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 !available_free_memory(sbi
, INO_ENTRIES
) ||
295 excess_prefree_segs(sbi
) ||
296 excess_dirty_nats(sbi
) ||
297 (is_idle(sbi
) && f2fs_time_over(sbi
, CP_TIME
))) {
298 if (test_opt(sbi
, DATA_FLUSH
))
299 sync_dirty_inodes(sbi
, FILE_INODE
);
300 f2fs_sync_fs(sbi
->sb
, true);
301 stat_inc_bg_cp_count(sbi
->stat_info
);
305 static int issue_flush_thread(void *data
)
307 struct f2fs_sb_info
*sbi
= data
;
308 struct flush_cmd_control
*fcc
= SM_I(sbi
)->cmd_control_info
;
309 wait_queue_head_t
*q
= &fcc
->flush_wait_queue
;
311 if (kthread_should_stop())
314 if (!llist_empty(&fcc
->issue_list
)) {
316 struct flush_cmd
*cmd
, *next
;
319 bio
= f2fs_bio_alloc(0);
321 fcc
->dispatch_list
= llist_del_all(&fcc
->issue_list
);
322 fcc
->dispatch_list
= llist_reverse_order(fcc
->dispatch_list
);
324 bio
->bi_bdev
= sbi
->sb
->s_bdev
;
325 ret
= submit_bio_wait(WRITE_FLUSH
, bio
);
327 llist_for_each_entry_safe(cmd
, next
,
328 fcc
->dispatch_list
, llnode
) {
330 complete(&cmd
->wait
);
333 fcc
->dispatch_list
= NULL
;
336 wait_event_interruptible(*q
,
337 kthread_should_stop() || !llist_empty(&fcc
->issue_list
));
341 int f2fs_issue_flush(struct f2fs_sb_info
*sbi
)
343 struct flush_cmd_control
*fcc
= SM_I(sbi
)->cmd_control_info
;
344 struct flush_cmd cmd
;
346 trace_f2fs_issue_flush(sbi
->sb
, test_opt(sbi
, NOBARRIER
),
347 test_opt(sbi
, FLUSH_MERGE
));
349 if (test_opt(sbi
, NOBARRIER
))
352 if (!test_opt(sbi
, FLUSH_MERGE
)) {
353 struct bio
*bio
= f2fs_bio_alloc(0);
356 bio
->bi_bdev
= sbi
->sb
->s_bdev
;
357 ret
= submit_bio_wait(WRITE_FLUSH
, bio
);
362 init_completion(&cmd
.wait
);
364 llist_add(&cmd
.llnode
, &fcc
->issue_list
);
366 if (!fcc
->dispatch_list
)
367 wake_up(&fcc
->flush_wait_queue
);
369 wait_for_completion(&cmd
.wait
);
374 int create_flush_cmd_control(struct f2fs_sb_info
*sbi
)
376 dev_t dev
= sbi
->sb
->s_bdev
->bd_dev
;
377 struct flush_cmd_control
*fcc
;
380 fcc
= kzalloc(sizeof(struct flush_cmd_control
), GFP_KERNEL
);
383 init_waitqueue_head(&fcc
->flush_wait_queue
);
384 init_llist_head(&fcc
->issue_list
);
385 SM_I(sbi
)->cmd_control_info
= fcc
;
386 fcc
->f2fs_issue_flush
= kthread_run(issue_flush_thread
, sbi
,
387 "f2fs_flush-%u:%u", MAJOR(dev
), MINOR(dev
));
388 if (IS_ERR(fcc
->f2fs_issue_flush
)) {
389 err
= PTR_ERR(fcc
->f2fs_issue_flush
);
391 SM_I(sbi
)->cmd_control_info
= NULL
;
398 void destroy_flush_cmd_control(struct f2fs_sb_info
*sbi
)
400 struct flush_cmd_control
*fcc
= SM_I(sbi
)->cmd_control_info
;
402 if (fcc
&& fcc
->f2fs_issue_flush
)
403 kthread_stop(fcc
->f2fs_issue_flush
);
405 SM_I(sbi
)->cmd_control_info
= NULL
;
408 static void __locate_dirty_segment(struct f2fs_sb_info
*sbi
, unsigned int segno
,
409 enum dirty_type dirty_type
)
411 struct dirty_seglist_info
*dirty_i
= DIRTY_I(sbi
);
413 /* need not be added */
414 if (IS_CURSEG(sbi
, segno
))
417 if (!test_and_set_bit(segno
, dirty_i
->dirty_segmap
[dirty_type
]))
418 dirty_i
->nr_dirty
[dirty_type
]++;
420 if (dirty_type
== DIRTY
) {
421 struct seg_entry
*sentry
= get_seg_entry(sbi
, segno
);
422 enum dirty_type t
= sentry
->type
;
424 if (unlikely(t
>= DIRTY
)) {
428 if (!test_and_set_bit(segno
, dirty_i
->dirty_segmap
[t
]))
429 dirty_i
->nr_dirty
[t
]++;
433 static void __remove_dirty_segment(struct f2fs_sb_info
*sbi
, unsigned int segno
,
434 enum dirty_type dirty_type
)
436 struct dirty_seglist_info
*dirty_i
= DIRTY_I(sbi
);
438 if (test_and_clear_bit(segno
, dirty_i
->dirty_segmap
[dirty_type
]))
439 dirty_i
->nr_dirty
[dirty_type
]--;
441 if (dirty_type
== DIRTY
) {
442 struct seg_entry
*sentry
= get_seg_entry(sbi
, segno
);
443 enum dirty_type t
= sentry
->type
;
445 if (test_and_clear_bit(segno
, dirty_i
->dirty_segmap
[t
]))
446 dirty_i
->nr_dirty
[t
]--;
448 if (get_valid_blocks(sbi
, segno
, sbi
->segs_per_sec
) == 0)
449 clear_bit(GET_SECNO(sbi
, segno
),
450 dirty_i
->victim_secmap
);
455 * Should not occur error such as -ENOMEM.
456 * Adding dirty entry into seglist is not critical operation.
457 * If a given segment is one of current working segments, it won't be added.
459 static void locate_dirty_segment(struct f2fs_sb_info
*sbi
, unsigned int segno
)
461 struct dirty_seglist_info
*dirty_i
= DIRTY_I(sbi
);
462 unsigned short valid_blocks
;
464 if (segno
== NULL_SEGNO
|| IS_CURSEG(sbi
, segno
))
467 mutex_lock(&dirty_i
->seglist_lock
);
469 valid_blocks
= get_valid_blocks(sbi
, segno
, 0);
471 if (valid_blocks
== 0) {
472 __locate_dirty_segment(sbi
, segno
, PRE
);
473 __remove_dirty_segment(sbi
, segno
, DIRTY
);
474 } else if (valid_blocks
< sbi
->blocks_per_seg
) {
475 __locate_dirty_segment(sbi
, segno
, DIRTY
);
477 /* Recovery routine with SSR needs this */
478 __remove_dirty_segment(sbi
, segno
, DIRTY
);
481 mutex_unlock(&dirty_i
->seglist_lock
);
484 static int f2fs_issue_discard(struct f2fs_sb_info
*sbi
,
485 block_t blkstart
, block_t blklen
)
487 sector_t start
= SECTOR_FROM_BLOCK(blkstart
);
488 sector_t len
= SECTOR_FROM_BLOCK(blklen
);
489 struct seg_entry
*se
;
493 for (i
= blkstart
; i
< blkstart
+ blklen
; i
++) {
494 se
= get_seg_entry(sbi
, GET_SEGNO(sbi
, i
));
495 offset
= GET_BLKOFF_FROM_SEG0(sbi
, i
);
497 if (!f2fs_test_and_set_bit(offset
, se
->discard_map
))
500 trace_f2fs_issue_discard(sbi
->sb
, blkstart
, blklen
);
501 return blkdev_issue_discard(sbi
->sb
->s_bdev
, start
, len
, GFP_NOFS
, 0);
504 bool discard_next_dnode(struct f2fs_sb_info
*sbi
, block_t blkaddr
)
508 if (test_opt(sbi
, DISCARD
)) {
509 struct seg_entry
*se
= get_seg_entry(sbi
,
510 GET_SEGNO(sbi
, blkaddr
));
511 unsigned int offset
= GET_BLKOFF_FROM_SEG0(sbi
, blkaddr
);
513 if (f2fs_test_bit(offset
, se
->discard_map
))
516 err
= f2fs_issue_discard(sbi
, blkaddr
, 1);
520 update_meta_page(sbi
, NULL
, blkaddr
);
526 static void __add_discard_entry(struct f2fs_sb_info
*sbi
,
527 struct cp_control
*cpc
, struct seg_entry
*se
,
528 unsigned int start
, unsigned int end
)
530 struct list_head
*head
= &SM_I(sbi
)->discard_list
;
531 struct discard_entry
*new, *last
;
533 if (!list_empty(head
)) {
534 last
= list_last_entry(head
, struct discard_entry
, list
);
535 if (START_BLOCK(sbi
, cpc
->trim_start
) + start
==
536 last
->blkaddr
+ last
->len
) {
537 last
->len
+= end
- start
;
542 new = f2fs_kmem_cache_alloc(discard_entry_slab
, GFP_NOFS
);
543 INIT_LIST_HEAD(&new->list
);
544 new->blkaddr
= START_BLOCK(sbi
, cpc
->trim_start
) + start
;
545 new->len
= end
- start
;
546 list_add_tail(&new->list
, head
);
548 SM_I(sbi
)->nr_discards
+= end
- start
;
551 static void add_discard_addrs(struct f2fs_sb_info
*sbi
, struct cp_control
*cpc
)
553 int entries
= SIT_VBLOCK_MAP_SIZE
/ sizeof(unsigned long);
554 int max_blocks
= sbi
->blocks_per_seg
;
555 struct seg_entry
*se
= get_seg_entry(sbi
, cpc
->trim_start
);
556 unsigned long *cur_map
= (unsigned long *)se
->cur_valid_map
;
557 unsigned long *ckpt_map
= (unsigned long *)se
->ckpt_valid_map
;
558 unsigned long *discard_map
= (unsigned long *)se
->discard_map
;
559 unsigned long *dmap
= SIT_I(sbi
)->tmp_map
;
560 unsigned int start
= 0, end
= -1;
561 bool force
= (cpc
->reason
== CP_DISCARD
);
564 if (se
->valid_blocks
== max_blocks
)
568 if (!test_opt(sbi
, DISCARD
) || !se
->valid_blocks
||
569 SM_I(sbi
)->nr_discards
>= SM_I(sbi
)->max_discards
)
573 /* SIT_VBLOCK_MAP_SIZE should be multiple of sizeof(unsigned long) */
574 for (i
= 0; i
< entries
; i
++)
575 dmap
[i
] = force
? ~ckpt_map
[i
] & ~discard_map
[i
] :
576 (cur_map
[i
] ^ ckpt_map
[i
]) & ckpt_map
[i
];
578 while (force
|| SM_I(sbi
)->nr_discards
<= SM_I(sbi
)->max_discards
) {
579 start
= __find_rev_next_bit(dmap
, max_blocks
, end
+ 1);
580 if (start
>= max_blocks
)
583 end
= __find_rev_next_zero_bit(dmap
, max_blocks
, start
+ 1);
584 __add_discard_entry(sbi
, cpc
, se
, start
, end
);
588 void release_discard_addrs(struct f2fs_sb_info
*sbi
)
590 struct list_head
*head
= &(SM_I(sbi
)->discard_list
);
591 struct discard_entry
*entry
, *this;
594 list_for_each_entry_safe(entry
, this, head
, list
) {
595 list_del(&entry
->list
);
596 kmem_cache_free(discard_entry_slab
, entry
);
601 * Should call clear_prefree_segments after checkpoint is done.
