4 * Copyright (c) 2012 Samsung Electronics Co., Ltd.
5 * http://www.samsung.com/
7 * This program is free software; you can redistribute it and/or modify
8 * it under the terms of the GNU General Public License version 2 as
9 * published by the Free Software Foundation.
12 #include <linux/f2fs_fs.h>
13 #include <linux/bio.h>
14 #include <linux/blkdev.h>
15 #include <linux/prefetch.h>
16 #include <linux/kthread.h>
17 #include <linux/swap.h>
18 #include <linux/timer.h>
24 #include <trace/events/f2fs.h>
26 #define __reverse_ffz(x) __reverse_ffs(~(x))
28 static struct kmem_cache
*discard_entry_slab
;
29 static struct kmem_cache
*sit_entry_set_slab
;
30 static struct kmem_cache
*inmem_entry_slab
;
32 static unsigned long __reverse_ulong(unsigned char *str
)
34 unsigned long tmp
= 0;
35 int shift
= 24, idx
= 0;
37 #if BITS_PER_LONG == 64
41 tmp
|= (unsigned long)str
[idx
++] << shift
;
42 shift
-= BITS_PER_BYTE
;
48 * __reverse_ffs is copied from include/asm-generic/bitops/__ffs.h since
49 * MSB and LSB are reversed in a byte by f2fs_set_bit.
51 static inline unsigned long __reverse_ffs(unsigned long word
)
55 #if BITS_PER_LONG == 64
56 if ((word
& 0xffffffff00000000UL
) == 0)
61 if ((word
& 0xffff0000) == 0)
66 if ((word
& 0xff00) == 0)
71 if ((word
& 0xf0) == 0)
76 if ((word
& 0xc) == 0)
81 if ((word
& 0x2) == 0)
87 * __find_rev_next(_zero)_bit is copied from lib/find_next_bit.c because
88 * f2fs_set_bit makes MSB and LSB reversed in a byte.
89 * @size must be integral times of unsigned long.
92 * f2fs_set_bit(0, bitmap) => 1000 0000
93 * f2fs_set_bit(7, bitmap) => 0000 0001
95 static unsigned long __find_rev_next_bit(const unsigned long *addr
,
96 unsigned long size
, unsigned long offset
)
98 const unsigned long *p
= addr
+ BIT_WORD(offset
);
99 unsigned long result
= size
;
105 size
-= (offset
& ~(BITS_PER_LONG
- 1));
106 offset
%= BITS_PER_LONG
;
112 tmp
= __reverse_ulong((unsigned char *)p
);
114 tmp
&= ~0UL >> offset
;
115 if (size
< BITS_PER_LONG
)
116 tmp
&= (~0UL << (BITS_PER_LONG
- size
));
120 if (size
<= BITS_PER_LONG
)
122 size
-= BITS_PER_LONG
;
128 return result
- size
+ __reverse_ffs(tmp
);
131 static unsigned long __find_rev_next_zero_bit(const unsigned long *addr
,
132 unsigned long size
, unsigned long offset
)
134 const unsigned long *p
= addr
+ BIT_WORD(offset
);
135 unsigned long result
= size
;
141 size
-= (offset
& ~(BITS_PER_LONG
- 1));
142 offset
%= BITS_PER_LONG
;
148 tmp
= __reverse_ulong((unsigned char *)p
);
151 tmp
|= ~0UL << (BITS_PER_LONG
- offset
);
152 if (size
< BITS_PER_LONG
)
157 if (size
<= BITS_PER_LONG
)
159 size
-= BITS_PER_LONG
;
165 return result
- size
+ __reverse_ffz(tmp
);
168 void register_inmem_page(struct inode
*inode
, struct page
*page
)
170 struct f2fs_inode_info
*fi
= F2FS_I(inode
);
171 struct inmem_pages
*new;
173 f2fs_trace_pid(page
);
175 set_page_private(page
, (unsigned long)ATOMIC_WRITTEN_PAGE
);
176 SetPagePrivate(page
);
178 new = f2fs_kmem_cache_alloc(inmem_entry_slab
, GFP_NOFS
);
180 /* add atomic page indices to the list */
182 INIT_LIST_HEAD(&new->list
);
184 /* increase reference count with clean state */
185 mutex_lock(&fi
->inmem_lock
);
187 list_add_tail(&new->list
, &fi
->inmem_pages
);
188 inc_page_count(F2FS_I_SB(inode
), F2FS_INMEM_PAGES
);
189 mutex_unlock(&fi
->inmem_lock
);
191 trace_f2fs_register_inmem_page(page
, INMEM
);
194 static int __revoke_inmem_pages(struct inode
*inode
,
195 struct list_head
*head
, bool drop
, bool recover
)
197 struct f2fs_sb_info
*sbi
= F2FS_I_SB(inode
);
198 struct inmem_pages
*cur
, *tmp
;
201 list_for_each_entry_safe(cur
, tmp
, head
, list
) {
202 struct page
*page
= cur
->page
;
205 trace_f2fs_commit_inmem_page(page
, INMEM_DROP
);
210 struct dnode_of_data dn
;
213 trace_f2fs_commit_inmem_page(page
, INMEM_REVOKE
);
215 set_new_dnode(&dn
, inode
, NULL
, NULL
, 0);
216 if (get_dnode_of_data(&dn
, page
->index
, LOOKUP_NODE
)) {
220 get_node_info(sbi
, dn
.nid
, &ni
);
221 f2fs_replace_block(sbi
, &dn
, dn
.data_blkaddr
,
222 cur
->old_addr
, ni
.version
, true, true);
226 /* we don't need to invalidate this in the sccessful status */
228 ClearPageUptodate(page
);
229 set_page_private(page
, 0);
230 ClearPagePrivate(page
);
231 f2fs_put_page(page
, 1);
233 list_del(&cur
->list
);
234 kmem_cache_free(inmem_entry_slab
, cur
);
235 dec_page_count(F2FS_I_SB(inode
), F2FS_INMEM_PAGES
);
240 void drop_inmem_pages(struct inode
*inode
)
242 struct f2fs_inode_info
*fi
= F2FS_I(inode
);
244 clear_inode_flag(inode
, FI_ATOMIC_FILE
);
246 mutex_lock(&fi
->inmem_lock
);
247 __revoke_inmem_pages(inode
, &fi
->inmem_pages
, true, false);
248 mutex_unlock(&fi
->inmem_lock
);
251 static int __commit_inmem_pages(struct inode
*inode
,
252 struct list_head
*revoke_list
)
254 struct f2fs_sb_info
*sbi
= F2FS_I_SB(inode
);
255 struct f2fs_inode_info
*fi
= F2FS_I(inode
);
256 struct inmem_pages
*cur
, *tmp
;
257 struct f2fs_io_info fio
= {
260 .rw
= WRITE_SYNC
| REQ_PRIO
,
261 .encrypted_page
= NULL
,
263 bool submit_bio
= false;
266 list_for_each_entry_safe(cur
, tmp
, &fi
->inmem_pages
, list
) {
267 struct page
*page
= cur
->page
;
270 if (page
->mapping
== inode
->i_mapping
) {
271 trace_f2fs_commit_inmem_page(page
, INMEM
);
273 set_page_dirty(page
);
274 f2fs_wait_on_page_writeback(page
, DATA
, true);
275 if (clear_page_dirty_for_io(page
))
276 inode_dec_dirty_pages(inode
);
279 err
= do_write_data_page(&fio
);
285 /* record old blkaddr for revoking */
286 cur
->old_addr
= fio
.old_blkaddr
;
288 clear_cold_data(page
);
292 list_move_tail(&cur
->list
, revoke_list
);
296 f2fs_submit_merged_bio_cond(sbi
, inode
, NULL
, 0, DATA
, WRITE
);
299 __revoke_inmem_pages(inode
, revoke_list
, false, false);
304 int commit_inmem_pages(struct inode
*inode
)
306 struct f2fs_sb_info
*sbi
= F2FS_I_SB(inode
);
307 struct f2fs_inode_info
*fi
= F2FS_I(inode
);
308 struct list_head revoke_list
;
311 INIT_LIST_HEAD(&revoke_list
);
312 f2fs_balance_fs(sbi
, true);
315 mutex_lock(&fi
->inmem_lock
);
316 err
= __commit_inmem_pages(inode
, &revoke_list
);
320 * try to revoke all committed pages, but still we could fail
321 * due to no memory or other reason, if that happened, EAGAIN
322 * will be returned, which means in such case, transaction is
323 * already not integrity, caller should use journal to do the
324 * recovery or rewrite & commit last transaction. For other
325 * error number, revoking was done by filesystem itself.
327 ret
= __revoke_inmem_pages(inode
, &revoke_list
, false, true);
331 /* drop all uncommitted pages */
332 __revoke_inmem_pages(inode
, &fi
->inmem_pages
, true, false);
334 mutex_unlock(&fi
->inmem_lock
);
341 * This function balances dirty node and dentry pages.
342 * In addition, it controls garbage collection.
344 void f2fs_balance_fs(struct f2fs_sb_info
*sbi
, bool need
)
349 /* balance_fs_bg is able to be pending */
350 if (excess_cached_nats(sbi
))
351 f2fs_balance_fs_bg(sbi
);
354 * We should do GC or end up with checkpoint, if there are so many dirty
355 * dir/node pages without enough free segments.
