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 f2fs_put_page(page
, 1);
232 list_del(&cur
->list
);
233 kmem_cache_free(inmem_entry_slab
, cur
);
234 dec_page_count(F2FS_I_SB(inode
), F2FS_INMEM_PAGES
);
239 void drop_inmem_pages(struct inode
*inode
)
241 struct f2fs_inode_info
*fi
= F2FS_I(inode
);
243 clear_inode_flag(F2FS_I(inode
), FI_ATOMIC_FILE
);
245 mutex_lock(&fi
->inmem_lock
);
246 __revoke_inmem_pages(inode
, &fi
->inmem_pages
, true, false);
247 mutex_unlock(&fi
->inmem_lock
);
250 static int __commit_inmem_pages(struct inode
*inode
,
251 struct list_head
*revoke_list
)
253 struct f2fs_sb_info
*sbi
= F2FS_I_SB(inode
);
254 struct f2fs_inode_info
*fi
= F2FS_I(inode
);
255 struct inmem_pages
*cur
, *tmp
;
256 struct f2fs_io_info fio
= {
259 .rw
= WRITE_SYNC
| REQ_PRIO
,
260 .encrypted_page
= NULL
,
262 bool submit_bio
= false;
265 list_for_each_entry_safe(cur
, tmp
, &fi
->inmem_pages
, list
) {
266 struct page
*page
= cur
->page
;
269 if (page
->mapping
== inode
->i_mapping
) {
270 trace_f2fs_commit_inmem_page(page
, INMEM
);
272 set_page_dirty(page
);
273 f2fs_wait_on_page_writeback(page
, DATA
, true);
274 if (clear_page_dirty_for_io(page
))
275 inode_dec_dirty_pages(inode
);
278 err
= do_write_data_page(&fio
);
284 /* record old blkaddr for revoking */
285 cur
->old_addr
= fio
.old_blkaddr
;
287 clear_cold_data(page
);
291 list_move_tail(&cur
->list
, revoke_list
);
295 f2fs_submit_merged_bio_cond(sbi
, inode
, NULL
, 0, DATA
, WRITE
);
298 __revoke_inmem_pages(inode
, revoke_list
, false, false);
303 int commit_inmem_pages(struct inode
*inode
)
305 struct f2fs_sb_info
*sbi
= F2FS_I_SB(inode
);
306 struct f2fs_inode_info
*fi
= F2FS_I(inode
);
307 struct list_head revoke_list
;
310 INIT_LIST_HEAD(&revoke_list
);
311 f2fs_balance_fs(sbi
, true);
314 mutex_lock(&fi
->inmem_lock
);
315 err
= __commit_inmem_pages(inode
, &revoke_list
);
319 * try to revoke all committed pages, but still we could fail
320 * due to no memory or other reason, if that happened, EAGAIN
321 * will be returned, which means in such case, transaction is
322 * already not integrity, caller should use journal to do the
323 * recovery or rewrite & commit last transaction. For other
324 * error number, revoking was done by filesystem itself.
326 ret
= __revoke_inmem_pages(inode
, &revoke_list
, false, true);
330 /* drop all uncommitted pages */
331 __revoke_inmem_pages(inode
, &fi
->inmem_pages
, true, false);
333 mutex_unlock(&fi
->inmem_lock
);
340 * This function balances dirty node and dentry pages.
341 * In addition, it controls garbage collection.
343 void f2fs_balance_fs(struct f2fs_sb_info
*sbi
, bool need
)
348 * We should do GC or end up with checkpoint, if there are so many dirty
349 * dir/node pages without enough free segments.
351 if (has_not_enough_free_secs(sbi
, 0)) {
352 mutex_lock(&sbi
->gc_mutex
);
357 void f2fs_balance_fs_bg(struct f2fs_sb_info
*sbi
)
359 /* try to shrink extent cache when there is no enough memory */
360 if (!available_free_memory(sbi
, EXTENT_CACHE
))
361 f2fs_shrink_extent_tree(sbi
, EXTENT_CACHE_SHRINK_NUMBER
);
363 /* check the # of cached NAT entries */
364 if (!available_free_memory(sbi
, NAT_ENTRIES
))
365 try_to_free_nats(sbi
, NAT_ENTRY_PER_BLOCK
);
367 if (!available_free_memory(sbi
, FREE_NIDS
))
368 try_to_free_nids(sbi
, NAT_ENTRY_PER_BLOCK
* FREE_NID_PAGES
);
370 /* checkpoint is the only way to shrink partial cached entries */
371 if (!available_free_memory(sbi
, NAT_ENTRIES
) ||
372 !available_free_memory(sbi
, INO_ENTRIES
) ||
373 excess_prefree_segs(sbi
) ||
374 excess_dirty_nats(sbi
) ||
375 (is_idle(sbi
) && f2fs_time_over(sbi
, CP_TIME
))) {
376 if (test_opt(sbi
, DATA_FLUSH
)) {
377 struct blk_plug plug
;
379 blk_start_plug(&plug
);
380 sync_dirty_inodes(sbi
, FILE_INODE
);
381 blk_finish_plug(&plug
);
383 f2fs_sync_fs(sbi
->sb
, true);
384 stat_inc_bg_cp_count(sbi
->stat_info
);
388 static int issue_flush_thread(void *data
)
390 struct f2fs_sb_info
*sbi
= data
;
391 struct flush_cmd_control
*fcc
= SM_I(sbi
)->cmd_control_info
;
392 wait_queue_head_t
*q
= &fcc
->flush_wait_queue
;
394 if (kthread_should_stop())
397 if (!llist_empty(&fcc
->issue_list
)) {
399 struct flush_cmd
*cmd
, *next
;
402 bio
= f2fs_bio_alloc(0);
404 fcc
->dispatch_list
= llist_del_all(&fcc
->issue_list
);
405 fcc
->dispatch_list
= llist_reverse_order(fcc
->dispatch_list
);
407 bio
->bi_bdev
= sbi
->sb
->s_bdev
;
408 ret
= submit_bio_wait(WRITE_FLUSH
, bio
);
410 llist_for_each_entry_safe(cmd
, next
,
411 fcc
->dispatch_list
, llnode
) {
413 complete(&cmd
->wait
);
416 fcc
->dispatch_list
= NULL
;
419 wait_event_interruptible(*q
,
420 kthread_should_stop() || !llist_empty(&fcc
->issue_list
));
424 int f2fs_issue_flush(struct f2fs_sb_info
*sbi
)
426 struct flush_cmd_control
*fcc
= SM_I(sbi
)->cmd_control_info
;
427 struct flush_cmd cmd
;
429 trace_f2fs_issue_flush(sbi
->sb
, test_opt(sbi
, NOBARRIER
),
430 test_opt(sbi
, FLUSH_MERGE
));
432 if (test_opt(sbi
, NOBARRIER
))
435 if (!test_opt(sbi
, FLUSH_MERGE
)) {
436 struct bio
*bio
= f2fs_bio_alloc(0);
439 bio
->bi_bdev
= sbi
->sb
->s_bdev
;
440 ret
= submit_bio_wait(WRITE_FLUSH
, bio
);
445 init_completion(&cmd
.wait
);
447 llist_add(&cmd
.llnode
, &fcc
->issue_list
);
449 if (!fcc
->dispatch_list
)
450 wake_up(&fcc
->flush_wait_queue
);
452 wait_for_completion(&cmd
.wait
);
457 int create_flush_cmd_control(struct f2fs_sb_info
*sbi
)
459 dev_t dev
= sbi
->sb
->s_bdev
->bd_dev
;
460 struct flush_cmd_control
*fcc
;
463 fcc
= kzalloc(sizeof(struct flush_cmd_control
), GFP_KERNEL
);
466 init_waitqueue_head(&fcc
->flush_wait_queue
);
467 init_llist_head(&fcc
->issue_list
);
468 SM_I(sbi
)->cmd_control_info
= fcc
;
469 fcc
->f2fs_issue_flush
= kthread_run(issue_flush_thread
, sbi
,
470 "f2fs_flush-%u:%u", MAJOR(dev
), MINOR(dev
));
471 if (IS_ERR(fcc
->f2fs_issue_flush
)) {
472 err
= PTR_ERR(fcc
->f2fs_issue_flush
);
474 SM_I(sbi
)->cmd_control_info
= NULL
;
481 void destroy_flush_cmd_control(struct f2fs_sb_info
*sbi
)
483 struct flush_cmd_control
*fcc
= SM_I(sbi
)->cmd_control_info
;
485 if (fcc
&& fcc
->f2fs_issue_flush
)
486 kthread_stop(fcc
->f2fs_issue_flush
);
488 SM_I(sbi
)->cmd_control_info
= NULL
;
491 static void __locate_dirty_segment(struct f2fs_sb_info
*sbi
, unsigned int segno
,
492 enum dirty_type dirty_type
)
494 struct dirty_seglist_info
*dirty_i
= DIRTY_I(sbi
);
496 /* need not be added */
497 if (IS_CURSEG(sbi
, segno
))
500 if (!test_and_set_bit(segno
, dirty_i
->dirty_segmap
[dirty_type
]))
501 dirty_i
->nr_dirty
[dirty_type
]++;
503 if (dirty_type
== DIRTY
) {
504 struct seg_entry
*sentry
= get_seg_entry(sbi
, segno
);
505 enum dirty_type t
= sentry
->type
;
507 if (unlikely(t
>= DIRTY
)) {
511 if (!test_and_set_bit(segno
, dirty_i
->dirty_segmap
[t
]))
512 dirty_i
->nr_dirty
[t
]++;
516 static void __remove_dirty_segment(struct f2fs_sb_info
*sbi
, unsigned int segno
,
517 enum dirty_type dirty_type
)
519 struct dirty_seglist_info
*dirty_i
= DIRTY_I(sbi
);
521 if (test_and_clear_bit(segno
, dirty_i
->dirty_segmap
[dirty_type
]))
522 dirty_i
->nr_dirty
[dirty_type
]--;
524 if (dirty_type
== DIRTY
) {
525 struct seg_entry
*sentry
= get_seg_entry(sbi
, segno
);
526 enum dirty_type t
= sentry
->type
;
528 if (test_and_clear_bit(segno
, dirty_i
->dirty_segmap
[t
]))
529 dirty_i
->nr_dirty
[t
]--;
531 if (get_valid_blocks(sbi
, segno
, sbi
->segs_per_sec
) == 0)
532 clear_bit(GET_SECNO(sbi
, segno
),
533 dirty_i
->victim_secmap
);
538 * Should not occur error such as -ENOMEM.
539 * Adding dirty entry into seglist is not critical operation.
540 * If a given segment is one of current working segments, it won't be added.