603 static void set_prefree_as_free_segments(struct f2fs_sb_info
*sbi
)
605 struct dirty_seglist_info
*dirty_i
= DIRTY_I(sbi
);
608 mutex_lock(&dirty_i
->seglist_lock
);
609 for_each_set_bit(segno
, dirty_i
->dirty_segmap
[PRE
], MAIN_SEGS(sbi
))
610 __set_test_and_free(sbi
, segno
);
611 mutex_unlock(&dirty_i
->seglist_lock
);
614 void clear_prefree_segments(struct f2fs_sb_info
*sbi
, struct cp_control
*cpc
)
616 struct list_head
*head
= &(SM_I(sbi
)->discard_list
);
617 struct discard_entry
*entry
, *this;
618 struct dirty_seglist_info
*dirty_i
= DIRTY_I(sbi
);
619 unsigned long *prefree_map
= dirty_i
->dirty_segmap
[PRE
];
620 unsigned int start
= 0, end
= -1;
622 mutex_lock(&dirty_i
->seglist_lock
);
626 start
= find_next_bit(prefree_map
, MAIN_SEGS(sbi
), end
+ 1);
627 if (start
>= MAIN_SEGS(sbi
))
629 end
= find_next_zero_bit(prefree_map
, MAIN_SEGS(sbi
),
632 for (i
= start
; i
< end
; i
++)
633 clear_bit(i
, prefree_map
);
635 dirty_i
->nr_dirty
[PRE
] -= end
- start
;
637 if (!test_opt(sbi
, DISCARD
))
640 f2fs_issue_discard(sbi
, START_BLOCK(sbi
, start
),
641 (end
- start
) << sbi
->log_blocks_per_seg
);
643 mutex_unlock(&dirty_i
->seglist_lock
);
645 /* send small discards */
646 list_for_each_entry_safe(entry
, this, head
, list
) {
647 if (cpc
->reason
== CP_DISCARD
&& entry
->len
< cpc
->trim_minlen
)
649 f2fs_issue_discard(sbi
, entry
->blkaddr
, entry
->len
);
650 cpc
->trimmed
+= entry
->len
;
652 list_del(&entry
->list
);
653 SM_I(sbi
)->nr_discards
-= entry
->len
;
654 kmem_cache_free(discard_entry_slab
, entry
);
658 static bool __mark_sit_entry_dirty(struct f2fs_sb_info
*sbi
, unsigned int segno
)
660 struct sit_info
*sit_i
= SIT_I(sbi
);
662 if (!__test_and_set_bit(segno
, sit_i
->dirty_sentries_bitmap
)) {
663 sit_i
->dirty_sentries
++;
670 static void __set_sit_entry_type(struct f2fs_sb_info
*sbi
, int type
,
671 unsigned int segno
, int modified
)
673 struct seg_entry
*se
= get_seg_entry(sbi
, segno
);
676 __mark_sit_entry_dirty(sbi
, segno
);
679 static void update_sit_entry(struct f2fs_sb_info
*sbi
, block_t blkaddr
, int del
)
681 struct seg_entry
*se
;
682 unsigned int segno
, offset
;
683 long int new_vblocks
;
685 segno
= GET_SEGNO(sbi
, blkaddr
);
687 se
= get_seg_entry(sbi
, segno
);
688 new_vblocks
= se
->valid_blocks
+ del
;
689 offset
= GET_BLKOFF_FROM_SEG0(sbi
, blkaddr
);
691 f2fs_bug_on(sbi
, (new_vblocks
>> (sizeof(unsigned short) << 3) ||
692 (new_vblocks
> sbi
->blocks_per_seg
)));
694 se
->valid_blocks
= new_vblocks
;
695 se
->mtime
= get_mtime(sbi
);
696 SIT_I(sbi
)->max_mtime
= se
->mtime
;
698 /* Update valid block bitmap */
700 if (f2fs_test_and_set_bit(offset
, se
->cur_valid_map
))
702 if (!f2fs_test_and_set_bit(offset
, se
->discard_map
))
705 if (!f2fs_test_and_clear_bit(offset
, se
->cur_valid_map
))
707 if (f2fs_test_and_clear_bit(offset
, se
->discard_map
))
710 if (!f2fs_test_bit(offset
, se
->ckpt_valid_map
))
711 se
->ckpt_valid_blocks
+= del
;
713 __mark_sit_entry_dirty(sbi
, segno
);
715 /* update total number of valid blocks to be written in ckpt area */
716 SIT_I(sbi
)->written_valid_blocks
+= del
;
718 if (sbi
->segs_per_sec
> 1)
719 get_sec_entry(sbi
, segno
)->valid_blocks
+= del
;
722 void refresh_sit_entry(struct f2fs_sb_info
*sbi
, block_t old
, block_t
new)
724 update_sit_entry(sbi
, new, 1);
725 if (GET_SEGNO(sbi
, old
) != NULL_SEGNO
)
726 update_sit_entry(sbi
, old
, -1);
728 locate_dirty_segment(sbi
, GET_SEGNO(sbi
, old
));
729 locate_dirty_segment(sbi
, GET_SEGNO(sbi
, new));
732 void invalidate_blocks(struct f2fs_sb_info
*sbi
, block_t addr
)
734 unsigned int segno
= GET_SEGNO(sbi
, addr
);
735 struct sit_info
*sit_i
= SIT_I(sbi
);
737 f2fs_bug_on(sbi
, addr
== NULL_ADDR
);
738 if (addr
== NEW_ADDR
)
741 /* add it into sit main buffer */
742 mutex_lock(&sit_i
->sentry_lock
);
744 update_sit_entry(sbi
, addr
, -1);
746 /* add it into dirty seglist */
747 locate_dirty_segment(sbi
, segno
);
749 mutex_unlock(&sit_i
->sentry_lock
);
752 bool is_checkpointed_data(struct f2fs_sb_info
*sbi
, block_t blkaddr
)
754 struct sit_info
*sit_i
= SIT_I(sbi
);
755 unsigned int segno
, offset
;
756 struct seg_entry
*se
;
759 if (blkaddr
== NEW_ADDR
|| blkaddr
== NULL_ADDR
)
762 mutex_lock(&sit_i
->sentry_lock
);
764 segno
= GET_SEGNO(sbi
, blkaddr
);
765 se
= get_seg_entry(sbi
, segno
);
766 offset
= GET_BLKOFF_FROM_SEG0(sbi
, blkaddr
);
768 if (f2fs_test_bit(offset
, se
->ckpt_valid_map
))
771 mutex_unlock(&sit_i
->sentry_lock
);
777 * This function should be resided under the curseg_mutex lock
779 static void __add_sum_entry(struct f2fs_sb_info
*sbi
, int type
,
780 struct f2fs_summary
*sum
)
782 struct curseg_info
*curseg
= CURSEG_I(sbi
, type
);
783 void *addr
= curseg
->sum_blk
;
784 addr
+= curseg
->next_blkoff
* sizeof(struct f2fs_summary
);
785 memcpy(addr
, sum
, sizeof(struct f2fs_summary
));
789 * Calculate the number of current summary pages for writing
791 int npages_for_summary_flush(struct f2fs_sb_info
*sbi
, bool for_ra
)
793 int valid_sum_count
= 0;
796 for (i
= CURSEG_HOT_DATA
; i
<= CURSEG_COLD_DATA
; i
++) {
797 if (sbi
->ckpt
->alloc_type
[i
] == SSR
)
798 valid_sum_count
+= sbi
->blocks_per_seg
;
801 valid_sum_count
+= le16_to_cpu(
802 F2FS_CKPT(sbi
)->cur_data_blkoff
[i
]);
804 valid_sum_count
+= curseg_blkoff(sbi
, i
);
808 sum_in_page
= (PAGE_CACHE_SIZE
- 2 * SUM_JOURNAL_SIZE
-
809 SUM_FOOTER_SIZE
) / SUMMARY_SIZE
;
810 if (valid_sum_count
<= sum_in_page
)
812 else if ((valid_sum_count
- sum_in_page
) <=
813 (PAGE_CACHE_SIZE
- SUM_FOOTER_SIZE
) / SUMMARY_SIZE
)
819 * Caller should put this summary page
821 struct page
*get_sum_page(struct f2fs_sb_info
*sbi
, unsigned int segno
)
823 return get_meta_page(sbi
, GET_SUM_BLOCK(sbi
, segno
));
826 void update_meta_page(struct f2fs_sb_info
*sbi
, void *src
, block_t blk_addr
)
828 struct page
*page
= grab_meta_page(sbi
, blk_addr
);
829 void *dst
= page_address(page
);
832 memcpy(dst
, src
, PAGE_CACHE_SIZE
);
834 memset(dst
, 0, PAGE_CACHE_SIZE
);
835 set_page_dirty(page
);
836 f2fs_put_page(page
, 1);
839 static void write_sum_page(struct f2fs_sb_info
*sbi
,
840 struct f2fs_summary_block
*sum_blk
, block_t blk_addr
)
842 update_meta_page(sbi
, (void *)sum_blk
, blk_addr
);
845 static int is_next_segment_free(struct f2fs_sb_info
*sbi
, int type
)
847 struct curseg_info
*curseg
= CURSEG_I(sbi
, type
);
848 unsigned int segno
= curseg
->segno
+ 1;
849 struct free_segmap_info
*free_i
= FREE_I(sbi
);
851 if (segno
< MAIN_SEGS(sbi
) && segno
% sbi
->segs_per_sec
)
852 return !test_bit(segno
, free_i
->free_segmap
);
857 * Find a new segment from the free segments bitmap to right order
858 * This function should be returned with success, otherwise BUG
