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 int commit_inmem_pages(struct inode
*inode
, bool abort
)
196 struct f2fs_sb_info
*sbi
= F2FS_I_SB(inode
);
197 struct f2fs_inode_info
*fi
= F2FS_I(inode
);
198 struct inmem_pages
*cur
, *tmp
;
199 bool submit_bio
= false;
200 struct f2fs_io_info fio
= {
203 .rw
= WRITE_SYNC
| REQ_PRIO
,
204 .encrypted_page
= NULL
,
209 * The abort is true only when f2fs_evict_inode is called.
210 * Basically, the f2fs_evict_inode doesn't produce any data writes, so
211 * that we don't need to call f2fs_balance_fs.
212 * Otherwise, f2fs_gc in f2fs_balance_fs can wait forever until this
213 * inode becomes free by iget_locked in f2fs_iget.
216 f2fs_balance_fs(sbi
, true);
220 mutex_lock(&fi
->inmem_lock
);
221 list_for_each_entry_safe(cur
, tmp
, &fi
->inmem_pages
, list
) {
222 lock_page(cur
->page
);
224 if (cur
->page
->mapping
== inode
->i_mapping
) {
225 set_page_dirty(cur
->page
);
226 f2fs_wait_on_page_writeback(cur
->page
, DATA
);
227 if (clear_page_dirty_for_io(cur
->page
))
228 inode_dec_dirty_pages(inode
);
229 trace_f2fs_commit_inmem_page(cur
->page
, INMEM
);
230 fio
.page
= cur
->page
;
231 err
= do_write_data_page(&fio
);
233 unlock_page(cur
->page
);
236 clear_cold_data(cur
->page
);
240 ClearPageUptodate(cur
->page
);
241 trace_f2fs_commit_inmem_page(cur
->page
, INMEM_DROP
);
243 set_page_private(cur
->page
, 0);
244 ClearPagePrivate(cur
->page
);
245 f2fs_put_page(cur
->page
, 1);
247 list_del(&cur
->list
);
248 kmem_cache_free(inmem_entry_slab
, cur
);
249 dec_page_count(F2FS_I_SB(inode
), F2FS_INMEM_PAGES
);
251 mutex_unlock(&fi
->inmem_lock
);
256 f2fs_submit_merged_bio(sbi
, DATA
, WRITE
);
262 * This function balances dirty node and dentry pages.
263 * In addition, it controls garbage collection.
265 void f2fs_balance_fs(struct f2fs_sb_info
*sbi
, bool need
)
270 * We should do GC or end up with checkpoint, if there are so many dirty
271 * dir/node pages without enough free segments.
273 if (has_not_enough_free_secs(sbi
, 0)) {
274 mutex_lock(&sbi
->gc_mutex
);
279 void f2fs_balance_fs_bg(struct f2fs_sb_info
*sbi
)
281 /* try to shrink extent cache when there is no enough memory */
282 if (!available_free_memory(sbi
, EXTENT_CACHE
))
283 f2fs_shrink_extent_tree(sbi
, EXTENT_CACHE_SHRINK_NUMBER
);
285 /* check the # of cached NAT entries */
286 if (!available_free_memory(sbi
, NAT_ENTRIES
))
287 try_to_free_nats(sbi
, NAT_ENTRY_PER_BLOCK
);
289 if (!available_free_memory(sbi
, FREE_NIDS
))
290 try_to_free_nids(sbi
, NAT_ENTRY_PER_BLOCK
* FREE_NID_PAGES
);
292 /* checkpoint is the only way to shrink partial cached entries */
293 if (!available_free_memory(sbi
, NAT_ENTRIES
) ||
294 excess_prefree_segs(sbi
) ||
295 !available_free_memory(sbi
, INO_ENTRIES
) ||
296 (is_idle(sbi
) && f2fs_time_over(sbi
, CP_TIME
))) {
297 if (test_opt(sbi
, DATA_FLUSH
))
298 sync_dirty_inodes(sbi
, FILE_INODE
);
299 f2fs_sync_fs(sbi
->sb
, true);
300 stat_inc_bg_cp_count(sbi
->stat_info
);
304 static int issue_flush_thread(void *data
)
306 struct f2fs_sb_info
*sbi
= data
;
307 struct flush_cmd_control
*fcc
= SM_I(sbi
)->cmd_control_info
;
308 wait_queue_head_t
*q
= &fcc
->flush_wait_queue
;
310 if (kthread_should_stop())
313 if (!llist_empty(&fcc
->issue_list
)) {
315 struct flush_cmd
*cmd
, *next
;
318 bio
= f2fs_bio_alloc(0);
320 fcc
->dispatch_list
= llist_del_all(&fcc
->issue_list
);
321 fcc
->dispatch_list
= llist_reverse_order(fcc
->dispatch_list
);
323 bio
->bi_bdev
= sbi
->sb
->s_bdev
;
324 ret
= submit_bio_wait(WRITE_FLUSH
, bio
);
326 llist_for_each_entry_safe(cmd
, next
,
327 fcc
->dispatch_list
, llnode
) {
329 complete(&cmd
->wait
);
332 fcc
->dispatch_list
= NULL
;
335 wait_event_interruptible(*q
,
336 kthread_should_stop() || !llist_empty(&fcc
->issue_list
));
340 int f2fs_issue_flush(struct f2fs_sb_info
*sbi
)
342 struct flush_cmd_control
*fcc
= SM_I(sbi
)->cmd_control_info
;
343 struct flush_cmd cmd
;
345 trace_f2fs_issue_flush(sbi
->sb
, test_opt(sbi
, NOBARRIER
),
346 test_opt(sbi
, FLUSH_MERGE
));
348 if (test_opt(sbi
, NOBARRIER
))
351 if (!test_opt(sbi
, FLUSH_MERGE
)) {
352 struct bio
*bio
= f2fs_bio_alloc(0);
355 bio
->bi_bdev
= sbi
->sb
->s_bdev
;
356 ret
= submit_bio_wait(WRITE_FLUSH
, bio
);
361 init_completion(&cmd
.wait
);
363 llist_add(&cmd
.llnode
, &fcc
->issue_list
);
365 if (!fcc
->dispatch_list
)
366 wake_up(&fcc
->flush_wait_queue
);
368 wait_for_completion(&cmd
.wait
);
373 int create_flush_cmd_control(struct f2fs_sb_info
*sbi
)
375 dev_t dev
= sbi
->sb
->s_bdev
->bd_dev
;
376 struct flush_cmd_control
*fcc
;
379 fcc
= kzalloc(sizeof(struct flush_cmd_control
), GFP_KERNEL
);
382 init_waitqueue_head(&fcc
->flush_wait_queue
);
383 init_llist_head(&fcc
->issue_list
);
384 SM_I(sbi
)->cmd_control_info
= fcc
;
385 fcc
->f2fs_issue_flush
= kthread_run(issue_flush_thread
, sbi
,
386 "f2fs_flush-%u:%u", MAJOR(dev
), MINOR(dev
));
387 if (IS_ERR(fcc
->f2fs_issue_flush
)) {
388 err
= PTR_ERR(fcc
->f2fs_issue_flush
);
390 SM_I(sbi
)->cmd_control_info
= NULL
;
397 void destroy_flush_cmd_control(struct f2fs_sb_info
*sbi
)
399 struct flush_cmd_control
*fcc
= SM_I(sbi
)->cmd_control_info
;
401 if (fcc
&& fcc
->f2fs_issue_flush
)
402 kthread_stop(fcc
->f2fs_issue_flush
);
404 SM_I(sbi
)->cmd_control_info
= NULL
;
407 static void __locate_dirty_segment(struct f2fs_sb_info
*sbi
, unsigned int segno
,
408 enum dirty_type dirty_type
)
410 struct dirty_seglist_info
*dirty_i
= DIRTY_I(sbi
);
412 /* need not be added */
413 if (IS_CURSEG(sbi
, segno
))
416 if (!test_and_set_bit(segno
, dirty_i
->dirty_segmap
[dirty_type
]))
417 dirty_i
->nr_dirty
[dirty_type
]++;
419 if (dirty_type
== DIRTY
) {
420 struct seg_entry
*sentry
= get_seg_entry(sbi
, segno
);
421 enum dirty_type t
= sentry
->type
;
423 if (unlikely(t
>= DIRTY
)) {
427 if (!test_and_set_bit(segno
, dirty_i
->dirty_segmap
[t
]))
428 dirty_i
->nr_dirty
[t
]++;
432 static void __remove_dirty_segment(struct f2fs_sb_info
*sbi
, unsigned int segno
,
433 enum dirty_type dirty_type
)
435 struct dirty_seglist_info
*dirty_i
= DIRTY_I(sbi
);
437 if (test_and_clear_bit(segno
, dirty_i
->dirty_segmap
[dirty_type
]))
438 dirty_i
->nr_dirty
[dirty_type
]--;
440 if (dirty_type
== DIRTY
) {
441 struct seg_entry
*sentry
= get_seg_entry(sbi
, segno
);
442 enum dirty_type t
= sentry
->type
;
444 if (test_and_clear_bit(segno
, dirty_i
->dirty_segmap
[t
]))
445 dirty_i
->nr_dirty
[t
]--;
447 if (get_valid_blocks(sbi
, segno
, sbi
->segs_per_sec
) == 0)
448 clear_bit(GET_SECNO(sbi
, segno
),
449 dirty_i
->victim_secmap
);
454 * Should not occur error such as -ENOMEM.
455 * Adding dirty entry into seglist is not critical operation.
456 * If a given segment is one of current working segments, it won't be added.