542 static void locate_dirty_segment(struct f2fs_sb_info
*sbi
, unsigned int segno
)
544 struct dirty_seglist_info
*dirty_i
= DIRTY_I(sbi
);
545 unsigned short valid_blocks
;
547 if (segno
== NULL_SEGNO
|| IS_CURSEG(sbi
, segno
))
550 mutex_lock(&dirty_i
->seglist_lock
);
552 valid_blocks
= get_valid_blocks(sbi
, segno
, 0);
554 if (valid_blocks
== 0) {
555 __locate_dirty_segment(sbi
, segno
, PRE
);
556 __remove_dirty_segment(sbi
, segno
, DIRTY
);
557 } else if (valid_blocks
< sbi
->blocks_per_seg
) {
558 __locate_dirty_segment(sbi
, segno
, DIRTY
);
560 /* Recovery routine with SSR needs this */
561 __remove_dirty_segment(sbi
, segno
, DIRTY
);
564 mutex_unlock(&dirty_i
->seglist_lock
);
567 static int f2fs_issue_discard(struct f2fs_sb_info
*sbi
,
568 block_t blkstart
, block_t blklen
)
570 sector_t start
= SECTOR_FROM_BLOCK(blkstart
);
571 sector_t len
= SECTOR_FROM_BLOCK(blklen
);
572 struct seg_entry
*se
;
576 for (i
= blkstart
; i
< blkstart
+ blklen
; i
++) {
577 se
= get_seg_entry(sbi
, GET_SEGNO(sbi
, i
));
578 offset
= GET_BLKOFF_FROM_SEG0(sbi
, i
);
580 if (!f2fs_test_and_set_bit(offset
, se
->discard_map
))
583 trace_f2fs_issue_discard(sbi
->sb
, blkstart
, blklen
);
584 return blkdev_issue_discard(sbi
->sb
->s_bdev
, start
, len
, GFP_NOFS
, 0);
587 bool discard_next_dnode(struct f2fs_sb_info
*sbi
, block_t blkaddr
)
589 int err
= -EOPNOTSUPP
;
591 if (test_opt(sbi
, DISCARD
)) {
592 struct seg_entry
*se
= get_seg_entry(sbi
,
593 GET_SEGNO(sbi
, blkaddr
));
594 unsigned int offset
= GET_BLKOFF_FROM_SEG0(sbi
, blkaddr
);
596 if (f2fs_test_bit(offset
, se
->discard_map
))
599 err
= f2fs_issue_discard(sbi
, blkaddr
, 1);
603 update_meta_page(sbi
, NULL
, blkaddr
);
609 static void __add_discard_entry(struct f2fs_sb_info
*sbi
,
610 struct cp_control
*cpc
, struct seg_entry
*se
,
611 unsigned int start
, unsigned int end
)
613 struct list_head
*head
= &SM_I(sbi
)->discard_list
;
614 struct discard_entry
*new, *last
;
616 if (!list_empty(head
)) {
617 last
= list_last_entry(head
, struct discard_entry
, list
);
618 if (START_BLOCK(sbi
, cpc
->trim_start
) + start
==
619 last
->blkaddr
+ last
->len
) {
620 last
->len
+= end
- start
;
625 new = f2fs_kmem_cache_alloc(discard_entry_slab
, GFP_NOFS
);
626 INIT_LIST_HEAD(&new->list
);
627 new->blkaddr
= START_BLOCK(sbi
, cpc
->trim_start
) + start
;
628 new->len
= end
- start
;
629 list_add_tail(&new->list
, head
);
631 SM_I(sbi
)->nr_discards
+= end
- start
;
634 static void add_discard_addrs(struct f2fs_sb_info
*sbi
, struct cp_control
*cpc
)
636 int entries
= SIT_VBLOCK_MAP_SIZE
/ sizeof(unsigned long);
637 int max_blocks
= sbi
->blocks_per_seg
;
638 struct seg_entry
*se
= get_seg_entry(sbi
, cpc
->trim_start
);
639 unsigned long *cur_map
= (unsigned long *)se
->cur_valid_map
;
640 unsigned long *ckpt_map
= (unsigned long *)se
->ckpt_valid_map
;
641 unsigned long *discard_map
= (unsigned long *)se
->discard_map
;
642 unsigned long *dmap
= SIT_I(sbi
)->tmp_map
;
643 unsigned int start
= 0, end
= -1;
644 bool force
= (cpc
->reason
== CP_DISCARD
);
647 if (se
->valid_blocks
== max_blocks
)
651 if (!test_opt(sbi
, DISCARD
) || !se
->valid_blocks
||
652 SM_I(sbi
)->nr_discards
>= SM_I(sbi
)->max_discards
)
656 /* SIT_VBLOCK_MAP_SIZE should be multiple of sizeof(unsigned long) */
657 for (i
= 0; i
< entries
; i
++)
658 dmap
[i
] = force
? ~ckpt_map
[i
] & ~discard_map
[i
] :
659 (cur_map
[i
] ^ ckpt_map
[i
]) & ckpt_map
[i
];
661 while (force
|| SM_I(sbi
)->nr_discards
<= SM_I(sbi
)->max_discards
) {
662 start
= __find_rev_next_bit(dmap
, max_blocks
, end
+ 1);
663 if (start
>= max_blocks
)
666 end
= __find_rev_next_zero_bit(dmap
, max_blocks
, start
+ 1);
667 __add_discard_entry(sbi
, cpc
, se
, start
, end
);
671 void release_discard_addrs(struct f2fs_sb_info
*sbi
)
673 struct list_head
*head
= &(SM_I(sbi
)->discard_list
);
674 struct discard_entry
*entry
, *this;
677 list_for_each_entry_safe(entry
, this, head
, list
) {
678 list_del(&entry
->list
);
679 kmem_cache_free(discard_entry_slab
, entry
);
684 * Should call clear_prefree_segments after checkpoint is done.
686 static void set_prefree_as_free_segments(struct f2fs_sb_info
*sbi
)
688 struct dirty_seglist_info
*dirty_i
= DIRTY_I(sbi
);
691 mutex_lock(&dirty_i
->seglist_lock
);
692 for_each_set_bit(segno
, dirty_i
->dirty_segmap
[PRE
], MAIN_SEGS(sbi
))
693 __set_test_and_free(sbi
, segno
);
694 mutex_unlock(&dirty_i
->seglist_lock
);
697 void clear_prefree_segments(struct f2fs_sb_info
*sbi
, struct cp_control
*cpc
)
699 struct list_head
*head
= &(SM_I(sbi
)->discard_list
);
700 struct discard_entry
*entry
, *this;
701 struct dirty_seglist_info
*dirty_i
= DIRTY_I(sbi
);
702 unsigned long *prefree_map
= dirty_i
->dirty_segmap
[PRE
];
703 unsigned int start
= 0, end
= -1;
705 mutex_lock(&dirty_i
->seglist_lock
);
709 start
= find_next_bit(prefree_map
, MAIN_SEGS(sbi
), end
+ 1);
710 if (start
>= MAIN_SEGS(sbi
))
712 end
= find_next_zero_bit(prefree_map
, MAIN_SEGS(sbi
),
715 for (i
= start
; i
< end
; i
++)
716 clear_bit(i
, prefree_map
);
718 dirty_i
->nr_dirty
[PRE
] -= end
- start
;
720 if (!test_opt(sbi
, DISCARD
))
723 f2fs_issue_discard(sbi
, START_BLOCK(sbi
, start
),
724 (end
- start
) << sbi
->log_blocks_per_seg
);
726 mutex_unlock(&dirty_i
->seglist_lock
);
728 /* send small discards */
729 list_for_each_entry_safe(entry
, this, head
, list
) {
730 if (cpc
->reason
== CP_DISCARD
&& entry
->len
< cpc
->trim_minlen
)
732 f2fs_issue_discard(sbi
, entry
->blkaddr
, entry
->len
);
733 cpc
->trimmed
+= entry
->len
;
735 list_del(&entry
->list
);
736 SM_I(sbi
)->nr_discards
-= entry
->len
;
737 kmem_cache_free(discard_entry_slab
, entry
);
741 static bool __mark_sit_entry_dirty(struct f2fs_sb_info
*sbi
, unsigned int segno
)
743 struct sit_info
*sit_i
= SIT_I(sbi
);
745 if (!__test_and_set_bit(segno
, sit_i
->dirty_sentries_bitmap
)) {
746 sit_i
->dirty_sentries
++;
753 static void __set_sit_entry_type(struct f2fs_sb_info
*sbi
, int type
,
754 unsigned int segno
, int modified
)
756 struct seg_entry
*se
= get_seg_entry(sbi
, segno
);
759 __mark_sit_entry_dirty(sbi
, segno
);
762 static void update_sit_entry(struct f2fs_sb_info
*sbi
, block_t blkaddr
, int del
)
764 struct seg_entry
*se
;
765 unsigned int segno
, offset
;
766 long int new_vblocks
;
768 segno
= GET_SEGNO(sbi
, blkaddr
);
770 se
= get_seg_entry(sbi
, segno
);
771 new_vblocks
= se
->valid_blocks
+ del
;
772 offset
= GET_BLKOFF_FROM_SEG0(sbi
, blkaddr
);
774 f2fs_bug_on(sbi
, (new_vblocks
>> (sizeof(unsigned short) << 3) ||
775 (new_vblocks
> sbi
->blocks_per_seg
)));
777 se
->valid_blocks
= new_vblocks
;
778 se
->mtime
= get_mtime(sbi
);
779 SIT_I(sbi
)->max_mtime
= se
->mtime
;
781 /* Update valid block bitmap */
783 if (f2fs_test_and_set_bit(offset
, se
->cur_valid_map
))
785 if (!f2fs_test_and_set_bit(offset
, se
->discard_map
))
788 if (!f2fs_test_and_clear_bit(offset
, se
->cur_valid_map
))
790 if (f2fs_test_and_clear_bit(offset
, se
->discard_map
))
793 if (!f2fs_test_bit(offset
, se
->ckpt_valid_map
))
794 se
->ckpt_valid_blocks
+= del
;
796 __mark_sit_entry_dirty(sbi
, segno
);
798 /* update total number of valid blocks to be written in ckpt area */
799 SIT_I(sbi