860 static void get_new_segment(struct f2fs_sb_info
*sbi
,
861 unsigned int *newseg
, bool new_sec
, int dir
)
863 struct free_segmap_info
*free_i
= FREE_I(sbi
);
864 unsigned int segno
, secno
, zoneno
;
865 unsigned int total_zones
= MAIN_SECS(sbi
) / sbi
->secs_per_zone
;
866 unsigned int hint
= *newseg
/ sbi
->segs_per_sec
;
867 unsigned int old_zoneno
= GET_ZONENO_FROM_SEGNO(sbi
, *newseg
);
868 unsigned int left_start
= hint
;
873 spin_lock(&free_i
->segmap_lock
);
875 if (!new_sec
&& ((*newseg
+ 1) % sbi
->segs_per_sec
)) {
876 segno
= find_next_zero_bit(free_i
->free_segmap
,
877 MAIN_SEGS(sbi
), *newseg
+ 1);
878 if (segno
- *newseg
< sbi
->segs_per_sec
-
879 (*newseg
% sbi
->segs_per_sec
))
883 secno
= find_next_zero_bit(free_i
->free_secmap
, MAIN_SECS(sbi
), hint
);
884 if (secno
>= MAIN_SECS(sbi
)) {
885 if (dir
== ALLOC_RIGHT
) {
886 secno
= find_next_zero_bit(free_i
->free_secmap
,
888 f2fs_bug_on(sbi
, secno
>= MAIN_SECS(sbi
));
891 left_start
= hint
- 1;
897 while (test_bit(left_start
, free_i
->free_secmap
)) {
898 if (left_start
> 0) {
902 left_start
= find_next_zero_bit(free_i
->free_secmap
,
904 f2fs_bug_on(sbi
, left_start
>= MAIN_SECS(sbi
));
910 segno
= secno
* sbi
->segs_per_sec
;
911 zoneno
= secno
/ sbi
->secs_per_zone
;
913 /* give up on finding another zone */
916 if (sbi
->secs_per_zone
== 1)
918 if (zoneno
== old_zoneno
)
920 if (dir
== ALLOC_LEFT
) {
921 if (!go_left
&& zoneno
+ 1 >= total_zones
)
923 if (go_left
&& zoneno
== 0)
926 for (i
= 0; i
< NR_CURSEG_TYPE
; i
++)
927 if (CURSEG_I(sbi
, i
)->zone
== zoneno
)
930 if (i
< NR_CURSEG_TYPE
) {
931 /* zone is in user, try another */
933 hint
= zoneno
* sbi
->secs_per_zone
- 1;
934 else if (zoneno
+ 1 >= total_zones
)
937 hint
= (zoneno
+ 1) * sbi
->secs_per_zone
;
939 goto find_other_zone
;
942 /* set it as dirty segment in free segmap */
943 f2fs_bug_on(sbi
, test_bit(segno
, free_i
->free_segmap
));
944 __set_inuse(sbi
, segno
);
946 spin_unlock(&free_i
->segmap_lock
);
949 static void reset_curseg(struct f2fs_sb_info
*sbi
, int type
, int modified
)
951 struct curseg_info
*curseg
= CURSEG_I(sbi
, type
);
952 struct summary_footer
*sum_footer
;
954 curseg
->segno
= curseg
->next_segno
;
955 curseg
->zone
= GET_ZONENO_FROM_SEGNO(sbi
, curseg
->segno
);
956 curseg
->next_blkoff
= 0;
957 curseg
->next_segno
= NULL_SEGNO
;
959 sum_footer
= &(curseg
->sum_blk
->footer
);
960 memset(sum_footer
, 0, sizeof(struct summary_footer
));
961 if (IS_DATASEG(type
))
962 SET_SUM_TYPE(sum_footer
, SUM_TYPE_DATA
);
963 if (IS_NODESEG(type
))
964 SET_SUM_TYPE(sum_footer
, SUM_TYPE_NODE
);
965 __set_sit_entry_type(sbi
, type
, curseg
->segno
, modified
);
969 * Allocate a current working segment.
970 * This function always allocates a free segment in LFS manner.
972 static void new_curseg(struct f2fs_sb_info
*sbi
, int type
, bool new_sec
)
974 struct curseg_info
*curseg
= CURSEG_I(sbi
, type
);
975 unsigned int segno
= curseg
->segno
;
976 int dir
= ALLOC_LEFT
;
978 write_sum_page(sbi
, curseg
->sum_blk
,
979 GET_SUM_BLOCK(sbi
, segno
));
980 if (type
== CURSEG_WARM_DATA
|| type
== CURSEG_COLD_DATA
)
983 if (test_opt(sbi
, NOHEAP
))
986 get_new_segment(sbi
, &segno
, new_sec
, dir
);
987 curseg
->next_segno
= segno
;
988 reset_curseg(sbi
, type
, 1);
989 curseg
->alloc_type
= LFS
;
992 static void __next_free_blkoff(struct f2fs_sb_info
*sbi
,
993 struct curseg_info
*seg
, block_t start
)
995 struct seg_entry
*se
= get_seg_entry(sbi
, seg
->segno
);
996 int entries
= SIT_VBLOCK_MAP_SIZE
/ sizeof(unsigned long);
997 unsigned long *target_map
= SIT_I(sbi
)->tmp_map
;
998 unsigned long *ckpt_map
= (unsigned long *)se
->ckpt_valid_map
;
999 unsigned long *cur_map
= (unsigned long *)se
->cur_valid_map
;
1002 for (i
= 0; i
< entries
; i
++)
1003 target_map
[i
] = ckpt_map
[i
] | cur_map
[i
];
1005 pos
= __find_rev_next_zero_bit(target_map
, sbi
->blocks_per_seg
, start
);
1007 seg
->next_blkoff
= pos
;
1011 * If a segment is written by LFS manner, next block offset is just obtained
1012 * by increasing the current block offset. However, if a segment is written by
1013 * SSR manner, next block offset obtained by calling __next_free_blkoff
1015 static void __refresh_next_blkoff(struct f2fs_sb_info
*sbi
,
1016 struct curseg_info
*seg
)
1018 if (seg
->alloc_type
== SSR
)
1019 __next_free_blkoff(sbi
, seg
, seg
->next_blkoff
+ 1);
1025 * This function always allocates a used segment(from dirty seglist) by SSR
1026 * manner, so it should recover the existing segment information of valid blocks
1028 static void change_curseg(struct f2fs_sb_info
*sbi
, int type
, bool reuse
)
1030 struct dirty_seglist_info
*dirty_i
= DIRTY_I(sbi
);
1031 struct curseg_info
*curseg
= CURSEG_I(sbi
, type
);
1032 unsigned int new_segno
= curseg
->next_segno
;
1033 struct f2fs_summary_block
*sum_node
;
1034 struct page
*sum_page
;
1036 write_sum_page(sbi
, curseg
->sum_blk
,
1037 GET_SUM_BLOCK(sbi
, curseg
->segno
));
1038 __set_test_and_inuse(sbi
, new_segno
);
1040 mutex_lock(&dirty_i
->seglist_lock
);
1041 __remove_dirty_segment(sbi
, new_segno
, PRE
);
1042 __remove_dirty_segment(sbi
, new_segno
, DIRTY
);
1043 mutex_unlock(&dirty_i
->seglist_lock
);
1045 reset_curseg(sbi
, type
, 1);
1046 curseg
->alloc_type
= SSR
;
1047 __next_free_blkoff(sbi
, curseg
, 0);
1050 sum_page
= get_sum_page(sbi
, new_segno
);
1051 sum_node
= (struct f2fs_summary_block
*)page_address(sum_page
);
1052 memcpy(curseg
->sum_blk
, sum_node
, SUM_ENTRY_SIZE
);
1053 f2fs_put_page(sum_page
, 1);
1057 static int get_ssr_segment(struct f2fs_sb_info
*sbi
, int type
)
1059 struct curseg_info
*curseg
= CURSEG_I(sbi
, type
);
1060 const struct victim_selection
*v_ops
= DIRTY_I(sbi
)->v_ops
;
1062 if (IS_NODESEG(type
) || !has_not_enough_free_secs(sbi
, 0))
1063 return v_ops
->get_victim(sbi
,
1064 &(curseg
)->next_segno
, BG_GC
, type
, SSR
);
1066 /* For data segments, let's do SSR more intensively */
1067 for (; type
>= CURSEG_HOT_DATA
; type
--)
1068 if (v_ops
->get_victim(sbi
, &(curseg
)->next_segno
,
1075 * flush out current segment and replace it with new segment
1076 * This function should be returned with success, otherwise BUG
1078 static void allocate_segment_by_default(struct f2fs_sb_info
*sbi
,
1079 int type
, bool force
)
1081 struct curseg_info
*curseg
= CURSEG_I(sbi
, type
);
1084 new_curseg(sbi
, type
, true);
1085 else if (type
== CURSEG_WARM_NODE
)
1086 new_curseg(sbi
, type
, false);
1087 else if (curseg
->alloc_type
== LFS
&& is_next_segment_free(sbi
, type
))
1088 new_curseg(sbi
, type
, false);
1089 else if (need_SSR(sbi
) && get_ssr_segment(sbi
, type
))
1090 change_curseg(sbi
, type
, true);
1092 new_curseg(sbi
, type
, false);
1094 stat_inc_seg_type(sbi