458 static void locate_dirty_segment(struct f2fs_sb_info
*sbi
, unsigned int segno
)
460 struct dirty_seglist_info
*dirty_i
= DIRTY_I(sbi
);
461 unsigned short valid_blocks
;
463 if (segno
== NULL_SEGNO
|| IS_CURSEG(sbi
, segno
))
466 mutex_lock(&dirty_i
->seglist_lock
);
468 valid_blocks
= get_valid_blocks(sbi
, segno
, 0);
470 if (valid_blocks
== 0) {
471 __locate_dirty_segment(sbi
, segno
, PRE
);
472 __remove_dirty_segment(sbi
, segno
, DIRTY
);
473 } else if (valid_blocks
< sbi
->blocks_per_seg
) {
474 __locate_dirty_segment(sbi
, segno
, DIRTY
);
476 /* Recovery routine with SSR needs this */
477 __remove_dirty_segment(sbi
, segno
, DIRTY
);
480 mutex_unlock(&dirty_i
->seglist_lock
);
483 static int f2fs_issue_discard(struct f2fs_sb_info
*sbi
,
484 block_t blkstart
, block_t blklen
)
486 sector_t start
= SECTOR_FROM_BLOCK(blkstart
);
487 sector_t len
= SECTOR_FROM_BLOCK(blklen
);
488 struct seg_entry
*se
;
492 for (i
= blkstart
; i
< blkstart
+ blklen
; i
++) {
493 se
= get_seg_entry(sbi
, GET_SEGNO(sbi
, i
));
494 offset
= GET_BLKOFF_FROM_SEG0(sbi
, i
);
496 if (!f2fs_test_and_set_bit(offset
, se
->discard_map
))
499 trace_f2fs_issue_discard(sbi
->sb
, blkstart
, blklen
);
500 return blkdev_issue_discard(sbi
->sb
->s_bdev
, start
, len
, GFP_NOFS
, 0);
503 bool discard_next_dnode(struct f2fs_sb_info
*sbi
, block_t blkaddr
)
507 if (test_opt(sbi
, DISCARD
)) {
508 struct seg_entry
*se
= get_seg_entry(sbi
,
509 GET_SEGNO(sbi
, blkaddr
));
510 unsigned int offset
= GET_BLKOFF_FROM_SEG0(sbi
, blkaddr
);
512 if (f2fs_test_bit(offset
, se
->discard_map
))
515 err
= f2fs_issue_discard(sbi
, blkaddr
, 1);
519 update_meta_page(sbi
, NULL
, blkaddr
);
525 static void __add_discard_entry(struct f2fs_sb_info
*sbi
,
526 struct cp_control
*cpc
, struct seg_entry
*se
,
527 unsigned int start
, unsigned int end
)
529 struct list_head
*head
= &SM_I(sbi
)->discard_list
;
530 struct discard_entry
*new, *last
;
532 if (!list_empty(head
)) {
533 last
= list_last_entry(head
, struct discard_entry
, list
);
534 if (START_BLOCK(sbi
, cpc
->trim_start
) + start
==
535 last
->blkaddr
+ last
->len
) {
536 last
->len
+= end
- start
;
541 new = f2fs_kmem_cache_alloc(discard_entry_slab
, GFP_NOFS
);
542 INIT_LIST_HEAD(&new->list
);
543 new->blkaddr
= START_BLOCK(sbi
, cpc
->trim_start
) + start
;
544 new->len
= end
- start
;
545 list_add_tail(&new->list
, head
);
547 SM_I(sbi
)->nr_discards
+= end
- start
;
550 static void add_discard_addrs(struct f2fs_sb_info
*sbi
, struct cp_control
*cpc
)
552 int entries
= SIT_VBLOCK_MAP_SIZE
/ sizeof(unsigned long);
553 int max_blocks
= sbi
->blocks_per_seg
;
554 struct seg_entry
*se
= get_seg_entry(sbi
, cpc
->trim_start
);
555 unsigned long *cur_map
= (unsigned long *)se
->cur_valid_map
;
556 unsigned long *ckpt_map
= (unsigned long *)se
->ckpt_valid_map
;
557 unsigned long *discard_map
= (unsigned long *)se
->discard_map
;
558 unsigned long *dmap
= SIT_I(sbi
)->tmp_map
;
559 unsigned int start
= 0, end
= -1;
560 bool force
= (cpc
->reason
== CP_DISCARD
);
563 if (se
->valid_blocks
== max_blocks
)
567 if (!test_opt(sbi
, DISCARD
) || !se
->valid_blocks
||
568 SM_I(sbi
)->nr_discards
>= SM_I(sbi
)->max_discards
)
572 /* SIT_VBLOCK_MAP_SIZE should be multiple of sizeof(unsigned long) */
573 for (i
= 0; i
< entries
; i
++)
574 dmap
[i
] = force
? ~ckpt_map
[i
] & ~discard_map
[i
] :
575 (cur_map
[i
] ^ ckpt_map
[i
]) & ckpt_map
[i
];
577 while (force
|| SM_I(sbi
)->nr_discards
<= SM_I(sbi
)->max_discards
) {
578 start
= __find_rev_next_bit(dmap
, max_blocks
, end
+ 1);
579 if (start
>= max_blocks
)
582 end
= __find_rev_next_zero_bit(dmap
, max_blocks
, start
+ 1);
583 __add_discard_entry(sbi
, cpc
, se
, start
, end
);
587 void release_discard_addrs(struct f2fs_sb_info
*sbi
)
589 struct list_head
*head
= &(SM_I(sbi
)->discard_list
);
590 struct discard_entry
*entry
, *this;
593 list_for_each_entry_safe(entry
, this, head
, list
) {
594 list_del(&entry
->list
);
595 kmem_cache_free(discard_entry_slab
, entry
);
600 * Should call clear_prefree_segments after checkpoint is done.
602 static void set_prefree_as_free_segments(struct f2fs_sb_info
*sbi
)
604 struct dirty_seglist_info
*dirty_i
= DIRTY_I(sbi
);
607 mutex_lock(&dirty_i
->seglist_lock
);
608 for_each_set_bit(segno
, dirty_i
->dirty_segmap
[PRE
], MAIN_SEGS(sbi
))
609 __set_test_and_free(sbi
, segno
);
610 mutex_unlock(&dirty_i
->seglist_lock
);
613 void clear_prefree_segments(struct f2fs_sb_info
*sbi
, struct cp_control
*cpc
)
615 struct list_head
*head
= &(SM_I(sbi
)->discard_list
);
616 struct discard_entry
*entry
, *this;
617 struct dirty_seglist_info
*dirty_i
= DIRTY_I(sbi
);
618 unsigned long *prefree_map
= dirty_i
->dirty_segmap
[PRE
];
619 unsigned int start
= 0, end
= -1;
621 mutex_lock(&dirty_i
->seglist_lock
);
625 start
= find_next_bit(prefree_map
, MAIN_SEGS(sbi
), end
+ 1);
626 if (start
>= MAIN_SEGS(sbi
))
628 end
= find_next_zero_bit(prefree_map
, MAIN_SEGS(sbi
),
631 for (i
= start
; i
< end
; i
++)
632 clear_bit(i
, prefree_map
);
634 dirty_i
->nr_dirty
[PRE
] -= end
- start
;
636 if (!test_opt(sbi
, DISCARD
))
639 f2fs_issue_discard(sbi
, START_BLOCK(sbi
, start
),
640 (end
- start
) << sbi
->log_blocks_per_seg
);
642 mutex_unlock(&dirty_i
->seglist_lock
);
644 /* send small discards */
645 list_for_each_entry_safe(entry
, this, head
, list
) {
646 if (cpc
->reason
== CP_DISCARD
&& entry
->len
< cpc
->trim_minlen
)
648 f2fs_issue_discard(sbi
, entry
->blkaddr
, entry
->len
);
649 cpc
->trimmed
+= entry
->len
;
651 list_del(&entry
->list
);
652 SM_I(sbi
)->nr_discards
-= entry
->len
;
653 kmem_cache_free(discard_entry_slab
, entry
);
657 static bool __mark_sit_entry_dirty(struct f2fs_sb_info
*sbi
, unsigned int segno
)
659 struct sit_info
*sit_i
= SIT_I(sbi
);
661 if (!__test_and_set_bit(segno
, sit_i
->dirty_sentries_bitmap
)) {
662 sit_i
->dirty_sentries
++;
669 static void __set_sit_entry_type(struct f2fs_sb_info
*sbi
, int type
,
670 unsigned int segno
, int modified
)
672 struct seg_entry
*se
= get_seg_entry(sbi
, segno
);
675 __mark_sit_entry_dirty(sbi
, segno
);
678 static void update_sit_entry(struct f2fs_sb_info
*sbi
, block_t blkaddr
, int del
)
680 struct seg_entry
*se
;
681 unsigned int segno
, offset
;
682 long int new_vblocks
;
684 segno
= GET_SEGNO(sbi
, blkaddr
);
686 se
= get_seg_entry(sbi
, segno
);
687 new_vblocks
= se
->valid_blocks
+ del
;
688 offset
= GET_BLKOFF_FROM_SEG0(sbi
, blkaddr
);
690 f2fs_bug_on(sbi
, (new_vblocks
>> (sizeof(unsigned short) << 3) ||
691 (new_vblocks
> sbi
->blocks_per_seg
)));
693 se
->valid_blocks
= new_vblocks
;
694 se
->mtime
= get_mtime(sbi
);
695 SIT_I(sbi
)->max_mtime
= se
->mtime
;
697 /* Update valid block bitmap */
699 if (f2fs_test_and_set_bit(offset
, se
->cur_valid_map
))
701 if (!f2fs_test_and_set_bit(offset
, se
->discard_map
))
704 if (!f2fs_test_and_clear_bit(offset
, se
->cur_valid_map
))
706 if (f2fs_test_and_clear_bit(offset
, se
->discard_map
))
709 if (!f2fs_test_bit(offset
, se
->ckpt_valid_map
))
710 se
->ckpt_valid_blocks
+= del
;
712 __mark_sit_entry_dirty(sbi
, segno
);
714 /* update total number of valid blocks to be written in ckpt area */
715 SIT_I(sbi
)->written_valid_blocks
+= del
;
717 if (sbi
->segs_per_sec
> 1)
718 get_sec_entry(sbi
, segno
)->valid_blocks
+= del
;
721 void refresh_sit_entry(struct f2fs_sb_info
*sbi
, block_t old
, block_t
new)
723 update_sit_entry(sbi
, new, 1);
