
< end
; i
++)
1793 write_current_sum_page(sbi
, i
, blkaddr
+ (i
- type
));
1796 void write_data_summaries(struct f2fs_sb_info
*sbi
, block_t start_blk
)
1798 if (is_set_ckpt_flags(F2FS_CKPT(sbi
), CP_COMPACT_SUM_FLAG
))
1799 write_compacted_summaries(sbi
, start_blk
);
1801 write_normal_summaries(sbi
, start_blk
, CURSEG_HOT_DATA
);
1804 void write_node_summaries(struct f2fs_sb_info
*sbi
, block_t start_blk
)
1806 write_normal_summaries(sbi
, start_blk
, CURSEG_HOT_NODE
);
1809 int lookup_journal_in_cursum(struct f2fs_journal
*journal
, int type
,
1810 unsigned int val
, int alloc
)
1814 if (type
== NAT_JOURNAL
) {
1815 for (i
= 0; i
< nats_in_cursum(journal
); i
++) {
1816 if (le32_to_cpu(nid_in_journal(journal
, i
)) == val
)
1819 if (alloc
&& __has_cursum_space(journal
, 1, NAT_JOURNAL
))
1820 return update_nats_in_cursum(journal
, 1);
1821 } else if (type
== SIT_JOURNAL
) {
1822 for (i
= 0; i
< sits_in_cursum(journal
); i
++)
1823 if (le32_to_cpu(segno_in_journal(journal
, i
)) == val
)
1825 if (alloc
&& __has_cursum_space(journal
, 1, SIT_JOURNAL
))
1826 return update_sits_in_cursum(journal
, 1);
1831 static struct page
*get_current_sit_page(struct f2fs_sb_info
*sbi
,
1834 return get_meta_page(sbi
, current_sit_addr(sbi
, segno
));
1837 static struct page
*get_next_sit_page(struct f2fs_sb_info
*sbi
,
1840 struct sit_info
*sit_i
= SIT_I(sbi
);
1841 struct page
*src_page
, *dst_page
;
1842 pgoff_t src_off
, dst_off
;
1843 void *src_addr
, *dst_addr
;
1845 src_off
= current_sit_addr(sbi
, start
);
1846 dst_off
= next_sit_addr(sbi
, src_off
);
1848 /* get current sit block page without lock */
1849 src_page
= get_meta_page(sbi
, src_off
);
1850 dst_page
= grab_meta_page(sbi
, dst_off
);
1851 f2fs_bug_on(sbi
, PageDirty(src_page
));
1853 src_addr
= page_address(src_page
);
1854 dst_addr
= page_address(dst_page
);
1855 memcpy(dst_addr
, src_addr
, PAGE_SIZE
);
1857 set_page_dirty(dst_page
);
1858 f2fs_put_page(src_page
, 1);
1860 set_to_next_sit(sit_i
, start
);
1865 static struct sit_entry_set
*grab_sit_entry_set(void)
1867 struct sit_entry_set
*ses
=
1868 f2fs_kmem_cache_alloc(sit_entry_set_slab
, GFP_NOFS
);
1871 INIT_LIST_HEAD(&ses
->set_list
);
1875 static void release_sit_entry_set(struct sit_entry_set
*ses
)
1877 list_del(&ses
->set_list
);
1878 kmem_cache_free(sit_entry_set_slab
, ses
);
1881 static void adjust_sit_entry_set(struct sit_entry_set
*ses
,
1882 struct list_head
*head
)
1884 struct sit_entry_set
*next
= ses
;
1886 if (list_is_last(&ses
->set_list
, head
))
1889 list_for_each_entry_continue(next
, head
, set_list
)
1890 if (ses
->entry_cnt
<= next
->entry_cnt
)
1893 list_move_tail(&ses
->set_list
, &next
->set_list
);
1896 static void add_sit_entry(unsigned int segno
, struct list_head
*head
)
1898 struct sit_entry_set
*ses
;
1899 unsigned int start_segno
= START_SEGNO(segno
);
1901 list_for_each_entry(ses
, head
, set_list
) {