357 if (has_not_enough_free_secs(sbi
, 0)) {
358 mutex_lock(&sbi
->gc_mutex
);
363 void f2fs_balance_fs_bg(struct f2fs_sb_info
*sbi
)
365 /* try to shrink extent cache when there is no enough memory */
366 if (!available_free_memory(sbi
, EXTENT_CACHE
))
367 f2fs_shrink_extent_tree(sbi
, EXTENT_CACHE_SHRINK_NUMBER
);
369 /* check the # of cached NAT entries */
370 if (!available_free_memory(sbi
, NAT_ENTRIES
))
371 try_to_free_nats(sbi
, NAT_ENTRY_PER_BLOCK
);
373 if (!available_free_memory(sbi
, FREE_NIDS
))
374 try_to_free_nids(sbi
, NAT_ENTRY_PER_BLOCK
* FREE_NID_PAGES
);
376 /* checkpoint is the only way to shrink partial cached entries */
377 if (!available_free_memory(sbi
, NAT_ENTRIES
) ||
378 !available_free_memory(sbi
, INO_ENTRIES
) ||
379 excess_prefree_segs(sbi
) ||
380 excess_dirty_nats(sbi
) ||
381 (is_idle(sbi
) && f2fs_time_over(sbi
, CP_TIME
))) {
382 if (test_opt(sbi
, DATA_FLUSH
)) {
383 struct blk_plug plug
;
385 blk_start_plug(&plug
);
386 sync_dirty_inodes(sbi
, FILE_INODE
);
387 blk_finish_plug(&plug
);
389 f2fs_sync_fs(sbi
->sb
, true);
390 stat_inc_bg_cp_count(sbi
->stat_info
);
394 static int issue_flush_thread(void *data
)
396 struct f2fs_sb_info
*sbi
= data
;
397 struct flush_cmd_control
*fcc
= SM_I(sbi
)->cmd_control_info
;
398 wait_queue_head_t
*q
= &fcc
->flush_wait_queue
;
400 if (kthread_should_stop())
403 if (!llist_empty(&fcc
->issue_list
)) {
405 struct flush_cmd
*cmd
, *next
;
408 bio
= f2fs_bio_alloc(0);
410 fcc
->dispatch_list
= llist_del_all(&fcc
->issue_list
);
411 fcc
->dispatch_list
= llist_reverse_order(fcc
->dispatch_list
);
413 bio
->bi_bdev
= sbi
->sb
->s_bdev
;
414 ret
= submit_bio_wait(WRITE_FLUSH
, bio
);
416 llist_for_each_entry_safe(cmd
, next
,
417 fcc
->dispatch_list
, llnode
) {
419 complete(&cmd
->wait
);
422 fcc
->dispatch_list
= NULL
;
425 wait_event_interruptible(*q
,
426 kthread_should_stop() || !llist_empty(&fcc
->issue_list
));
430 int f2fs_issue_flush(struct f2fs_sb_info
*sbi
)
432 struct flush_cmd_control
*fcc
= SM_I(sbi
)->cmd_control_info
;
433 struct flush_cmd cmd
;
435 trace_f2fs_issue_flush(sbi
->sb
, test_opt(sbi
, NOBARRIER
),
436 test_opt(sbi
, FLUSH_MERGE
));
438 if (test_opt(sbi
, NOBARRIER
))
441 if (!test_opt(sbi
, FLUSH_MERGE
) || !atomic_read(&fcc
->submit_flush
)) {
442 struct bio
*bio
= f2fs_bio_alloc(0);
445 atomic_inc(&fcc
->submit_flush
);
446 bio
->bi_bdev
= sbi
->sb
->s_bdev
;
447 ret
= submit_bio_wait(WRITE_FLUSH
, bio
);
448 atomic_dec(&fcc
->submit_flush
);
453 init_completion(&cmd
.wait
);
455 atomic_inc(&fcc
->submit_flush
);
456 llist_add(&cmd
.llnode
, &fcc
->issue_list
);
458 if (!fcc
->dispatch_list
)
459 wake_up(&fcc
->flush_wait_queue
);
461 wait_for_completion(&cmd
.wait
);
462 atomic_dec(&fcc
->submit_flush
);
467 int create_flush_cmd_control(struct f2fs_sb_info
*sbi
)
469 dev_t dev
= sbi
->sb
->s_bdev
->bd_dev
;
470 struct flush_cmd_control
*fcc
;
473 fcc
= kzalloc(sizeof(struct flush_cmd_control
), GFP_KERNEL
);
476 atomic_set(&fcc
->submit_flush
, 0);
477 init_waitqueue_head(&fcc
->flush_wait_queue
);
478 init_llist_head(&fcc
->issue_list
);
479 SM_I(sbi
)->cmd_control_info
= fcc
;
480 fcc
->f2fs_issue_flush
= kthread_run(issue_flush_thread
, sbi
,
481 "f2fs_flush-%u:%u", MAJOR(dev
), MINOR(dev
));
482 if (IS_ERR(fcc
->f2fs_issue_flush
)) {
483 err
= PTR_ERR(fcc
->f2fs_issue_flush
);
485 SM_I(sbi
)->cmd_control_info
= NULL
;
492 void destroy_flush_cmd_control(struct f2fs_sb_info
*sbi
)
494 struct flush_cmd_control
*fcc
= SM_I(sbi
)->cmd_control_info
;
496 if (fcc
&& fcc
->f2fs_issue_flush
)
497 kthread_stop(fcc
->f2fs_issue_flush
);
499 SM_I(sbi
)->cmd_control_info
= NULL
;
502 static void __locate_dirty_segment(struct f2fs_sb_info
*sbi
, unsigned int segno
,
503 enum dirty_type dirty_type
)
505 struct dirty_seglist_info
*dirty_i
= DIRTY_I(sbi
);
507 /* need not be added */
508 if (IS_CURSEG(sbi
, segno
))
511 if (!test_and_set_bit(segno
, dirty_i
->dirty_segmap
[dirty_type
]))
512 dirty_i
->nr_dirty
[dirty_type
]++;
514 if (dirty_type
== DIRTY
) {
515 struct seg_entry
*sentry
= get_seg_entry(sbi
, segno
);
516 enum dirty_type t
= sentry
->type
;
518 if (unlikely(t
>= DIRTY
)) {
522 if (!test_and_set_bit(segno
, dirty_i
->dirty_segmap
[t
]))
523 dirty_i
->nr_dirty
[t
]++;
527 static void __remove_dirty_segment(struct f2fs_sb_info
*sbi
, unsigned int segno
,
528 enum dirty_type dirty_type
)
530 struct dirty_seglist_info
*dirty_i
= DIRTY_I(sbi
);
532 if (test_and_clear_bit(segno
, dirty_i
->dirty_segmap
[dirty_type
]))
533 dirty_i
->nr_dirty
[dirty_type
]--;
535 if (dirty_type
== DIRTY
) {
536 struct seg_entry
*sentry
= get_seg_entry(sbi
, segno
);
537 enum dirty_type t
= sentry
->type
;
539 if (test_and_clear_bit(segno
, dirty_i
->dirty_segmap
[t
]))
540 dirty_i
->nr_dirty
[t
]--;
542 if (get_valid_blocks(sbi
, segno
, sbi
->segs_per_sec
) == 0)
543 clear_bit(GET_SECNO(sbi
, segno
),
544 dirty_i
->victim_secmap
);
549 * Should not occur error such as -ENOMEM.
550 * Adding dirty entry into seglist is not critical operation.
551 * If a given segment is one of current working segments, it won't be added.
553 static void locate_dirty_segment(struct f2fs_sb_info
*sbi
, unsigned int segno
)
555 struct dirty_seglist_info
*dirty_i
= DIRTY_I(sbi
);
556 unsigned short valid_blocks
;
558 if (segno
== NULL_SEGNO
|| IS_CURSEG(sbi
, segno
))
561 mutex_lock(&dirty_i
->seglist_lock
);
563 valid_blocks
= get_valid_blocks(sbi
, segno
, 0);
565 if (valid_blocks
== 0) {
566 __locate_dirty_segment(sbi
, segno
, PRE
);
567 __remove_dirty_segment(sbi
, segno
, DIRTY
);
568 } else if (valid_blocks
< sbi
->blocks_per_seg
) {
569 __locate_dirty_segment(sbi
, segno
, DIRTY
);
571 /* Recovery routine with SSR needs this */
572 __remove_dirty_segment(sbi
, segno
, DIRTY
);
575 mutex_unlock(&dirty_i
->seglist_lock
);
578 static int f2fs_issue_discard(struct f2fs_sb_info
*sbi
,
579 block_t blkstart
, block_t blklen
)
581 sector_t start
= SECTOR_FROM_BLOCK(blkstart
);
582 sector_t len
= SECTOR_FROM_BLOCK(blklen
);
583 struct seg_entry
*se
;
587 for (i
= blkstart
; i
< blkstart
+ blklen
; i
++) {
588 se
= get_seg_entry(sbi
, GET_SEGNO(sbi
, i
));
589 offset
= GET_BLKOFF_FROM_SEG0(sbi
, i
);
591 if (!f2fs_test_and_set_bit(offset
, se
->discard_map
))
594 trace_f2fs_issue_discard(sbi
->sb
, blkstart
, blklen
);
595 return blkdev_issue_discard(sbi
->sb
->s_bdev
, start
, len
, GFP_NOFS
, 0);
598 bool discard_next_dnode(struct f2fs_sb_info
*sbi
, block_t blkaddr
)
600 int err
= -EOPNOTSUPP
;
602 if (test_opt(sbi
, DISCARD
)) {
603 struct seg_entry
*se
= get_seg_entry(sbi
,
604 GET_SEGNO(sbi
, blkaddr
));
605 unsigned int offset
= GET_BLKOFF_FROM_SEG0(sbi
, blkaddr
);
607 if (f2fs_test_bit(offset
, se
->discard_map
))
610 err
= f2fs_issue_discard(sbi
, blkaddr
, 1);
614 update_meta_page(sbi
, NULL
, blkaddr
);
620 static void __add_discard_entry(struct f2fs_sb_info
*sbi
,
621 struct cp_control
*cpc
, struct seg_entry
*se
,
622 unsigned int start
, unsigned int end
)
624 struct list_head
*head
= &SM_I(sbi
)->discard_list
;
625 struct discard_entry
*new, *last
;
627 if (!list_empty(head
)) {
628 last
= list_last_entry(head
, struct discard_entry
, list
);
629 if (START_BLOCK(sbi
, cpc
->trim_start
) + start
==
630 last
->blkaddr
+ last
->len
) {
631 last
->len
+= end
- start
;
636 new = f2fs_kmem_cache_alloc(discard_entry_slab
, GFP_NOFS
);
637 INIT_LIST_HEAD(&new->list
);
638 new->blkaddr
= START_BLOCK(sbi
, cpc
->trim_start
) + start
;
639 new->len
= end
- start
;
640 list_add_tail(&new->list
, head
);
642 SM_I(sbi
)->nr_discards
+= end
- start
;
645 static void add_discard_addrs(struct f2fs_sb_info
*sbi
, struct cp_control
*cpc
)
647 int entries
= SIT_VBLOCK_MAP_SIZE
/ sizeof(unsigned long);
648 int max_blocks
= sbi
->blocks_per_seg
;
649 struct seg_entry
*se
= get_seg_entry(sbi
, cpc
->trim_start
);
650 unsigned long *cur_map
= (unsigned long *)se
->cur_valid_map
;
651 unsigned long *ckpt_map
= (unsigned long *)se
->ckpt_valid_map
;
652 unsigned long *discard_map
= (unsigned long *)se
->discard_map
;
653 unsigned long *dmap
= SIT_I(sbi
)->tmp_map
;
654 unsigned int start
= 0, end
= -1;
655 bool force
= (cpc
->reason
== CP_DISCARD
);
658 if (se
->valid_blocks
== max_blocks
)
662 if (!test_opt(sbi
, DISCARD
) || !se
->valid_blocks
||
663 SM_I(sbi
)->nr_discards
>= SM_I(sbi
)->max_discards
)
667 /* SIT_VBLOCK_MAP_SIZE should be multiple of sizeof(unsigned long) */
668 for (i
= 0; i
< entries
; i
++)
669 dmap
[i
] = force
? ~ckpt_map
[i
] & ~discard_map
[i
] :
670 (cur_map
[i
] ^ ckpt_map
[i
]) & ckpt_map
[i
];
672 while (force
|| SM_I(sbi
)->nr_discards
<= SM_I(sbi
)->max_discards
) {
673 start
= __find_rev_next_bit(dmap
, max_blocks
, end
+ 1);
674 if (start
>= max_blocks
)
677 end
= __find_rev_next_zero_bit(dmap
, max_blocks
, start
+ 1);
678 __add_discard_entry(sbi
, cpc
, se
, start
, end
);
682 void release_discard_addrs(struct f2fs_sb_info
*sbi
)
684 struct list_head
*head
= &(SM_I(sbi
)->discard_list
);
685 struct discard_entry
*entry
, *this;
688 list_for_each_entry_safe(entry
, this, head
, list
) {
689 list_del(&entry
->list
);
690 kmem_cache_free(discard_entry_slab
, entry
);
695 * Should call clear_prefree_segments after checkpoint is done.