, curseg
);
1097 static void __allocate_new_segments(struct f2fs_sb_info
*sbi
, int type
)
1099 struct curseg_info
*curseg
= CURSEG_I(sbi
, type
);
1100 unsigned int old_segno
;
1102 old_segno
= curseg
->segno
;
1103 SIT_I(sbi
)->s_ops
->allocate_segment(sbi
, type
, true);
1104 locate_dirty_segment(sbi
, old_segno
);
1107 void allocate_new_segments(struct f2fs_sb_info
*sbi
)
1111 for (i
= CURSEG_HOT_DATA
; i
<= CURSEG_COLD_DATA
; i
++)
1112 __allocate_new_segments(sbi
, i
);
1115 static const struct segment_allocation default_salloc_ops
= {
1116 .allocate_segment
= allocate_segment_by_default
,
1119 int f2fs_trim_fs(struct f2fs_sb_info
*sbi
, struct fstrim_range
*range
)
1121 __u64 start
= F2FS_BYTES_TO_BLK(range
->start
);
1122 __u64 end
= start
+ F2FS_BYTES_TO_BLK(range
->len
) - 1;
1123 unsigned int start_segno
, end_segno
;
1124 struct cp_control cpc
;
1127 if (start
>= MAX_BLKADDR(sbi
) || range
->len
< sbi
->blocksize
)
1131 if (end
<= MAIN_BLKADDR(sbi
))
1134 /* start/end segment number in main_area */
1135 start_segno
= (start
<= MAIN_BLKADDR(sbi
)) ? 0 : GET_SEGNO(sbi
, start
);
1136 end_segno
= (end
>= MAX_BLKADDR(sbi
)) ? MAIN_SEGS(sbi
) - 1 :
1137 GET_SEGNO(sbi
, end
);
1138 cpc
.reason
= CP_DISCARD
;
1139 cpc
.trim_minlen
= max_t(__u64
, 1, F2FS_BYTES_TO_BLK(range
->minlen
));
1141 /* do checkpoint to issue discard commands safely */
1142 for (; start_segno
<= end_segno
; start_segno
= cpc
.trim_end
+ 1) {
1143 cpc
.trim_start
= start_segno
;
1145 if (sbi
->discard_blks
== 0)
1147 else if (sbi
->discard_blks
< BATCHED_TRIM_BLOCKS(sbi
))
1148 cpc
.trim_end
= end_segno
;
1150 cpc
.trim_end
= min_t(unsigned int,
1151 rounddown(start_segno
+
1152 BATCHED_TRIM_SEGMENTS(sbi
),
1153 sbi
->segs_per_sec
) - 1, end_segno
);
1155 mutex_lock(&sbi
->gc_mutex
);
1156 err
= write_checkpoint(sbi
, &cpc
);
1157 mutex_unlock(&sbi
->gc_mutex
);
1160 range
->len
= F2FS_BLK_TO_BYTES(cpc
.trimmed
);
1164 static bool __has_curseg_space(struct f2fs_sb_info
*sbi
, int type
)
1166 struct curseg_info
*curseg
= CURSEG_I(sbi
, type
);
1167 if (curseg
->next_blkoff
< sbi
->blocks_per_seg
)
1172 static int __get_segment_type_2(struct page
*page
, enum page_type p_type
)
1175 return CURSEG_HOT_DATA
;
1177 return CURSEG_HOT_NODE
;
1180 static int __get_segment_type_4(struct page
*page
, enum page_type p_type
)
1182 if (p_type
== DATA
) {
1183 struct inode
*inode
= page
->mapping
->host
;
1185 if (S_ISDIR(inode
->i_mode
))
1186 return CURSEG_HOT_DATA
;
1188 return CURSEG_COLD_DATA
;
1190 if (IS_DNODE(page
) && is_cold_node(page
))
1191 return CURSEG_WARM_NODE
;
1193 return CURSEG_COLD_NODE
;
1197 static int __get_segment_type_6(struct page
*page
, enum page_type p_type
)
1199 if (p_type
== DATA
) {
1200 struct inode
*inode
= page
->mapping
->host
;
1202 if (S_ISDIR(inode
->i_mode
))
1203 return CURSEG_HOT_DATA
;
1204 else if (is_cold_data(page
) || file_is_cold(inode
))
1205 return CURSEG_COLD_DATA
;
1207 return CURSEG_WARM_DATA
;
1210 return is_cold_node(page
) ? CURSEG_WARM_NODE
:
1213 return CURSEG_COLD_NODE
;
1217 static int __get_segment_type(struct page
*page
, enum page_type p_type
)
1219 switch (F2FS_P_SB(page
)->active_logs
) {
1221 return __get_segment_type_2(page
, p_type
);
1223 return __get_segment_type_4(page
, p_type
);
1225 /* NR_CURSEG_TYPE(6) logs by default */
1226 f2fs_bug_on(F2FS_P_SB(page
),
1227 F2FS_P_SB(page
)->active_logs
!= NR_CURSEG_TYPE
);
1228 return __get_segment_type_6(page
, p_type
);
1231 void allocate_data_block(struct f2fs_sb_info
*sbi
, struct page
*page
,
1232 block_t old_blkaddr
, block_t
*new_blkaddr
,
1233 struct f2fs_summary
*sum
, int type
)
1235 struct sit_info
*sit_i
= SIT_I(sbi
);
1236 struct curseg_info
*curseg
;
1237 bool direct_io
= (type
== CURSEG_DIRECT_IO
);
1239 type
= direct_io
? CURSEG_WARM_DATA
: type
;
1241 curseg
= CURSEG_I(sbi
, type
);
1243 mutex_lock(&curseg
->curseg_mutex
);
1244 mutex_lock(&sit_i
->sentry_lock
);
1246 /* direct_io'ed data is aligned to the segment for better performance */
1247 if (direct_io
&& curseg
->next_blkoff
&&
1248 !has_not_enough_free_secs(sbi
, 0))
1249 __allocate_new_segments(sbi
, type
);
1251 *new_blkaddr
= NEXT_FREE_BLKADDR(sbi
, curseg
);
1254 * __add_sum_entry should be resided under the curseg_mutex
1255 * because, this function updates a summary entry in the
1256 * current summary block.
1258 __add_sum_entry(sbi
, type
, sum
);
1260 __refresh_next_blkoff(sbi
, curseg
);
1262 stat_inc_block_count(sbi
, curseg
);
1264 if (!__has_curseg_space(sbi
, type
))
1265 sit_i
->s_ops
->allocate_segment(sbi
, type
, false);
1267 * SIT information should be updated before segment allocation,
1268 * since SSR needs latest valid block information.
1270 refresh_sit_entry(sbi
, old_blkaddr
, *new_blkaddr
);
1272 mutex_unlock(&sit_i
->sentry_lock
);
1274 if (page
&& IS_NODESEG(type
))
1275 fill_node_footer_blkaddr(page
, NEXT_FREE_BLKADDR(sbi
, curseg
));
1277 mutex_unlock(&curseg
->curseg_mutex
);
1280 static void do_write_page(struct f2fs_summary
*sum
, struct f2fs_io_info
*fio
)
1282 int type
= __get_segment_type(fio
->page
, fio
->type
);
1284 allocate_data_block(fio
->sbi
, fio
->page
, fio
->blk_addr
,
1285 &fio
->blk_addr
, sum
, type
);
1287 /* writeout dirty page into bdev */
1288 f2fs_submit_page_mbio(fio
);
1291 void write_meta_page(struct f2fs_sb_info
*sbi
, struct page
*page
)
1293 struct f2fs_io_info fio
= {
1296 .rw
= WRITE_SYNC
| REQ_META
| REQ_PRIO
,
1297 .blk_addr
= page
->index
,
1299 .encrypted_page
= NULL
,
1302 if (unlikely(page
->index
>= MAIN_BLKADDR(sbi
)))
1303 fio
.rw
&= ~REQ_META
;
1305 set_page_writeback(page
);
1306 f2fs_submit_page_mbio(&fio
);
1309 void write_node_page(unsigned int nid
, struct f2fs_io_info
*fio
)
1311 struct f2fs_summary sum
;
1313 set_summary(&sum
, nid
, 0, 0);
1314 do_write_page(&sum
, fio
);
1317 void write_data_page(struct dnode_of_data
*dn
, struct f2fs_io_info
*fio
)
1319 struct f2fs_sb_info
*sbi
= fio
->sbi
;
1320 struct f2fs_summary sum
;
1321 struct node_info ni
;
1323 f2fs_bug_on(sbi
, dn
->data_blkaddr
== NULL_ADDR
);
1324 get_node_info(sbi
, dn
->nid
, &ni
);
1325 set_summary(&sum
, dn
->nid
, dn
->ofs_in_node
, ni
.version
);
1326 do_write_page(&sum
, fio
);
1327 dn
->data_blkaddr
= fio
->blk_addr
;
1330 void rewrite_data_page(struct f2fs_io_info
*fio
)
1332 stat_inc_inplace_blocks(fio
->sbi
);
1333 f2fs_submit_page_mbio(fio
);
1336 static void __f2fs_replace_block(struct f2fs_sb_info
*sbi
,
1337 struct f2fs_summary
*sum
,
1338 block_t old_blkaddr
, block_t new_blkaddr
,
1339 bool recover_curseg
)
1341 struct sit_info
*sit_i
= SIT_I(sbi
);
1342 struct curseg_info
*curseg
;
1343 unsigned int segno
, old_cursegno
;
1344 struct seg_entry
*se