)->written_valid_blocks
+= del
;
801 if (sbi
->segs_per_sec
> 1)
802 get_sec_entry(sbi
, segno
)->valid_blocks
+= del
;
805 void refresh_sit_entry(struct f2fs_sb_info
*sbi
, block_t old
, block_t
new)
807 update_sit_entry(sbi
, new, 1);
808 if (GET_SEGNO(sbi
, old
) != NULL_SEGNO
)
809 update_sit_entry(sbi
, old
, -1);
811 locate_dirty_segment(sbi
, GET_SEGNO(sbi
, old
));
812 locate_dirty_segment(sbi
, GET_SEGNO(sbi
, new));
815 void invalidate_blocks(struct f2fs_sb_info
*sbi
, block_t addr
)
817 unsigned int segno
= GET_SEGNO(sbi
, addr
);
818 struct sit_info
*sit_i
= SIT_I(sbi
);
820 f2fs_bug_on(sbi
, addr
== NULL_ADDR
);
821 if (addr
== NEW_ADDR
)
824 /* add it into sit main buffer */
825 mutex_lock(&sit_i
->sentry_lock
);
827 update_sit_entry(sbi
, addr
, -1);
829 /* add it into dirty seglist */
830 locate_dirty_segment(sbi
, segno
);
832 mutex_unlock(&sit_i
->sentry_lock
);
835 bool is_checkpointed_data(struct f2fs_sb_info
*sbi
, block_t blkaddr
)
837 struct sit_info
*sit_i
= SIT_I(sbi
);
838 unsigned int segno
, offset
;
839 struct seg_entry
*se
;
842 if (blkaddr
== NEW_ADDR
|| blkaddr
== NULL_ADDR
)
845 mutex_lock(&sit_i
->sentry_lock
);
847 segno
= GET_SEGNO(sbi
, blkaddr
);
848 se
= get_seg_entry(sbi
, segno
);
849 offset
= GET_BLKOFF_FROM_SEG0(sbi
, blkaddr
);
851 if (f2fs_test_bit(offset
, se
->ckpt_valid_map
))
854 mutex_unlock(&sit_i
->sentry_lock
);
860 * This function should be resided under the curseg_mutex lock
862 static void __add_sum_entry(struct f2fs_sb_info
*sbi
, int type
,
863 struct f2fs_summary
*sum
)
865 struct curseg_info
*curseg
= CURSEG_I(sbi
, type
);
866 void *addr
= curseg
->sum_blk
;
867 addr
+= curseg
->next_blkoff
* sizeof(struct f2fs_summary
);
868 memcpy(addr
, sum
, sizeof(struct f2fs_summary
));
872 * Calculate the number of current summary pages for writing
874 int npages_for_summary_flush(struct f2fs_sb_info
*sbi
, bool for_ra
)
876 int valid_sum_count
= 0;
879 for (i
= CURSEG_HOT_DATA
; i
<= CURSEG_COLD_DATA
; i
++) {
880 if (sbi
->ckpt
->alloc_type
[i
] == SSR
)
881 valid_sum_count
+= sbi
->blocks_per_seg
;
884 valid_sum_count
+= le16_to_cpu(
885 F2FS_CKPT(sbi
)->cur_data_blkoff
[i
]);
887 valid_sum_count
+= curseg_blkoff(sbi
, i
);
891 sum_in_page
= (PAGE_SIZE
- 2 * SUM_JOURNAL_SIZE
-
892 SUM_FOOTER_SIZE
) / SUMMARY_SIZE
;
893 if (valid_sum_count
<= sum_in_page
)
895 else if ((valid_sum_count
- sum_in_page
) <=
896 (PAGE_SIZE
- SUM_FOOTER_SIZE
) / SUMMARY_SIZE
)
902 * Caller should put this summary page
904 struct page
*get_sum_page(struct f2fs_sb_info
*sbi
, unsigned int segno
)
906 return get_meta_page(sbi
, GET_SUM_BLOCK(sbi
, segno
));
909 void update_meta_page(struct f2fs_sb_info
*sbi
, void *src
, block_t blk_addr
)
911 struct page
*page
= grab_meta_page(sbi
, blk_addr
);
912 void *dst
= page_address(page
);
915 memcpy(dst
, src
, PAGE_SIZE
);
917 memset(dst
, 0, PAGE_SIZE
);
918 set_page_dirty(page
);
919 f2fs_put_page(page
, 1);
922 static void write_sum_page(struct f2fs_sb_info
*sbi
,
923 struct f2fs_summary_block
*sum_blk
, block_t blk_addr
)
925 update_meta_page(sbi
, (void *)sum_blk
, blk_addr
);
928 static void write_current_sum_page(struct f2fs_sb_info
*sbi
,
929 int type
, block_t blk_addr
)
931 struct curseg_info
*curseg
= CURSEG_I(sbi
, type
);
932 struct page
*page
= grab_meta_page(sbi
, blk_addr
);
933 struct f2fs_summary_block
*src
= curseg
->sum_blk
;
934 struct f2fs_summary_block
*dst
;
936 dst
= (struct f2fs_summary_block
*)page_address(page
);
938 mutex_lock(&curseg
->curseg_mutex
);
940 down_read(&curseg
->journal_rwsem
);
941 memcpy(&dst
->journal
, curseg
->journal
, SUM_JOURNAL_SIZE
);
942 up_read(&curseg
->journal_rwsem
);
944 memcpy(dst
->entries
, src
->entries
, SUM_ENTRY_SIZE
);
945 memcpy(&dst
->footer
, &src
->footer
, SUM_FOOTER_SIZE
);
947 mutex_unlock(&curseg
->curseg_mutex
);
949 set_page_dirty(page
);
950 f2fs_put_page(page
, 1);
953 static int is_next_segment_free(struct f2fs_sb_info
*sbi
, int type
)
955 struct curseg_info
*curseg
= CURSEG_I(sbi
, type
);
956 unsigned int segno
= curseg
->segno
+ 1;
957 struct free_segmap_info
*free_i
= FREE_I(sbi
);
959 if (segno
< MAIN_SEGS(sbi
) && segno
% sbi
->segs_per_sec
)
960 return !test_bit(segno
, free_i
->free_segmap
);
965 * Find a new segment from the free segments bitmap to right order
966 * This function should be returned with success, otherwise BUG
968 static void get_new_segment(struct f2fs_sb_info
*sbi
,
969 unsigned int *newseg
, bool new_sec
, int dir
)
971 struct free_segmap_info
*free_i
= FREE_I(sbi
);
972 unsigned int segno
, secno
, zoneno
;
973 unsigned int total_zones
= MAIN_SECS(sbi
) / sbi
->secs_per_zone
;
974 unsigned int hint
= *newseg
/ sbi
->segs_per_sec
;
975 unsigned int old_zoneno
= GET_ZONENO_FROM_SEGNO(sbi
, *newseg
);
976 unsigned int left_start
= hint
;
981 spin_lock(&free_i
->segmap_lock
);
983 if (!new_sec
&& ((*newseg
+ 1) % sbi
->segs_per_sec
)) {
984 segno
= find_next_zero_bit(free_i
->free_segmap
,
985 (hint
+ 1) * sbi
->segs_per_sec
, *newseg
+ 1);
986 if (segno
< (hint
+ 1) * sbi
->segs_per_sec
)
990 secno
= find_next_zero_bit(free_i
->free_secmap
, MAIN_SECS(sbi
), hint
);
991 if (secno
>= MAIN_SECS(sbi
)) {
992 if (dir
== ALLOC_RIGHT
) {
993 secno
= find_next_zero_bit(free_i
->free_secmap
,
995 f2fs_bug_on(sbi
, secno
>= MAIN_SECS(sbi
));
998 left_start
= hint
- 1;
1004 while (test_bit(left_start
, free_i
->free_secmap
)) {
1005 if (left_start
> 0) {
1009 left_start
= find_next_zero_bit(free_i
->free_secmap
,
1011 f2fs_bug_on(sbi
, left_start
>= MAIN_SECS(sbi
));
1017 segno
= secno
* sbi
->segs_per_sec
;
1018 zoneno
= secno
/ sbi
->secs_per_zone
;
1020 /* give up on finding another zone */
1023 if (sbi
->secs_per_zone
== 1)
1025 if (zoneno
== old_zoneno
)
1027 if (dir
== ALLOC_LEFT
) {
1028 if (!go_left
&& zoneno
+ 1 >= total_zones
)
1030 if (go_left
&& zoneno
== 0)
1033 for (i
= 0; i
< NR_CURSEG_TYPE
; i
++)
1034 if (CURSEG_I(sbi
, i
)->zone
== zoneno
)
1037 if (i
< NR_CURSEG_TYPE
) {
1038 /* zone is in user, try another */
1040 hint
= zoneno
* sbi
->secs_per_zone
- 1;
1041 else if (zoneno
+ 1 >= total_zones
)
1044 hint
= (zoneno
+ 1) * sbi
->secs_per_zone
;
1046 goto find_other_zone
;
1049 /* set it as dirty segment in free segmap */
1050 f2fs_bug_on(sbi
, test_bit(segno
, free_i
->free_segmap
));
1051 __set_inuse(sbi
, segno
);
1053 spin_unlock(&free_i
->segmap_lock
);
1056 static void reset_curseg(struct f2fs_sb_info
*sbi
, int type
, int modified
)
1058 struct curseg_info
*curseg
= CURSEG_I(sbi
, type
);
1059 struct summary_footer
*sum_footer
;
1061 curseg
->segno
= curseg
->next_segno
;
1062 curseg
->zone
= GET_ZONENO_FROM_SEGNO(sbi
, curseg
->segno
);
1063 curseg
->next_blkoff
= 0;
1064 curseg
->next_segno
= NULL_SEGNO
;
1066 sum_footer
= &(curseg
->sum_blk
->footer
);
1067 memset(sum_footer
, 0, sizeof(struct summary_footer
));
1068 if (IS_DATASEG(type
))
1069 SET_SUM_TYPE(sum_footer
, SUM_TYPE_DATA
);
1070 if (IS_NODESEG(type
))
1071 SET_SUM_TYPE(sum_footer
, SUM_TYPE_NODE
);
1072 __set_sit_entry_type(sbi
, type
, curseg
->segno
, modified
);
1076 * Allocate a current working segment.
1077 * This function always allocates a free segment in LFS manner.