697 static void set_prefree_as_free_segments(struct f2fs_sb_info
*sbi
)
699 struct dirty_seglist_info
*dirty_i
= DIRTY_I(sbi
);
702 mutex_lock(&dirty_i
->seglist_lock
);
703 for_each_set_bit(segno
, dirty_i
->dirty_segmap
[PRE
], MAIN_SEGS(sbi
))
704 __set_test_and_free(sbi
, segno
);
705 mutex_unlock(&dirty_i
->seglist_lock
);
708 void clear_prefree_segments(struct f2fs_sb_info
*sbi
, struct cp_control
*cpc
)
710 struct list_head
*head
= &(SM_I(sbi
)->discard_list
);
711 struct discard_entry
*entry
, *this;
712 struct dirty_seglist_info
*dirty_i
= DIRTY_I(sbi
);
713 unsigned long *prefree_map
= dirty_i
->dirty_segmap
[PRE
];
714 unsigned int start
= 0, end
= -1;
716 mutex_lock(&dirty_i
->seglist_lock
);
720 start
= find_next_bit(prefree_map
, MAIN_SEGS(sbi
), end
+ 1);
721 if (start
>= MAIN_SEGS(sbi
))
723 end
= find_next_zero_bit(prefree_map
, MAIN_SEGS(sbi
),
726 for (i
= start
; i
< end
; i
++)
727 clear_bit(i
, prefree_map
);
729 dirty_i
->nr_dirty
[PRE
] -= end
- start
;
731 if (!test_opt(sbi
, DISCARD
))
734 f2fs_issue_discard(sbi
, START_BLOCK(sbi
, start
),
735 (end
- start
) << sbi
->log_blocks_per_seg
);
737 mutex_unlock(&dirty_i
->seglist_lock
);
739 /* send small discards */
740 list_for_each_entry_safe(entry
, this, head
, list
) {
741 if (cpc
->reason
== CP_DISCARD
&& entry
->len
< cpc
->trim_minlen
)
743 f2fs_issue_discard(sbi
, entry
->blkaddr
, entry
->len
);
744 cpc
->trimmed
+= entry
->len
;
746 list_del(&entry
->list
);
747 SM_I(sbi
)->nr_discards
-= entry
->len
;
748 kmem_cache_free(discard_entry_slab
, entry
);
752 static bool __mark_sit_entry_dirty(struct f2fs_sb_info
*sbi
, unsigned int segno
)
754 struct sit_info
*sit_i
= SIT_I(sbi
);
756 if (!__test_and_set_bit(segno
, sit_i
->dirty_sentries_bitmap
)) {
757 sit_i
->dirty_sentries
++;
764 static void __set_sit_entry_type(struct f2fs_sb_info
*sbi
, int type
,
765 unsigned int segno
, int modified
)
767 struct seg_entry
*se
= get_seg_entry(sbi
, segno
);
770 __mark_sit_entry_dirty(sbi
, segno
);
773 static void update_sit_entry(struct f2fs_sb_info
*sbi
, block_t blkaddr
, int del
)
775 struct seg_entry
*se
;
776 unsigned int segno
, offset
;
777 long int new_vblocks
;
779 segno
= GET_SEGNO(sbi
, blkaddr
);
781 se
= get_seg_entry(sbi
, segno
);
782 new_vblocks
= se
->valid_blocks
+ del
;
783 offset
= GET_BLKOFF_FROM_SEG0(sbi
, blkaddr
);
785 f2fs_bug_on(sbi
, (new_vblocks
>> (sizeof(unsigned short) << 3) ||
786 (new_vblocks
> sbi
->blocks_per_seg
)));
788 se
->valid_blocks
= new_vblocks
;
789 se
->mtime
= get_mtime(sbi
);
790 SIT_I(sbi
)->max_mtime
= se
->mtime
;
792 /* Update valid block bitmap */
794 if (f2fs_test_and_set_bit(offset
, se
->cur_valid_map
))
796 if (!f2fs_test_and_set_bit(offset
, se
->discard_map
))
799 if (!f2fs_test_and_clear_bit(offset
, se
->cur_valid_map
))
801 if (f2fs_test_and_clear_bit(offset
, se
->discard_map
))
804 if (!f2fs_test_bit(offset
, se
->ckpt_valid_map
))
805 se
->ckpt_valid_blocks
+= del
;
807 __mark_sit_entry_dirty(sbi
, segno
);
809 /* update total number of valid blocks to be written in ckpt area */
810 SIT_I(sbi
)->written_valid_blocks
+= del
;
812 if (sbi
->segs_per_sec
> 1)
813 get_sec_entry(sbi
, segno
)->valid_blocks
+= del
;
816 void refresh_sit_entry(struct f2fs_sb_info
*sbi
, block_t old
, block_t
new)
818 update_sit_entry(sbi
, new, 1);
819 if (GET_SEGNO(sbi
, old
) != NULL_SEGNO
)
820 update_sit_entry(sbi
, old
, -1);
822 locate_dirty_segment(sbi
, GET_SEGNO(sbi
, old
));
823 locate_dirty_segment(sbi
, GET_SEGNO(sbi
, new));
826 void invalidate_blocks(struct f2fs_sb_info
*sbi
, block_t addr
)
828 unsigned int segno
= GET_SEGNO(sbi
, addr
);
829 struct sit_info
*sit_i
= SIT_I(sbi
);
831 f2fs_bug_on(sbi
, addr
== NULL_ADDR
);
832 if (addr
== NEW_ADDR
)
835 /* add it into sit main buffer */
836 mutex_lock(&sit_i
->sentry_lock
);
838 update_sit_entry(sbi
, addr
, -1);
840 /* add it into dirty seglist */
841 locate_dirty_segment(sbi
, segno
);
843 mutex_unlock(&sit_i
->sentry_lock
);
846 bool is_checkpointed_data(struct f2fs_sb_info
*sbi
, block_t blkaddr
)
848 struct sit_info
*sit_i
= SIT_I(sbi
);
849 unsigned int segno
, offset
;
850 struct seg_entry
*se
;
853 if (blkaddr
== NEW_ADDR
|| blkaddr
== NULL_ADDR
)
856 mutex_lock(&sit_i
->sentry_lock
);
858 segno
= GET_SEGNO(sbi
, blkaddr
);
859 se
= get_seg_entry(sbi
, segno
);
860 offset
= GET_BLKOFF_FROM_SEG0(sbi
, blkaddr
);
862 if (f2fs_test_bit(offset
, se
->ckpt_valid_map
))
865 mutex_unlock(&sit_i
->sentry_lock
);
871 * This function should be resided under the curseg_mutex lock
873 static void __add_sum_entry(struct f2fs_sb_info
*sbi
, int type
,
874 struct f2fs_summary
*sum
)
876 struct curseg_info
*curseg
= CURSEG_I(sbi
, type
);
877 void *addr
= curseg
->sum_blk
;
878 addr
+= curseg
->next_blkoff
* sizeof(struct f2fs_summary
);
879 memcpy(addr
, sum
, sizeof(struct f2fs_summary
));
883 * Calculate the number of current summary pages for writing
885 int npages_for_summary_flush(struct f2fs_sb_info
*sbi
, bool for_ra
)
887 int valid_sum_count
= 0;
890 for (i
= CURSEG_HOT_DATA
; i
<= CURSEG_COLD_DATA
; i
++) {
891 if (sbi
->ckpt
->alloc_type
[i
] == SSR
)
892 valid_sum_count
+= sbi
->blocks_per_seg
;
895 valid_sum_count
+= le16_to_cpu(
896 F2FS_CKPT(sbi
)->cur_data_blkoff
[i
]);
898 valid_sum_count
+= curseg_blkoff(sbi
, i
);
902 sum_in_page
= (PAGE_SIZE
- 2 * SUM_JOURNAL_SIZE
-
903 SUM_FOOTER_SIZE
) / SUMMARY_SIZE
;
904 if (valid_sum_count
<= sum_in_page
)
906 else if ((valid_sum_count
- sum_in_page
) <=
907 (PAGE_SIZE
- SUM_FOOTER_SIZE
) / SUMMARY_SIZE
)
913 * Caller should put this summary page
915 struct page
*get_sum_page(struct f2fs_sb_info
*sbi
, unsigned int segno
)
917 return get_meta_page(sbi
, GET_SUM_BLOCK(sbi
, segno
));
920 void update_meta_page(struct f2fs_sb_info
*sbi
, void *src
, block_t blk_addr
)
922 struct page
*page
= grab_meta_page(sbi
, blk_addr
);
923 void *dst
= page_address(page
);
926 memcpy(dst
, src
, PAGE_SIZE
);
928 memset(dst
, 0, PAGE_SIZE
);
929 set_page_dirty(page
);
930 f2fs_put_page(page
, 1);
933 static void write_sum_page(struct f2fs_sb_info
*sbi
,
934 struct f2fs_summary_block
*sum_blk
, block_t blk_addr
)
936 update_meta_page(sbi
, (void *)sum_blk
, blk_addr
);
939 static void write_current_sum_page(struct f2fs_sb_info
*sbi
,
940 int type
, block_t blk_addr
)
942 struct curseg_info
*curseg
= CURSEG_I(sbi
, type
);
943 struct page
*page
= grab_meta_page(sbi
, blk_addr
);
944 struct f2fs_summary_block
*src
= curseg
->sum_blk
;
945 struct f2fs_summary_block
*dst
;
947 dst
= (struct f2fs_summary_block
*)page_address(page
);
949 mutex_lock(&curseg
->curseg_mutex
);
951 down_read(&curseg
->journal_rwsem
);
952 memcpy(&dst
->journal
, curseg
->journal
, SUM_JOURNAL_SIZE
);
953 up_read(&curseg
->journal_rwsem
);
955 memcpy(dst
->entries
, src
->entries
, SUM_ENTRY_SIZE
);
956 memcpy(&dst
->footer
, &src
->footer
, SUM_FOOTER_SIZE
);
958 mutex_unlock(&curseg
->curseg_mutex
);
960 set_page_dirty(page
);
961 f2fs_put_page(page
, 1);
964 static int is_next_segment_free(struct f2fs_sb_info
*sbi
, int type
)
966 struct curseg_info
*curseg
= CURSEG_I(sbi
, type
);
967 unsigned int segno
= curseg
->segno
+ 1;
968 struct free_segmap_info
*free_i
= FREE_I(sbi
);
970 if (segno
< MAIN_SEGS(sbi
) && segno
% sbi
->segs_per_sec
)
971 return !test_bit(segno
, free_i
->free_segmap
);
976 * Find a new segment from the free segments bitmap to right order
977 * This function should be returned with success, otherwise BUG
979 static void get_new_segment(struct f2fs_sb_info
*sbi
,
980 unsigned int *newseg
, bool new_sec
, int dir
)
982 struct free_segmap_info
*free_i
= FREE_I(sbi
);
983 unsigned int segno
, secno
, zoneno
;
984 unsigned int total_zones
= MAIN_SECS(sbi
) / sbi
->secs_per_zone
;
985 unsigned int hint
= *newseg
/ sbi
->segs_per_sec
;
986 unsigned int old_zoneno
= GET_ZONENO_FROM_SEGNO(sbi
, *newseg
);
987 unsigned int left_start
= hint
;
992 spin_lock(&free_i
->segmap_lock
);
994 if (!new_sec
&& ((*newseg
+ 1) % sbi
->segs_per_sec
)) {
995 segno
= find_next_zero_bit(free_i
->free_segmap
,
996 (hint
+ 1) * sbi
->segs_per_sec
, *newseg
+ 1);
997 if (segno
< (hint
+ 1) * sbi
->segs_per_sec
)
1001 secno
= find_next_zero_bit(free_i
->free_secmap
, MAIN_SECS(sbi
), hint
);
1002 if (secno
>= MAIN_SECS(sbi
)) {
1003 if (dir
== ALLOC_RIGHT
) {
1004 secno
= find_next_zero_bit(free_i
->free_secmap
,
1006 f2fs_bug_on(sbi
, secno
>= MAIN_SECS(sbi
));
1009 left_start
= hint
- 1;
1015 while (test_bit(left_start
, free_i
->free_secmap
)) {
1016 if (left_start
> 0) {
1020 left_start
= find_next_zero_bit(free_i
->free_secmap
,
1022 f2fs_bug_on(sbi
, left_start
>= MAIN_SECS(sbi
));
1028 segno
= secno
* sbi
->segs_per_sec
;
1029 zoneno
= secno
/ sbi
->secs_per_zone
;
1031 /* give up on finding another zone */
1034 if (sbi
->secs_per_zone
== 1)
1036 if (zoneno
== old_zoneno
)
1038 if (dir
== ALLOC_LEFT
) {
1039 if (!go_left
&& zoneno
+ 1 >= total_zones
)
1041 if (go_left
&& zoneno
== 0)
1044 for (i
= 0; i
< NR_CURSEG_TYPE
; i
++)
1045 if (CURSEG_I(sbi
, i
)->zone
== zoneno
)
1048 if (i
< NR_CURSEG_TYPE
) {
1049 /* zone is in user, try another */
1051 hint
= zoneno
* sbi
->secs_per_zone
- 1;
1052 else if (zoneno
+ 1 >= total_zones
)
1055 hint
= (zoneno
+ 1) * sbi
->secs_per_zone
;
1057 goto find_other_zone
;
1060 /* set it as dirty segment in free segmap */
1061 f2fs_bug_on(sbi
, test_bit(segno
, free_i
->free_segmap
));
1062 __set_inuse(sbi
, segno
);
1064 spin_unlock(&free_i
->segmap_lock
);
1067 static void reset_curseg(struct f2fs_sb_info
*sbi
, int type
, int modified
)
1069 struct curseg_info
*curseg
= CURSEG_I(sbi
, type
);
1070 struct summary_footer
*sum_footer
;
1072 curseg
->segno
= curseg
->next_segno
;
1073 curseg
->zone
= GET_ZONENO_FROM_SEGNO(sbi
, curseg
->segno
);
1074 curseg
->next_blkoff
= 0;
1075 curseg
->next_segno
= NULL_SEGNO
;
1077 sum_footer
= &(curseg
->sum_blk
->footer
);
1078 memset(sum_footer
, 0, sizeof(struct summary_footer
));
1079 if (IS_DATASEG(type
))
1080 SET_SUM_TYPE(sum_footer
, SUM_TYPE_DATA
);
1081 if (IS_NODESEG(type
))
1082 SET_SUM_TYPE(sum_footer
, SUM_TYPE_NODE
);
1083 __set_sit_entry_type(sbi
, type
, curseg
->segno
, modified
);
1087 * Allocate a current working segment.