724 if (GET_SEGNO(sbi
, old
) != NULL_SEGNO
)
725 update_sit_entry(sbi
, old
, -1);
727 locate_dirty_segment(sbi
, GET_SEGNO(sbi
, old
));
728 locate_dirty_segment(sbi
, GET_SEGNO(sbi
, new));
731 void invalidate_blocks(struct f2fs_sb_info
*sbi
, block_t addr
)
733 unsigned int segno
= GET_SEGNO(sbi
, addr
);
734 struct sit_info
*sit_i
= SIT_I(sbi
);
736 f2fs_bug_on(sbi
, addr
== NULL_ADDR
);
737 if (addr
== NEW_ADDR
)
740 /* add it into sit main buffer */
741 mutex_lock(&sit_i
->sentry_lock
);
743 update_sit_entry(sbi
, addr
, -1);
745 /* add it into dirty seglist */
746 locate_dirty_segment(sbi
, segno
);
748 mutex_unlock(&sit_i
->sentry_lock
);
751 bool is_checkpointed_data(struct f2fs_sb_info
*sbi
, block_t blkaddr
)
753 struct sit_info
*sit_i
= SIT_I(sbi
);
754 unsigned int segno
, offset
;
755 struct seg_entry
*se
;
758 if (blkaddr
== NEW_ADDR
|| blkaddr
== NULL_ADDR
)
761 mutex_lock(&sit_i
->sentry_lock
);
763 segno
= GET_SEGNO(sbi
, blkaddr
);
764 se
= get_seg_entry(sbi
, segno
);
765 offset
= GET_BLKOFF_FROM_SEG0(sbi
, blkaddr
);
767 if (f2fs_test_bit(offset
, se
->ckpt_valid_map
))
770 mutex_unlock(&sit_i
->sentry_lock
);
776 * This function should be resided under the curseg_mutex lock
778 static void __add_sum_entry(struct f2fs_sb_info
*sbi
, int type
,
779 struct f2fs_summary
*sum
)
781 struct curseg_info
*curseg
= CURSEG_I(sbi
, type
);
782 void *addr
= curseg
->sum_blk
;
783 addr
+= curseg
->next_blkoff
* sizeof(struct f2fs_summary
);
784 memcpy(addr
, sum
, sizeof(struct f2fs_summary
));
788 * Calculate the number of current summary pages for writing
790 int npages_for_summary_flush(struct f2fs_sb_info
*sbi
, bool for_ra
)
792 int valid_sum_count
= 0;
795 for (i
= CURSEG_HOT_DATA
; i
<= CURSEG_COLD_DATA
; i
++) {
796 if (sbi
->ckpt
->alloc_type
[i
] == SSR
)
797 valid_sum_count
+= sbi
->blocks_per_seg
;
800 valid_sum_count
+= le16_to_cpu(
801 F2FS_CKPT(sbi
)->cur_data_blkoff
[i
]);
803 valid_sum_count
+= curseg_blkoff(sbi
, i
);
807 sum_in_page
= (PAGE_CACHE_SIZE
- 2 * SUM_JOURNAL_SIZE
-
808 SUM_FOOTER_SIZE
) / SUMMARY_SIZE
;
809 if (valid_sum_count
<= sum_in_page
)
811 else if ((valid_sum_count
- sum_in_page
) <=
812 (PAGE_CACHE_SIZE
- SUM_FOOTER_SIZE
) / SUMMARY_SIZE
)
818 * Caller should put this summary page
820 struct page
*get_sum_page(struct f2fs_sb_info
*sbi
, unsigned int segno
)
822 return get_meta_page(sbi
, GET_SUM_BLOCK(sbi
, segno
));
825 void update_meta_page(struct f2fs_sb_info
*sbi
, void *src
, block_t blk_addr
)
827 struct page
*page
= grab_meta_page(sbi
, blk_addr
);
828 void *dst
= page_address(page
);
831 memcpy(dst
, src
, PAGE_CACHE_SIZE
);
833 memset(dst
, 0, PAGE_CACHE_SIZE
);
834 set_page_dirty(page
);
835 f2fs_put_page(page
, 1);
838 static void write_sum_page(struct f2fs_sb_info
*sbi
,
839 struct f2fs_summary_block
*sum_blk
, block_t blk_addr
)
841 update_meta_page(sbi
, (void *)sum_blk
, blk_addr
);
844 static int is_next_segment_free(struct f2fs_sb_info
*sbi
, int type
)
846 struct curseg_info
*curseg
= CURSEG_I(sbi
, type
);
847 unsigned int segno
= curseg
->segno
+ 1;
848 struct free_segmap_info
*free_i
= FREE_I(sbi
);
850 if (segno
< MAIN_SEGS(sbi
) && segno
% sbi
->segs_per_sec
)
851 return !test_bit(segno
, free_i
->free_segmap
);
856 * Find a new segment from the free segments bitmap to right order
857 * This function should be returned with success, otherwise BUG
859 static void get_new_segment(struct f2fs_sb_info
*sbi
,
860 unsigned int *newseg
, bool new_sec
, int dir
)
862 struct free_segmap_info
*free_i
= FREE_I(sbi
);
863 unsigned int segno
, secno
, zoneno
;
864 unsigned int total_zones
= MAIN_SECS(sbi
) / sbi
->secs_per_zone
;
865 unsigned int hint
= *newseg
/ sbi
->segs_per_sec
;
866 unsigned int old_zoneno
= GET_ZONENO_FROM_SEGNO(sbi
, *newseg
);
867 unsigned int left_start
= hint
;
872 spin_lock(&free_i
->segmap_lock
);
874 if (!new_sec
&& ((*newseg
+ 1) % sbi
->segs_per_sec
)) {
875 segno
= find_next_zero_bit(free_i
->free_segmap
,
876 MAIN_SEGS(sbi
), *newseg
+ 1);
877 if (segno
- *newseg
< sbi
->segs_per_sec
-
878 (*newseg
% sbi
->segs_per_sec
))
882 secno
= find_next_zero_bit(free_i
->free_secmap
, MAIN_SECS(sbi
), hint
);
883 if (secno
>= MAIN_SECS(sbi
)) {
884 if (dir
== ALLOC_RIGHT
) {
885 secno
= find_next_zero_bit(free_i
->free_secmap
,
887 f2fs_bug_on(sbi
, secno
>= MAIN_SECS(sbi
));
890 left_start
= hint
- 1;
896 while (test_bit(left_start
, free_i
->free_secmap
)) {
897 if (left_start
> 0) {
901 left_start
= find_next_zero_bit(free_i
->free_secmap
,
903 f2fs_bug_on(sbi
, left_start
>= MAIN_SECS(sbi
));
909 segno
= secno
* sbi
->segs_per_sec
;
910 zoneno
= secno
/ sbi
->secs_per_zone
;
912 /* give up on finding another zone */
915 if (sbi
->secs_per_zone
== 1)
917 if (zoneno
== old_zoneno
)
919 if (dir
== ALLOC_LEFT
) {
920 if (!go_left
&& zoneno
+ 1 >= total_zones
)
922 if (go_left
&& zoneno
== 0)
925 for (i
= 0; i
< NR_CURSEG_TYPE
; i
++)
926 if (CURSEG_I(sbi
, i
)->zone
== zoneno
)
929 if (i
< NR_CURSEG_TYPE
) {
930 /* zone is in user, try another */
932 hint
= zoneno
* sbi
->secs_per_zone
- 1;
933 else if (zoneno
+ 1 >= total_zones
)
936 hint
= (zoneno
+ 1) * sbi
->secs_per_zone
;
938 goto find_other_zone
;
941 /* set it as dirty segment in free segmap */
942 f2fs_bug_on(sbi
, test_bit(segno
, free_i
->free_segmap
));
943 __set_inuse(sbi
, segno
);
945 spin_unlock(&free_i
->segmap_lock
);
948 static void reset_curseg(struct f2fs_sb_info
*sbi
, int type
, int modified
)
950 struct curseg_info
*curseg
= CURSEG_I(sbi
, type
);
951 struct summary_footer
*sum_footer
;
953 curseg
->segno
= curseg
->next_segno
;
954 curseg
->zone
= GET_ZONENO_FROM_SEGNO(sbi
, curseg
->segno
);
955 curseg
->next_blkoff
= 0;
956 curseg
->next_segno
= NULL_SEGNO
;
958 sum_footer
= &(curseg
->sum_blk
->footer
);
959 memset(sum_footer
, 0, sizeof(struct summary_footer
));
960 if (IS_DATASEG(type
))
961 SET_SUM_TYPE(sum_footer
, SUM_TYPE_DATA
);
962 if (IS_NODESEG(type
))
963 SET_SUM_TYPE(sum_footer
, SUM_TYPE_NODE
);
964 __set_sit_entry_type(sbi
, type
, curseg
->segno
, modified
);
968 * Allocate a current working segment.
969 * This function always allocates a free segment in LFS manner.
971 static void new_curseg(struct f2fs_sb_info
*sbi
, int type
, bool new_sec
)
973 struct curseg_info
*curseg
= CURSEG_I(sbi
, type
);
974 unsigned int segno
= curseg
->segno
;
975 int dir
= ALLOC_LEFT
;
977 write_sum_page(sbi
, curseg
->sum_blk
,
978 GET_SUM_BLOCK(sbi
, segno
));
979 if (type
== CURSEG_WARM_DATA
|| type
== CURSEG_COLD_DATA
)
982 if (test_opt(sbi
, NOHEAP
))
985 get_new_segment(sbi
, &segno
, new_sec
, dir
);
986 curseg
->next_segno
= segno
;
987 reset_curseg(sbi
, type
, 1);
988 curseg
->alloc_type
= LFS
;
991 static void __next_free_blkoff(struct f2fs_sb_info
*sbi
,
992 struct curseg_info
*seg
, block_t start
)
994 struct seg_entry
*se
= get_seg_entry(sbi
, seg
->segno
);
995 int entries
= SIT_VBLOCK_MAP_SIZE
/ sizeof(unsigned long);
996 unsigned long *target_map
= SIT_I(sbi
)->tmp_map
;
997 unsigned long *ckpt_map
= (unsigned long *)se
->ckpt_valid_map
;
998 unsigned long *cur_map
= (unsigned long *)se
->cur_valid_map
;
1001 for (i
= 0; i
< entries
; i
++)
1002 target_map
[i
] = ckpt_map
[i
] | cur_map
[i
];
1004 pos
= __find_rev_next_zero_bit(target_map
, sbi
->blocks_per_seg
, start
);
1006 seg
->next_blkoff
= pos
;
1010 * If a segment is written by LFS manner, next block offset is just obtained