1902 if (ses
->start_segno
== start_segno
) {
1904 adjust_sit_entry_set(ses
, head
);
1909 ses
= grab_sit_entry_set();
1911 ses
->start_segno
= start_segno
;
1913 list_add(&ses
->set_list
, head
);
1916 static void add_sits_in_set(struct f2fs_sb_info
*sbi
)
1918 struct f2fs_sm_info
*sm_info
= SM_I(sbi
);
1919 struct list_head
*set_list
= &sm_info
->sit_entry_set
;
1920 unsigned long *bitmap
= SIT_I(sbi
)->dirty_sentries_bitmap
;
1923 for_each_set_bit(segno
, bitmap
, MAIN_SEGS(sbi
))
1924 add_sit_entry(segno
, set_list
);
1927 static void remove_sits_in_journal(struct f2fs_sb_info
*sbi
)
1929 struct curseg_info
*curseg
= CURSEG_I(sbi
, CURSEG_COLD_DATA
);
1930 struct f2fs_journal
*journal
= curseg
->journal
;
1933 down_write(&curseg
->journal_rwsem
);
1934 for (i
= 0; i
< sits_in_cursum(journal
); i
++) {
1938 segno
= le32_to_cpu(segno_in_journal(journal
, i
));
1939 dirtied
= __mark_sit_entry_dirty(sbi
, segno
);
1942 add_sit_entry(segno
, &SM_I(sbi
)->sit_entry_set
);
1944 update_sits_in_cursum(journal
, -i
);
1945 up_write(&curseg
->journal_rwsem
);
1949 * CP calls this function, which flushes SIT entries including sit_journal,
1950 * and moves prefree segs to free segs.
1952 void flush_sit_entries(struct f2fs_sb_info
*sbi
, struct cp_control
*cpc
)
1954 struct sit_info
*sit_i
= SIT_I(sbi
);
1955 unsigned long *bitmap
= sit_i
->dirty_sentries_bitmap
;
1956 struct curseg_info
*curseg
= CURSEG_I(sbi
, CURSEG_COLD_DATA
);
1957 struct f2fs_journal
*journal
= curseg
->journal
;
1958 struct sit_entry_set
*ses
, *tmp
;
1959 struct list_head
*head
= &SM_I(sbi
)->sit_entry_set
;
1960 bool to_journal
= true;
1961 struct seg_entry
*se
;
1963 mutex_lock(&sit_i
->sentry_lock
);
1965 if (!sit_i
->dirty_sentries
)
1969 * add and account sit entries of dirty bitmap in sit entry
1972 add_sits_in_set(sbi
);
1975 * if there are no enough space in journal to store dirty sit
1976 * entries, remove all entries from journal and add and account
1977 * them in sit entry set.
1979 if (!__has_cursum_space(journal
, sit_i
->dirty_sentries
, SIT_JOURNAL
))
1980 remove_sits_in_journal(sbi
);
1983 * there are two steps to flush sit entries:
1984 * #1, flush sit entries to journal in current cold data summary block.
1985 * #2, flush sit entries to sit page.
1987 list_for_each_entry_safe(ses
, tmp
, head
, set_list
) {
1988 struct page
*page
= NULL
;
1989 struct f2fs_sit_block
*raw_sit
= NULL
;
1990 unsigned int start_segno
= ses
->start_segno
;
1991 unsigned int end
= min(start_segno
+ SIT_ENTRY_PER_BLOCK
,
1992 (unsigned long)MAIN_SEGS(sbi
));
1993 unsigned int segno
= start_segno
;
1996 !__has_cursum_space(journal
, ses
->entry_cnt
, SIT_JOURNAL
))
2000 down_write(&curseg
->journal_rwsem
);
2002 page
= get_next_sit_page(sbi
, start_segno
);
2003 raw_sit
= page_address(page
);
2006 /* flush dirty sit entries in region of current sit set */
2007 for_each_set_bit_from(segno