1088 * This function always allocates a free segment in LFS manner.
1090 static void new_curseg(struct f2fs_sb_info
*sbi
, int type
, bool new_sec
)
1092 struct curseg_info
*curseg
= CURSEG_I(sbi
, type
);
1093 unsigned int segno
= curseg
->segno
;
1094 int dir
= ALLOC_LEFT
;
1096 write_sum_page(sbi
, curseg
->sum_blk
,
1097 GET_SUM_BLOCK(sbi
, segno
));
1098 if (type
== CURSEG_WARM_DATA
|| type
== CURSEG_COLD_DATA
)
1101 if (test_opt(sbi
, NOHEAP
))
1104 get_new_segment(sbi
, &segno
, new_sec
, dir
);
1105 curseg
->next_segno
= segno
;
1106 reset_curseg(sbi
, type
, 1);
1107 curseg
->alloc_type
= LFS
;
1110 static void __next_free_blkoff(struct f2fs_sb_info
*sbi
,
1111 struct curseg_info
*seg
, block_t start
)
1113 struct seg_entry
*se
= get_seg_entry(sbi
, seg
->segno
);
1114 int entries
= SIT_VBLOCK_MAP_SIZE
/ sizeof(unsigned long);
1115 unsigned long *target_map
= SIT_I(sbi
)->tmp_map
;
1116 unsigned long *ckpt_map
= (unsigned long *)se
->ckpt_valid_map
;
1117 unsigned long *cur_map
= (unsigned long *)se
->cur_valid_map
;
1120 for (i
= 0; i
< entries
; i
++)
1121 target_map
[i
] = ckpt_map
[i
] | cur_map
[i
];
1123 pos
= __find_rev_next_zero_bit(target_map
, sbi
->blocks_per_seg
, start
);
1125 seg
->next_blkoff
= pos
;
1129 * If a segment is written by LFS manner, next block offset is just obtained
1130 * by increasing the current block offset. However, if a segment is written by
1131 * SSR manner, next block offset obtained by calling __next_free_blkoff
1133 static void __refresh_next_blkoff(struct f2fs_sb_info
*sbi
,
1134 struct curseg_info
*seg
)
1136 if (seg
->alloc_type
== SSR
)
1137 __next_free_blkoff(sbi
, seg
, seg
->next_blkoff
+ 1);
1143 * This function always allocates a used segment(from dirty seglist) by SSR
1144 * manner, so it should recover the existing segment information of valid blocks
1146 static void change_curseg(struct f2fs_sb_info
*sbi
, int type
, bool reuse
)
1148 struct dirty_seglist_info
*dirty_i
= DIRTY_I(sbi
);
1149 struct curseg_info
*curseg
= CURSEG_I(sbi
, type
);
1150 unsigned int new_segno
= curseg
->next_segno
;
1151 struct f2fs_summary_block
*sum_node
;
1152 struct page
*sum_page
;
1154 write_sum_page(sbi
, curseg
->sum_blk
,
1155 GET_SUM_BLOCK(sbi
, curseg
->segno
));
1156 __set_test_and_inuse(sbi
, new_segno
);
1158 mutex_lock(&dirty_i
->seglist_lock
);
1159 __remove_dirty_segment(sbi
, new_segno
, PRE
);
1160 __remove_dirty_segment(sbi
, new_segno
, DIRTY
);
1161 mutex_unlock(&dirty_i
->seglist_lock
);
1163 reset_curseg(sbi
, type
, 1);
1164 curseg
->alloc_type
= SSR
;
1165 __next_free_blkoff(sbi
, curseg
, 0);
1168 sum_page
= get_sum_page(sbi
, new_segno
);
1169 sum_node
= (struct f2fs_summary_block
*)page_address(sum_page
);
1170 memcpy(curseg
->sum_blk
, sum_node
, SUM_ENTRY_SIZE
);
1171 f2fs_put_page(sum_page
, 1);
1175 static int get_ssr_segment(struct f2fs_sb_info
*sbi
, int type
)
1177 struct curseg_info
*curseg
= CURSEG_I(sbi
, type
);
1178 const struct victim_selection
*v_ops
= DIRTY_I(sbi
)->v_ops
;
1180 if (IS_NODESEG(type
) || !has_not_enough_free_secs(sbi
, 0))
1181 return v_ops
->get_victim(sbi
,
1182 &(curseg
)->next_segno
, BG_GC
, type
, SSR
);
1184 /* For data segments, let's do SSR more intensively */
1185 for (; type
>= CURSEG_HOT_DATA
; type
--)
1186 if (v_ops
->get_victim(sbi
, &(curseg
)->next_segno
,
1193 * flush out current segment and replace it with new segment
1194 * This function should be returned with success, otherwise BUG
1196 static void allocate_segment_by_default(struct f2fs_sb_info
*sbi
,
1197 int type
, bool force
)
1199 struct curseg_info
*curseg
= CURSEG_I(sbi
, type
);
1202 new_curseg(sbi
, type
, true);
1203 else if (type
== CURSEG_WARM_NODE
)
1204 new_curseg(sbi
, type
, false);
1205 else if (curseg
->alloc_type
== LFS
&& is_next_segment_free(sbi
, type
))
1206 new_curseg(sbi
, type
, false);
1207 else if (need_SSR(sbi
) && get_ssr_segment(sbi
, type
))
1208 change_curseg(sbi
, type
, true);
1210 new_curseg(sbi
, type
, false);
1212 stat_inc_seg_type(sbi
, curseg
);
1215 static void __allocate_new_segments(struct f2fs_sb_info
*sbi
, int type
)
1217 struct curseg_info
*curseg
= CURSEG_I(sbi
, type
);
1218 unsigned int old_segno
;
1220 old_segno
= curseg
->segno
;
1221 SIT_I(sbi
)->s_ops
->allocate_segment(sbi
, type
, true);
1222 locate_dirty_segment(sbi
, old_segno
);
1225 void allocate_new_segments(struct f2fs_sb_info
*sbi
)
1229 for (i
= CURSEG_HOT_DATA
; i
<= CURSEG_COLD_DATA
; i
++)
1230 __allocate_new_segments(sbi
, i
);
1233 static const struct segment_allocation default_salloc_ops
= {
1234 .allocate_segment
= allocate_segment_by_default
,
1237 int f2fs_trim_fs(struct f2fs_sb_info
*sbi
, struct fstrim_range
*range
)
1239 __u64 start
= F2FS_BYTES_TO_BLK(range
->start
);
1240 __u64 end
= start
+ F2FS_BYTES_TO_BLK(range
->len
) - 1;
1241 unsigned int start_segno
, end_segno
;
1242 struct cp_control cpc
;
1245 if (start
>= MAX_BLKADDR(sbi
) || range
->len
< sbi
->blocksize
)
1249 if (end
<= MAIN_BLKADDR(sbi
))
1252 /* start/end segment number in main_area */
1253 start_segno
= (start
<= MAIN_BLKADDR(sbi
)) ? 0 : GET_SEGNO(sbi
, start
);
1254 end_segno
= (end
>= MAX_BLKADDR(sbi
)) ? MAIN_SEGS(sbi
) - 1 :
1255 GET_SEGNO(sbi
, end
);
1256 cpc
.reason
= CP_DISCARD
;
1257 cpc
.trim_minlen
= max_t(__u64
, 1, F2FS_BYTES_TO_BLK(range
->minlen
));
1259 /* do checkpoint to issue discard commands safely */
1260 for (; start_segno
<= end_segno
; start_segno
= cpc
.trim_end
+ 1) {
1261 cpc
.trim_start
= start_segno
;
1263 if (sbi
->discard_blks
== 0)
1265 else if (sbi
->discard_blks
< BATCHED_TRIM_BLOCKS(sbi
))
1266 cpc
.trim_end
= end_segno
;
1268 cpc
.trim_end
= min_t(unsigned int,
1269 rounddown(start_segno
+
1270 BATCHED_TRIM_SEGMENTS(sbi
),
1271 sbi
->segs_per_sec
) - 1, end_segno
);
1273 mutex_lock(&sbi
->gc_mutex
);
1274 err
= write_checkpoint(sbi
, &cpc
);
1275 mutex_unlock(&sbi
->gc_mutex
);
1278 range
->len
= F2FS_BLK_TO_BYTES(cpc
.trimmed
);
1282 static bool __has_curseg_space(struct f2fs_sb_info
*sbi
, int type
)
1284 struct curseg_info
*curseg
= CURSEG_I(sbi
, type
);
1285 if (curseg
->next_blkoff
< sbi
->blocks_per_seg
)
1290 static int __get_segment_type_2(struct page
*page
, enum page_type p_type
)
1293 return CURSEG_HOT_DATA
;
1295 return CURSEG_HOT_NODE
;
1298 static int __get_segment_type_4(struct page
*page
, enum page_type p_type
)
1300 if (p_type
== DATA
) {
1301 struct inode
*inode
= page
->mapping
->host
;
1303 if (S_ISDIR(inode
->i_mode
))
1304 return CURSEG_HOT_DATA
;
1306 return CURSEG_COLD_DATA
;
1308 if (IS_DNODE(page
) && is_cold_node(page
))
1309 return CURSEG_WARM_NODE
;
1311 return CURSEG_COLD_NODE
;
1315 static int __get_segment_type_6(struct page
*page
, enum page_type p_type
)
1317 if (p_type
== DATA
) {
1318 struct inode
*inode
= page
->mapping
->host
;
1320 if (S_ISDIR(inode
->i_mode
))
1321 return CURSEG_HOT_DATA
;
1322 else if (is_cold_data(page
) || file_is_cold(inode
))
1323 return CURSEG_COLD_DATA
;
1325 return CURSEG_WARM_DATA
;
1328 return is_cold_node(page
) ? CURSEG_WARM_NODE
:
1331 return CURSEG_COLD_NODE
;
1335 static int __get_segment_type(struct page
*page
, enum page_type p_type
)
1337 switch (F2FS_P_SB(page
)->active_logs
) {
1339 return __get_segment_type_2(page
, p_type
);
1341 return __get_segment_type_4(page
, p_type
);
1343 /* NR_CURSEG_TYPE(6) logs by default */
1344 f2fs_bug_on(F2FS_P_SB(page
),
1345 F2FS_P_SB(page
)->active_logs
!= NR_CURSEG_TYPE
);
1346 return __get_segment_type_6(page
, p_type
);
1349 void allocate_data_block(struct f2fs_sb_info
*sbi
, struct page
*page
,
1350 block_t old_blkaddr
, block_t
*new_blkaddr
,
1351 struct f2fs_summary
*sum
, int type
)
1353 struct sit_info
*sit_i
= SIT_I(sbi
);
1354 struct curseg_info
*curseg
;
1355 bool direct_io
= (type
== CURSEG_DIRECT_IO
);
1357 type
= direct_io
? CURSEG_WARM_DATA
: type
;
1359 curseg
= CURSEG_I(sbi
, type
);
1361 mutex_lock(&curseg
->curseg_mutex
);
1362 mutex_lock(&sit_i
->sentry_lock
);
1364 /* direct_io'ed data is aligned to the segment for better performance */
1365 if (direct_io
&& curseg
->next_blkoff
&&
1366 !has_not_enough_free_secs(sbi
, 0))
1367 __allocate_new_segments(sbi
, type
);
1369 *new_blkaddr
= NEXT_FREE_BLKADDR(sbi
, curseg
);
1372 * __add_sum_entry should be resided under the curseg_mutex
1373 * because, this function updates a summary entry in the
1374 * current summary block.