1011 * by increasing the current block offset. However, if a segment is written by
1012 * SSR manner, next block offset obtained by calling __next_free_blkoff
1014 static void __refresh_next_blkoff(struct f2fs_sb_info
*sbi
,
1015 struct curseg_info
*seg
)
1017 if (seg
->alloc_type
== SSR
)
1018 __next_free_blkoff(sbi
, seg
, seg
->next_blkoff
+ 1);
1024 * This function always allocates a used segment(from dirty seglist) by SSR
1025 * manner, so it should recover the existing segment information of valid blocks
1027 static void change_curseg(struct f2fs_sb_info
*sbi
, int type
, bool reuse
)
1029 struct dirty_seglist_info
*dirty_i
= DIRTY_I(sbi
);
1030 struct curseg_info
*curseg
= CURSEG_I(sbi
, type
);
1031 unsigned int new_segno
= curseg
->next_segno
;
1032 struct f2fs_summary_block
*sum_node
;
1033 struct page
*sum_page
;
1035 write_sum_page(sbi
, curseg
->sum_blk
,
1036 GET_SUM_BLOCK(sbi
, curseg
->segno
));
1037 __set_test_and_inuse(sbi
, new_segno
);
1039 mutex_lock(&dirty_i
->seglist_lock
);
1040 __remove_dirty_segment(sbi
, new_segno
, PRE
);
1041 __remove_dirty_segment(sbi
, new_segno
, DIRTY
);
1042 mutex_unlock(&dirty_i
->seglist_lock
);
1044 reset_curseg(sbi
, type
, 1);
1045 curseg
->alloc_type
= SSR
;
1046 __next_free_blkoff(sbi
, curseg
, 0);
1049 sum_page
= get_sum_page(sbi
, new_segno
);
1050 sum_node
= (struct f2fs_summary_block
*)page_address(sum_page
);
1051 memcpy(curseg
->sum_blk
, sum_node
, SUM_ENTRY_SIZE
);
1052 f2fs_put_page(sum_page
, 1);
1056 static int get_ssr_segment(struct f2fs_sb_info
*sbi
, int type
)
1058 struct curseg_info
*curseg
= CURSEG_I(sbi
, type
);
1059 const struct victim_selection
*v_ops
= DIRTY_I(sbi
)->v_ops
;
1061 if (IS_NODESEG(type
) || !has_not_enough_free_secs(sbi
, 0))
1062 return v_ops
->get_victim(sbi
,
1063 &(curseg
)->next_segno
, BG_GC
, type
, SSR
);
1065 /* For data segments, let's do SSR more intensively */
1066 for (; type
>= CURSEG_HOT_DATA
; type
--)
1067 if (v_ops
->get_victim(sbi
, &(curseg
)->next_segno
,
1074 * flush out current segment and replace it with new segment
1075 * This function should be returned with success, otherwise BUG
1077 static void allocate_segment_by_default(struct f2fs_sb_info
*sbi
,
1078 int type
, bool force
)
1080 struct curseg_info
*curseg
= CURSEG_I(sbi
, type
);
1083 new_curseg(sbi
, type
, true);
1084 else if (type
== CURSEG_WARM_NODE
)
1085 new_curseg(sbi
, type
, false);
1086 else if (curseg
->alloc_type
== LFS
&& is_next_segment_free(sbi
, type
))
1087 new_curseg(sbi
, type
, false);
1088 else if (need_SSR(sbi
) && get_ssr_segment(sbi
, type
))
1089 change_curseg(sbi
, type
, true);
1091 new_curseg(sbi
, type
, false);
1093 stat_inc_seg_type(sbi
, curseg
);
1096 static void __allocate_new_segments(struct f2fs_sb_info
*sbi
, int type
)
1098 struct curseg_info
*curseg
= CURSEG_I(sbi
, type
);
1099 unsigned int old_segno
;
1101 old_segno
= curseg
->segno
;
1102 SIT_I(sbi
)->s_ops
->allocate_segment(sbi
, type
, true);
1103 locate_dirty_segment(sbi
, old_segno
);
1106 void allocate_new_segments(struct f2fs_sb_info
*sbi
)
1110 for (i
= CURSEG_HOT_DATA
; i
<= CURSEG_COLD_DATA
; i
++)
1111 __allocate_new_segments(sbi
, i
);
1114 static const struct segment_allocation default_salloc_ops
= {
1115 .allocate_segment
= allocate_segment_by_default
,
1118 int f2fs_trim_fs(struct f2fs_sb_info
*sbi
, struct fstrim_range
*range
)
1120 __u64 start
= F2FS_BYTES_TO_BLK(range
->start
);
1121 __u64 end
= start
+ F2FS_BYTES_TO_BLK(range
->len
) - 1;
1122 unsigned int start_segno
, end_segno
;
1123 struct cp_control cpc
;
1126 if (start
>= MAX_BLKADDR(sbi
) || range
->len
< sbi
->blocksize
)
1130 if (end
<= MAIN_BLKADDR(sbi
))
1133 /* start/end segment number in main_area */
1134 start_segno
= (start
<= MAIN_BLKADDR(sbi
)) ? 0 : GET_SEGNO(sbi
, start
);
1135 end_segno
= (end
>= MAX_BLKADDR(sbi
)) ? MAIN_SEGS(sbi
) - 1 :
1136 GET_SEGNO(sbi
, end
);
1137 cpc
.reason
= CP_DISCARD
;
1138 cpc
.trim_minlen
= max_t(__u64
, 1, F2FS_BYTES_TO_BLK(range
->minlen
));
1140 /* do checkpoint to issue discard commands safely */
1141 for (; start_segno
<= end_segno
; start_segno
= cpc
.trim_end
+ 1) {
1142 cpc
.trim_start
= start_segno
;
1144 if (sbi
->discard_blks
== 0)
1146 else if (sbi
->discard_blks
< BATCHED_TRIM_BLOCKS(sbi
))
1147 cpc
.trim_end
= end_segno
;
1149 cpc
.trim_end
= min_t(unsigned int,
1150 rounddown(start_segno
+
1151 BATCHED_TRIM_SEGMENTS(sbi
),
1152 sbi
->segs_per_sec
) - 1, end_segno
);
1154 mutex_lock(&sbi
->gc_mutex
);
1155 err
= write_checkpoint(sbi
, &cpc
);
1156 mutex_unlock(&sbi
->gc_mutex
);
1159 range
->len
= F2FS_BLK_TO_BYTES(cpc
.trimmed
);
1163 static bool __has_curseg_space(struct f2fs_sb_info
*sbi
, int type
)
1165 struct curseg_info
*curseg
= CURSEG_I(sbi
, type
);
1166 if (curseg
->next_blkoff
< sbi
->blocks_per_seg
)
1171 static int __get_segment_type_2(struct page
*page
, enum page_type p_type
)
1174 return CURSEG_HOT_DATA
;
1176 return CURSEG_HOT_NODE
;
1179 static int __get_segment_type_4(struct page
*page
, enum page_type p_type
)
1181 if (p_type
== DATA
) {
1182 struct inode
*inode
= page
->mapping
->host
;
1184 if (S_ISDIR(inode
->i_mode
))
1185 return CURSEG_HOT_DATA
;
1187 return CURSEG_COLD_DATA
;
1189 if (IS_DNODE(page
) && is_cold_node(page
))
1190 return CURSEG_WARM_NODE
;
1192 return CURSEG_COLD_NODE
;
1196 static int __get_segment_type_6(struct page
*page
, enum page_type p_type
)
1198 if (p_type
== DATA
) {
1199 struct inode
*inode
= page
->mapping
->host
;
1201 if (S_ISDIR(inode
->i_mode
))
1202 return CURSEG_HOT_DATA
;
1203 else if (is_cold_data(page
) || file_is_cold(inode
))
1204 return CURSEG_COLD_DATA
;
1206 return CURSEG_WARM_DATA
;
1209 return is_cold_node(page
) ? CURSEG_WARM_NODE
:
1212 return CURSEG_COLD_NODE
;
1216 static int __get_segment_type(struct page
*page
, enum page_type p_type
)
1218 switch (F2FS_P_SB(page
)->active_logs
) {
1220 return __get_segment_type_2(page
, p_type
);
1222 return __get_segment_type_4(page
, p_type
);
1224 /* NR_CURSEG_TYPE(6) logs by default */
1225 f2fs_bug_on(F2FS_P_SB(page
),
1226 F2FS_P_SB(page
)->active_logs
!= NR_CURSEG_TYPE
);
1227 return __get_segment_type_6(page
, p_type
);
1230 void allocate_data_block(struct f2fs_sb_info
*sbi
, struct page
*page
,
1231 block_t old_blkaddr
, block_t
*new_blkaddr
,
1232 struct f2fs_summary
*sum
, int type
)
1234 struct sit_info
*sit_i
= SIT_I(sbi
);
1235 struct curseg_info
*curseg
;
1236 bool direct_io
= (type
== CURSEG_DIRECT_IO
);
1238 type
= direct_io
? CURSEG_WARM_DATA
: type
;
1240 curseg
= CURSEG_I(sbi
, type
);
1242 mutex_lock(&curseg
->curseg_mutex
);
1243 mutex_lock(&sit_i
->sentry_lock
);
1245 /* direct_io'ed data is aligned to the segment for better performance */
1246 if (direct_io
&& curseg
->next_blkoff
&&
1247 !has_not_enough_free_secs(sbi
, 0))
1248 __allocate_new_segments(sbi
, type
);
1250 *new_blkaddr
= NEXT_FREE_BLKADDR(sbi
, curseg
);
1253 * __add_sum_entry should be resided under the curseg_mutex
1254 * because, this function updates a summary entry in the
1255 * current summary block.
1257 __add_sum_entry(sbi
, type
, sum
);
1259 __refresh_next_blkoff(sbi
, curseg
);
1261 stat_inc_block_count(sbi
, curseg
);
1263 if (!__has_curseg_space(sbi
, type
))
1264 sit_i
->s_ops
->allocate_segment(sbi
, type
, false);
1266 * SIT information should be updated before segment allocation,
1267 * since SSR needs latest valid block information.