, bitmap
, end
) {
2008 int offset
, sit_offset
;
2010 se
= get_seg_entry(sbi
, segno
);
2012 /* add discard candidates */
2013 if (cpc
->reason
!= CP_DISCARD
) {
2014 cpc
->trim_start
= segno
;
2015 add_discard_addrs(sbi
, cpc
);
2019 offset
= lookup_journal_in_cursum(journal
,
2020 SIT_JOURNAL
, segno
, 1);
2021 f2fs_bug_on(sbi
, offset
< 0);
2022 segno_in_journal(journal
, offset
) =
2024 seg_info_to_raw_sit(se
,
2025 &sit_in_journal(journal
, offset
));
2027 sit_offset
= SIT_ENTRY_OFFSET(sit_i
, segno
);
2028 seg_info_to_raw_sit(se
,
2029 &raw_sit
->entries
[sit_offset
]);
2032 __clear_bit(segno
, bitmap
);
2033 sit_i
->dirty_sentries
--;
2038 up_write(&curseg
->journal_rwsem
);
2040 f2fs_put_page(page
, 1);
2042 f2fs_bug_on(sbi
, ses
->entry_cnt
);
2043 release_sit_entry_set(ses
);
2046 f2fs_bug_on(sbi
, !list_empty(head
));
2047 f2fs_bug_on(sbi
, sit_i
->dirty_sentries
);
2049 if (cpc
->reason
== CP_DISCARD
) {
2050 for (; cpc
->trim_start
<= cpc
->trim_end
; cpc
->trim_start
++)
2051 add_discard_addrs(sbi
, cpc
);
2053 mutex_unlock(&sit_i
->sentry_lock
);
2055 set_prefree_as_free_segments(sbi
);
2058 static int build_sit_info(struct f2fs_sb_info
*sbi
)
2060 struct f2fs_super_block
*raw_super
= F2FS_RAW_SUPER(sbi
);
2061 struct f2fs_checkpoint
*ckpt
= F2FS_CKPT(sbi
);
2062 struct sit_info
*sit_i
;
2063 unsigned int sit_segs
, start
;
2064 char *src_bitmap
, *dst_bitmap
;
2065 unsigned int bitmap_size
;
2067 /* allocate memory for SIT information */
2068 sit_i
= kzalloc(sizeof(struct sit_info
), GFP_KERNEL
);
2072 SM_I(sbi
)->sit_info
= sit_i
;
2074 sit_i
->sentries
= f2fs_kvzalloc(MAIN_SEGS(sbi
) *
2075 sizeof(struct seg_entry
), GFP_KERNEL
);
2076 if (!sit_i
->sentries
)
2079 bitmap_size
= f2fs_bitmap_size(MAIN_SEGS(sbi
));
2080 sit_i
->dirty_sentries_bitmap
= f2fs_kvzalloc(bitmap_size
, GFP_KERNEL
);
2081 if (!sit_i
->dirty_sentries_bitmap
)
2084 for (start
= 0; start
< MAIN_SEGS(sbi
); start
++) {
2085 sit_i
->sentries
[start
].cur_valid_map
2086 = kzalloc(SIT_VBLOCK_MAP_SIZE
, GFP_KERNEL
);
2087 sit_i
->sentries
[start
].ckpt_valid_map
2088 = kzalloc(SIT_VBLOCK_MAP_SIZE
, GFP_KERNEL
);
2089 sit_i
->sentries
[start
].discard_map
2090 = kzalloc(SIT_VBLOCK_MAP_SIZE
, GFP_KERNEL
);
2091 if (!sit_i
->sentries
[start
].cur_valid_map
||
2092 !sit_i
->sentries
[start
].ckpt_valid_map
||
2093 !sit_i
->sentries
[start
].discard_map
)
2097 sit_i
->tmp_map
= kzalloc(SIT_VBLOCK_MAP_SIZE
, GFP_KERNEL
);
2098 if (!sit_i
->tmp_map
)
2101 if (sbi
->segs_per_sec
> 1) {
2102 sit_i
->sec_entries
= f2fs_kvzalloc(MAIN_SECS(sbi
) *
2103 sizeof(struct sec_entry
), GFP_KERNEL
);
2104 if (!sit_i
->sec_entries
)
2108 /* get information related with SIT */
2109 sit_segs
= le32_to_cpu(raw_super
->segment_count_sit
) >> 1;
2111 /* setup SIT bitmap from ckeckpoint pack */
2112 bitmap_size
= __bitmap_size(sbi
, SIT_BITMAP
);
2113 src_bitmap
= __bitmap_ptr(sbi
, SIT_BITMAP