1079 static void new_curseg(struct f2fs_sb_info
*sbi
, int type
, bool new_sec
)
1081 struct curseg_info
*curseg
= CURSEG_I(sbi
, type
);
1082 unsigned int segno
= curseg
->segno
;
1083 int dir
= ALLOC_LEFT
;
1085 write_sum_page(sbi
, curseg
->sum_blk
,
1086 GET_SUM_BLOCK(sbi
, segno
));
1087 if (type
== CURSEG_WARM_DATA
|| type
== CURSEG_COLD_DATA
)
1090 if (test_opt(sbi
, NOHEAP
))
1093 get_new_segment(sbi
, &segno
, new_sec
, dir
);
1094 curseg
->next_segno
= segno
;
1095 reset_curseg(sbi
, type
, 1);
1096 curseg
->alloc_type
= LFS
;
1099 static void __next_free_blkoff(struct f2fs_sb_info
*sbi
,
1100 struct curseg_info
*seg
, block_t start
)
1102 struct seg_entry
*se
= get_seg_entry(sbi
, seg
->segno
);
1103 int entries
= SIT_VBLOCK_MAP_SIZE
/ sizeof(unsigned long);
1104 unsigned long *target_map
= SIT_I(sbi
)->tmp_map
;
1105 unsigned long *ckpt_map
= (unsigned long *)se
->ckpt_valid_map
;
1106 unsigned long *cur_map
= (unsigned long *)se
->cur_valid_map
;
1109 for (i
= 0; i
< entries
; i
++)
1110 target_map
[i
] = ckpt_map
[i
] | cur_map
[i
];
1112 pos
= __find_rev_next_zero_bit(target_map
, sbi
->blocks_per_seg
, start
);
1114 seg
->next_blkoff
= pos
;
1118 * If a segment is written by LFS manner, next block offset is just obtained
1119 * by increasing the current block offset. However, if a segment is written by
1120 * SSR manner, next block offset obtained by calling __next_free_blkoff
1122 static void __refresh_next_blkoff(struct f2fs_sb_info
*sbi
,
1123 struct curseg_info
*seg
)
1125 if (seg
->alloc_type
== SSR
)
1126 __next_free_blkoff(sbi
, seg
, seg
->next_blkoff
+ 1);
1132 * This function always allocates a used segment(from dirty seglist) by SSR
1133 * manner, so it should recover the existing segment information of valid blocks
1135 static void change_curseg(struct f2fs_sb_info
*sbi
, int type
, bool reuse
)
1137 struct dirty_seglist_info
*dirty_i
= DIRTY_I(sbi
);
1138 struct curseg_info
*curseg
= CURSEG_I(sbi
, type
);
1139 unsigned int new_segno
= curseg
->next_segno
;
1140 struct f2fs_summary_block
*sum_node
;
1141 struct page
*sum_page
;
1143 write_sum_page(sbi
, curseg
->sum_blk
,
1144 GET_SUM_BLOCK(sbi
, curseg
->segno
));
1145 __set_test_and_inuse(sbi
, new_segno
);
1147 mutex_lock(&dirty_i
->seglist_lock
);
1148 __remove_dirty_segment(sbi
, new_segno
, PRE
);
1149 __remove_dirty_segment(sbi
, new_segno
, DIRTY
);
1150 mutex_unlock(&dirty_i
->seglist_lock
);
1152 reset_curseg(sbi
, type
, 1);
1153 curseg
->alloc_type
= SSR
;
1154 __next_free_blkoff(sbi
, curseg
, 0);
1157 sum_page
= get_sum_page(sbi
, new_segno
);
1158 sum_node
= (struct f2fs_summary_block
*)page_address(sum_page
);
1159 memcpy(curseg
->sum_blk
, sum_node
, SUM_ENTRY_SIZE
);
1160 f2fs_put_page(sum_page
, 1);
1164 static int get_ssr_segment(struct f2fs_sb_info
*sbi
, int type
)
1166 struct curseg_info
*curseg
= CURSEG_I(sbi
, type
);
1167 const struct victim_selection
*v_ops
= DIRTY_I(sbi
)->v_ops
;
1169 if (IS_NODESEG(type
) || !has_not_enough_free_secs(sbi
, 0))
1170 return v_ops
->get_victim(sbi
,
1171 &(curseg
)->next_segno
, BG_GC
, type
, SSR
);
1173 /* For data segments, let's do SSR more intensively */
1174 for (; type
>= CURSEG_HOT_DATA
; type
--)
1175 if (v_ops
->get_victim(sbi
, &(curseg
)->next_segno
,
1182 * flush out current segment and replace it with new segment
1183 * This function should be returned with success, otherwise BUG
1185 static void allocate_segment_by_default(struct f2fs_sb_info
*sbi
,
1186 int type
, bool force
)
1188 struct curseg_info
*curseg
= CURSEG_I(sbi
, type
);
1191 new_curseg(sbi
, type
, true);
1192 else if (type
== CURSEG_WARM_NODE
)
1193 new_curseg(sbi
, type
, false);
1194 else if (curseg
->alloc_type
== LFS
&& is_next_segment_free(sbi
, type
))
1195 new_curseg(sbi
, type
, false);
1196 else if (need_SSR(sbi
) && get_ssr_segment(sbi
, type
))
1197 change_curseg(sbi
, type
, true);
1199 new_curseg(sbi
, type
, false);
1201 stat_inc_seg_type(sbi
, curseg
);
1204 static void __allocate_new_segments(struct f2fs_sb_info
*sbi
, int type
)
1206 struct curseg_info
*curseg
= CURSEG_I(sbi
, type
);
1207 unsigned int old_segno
;
1209 old_segno
= curseg
->segno
;
1210 SIT_I(sbi
)->s_ops
->allocate_segment(sbi
, type
, true);
1211 locate_dirty_segment(sbi
, old_segno
);
1214 void allocate_new_segments(struct f2fs_sb_info
*sbi
)
1218 for (i
= CURSEG_HOT_DATA
; i
<= CURSEG_COLD_DATA
; i
++)
1219 __allocate_new_segments(sbi
, i
);
1222 static const struct segment_allocation default_salloc_ops
= {
1223 .allocate_segment
= allocate_segment_by_default
,
1226 int f2fs_trim_fs(struct f2fs_sb_info
*sbi
, struct fstrim_range
*range
)
1228 __u64 start
= F2FS_BYTES_TO_BLK(range
->start
);
1229 __u64 end
= start
+ F2FS_BYTES_TO_BLK(range
->len
) - 1;
1230 unsigned int start_segno
, end_segno
;
1231 struct cp_control cpc
;
1234 if (start
>= MAX_BLKADDR(sbi
) || range
->len
< sbi
->blocksize
)
1238 if (end
<= MAIN_BLKADDR(sbi
))
1241 /* start/end segment number in main_area */
1242 start_segno
= (start
<= MAIN_BLKADDR(sbi
)) ? 0 : GET_SEGNO(sbi
, start
);
1243 end_segno
= (end
>= MAX_BLKADDR(sbi
)) ? MAIN_SEGS(sbi
) - 1 :
1244 GET_SEGNO(sbi
, end
);
1245 cpc
.reason
= CP_DISCARD
;
1246 cpc
.trim_minlen
= max_t(__u64
, 1, F2FS_BYTES_TO_BLK(range
->minlen
));
1248 /* do checkpoint to issue discard commands safely */
1249 for (; start_segno
<= end_segno
; start_segno
= cpc
.trim_end
+ 1) {
1250 cpc
.trim_start
= start_segno
;
1252 if (sbi
->discard_blks
== 0)
1254 else if (sbi
->discard_blks
< BATCHED_TRIM_BLOCKS(sbi
))
1255 cpc
.trim_end
= end_segno
;
1257 cpc
.trim_end
= min_t(unsigned int,
1258 rounddown(start_segno
+
1259 BATCHED_TRIM_SEGMENTS(sbi
),
1260 sbi
->segs_per_sec
) - 1, end_segno
);
1262 mutex_lock(&sbi
->gc_mutex
);
1263 err
= write_checkpoint(sbi
, &cpc
);
1264 mutex_unlock(&sbi
->gc_mutex
);
1267 range
->len
= F2FS_BLK_TO_BYTES(cpc
.trimmed
);
1271 static bool __has_curseg_space(struct f2fs_sb_info
*sbi
, int type
)
1273 struct curseg_info
*curseg
= CURSEG_I(sbi
, type
);
1274 if (curseg
->next_blkoff
< sbi
->blocks_per_seg
)
1279 static int __get_segment_type_2(struct page
*page
, enum page_type p_type
)
1282 return CURSEG_HOT_DATA
;
1284 return CURSEG_HOT_NODE
;
1287 static int __get_segment_type_4(struct page
*page
, enum page_type p_type
)
1289 if (p_type
== DATA
) {
1290 struct inode
*inode
= page
->mapping
->host
;
1292 if (S_ISDIR(inode
->i_mode
))
1293 return CURSEG_HOT_DATA
;
1295 return CURSEG_COLD_DATA
;
1297 if (IS_DNODE(page
) && is_cold_node(page
))
1298 return CURSEG_WARM_NODE
;
1300 return CURSEG_COLD_NODE
;
1304 static int __get_segment_type_6(struct page
*page
, enum page_type p_type
)
1306 if (p_type
== DATA
) {
1307 struct inode
*inode
= page
->mapping
->host
;
1309 if (S_ISDIR(inode
->i_mode
))
1310 return CURSEG_HOT_DATA
;
1311 else if (is_cold_data(page
) || file_is_cold(inode
))
1312 return CURSEG_COLD_DATA
;
1314 return CURSEG_WARM_DATA
;
1317 return is_cold_node(page
) ? CURSEG_WARM_NODE
:
1320 return CURSEG_COLD_NODE
;
1324 static int __get_segment_type(struct page
*page
, enum page_type p_type
)
1326 switch (F2FS_P_SB(page
)->active_logs
) {
1328 return __get_segment_type_2(page
, p_type
);
1330 return __get_segment_type_4(page
, p_type
);
1332 /* NR_CURSEG_TYPE(6) logs by default */
1333 f2fs_bug_on(F2FS_P_SB(page
),
1334 F2FS_P_SB(page
)->active_logs
!= NR_CURSEG_TYPE
);
1335 return __get_segment_type_6(page
, p_type
);
1338 void allocate_data_block(struct f2fs_sb_info
*sbi
, struct page
*page
,
1339 block_t old_blkaddr
, block_t
*new_blkaddr
,
1340 struct f2fs_summary
*sum
, int type
)
1342 struct sit_info
*sit_i
= SIT_I(sbi
);
1343 struct curseg_info
*curseg
;
1344 bool direct_io
= (type
== CURSEG_DIRECT_IO
);
1346 type
= direct_io
? CURSEG_WARM_DATA
: type
;
1348 curseg
= CURSEG_I(sbi
, type
);
1350 mutex_lock(&curseg
->curseg_mutex
);
1351 mutex_lock(&sit_i
->sentry_lock
);
1353 /* direct_io'ed data is aligned to the segment for better performance */
1354 if (direct_io
&& curseg
->next_blkoff
&&
1355 !has_not_enough_free_secs(sbi
, 0))
1356 __allocate_new_segments(sbi
, type
);
1358 *new_blkaddr
= NEXT_FREE_BLKADDR(sbi
, curseg
);
1361 * __add_sum_entry should be resided under the curseg_mutex
1362 * because, this function updates a summary entry in the
1363 * current summary block.
1365 __add_sum_entry(sbi
, type
, sum
);
1367 __refresh_next_blkoff(sbi
, curseg
);
1369 stat_inc_block_count(sbi
, curseg
);
1371 if (!__has_curseg_space(sbi
, type
))
1372 sit_i
->s_ops
->allocate_segment(sbi
, type
, false);
1374 * SIT information should be updated before segment allocation,
1375 * since SSR needs latest valid block information.