1269 refresh_sit_entry(sbi
, old_blkaddr
, *new_blkaddr
);
1271 mutex_unlock(&sit_i
->sentry_lock
);
1273 if (page
&& IS_NODESEG(type
))
1274 fill_node_footer_blkaddr(page
, NEXT_FREE_BLKADDR(sbi
, curseg
));
1276 mutex_unlock(&curseg
->curseg_mutex
);
1279 static void do_write_page(struct f2fs_summary
*sum
, struct f2fs_io_info
*fio
)
1281 int type
= __get_segment_type(fio
->page
, fio
->type
);
1283 allocate_data_block(fio
->sbi
, fio
->page
, fio
->blk_addr
,
1284 &fio
->blk_addr
, sum
, type
);
1286 /* writeout dirty page into bdev */
1287 f2fs_submit_page_mbio(fio
);
1290 void write_meta_page(struct f2fs_sb_info
*sbi
, struct page
*page
)
1292 struct f2fs_io_info fio
= {
1295 .rw
= WRITE_SYNC
| REQ_META
| REQ_PRIO
,
1296 .blk_addr
= page
->index
,
1298 .encrypted_page
= NULL
,
1301 if (unlikely(page
->index
>= MAIN_BLKADDR(sbi
)))
1302 fio
.rw
&= ~REQ_META
;
1304 set_page_writeback(page
);
1305 f2fs_submit_page_mbio(&fio
);
1308 void write_node_page(unsigned int nid
, struct f2fs_io_info
*fio
)
1310 struct f2fs_summary sum
;
1312 set_summary(&sum
, nid
, 0, 0);
1313 do_write_page(&sum
, fio
);
1316 void write_data_page(struct dnode_of_data
*dn
, struct f2fs_io_info
*fio
)
1318 struct f2fs_sb_info
*sbi
= fio
->sbi
;
1319 struct f2fs_summary sum
;
1320 struct node_info ni
;
1322 f2fs_bug_on(sbi
, dn
->data_blkaddr
== NULL_ADDR
);
1323 get_node_info(sbi
, dn
->nid
, &ni
);
1324 set_summary(&sum
, dn
->nid
, dn
->ofs_in_node
, ni
.version
);
1325 do_write_page(&sum
, fio
);
1326 dn
->data_blkaddr
= fio
->blk_addr
;
1329 void rewrite_data_page(struct f2fs_io_info
*fio
)
1331 stat_inc_inplace_blocks(fio
->sbi
);
1332 f2fs_submit_page_mbio(fio
);
1335 static void __f2fs_replace_block(struct f2fs_sb_info
*sbi
,
1336 struct f2fs_summary
*sum
,
1337 block_t old_blkaddr
, block_t new_blkaddr
,
1338 bool recover_curseg
)
1340 struct sit_info
*sit_i
= SIT_I(sbi
);
1341 struct curseg_info
*curseg
;
1342 unsigned int segno
, old_cursegno
;
1343 struct seg_entry
*se
;
1345 unsigned short old_blkoff
;
1347 segno
= GET_SEGNO(sbi
, new_blkaddr
);
1348 se
= get_seg_entry(sbi
, segno
);
1351 if (!recover_curseg
) {
1352 /* for recovery flow */
1353 if (se
->valid_blocks
== 0 && !IS_CURSEG(sbi
, segno
)) {
1354 if (old_blkaddr
== NULL_ADDR
)
1355 type
= CURSEG_COLD_DATA
;
1357 type
= CURSEG_WARM_DATA
;
1360 if (!IS_CURSEG(sbi
, segno
))
1361 type
= CURSEG_WARM_DATA
;
1364 curseg
= CURSEG_I(sbi
, type
);
1366 mutex_lock(&curseg
->curseg_mutex
);
1367 mutex_lock(&sit_i
->sentry_lock
);
1369 old_cursegno
= curseg
->segno
;
1370 old_blkoff
= curseg
->next_blkoff
;
1372 /* change the current segment */
1373 if (segno
!= curseg
->segno
) {
1374 curseg
->next_segno
= segno
;
1375 change_curseg(sbi
, type
, true);
1378 curseg
->next_blkoff
= GET_BLKOFF_FROM_SEG0(sbi
, new_blkaddr
);
1379 __add_sum_entry(sbi
, type
, sum
);
1381 if (!recover_curseg
)
1382 update_sit_entry(sbi
, new_blkaddr
, 1);
1383 if (GET_SEGNO(sbi
, old_blkaddr
) != NULL_SEGNO
)
1384 update_sit_entry(sbi
, old_blkaddr
, -1);
1386 locate_dirty_segment(sbi
, GET_SEGNO(sbi
, old_blkaddr
));
1387 locate_dirty_segment(sbi
, GET_SEGNO(sbi
, new_blkaddr
));
1389 locate_dirty_segment(sbi
, old_cursegno
);
1391 if (recover_curseg
) {
1392 if (old_cursegno
!= curseg
->segno
) {
1393 curseg
->next_segno
= old_cursegno
;
1394 change_curseg(sbi
, type
, true);
1396 curseg
->next_blkoff
= old_blkoff
;
1399 mutex_unlock(&sit_i
->sentry_lock
);
1400 mutex_unlock(&curseg
->curseg_mutex
);
1403 void f2fs_replace_block(struct f2fs_sb_info
*sbi
, struct dnode_of_data
*dn
,
1404 block_t old_addr
, block_t new_addr
,
1405 unsigned char version
, bool recover_curseg
)
1407 struct f2fs_summary sum
;
1409 set_summary(&sum
, dn
->nid
, dn
->ofs_in_node
, version
);
1411 __f2fs_replace_block(sbi
, &sum
, old_addr
, new_addr
, recover_curseg
);
1413 dn
->data_blkaddr
= new_addr
;
1414 set_data_blkaddr(dn
);
1415 f2fs_update_extent_cache(dn
);
1418 static inline bool is_merged_page(struct f2fs_sb_info
*sbi
,
1419 struct page
*page
, enum page_type type
)
1421 enum page_type btype
= PAGE_TYPE_OF_BIO(type
);
1422 struct f2fs_bio_info
*io
= &sbi
->write_io
[btype
];
1423 struct bio_vec
*bvec
;
1424 struct page
*target
;
1427 down_read(&io
->io_rwsem
);
1429 up_read(&io
->io_rwsem
);
1433 bio_for_each_segment_all(bvec
, io
->bio
, i
) {
1435 if (bvec
->bv_page
->mapping
) {
1436 target
= bvec
->bv_page
;
1438 struct f2fs_crypto_ctx
*ctx
;
1440 /* encrypted page */
1441 ctx
= (struct f2fs_crypto_ctx
*)page_private(
1443 target
= ctx
->w
.control_page
;
1446 if (page
== target
) {
1447 up_read(&io
->io_rwsem
);
1452 up_read(&io
->io_rwsem
);
1456 void f2fs_wait_on_page_writeback(struct page
*page
,
1457 enum page_type type
)
1459 if (PageWriteback(page
)) {
1460 struct f2fs_sb_info
*sbi
= F2FS_P_SB(page
);
1462 if (is_merged_page(sbi
, page
, type
))
1463 f2fs_submit_merged_bio(sbi
, type
, WRITE
);
1464 wait_on_page_writeback(page
);
1468 void f2fs_wait_on_encrypted_page_writeback(struct f2fs_sb_info
*sbi
,
1473 if (blkaddr
== NEW_ADDR
)
1476 f2fs_bug_on(sbi
, blkaddr
== NULL_ADDR
);
1478 cpage
= find_lock_page(META_MAPPING(sbi
), blkaddr
);
1480 f2fs_wait_on_page_writeback(cpage
, DATA
);
1481 f2fs_put_page(cpage
, 1);
1485 static int read_compacted_summaries(struct f2fs_sb_info
*sbi
)
1487 struct f2fs_checkpoint
*ckpt
= F2FS_CKPT(sbi
);
1488 struct curseg_info
*seg_i
;
1489 unsigned char *kaddr
;
1494 start
= start_sum_block(sbi
);
1496 page
= get_meta_page(sbi
, start
++);
1497 kaddr
= (unsigned char *)page_address(page
);
1499 /* Step 1: restore nat cache */
1500 seg_i
= CURSEG_I(sbi
, CURSEG_HOT_DATA
);
1501 memcpy(&seg_i
->sum_blk
->n_nats
, kaddr
, SUM_JOURNAL_SIZE
);
1503 /* Step 2: restore sit cache */
1504 seg_i
= CURSEG_I(sbi
, CURSEG_COLD_DATA
);
1505 memcpy(&seg_i
->sum_blk
->n_sits
, kaddr
+ SUM_JOURNAL_SIZE
,
1507 offset
= 2 * SUM_JOURNAL_SIZE
;
1509 /* Step 3: restore summary entries */
1510 for (i
= CURSEG_HOT_DATA
; i
<= CURSEG_COLD_DATA
; i
++) {
1511 unsigned short blk_off
;
1514 seg_i
= CURSEG_I(sbi
, i
);
1515 segno
= le32_to_cpu(ckpt
->cur_data_segno
[i
]);
1516 blk_off
= le16_to_cpu(ckpt
->cur_data_blkoff
[i
]);
1517 seg_i
->next_segno
= segno
;
1518 reset_curseg(sbi
, i
, 0);
1519 seg_i
->alloc_type
= ckpt
->alloc_type
[i
];
1520 seg_i
->next_blkoff
= blk_off
;
1522 if (seg_i
->alloc_type
== SSR
)
1523 blk_off
= sbi
->blocks_per_seg
;
1525 for (j
= 0; j
< blk_off
; j
++) {
1526 struct f2fs_summary
*s
;
1527 s
= (struct f2fs_summary
*)(kaddr
+ offset
);
1528 seg_i
->sum_blk
->entries
[j
] = *s
;
1529 offset
+= SUMMARY_SIZE
;
1530 if (offset
+ SUMMARY_SIZE
<= PAGE_CACHE_SIZE
-
1534 f2fs_put_page(page
, 1);
1537 page
= get_meta_page(sbi
, start
++);
1538 kaddr
= (unsigned char *)page_address(page
);
1542 f2fs_put_page(page
, 1);
1546 static int read_normal_summaries(struct f2fs_sb_info
*sbi
, int type
)
1548 struct f2fs_checkpoint
*ckpt
= F2FS_CKPT(sbi
);
1549 struct f2fs_summary_block
*sum
;
1550 struct curseg_info
*curseg
;
1552 unsigned short blk_off
;
1553 unsigned int segno
= 0;
1554 block_t blk_addr
= 0;
1556 /* get segment number and block addr */
1557 if (IS_DATASEG(type
)) {
1558 segno
= le32_to_cpu(ckpt
->cur_data_segno
[type
]);
1559 blk_off
= le16_to_cpu(ckpt
->cur_data_blkoff
[type
-
1561 if (__exist_node_summaries(sbi
))
1562 blk_addr
= sum_blk_addr(sbi
, NR_CURSEG_TYPE
, type
);
1564 blk_addr
= sum_blk_addr(sbi
, NR_CURSEG_DATA_TYPE
, type
);
1566 segno
= le32_to_cpu(ckpt
->cur_node_segno
[type
-
1568 blk_off
= le16_to_cpu(ckpt
->cur_node_blkoff
[type
-
1570 if (__exist_node_summaries(sbi
))
1571 blk_addr
= sum_blk_addr(sbi
, NR_CURSEG_NODE_TYPE
,
1572 type
- CURSEG_HOT_NODE
);
1574 blk_addr
= GET_SUM_BLOCK(sbi
, segno
);
1577 new = get_meta_page(sbi
, blk_addr
);
1578 sum
= (struct f2fs_summary_block
*)page_address(new);
1580 if (IS_NODESEG(type
)) {
1581 if (__exist_node_summaries(sbi
)) {
1582 struct f2fs_summary
*ns
= &sum
->entries
[0];
1584 for (i
= 0; i
< sbi
->blocks_per_seg
; i
++, ns
++) {
1586 ns
->ofs_in_node
= 0;
1591 err