);
2115 dst_bitmap
= kmemdup(src_bitmap
, bitmap_size
, GFP_KERNEL
);
2119 /* init SIT information */
2120 sit_i
->s_ops
= &default_salloc_ops
;
2122 sit_i
->sit_base_addr
= le32_to_cpu(raw_super
->sit_blkaddr
);
2123 sit_i
->sit_blocks
= sit_segs
<< sbi
->log_blocks_per_seg
;
2124 sit_i
->written_valid_blocks
= le64_to_cpu(ckpt
->valid_block_count
);
2125 sit_i
->sit_bitmap
= dst_bitmap
;
2126 sit_i
->bitmap_size
= bitmap_size
;
2127 sit_i
->dirty_sentries
= 0;
2128 sit_i
->sents_per_block
= SIT_ENTRY_PER_BLOCK
;
2129 sit_i
->elapsed_time
= le64_to_cpu(sbi
->ckpt
->elapsed_time
);
2130 sit_i
->mounted_time
= CURRENT_TIME_SEC
.tv_sec
;
2131 mutex_init(&sit_i
->sentry_lock
);
2135 static int build_free_segmap(struct f2fs_sb_info
*sbi
)
2137 struct free_segmap_info
*free_i
;
2138 unsigned int bitmap_size
, sec_bitmap_size
;
2140 /* allocate memory for free segmap information */
2141 free_i
= kzalloc(sizeof(struct free_segmap_info
), GFP_KERNEL
);
2145 SM_I(sbi
)->free_info
= free_i
;
2147 bitmap_size
= f2fs_bitmap_size(MAIN_SEGS(sbi
));
2148 free_i
->free_segmap
= f2fs_kvmalloc(bitmap_size
, GFP_KERNEL
);
2149 if (!free_i
->free_segmap
)
2152 sec_bitmap_size
= f2fs_bitmap_size(MAIN_SECS(sbi
));
2153 free_i
->free_secmap
= f2fs_kvmalloc(sec_bitmap_size
, GFP_KERNEL
);
2154 if (!free_i
->free_secmap
)
2157 /* set all segments as dirty temporarily */
2158 memset(free_i
->free_segmap
, 0xff, bitmap_size
);
2159 memset(free_i
->free_secmap
, 0xff, sec_bitmap_size
);
2161 /* init free segmap information */
2162 free_i
->start_segno
= GET_SEGNO_FROM_SEG0(sbi
, MAIN_BLKADDR(sbi
));
2163 free_i
->free_segments
= 0;
2164 free_i
->free_sections
= 0;
2165 spin_lock_init(&free_i
->segmap_lock
);
2169 static int build_curseg(struct f2fs_sb_info
*sbi
)
2171 struct curseg_info
*array
;
2174 array
= kcalloc(NR_CURSEG_TYPE
, sizeof(*array
), GFP_KERNEL
);
2178 SM_I(sbi
)->curseg_array
= array
;
2180 for (i
= 0; i
< NR_CURSEG_TYPE
; i
++) {
2181 mutex_init(&array
[i
].curseg_mutex
);
2182 array
[i
].sum_blk
= kzalloc(PAGE_SIZE
, GFP_KERNEL
);
2183 if (!array
[i
].sum_blk
)
2185 init_rwsem(&array
[i
].journal_rwsem
);
2186 array
[i
].journal
= kzalloc(sizeof(struct f2fs_journal
),
2188 if (!array
[i
].journal
)
2190 array
[i
].segno
= NULL_SEGNO
;
2191 array
[i
].next_blkoff
= 0;
2193 return restore_curseg_summaries(sbi
);
2196 static void build_sit_entries(struct f2fs_sb_info
*sbi
)
2198 struct sit_info
*sit_i
= SIT_I(sbi
);
2199 struct curseg_info
*curseg
= CURSEG_I(sbi
, CURSEG_COLD_DATA
);
2200 struct f2fs_journal
*journal
= curseg
->journal
;
2201 int sit_blk_cnt
= SIT_BLK_CNT(sbi
);
2202 unsigned int i
, start
, end
;
2203 unsigned int readed
, start_blk
= 0;
2204 int nrpages
= MAX_BIO_BLOCKS(sbi
) * 8;
2207 readed
= ra_meta_pages(sbi
, start_blk
, nrpages
, META_SIT
, true);
2209 start
= start_blk
* sit_i
->sents_per_block
;
2210 end
= (start_blk
+ readed
) * sit_i