1377 refresh_sit_entry(sbi
, old_blkaddr
, *new_blkaddr
);
1379 mutex_unlock(&sit_i
->sentry_lock
);
1381 if (page
&& IS_NODESEG(type
))
1382 fill_node_footer_blkaddr(page
, NEXT_FREE_BLKADDR(sbi
, curseg
));
1384 mutex_unlock(&curseg
->curseg_mutex
);
1387 static void do_write_page(struct f2fs_summary
*sum
, struct f2fs_io_info
*fio
)
1389 int type
= __get_segment_type(fio
->page
, fio
->type
);
1391 allocate_data_block(fio
->sbi
, fio
->page
, fio
->old_blkaddr
,
1392 &fio
->new_blkaddr
, sum
, type
);
1394 /* writeout dirty page into bdev */
1395 f2fs_submit_page_mbio(fio
);
1398 void write_meta_page(struct f2fs_sb_info
*sbi
, struct page
*page
)
1400 struct f2fs_io_info fio
= {
1403 .rw
= WRITE_SYNC
| REQ_META
| REQ_PRIO
,
1404 .old_blkaddr
= page
->index
,
1405 .new_blkaddr
= page
->index
,
1407 .encrypted_page
= NULL
,
1410 if (unlikely(page
->index
>= MAIN_BLKADDR(sbi
)))
1411 fio
.rw
&= ~REQ_META
;
1413 set_page_writeback(page
);
1414 f2fs_submit_page_mbio(&fio
);
1417 void write_node_page(unsigned int nid
, struct f2fs_io_info
*fio
)
1419 struct f2fs_summary sum
;
1421 set_summary(&sum
, nid
, 0, 0);
1422 do_write_page(&sum
, fio
);
1425 void write_data_page(struct dnode_of_data
*dn
, struct f2fs_io_info
*fio
)
1427 struct f2fs_sb_info
*sbi
= fio
->sbi
;
1428 struct f2fs_summary sum
;
1429 struct node_info ni
;
1431 f2fs_bug_on(sbi
, dn
->data_blkaddr
== NULL_ADDR
);
1432 get_node_info(sbi
, dn
->nid
, &ni
);
1433 set_summary(&sum
, dn
->nid
, dn
->ofs_in_node
, ni
.version
);
1434 do_write_page(&sum
, fio
);
1435 f2fs_update_data_blkaddr(dn
, fio
->new_blkaddr
);
1438 void rewrite_data_page(struct f2fs_io_info
*fio
)
1440 fio
->new_blkaddr
= fio
->old_blkaddr
;
1441 stat_inc_inplace_blocks(fio
->sbi
);
1442 f2fs_submit_page_mbio(fio
);
1445 void __f2fs_replace_block(struct f2fs_sb_info
*sbi
, struct f2fs_summary
*sum
,
1446 block_t old_blkaddr
, block_t new_blkaddr
,
1447 bool recover_curseg
, bool recover_newaddr
)
1449 struct sit_info
*sit_i
= SIT_I(sbi
);
1450 struct curseg_info
*curseg
;
1451 unsigned int segno
, old_cursegno
;
1452 struct seg_entry
*se
;
1454 unsigned short old_blkoff
;
1456 segno
= GET_SEGNO(sbi
, new_blkaddr
);
1457 se
= get_seg_entry(sbi
, segno
);
1460 if (!recover_curseg
) {
1461 /* for recovery flow */
1462 if (se
->valid_blocks
== 0 && !IS_CURSEG(sbi
, segno
)) {
1463 if (old_blkaddr
== NULL_ADDR
)
1464 type
= CURSEG_COLD_DATA
;
1466 type
= CURSEG_WARM_DATA
;
1469 if (!IS_CURSEG(sbi
, segno
))
1470 type
= CURSEG_WARM_DATA
;
1473 curseg
= CURSEG_I(sbi
, type
);
1475 mutex_lock(&curseg
->curseg_mutex
);
1476 mutex_lock(&sit_i
->sentry_lock
);
1478 old_cursegno
= curseg
->segno
;
1479 old_blkoff
= curseg
->next_blkoff
;
1481 /* change the current segment */
1482 if (segno
!= curseg
->segno
) {
1483 curseg
->next_segno
= segno
;
1484 change_curseg(sbi
, type
, true);
1487 curseg
->next_blkoff
= GET_BLKOFF_FROM_SEG0(sbi
, new_blkaddr
);
1488 __add_sum_entry(sbi
, type
, sum
);
1490 if (!recover_curseg
|| recover_newaddr
)
1491 update_sit_entry(sbi
, new_blkaddr
, 1);
1492 if (GET_SEGNO(sbi
, old_blkaddr
) != NULL_SEGNO
)
1493 update_sit_entry(sbi
, old_blkaddr
, -1);
1495 locate_dirty_segment(sbi
, GET_SEGNO(sbi
, old_blkaddr
));
1496 locate_dirty_segment(sbi
, GET_SEGNO(sbi
, new_blkaddr
));
1498 locate_dirty_segment(sbi
, old_cursegno
);
1500 if (recover_curseg
) {
1501 if (old_cursegno
!= curseg
->segno
) {
1502 curseg
->next_segno
= old_cursegno
;
1503 change_curseg(sbi
, type
, true);
1505 curseg
->next_blkoff
= old_blkoff
;
1508 mutex_unlock(&sit_i
->sentry_lock
);
1509 mutex_unlock(&curseg
->curseg_mutex
);
1512 void f2fs_replace_block(struct f2fs_sb_info
*sbi
, struct dnode_of_data
*dn
,
1513 block_t old_addr
, block_t new_addr
,
1514 unsigned char version
, bool recover_curseg
,
1515 bool recover_newaddr
)
1517 struct f2fs_summary sum
;
1519 set_summary(&sum
, dn
->nid
, dn
->ofs_in_node
, version
);
1521 __f2fs_replace_block(sbi
, &sum
, old_addr
, new_addr
,
1522 recover_curseg
, recover_newaddr
);
1524 f2fs_update_data_blkaddr(dn
, new_addr
);
1527 void f2fs_wait_on_page_writeback(struct page
*page
,
1528 enum page_type type
, bool ordered
)
1530 if (PageWriteback(page
)) {
1531 struct f2fs_sb_info
*sbi
= F2FS_P_SB(page
);
1533 f2fs_submit_merged_bio_cond(sbi
, NULL
, page
, 0, type
, WRITE
);
1535 wait_on_page_writeback(page
);
1537 wait_for_stable_page(page
);
1541 void f2fs_wait_on_encrypted_page_writeback(struct f2fs_sb_info
*sbi
,
1546 if (blkaddr
== NEW_ADDR
)
1549 f2fs_bug_on(sbi
, blkaddr
== NULL_ADDR
);
1551 cpage
= find_lock_page(META_MAPPING(sbi
), blkaddr
);
1553 f2fs_wait_on_page_writeback(cpage
, DATA
, true);
1554 f2fs_put_page(cpage
, 1);
1558 static int read_compacted_summaries(struct f2fs_sb_info
*sbi
)
1560 struct f2fs_checkpoint
*ckpt
= F2FS_CKPT(sbi
);
1561 struct curseg_info
*seg_i
;
1562 unsigned char *kaddr
;
1567 start
= start_sum_block(sbi
);
1569 page
= get_meta_page(sbi
, start
++);
1570 kaddr
= (unsigned char *)page_address(page
);
1572 /* Step 1: restore nat cache */
1573 seg_i
= CURSEG_I(sbi
, CURSEG_HOT_DATA
);
1574 memcpy(seg_i
->journal
, kaddr
, SUM_JOURNAL_SIZE
);
1576 /* Step 2: restore sit cache */
1577 seg_i
= CURSEG_I(sbi
, CURSEG_COLD_DATA
);
1578 memcpy(seg_i
->journal
, kaddr
+ SUM_JOURNAL_SIZE
, SUM_JOURNAL_SIZE
);
1579 offset
= 2 * SUM_JOURNAL_SIZE
;
1581 /* Step 3: restore summary entries */
1582 for (i
= CURSEG_HOT_DATA
; i
<= CURSEG_COLD_DATA
; i
++) {
1583 unsigned short blk_off
;
1586 seg_i
= CURSEG_I(sbi
, i
);
1587 segno
= le32_to_cpu(ckpt
->cur_data_segno
[i
]);
1588 blk_off
= le16_to_cpu(ckpt
->cur_data_blkoff
[i
]);
1589 seg_i
->next_segno
= segno
;
1590 reset_curseg(sbi
, i
, 0);
1591 seg_i
->alloc_type
= ckpt
->alloc_type
[i
];
1592 seg_i
->next_blkoff
= blk_off
;
1594 if (seg_i
->alloc_type
== SSR
)
1595 blk_off
= sbi
->blocks_per_seg
;
1597 for (j
= 0; j
< blk_off
; j
++) {
1598 struct f2fs_summary
*s
;
1599 s
= (struct f2fs_summary
*)(kaddr
+ offset
);
1600 seg_i
->sum_blk
->entries
[j
] = *s
;
1601 offset
+= SUMMARY_SIZE
;
1602 if (offset
+ SUMMARY_SIZE
<= PAGE_SIZE
-
1606 f2fs_put_page(page
, 1);
1609 page
= get_meta_page(sbi
, start
++);
1610 kaddr
= (unsigned char *)page_address(page
);
1614 f2fs_put_page(page
, 1);
1618 static int read_normal_summaries(struct f2fs_sb_info
*sbi
, int type
)
1620 struct f2fs_checkpoint
*ckpt
= F2FS_CKPT(sbi
);
1621 struct f2fs_summary_block
*sum
;
1622 struct curseg_info
*curseg
;
1624 unsigned short blk_off
;
1625 unsigned int segno
= 0;
1626 block_t blk_addr
= 0;
1628 /* get segment number and block addr */
1629 if (IS_DATASEG(type
)) {
1630 segno
= le32_to_cpu(ckpt
->cur_data_segno
[type
]);
1631 blk_off
= le16_to_cpu(ckpt
->cur_data_blkoff
[type
-
1633 if (__exist_node_summaries(sbi
))
1634 blk_addr
= sum_blk_addr(sbi
, NR_CURSEG_TYPE
, type
);
1636 blk_addr
= sum_blk_addr(sbi
, NR_CURSEG_DATA_TYPE
, type
);
1638 segno
= le32_to_cpu(ckpt
->cur_node_segno
[type
-
1640 blk_off
= le16_to_cpu(ckpt
->cur_node_blkoff
[type
-
1642 if (__exist_node_summaries(sbi
))
1643 blk_addr
= sum_blk_addr(sbi
, NR_CURSEG_NODE_TYPE
,
1644 type
- CURSEG_HOT_NODE
);
1646 blk_addr
= GET_SUM_BLOCK(sbi
, segno
);
1649 new = get_meta_page(sbi
, blk_addr
);
1650 sum
= (struct f2fs_summary_block
*)page_address(new);
1652 if (IS_NODESEG(type
)) {
1653 if (__exist_node_summaries(sbi
)) {
1654 struct f2fs_summary
*ns
= &sum
->entries
[0];
1656 for (i
= 0; i
< sbi
->blocks_per_seg
; i
++, ns
++) {
1658 ns
->ofs_in_node
= 0;
1663 err
= restore_node_summary(sbi
, segno
, sum
);
1665 f2fs_put_page(new, 1);
1671 /* set uncompleted segment to curseg */
1672 curseg
= CURSEG_I(sbi
, type
);
1673 mutex_lock(&curseg
->curseg_mutex
);
1675 /* update journal info */
1676 down_write(&curseg
->journal_rwsem
);
1677 memcpy(curseg
->journal
, &sum
->journal
, SUM_JOURNAL_SIZE
);
1678 up_write(&curseg
->journal_rwsem