= restore_node_summary(sbi
, segno
, sum
);
1593 f2fs_put_page(new, 1);
1599 /* set uncompleted segment to curseg */
1600 curseg
= CURSEG_I(sbi
, type
);
1601 mutex_lock(&curseg
->curseg_mutex
);
1602 memcpy(curseg
->sum_blk
, sum
, PAGE_CACHE_SIZE
);
1603 curseg
->next_segno
= segno
;
1604 reset_curseg(sbi
, type
, 0);
1605 curseg
->alloc_type
= ckpt
->alloc_type
[type
];
1606 curseg
->next_blkoff
= blk_off
;
1607 mutex_unlock(&curseg
->curseg_mutex
);
1608 f2fs_put_page(new, 1);
1612 static int restore_curseg_summaries(struct f2fs_sb_info
*sbi
)
1614 int type
= CURSEG_HOT_DATA
;
1617 if (is_set_ckpt_flags(F2FS_CKPT(sbi
), CP_COMPACT_SUM_FLAG
)) {
1618 int npages
= npages_for_summary_flush(sbi
, true);
1621 ra_meta_pages(sbi
, start_sum_block(sbi
), npages
,
1624 /* restore for compacted data summary */
1625 if (read_compacted_summaries(sbi
))
1627 type
= CURSEG_HOT_NODE
;
1630 if (__exist_node_summaries(sbi
))
1631 ra_meta_pages(sbi
, sum_blk_addr(sbi
, NR_CURSEG_TYPE
, type
),
1632 NR_CURSEG_TYPE
- type
, META_CP
, true);
1634 for (; type
<= CURSEG_COLD_NODE
; type
++) {
1635 err
= read_normal_summaries(sbi
, type
);
1643 static void write_compacted_summaries(struct f2fs_sb_info
*sbi
, block_t blkaddr
)
1646 unsigned char *kaddr
;
1647 struct f2fs_summary
*summary
;
1648 struct curseg_info
*seg_i
;
1649 int written_size
= 0;
1652 page
= grab_meta_page(sbi
, blkaddr
++);
1653 kaddr
= (unsigned char *)page_address(page
);
1655 /* Step 1: write nat cache */
1656 seg_i
= CURSEG_I(sbi
, CURSEG_HOT_DATA
);
1657 memcpy(kaddr
, &seg_i
->sum_blk
->n_nats
, SUM_JOURNAL_SIZE
);
1658 written_size
+= SUM_JOURNAL_SIZE
;
1660 /* Step 2: write sit cache */
1661 seg_i
= CURSEG_I(sbi
, CURSEG_COLD_DATA
);
1662 memcpy(kaddr
+ written_size
, &seg_i
->sum_blk
->n_sits
,
1664 written_size
+= SUM_JOURNAL_SIZE
;
1666 /* Step 3: write summary entries */
1667 for (i
= CURSEG_HOT_DATA
; i
<= CURSEG_COLD_DATA
; i
++) {
1668 unsigned short blkoff
;
1669 seg_i
= CURSEG_I(sbi
, i
);
1670 if (sbi
->ckpt
->alloc_type
[i
] == SSR
)
1671 blkoff
= sbi
->blocks_per_seg
;
1673 blkoff
= curseg_blkoff(sbi
, i
);
1675 for (j
= 0; j
< blkoff
; j
++) {
1677 page
= grab_meta_page(sbi
, blkaddr
++);
1678 kaddr
= (unsigned char *)page_address(page
);
1681 summary
= (struct f2fs_summary
*)(kaddr
+ written_size
);
1682 *summary
= seg_i
->sum_blk
->entries
[j
];
1683 written_size
+= SUMMARY_SIZE
;
1685 if (written_size
+ SUMMARY_SIZE
<= PAGE_CACHE_SIZE
-
1689 set_page_dirty(page
);
1690 f2fs_put_page(page
, 1);
1695 set_page_dirty(page
);
1696 f2fs_put_page(page
, 1);
1700 static void write_normal_summaries(struct f2fs_sb_info
*sbi
,
1701 block_t blkaddr
, int type
)
1704 if (IS_DATASEG(type
))
1705 end
= type
+ NR_CURSEG_DATA_TYPE
;
1707 end
= type
+ NR_CURSEG_NODE_TYPE
;
1709 for (i
= type
; i
< end
; i
++) {
1710 struct curseg_info
*sum
= CURSEG_I(sbi
, i
);
1711 mutex_lock(&sum
->curseg_mutex
);
1712 write_sum_page(sbi
, sum
->sum_blk
, blkaddr
+ (i
- type
));
1713 mutex_unlock(&sum
->curseg_mutex
);
1717 void write_data_summaries(struct f2fs_sb_info
*sbi
, block_t start_blk
)
1719 if (is_set_ckpt_flags(F2FS_CKPT(sbi
), CP_COMPACT_SUM_FLAG
))
1720 write_compacted_summaries(sbi
, start_blk
);
1722 write_normal_summaries(sbi
, start_blk
, CURSEG_HOT_DATA
);
1725 void write_node_summaries(struct f2fs_sb_info
*sbi
, block_t start_blk
)
1727 write_normal_summaries(sbi
, start_blk
, CURSEG_HOT_NODE
);
1730 int lookup_journal_in_cursum(struct f2fs_summary_block
*sum
, int type
,
1731 unsigned int val
, int alloc
)
1735 if (type
== NAT_JOURNAL
) {
1736 for (i
= 0; i
< nats_in_cursum(sum
); i
++) {
1737 if (le32_to_cpu(nid_in_journal(sum
, i
)) == val
)
1740 if (alloc
&& __has_cursum_space(sum
, 1, NAT_JOURNAL
))
1741 return update_nats_in_cursum(sum
, 1);
1742 } else if (type
== SIT_JOURNAL
) {
1743 for (i
= 0; i
< sits_in_cursum(sum
); i
++)
1744 if (le32_to_cpu(segno_in_journal(sum
, i
)) == val
)
1746 if (alloc
&& __has_cursum_space(sum
, 1, SIT_JOURNAL
))
1747 return update_sits_in_cursum(sum
, 1);
1752 static struct page
*get_current_sit_page(struct f2fs_sb_info
*sbi
,
1755 return get_meta_page(sbi
, current_sit_addr(sbi
, segno
));
1758 static struct page
*get_next_sit_page(struct f2fs_sb_info
*sbi
,
1761 struct sit_info
*sit_i
= SIT_I(sbi
);
1762 struct page
*src_page
, *dst_page
;
1763 pgoff_t src_off
, dst_off
;
1764 void *src_addr
, *dst_addr
;
1766 src_off
= current_sit_addr(sbi
, start
);
1767 dst_off
= next_sit_addr(sbi
, src_off
);
1769 /* get current sit block page without lock */
1770 src_page
= get_meta_page(sbi
, src_off
);
1771 dst_page
= grab_meta_page(sbi
, dst_off
);
1772 f2fs_bug_on(sbi
, PageDirty(src_page
));
1774 src_addr
= page_address(src_page
);
1775 dst_addr
= page_address(dst_page
);
1776 memcpy(dst_addr
, src_addr
, PAGE_CACHE_SIZE
);
1778 set_page_dirty(dst_page
);
1779 f2fs_put_page(src_page
, 1);
1781 set_to_next_sit(sit_i
, start
);
1786 static struct sit_entry_set
*grab_sit_entry_set(void)
1788 struct sit_entry_set
*ses
=
1789 f2fs_kmem_cache_alloc(sit_entry_set_slab
, GFP_NOFS
);
1792 INIT_LIST_HEAD(&ses
->set_list
);
1796 static void release_sit_entry_set(struct sit_entry_set
*ses
)
1798 list_del(&ses
->set_list
);
1799 kmem_cache_free(sit_entry_set_slab
, ses
);
1802 static void adjust_sit_entry_set(struct sit_entry_set
*ses
,
1803 struct list_head
*head
)
1805 struct sit_entry_set
*next
= ses
;
1807 if (list_is_last(&ses
->set_list
, head
))
1810 list_for_each_entry_continue(next
, head
, set_list
)
1811 if (ses
->entry_cnt
<= next
->entry_cnt
)
1814 list_move_tail(&ses
->set_list
, &next
->set_list
);
1817 static void add_sit_entry(unsigned int segno
, struct list_head
*head
)
1819 struct sit_entry_set
*ses
;
1820 unsigned int start_segno
= START_SEGNO(segno
);
1822 list_for_each_entry(ses
, head
, set_list
) {
1823 if (ses
->start_segno
== start_segno
) {
1825 adjust_sit_entry_set(ses
, head
);
1830 ses
= grab_sit_entry_set();
1832 ses
->start_segno
= start_segno
;
1834 list_add(&ses
->set_list
, head
);
1837 static void add_sits_in_set(struct f2fs_sb_info
*sbi
)
1839 struct f2fs_sm_info
*sm_info
= SM_I(sbi
);
1840 struct list_head
*set_list
= &sm_info
->sit_entry_set
;
1841 unsigned long *bitmap
= SIT_I(sbi
)->dirty_sentries_bitmap
;
1844 for_each_set_bit(segno
, bitmap
, MAIN_SEGS(sbi
))
1845 add_sit_entry(segno
, set_list
);
1848 static void remove_sits_in_journal(struct f2fs_sb_info
*sbi
)
1850 struct curseg_info
*curseg
= CURSEG_I(sbi
, CURSEG_COLD_DATA
);
1851 struct f2fs_summary_block
*sum
= curseg
->sum_blk
;
1854 for (i
= sits_in_cursum(sum
) - 1; i
>= 0; i
--) {
1858 segno
= le32_to_cpu(segno_in_journal(sum
, i
));
1859 dirtied
= __mark_sit_entry_dirty(sbi
, segno
);
1862 add_sit_entry(segno
, &SM_I(sbi
)->sit_entry_set
);
1864 update_sits_in_cursum(sum
, -sits_in_cursum(sum
));
1868 * CP calls this function, which flushes SIT entries including sit_journal,
1869 * and moves prefree segs to free segs.
1871 void flush_sit_entries(struct f2fs_sb_info
*sbi
, struct cp_control
*cpc
)
1873 struct sit_info
*sit_i
= SIT_I(sbi
);
1874 unsigned long *bitmap
= sit_i
->dirty_sentries_bitmap
;
1875 struct curseg_info
*curseg
= CURSEG_I(sbi
, CURSEG_COLD_DATA
);
1876 struct f2fs_summary_block
*sum
= curseg
->sum_blk
;
1877 struct sit_entry_set
*ses
, *tmp
;
1878 struct list_head
*head
= &SM_I(sbi
)->sit_entry_set
;
1879 bool to_journal
= true;
1880 struct seg_entry
*se
;
1882 mutex_lock(&curseg
->curseg_mutex
);
1883 mutex_lock(&sit_i
->sentry_lock
);
1885 if (!sit_i
->dirty_sentries
)
1889 * add and account sit entries of dirty bitmap in sit entry
1892 add_sits_in_set(sbi
);
1895 * if there are no enough space in journal to store dirty sit
1896 * entries, remove all entries from journal and add and account
1897 * them in sit entry set.
1899 if (!__has_cursum_space(sum
, sit_i
->dirty_sentries
, SIT_JOURNAL
))
1900 remove_sits_in_journal(sbi
);
1903 * there are two steps to flush sit entries:
1904 * #1, flush sit entries to journal in current cold data summary block.
1905 * #2, flush sit entries to sit page.