->sents_per_block
;
2212 for (; start
< end
&& start
< MAIN_SEGS(sbi
); start
++) {
2213 struct seg_entry
*se
= &sit_i
->sentries
[start
];
2214 struct f2fs_sit_block
*sit_blk
;
2215 struct f2fs_sit_entry sit
;
2218 down_read(&curseg
->journal_rwsem
);
2219 for (i
= 0; i
< sits_in_cursum(journal
); i
++) {
2220 if (le32_to_cpu(segno_in_journal(journal
, i
))
2222 sit
= sit_in_journal(journal
, i
);
2223 up_read(&curseg
->journal_rwsem
);
2227 up_read(&curseg
->journal_rwsem
);
2229 page
= get_current_sit_page(sbi
, start
);
2230 sit_blk
= (struct f2fs_sit_block
*)page_address(page
);
2231 sit
= sit_blk
->entries
[SIT_ENTRY_OFFSET(sit_i
, start
)];
2232 f2fs_put_page(page
, 1);
2234 check_block_count(sbi
, start
, &sit
);
2235 seg_info_from_raw_sit(se
, &sit
);
2237 /* build discard map only one time */
2238 memcpy(se
->discard_map
, se
->cur_valid_map
, SIT_VBLOCK_MAP_SIZE
);
2239 sbi
->discard_blks
+= sbi
->blocks_per_seg
- se
->valid_blocks
;
2241 if (sbi
->segs_per_sec
> 1) {
2242 struct sec_entry
*e
= get_sec_entry(sbi
, start
);
2243 e
->valid_blocks
+= se
->valid_blocks
;
2246 start_blk
+= readed
;
2247 } while (start_blk
< sit_blk_cnt
);
2250 static void init_free_segmap(struct f2fs_sb_info
*sbi
)
2255 for (start
= 0; start
< MAIN_SEGS(sbi
); start
++) {
2256 struct seg_entry
*sentry
= get_seg_entry(sbi
, start
);
2257 if (!sentry
->valid_blocks
)
2258 __set_free(sbi
, start
);
2261 /* set use the current segments */
2262 for (type
= CURSEG_HOT_DATA
; type
<= CURSEG_COLD_NODE
; type
++) {
2263 struct curseg_info
*curseg_t
= CURSEG_I(sbi
, type
);
2264 __set_test_and_inuse(sbi
, curseg_t
->segno
);
2268 static void init_dirty_segmap(struct f2fs_sb_info
*sbi
)
2270 struct dirty_seglist_info
*dirty_i
= DIRTY_I(sbi
);
2271 struct free_segmap_info
*free_i
= FREE_I(sbi
);
2272 unsigned int segno
= 0, offset
= 0;
2273 unsigned short valid_blocks
;
2276 /* find dirty segment based on free segmap */
2277 segno
= find_next_inuse(free_i
, MAIN_SEGS(sbi
), offset
);
2278 if (segno
>= MAIN_SEGS(sbi
))
2281 valid_blocks
= get_valid_blocks(sbi
, segno
, 0);
2282 if (valid_blocks
== sbi
->blocks_per_seg
|| !valid_blocks
)
2284 if (valid_blocks
> sbi
->blocks_per_seg
) {
2285 f2fs_bug_on(sbi
, 1);
2288 mutex_lock(&dirty_i
->seglist_lock
);
2289 __locate_dirty_segment(sbi
, segno
, DIRTY
);
2290 mutex_unlock(&dirty_i
->seglist_lock
);
2294 static int init_victim_secmap(struct f2fs_sb_info
*sbi
)
2296 struct dirty_seglist_info
*dirty_i
= DIRTY_I(sbi
);
2297 unsigned int bitmap_size
= f2fs_bitmap_size(MAIN_SECS(sbi
));
2299 dirty_i
->victim_secmap
= f2fs_kvzalloc(bitmap_size
, GFP_KERNEL
);
2300 if (!dirty_i
->victim_secmap
)
2305 static int build_dirty_segmap(struct f2fs_sb_info
*sbi
)
2307 struct dirty_seglist_info
*dirty_i
;
2308 unsigned int bitmap_size
, i
;
2310 /* allocate memory for dirty segments list information */
2311 dirty_i