;
1346 unsigned short old_blkoff
;
1348 segno
= GET_SEGNO(sbi
, new_blkaddr
);
1349 se
= get_seg_entry(sbi
, segno
);
1352 if (!recover_curseg
) {
1353 /* for recovery flow */
1354 if (se
->valid_blocks
== 0 && !IS_CURSEG(sbi
, segno
)) {
1355 if (old_blkaddr
== NULL_ADDR
)
1356 type
= CURSEG_COLD_DATA
;
1358 type
= CURSEG_WARM_DATA
;
1361 if (!IS_CURSEG(sbi
, segno
))
1362 type
= CURSEG_WARM_DATA
;
1365 curseg
= CURSEG_I(sbi
, type
);
1367 mutex_lock(&curseg
->curseg_mutex
);
1368 mutex_lock(&sit_i
->sentry_lock
);
1370 old_cursegno
= curseg
->segno
;
1371 old_blkoff
= curseg
->next_blkoff
;
1373 /* change the current segment */
1374 if (segno
!= curseg
->segno
) {
1375 curseg
->next_segno
= segno
;
1376 change_curseg(sbi
, type
, true);
1379 curseg
->next_blkoff
= GET_BLKOFF_FROM_SEG0(sbi
, new_blkaddr
);
1380 __add_sum_entry(sbi
, type
, sum
);
1382 if (!recover_curseg
)
1383 update_sit_entry(sbi
, new_blkaddr
, 1);
1384 if (GET_SEGNO(sbi
, old_blkaddr
) != NULL_SEGNO
)
1385 update_sit_entry(sbi
, old_blkaddr
, -1);
1387 locate_dirty_segment(sbi
, GET_SEGNO(sbi
, old_blkaddr
));
1388 locate_dirty_segment(sbi
, GET_SEGNO(sbi
, new_blkaddr
));
1390 locate_dirty_segment(sbi
, old_cursegno
);
1392 if (recover_curseg
) {
1393 if (old_cursegno
!= curseg
->segno
) {
1394 curseg
->next_segno
= old_cursegno
;
1395 change_curseg(sbi
, type
, true);
1397 curseg
->next_blkoff
= old_blkoff
;
1400 mutex_unlock(&sit_i
->sentry_lock
);
1401 mutex_unlock(&curseg
->curseg_mutex
);
1404 void f2fs_replace_block(struct f2fs_sb_info
*sbi
, struct dnode_of_data
*dn
,
1405 block_t old_addr
, block_t new_addr
,
1406 unsigned char version
, bool recover_curseg
)
1408 struct f2fs_summary sum
;
1410 set_summary(&sum
, dn
->nid
, dn
->ofs_in_node
, version
);
1412 __f2fs_replace_block(sbi
, &sum
, old_addr
, new_addr
, recover_curseg
);
1414 dn
->data_blkaddr
= new_addr
;
1415 set_data_blkaddr(dn
);
1416 f2fs_update_extent_cache(dn
);
1419 void f2fs_wait_on_page_writeback(struct page
*page
,
1420 enum page_type type
)
1422 if (PageWriteback(page
)) {
1423 struct f2fs_sb_info
*sbi
= F2FS_P_SB(page
);
1425 if (is_merged_page(sbi
, page
, type
))
1426 f2fs_submit_merged_bio(sbi
, type
, WRITE
);
1427 wait_on_page_writeback(page
);
1431 void f2fs_wait_on_encrypted_page_writeback(struct f2fs_sb_info
*sbi
,
1436 if (blkaddr
== NEW_ADDR
)
1439 f2fs_bug_on(sbi
, blkaddr
== NULL_ADDR
);
1441 cpage
= find_lock_page(META_MAPPING(sbi
), blkaddr
);
1443 f2fs_wait_on_page_writeback(cpage
, DATA
);
1444 f2fs_put_page(cpage
, 1);
1448 static int read_compacted_summaries(struct f2fs_sb_info
*sbi
)
1450 struct f2fs_checkpoint
*ckpt
= F2FS_CKPT(sbi
);
1451 struct curseg_info
*seg_i
;
1452 unsigned char *kaddr
;
1457 start
= start_sum_block(sbi
);
1459 page
= get_meta_page(sbi
, start
++);
1460 kaddr
= (unsigned char *)page_address(page
);
1462 /* Step 1: restore nat cache */
1463 seg_i
= CURSEG_I(sbi
, CURSEG_HOT_DATA
);
1464 memcpy(&seg_i
->sum_blk
->n_nats
, kaddr
, SUM_JOURNAL_SIZE
);
1466 /* Step 2: restore sit cache */
1467 seg_i
= CURSEG_I(sbi
, CURSEG_COLD_DATA
);
1468 memcpy(&seg_i
->sum_blk
->n_sits
, kaddr
+ SUM_JOURNAL_SIZE
,
1470 offset
= 2 * SUM_JOURNAL_SIZE
;
1472 /* Step 3: restore summary entries */
1473 for (i
= CURSEG_HOT_DATA
; i
<= CURSEG_COLD_DATA
; i
++) {
1474 unsigned short blk_off
;
1477 seg_i
= CURSEG_I(sbi
, i
);
1478 segno
= le32_to_cpu(ckpt
->cur_data_segno
[i
]);
1479 blk_off
= le16_to_cpu(ckpt
->cur_data_blkoff
[i
]);
1480 seg_i
->next_segno
= segno
;
1481 reset_curseg(sbi
, i
, 0);
1482 seg_i
->alloc_type
= ckpt
->alloc_type
[i
];
1483 seg_i
->next_blkoff
= blk_off
;
1485 if (seg_i
->alloc_type
== SSR
)
1486 blk_off
= sbi
->blocks_per_seg
;
1488 for (j
= 0; j
< blk_off
; j
++) {
1489 struct f2fs_summary
*s
;
1490 s
= (struct f2fs_summary
*)(kaddr
+ offset
);
1491 seg_i
->sum_blk
->entries
[j
] = *s
;
1492 offset
+= SUMMARY_SIZE
;
1493 if (offset
+ SUMMARY_SIZE
<= PAGE_CACHE_SIZE
-
1497 f2fs_put_page(page
, 1);
1500 page
= get_meta_page(sbi
, start
++);
1501 kaddr
= (unsigned char *)page_address(page
);
1505 f2fs_put_page(page
, 1);
1509 static int read_normal_summaries(struct f2fs_sb_info
*sbi
, int type
)
1511 struct f2fs_checkpoint
*ckpt
= F2FS_CKPT(sbi
);
1512 struct f2fs_summary_block
*sum
;
1513 struct curseg_info
*curseg
;
1515 unsigned short blk_off
;
1516 unsigned int segno
= 0;
1517 block_t blk_addr
= 0;
1519 /* get segment number and block addr */
1520 if (IS_DATASEG(type
)) {
1521 segno
= le32_to_cpu(ckpt
->cur_data_segno
[type
]);
1522 blk_off
= le16_to_cpu(ckpt
->cur_data_blkoff
[type
-
1524 if (__exist_node_summaries(sbi
))
1525 blk_addr
= sum_blk_addr(sbi
, NR_CURSEG_TYPE
, type
);
1527 blk_addr
= sum_blk_addr(sbi
, NR_CURSEG_DATA_TYPE
, type
);
1529 segno
= le32_to_cpu(ckpt
->cur_node_segno
[type
-
1531 blk_off
= le16_to_cpu(ckpt
->cur_node_blkoff
[type
-
1533 if (__exist_node_summaries(sbi
))
1534 blk_addr
= sum_blk_addr(sbi
, NR_CURSEG_NODE_TYPE
,
1535 type
- CURSEG_HOT_NODE
);
1537 blk_addr
= GET_SUM_BLOCK(sbi
, segno
);
1540 new = get_meta_page(sbi
, blk_addr
);
1541 sum
= (struct f2fs_summary_block
*)page_address(new);
1543 if (IS_NODESEG(type
)) {
1544 if (__exist_node_summaries(sbi
)) {
1545 struct f2fs_summary
*ns
= &sum
->entries
[0];
1547 for (i
= 0; i
< sbi
->blocks_per_seg
; i
++, ns
++) {
1549 ns
->ofs_in_node
= 0;
1554 err
= restore_node_summary(sbi
, segno
, sum
);
1556 f2fs_put_page(new, 1);
1562 /* set uncompleted segment to curseg */
1563 curseg
= CURSEG_I(sbi
, type
);
1564 mutex_lock(&curseg
->curseg_mutex
);
1565 memcpy(curseg
->sum_blk
, sum
, PAGE_CACHE_SIZE
);
1566 curseg
->next_segno
= segno
;
1567 reset_curseg(sbi
, type
, 0);
1568 curseg
->alloc_type
= ckpt
->alloc_type
[type
];
1569 curseg
->next_blkoff
= blk_off
;
1570 mutex_unlock(&curseg
->curseg_mutex
);
1571 f2fs_put_page(new, 1);
1575 static int restore_curseg_summaries(struct f2fs_sb_info
*sbi
)
1577 int type
= CURSEG_HOT_DATA
;
1580 if (is_set_ckpt_flags(F2FS_CKPT(sbi
), CP_COMPACT_SUM_FLAG
)) {
1581 int npages
= npages_for_summary_flush(sbi
, true);
1584 ra_meta_pages(sbi
, start_sum_block(sbi
), npages
,
1587 /* restore for compacted data summary */
1588 if (read_compacted_summaries(sbi
))
1590 type
= CURSEG_HOT_NODE
;
1593 if (__exist_node_summaries(sbi
))
1594 ra_meta_pages(sbi
, sum_blk_addr(sbi
, NR_CURSEG_TYPE
, type
),
1595 NR_CURSEG_TYPE
- type
, META_CP
, true);
1597 for (; type
<= CURSEG_COLD_NODE
; type
++) {
1598 err
= read_normal_summaries(sbi
, type
);
1606 static void write_compacted_summaries(struct f2fs_sb_info
*sbi
, block_t blkaddr
)
1609 unsigned char *kaddr
;
1610 struct f2fs_summary
*summary
;
1611 struct curseg_info
*seg_i
;
1612 int written_size
= 0;