1376 __add_sum_entry(sbi
, type
, sum
);
1378 __refresh_next_blkoff(sbi
, curseg
);
1380 stat_inc_block_count(sbi
, curseg
);
1382 if (!__has_curseg_space(sbi
, type
))
1383 sit_i
->s_ops
->allocate_segment(sbi
, type
, false);
1385 * SIT information should be updated before segment allocation,
1386 * since SSR needs latest valid block information.
1388 refresh_sit_entry(sbi
, old_blkaddr
, *new_blkaddr
);
1390 mutex_unlock(&sit_i
->sentry_lock
);
1392 if (page
&& IS_NODESEG(type
))
1393 fill_node_footer_blkaddr(page
, NEXT_FREE_BLKADDR(sbi
, curseg
));
1395 mutex_unlock(&curseg
->curseg_mutex
);
1398 static void do_write_page(struct f2fs_summary
*sum
, struct f2fs_io_info
*fio
)
1400 int type
= __get_segment_type(fio
->page
, fio
->type
);
1402 if (fio
->type
== NODE
|| fio
->type
== DATA
)
1403 mutex_lock(&fio
->sbi
->wio_mutex
[fio
->type
]);
1405 allocate_data_block(fio
->sbi
, fio
->page
, fio
->old_blkaddr
,
1406 &fio
->new_blkaddr
, sum
, type
);
1408 /* writeout dirty page into bdev */
1409 f2fs_submit_page_mbio(fio
);
1411 if (fio
->type
== NODE
|| fio
->type
== DATA
)
1412 mutex_unlock(&fio
->sbi
->wio_mutex
[fio
->type
]);
1415 void write_meta_page(struct f2fs_sb_info
*sbi
, struct page
*page
)
1417 struct f2fs_io_info fio
= {
1420 .rw
= WRITE_SYNC
| REQ_META
| REQ_PRIO
,
1421 .old_blkaddr
= page
->index
,
1422 .new_blkaddr
= page
->index
,
1424 .encrypted_page
= NULL
,
1427 if (unlikely(page
->index
>= MAIN_BLKADDR(sbi
)))
1428 fio
.rw
&= ~REQ_META
;
1430 set_page_writeback(page
);
1431 f2fs_submit_page_mbio(&fio
);
1434 void write_node_page(unsigned int nid
, struct f2fs_io_info
*fio
)
1436 struct f2fs_summary sum
;
1438 set_summary(&sum
, nid
, 0, 0);
1439 do_write_page(&sum
, fio
);
1442 void write_data_page(struct dnode_of_data
*dn
, struct f2fs_io_info
*fio
)
1444 struct f2fs_sb_info
*sbi
= fio
->sbi
;
1445 struct f2fs_summary sum
;
1446 struct node_info ni
;
1448 f2fs_bug_on(sbi
, dn
->data_blkaddr
== NULL_ADDR
);
1449 get_node_info(sbi
, dn
->nid
, &ni
);
1450 set_summary(&sum
, dn
->nid
, dn
->ofs_in_node
, ni
.version
);
1451 do_write_page(&sum
, fio
);
1452 f2fs_update_data_blkaddr(dn
, fio
->new_blkaddr
);
1455 void rewrite_data_page(struct f2fs_io_info
*fio
)
1457 fio
->new_blkaddr
= fio
->old_blkaddr
;
1458 stat_inc_inplace_blocks(fio
->sbi
);
1459 f2fs_submit_page_mbio(fio
);
1462 void __f2fs_replace_block(struct f2fs_sb_info
*sbi
, struct f2fs_summary
*sum
,
1463 block_t old_blkaddr
, block_t new_blkaddr
,
1464 bool recover_curseg
, bool recover_newaddr
)
1466 struct sit_info
*sit_i
= SIT_I(sbi
);
1467 struct curseg_info
*curseg
;
1468 unsigned int segno
, old_cursegno
;
1469 struct seg_entry
*se
;
1471 unsigned short old_blkoff
;
1473 segno
= GET_SEGNO(sbi
, new_blkaddr
);
1474 se
= get_seg_entry(sbi
, segno
);
1477 if (!recover_curseg
) {
1478 /* for recovery flow */
1479 if (se
->valid_blocks
== 0 && !IS_CURSEG(sbi
, segno
)) {
1480 if (old_blkaddr
== NULL_ADDR
)
1481 type
= CURSEG_COLD_DATA
;
1483 type
= CURSEG_WARM_DATA
;
1486 if (!IS_CURSEG(sbi
, segno
))
1487 type
= CURSEG_WARM_DATA
;
1490 curseg
= CURSEG_I(sbi
, type
);
1492 mutex_lock(&curseg
->curseg_mutex
);
1493 mutex_lock(&sit_i
->sentry_lock
);
1495 old_cursegno
= curseg
->segno
;
1496 old_blkoff
= curseg
->next_blkoff
;
1498 /* change the current segment */
1499 if (segno
!= curseg
->segno
) {
1500 curseg
->next_segno
= segno
;
1501 change_curseg(sbi
, type
, true);
1504 curseg
->next_blkoff
= GET_BLKOFF_FROM_SEG0(sbi
, new_blkaddr
);
1505 __add_sum_entry(sbi
, type
, sum
);
1507 if (!recover_curseg
|| recover_newaddr
)
1508 update_sit_entry(sbi
, new_blkaddr
, 1);
1509 if (GET_SEGNO(sbi
, old_blkaddr
) != NULL_SEGNO
)
1510 update_sit_entry(sbi
, old_blkaddr
, -1);
1512 locate_dirty_segment(sbi
, GET_SEGNO(sbi
, old_blkaddr
));
1513 locate_dirty_segment(sbi
, GET_SEGNO(sbi
, new_blkaddr
));
1515 locate_dirty_segment(sbi
, old_cursegno
);
1517 if (recover_curseg
) {
1518 if (old_cursegno
!= curseg
->segno
) {
1519 curseg
->next_segno
= old_cursegno
;
1520 change_curseg(sbi
, type
, true);
1522 curseg
->next_blkoff
= old_blkoff
;
1525 mutex_unlock(&sit_i
->sentry_lock
);
1526 mutex_unlock(&curseg
->curseg_mutex
);
1529 void f2fs_replace_block(struct f2fs_sb_info
*sbi
, struct dnode_of_data
*dn
,
1530 block_t old_addr
, block_t new_addr
,
1531 unsigned char version
, bool recover_curseg
,
1532 bool recover_newaddr
)
1534 struct f2fs_summary sum
;
1536 set_summary(&sum
, dn
->nid
, dn
->ofs_in_node
, version
);
1538 __f2fs_replace_block(sbi
, &sum
, old_addr
, new_addr
,
1539 recover_curseg
, recover_newaddr
);
1541 f2fs_update_data_blkaddr(dn
, new_addr
);
1544 void f2fs_wait_on_page_writeback(struct page
*page
,
1545 enum page_type type
, bool ordered
)
1547 if (PageWriteback(page
)) {
1548 struct f2fs_sb_info
*sbi
= F2FS_P_SB(page
);
1550 f2fs_submit_merged_bio_cond(sbi
, NULL
, page
, 0, type
, WRITE
);
1552 wait_on_page_writeback(page
);
1554 wait_for_stable_page(page
);
1558 void f2fs_wait_on_encrypted_page_writeback(struct f2fs_sb_info
*sbi
,
1563 if (blkaddr
== NEW_ADDR
)
1566 f2fs_bug_on(sbi
, blkaddr
== NULL_ADDR
);
1568 cpage
= find_lock_page(META_MAPPING(sbi
), blkaddr
);
1570 f2fs_wait_on_page_writeback(cpage
, DATA
, true);
1571 f2fs_put_page(cpage
, 1);
1575 static int read_compacted_summaries(struct f2fs_sb_info
*sbi
)
1577 struct f2fs_checkpoint
*ckpt
= F2FS_CKPT(sbi
);
1578 struct curseg_info
*seg_i
;
1579 unsigned char *kaddr
;
1584 start
= start_sum_block(sbi
);
1586 page
= get_meta_page(sbi
, start
++);
1587 kaddr
= (unsigned char *)page_address(page
);
1589 /* Step 1: restore nat cache */
1590 seg_i
= CURSEG_I(sbi
, CURSEG_HOT_DATA
);
1591 memcpy(seg_i
->journal
, kaddr
, SUM_JOURNAL_SIZE
);
1593 /* Step 2: restore sit cache */
1594 seg_i
= CURSEG_I(sbi
, CURSEG_COLD_DATA
);
1595 memcpy(seg_i
->journal
, kaddr
+ SUM_JOURNAL_SIZE
, SUM_JOURNAL_SIZE
);
1596 offset
= 2 * SUM_JOURNAL_SIZE
;
1598 /* Step 3: restore summary entries */
1599 for (i
= CURSEG_HOT_DATA
; i
<= CURSEG_COLD_DATA
; i
++) {
1600 unsigned short blk_off
;
1603 seg_i
= CURSEG_I(sbi
, i
);
1604 segno
= le32_to_cpu(ckpt
->cur_data_segno
[i
]);
1605 blk_off
= le16_to_cpu(ckpt
->cur_data_blkoff
[i
]);
1606 seg_i
->next_segno
= segno
;
1607 reset_curseg(sbi
, i
, 0);
1608 seg_i
->alloc_type
= ckpt
->alloc_type
[i
];
1609 seg_i
->next_blkoff
= blk_off
;
1611 if (seg_i
->alloc_type
== SSR
)
1612 blk_off
= sbi
->blocks_per_seg
;
1614 for (j
= 0; j
< blk_off
; j
++) {
1615 struct f2fs_summary
*s
;
1616 s
= (struct f2fs_summary
*)(kaddr
+ offset
);
1617 seg_i
->sum_blk
->entries
[j
] = *s
;
1618 offset
+= SUMMARY_SIZE
;
1619 if (offset
+ SUMMARY_SIZE
<= PAGE_SIZE
-
1623 f2fs_put_page(page
, 1);
1626 page
= get_meta_page(sbi
, start
++);
1627 kaddr
= (unsigned char *)page_address(page
);
1631 f2fs_put_page(page
, 1);
1635 static int read_normal_summaries(struct f2fs_sb_info
*sbi
, int type
)
1637 struct f2fs_checkpoint
*ckpt
= F2FS_CKPT(sbi
);
1638 struct f2fs_summary_block
*sum
;
1639 struct curseg_info
*curseg
;
1641 unsigned short blk_off
;
1642 unsigned int segno
= 0;
1643 block_t blk_addr
= 0;
1645 /* get segment number and block addr */
1646 if (IS_DATASEG(type
)) {
1647 segno
= le32_to_cpu(ckpt
->cur_data_segno
[type
]);
1648 blk_off
= le16_to_cpu(ckpt
->cur_data_blkoff
[type
-
1650 if (__exist_node_summaries(sbi
))
1651 blk_addr
= sum_blk_addr(sbi
, NR_CURSEG_TYPE
, type
);
1653 blk_addr
= sum_blk_addr(sbi
, NR_CURSEG_DATA_TYPE
, type
);
1655 segno
= le32_to_cpu(ckpt
->cur_node_segno
[type
-
1657 blk_off
= le16_to_cpu(ckpt
->cur_node_blkoff
[type
-
1659 if (__exist_node_summaries(sbi
))
1660 blk_addr