= kzalloc(sizeof(struct dirty_seglist_info
), GFP_KERNEL
);
2315 SM_I(sbi
)->dirty_info
= dirty_i
;
2316 mutex_init(&dirty_i
->seglist_lock
);
2318 bitmap_size
= f2fs_bitmap_size(MAIN_SEGS(sbi
));
2320 for (i
= 0; i
< NR_DIRTY_TYPE
; i
++) {
2321 dirty_i
->dirty_segmap
[i
] = f2fs_kvzalloc(bitmap_size
, GFP_KERNEL
);
2322 if (!dirty_i
->dirty_segmap
[i
])
2326 init_dirty_segmap(sbi
);
2327 return init_victim_secmap(sbi
);
2331 * Update min, max modified time for cost-benefit GC algorithm
2333 static void init_min_max_mtime(struct f2fs_sb_info
*sbi
)
2335 struct sit_info
*sit_i
= SIT_I(sbi
);
2338 mutex_lock(&sit_i
->sentry_lock
);
2340 sit_i
->min_mtime
= LLONG_MAX
;
2342 for (segno
= 0; segno
< MAIN_SEGS(sbi
); segno
+= sbi
->segs_per_sec
) {
2344 unsigned long long mtime
= 0;
2346 for (i
= 0; i
< sbi
->segs_per_sec
; i
++)
2347 mtime
+= get_seg_entry(sbi
, segno
+ i
)->mtime
;
2349 mtime
= div_u64(mtime
, sbi
->segs_per_sec
);
2351 if (sit_i
->min_mtime
> mtime
)
2352 sit_i
->min_mtime
= mtime
;
2354 sit_i
->max_mtime
= get_mtime(sbi
);
2355 mutex_unlock(&sit_i
->sentry_lock
);
2358 int build_segment_manager(struct f2fs_sb_info
*sbi
)
2360 struct f2fs_super_block
*raw_super
= F2FS_RAW_SUPER(sbi
);
2361 struct f2fs_checkpoint
*ckpt
= F2FS_CKPT(sbi
);
2362 struct f2fs_sm_info
*sm_info
;
2365 sm_info
= kzalloc(sizeof(struct f2fs_sm_info
), GFP_KERNEL
);
2370 sbi
->sm_info
= sm_info
;
2371 sm_info
->seg0_blkaddr
= le32_to_cpu(raw_super
->segment0_blkaddr
);
2372 sm_info
->main_blkaddr
= le32_to_cpu(raw_super
->main_blkaddr
);
2373 sm_info
->segment_count
= le32_to_cpu(raw_super
->segment_count
);
2374 sm_info
->reserved_segments
= le32_to_cpu(ckpt
->rsvd_segment_count
);
2375 sm_info
->ovp_segments
= le32_to_cpu(ckpt
->overprov_segment_count
);
2376 sm_info
->main_segments
= le32_to_cpu(raw_super
->segment_count_main
);
2377 sm_info
->ssa_blkaddr
= le32_to_cpu(raw_super
->ssa_blkaddr
);
2378 sm_info
->rec_prefree_segments
= sm_info
->main_segments
*
2379 DEF_RECLAIM_PREFREE_SEGMENTS
/ 100;
2380 sm_info
->ipu_policy
= 1 << F2FS_IPU_FSYNC
;
2381 sm_info
->min_ipu_util
= DEF_MIN_IPU_UTIL
;
2382 sm_info
->min_fsync_blocks
= DEF_MIN_FSYNC_BLOCKS
;
2384 INIT_LIST_HEAD(&sm_info
->discard_list
);
2385 sm_info
->nr_discards
= 0;
2386 sm_info
->max_discards
= 0;
2388 sm_info
->trim_sections
= DEF_BATCHED_TRIM_SECTIONS
;
2390 INIT_LIST_HEAD(&sm_info
->sit_entry_set
);
2392 if (test_opt(sbi
, FLUSH_MERGE
) && !f2fs_readonly(sbi
->sb
)) {
2393 err
= create_flush_cmd_control(sbi
);
2398 err
= build_sit_info(sbi
);
2401 err
= build_free_segmap(sbi
);
2404 err
= build_curseg(sbi
);
2408 /* reinit free segmap based on SIT */
2409 build_sit_entries(sbi
);
2411 init_free_segmap(sbi
);
2412 err
= build_dirty_segmap(sbi
);
2416 init_min_max_mtime(sbi
);
2420 static void discard_dirty_segmap(struct f2fs_sb_info
*sbi
,
2421 enum dirty_type dirty_type
)
2423 struct dirty_seglist_info
*dirty_i