);
1680 memcpy(curseg
->sum_blk
->entries
, sum
->entries
, SUM_ENTRY_SIZE
);
1681 memcpy(&curseg
->sum_blk
->footer
, &sum
->footer
, SUM_FOOTER_SIZE
);
1682 curseg
->next_segno
= segno
;
1683 reset_curseg(sbi
, type
, 0);
1684 curseg
->alloc_type
= ckpt
->alloc_type
[type
];
1685 curseg
->next_blkoff
= blk_off
;
1686 mutex_unlock(&curseg
->curseg_mutex
);
1687 f2fs_put_page(new, 1);
1691 static int restore_curseg_summaries(struct f2fs_sb_info
*sbi
)
1693 int type
= CURSEG_HOT_DATA
;
1696 if (is_set_ckpt_flags(F2FS_CKPT(sbi
), CP_COMPACT_SUM_FLAG
)) {
1697 int npages
= npages_for_summary_flush(sbi
, true);
1700 ra_meta_pages(sbi
, start_sum_block(sbi
), npages
,
1703 /* restore for compacted data summary */
1704 if (read_compacted_summaries(sbi
))
1706 type
= CURSEG_HOT_NODE
;
1709 if (__exist_node_summaries(sbi
))
1710 ra_meta_pages(sbi
, sum_blk_addr(sbi
, NR_CURSEG_TYPE
, type
),
1711 NR_CURSEG_TYPE
- type
, META_CP
, true);
1713 for (; type
<= CURSEG_COLD_NODE
; type
++) {
1714 err
= read_normal_summaries(sbi
, type
);
1722 static void write_compacted_summaries(struct f2fs_sb_info
*sbi
, block_t blkaddr
)
1725 unsigned char *kaddr
;
1726 struct f2fs_summary
*summary
;
1727 struct curseg_info
*seg_i
;
1728 int written_size
= 0;
1731 page
= grab_meta_page(sbi
, blkaddr
++);
1732 kaddr
= (unsigned char *)page_address(page
);
1734 /* Step 1: write nat cache */
1735 seg_i
= CURSEG_I(sbi
, CURSEG_HOT_DATA
);
1736 memcpy(kaddr
, seg_i
->journal
, SUM_JOURNAL_SIZE
);
1737 written_size
+= SUM_JOURNAL_SIZE
;
1739 /* Step 2: write sit cache */
1740 seg_i
= CURSEG_I(sbi
, CURSEG_COLD_DATA
);
1741 memcpy(kaddr
+ written_size
, seg_i
->journal
, SUM_JOURNAL_SIZE
);
1742 written_size
+= SUM_JOURNAL_SIZE
;
1744 /* Step 3: write summary entries */
1745 for (i
= CURSEG_HOT_DATA
; i
<= CURSEG_COLD_DATA
; i
++) {
1746 unsigned short blkoff
;
1747 seg_i
= CURSEG_I(sbi
, i
);
1748 if (sbi
->ckpt
->alloc_type
[i
] == SSR
)
1749 blkoff
= sbi
->blocks_per_seg
;
1751 blkoff
= curseg_blkoff(sbi
, i
);
1753 for (j
= 0; j
< blkoff
; j
++) {
1755 page
= grab_meta_page(sbi
, blkaddr
++);
1756 kaddr
= (unsigned char *)page_address(page
);
1759 summary
= (struct f2fs_summary
*)(kaddr
+ written_size
);
1760 *summary
= seg_i
->sum_blk
->entries
[j
];
1761 written_size
+= SUMMARY_SIZE
;
1763 if (written_size
+ SUMMARY_SIZE
<= PAGE_SIZE
-
1767 set_page_dirty(page
);
1768 f2fs_put_page(page
, 1);
1773 set_page_dirty(page
);
1774 f2fs_put_page(page
, 1);
1778 static void write_normal_summaries(struct f2fs_sb_info
*sbi
,
1779 block_t blkaddr
, int type
)
1782 if (IS_DATASEG(type
))
1783 end
= type
+ NR_CURSEG_DATA_TYPE
;
1785 end
= type
+ NR_CURSEG_NODE_TYPE
;
1787 for (i
= type
; i
< end
; i
++)
1788 write_current_sum_page(sbi
, i
, blkaddr
+ (i
- type
));
1791 void write_data_summaries(struct f2fs_sb_info
*sbi
, block_t start_blk
)
1793 if (is_set_ckpt_flags(F2FS_CKPT(sbi
), CP_COMPACT_SUM_FLAG
))
1794 write_compacted_summaries(sbi
, start_blk
);
1796 write_normal_summaries(sbi
, start_blk
, CURSEG_HOT_DATA
);
1799 void write_node_summaries(struct f2fs_sb_info
*sbi
, block_t start_blk
)
1801 write_normal_summaries(sbi
, start_blk
, CURSEG_HOT_NODE
);
1804 int lookup_journal_in_cursum(struct f2fs_journal
*journal
, int type
,
1805 unsigned int val
, int alloc
)
1809 if (type
== NAT_JOURNAL
) {
1810 for (i
= 0; i
< nats_in_cursum(journal
); i
++) {
1811 if (le32_to_cpu(nid_in_journal(journal
, i
)) == val
)
1814 if (alloc
&& __has_cursum_space(journal
, 1, NAT_JOURNAL
))
1815 return update_nats_in_cursum(journal
, 1);
1816 } else if (type
== SIT_JOURNAL
) {
1817 for (i
= 0; i
< sits_in_cursum(journal
); i
++)
1818 if (le32_to_cpu(segno_in_journal(journal
, i
)) == val
)
1820 if (alloc
&& __has_cursum_space(journal
, 1, SIT_JOURNAL
))
1821 return update_sits_in_cursum(journal
, 1);
1826 static struct page
*get_current_sit_page(struct f2fs_sb_info
*sbi
,
1829 return get_meta_page(sbi
, current_sit_addr(sbi
, segno
));
1832 static struct page
*get_next_sit_page(struct f2fs_sb_info
*sbi
,
1835 struct sit_info
*sit_i
= SIT_I(sbi
);
1836 struct page
*src_page
, *dst_page
;
1837 pgoff_t src_off
, dst_off
;
1838 void *src_addr
, *dst_addr
;
1840 src_off
= current_sit_addr(sbi
, start
);
1841 dst_off
= next_sit_addr(sbi
, src_off
);
1843 /* get current sit block page without lock */
1844 src_page
= get_meta_page(sbi
, src_off
);
1845 dst_page
= grab_meta_page(sbi
, dst_off
);
1846 f2fs_bug_on(sbi
, PageDirty(src_page
));
1848 src_addr
= page_address(src_page
);
1849 dst_addr
= page_address(dst_page
);
1850 memcpy(dst_addr
, src_addr
, PAGE_SIZE
);
1852 set_page_dirty(dst_page
);
1853 f2fs_put_page(src_page
, 1);
1855 set_to_next_sit(sit_i
, start
);
1860 static struct sit_entry_set
*grab_sit_entry_set(void)
1862 struct sit_entry_set
*ses
=
1863 f2fs_kmem_cache_alloc(sit_entry_set_slab
, GFP_NOFS
);
1866 INIT_LIST_HEAD(&ses
->set_list
);
1870 static void release_sit_entry_set(struct sit_entry_set
*ses
)
1872 list_del(&ses
->set_list
);
1873 kmem_cache_free(sit_entry_set_slab
, ses
);
1876 static void adjust_sit_entry_set(struct sit_entry_set
*ses
,
1877 struct list_head
*head
)
1879 struct sit_entry_set
*next
= ses
;
1881 if (list_is_last(&ses
->set_list
, head
))
1884 list_for_each_entry_continue(next
, head
, set_list
)
1885 if (ses
->entry_cnt
<= next
->entry_cnt
)
1888 list_move_tail(&ses
->set_list
, &next
->set_list
);
1891 static void add_sit_entry(unsigned int segno
, struct list_head
*head
)
1893 struct sit_entry_set
*ses
;
1894 unsigned int start_segno
= START_SEGNO(segno
);
1896 list_for_each_entry(ses
, head
, set_list
) {
1897 if (ses
->start_segno
== start_segno
) {
1899 adjust_sit_entry_set(ses
, head
);
1904 ses
= grab_sit_entry_set();
1906 ses
->start_segno
= start_segno
;
1908 list_add(&ses
->set_list
, head
);
1911 static void add_sits_in_set(struct f2fs_sb_info
*sbi
)
1913 struct f2fs_sm_info
*sm_info
= SM_I(sbi
);
1914 struct list_head
*set_list
= &sm_info
->sit_entry_set
;
1915 unsigned long *bitmap
= SIT_I(sbi
)->dirty_sentries_bitmap
;
1918 for_each_set_bit(segno
, bitmap
, MAIN_SEGS(sbi
))
1919 add_sit_entry(segno
, set_list
);
1922 static void remove_sits_in_journal(struct f2fs_sb_info
*sbi
)
1924 struct curseg_info
*curseg
= CURSEG_I(sbi
, CURSEG_COLD_DATA
);
1925 struct f2fs_journal
*journal
= curseg
->journal
;
1928 down_write(&curseg
->journal_rwsem
);
1929 for (i
= 0; i
< sits_in_cursum(journal
); i
++) {
1933 segno
= le32_to_cpu(segno_in_journal(journal
, i
));
1934 dirtied
= __mark_sit_entry_dirty(sbi
, segno
);
1937 add_sit_entry(segno
, &SM_I(sbi
)->sit_entry_set
);
1939 update_sits_in_cursum(journal
, -i
);
1940 up_write(&curseg
->journal_rwsem
);
1944 * CP calls this function, which flushes SIT entries including sit_journal,
1945 * and moves prefree segs to free segs.
1947 void flush_sit_entries(struct f2fs_sb_info
*sbi
, struct cp_control
*cpc
)
1949 struct sit_info
*sit_i
= SIT_I(sbi
);
1950 unsigned long *bitmap
= sit_i
->dirty_sentries_bitmap
;
1951 struct curseg_info
*curseg
= CURSEG_I(sbi
, CURSEG_COLD_DATA
);
1952 struct f2fs_journal
*journal
= curseg
->journal
;
1953 struct sit_entry_set
*ses
, *tmp
;
1954 struct list_head
*head
= &SM_I(sbi
)->sit_entry_set
;
1955 bool to_journal
= true;
1956 struct seg_entry
*se
;
1958 mutex_lock(&sit_i
->sentry_lock
);
1960 if (!sit_i
->dirty_sentries
)
1964 * add and account sit entries of dirty bitmap in sit entry
1967 add_sits_in_set(sbi
);
1970 * if there are no enough space in journal to store dirty sit
1971 * entries, remove all entries from journal and add and account
1972 * them in sit entry set.
1974 if (!__has_cursum_space(journal
, sit_i
->dirty_sentries
, SIT_JOURNAL
))
1975 remove_sits_in_journal(sbi
);
1978 * there are two steps to flush sit entries:
1979 * #1, flush sit entries to journal in current cold data summary block.
1980 * #2, flush sit entries to sit page.