= sum_blk_addr(sbi
, NR_CURSEG_NODE_TYPE
,
1661 type
- CURSEG_HOT_NODE
);
1663 blk_addr
= GET_SUM_BLOCK(sbi
, segno
);
1666 new = get_meta_page(sbi
, blk_addr
);
1667 sum
= (struct f2fs_summary_block
*)page_address(new);
1669 if (IS_NODESEG(type
)) {
1670 if (__exist_node_summaries(sbi
)) {
1671 struct f2fs_summary
*ns
= &sum
->entries
[0];
1673 for (i
= 0; i
< sbi
->blocks_per_seg
; i
++, ns
++) {
1675 ns
->ofs_in_node
= 0;
1680 err
= restore_node_summary(sbi
, segno
, sum
);
1682 f2fs_put_page(new, 1);
1688 /* set uncompleted segment to curseg */
1689 curseg
= CURSEG_I(sbi
, type
);
1690 mutex_lock(&curseg
->curseg_mutex
);
1692 /* update journal info */
1693 down_write(&curseg
->journal_rwsem
);
1694 memcpy(curseg
->journal
, &sum
->journal
, SUM_JOURNAL_SIZE
);
1695 up_write(&curseg
->journal_rwsem
);
1697 memcpy(curseg
->sum_blk
->entries
, sum
->entries
, SUM_ENTRY_SIZE
);
1698 memcpy(&curseg
->sum_blk
->footer
, &sum
->footer
, SUM_FOOTER_SIZE
);
1699 curseg
->next_segno
= segno
;
1700 reset_curseg(sbi
, type
, 0);
1701 curseg
->alloc_type
= ckpt
->alloc_type
[type
];
1702 curseg
->next_blkoff
= blk_off
;
1703 mutex_unlock(&curseg
->curseg_mutex
);
1704 f2fs_put_page(new, 1);
1708 static int restore_curseg_summaries(struct f2fs_sb_info
*sbi
)
1710 int type
= CURSEG_HOT_DATA
;
1713 if (is_set_ckpt_flags(F2FS_CKPT(sbi
), CP_COMPACT_SUM_FLAG
)) {
1714 int npages
= npages_for_summary_flush(sbi
, true);
1717 ra_meta_pages(sbi
, start_sum_block(sbi
), npages
,
1720 /* restore for compacted data summary */
1721 if (read_compacted_summaries(sbi
))
1723 type
= CURSEG_HOT_NODE
;
1726 if (__exist_node_summaries(sbi
))
1727 ra_meta_pages(sbi
, sum_blk_addr(sbi
, NR_CURSEG_TYPE
, type
),
1728 NR_CURSEG_TYPE
- type
, META_CP
, true);
1730 for (; type
<= CURSEG_COLD_NODE
; type
++) {
1731 err
= read_normal_summaries(sbi
, type
);
1739 static void write_compacted_summaries(struct f2fs_sb_info
*sbi
, block_t blkaddr
)
1742 unsigned char *kaddr
;
1743 struct f2fs_summary
*summary
;
1744 struct curseg_info
*seg_i
;
1745 int written_size
= 0;
1748 page
= grab_meta_page(sbi
, blkaddr
++);
1749 kaddr
= (unsigned char *)page_address(page
);
1751 /* Step 1: write nat cache */
1752 seg_i
= CURSEG_I(sbi
, CURSEG_HOT_DATA
);
1753 memcpy(kaddr
, seg_i
->journal
, SUM_JOURNAL_SIZE
);
1754 written_size
+= SUM_JOURNAL_SIZE
;
1756 /* Step 2: write sit cache */
1757 seg_i
= CURSEG_I(sbi
, CURSEG_COLD_DATA
);
1758 memcpy(kaddr
+ written_size
, seg_i
->journal
, SUM_JOURNAL_SIZE
);
1759 written_size
+= SUM_JOURNAL_SIZE
;
1761 /* Step 3: write summary entries */
1762 for (i
= CURSEG_HOT_DATA
; i
<= CURSEG_COLD_DATA
; i
++) {
1763 unsigned short blkoff
;
1764 seg_i
= CURSEG_I(sbi
, i
);
1765 if (sbi
->ckpt
->alloc_type
[i
] == SSR
)
1766 blkoff
= sbi
->blocks_per_seg
;
1768 blkoff
= curseg_blkoff(sbi
, i
);
1770 for (j
= 0; j
< blkoff
; j
++) {
1772 page
= grab_meta_page(sbi
, blkaddr
++);
1773 kaddr
= (unsigned char *)page_address(page
);
1776 summary
= (struct f2fs_summary
*)(kaddr
+ written_size
);
1777 *summary
= seg_i
->sum_blk
->entries
[j
];
1778 written_size
+= SUMMARY_SIZE
;
1780 if (written_size
+ SUMMARY_SIZE
<= PAGE_SIZE
-
1784 set_page_dirty(page
);
1785 f2fs_put_page(page
, 1);
1790 set_page_dirty(page
);
1791 f2fs_put_page(page
, 1);
1795 static void write_normal_summaries(struct f2fs_sb_info
*sbi
,
1796 block_t blkaddr
, int type
)
1799 if (IS_DATASEG(type
))
1800 end
= type
+ NR_CURSEG_DATA_TYPE
;
1802 end
= type
+ NR_CURSEG_NODE_TYPE
;
1804 for (i
= type
; i
< end
; i
++)
1805 write_current_sum_page(sbi
, i
, blkaddr
+ (i
- type
));
1808 void write_data_summaries(struct f2fs_sb_info
*sbi
, block_t start_blk
)
1810 if (is_set_ckpt_flags(F2FS_CKPT(sbi
), CP_COMPACT_SUM_FLAG
))
1811 write_compacted_summaries(sbi
, start_blk
);
1813 write_normal_summaries(sbi
, start_blk
, CURSEG_HOT_DATA
);
1816 void write_node_summaries(struct f2fs_sb_info
*sbi
, block_t start_blk
)
1818 write_normal_summaries(sbi
, start_blk
, CURSEG_HOT_NODE
);
1821 int lookup_journal_in_cursum(struct f2fs_journal
*journal
, int type
,
1822 unsigned int val
, int alloc
)
1826 if (type
== NAT_JOURNAL
) {
1827 for (i
= 0; i
< nats_in_cursum(journal
); i
++) {
1828 if (le32_to_cpu(nid_in_journal(journal
, i
)) == val
)
1831 if (alloc
&& __has_cursum_space(journal
, 1, NAT_JOURNAL
))
1832 return update_nats_in_cursum(journal
, 1);
1833 } else if (type
== SIT_JOURNAL
) {
1834 for (i
= 0; i
< sits_in_cursum(journal
); i
++)
1835 if (le32_to_cpu(segno_in_journal(journal
, i
)) == val
)
1837 if (alloc
&& __has_cursum_space(journal
, 1, SIT_JOURNAL
))
1838 return update_sits_in_cursum(journal
, 1);
1843 static struct page
*get_current_sit_page(struct f2fs_sb_info
*sbi
,
1846 return get_meta_page(sbi
, current_sit_addr(sbi
, segno
));
1849 static struct page
*get_next_sit_page(struct f2fs_sb_info
*sbi
,
1852 struct sit_info
*sit_i
= SIT_I(sbi
);
1853 struct page
*src_page
, *dst_page
;
1854 pgoff_t src_off
, dst_off
;
1855 void *src_addr
, *dst_addr
;
1857 src_off
= current_sit_addr(sbi
, start
);
1858 dst_off
= next_sit_addr(sbi
, src_off
);
1860 /* get current sit block page without lock */
1861 src_page
= get_meta_page(sbi
, src_off
);
1862 dst_page
= grab_meta_page(sbi
, dst_off
);
1863 f2fs_bug_on(sbi
, PageDirty(src_page
));
1865 src_addr
= page_address(src_page
);
1866 dst_addr
= page_address(dst_page
);
1867 memcpy(dst_addr
, src_addr
, PAGE_SIZE
);
1869 set_page_dirty(dst_page
);
1870 f2fs_put_page(src_page
, 1);
1872 set_to_next_sit(sit_i
, start
);
1877 static struct sit_entry_set
*grab_sit_entry_set(void)
1879 struct sit_entry_set
*ses
=
1880 f2fs_kmem_cache_alloc(sit_entry_set_slab
, GFP_NOFS
);
1883 INIT_LIST_HEAD(&ses
->set_list
);
1887 static void release_sit_entry_set(struct sit_entry_set
*ses
)
1889 list_del(&ses
->set_list
);
1890 kmem_cache_free(sit_entry_set_slab
, ses
);
1893 static void adjust_sit_entry_set(struct sit_entry_set
*ses
,
1894 struct list_head
*head
)
1896 struct sit_entry_set
*next
= ses
;
1898 if (list_is_last(&ses
->set_list
, head
))
1901 list_for_each_entry_continue(next
, head
, set_list
)
1902 if (ses
->entry_cnt
<= next
->entry_cnt
)
1905 list_move_tail(&ses
->set_list
, &next
->set_list
);
1908 static void add_sit_entry(unsigned int segno
, struct list_head
*head
)
1910 struct sit_entry_set
*ses
;
1911 unsigned int start_segno
= START_SEGNO(segno
);
1913 list_for_each_entry(ses
, head
, set_list
) {
1914 if (ses
->start_segno
== start_segno
) {
1916 adjust_sit_entry_set(ses
, head
);
1921 ses
= grab_sit_entry_set();
1923 ses
->start_segno
= start_segno
;
1925 list_add(&ses
->set_list
, head
);
1928 static void add_sits_in_set(struct f2fs_sb_info
*sbi
)
1930 struct f2fs_sm_info
*sm_info
= SM_I(sbi
);
1931 struct list_head
*set_list
= &sm_info
->sit_entry_set
;
1932 unsigned long *bitmap
= SIT_I(sbi
)->dirty_sentries_bitmap
;
1935 for_each_set_bit(segno
, bitmap
, MAIN_SEGS(sbi
))
1936 add_sit_entry(segno
, set_list
);
1939 static void remove_sits_in_journal(struct f2fs_sb_info
*sbi
)
1941 struct curseg_info
*curseg
= CURSEG_I(sbi
, CURSEG_COLD_DATA
);
1942 struct f2fs_journal
*journal
= curseg
->journal
;
1945 down_write(&curseg
->journal_rwsem
);
1946 for (i
= 0; i
< sits_in_cursum(journal
); i
++) {
1950 segno
= le32_to_cpu(segno_in_journal(journal
, i
));
1951 dirtied
= __mark_sit_entry_dirty(sbi
, segno
);
1954 add_sit_entry(segno
, &SM_I(sbi
)->sit_entry_set
);
1956 update_sits_in_cursum(journal
, -i
);
1957 up_write(&curseg
->journal_rwsem
);
1961 * CP calls this function, which flushes SIT entries including sit_journal,
1962 * and moves prefree segs to free segs.