= DIRTY_I(sbi
);
2425 mutex_lock(&dirty_i
->seglist_lock
);
2426 kvfree(dirty_i
->dirty_segmap
[dirty_type
]);
2427 dirty_i
->nr_dirty
[dirty_type
] = 0;
2428 mutex_unlock(&dirty_i
->seglist_lock
);
2431 static void destroy_victim_secmap(struct f2fs_sb_info
*sbi
)
2433 struct dirty_seglist_info
*dirty_i
= DIRTY_I(sbi
);
2434 kvfree(dirty_i
->victim_secmap
);
2437 static void destroy_dirty_segmap(struct f2fs_sb_info
*sbi
)
2439 struct dirty_seglist_info
*dirty_i
= DIRTY_I(sbi
);
2445 /* discard pre-free/dirty segments list */
2446 for (i
= 0; i
< NR_DIRTY_TYPE
; i
++)
2447 discard_dirty_segmap(sbi
, i
);
2449 destroy_victim_secmap(sbi
);
2450 SM_I(sbi
)->dirty_info
= NULL
;
2454 static void destroy_curseg(struct f2fs_sb_info
*sbi
)
2456 struct curseg_info
*array
= SM_I(sbi
)->curseg_array
;
2461 SM_I(sbi
)->curseg_array
= NULL
;
2462 for (i
= 0; i
< NR_CURSEG_TYPE
; i
++) {
2463 kfree(array
[i
].sum_blk
);
2464 kfree(array
[i
].journal
);
2469 static void destroy_free_segmap(struct f2fs_sb_info
*sbi
)
2471 struct free_segmap_info
*free_i
= SM_I(sbi
)->free_info
;
2474 SM_I(sbi
)->free_info
= NULL
;
2475 kvfree(free_i
->free_segmap
);
2476 kvfree(free_i
->free_secmap
);
2480 static void destroy_sit_info(struct f2fs_sb_info
*sbi
)
2482 struct sit_info
*sit_i
= SIT_I(sbi
);
2488 if (sit_i
->sentries
) {
2489 for (start
= 0; start
< MAIN_SEGS(sbi
); start
++) {
2490 kfree(sit_i
->sentries
[start
].cur_valid_map
);
2491 kfree(sit_i
->sentries
[start
].ckpt_valid_map
);
2492 kfree(sit_i
->sentries
[start
].discard_map
);
2495 kfree(sit_i
->tmp_map
);
2497 kvfree(sit_i
->sentries
);
2498 kvfree(sit_i
->sec_entries
);
2499 kvfree(sit_i
->dirty_sentries_bitmap
);
2501 SM_I(sbi
)->sit_info
= NULL
;
2502 kfree(sit_i
->sit_bitmap
);
2506 void destroy_segment_manager(struct f2fs_sb_info
*sbi
)
2508 struct f2fs_sm_info
*sm_info
= SM_I(sbi
);
2512 destroy_flush_cmd_control(sbi
);
2513 destroy_dirty_segmap(sbi
);
2514 destroy_curseg(sbi
);
2515 destroy_free_segmap(sbi
);
2516 destroy_sit_info(sbi
);
2517 sbi
->sm_info
= NULL
;
2521 int __init
create_segment_manager_caches(void)
2523 discard_entry_slab
= f2fs_kmem_cache_create("discard_entry",
2524 sizeof(struct discard_entry
));
2525 if (!discard_entry_slab
)
2528 sit_entry_set_slab
= f2fs_kmem_cache_create("sit_entry_set",
2529 sizeof(struct sit_entry_set
));
2530 if (!sit_entry_set_slab
)
2531 goto destory_discard_entry
;
2533 inmem_entry_slab
= f2fs_kmem_cache_create("inmem_page_entry",
2534 sizeof(struct inmem_pages
));
2535 if (!inmem_entry_slab
)
2536 goto destroy_sit_entry_set
;
2539 destroy_sit_entry_set
:
2540 kmem_cache_destroy(sit_entry_set_slab
);
2541 destory_discard_entry
:
2542 kmem_cache_destroy(discard_entry_slab
);
2547 void destroy_segment_manager_caches(void)
2549 kmem_cache_destroy(sit_entry_set_slab
);
2550 kmem_cache_destroy(discard_entry_slab
);
2551 kmem_cache_destroy(inmem_entry_slab
);