1615 page
= grab_meta_page(sbi
, blkaddr
++);
1616 kaddr
= (unsigned char *)page_address(page
);
1618 /* Step 1: write nat cache */
1619 seg_i
= CURSEG_I(sbi
, CURSEG_HOT_DATA
);
1620 memcpy(kaddr
, &seg_i
->sum_blk
->n_nats
, SUM_JOURNAL_SIZE
);
1621 written_size
+= SUM_JOURNAL_SIZE
;
1623 /* Step 2: write sit cache */
1624 seg_i
= CURSEG_I(sbi
, CURSEG_COLD_DATA
);
1625 memcpy(kaddr
+ written_size
, &seg_i
->sum_blk
->n_sits
,
1627 written_size
+= SUM_JOURNAL_SIZE
;
1629 /* Step 3: write summary entries */
1630 for (i
= CURSEG_HOT_DATA
; i
<= CURSEG_COLD_DATA
; i
++) {
1631 unsigned short blkoff
;
1632 seg_i
= CURSEG_I(sbi
, i
);
1633 if (sbi
->ckpt
->alloc_type
[i
] == SSR
)
1634 blkoff
= sbi
->blocks_per_seg
;
1636 blkoff
= curseg_blkoff(sbi
, i
);
1638 for (j
= 0; j
< blkoff
; j
++) {
1640 page
= grab_meta_page(sbi
, blkaddr
++);
1641 kaddr
= (unsigned char *)page_address(page
);
1644 summary
= (struct f2fs_summary
*)(kaddr
+ written_size
);
1645 *summary
= seg_i
->sum_blk
->entries
[j
];
1646 written_size
+= SUMMARY_SIZE
;
1648 if (written_size
+ SUMMARY_SIZE
<= PAGE_CACHE_SIZE
-
1652 set_page_dirty(page
);
1653 f2fs_put_page(page
, 1);
1658 set_page_dirty(page
);
1659 f2fs_put_page(page
, 1);
1663 static void write_normal_summaries(struct f2fs_sb_info
*sbi
,
1664 block_t blkaddr
, int type
)
1667 if (IS_DATASEG(type
))
1668 end
= type
+ NR_CURSEG_DATA_TYPE
;
1670 end
= type
+ NR_CURSEG_NODE_TYPE
;
1672 for (i
= type
; i
< end
; i
++) {
1673 struct curseg_info
*sum
= CURSEG_I(sbi
, i
);
1674 mutex_lock(&sum
->curseg_mutex
);
1675 write_sum_page(sbi
, sum
->sum_blk
, blkaddr
+ (i
- type
));
1676 mutex_unlock(&sum
->curseg_mutex
);
1680 void write_data_summaries(struct f2fs_sb_info
*sbi
, block_t start_blk
)
1682 if (is_set_ckpt_flags(F2FS_CKPT(sbi
), CP_COMPACT_SUM_FLAG
))
1683 write_compacted_summaries(sbi
, start_blk
);
1685 write_normal_summaries(sbi
, start_blk
, CURSEG_HOT_DATA
);
1688 void write_node_summaries(struct f2fs_sb_info
*sbi
, block_t start_blk
)
1690 write_normal_summaries(sbi
, start_blk
, CURSEG_HOT_NODE
);
1693 int lookup_journal_in_cursum(struct f2fs_summary_block
*sum
, int type
,
1694 unsigned int val
, int alloc
)
1698 if (type
== NAT_JOURNAL
) {
1699 for (i
= 0; i
< nats_in_cursum(sum
); i
++) {
1700 if (le32_to_cpu(nid_in_journal(sum
, i
)) == val
)
1703 if (alloc
&& __has_cursum_space(sum
, 1, NAT_JOURNAL
))
1704 return update_nats_in_cursum(sum
, 1);
1705 } else if (type
== SIT_JOURNAL
) {
1706 for (i
= 0; i
< sits_in_cursum(sum
); i
++)
1707 if (le32_to_cpu(segno_in_journal(sum
, i
)) == val
)
1709 if (alloc
&& __has_cursum_space(sum
, 1, SIT_JOURNAL
))
1710 return update_sits_in_cursum(sum
, 1);
1715 static struct page
*get_current_sit_page(struct f2fs_sb_info
*sbi
,
1718 return get_meta_page(sbi
, current_sit_addr(sbi
, segno
));
1721 static struct page
*get_next_sit_page(struct f2fs_sb_info
*sbi
,
1724 struct sit_info
*sit_i
= SIT_I(sbi
);
1725 struct page
*src_page
, *dst_page
;
1726 pgoff_t src_off
, dst_off
;
1727 void *src_addr
, *dst_addr
;
1729 src_off
= current_sit_addr(sbi
, start
);
1730 dst_off
= next_sit_addr(sbi
, src_off
);
1732 /* get current sit block page without lock */
1733 src_page
= get_meta_page(sbi
, src_off
);
1734 dst_page
= grab_meta_page(sbi
, dst_off
);
1735 f2fs_bug_on(sbi
, PageDirty(src_page
));
1737 src_addr
= page_address(src_page
);
1738 dst_addr
= page_address(dst_page
);
1739 memcpy(dst_addr
, src_addr
, PAGE_CACHE_SIZE
);
1741 set_page_dirty(dst_page
);
1742 f2fs_put_page(src_page
, 1);
1744 set_to_next_sit(sit_i
, start
);
1749 static struct sit_entry_set
*grab_sit_entry_set(void)
1751 struct sit_entry_set
*ses
=
1752 f2fs_kmem_cache_alloc(sit_entry_set_slab
, GFP_NOFS
);
1755 INIT_LIST_HEAD(&ses
->set_list
);
1759 static void release_sit_entry_set(struct sit_entry_set
*ses
)
1761 list_del(&ses
->set_list
);
1762 kmem_cache_free(sit_entry_set_slab
, ses
);
1765 static void adjust_sit_entry_set(struct sit_entry_set
*ses
,
1766 struct list_head
*head
)
1768 struct sit_entry_set
*next
= ses
;
1770 if (list_is_last(&ses
->set_list
, head
))
1773 list_for_each_entry_continue(next
, head
, set_list
)
1774 if (ses
->entry_cnt
<= next
->entry_cnt
)
1777 list_move_tail(&ses
->set_list
, &next
->set_list
);
1780 static void add_sit_entry(unsigned int segno
, struct list_head
*head
)
1782 struct sit_entry_set
*ses
;
1783 unsigned int start_segno
= START_SEGNO(segno
);
1785 list_for_each_entry(ses
, head
, set_list
) {
1786 if (ses
->start_segno
== start_segno
) {
1788 adjust_sit_entry_set(ses
, head
);
1793 ses
= grab_sit_entry_set();
1795 ses
->start_segno
= start_segno
;
1797 list_add(&ses
->set_list
, head
);
1800 static void add_sits_in_set(struct f2fs_sb_info
*sbi
)
1802 struct f2fs_sm_info
*sm_info
= SM_I(sbi
);
1803 struct list_head
*set_list
= &sm_info
->sit_entry_set
;
1804 unsigned long *bitmap
= SIT_I(sbi
)->dirty_sentries_bitmap
;
1807 for_each_set_bit(segno
, bitmap
, MAIN_SEGS(sbi
))
1808 add_sit_entry(segno
, set_list
);
1811 static void remove_sits_in_journal(struct f2fs_sb_info
*sbi
)
1813 struct curseg_info
*curseg
= CURSEG_I(sbi
, CURSEG_COLD_DATA
);
1814 struct f2fs_summary_block
*sum
= curseg
->sum_blk
;
1817 for (i
= sits_in_cursum(sum
) - 1; i
>= 0; i
--) {
1821 segno
= le32_to_cpu(segno_in_journal(sum
, i
));
1822 dirtied
= __mark_sit_entry_dirty(sbi
, segno
);
1825 add_sit_entry(segno
, &SM_I(sbi
)->sit_entry_set
);
1827 update_sits_in_cursum(sum
, -sits_in_cursum(sum
));
1831 * CP calls this function, which flushes SIT entries including sit_journal,
1832 * and moves prefree segs to free segs.
1834 void flush_sit_entries(struct f2fs_sb_info
*sbi
, struct cp_control
*cpc
)
1836 struct sit_info
*sit_i
= SIT_I(sbi
);
1837 unsigned long *bitmap
= sit_i
->dirty_sentries_bitmap
;
1838 struct curseg_info
*curseg
= CURSEG_I(sbi
, CURSEG_COLD_DATA
);
1839 struct f2fs_summary_block
*sum
= curseg
->sum_blk
;
1840 struct sit_entry_set
*ses
, *tmp
;
1841 struct list_head
*head
= &SM_I(sbi
)->sit_entry_set
;
1842 bool to_journal
= true;
1843 struct seg_entry
*se
;
1845 mutex_lock(&curseg
->curseg_mutex
);
1846 mutex_lock(&sit_i
->sentry_lock
);
1848 if (!sit_i
->dirty_sentries
)
1852 * add and account sit entries of dirty bitmap in sit entry
1855 add_sits_in_set(sbi
);
1858 * if there are no enough space in journal to store dirty sit
1859 * entries, remove all entries from journal and add and account
1860 * them in sit entry set.
1862 if (!__has_cursum_space(sum
, sit_i
->dirty_sentries
, SIT_JOURNAL
))
1863 remove_sits_in_journal(sbi
);
1866 * there are two steps to flush sit entries:
1867 * #1, flush sit entries to journal in current cold data summary block.
1868 * #2, flush sit entries to sit page.