1982 list_for_each_entry_safe(ses
, tmp
, head
, set_list
) {
1983 struct page
*page
= NULL
;
1984 struct f2fs_sit_block
*raw_sit
= NULL
;
1985 unsigned int start_segno
= ses
->start_segno
;
1986 unsigned int end
= min(start_segno
+ SIT_ENTRY_PER_BLOCK
,
1987 (unsigned long)MAIN_SEGS(sbi
));
1988 unsigned int segno
= start_segno
;
1991 !__has_cursum_space(journal
, ses
->entry_cnt
, SIT_JOURNAL
))
1995 down_write(&curseg
->journal_rwsem
);
1997 page
= get_next_sit_page(sbi
, start_segno
);
1998 raw_sit
= page_address(page
);
2001 /* flush dirty sit entries in region of current sit set */
2002 for_each_set_bit_from(segno
, bitmap
, end
) {
2003 int offset
, sit_offset
;
2005 se
= get_seg_entry(sbi
, segno
);
2007 /* add discard candidates */
2008 if (cpc
->reason
!= CP_DISCARD
) {
2009 cpc
->trim_start
= segno
;
2010 add_discard_addrs(sbi
, cpc
);
2014 offset
= lookup_journal_in_cursum(journal
,
2015 SIT_JOURNAL
, segno
, 1);
2016 f2fs_bug_on(sbi
, offset
< 0);
2017 segno_in_journal(journal
, offset
) =
2019 seg_info_to_raw_sit(se
,
2020 &sit_in_journal(journal
, offset
));
2022 sit_offset
= SIT_ENTRY_OFFSET(sit_i
, segno
);
2023 seg_info_to_raw_sit(se
,
2024 &raw_sit
->entries
[sit_offset
]);
2027 __clear_bit(segno
, bitmap
);
2028 sit_i
->dirty_sentries
--;
2033 up_write(&curseg
->journal_rwsem
);
2035 f2fs_put_page(page
, 1);
2037 f2fs_bug_on(sbi
, ses
->entry_cnt
);
2038 release_sit_entry_set(ses
);
2041 f2fs_bug_on(sbi
, !list_empty(head
));
2042 f2fs_bug_on(sbi
, sit_i
->dirty_sentries
);
2044 if (cpc
->reason
== CP_DISCARD
) {
2045 for (; cpc
->trim_start
<= cpc
->trim_end
; cpc
->trim_start
++)
2046 add_discard_addrs(sbi
, cpc
);
2048 mutex_unlock(&sit_i
->sentry_lock
);
2050 set_prefree_as_free_segments(sbi
);
2053 static int build_sit_info(struct f2fs_sb_info
*sbi
)
2055 struct f2fs_super_block
*raw_super
= F2FS_RAW_SUPER(sbi
);
2056 struct f2fs_checkpoint
*ckpt
= F2FS_CKPT(sbi
);
2057 struct sit_info
*sit_i
;
2058 unsigned int sit_segs
, start
;
2059 char *src_bitmap
, *dst_bitmap
;
2060 unsigned int bitmap_size
;
2062 /* allocate memory for SIT information */
2063 sit_i
= kzalloc(sizeof(struct sit_info
), GFP_KERNEL
);
2067 SM_I(sbi
)->sit_info
= sit_i
;
2069 sit_i
->sentries
= f2fs_kvzalloc(MAIN_SEGS(sbi
) *
2070 sizeof(struct seg_entry
), GFP_KERNEL
);
2071 if (!sit_i
->sentries
)
2074 bitmap_size
= f2fs_bitmap_size(MAIN_SEGS(sbi
));
2075 sit_i
->dirty_sentries_bitmap
= f2fs_kvzalloc(bitmap_size
, GFP_KERNEL
);
2076 if (!sit_i
->dirty_sentries_bitmap
)
2079 for (start
= 0; start
< MAIN_SEGS(sbi
); start
++) {
2080 sit_i
->sentries
[start
].cur_valid_map
2081 = kzalloc(SIT_VBLOCK_MAP_SIZE
, GFP_KERNEL
);
2082 sit_i
->sentries
[start
].ckpt_valid_map
2083 = kzalloc(SIT_VBLOCK_MAP_SIZE
, GFP_KERNEL
);
2084 sit_i
->sentries
[start
].discard_map
2085 = kzalloc(SIT_VBLOCK_MAP_SIZE
, GFP_KERNEL
);
2086 if (!sit_i
->sentries
[start
].cur_valid_map
||
2087 !sit_i
->sentries
[start
].ckpt_valid_map
||
2088 !sit_i
->sentries
[start
].discard_map
)
2092 sit_i
->tmp_map
= kzalloc(SIT_VBLOCK_MAP_SIZE
, GFP_KERNEL
);
2093 if (!sit_i
->tmp_map
)
2096 if (sbi
->segs_per_sec
> 1) {
2097 sit_i
->sec_entries
= f2fs_kvzalloc(MAIN_SECS(sbi
) *
2098 sizeof(struct sec_entry
), GFP_KERNEL
);
2099 if (!sit_i
->sec_entries
)
2103 /* get information related with SIT */
2104 sit_segs
= le32_to_cpu(raw_super
->segment_count_sit
) >> 1;
2106 /* setup SIT bitmap from ckeckpoint pack */
2107 bitmap_size
= __bitmap_size(sbi
, SIT_BITMAP
);
2108 src_bitmap
= __bitmap_ptr(sbi
, SIT_BITMAP
);
2110 dst_bitmap
= kmemdup(src_bitmap
, bitmap_size
, GFP_KERNEL
);
2114 /* init SIT information */
2115 sit_i
->s_ops
= &default_salloc_ops
;
2117 sit_i
->sit_base_addr
= le32_to_cpu(raw_super
->sit_blkaddr
);
2118 sit_i
->sit_blocks
= sit_segs
<< sbi
->log_blocks_per_seg
;
2119 sit_i
->written_valid_blocks
= le64_to_cpu(ckpt
->valid_block_count
);
2120 sit_i
->sit_bitmap
= dst_bitmap
;
2121 sit_i
->bitmap_size
= bitmap_size
;
2122 sit_i
->dirty_sentries
= 0;
2123 sit_i
->sents_per_block
= SIT_ENTRY_PER_BLOCK
;
2124 sit_i
->elapsed_time
= le64_to_cpu(sbi
->ckpt
->elapsed_time
);
2125 sit_i
->mounted_time
= CURRENT_TIME_SEC
.tv_sec
;
2126 mutex_init(&sit_i
->sentry_lock
);
2130 static int build_free_segmap(struct f2fs_sb_info
*sbi
)
2132 struct free_segmap_info
*free_i
;
2133 unsigned int bitmap_size
, sec_bitmap_size
;
2135 /* allocate memory for free segmap information */
2136 free_i
= kzalloc(sizeof(struct free_segmap_info
), GFP_KERNEL
);
2140 SM_I(sbi
)->free_info
= free_i
;
2142 bitmap_size
= f2fs_bitmap_size(MAIN_SEGS(sbi
));
2143 free_i
->free_segmap
= f2fs_kvmalloc(bitmap_size
, GFP_KERNEL
);
2144 if (!free_i
->free_segmap
)
2147 sec_bitmap_size
= f2fs_bitmap_size(MAIN_SECS(sbi
));
2148 free_i
->free_secmap
= f2fs_kvmalloc(sec_bitmap_size
, GFP_KERNEL
);
2149 if (!free_i
->free_secmap
)
2152 /* set all segments as dirty temporarily */
2153 memset(free_i
->free_segmap
, 0xff, bitmap_size
);
2154 memset(free_i
->free_secmap
, 0xff, sec_bitmap_size
);
2156 /* init free segmap information */
2157 free_i
->start_segno
= GET_SEGNO_FROM_SEG0(sbi
, MAIN_BLKADDR(sbi
));
2158 free_i
->free_segments
= 0;
2159 free_i
->free_sections
= 0;
2160 spin_lock_init(&free_i
->segmap_lock
);
2164 static int build_curseg(struct f2fs_sb_info
*sbi
)
2166 struct curseg_info
*array
;
2169 array
= kcalloc(NR_CURSEG_TYPE
, sizeof(*array
), GFP_KERNEL
);
2173 SM_I(sbi
)->curseg_array
= array
;
2175 for (i
= 0; i
< NR_CURSEG_TYPE
; i
++) {
2176 mutex_init(&array
[i
].curseg_mutex
);
2177 array
[i
].sum_blk
= kzalloc(PAGE_SIZE
, GFP_KERNEL
);
2178 if (!array
[i
].sum_blk
)
2180 init_rwsem(&array
[i
].journal_rwsem
);
2181 array
[i
].journal
= kzalloc(sizeof(struct f2fs_journal
),
2183 if (!array
[i
].journal
)
2185 array
[i
].segno
= NULL_SEGNO
;
2186 array
[i
].next_blkoff
= 0;
2188 return restore_curseg_summaries(sbi
);
2191 static void build_sit_entries(struct f2fs_sb_info
*sbi
)
2193 struct sit_info
*sit_i
= SIT_I(sbi
);
2194 struct curseg_info
*curseg
= CURSEG_I(sbi
, CURSEG_COLD_DATA
);
2195 struct f2fs_journal
*journal
= curseg
->journal
;
2196 int sit_blk_cnt
= SIT_BLK_CNT(sbi
);
2197 unsigned int i
, start
, end
;
2198 unsigned int readed
, start_blk
= 0;
2199 int nrpages
= MAX_BIO_BLOCKS(sbi
) * 8;
2202 readed
= ra_meta_pages(sbi
, start_blk
, nrpages
, META_SIT
, true);
2204 start
= start_blk
* sit_i
->sents_per_block
;
2205 end
= (start_blk
+ readed
) * sit_i
->sents_per_block
;
2207 for (; start
< end
&& start
< MAIN_SEGS(sbi
); start
++) {
2208 struct seg_entry
*se
= &sit_i
->sentries
[start
];
2209 struct f2fs_sit_block
*sit_blk
;
2210 struct f2fs_sit_entry sit
;
2213 down_read(&curseg
->journal_rwsem
);
2214 for (i
= 0; i
< sits_in_cursum(journal
); i
++) {
2215 if (le32_to_cpu(segno_in_journal(journal
, i
))
2217 sit
= sit_in_journal(journal
, i
);
2218 up_read(&curseg
->journal_rwsem
);
2222 up_read(&curseg
->journal_rwsem
);
2224 page
= get_current_sit_page(sbi
, start
);
2225 sit_blk
= (struct f2fs_sit_block
*)page_address(page
);
2226 sit
= sit_blk
->entries
[SIT_ENTRY_OFFSET(sit_i
, start
)];
2227 f2fs_put_page(page
, 1);
2229 check_block_count(sbi
, start
, &sit
);
2230 seg_info_from_raw_sit(se
, &sit
);
2232 /* build discard map only one time */
2233 memcpy(se
->discard_map
, se
->cur_valid_map
, SIT_VBLOCK_MAP_SIZE
);
2234 sbi
->discard_blks
+= sbi
->blocks_per_seg
- se
->valid_blocks
;
2236 if (sbi
->segs_per_sec
> 1) {
2237 struct sec_entry
*e
= get_sec_entry(sbi
, start
);
2238 e
->valid_blocks
+= se
->valid_blocks
;
2241 start_blk
+= readed
;