1907 list_for_each_entry_safe(ses
, tmp
, head
, set_list
) {
1908 struct page
*page
= NULL
;
1909 struct f2fs_sit_block
*raw_sit
= NULL
;
1910 unsigned int start_segno
= ses
->start_segno
;
1911 unsigned int end
= min(start_segno
+ SIT_ENTRY_PER_BLOCK
,
1912 (unsigned long)MAIN_SEGS(sbi
));
1913 unsigned int segno
= start_segno
;
1916 !__has_cursum_space(sum
, ses
->entry_cnt
, SIT_JOURNAL
))
1920 page
= get_next_sit_page(sbi
, start_segno
);
1921 raw_sit
= page_address(page
);
1924 /* flush dirty sit entries in region of current sit set */
1925 for_each_set_bit_from(segno
, bitmap
, end
) {
1926 int offset
, sit_offset
;
1928 se
= get_seg_entry(sbi
, segno
);
1930 /* add discard candidates */
1931 if (cpc
->reason
!= CP_DISCARD
) {
1932 cpc
->trim_start
= segno
;
1933 add_discard_addrs(sbi
, cpc
);
1937 offset
= lookup_journal_in_cursum(sum
,
1938 SIT_JOURNAL
, segno
, 1);
1939 f2fs_bug_on(sbi
, offset
< 0);
1940 segno_in_journal(sum
, offset
) =
1942 seg_info_to_raw_sit(se
,
1943 &sit_in_journal(sum
, offset
));
1945 sit_offset
= SIT_ENTRY_OFFSET(sit_i
, segno
);
1946 seg_info_to_raw_sit(se
,
1947 &raw_sit
->entries
[sit_offset
]);
1950 __clear_bit(segno
, bitmap
);
1951 sit_i
->dirty_sentries
--;
1956 f2fs_put_page(page
, 1);
1958 f2fs_bug_on(sbi
, ses
->entry_cnt
);
1959 release_sit_entry_set(ses
);
1962 f2fs_bug_on(sbi
, !list_empty(head
));
1963 f2fs_bug_on(sbi
, sit_i
->dirty_sentries
);
1965 if (cpc
->reason
== CP_DISCARD
) {
1966 for (; cpc
->trim_start
<= cpc
->trim_end
; cpc
->trim_start
++)
1967 add_discard_addrs(sbi
, cpc
);
1969 mutex_unlock(&sit_i
->sentry_lock
);
1970 mutex_unlock(&curseg
->curseg_mutex
);
1972 set_prefree_as_free_segments(sbi
);
1975 static int build_sit_info(struct f2fs_sb_info
*sbi
)
1977 struct f2fs_super_block
*raw_super
= F2FS_RAW_SUPER(sbi
);
1978 struct f2fs_checkpoint
*ckpt
= F2FS_CKPT(sbi
);
1979 struct sit_info
*sit_i
;
1980 unsigned int sit_segs
, start
;
1981 char *src_bitmap
, *dst_bitmap
;
1982 unsigned int bitmap_size
;
1984 /* allocate memory for SIT information */
1985 sit_i
= kzalloc(sizeof(struct sit_info
), GFP_KERNEL
);
1989 SM_I(sbi
)->sit_info
= sit_i
;
1991 sit_i
->sentries
= f2fs_kvzalloc(MAIN_SEGS(sbi
) *
1992 sizeof(struct seg_entry
), GFP_KERNEL
);
1993 if (!sit_i
->sentries
)
1996 bitmap_size
= f2fs_bitmap_size(MAIN_SEGS(sbi
));
1997 sit_i
->dirty_sentries_bitmap
= f2fs_kvzalloc(bitmap_size
, GFP_KERNEL
);
1998 if (!sit_i
->dirty_sentries_bitmap
)
2001 for (start
= 0; start
< MAIN_SEGS(sbi
); start
++) {
2002 sit_i
->sentries
[start
].cur_valid_map
2003 = kzalloc(SIT_VBLOCK_MAP_SIZE
, GFP_KERNEL
);
2004 sit_i
->sentries
[start
].ckpt_valid_map
2005 = kzalloc(SIT_VBLOCK_MAP_SIZE
, GFP_KERNEL
);
2006 sit_i
->sentries
[start
].discard_map
2007 = kzalloc(SIT_VBLOCK_MAP_SIZE
, GFP_KERNEL
);
2008 if (!sit_i
->sentries
[start
].cur_valid_map
||
2009 !sit_i
->sentries
[start
].ckpt_valid_map
||
2010 !sit_i
->sentries
[start
].discard_map
)
2014 sit_i
->tmp_map
= kzalloc(SIT_VBLOCK_MAP_SIZE
, GFP_KERNEL
);
2015 if (!sit_i
->tmp_map
)
2018 if (sbi
->segs_per_sec
> 1) {
2019 sit_i
->sec_entries
= f2fs_kvzalloc(MAIN_SECS(sbi
) *
2020 sizeof(struct sec_entry
), GFP_KERNEL
);
2021 if (!sit_i
->sec_entries
)
2025 /* get information related with SIT */
2026 sit_segs
= le32_to_cpu(raw_super
->segment_count_sit
) >> 1;
2028 /* setup SIT bitmap from ckeckpoint pack */
2029 bitmap_size
= __bitmap_size(sbi
, SIT_BITMAP
);
2030 src_bitmap
= __bitmap_ptr(sbi
, SIT_BITMAP
);
2032 dst_bitmap
= kmemdup(src_bitmap
, bitmap_size
, GFP_KERNEL
);
2036 /* init SIT information */
2037 sit_i
->s_ops
= &default_salloc_ops
;
2039 sit_i
->sit_base_addr
= le32_to_cpu(raw_super
->sit_blkaddr
);
2040 sit_i
->sit_blocks
= sit_segs
<< sbi
->log_blocks_per_seg
;
2041 sit_i
->written_valid_blocks
= le64_to_cpu(ckpt
->valid_block_count
);
2042 sit_i
->sit_bitmap
= dst_bitmap
;
2043 sit_i
->bitmap_size
= bitmap_size
;
2044 sit_i
->dirty_sentries
= 0;
2045 sit_i
->sents_per_block
= SIT_ENTRY_PER_BLOCK
;
2046 sit_i
->elapsed_time
= le64_to_cpu(sbi
->ckpt
->elapsed_time
);
2047 sit_i
->mounted_time
= CURRENT_TIME_SEC
.tv_sec
;
2048 mutex_init(&sit_i
->sentry_lock
);
2052 static int build_free_segmap(struct f2fs_sb_info
*sbi
)
2054 struct free_segmap_info
*free_i
;
2055 unsigned int bitmap_size
, sec_bitmap_size
;
2057 /* allocate memory for free segmap information */
2058 free_i
= kzalloc(sizeof(struct free_segmap_info
), GFP_KERNEL
);
2062 SM_I(sbi
)->free_info
= free_i
;
2064 bitmap_size
= f2fs_bitmap_size(MAIN_SEGS(sbi
));
2065 free_i
->free_segmap
= f2fs_kvmalloc(bitmap_size
, GFP_KERNEL
);
2066 if (!free_i
->free_segmap
)
2069 sec_bitmap_size
= f2fs_bitmap_size(MAIN_SECS(sbi
));
2070 free_i
->free_secmap
= f2fs_kvmalloc(sec_bitmap_size
, GFP_KERNEL
);
2071 if (!free_i
->free_secmap
)
2074 /* set all segments as dirty temporarily */
2075 memset(free_i
->free_segmap
, 0xff, bitmap_size
);
2076 memset(free_i
->free_secmap
, 0xff, sec_bitmap_size
);
2078 /* init free segmap information */
2079 free_i
->start_segno
= GET_SEGNO_FROM_SEG0(sbi
, MAIN_BLKADDR(sbi
));
2080 free_i
->free_segments
= 0;
2081 free_i
->free_sections
= 0;
2082 spin_lock_init(&free_i
->segmap_lock
);
2086 static int build_curseg(struct f2fs_sb_info
*sbi
)
2088 struct curseg_info
*array
;
2091 array
= kcalloc(NR_CURSEG_TYPE
, sizeof(*array
), GFP_KERNEL
);
2095 SM_I(sbi
)->curseg_array
= array
;
2097 for (i
= 0; i
< NR_CURSEG_TYPE
; i
++) {
2098 mutex_init(&array
[i
].curseg_mutex
);
2099 array
[i
].sum_blk
= kzalloc(PAGE_CACHE_SIZE
, GFP_KERNEL
);
2100 if (!array
[i
].sum_blk
)
2102 array
[i
].segno
= NULL_SEGNO
;
2103 array
[i
].next_blkoff
= 0;
2105 return restore_curseg_summaries(sbi
);
2108 static void build_sit_entries(struct f2fs_sb_info
*sbi
)
2110 struct sit_info
*sit_i
= SIT_I(sbi
);
2111 struct curseg_info
*curseg
= CURSEG_I(sbi
, CURSEG_COLD_DATA
);
2112 struct f2fs_summary_block
*sum
= curseg
->sum_blk
;
2113 int sit_blk_cnt
= SIT_BLK_CNT(sbi
);
2114 unsigned int i
, start
, end
;
2115 unsigned int readed
, start_blk
= 0;
2116 int nrpages
= MAX_BIO_BLOCKS(sbi
);
2119 readed
= ra_meta_pages(sbi
, start_blk
, nrpages
, META_SIT
, true);
2121 start
= start_blk
* sit_i
->sents_per_block
;
2122 end
= (start_blk
+ readed
) * sit_i
->sents_per_block
;
2124 for (; start
< end
&& start
< MAIN_SEGS(sbi
); start
++) {
2125 struct seg_entry
*se
= &sit_i
->sentries
[start
];
2126 struct f2fs_sit_block
*sit_blk
;
2127 struct f2fs_sit_entry sit
;
2130 mutex_lock(&curseg
->curseg_mutex
);
2131 for (i
= 0; i
< sits_in_cursum(sum
); i
++) {
2132 if (le32_to_cpu(segno_in_journal(sum
, i
))
2134 sit
= sit_in_journal(sum
, i
);
2135 mutex_unlock(&curseg
->curseg_mutex
);
2139 mutex_unlock(&curseg
->curseg_mutex
);
2141 page
= get_current_sit_page(sbi
, start
);
2142 sit_blk
= (struct f2fs_sit_block
*)page_address(page
);
2143 sit
= sit_blk
->entries
[SIT_ENTRY_OFFSET(sit_i
, start
)];
2144 f2fs_put_page(page
, 1);
2146 check_block_count(sbi
, start
, &sit
);
2147 seg_info_from_raw_sit(se
, &sit
);
2149 /* build discard map only one time */
2150 memcpy(se
->discard_map
, se
->cur_valid_map
, SIT_VBLOCK_MAP_SIZE
);
2151 sbi
->discard_blks
+= sbi
->blocks_per_seg
- se
->valid_blocks
;
2153 if (sbi
->segs_per_sec
> 1) {
2154 struct sec_entry
*e
= get_sec_entry(sbi
, start
);
2155 e
->valid_blocks
+= se
->valid_blocks
;
2158 start_blk
+= readed
;
2159 } while (start_blk
< sit_blk_cnt
);
2162 static void init_free_segmap(struct f2fs_sb_info