1870 list_for_each_entry_safe(ses
, tmp
, head
, set_list
) {
1871 struct page
*page
= NULL
;
1872 struct f2fs_sit_block
*raw_sit
= NULL
;
1873 unsigned int start_segno
= ses
->start_segno
;
1874 unsigned int end
= min(start_segno
+ SIT_ENTRY_PER_BLOCK
,
1875 (unsigned long)MAIN_SEGS(sbi
));
1876 unsigned int segno
= start_segno
;
1879 !__has_cursum_space(sum
, ses
->entry_cnt
, SIT_JOURNAL
))
1883 page
= get_next_sit_page(sbi
, start_segno
);
1884 raw_sit
= page_address(page
);
1887 /* flush dirty sit entries in region of current sit set */
1888 for_each_set_bit_from(segno
, bitmap
, end
) {
1889 int offset
, sit_offset
;
1891 se
= get_seg_entry(sbi
, segno
);
1893 /* add discard candidates */
1894 if (cpc
->reason
!= CP_DISCARD
) {
1895 cpc
->trim_start
= segno
;
1896 add_discard_addrs(sbi
, cpc
);
1900 offset
= lookup_journal_in_cursum(sum
,
1901 SIT_JOURNAL
, segno
, 1);
1902 f2fs_bug_on(sbi
, offset
< 0);
1903 segno_in_journal(sum
, offset
) =
1905 seg_info_to_raw_sit(se
,
1906 &sit_in_journal(sum
, offset
));
1908 sit_offset
= SIT_ENTRY_OFFSET(sit_i
, segno
);
1909 seg_info_to_raw_sit(se
,
1910 &raw_sit
->entries
[sit_offset
]);
1913 __clear_bit(segno
, bitmap
);
1914 sit_i
->dirty_sentries
--;
1919 f2fs_put_page(page
, 1);
1921 f2fs_bug_on(sbi
, ses
->entry_cnt
);
1922 release_sit_entry_set(ses
);
1925 f2fs_bug_on(sbi
, !list_empty(head
));
1926 f2fs_bug_on(sbi
, sit_i
->dirty_sentries
);
1928 if (cpc
->reason
== CP_DISCARD
) {
1929 for (; cpc
->trim_start
<= cpc
->trim_end
; cpc
->trim_start
++)
1930 add_discard_addrs(sbi
, cpc
);
1932 mutex_unlock(&sit_i
->sentry_lock
);
1933 mutex_unlock(&curseg
->curseg_mutex
);
1935 set_prefree_as_free_segments(sbi
);
1938 static int build_sit_info(struct f2fs_sb_info
*sbi
)
1940 struct f2fs_super_block
*raw_super
= F2FS_RAW_SUPER(sbi
);
1941 struct f2fs_checkpoint
*ckpt
= F2FS_CKPT(sbi
);
1942 struct sit_info
*sit_i
;
1943 unsigned int sit_segs
, start
;
1944 char *src_bitmap
, *dst_bitmap
;
1945 unsigned int bitmap_size
;
1947 /* allocate memory for SIT information */
1948 sit_i
= kzalloc(sizeof(struct sit_info
), GFP_KERNEL
);
1952 SM_I(sbi
)->sit_info
= sit_i
;
1954 sit_i
->sentries
= f2fs_kvzalloc(MAIN_SEGS(sbi
) *
1955 sizeof(struct seg_entry
), GFP_KERNEL
);
1956 if (!sit_i
->sentries
)
1959 bitmap_size
= f2fs_bitmap_size(MAIN_SEGS(sbi
));
1960 sit_i
->dirty_sentries_bitmap
= f2fs_kvzalloc(bitmap_size
, GFP_KERNEL
);
1961 if (!sit_i
->dirty_sentries_bitmap
)
1964 for (start
= 0; start
< MAIN_SEGS(sbi
); start
++) {
1965 sit_i
->sentries
[start
].cur_valid_map
1966 = kzalloc(SIT_VBLOCK_MAP_SIZE
, GFP_KERNEL
);
1967 sit_i
->sentries
[start
].ckpt_valid_map
1968 = kzalloc(SIT_VBLOCK_MAP_SIZE
, GFP_KERNEL
);
1969 sit_i
->sentries
[start
].discard_map
1970 = kzalloc(SIT_VBLOCK_MAP_SIZE
, GFP_KERNEL
);
1971 if (!sit_i
->sentries
[start
].cur_valid_map
||
1972 !sit_i
->sentries
[start
].ckpt_valid_map
||
1973 !sit_i
->sentries
[start
].discard_map
)
1977 sit_i
->tmp_map
= kzalloc(SIT_VBLOCK_MAP_SIZE
, GFP_KERNEL
);
1978 if (!sit_i
->tmp_map
)
1981 if (sbi
->segs_per_sec
> 1) {
1982 sit_i
->sec_entries
= f2fs_kvzalloc(MAIN_SECS(sbi
) *
1983 sizeof(struct sec_entry
), GFP_KERNEL
);
1984 if (!sit_i
->sec_entries
)
1988 /* get information related with SIT */
1989 sit_segs
= le32_to_cpu(raw_super
->segment_count_sit
) >> 1;
1991 /* setup SIT bitmap from ckeckpoint pack */
1992 bitmap_size
= __bitmap_size(sbi
, SIT_BITMAP
);
1993 src_bitmap
= __bitmap_ptr(sbi
, SIT_BITMAP
);
1995 dst_bitmap
= kmemdup(src_bitmap
, bitmap_size
, GFP_KERNEL
);
1999 /* init SIT information */
2000 sit_i
->s_ops
= &default_salloc_ops
;
2002 sit_i
->sit_base_addr
= le32_to_cpu(raw_super
->sit_blkaddr
);
2003 sit_i
->sit_blocks
= sit_segs
<< sbi
->log_blocks_per_seg
;
2004 sit_i
->written_valid_blocks
= le64_to_cpu(ckpt
->valid_block_count
);
2005 sit_i
->sit_bitmap
= dst_bitmap
;
2006 sit_i
->bitmap_size
= bitmap_size
;
2007 sit_i
->dirty_sentries
= 0;
2008 sit_i
->sents_per_block
= SIT_ENTRY_PER_BLOCK
;
2009 sit_i
->elapsed_time
= le64_to_cpu(sbi
->ckpt
->elapsed_time
);
2010 sit_i
->mounted_time
= CURRENT_TIME_SEC
.tv_sec
;
2011 mutex_init(&sit_i
->sentry_lock
);
2015 static int build_free_segmap(struct f2fs_sb_info
*sbi
)
2017 struct free_segmap_info
*free_i
;
2018 unsigned int bitmap_size
, sec_bitmap_size
;
2020 /* allocate memory for free segmap information */
2021 free_i
= kzalloc(sizeof(struct free_segmap_info
), GFP_KERNEL
);
2025 SM_I(sbi
)->free_info
= free_i
;
2027 bitmap_size
= f2fs_bitmap_size(MAIN_SEGS(sbi
));
2028 free_i
->free_segmap
= f2fs_kvmalloc(bitmap_size
, GFP_KERNEL
);
2029 if (!free_i
->free_segmap
)
2032 sec_bitmap_size
= f2fs_bitmap_size(MAIN_SECS(sbi
));
2033 free_i
->free_secmap
= f2fs_kvmalloc(sec_bitmap_size
, GFP_KERNEL
);
2034 if (!free_i
->free_secmap
)
2037 /* set all segments as dirty temporarily */
2038 memset(free_i
->free_segmap
, 0xff, bitmap_size
);
2039 memset(free_i
->free_secmap
, 0xff, sec_bitmap_size
);
2041 /* init free segmap information */
2042 free_i
->start_segno
= GET_SEGNO_FROM_SEG0(sbi
, MAIN_BLKADDR(sbi
));
2043 free_i
->free_segments
= 0;
2044 free_i
->free_sections
= 0;
2045 spin_lock_init(&free_i
->segmap_lock
);
2049 static int build_curseg(struct f2fs_sb_info
*sbi
)
2051 struct curseg_info
*array
;
2054 array
= kcalloc(NR_CURSEG_TYPE
, sizeof(*array
), GFP_KERNEL
);
2058 SM_I(sbi
)->curseg_array
= array
;
2060 for (i
= 0; i
< NR_CURSEG_TYPE
; i
++) {
2061 mutex_init(&array
[i
].curseg_mutex
);
2062 array
[i
].sum_blk
= kzalloc(PAGE_CACHE_SIZE
, GFP_KERNEL
);
2063 if (!array
[i
].sum_blk
)
2065 array
[i
].segno
= NULL_SEGNO
;
2066 array
[i
].next_blkoff
= 0;
2068 return restore_curseg_summaries(sbi
);
2071 static void build_sit_entries(struct f2fs_sb_info
*sbi
)
2073 struct sit_info
*sit_i
= SIT_I(sbi
);
2074 struct curseg_info
*curseg
= CURSEG_I(sbi
, CURSEG_COLD_DATA
);
2075 struct f2fs_summary_block
*sum
= curseg
->sum_blk
;
2076 int sit_blk_cnt
= SIT_BLK_CNT(sbi
);
2077 unsigned int i
, start
, end
;
2078 unsigned int readed
, start_blk
= 0;
2079 int nrpages
= MAX_BIO_BLOCKS(sbi
);
2082 readed
= ra_meta_pages(sbi
, start_blk
, nrpages
, META_SIT
, true);
2084 start
= start_blk
* sit_i
->sents_per_block
;
2085 end
= (start_blk
+ readed
) * sit_i
->sents_per_block
;
2087 for (; start
< end
&& start
< MAIN_SEGS(sbi
); start
++) {
2088 struct seg_entry
*se
= &sit_i
->sentries
[start
];
2089 struct f2fs_sit_block
*sit_blk
;
2090 struct f2fs_sit_entry sit
;
2093 mutex_lock(&curseg
->curseg_mutex
);
2094 for (i
= 0; i
< sits_in_cursum(sum
); i
++) {
2095 if (le32_to_cpu(segno_in_journal(sum
, i
))
2097 sit
= sit_in_journal(sum
, i
);
2098 mutex_unlock(&curseg
->curseg_mutex
);
2102 mutex_unlock(&curseg
->curseg_mutex
);
2104 page
= get_current_sit_page(sbi
, start
);
2105 sit_blk
= (struct f2fs_sit_block
*)page_address(page
);
2106 sit
= sit_blk
->entries
[SIT_ENTRY_OFFSET(sit_i
, start
)];
2107 f2fs_put_page(page
, 1);
2109 check_block_count(sbi
, start
, &sit
);
2110 seg_info_from_raw_sit(se
, &sit
);
2112 /* build discard map only one time */
2113 memcpy(se
->discard_map
, se
->cur_valid_map
, SIT_VBLOCK_MAP_SIZE
);
2114 sbi
->discard_blks
+= sbi
->blocks_per_seg
- se
->valid_blocks
;
2116 if (sbi
->segs_per_sec
> 1) {
2117 struct sec_entry
*e
= get_sec_entry(sbi
, start
);
2118 e
->valid_blocks
+= se
->valid_blocks
;
2121 start_blk
+= readed
;
2122 } while (start_blk
< sit_blk_cnt
);
2125 static void init_free_segmap(struct f2fs_sb_info