2242 } while (start_blk
< sit_blk_cnt
);
2245 static void init_free_segmap(struct f2fs_sb_info
*sbi
)
2250 for (start
= 0; start
< MAIN_SEGS(sbi
); start
++) {
2251 struct seg_entry
*sentry
= get_seg_entry(sbi
, start
);
2252 if (!sentry
->valid_blocks
)
2253 __set_free(sbi
, start
);
2256 /* set use the current segments */
2257 for (type
= CURSEG_HOT_DATA
; type
<= CURSEG_COLD_NODE
; type
++) {
2258 struct curseg_info
*curseg_t
= CURSEG_I(sbi
, type
);
2259 __set_test_and_inuse(sbi
, curseg_t
->segno
);
2263 static void init_dirty_segmap(struct f2fs_sb_info
*sbi
)
2265 struct dirty_seglist_info
*dirty_i
= DIRTY_I(sbi
);
2266 struct free_segmap_info
*free_i
= FREE_I(sbi
);
2267 unsigned int segno
= 0, offset
= 0;
2268 unsigned short valid_blocks
;
2271 /* find dirty segment based on free segmap */
2272 segno
= find_next_inuse(free_i
, MAIN_SEGS(sbi
), offset
);
2273 if (segno
>= MAIN_SEGS(sbi
))
2276 valid_blocks
= get_valid_blocks(sbi
, segno
, 0);
2277 if (valid_blocks
== sbi
->blocks_per_seg
|| !valid_blocks
)
2279 if (valid_blocks
> sbi
->blocks_per_seg
) {
2280 f2fs_bug_on(sbi
, 1);
2283 mutex_lock(&dirty_i
->seglist_lock
);
2284 __locate_dirty_segment(sbi
, segno
, DIRTY
);
2285 mutex_unlock(&dirty_i
->seglist_lock
);
2289 static int init_victim_secmap(struct f2fs_sb_info
*sbi
)
2291 struct dirty_seglist_info
*dirty_i
= DIRTY_I(sbi
);
2292 unsigned int bitmap_size
= f2fs_bitmap_size(MAIN_SECS(sbi
));
2294 dirty_i
->victim_secmap
= f2fs_kvzalloc(bitmap_size
, GFP_KERNEL
);
2295 if (!dirty_i
->victim_secmap
)
2300 static int build_dirty_segmap(struct f2fs_sb_info
*sbi
)
2302 struct dirty_seglist_info
*dirty_i
;
2303 unsigned int bitmap_size
, i
;
2305 /* allocate memory for dirty segments list information */
2306 dirty_i
= kzalloc(sizeof(struct dirty_seglist_info
), GFP_KERNEL
);
2310 SM_I(sbi
)->dirty_info
= dirty_i
;
2311 mutex_init(&dirty_i
->seglist_lock
);
2313 bitmap_size
= f2fs_bitmap_size(MAIN_SEGS(sbi
));
2315 for (i
= 0; i
< NR_DIRTY_TYPE
; i
++) {
2316 dirty_i
->dirty_segmap
[i
] = f2fs_kvzalloc(bitmap_size
, GFP_KERNEL
);
2317 if (!dirty_i
->dirty_segmap
[i
])
2321 init_dirty_segmap(sbi
);
2322 return init_victim_secmap(sbi
);
2326 * Update min, max modified time for cost-benefit GC algorithm
2328 static void init_min_max_mtime(struct f2fs_sb_info
*sbi
)
2330 struct sit_info
*sit_i
= SIT_I(sbi
);
2333 mutex_lock(&sit_i
->sentry_lock
);
2335 sit_i
->min_mtime
= LLONG_MAX
;
2337 for (segno
= 0; segno
< MAIN_SEGS(sbi
); segno
+= sbi
->segs_per_sec
) {
2339 unsigned long long mtime
= 0;
2341 for (i
= 0; i
< sbi
->segs_per_sec
; i
++)
2342 mtime
+= get_seg_entry(sbi
, segno
+ i
)->mtime
;
2344 mtime
= div_u64(mtime
, sbi
->segs_per_sec
);
2346 if (sit_i
->min_mtime
> mtime
)
2347 sit_i
->min_mtime
= mtime
;
2349 sit_i
->max_mtime
= get_mtime(sbi
);
2350 mutex_unlock(&sit_i
->sentry_lock
);
2353 int build_segment_manager(struct f2fs_sb_info
*sbi
)
2355 struct f2fs_super_block
*raw_super
= F2FS_RAW_SUPER(sbi
);
2356 struct f2fs_checkpoint
*ckpt
= F2FS_CKPT(sbi
);
2357 struct f2fs_sm_info
*sm_info
;
2360 sm_info
= kzalloc(sizeof(struct f2fs_sm_info
), GFP_KERNEL
);
2365 sbi
->sm_info
= sm_info
;
2366 sm_info
->seg0_blkaddr
= le32_to_cpu(raw_super
->segment0_blkaddr
);
2367 sm_info
->main_blkaddr
= le32_to_cpu(raw_super
->main_blkaddr
);
2368 sm_info
->segment_count
= le32_to_cpu(raw_super
->segment_count
);
2369 sm_info
->reserved_segments
= le32_to_cpu(ckpt
->rsvd_segment_count
);
2370 sm_info
->ovp_segments
= le32_to_cpu(ckpt
->overprov_segment_count
);
2371 sm_info
->main_segments
= le32_to_cpu(raw_super
->segment_count_main
);
2372 sm_info
->ssa_blkaddr
= le32_to_cpu(raw_super
->ssa_blkaddr
);
2373 sm_info
->rec_prefree_segments
= sm_info
->main_segments
*
2374 DEF_RECLAIM_PREFREE_SEGMENTS
/ 100;
2375 sm_info
->ipu_policy
= 1 << F2FS_IPU_FSYNC
;
2376 sm_info
->min_ipu_util
= DEF_MIN_IPU_UTIL
;
2377 sm_info
->min_fsync_blocks
= DEF_MIN_FSYNC_BLOCKS
;
2379 INIT_LIST_HEAD(&sm_info
->discard_list
);
2380 sm_info
->nr_discards
= 0;
2381 sm_info
->max_discards
= 0;
2383 sm_info
->trim_sections
= DEF_BATCHED_TRIM_SECTIONS
;
2385 INIT_LIST_HEAD(&sm_info
->sit_entry_set
);
2387 if (test_opt(sbi
, FLUSH_MERGE
) && !f2fs_readonly(sbi
->sb
)) {
2388 err
= create_flush_cmd_control(sbi
);
2393 err
= build_sit_info(sbi
);
2396 err
= build_free_segmap(sbi
);
2399 err
= build_curseg(sbi
);
2403 /* reinit free segmap based on SIT */
2404 build_sit_entries(sbi
);
2406 init_free_segmap(sbi
);
2407 err
= build_dirty_segmap(sbi
);
2411 init_min_max_mtime(sbi
);
2415 static void discard_dirty_segmap(struct f2fs_sb_info
*sbi
,
2416 enum dirty_type dirty_type
)
2418 struct dirty_seglist_info
*dirty_i
= DIRTY_I(sbi
);
2420 mutex_lock(&dirty_i
->seglist_lock
);
2421 kvfree(dirty_i
->dirty_segmap
[dirty_type
]);
2422 dirty_i
->nr_dirty
[dirty_type
] = 0;
2423 mutex_unlock(&dirty_i
->seglist_lock
);
2426 static void destroy_victim_secmap(struct f2fs_sb_info
*sbi
)
2428 struct dirty_seglist_info
*dirty_i
= DIRTY_I(sbi
);
2429 kvfree(dirty_i
->victim_secmap
);
2432 static void destroy_dirty_segmap(struct f2fs_sb_info
*sbi
)
2434 struct dirty_seglist_info
*dirty_i
= DIRTY_I(sbi
);
2440 /* discard pre-free/dirty segments list */
2441 for (i
= 0; i
< NR_DIRTY_TYPE
; i
++)
2442 discard_dirty_segmap(sbi
, i
);
2444 destroy_victim_secmap(sbi
);
2445 SM_I(sbi
)->dirty_info
= NULL
;
2449 static void destroy_curseg(struct f2fs_sb_info
*sbi
)
2451 struct curseg_info
*array
= SM_I(sbi
)->curseg_array
;
2456 SM_I(sbi
)->curseg_array
= NULL
;
2457 for (i
= 0; i
< NR_CURSEG_TYPE
; i
++) {
2458 kfree(array
[i
].sum_blk
);
2459 kfree(array
[i
].journal
);
2464 static void destroy_free_segmap(struct f2fs_sb_info
*sbi
)
2466 struct free_segmap_info
*free_i
= SM_I(sbi
)->free_info
;
2469 SM_I(sbi
)->free_info
= NULL
;
2470 kvfree(free_i
->free_segmap
);
2471 kvfree(free_i
->free_secmap
);
2475 static void destroy_sit_info(struct f2fs_sb_info
*sbi
)
2477 struct sit_info
*sit_i
= SIT_I(sbi
);
2483 if (sit_i
->sentries
) {
2484 for (start
= 0; start
< MAIN_SEGS(sbi
); start
++) {
2485 kfree(sit_i
->sentries
[start
].cur_valid_map
);
2486 kfree(sit_i
->sentries
[start
].ckpt_valid_map
);
2487 kfree(sit_i
->sentries
[start
].discard_map
);
2490 kfree(sit_i
->tmp_map
);
2492 kvfree(sit_i
->sentries
);
2493 kvfree(sit_i
->sec_entries
);
2494 kvfree(sit_i
->dirty_sentries_bitmap
);
2496 SM_I(sbi
)->sit_info
= NULL
;
2497 kfree(sit_i
->sit_bitmap
);
2501 void destroy_segment_manager(struct f2fs_sb_info
*sbi
)
2503 struct f2fs_sm_info
*sm_info
= SM_I(sbi
);
2507 destroy_flush_cmd_control(sbi
);
2508 destroy_dirty_segmap(sbi
);
2509 destroy_curseg(sbi
);
2510 destroy_free_segmap(sbi
);
2511 destroy_sit_info(sbi
);
2512 sbi
->sm_info
= NULL
;
2516 int __init
create_segment_manager_caches(void)
2518 discard_entry_slab
= f2fs_kmem_cache_create("discard_entry",
2519 sizeof(struct discard_entry
));
2520 if (!discard_entry_slab
)
2523 sit_entry_set_slab
= f2fs_kmem_cache_create("sit_entry_set",
2524 sizeof(struct sit_entry_set
));
2525 if (!sit_entry_set_slab
)
2526 goto destory_discard_entry
;
2528 inmem_entry_slab
= f2fs_kmem_cache_create("inmem_page_entry",
2529 sizeof(struct inmem_pages
));
2530 if (!inmem_entry_slab
)
2531 goto destroy_sit_entry_set
;
2534 destroy_sit_entry_set
:
2535 kmem_cache_destroy(sit_entry_set_slab
);
2536 destory_discard_entry
:
2537 kmem_cache_destroy(discard_entry_slab
);
2542 void destroy_segment_manager_caches(void)
2544 kmem_cache_destroy(sit_entry_set_slab
);
2545 kmem_cache_destroy(discard_entry_slab
);
2546 kmem_cache_destroy(inmem_entry_slab
);