1964 void flush_sit_entries(struct f2fs_sb_info
*sbi
, struct cp_control
*cpc
)
1966 struct sit_info
*sit_i
= SIT_I(sbi
);
1967 unsigned long *bitmap
= sit_i
->dirty_sentries_bitmap
;
1968 struct curseg_info
*curseg
= CURSEG_I(sbi
, CURSEG_COLD_DATA
);
1969 struct f2fs_journal
*journal
= curseg
->journal
;
1970 struct sit_entry_set
*ses
, *tmp
;
1971 struct list_head
*head
= &SM_I(sbi
)->sit_entry_set
;
1972 bool to_journal
= true;
1973 struct seg_entry
*se
;
1975 mutex_lock(&sit_i
->sentry_lock
);
1977 if (!sit_i
->dirty_sentries
)
1981 * add and account sit entries of dirty bitmap in sit entry
1984 add_sits_in_set(sbi
);
1987 * if there are no enough space in journal to store dirty sit
1988 * entries, remove all entries from journal and add and account
1989 * them in sit entry set.
1991 if (!__has_cursum_space(journal
, sit_i
->dirty_sentries
, SIT_JOURNAL
))
1992 remove_sits_in_journal(sbi
);
1995 * there are two steps to flush sit entries:
1996 * #1, flush sit entries to journal in current cold data summary block.
1997 * #2, flush sit entries to sit page.
1999 list_for_each_entry_safe(ses
, tmp
, head
, set_list
) {
2000 struct page
*page
= NULL
;
2001 struct f2fs_sit_block
*raw_sit
= NULL
;
2002 unsigned int start_segno
= ses
->start_segno
;
2003 unsigned int end
= min(start_segno
+ SIT_ENTRY_PER_BLOCK
,
2004 (unsigned long)MAIN_SEGS(sbi
));
2005 unsigned int segno
= start_segno
;
2008 !__has_cursum_space(journal
, ses
->entry_cnt
, SIT_JOURNAL
))
2012 down_write(&curseg
->journal_rwsem
);
2014 page
= get_next_sit_page(sbi
, start_segno
);
2015 raw_sit
= page_address(page
);
2018 /* flush dirty sit entries in region of current sit set */
2019 for_each_set_bit_from(segno
, bitmap
, end
) {
2020 int offset
, sit_offset
;
2022 se
= get_seg_entry(sbi
, segno
);
2024 /* add discard candidates */
2025 if (cpc
->reason
!= CP_DISCARD
) {
2026 cpc
->trim_start
= segno
;
2027 add_discard_addrs(sbi
, cpc
);
2031 offset
= lookup_journal_in_cursum(journal
,
2032 SIT_JOURNAL
, segno
, 1);
2033 f2fs_bug_on(sbi
, offset
< 0);
2034 segno_in_journal(journal
, offset
) =
2036 seg_info_to_raw_sit(se
,
2037 &sit_in_journal(journal
, offset
));
2039 sit_offset
= SIT_ENTRY_OFFSET(sit_i
, segno
);
2040 seg_info_to_raw_sit(se
,
2041 &raw_sit
->entries
[sit_offset
]);
2044 __clear_bit(segno
, bitmap
);
2045 sit_i
->dirty_sentries
--;
2050 up_write(&curseg
->journal_rwsem
);
2052 f2fs_put_page(page
, 1);
2054 f2fs_bug_on(sbi
, ses
->entry_cnt
);
2055 release_sit_entry_set(ses
);
2058 f2fs_bug_on(sbi
, !list_empty(head
));
2059 f2fs_bug_on(sbi
, sit_i
->dirty_sentries
);
2061 if (cpc
->reason
== CP_DISCARD
) {
2062 for (; cpc
->trim_start
<= cpc
->trim_end
; cpc
->trim_start
++)
2063 add_discard_addrs(sbi
, cpc
);
2065 mutex_unlock(&sit_i
->sentry_lock
);
2067 set_prefree_as_free_segments(sbi
);
2070 static int build_sit_info(struct f2fs_sb_info
*sbi
)
2072 struct f2fs_super_block
*raw_super
= F2FS_RAW_SUPER(sbi
);
2073 struct f2fs_checkpoint
*ckpt
= F2FS_CKPT(sbi
);
2074 struct sit_info
*sit_i
;
2075 unsigned int sit_segs
, start
;
2076 char *src_bitmap
, *dst_bitmap
;
2077 unsigned int bitmap_size
;
2079 /* allocate memory for SIT information */
2080 sit_i
= kzalloc(sizeof(struct sit_info
), GFP_KERNEL
);
2084 SM_I(sbi
)->sit_info
= sit_i
;
2086 sit_i
->sentries
= f2fs_kvzalloc(MAIN_SEGS(sbi
) *
2087 sizeof(struct seg_entry
), GFP_KERNEL
);
2088 if (!sit_i
->sentries
)
2091 bitmap_size
= f2fs_bitmap_size(MAIN_SEGS(sbi
));
2092 sit_i
->dirty_sentries_bitmap
= f2fs_kvzalloc(bitmap_size
, GFP_KERNEL
);
2093 if (!sit_i
->dirty_sentries_bitmap
)
2096 for (start
= 0; start
< MAIN_SEGS(sbi
); start
++) {
2097 sit_i
->sentries
[start
].cur_valid_map
2098 = kzalloc(SIT_VBLOCK_MAP_SIZE
, GFP_KERNEL
);
2099 sit_i
->sentries
[start
].ckpt_valid_map
2100 = kzalloc(SIT_VBLOCK_MAP_SIZE
, GFP_KERNEL
);
2101 sit_i
->sentries
[start
].discard_map
2102 = kzalloc(SIT_VBLOCK_MAP_SIZE
, GFP_KERNEL
);
2103 if (!sit_i
->sentries
[start
].cur_valid_map
||
2104 !sit_i
->sentries
[start
].ckpt_valid_map
||
2105 !sit_i
->sentries
[start
].discard_map
)
2109 sit_i
->tmp_map
= kzalloc(SIT_VBLOCK_MAP_SIZE
, GFP_KERNEL
);
2110 if (!sit_i
->tmp_map
)
2113 if (sbi
->segs_per_sec
> 1) {
2114 sit_i
->sec_entries
= f2fs_kvzalloc(MAIN_SECS(sbi
) *
2115 sizeof(struct sec_entry
), GFP_KERNEL
);
2116 if (!sit_i
->sec_entries
)
2120 /* get information related with SIT */
2121 sit_segs
= le32_to_cpu(raw_super
->segment_count_sit
) >> 1;
2123 /* setup SIT bitmap from ckeckpoint pack */
2124 bitmap_size
= __bitmap_size(sbi
, SIT_BITMAP
);
2125 src_bitmap
= __bitmap_ptr(sbi
, SIT_BITMAP
);
2127 dst_bitmap
= kmemdup(src_bitmap
, bitmap_size
, GFP_KERNEL
);
2131 /* init SIT information */
2132 sit_i
->s_ops
= &default_salloc_ops
;
2134 sit_i
->sit_base_addr
= le32_to_cpu(raw_super
->sit_blkaddr
);
2135 sit_i
->sit_blocks
= sit_segs
<< sbi
->log_blocks_per_seg
;
2136 sit_i
->written_valid_blocks
= le64_to_cpu(ckpt
->valid_block_count
);
2137 sit_i
->sit_bitmap
= dst_bitmap
;
2138 sit_i
->bitmap_size
= bitmap_size
;
2139 sit_i
->dirty_sentries
= 0;
2140 sit_i
->sents_per_block
= SIT_ENTRY_PER_BLOCK
;
2141 sit_i
->elapsed_time
= le64_to_cpu(sbi
->ckpt
->elapsed_time
);
2142 sit_i
->mounted_time
= CURRENT_TIME_SEC
.tv_sec
;
2143 mutex_init(&sit_i
->sentry_lock
);
2147 static int build_free_segmap(struct f2fs_sb_info
*sbi
)
2149 struct free_segmap_info
*free_i
;
2150 unsigned int bitmap_size
, sec_bitmap_size
;
2152 /* allocate memory for free segmap information */
2153 free_i
= kzalloc(sizeof(struct free_segmap_info
), GFP_KERNEL
);
2157 SM_I(sbi
)->free_info
= free_i
;
2159 bitmap_size
= f2fs_bitmap_size(MAIN_SEGS(sbi
));
2160 free_i
->free_segmap
= f2fs_kvmalloc(bitmap_size
, GFP_KERNEL
);
2161 if (!free_i
->free_segmap
)
2164 sec_bitmap_size
= f2fs_bitmap_size(MAIN_SECS(sbi
));
2165 free_i
->free_secmap
= f2fs_kvmalloc(sec_bitmap_size
, GFP_KERNEL
);
2166 if (!free_i
->free_secmap
)
2169 /* set all segments as dirty temporarily */
2170 memset(free_i
->free_segmap
, 0xff, bitmap_size
);
2171 memset(free_i
->free_secmap
, 0xff, sec_bitmap_size
);
2173 /* init free segmap information */
2174 free_i
->start_segno
= GET_SEGNO_FROM_SEG0(sbi
, MAIN_BLKADDR(sbi
));
2175 free_i
->free_segments
= 0;
2176 free_i
->free_sections
= 0;
2177 spin_lock_init(&free_i
->segmap_lock
);
2181 static int build_curseg(struct f2fs_sb_info
*sbi
)
2183 struct curseg_info
*array
;
2186 array
= kcalloc(NR_CURSEG_TYPE
, sizeof(*array
), GFP_KERNEL
);
2190 SM_I(sbi
)->curseg_array
= array
;
2192 for (i
= 0; i
< NR_CURSEG_TYPE
; i
++) {
2193 mutex_init(&array
[i
].curseg_mutex
);
2194 array
[i
].sum_blk
= kzalloc(PAGE_SIZE
, GFP_KERNEL
);
2195 if (!array
[i
].sum_blk
)
2197 init_rwsem(&array
[i
].journal_rwsem
);
2198 array
[i
].journal
= kzalloc(sizeof(struct f2fs_journal
),
2200 if (!array
[i
].journal
)
2202 array
[i
].segno
= NULL_SEGNO
;
2203 array
[i
].next_blkoff
= 0;
2205 return restore_curseg_summaries(sbi
);
2208 static void build_sit_entries(struct f2fs_sb_info
*sbi
)
2210 struct sit_info
*sit_i
= SIT_I(sbi
);
2211 struct curseg_info
*curseg
= CURSEG_I(sbi
, CURSEG_COLD_DATA
);
2212 struct f2fs_journal
*journal
= curseg
->journal
;
2213 int sit_blk_cnt
= SIT_BLK_CNT(sbi
);
2214 unsigned int i
, start
, end
;
2215 unsigned int readed
, start_blk
= 0;
2216 int nrpages
= MAX_BIO_BLOCKS(sbi
) * 8;
2219 readed
= ra_meta_pages(sbi
, start_blk
, nrpages
, META_SIT
, true);
2221 start
= start_blk
* sit_i
->sents_per_block
;
2222 end
= (start_blk
+ readed
) * sit_i
->sents_per_block
;
2224 for (; start
< end
&& start
< MAIN_SEGS(sbi
); start
++) {
2225 struct seg_entry
*se
= &sit_i
->sentries
[start
];
2226 struct f2fs_sit_block
*sit_blk
;
2227 struct f2fs_sit_entry sit
;
2230 down_read(&curseg
->journal_rwsem
);
2231 for (i
= 0; i
< sits_in_cursum(journal
); i
++) {
2232 if (le32_to_cpu(segno_in_journal(journal
, i
))
2234 sit
= sit_in_journal(journal
, i
);
2235 up_read(&curseg
->journal_rwsem
);
2239 up_read(&curseg
->journal_rwsem
);
2241 page
= get_current_sit_page(sbi
, start
);
2242 sit_blk
= (struct f2fs_sit_block
*)page_address(page
);