*sbi
)
2130 for (start
= 0; start
< MAIN_SEGS(sbi
); start
++) {
2131 struct seg_entry
*sentry
= get_seg_entry(sbi
, start
);
2132 if (!sentry
->valid_blocks
)
2133 __set_free(sbi
, start
);
2136 /* set use the current segments */
2137 for (type
= CURSEG_HOT_DATA
; type
<= CURSEG_COLD_NODE
; type
++) {
2138 struct curseg_info
*curseg_t
= CURSEG_I(sbi
, type
);
2139 __set_test_and_inuse(sbi
, curseg_t
->segno
);
2143 static void init_dirty_segmap(struct f2fs_sb_info
*sbi
)
2145 struct dirty_seglist_info
*dirty_i
= DIRTY_I(sbi
);
2146 struct free_segmap_info
*free_i
= FREE_I(sbi
);
2147 unsigned int segno
= 0, offset
= 0;
2148 unsigned short valid_blocks
;
2151 /* find dirty segment based on free segmap */
2152 segno
= find_next_inuse(free_i
, MAIN_SEGS(sbi
), offset
);
2153 if (segno
>= MAIN_SEGS(sbi
))
2156 valid_blocks
= get_valid_blocks(sbi
, segno
, 0);
2157 if (valid_blocks
== sbi
->blocks_per_seg
|| !valid_blocks
)
2159 if (valid_blocks
> sbi
->blocks_per_seg
) {
2160 f2fs_bug_on(sbi
, 1);
2163 mutex_lock(&dirty_i
->seglist_lock
);
2164 __locate_dirty_segment(sbi
, segno
, DIRTY
);
2165 mutex_unlock(&dirty_i
->seglist_lock
);
2169 static int init_victim_secmap(struct f2fs_sb_info
*sbi
)
2171 struct dirty_seglist_info
*dirty_i
= DIRTY_I(sbi
);
2172 unsigned int bitmap_size
= f2fs_bitmap_size(MAIN_SECS(sbi
));
2174 dirty_i
->victim_secmap
= f2fs_kvzalloc(bitmap_size
, GFP_KERNEL
);
2175 if (!dirty_i
->victim_secmap
)
2180 static int build_dirty_segmap(struct f2fs_sb_info
*sbi
)
2182 struct dirty_seglist_info
*dirty_i
;
2183 unsigned int bitmap_size
, i
;
2185 /* allocate memory for dirty segments list information */
2186 dirty_i
= kzalloc(sizeof(struct dirty_seglist_info
), GFP_KERNEL
);
2190 SM_I(sbi
)->dirty_info
= dirty_i
;
2191 mutex_init(&dirty_i
->seglist_lock
);
2193 bitmap_size
= f2fs_bitmap_size(MAIN_SEGS(sbi
));
2195 for (i
= 0; i
< NR_DIRTY_TYPE
; i
++) {
2196 dirty_i
->dirty_segmap
[i
] = f2fs_kvzalloc(bitmap_size
, GFP_KERNEL
);
2197 if (!dirty_i
->dirty_segmap
[i
])
2201 init_dirty_segmap(sbi
);
2202 return init_victim_secmap(sbi
);
2206 * Update min, max modified time for cost-benefit GC algorithm
2208 static void init_min_max_mtime(struct f2fs_sb_info
*sbi
)
2210 struct sit_info
*sit_i
= SIT_I(sbi
);
2213 mutex_lock(&sit_i
->sentry_lock
);
2215 sit_i
->min_mtime
= LLONG_MAX
;
2217 for (segno
= 0; segno
< MAIN_SEGS(sbi
); segno
+= sbi
->segs_per_sec
) {
2219 unsigned long long mtime
= 0;
2221 for (i
= 0; i
< sbi
->segs_per_sec
; i
++)
2222 mtime
+= get_seg_entry(sbi
, segno
+ i
)->mtime
;
2224 mtime
= div_u64(mtime
, sbi
->segs_per_sec
);
2226 if (sit_i
->min_mtime
> mtime
)
2227 sit_i
->min_mtime
= mtime
;
2229 sit_i
->max_mtime
= get_mtime(sbi
);
2230 mutex_unlock(&sit_i
->sentry_lock
);
2233 int build_segment_manager(struct f2fs_sb_info
*sbi
)
2235 struct f2fs_super_block
*raw_super
= F2FS_RAW_SUPER(sbi
);
2236 struct f2fs_checkpoint
*ckpt
= F2FS_CKPT(sbi
);
2237 struct f2fs_sm_info
*sm_info
;
2240 sm_info
= kzalloc(sizeof(struct f2fs_sm_info
), GFP_KERNEL
);
2245 sbi
->sm_info
= sm_info
;
2246 sm_info
->seg0_blkaddr
= le32_to_cpu(raw_super
->segment0_blkaddr
);
2247 sm_info
->main_blkaddr
= le32_to_cpu(raw_super
->main_blkaddr
);
2248 sm_info
->segment_count
= le32_to_cpu(raw_super
->segment_count
);
2249 sm_info
->reserved_segments
= le32_to_cpu(ckpt
->rsvd_segment_count
);
2250 sm_info
->ovp_segments
= le32_to_cpu(ckpt
->overprov_segment_count
);
2251 sm_info
->main_segments
= le32_to_cpu(raw_super
->segment_count_main
);
2252 sm_info
->ssa_blkaddr
= le32_to_cpu(raw_super
->ssa_blkaddr
);
2253 sm_info
->rec_prefree_segments
= sm_info
->main_segments
*
2254 DEF_RECLAIM_PREFREE_SEGMENTS
/ 100;
2255 sm_info
->ipu_policy
= 1 << F2FS_IPU_FSYNC
;
2256 sm_info
->min_ipu_util
= DEF_MIN_IPU_UTIL
;
2257 sm_info
->min_fsync_blocks
= DEF_MIN_FSYNC_BLOCKS
;
2259 INIT_LIST_HEAD(&sm_info
->discard_list
);
2260 sm_info
->nr_discards
= 0;
2261 sm_info
->max_discards
= 0;
2263 sm_info
->trim_sections
= DEF_BATCHED_TRIM_SECTIONS
;
2265 INIT_LIST_HEAD(&sm_info
->sit_entry_set
);
2267 if (test_opt(sbi
, FLUSH_MERGE
) && !f2fs_readonly(sbi
->sb
)) {
2268 err
= create_flush_cmd_control(sbi
);
2273 err
= build_sit_info(sbi
);
2276 err
= build_free_segmap(sbi
);
2279 err
= build_curseg(sbi
);
2283 /* reinit free segmap based on SIT */
2284 build_sit_entries(sbi
);
2286 init_free_segmap(sbi
);
2287 err
= build_dirty_segmap(sbi
);
2291 init_min_max_mtime(sbi
);
2295 static void discard_dirty_segmap(struct f2fs_sb_info
*sbi
,
2296 enum dirty_type dirty_type
)
2298 struct dirty_seglist_info
*dirty_i
= DIRTY_I(sbi
);
2300 mutex_lock(&dirty_i
->seglist_lock
);
2301 kvfree(dirty_i
->dirty_segmap
[dirty_type
]);
2302 dirty_i
->nr_dirty
[dirty_type
] = 0;
2303 mutex_unlock(&dirty_i
->seglist_lock
);
2306 static void destroy_victim_secmap(struct f2fs_sb_info
*sbi
)
2308 struct dirty_seglist_info
*dirty_i
= DIRTY_I(sbi
);
2309 kvfree(dirty_i
->victim_secmap
);
2312 static void destroy_dirty_segmap(struct f2fs_sb_info
*sbi
)
2314 struct dirty_seglist_info
*dirty_i
= DIRTY_I(sbi
);
2320 /* discard pre-free/dirty segments list */
2321 for (i
= 0; i
< NR_DIRTY_TYPE
; i
++)
2322 discard_dirty_segmap(sbi
, i
);
2324 destroy_victim_secmap(sbi
);
2325 SM_I(sbi
)->dirty_info
= NULL
;
2329 static void destroy_curseg(struct f2fs_sb_info
*sbi
)
2331 struct curseg_info
*array
= SM_I(sbi
)->curseg_array
;
2336 SM_I(sbi
)->curseg_array
= NULL
;
2337 for (i
= 0; i
< NR_CURSEG_TYPE
; i
++)
2338 kfree(array
[i
].sum_blk
);
2342 static void destroy_free_segmap(struct f2fs_sb_info
*sbi
)
2344 struct free_segmap_info
*free_i
= SM_I(sbi
)->free_info
;
2347 SM_I(sbi
)->free_info
= NULL
;
2348 kvfree(free_i
->free_segmap
);
2349 kvfree(free_i
->free_secmap
);
2353 static void destroy_sit_info(struct f2fs_sb_info
*sbi
)
2355 struct sit_info
*sit_i
= SIT_I(sbi
);
2361 if (sit_i
->sentries
) {
2362 for (start
= 0; start
< MAIN_SEGS(sbi
); start
++) {
2363 kfree(sit_i
->sentries
[start
].cur_valid_map
);
2364 kfree(sit_i
->sentries
[start
].ckpt_valid_map
);
2365 kfree(sit_i
->sentries
[start
].discard_map
);
2368 kfree(sit_i
->tmp_map
);
2370 kvfree(sit_i
->sentries
);
2371 kvfree(sit_i
->sec_entries
);
2372 kvfree(sit_i
->dirty_sentries_bitmap
);
2374 SM_I(sbi
)->sit_info
= NULL
;
2375 kfree(sit_i
->sit_bitmap
);
2379 void destroy_segment_manager(struct f2fs_sb_info
*sbi
)
2381 struct f2fs_sm_info
*sm_info
= SM_I(sbi
);
2385 destroy_flush_cmd_control(sbi
);
2386 destroy_dirty_segmap(sbi
);
2387 destroy_curseg(sbi
);
2388 destroy_free_segmap(sbi
);
2389 destroy_sit_info(sbi
);
2390 sbi
->sm_info
= NULL
;
2394 int __init
create_segment_manager_caches(void)
2396 discard_entry_slab
= f2fs_kmem_cache_create("discard_entry",
2397 sizeof(struct discard_entry
));
2398 if (!discard_entry_slab
)
2401 sit_entry_set_slab
= f2fs_kmem_cache_create("sit_entry_set",
2402 sizeof(struct sit_entry_set
));
2403 if (!sit_entry_set_slab
)
2404 goto destory_discard_entry
;
2406 inmem_entry_slab
= f2fs_kmem_cache_create("inmem_page_entry",
2407 sizeof(struct inmem_pages
));
2408 if (!inmem_entry_slab
)
2409 goto destroy_sit_entry_set
;
2412 destroy_sit_entry_set
:
2413 kmem_cache_destroy(sit_entry_set_slab
);
2414 destory_discard_entry
:
2415 kmem_cache_destroy(discard_entry_slab
);
2420 void destroy_segment_manager_caches(void)
2422 kmem_cache_destroy(sit_entry_set_slab
);
2423 kmem_cache_destroy(discard_entry_slab
);
2424 kmem_cache_destroy(inmem_entry_slab
);