*sbi
)
2167 for (start
= 0; start
< MAIN_SEGS(sbi
); start
++) {
2168 struct seg_entry
*sentry
= get_seg_entry(sbi
, start
);
2169 if (!sentry
->valid_blocks
)
2170 __set_free(sbi
, start
);
2173 /* set use the current segments */
2174 for (type
= CURSEG_HOT_DATA
; type
<= CURSEG_COLD_NODE
; type
++) {
2175 struct curseg_info
*curseg_t
= CURSEG_I(sbi
, type
);
2176 __set_test_and_inuse(sbi
, curseg_t
->segno
);
2180 static void init_dirty_segmap(struct f2fs_sb_info
*sbi
)
2182 struct dirty_seglist_info
*dirty_i
= DIRTY_I(sbi
);
2183 struct free_segmap_info
*free_i
= FREE_I(sbi
);
2184 unsigned int segno
= 0, offset
= 0;
2185 unsigned short valid_blocks
;
2188 /* find dirty segment based on free segmap */
2189 segno
= find_next_inuse(free_i
, MAIN_SEGS(sbi
), offset
);
2190 if (segno
>= MAIN_SEGS(sbi
))
2193 valid_blocks
= get_valid_blocks(sbi
, segno
, 0);
2194 if (valid_blocks
== sbi
->blocks_per_seg
|| !valid_blocks
)
2196 if (valid_blocks
> sbi
->blocks_per_seg
) {
2197 f2fs_bug_on(sbi
, 1);
2200 mutex_lock(&dirty_i
->seglist_lock
);
2201 __locate_dirty_segment(sbi
, segno
, DIRTY
);
2202 mutex_unlock(&dirty_i
->seglist_lock
);
2206 static int init_victim_secmap(struct f2fs_sb_info
*sbi
)
2208 struct dirty_seglist_info
*dirty_i
= DIRTY_I(sbi
);
2209 unsigned int bitmap_size
= f2fs_bitmap_size(MAIN_SECS(sbi
));
2211 dirty_i
->victim_secmap
= f2fs_kvzalloc(bitmap_size
, GFP_KERNEL
);
2212 if (!dirty_i
->victim_secmap
)
2217 static int build_dirty_segmap(struct f2fs_sb_info
*sbi
)
2219 struct dirty_seglist_info
*dirty_i
;
2220 unsigned int bitmap_size
, i
;
2222 /* allocate memory for dirty segments list information */
2223 dirty_i
= kzalloc(sizeof(struct dirty_seglist_info
), GFP_KERNEL
);
2227 SM_I(sbi
)->dirty_info
= dirty_i
;
2228 mutex_init(&dirty_i
->seglist_lock
);
2230 bitmap_size
= f2fs_bitmap_size(MAIN_SEGS(sbi
));
2232 for (i
= 0; i
< NR_DIRTY_TYPE
; i
++) {
2233 dirty_i
->dirty_segmap
[i
] = f2fs_kvzalloc(bitmap_size
, GFP_KERNEL
);
2234 if (!dirty_i
->dirty_segmap
[i
])
2238 init_dirty_segmap(sbi
);
2239 return init_victim_secmap(sbi
);
2243 * Update min, max modified time for cost-benefit GC algorithm
2245 static void init_min_max_mtime(struct f2fs_sb_info
*sbi
)
2247 struct sit_info
*sit_i
= SIT_I(sbi
);
2250 mutex_lock(&sit_i
->sentry_lock
);
2252 sit_i
->min_mtime
= LLONG_MAX
;
2254 for (segno
= 0; segno
< MAIN_SEGS(sbi
); segno
+= sbi
->segs_per_sec
) {
2256 unsigned long long mtime
= 0;
2258 for (i
= 0; i
< sbi
->segs_per_sec
; i
++)
2259 mtime
+= get_seg_entry(sbi
, segno
+ i
)->mtime
;
2261 mtime
= div_u64(mtime
, sbi
->segs_per_sec
);
2263 if (sit_i
->min_mtime
> mtime
)
2264 sit_i
->min_mtime
= mtime
;
2266 sit_i
->max_mtime
= get_mtime(sbi
);
2267 mutex_unlock(&sit_i
->sentry_lock
);
2270 int build_segment_manager(struct f2fs_sb_info
*sbi
)
2272 struct f2fs_super_block
*raw_super
= F2FS_RAW_SUPER(sbi
);
2273 struct f2fs_checkpoint
*ckpt
= F2FS_CKPT(sbi
);
2274 struct f2fs_sm_info
*sm_info
;
2277 sm_info
= kzalloc(sizeof(struct f2fs_sm_info
), GFP_KERNEL
);
2282 sbi
->sm_info
= sm_info
;
2283 sm_info
->seg0_blkaddr
= le32_to_cpu(raw_super
->segment0_blkaddr
);
2284 sm_info
->main_blkaddr
= le32_to_cpu(raw_super
->main_blkaddr
);
2285 sm_info
->segment_count
= le32_to_cpu(raw_super
->segment_count
);
2286 sm_info
->reserved_segments
= le32_to_cpu(ckpt
->rsvd_segment_count
);
2287 sm_info
->ovp_segments
= le32_to_cpu(ckpt
->overprov_segment_count
);
2288 sm_info
->main_segments
= le32_to_cpu(raw_super
->segment_count_main
);
2289 sm_info
->ssa_blkaddr
= le32_to_cpu(raw_super
->ssa_blkaddr
);
2290 sm_info
->rec_prefree_segments
= sm_info
->main_segments
*
2291 DEF_RECLAIM_PREFREE_SEGMENTS
/ 100;
2292 sm_info
->ipu_policy
= 1 << F2FS_IPU_FSYNC
;
2293 sm_info
->min_ipu_util
= DEF_MIN_IPU_UTIL
;
2294 sm_info
->min_fsync_blocks
= DEF_MIN_FSYNC_BLOCKS
;
2296 INIT_LIST_HEAD(&sm_info
->discard_list
);
2297 sm_info
->nr_discards
= 0;
2298 sm_info
->max_discards
= 0;
2300 sm_info
->trim_sections
= DEF_BATCHED_TRIM_SECTIONS
;
2302 INIT_LIST_HEAD(&sm_info
->sit_entry_set
);
2304 if (test_opt(sbi
, FLUSH_MERGE
) && !f2fs_readonly(sbi
->sb
)) {
2305 err
= create_flush_cmd_control(sbi
);
2310 err
= build_sit_info(sbi
);
2313 err
= build_free_segmap(sbi
);
2316 err
= build_curseg(sbi
);
2320 /* reinit free segmap based on SIT */
2321 build_sit_entries(sbi
);
2323 init_free_segmap(sbi
);
2324 err
= build_dirty_segmap(sbi
);
2328 init_min_max_mtime(sbi
);
2332 static void discard_dirty_segmap(struct f2fs_sb_info
*sbi
,
2333 enum dirty_type dirty_type
)
2335 struct dirty_seglist_info
*dirty_i
= DIRTY_I(sbi
);
2337 mutex_lock(&dirty_i
->seglist_lock
);
2338 kvfree(dirty_i
->dirty_segmap
[dirty_type
]);
2339 dirty_i
->nr_dirty
[dirty_type
] = 0;
2340 mutex_unlock(&dirty_i
->seglist_lock
);
2343 static void destroy_victim_secmap(struct f2fs_sb_info
*sbi
)
2345 struct dirty_seglist_info
*dirty_i
= DIRTY_I(sbi
);
2346 kvfree(dirty_i
->victim_secmap
);
2349 static void destroy_dirty_segmap(struct f2fs_sb_info
*sbi
)
2351 struct dirty_seglist_info
*dirty_i
= DIRTY_I(sbi
);
2357 /* discard pre-free/dirty segments list */
2358 for (i
= 0; i
< NR_DIRTY_TYPE
; i
++)
2359 discard_dirty_segmap(sbi
, i
);
2361 destroy_victim_secmap(sbi
);
2362 SM_I(sbi
)->dirty_info
= NULL
;
2366 static void destroy_curseg(struct f2fs_sb_info
*sbi
)
2368 struct curseg_info
*array
= SM_I(sbi
)->curseg_array
;
2373 SM_I(sbi
)->curseg_array
= NULL
;
2374 for (i
= 0; i
< NR_CURSEG_TYPE
; i
++)
2375 kfree(array
[i
].sum_blk
);
2379 static void destroy_free_segmap(struct f2fs_sb_info
*sbi
)
2381 struct free_segmap_info
*free_i
= SM_I(sbi
)->free_info
;
2384 SM_I(sbi
)->free_info
= NULL
;
2385 kvfree(free_i
->free_segmap
);
2386 kvfree(free_i
->free_secmap
);
2390 static void destroy_sit_info(struct f2fs_sb_info
*sbi
)
2392 struct sit_info
*sit_i
= SIT_I(sbi
);
2398 if (sit_i
->sentries
) {
2399 for (start
= 0; start
< MAIN_SEGS(sbi
); start
++) {
2400 kfree(sit_i
->sentries
[start
].cur_valid_map
);
2401 kfree(sit_i
->sentries
[start
].ckpt_valid_map
);
2402 kfree(sit_i
->sentries
[start
].discard_map
);
2405 kfree(sit_i
->tmp_map
);
2407 kvfree(sit_i
->sentries
);
2408 kvfree(sit_i
->sec_entries
);
2409 kvfree(sit_i
->dirty_sentries_bitmap
);
2411 SM_I(sbi
)->sit_info
= NULL
;
2412 kfree(sit_i
->sit_bitmap
);
2416 void destroy_segment_manager(struct f2fs_sb_info
*sbi
)
2418 struct f2fs_sm_info
*sm_info
= SM_I(sbi
);
2422 destroy_flush_cmd_control(sbi
);
2423 destroy_dirty_segmap(sbi
);
2424 destroy_curseg(sbi
);
2425 destroy_free_segmap(sbi
);
2426 destroy_sit_info(sbi
);
2427 sbi
->sm_info
= NULL
;
2431 int __init
create_segment_manager_caches(void)
2433 discard_entry_slab
= f2fs_kmem_cache_create("discard_entry",
2434 sizeof(struct discard_entry
));
2435 if (!discard_entry_slab
)
2438 sit_entry_set_slab
= f2fs_kmem_cache_create("sit_entry_set",
2439 sizeof(struct sit_entry_set
));
2440 if (!sit_entry_set_slab
)
2441 goto destory_discard_entry
;
2443 inmem_entry_slab
= f2fs_kmem_cache_create("inmem_page_entry",
2444 sizeof(struct inmem_pages
));
2445 if (!inmem_entry_slab
)
2446 goto destroy_sit_entry_set
;
2449 destroy_sit_entry_set
:
2450 kmem_cache_destroy(sit_entry_set_slab
);
2451 destory_discard_entry
:
2452 kmem_cache_destroy(discard_entry_slab
);
2457 void destroy_segment_manager_caches(void)
2459 kmem_cache_destroy(sit_entry_set_slab
);
2460 kmem_cache_destroy(discard_entry_slab
);
2461 kmem_cache_destroy(inmem_entry_slab
);