2243 sit
= sit_blk
->entries
[SIT_ENTRY_OFFSET(sit_i
, start
)];
2244 f2fs_put_page(page
, 1);
2246 check_block_count(sbi
, start
, &sit
);
2247 seg_info_from_raw_sit(se
, &sit
);
2249 /* build discard map only one time */
2250 memcpy(se
->discard_map
, se
->cur_valid_map
, SIT_VBLOCK_MAP_SIZE
);
2251 sbi
->discard_blks
+= sbi
->blocks_per_seg
- se
->valid_blocks
;
2253 if (sbi
->segs_per_sec
> 1) {
2254 struct sec_entry
*e
= get_sec_entry(sbi
, start
);
2255 e
->valid_blocks
+= se
->valid_blocks
;
2258 start_blk
+= readed
;
2259 } while (start_blk
< sit_blk_cnt
);
2262 static void init_free_segmap(struct f2fs_sb_info
*sbi
)
2267 for (start
= 0; start
< MAIN_SEGS(sbi
); start
++) {
2268 struct seg_entry
*sentry
= get_seg_entry(sbi
, start
);
2269 if (!sentry
->valid_blocks
)
2270 __set_free(sbi
, start
);
2273 /* set use the current segments */
2274 for (type
= CURSEG_HOT_DATA
; type
<= CURSEG_COLD_NODE
; type
++) {
2275 struct curseg_info
*curseg_t
= CURSEG_I(sbi
, type
);
2276 __set_test_and_inuse(sbi
, curseg_t
->segno
);
2280 static void init_dirty_segmap(struct f2fs_sb_info
*sbi
)
2282 struct dirty_seglist_info
*dirty_i
= DIRTY_I(sbi
);
2283 struct free_segmap_info
*free_i
= FREE_I(sbi
);
2284 unsigned int segno
= 0, offset
= 0;
2285 unsigned short valid_blocks
;
2288 /* find dirty segment based on free segmap */
2289 segno
= find_next_inuse(free_i
, MAIN_SEGS(sbi
), offset
);
2290 if (segno
>= MAIN_SEGS(sbi
))
2293 valid_blocks
= get_valid_blocks(sbi
, segno
, 0);
2294 if (valid_blocks
== sbi
->blocks_per_seg
|| !valid_blocks
)
2296 if (valid_blocks
> sbi
->blocks_per_seg
) {
2297 f2fs_bug_on(sbi
, 1);
2300 mutex_lock(&dirty_i
->seglist_lock
);
2301 __locate_dirty_segment(sbi
, segno
, DIRTY
);
2302 mutex_unlock(&dirty_i
->seglist_lock
);
2306 static int init_victim_secmap(struct f2fs_sb_info
*sbi
)
2308 struct dirty_seglist_info
*dirty_i
= DIRTY_I(sbi
);
2309 unsigned int bitmap_size
= f2fs_bitmap_size(MAIN_SECS(sbi
));
2311 dirty_i
->victim_secmap
= f2fs_kvzalloc(bitmap_size
, GFP_KERNEL
);
2312 if (!dirty_i
->victim_secmap
)
2317 static int build_dirty_segmap(struct f2fs_sb_info
*sbi
)
2319 struct dirty_seglist_info
*dirty_i
;
2320 unsigned int bitmap_size
, i
;
2322 /* allocate memory for dirty segments list information */
2323 dirty_i
= kzalloc(sizeof(struct dirty_seglist_info
), GFP_KERNEL
);
2327 SM_I(sbi
)->dirty_info
= dirty_i
;
2328 mutex_init(&dirty_i
->seglist_lock
);
2330 bitmap_size
= f2fs_bitmap_size(MAIN_SEGS(sbi
));
2332 for (i
= 0; i
< NR_DIRTY_TYPE
; i
++) {
2333 dirty_i
->dirty_segmap
[i
] = f2fs_kvzalloc(bitmap_size
, GFP_KERNEL
);
2334 if (!dirty_i
->dirty_segmap
[i
])
2338 init_dirty_segmap(sbi
);
2339 return init_victim_secmap(sbi
);
2343 * Update min, max modified time for cost-benefit GC algorithm
2345 static void init_min_max_mtime(struct f2fs_sb_info
*sbi
)
2347 struct sit_info
*sit_i
= SIT_I(sbi
);
2350 mutex_lock(&sit_i
->sentry_lock
);
2352 sit_i
->min_mtime
= LLONG_MAX
;
2354 for (segno
= 0; segno
< MAIN_SEGS(sbi
); segno
+= sbi
->segs_per_sec
) {
2356 unsigned long long mtime
= 0;
2358 for (i
= 0; i
< sbi
->segs_per_sec
; i
++)
2359 mtime
+= get_seg_entry(sbi
, segno
+ i
)->mtime
;
2361 mtime
= div_u64(mtime
, sbi
->segs_per_sec
);
2363 if (sit_i
->min_mtime
> mtime
)
2364 sit_i
->min_mtime
= mtime
;
2366 sit_i
->max_mtime
= get_mtime(sbi
);
2367 mutex_unlock(&sit_i
->sentry_lock
);
2370 int build_segment_manager(struct f2fs_sb_info
*sbi
)
2372 struct f2fs_super_block
*raw_super
= F2FS_RAW_SUPER(sbi
);
2373 struct f2fs_checkpoint
*ckpt
= F2FS_CKPT(sbi
);
2374 struct f2fs_sm_info
*sm_info
;
2377 sm_info
= kzalloc(sizeof(struct f2fs_sm_info
), GFP_KERNEL
);
2382 sbi
->sm_info
= sm_info
;
2383 sm_info
->seg0_blkaddr
= le32_to_cpu(raw_super
->segment0_blkaddr
);
2384 sm_info
->main_blkaddr
= le32_to_cpu(raw_super
->main_blkaddr
);
2385 sm_info
->segment_count
= le32_to_cpu(raw_super
->segment_count
);
2386 sm_info
->reserved_segments
= le32_to_cpu(ckpt
->rsvd_segment_count
);
2387 sm_info
->ovp_segments
= le32_to_cpu(ckpt
->overprov_segment_count
);
2388 sm_info
->main_segments
= le32_to_cpu(raw_super
->segment_count_main
);
2389 sm_info
->ssa_blkaddr
= le32_to_cpu(raw_super
->ssa_blkaddr
);
2390 sm_info
->rec_prefree_segments
= sm_info
->main_segments
*
2391 DEF_RECLAIM_PREFREE_SEGMENTS
/ 100;
2392 sm_info
->ipu_policy
= 1 << F2FS_IPU_FSYNC
;
2393 sm_info
->min_ipu_util
= DEF_MIN_IPU_UTIL
;
2394 sm_info
->min_fsync_blocks
= DEF_MIN_FSYNC_BLOCKS
;
2396 INIT_LIST_HEAD(&sm_info
->discard_list
);
2397 sm_info
->nr_discards
= 0;
2398 sm_info
->max_discards
= 0;
2400 sm_info
->trim_sections
= DEF_BATCHED_TRIM_SECTIONS
;
2402 INIT_LIST_HEAD(&sm_info
->sit_entry_set
);
2404 if (test_opt(sbi
, FLUSH_MERGE
) && !f2fs_readonly(sbi
->sb
)) {
2405 err
= create_flush_cmd_control(sbi
);
2410 err
= build_sit_info(sbi
);
2413 err
= build_free_segmap(sbi
);
2416 err
= build_curseg(sbi
);
2420 /* reinit free segmap based on SIT */
2421 build_sit_entries(sbi
);
2423 init_free_segmap(sbi
);
2424 err
= build_dirty_segmap(sbi
);
2428 init_min_max_mtime(sbi
);
2432 static void discard_dirty_segmap(struct f2fs_sb_info
*sbi
,
2433 enum dirty_type dirty_type
)
2435 struct dirty_seglist_info
*dirty_i
= DIRTY_I(sbi
);
2437 mutex_lock(&dirty_i
->seglist_lock
);
2438 kvfree(dirty_i
->dirty_segmap
[dirty_type
]);
2439 dirty_i
->nr_dirty
[dirty_type
] = 0;
2440 mutex_unlock(&dirty_i
->seglist_lock
);
2443 static void destroy_victim_secmap(struct f2fs_sb_info
*sbi
)
2445 struct dirty_seglist_info
*dirty_i
= DIRTY_I(sbi
);
2446 kvfree(dirty_i
->victim_secmap
);
2449 static void destroy_dirty_segmap(struct f2fs_sb_info
*sbi
)
2451 struct dirty_seglist_info
*dirty_i
= DIRTY_I(sbi
);
2457 /* discard pre-free/dirty segments list */
2458 for (i
= 0; i
< NR_DIRTY_TYPE
; i
++)
2459 discard_dirty_segmap(sbi
, i
);
2461 destroy_victim_secmap(sbi
);
2462 SM_I(sbi
)->dirty_info
= NULL
;
2466 static void destroy_curseg(struct f2fs_sb_info
*sbi
)
2468 struct curseg_info
*array
= SM_I(sbi
)->curseg_array
;
2473 SM_I(sbi
)->curseg_array
= NULL
;
2474 for (i
= 0; i
< NR_CURSEG_TYPE
; i
++) {
2475 kfree(array
[i
].sum_blk
);
2476 kfree(array
[i
].journal
);
2481 static void destroy_free_segmap(struct f2fs_sb_info
*sbi
)
2483 struct free_segmap_info
*free_i
= SM_I(sbi
)->free_info
;
2486 SM_I(sbi
)->free_info
= NULL
;
2487 kvfree(free_i
->free_segmap
);
2488 kvfree(free_i
->free_secmap
);
2492 static void destroy_sit_info(struct f2fs_sb_info
*sbi
)
2494 struct sit_info
*sit_i
= SIT_I(sbi
);
2500 if (sit_i
->sentries
) {
2501 for (start
= 0; start
< MAIN_SEGS(sbi
); start
++) {
2502 kfree(sit_i
->sentries
[start
].cur_valid_map
);
2503 kfree(sit_i
->sentries
[start
].ckpt_valid_map
);
2504 kfree(sit_i
->sentries
[start
].discard_map
);
2507 kfree(sit_i
->tmp_map
);
2509 kvfree(sit_i
->sentries
);
2510 kvfree(sit_i
->sec_entries
);
2511 kvfree(sit_i
->dirty_sentries_bitmap
);
2513 SM_I(sbi
)->sit_info
= NULL
;
2514 kfree(sit_i
->sit_bitmap
);
2518 void destroy_segment_manager(struct f2fs_sb_info
*sbi
)
2520 struct f2fs_sm_info
*sm_info
= SM_I(sbi
);
2524 destroy_flush_cmd_control(sbi
);
2525 destroy_dirty_segmap(sbi
);
2526 destroy_curseg(sbi
);
2527 destroy_free_segmap(sbi
);
2528 destroy_sit_info(sbi
);
2529 sbi
->sm_info
= NULL
;
2533 int __init
create_segment_manager_caches(void)
2535 discard_entry_slab
= f2fs_kmem_cache_create("discard_entry",
2536 sizeof(struct discard_entry
));
2537 if (!discard_entry_slab
)
2540 sit_entry_set_slab
= f2fs_kmem_cache_create("sit_entry_set",
2541 sizeof(struct sit_entry_set
));
2542 if (!sit_entry_set_slab
)
2543 goto destory_discard_entry
;
2545 inmem_entry_slab
= f2fs_kmem_cache_create("inmem_page_entry",
2546 sizeof(struct inmem_pages
));
2547 if (!inmem_entry_slab
)
2548 goto destroy_sit_entry_set
;
2551 destroy_sit_entry_set
:
2552 kmem_cache_destroy(sit_entry_set_slab
);
2553 destory_discard_entry
:
2554 kmem_cache_destroy(discard_entry_slab
);
2559 void destroy_segment_manager_caches(void)
2561 kmem_cache_destroy(sit_entry_set_slab
);
2562 kmem_cache_destroy(discard_entry_slab
);
2563 kmem_cache_destroy(inmem_entry_slab
);