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>
19 #include <linux/freezer.h>
20 #include <linux/sched/signal.h>
27 #include <trace/events/f2fs.h>
29 #define __reverse_ffz(x) __reverse_ffs(~(x))
31 static struct kmem_cache
*discard_entry_slab
;
32 static struct kmem_cache
*discard_cmd_slab
;
33 static struct kmem_cache
*sit_entry_set_slab
;
34 static struct kmem_cache
*inmem_entry_slab
;
36 static unsigned long __reverse_ulong(unsigned char *str
)
38 unsigned long tmp
= 0;
39 int shift
= 24, idx
= 0;
41 #if BITS_PER_LONG == 64
45 tmp
|= (unsigned long)str
[idx
++] << shift
;
46 shift
-= BITS_PER_BYTE
;
52 * __reverse_ffs is copied from include/asm-generic/bitops/__ffs.h since
53 * MSB and LSB are reversed in a byte by f2fs_set_bit.
55 static inline unsigned long __reverse_ffs(unsigned long word
)
59 #if BITS_PER_LONG == 64
60 if ((word
& 0xffffffff00000000UL
) == 0)
65 if ((word
& 0xffff0000) == 0)
70 if ((word
& 0xff00) == 0)
75 if ((word
& 0xf0) == 0)
80 if ((word
& 0xc) == 0)
85 if ((word
& 0x2) == 0)
91 * __find_rev_next(_zero)_bit is copied from lib/find_next_bit.c because
92 * f2fs_set_bit makes MSB and LSB reversed in a byte.
93 * @size must be integral times of unsigned long.
96 * f2fs_set_bit(0, bitmap) => 1000 0000
97 * f2fs_set_bit(7, bitmap) => 0000 0001
99 static unsigned long __find_rev_next_bit(const unsigned long *addr
,
100 unsigned long size
, unsigned long offset
)
102 const unsigned long *p
= addr
+ BIT_WORD(offset
);
103 unsigned long result
= size
;
109 size
-= (offset
& ~(BITS_PER_LONG
- 1));
110 offset
%= BITS_PER_LONG
;
116 tmp
= __reverse_ulong((unsigned char *)p
);
118 tmp
&= ~0UL >> offset
;
119 if (size
< BITS_PER_LONG
)
120 tmp
&= (~0UL << (BITS_PER_LONG
- size
));
124 if (size
<= BITS_PER_LONG
)
126 size
-= BITS_PER_LONG
;
132 return result
- size
+ __reverse_ffs(tmp
);
135 static unsigned long __find_rev_next_zero_bit(const unsigned long *addr
,
136 unsigned long size
, unsigned long offset
)
138 const unsigned long *p
= addr
+ BIT_WORD(offset
);
139 unsigned long result
= size
;
145 size
-= (offset
& ~(BITS_PER_LONG
- 1));
146 offset
%= BITS_PER_LONG
;
152 tmp
= __reverse_ulong((unsigned char *)p
);
155 tmp
|= ~0UL << (BITS_PER_LONG
- offset
);
156 if (size
< BITS_PER_LONG
)
161 if (size
<= BITS_PER_LONG
)
163 size
-= BITS_PER_LONG
;
169 return result
- size
+ __reverse_ffz(tmp
);
172 bool need_SSR(struct f2fs_sb_info
*sbi
)
174 int node_secs
= get_blocktype_secs(sbi
, F2FS_DIRTY_NODES
);
175 int dent_secs
= get_blocktype_secs(sbi
, F2FS_DIRTY_DENTS
);
176 int imeta_secs
= get_blocktype_secs(sbi
, F2FS_DIRTY_IMETA
);
178 if (test_opt(sbi
, LFS
))
180 if (sbi
->gc_thread
&& sbi
->gc_thread
->gc_urgent
)
183 return free_sections(sbi
) <= (node_secs
+ 2 * dent_secs
+ imeta_secs
+
184 SM_I(sbi
)->min_ssr_sections
+ reserved_sections(sbi
));
187 void register_inmem_page(struct inode
*inode
, struct page
*page
)
189 struct f2fs_sb_info
*sbi
= F2FS_I_SB(inode
);
190 struct f2fs_inode_info
*fi
= F2FS_I(inode
);
191 struct inmem_pages
*new;
193 f2fs_trace_pid(page
);
195 set_page_private(page
, (unsigned long)ATOMIC_WRITTEN_PAGE
);
196 SetPagePrivate(page
);
198 new = f2fs_kmem_cache_alloc(inmem_entry_slab
, GFP_NOFS
);
200 /* add atomic page indices to the list */
202 INIT_LIST_HEAD(&new->list
);
204 /* increase reference count with clean state */
205 mutex_lock(&fi
->inmem_lock
);
207 list_add_tail(&new->list
, &fi
->inmem_pages
);
208 spin_lock(&sbi
->inode_lock
[ATOMIC_FILE
]);
209 if (list_empty(&fi
->inmem_ilist
))
210 list_add_tail(&fi
->inmem_ilist
, &sbi
->inode_list
[ATOMIC_FILE
]);
211 spin_unlock(&sbi
->inode_lock
[ATOMIC_FILE
]);
212 inc_page_count(F2FS_I_SB(inode
), F2FS_INMEM_PAGES
);
213 mutex_unlock(&fi
->inmem_lock
);
215 trace_f2fs_register_inmem_page(page
, INMEM
);
218 static int __revoke_inmem_pages(struct inode
*inode
,
219 struct list_head
*head
, bool drop
, bool recover
)
221 struct f2fs_sb_info
*sbi
= F2FS_I_SB(inode
);
222 struct inmem_pages
*cur
, *tmp
;
225 list_for_each_entry_safe(cur
, tmp
, head
, list
) {
226 struct page
*page
= cur
->page
;
229 trace_f2fs_commit_inmem_page(page
, INMEM_DROP
);
234 struct dnode_of_data dn
;
237 trace_f2fs_commit_inmem_page(page
, INMEM_REVOKE
);
239 set_new_dnode(&dn
, inode
, NULL
, NULL
, 0);
240 err
= get_dnode_of_data(&dn
, page
->index
, LOOKUP_NODE
);
242 if (err
== -ENOMEM
) {
243 congestion_wait(BLK_RW_ASYNC
, HZ
/50);
250 get_node_info(sbi
, dn
.nid
, &ni
);
251 f2fs_replace_block(sbi
, &dn
, dn
.data_blkaddr
,
252 cur
->old_addr
, ni
.version
, true, true);
256 /* we don't need to invalidate this in the sccessful status */
258 ClearPageUptodate(page
);
259 set_page_private(page
, 0);
260 ClearPagePrivate(page
);
261 f2fs_put_page(page
, 1);
263 list_del(&cur
->list
);
264 kmem_cache_free(inmem_entry_slab
, cur
);
265 dec_page_count(F2FS_I_SB(inode
), F2FS_INMEM_PAGES
);
270 void drop_inmem_pages_all(struct f2fs_sb_info
*sbi
)
272 struct list_head
*head
= &sbi
->inode_list
[ATOMIC_FILE
];
274 struct f2fs_inode_info
*fi
;
276 spin_lock(&sbi
->inode_lock
[ATOMIC_FILE
]);
277 if (list_empty(head
)) {
278 spin_unlock(&sbi
->inode_lock
[ATOMIC_FILE
]);
281 fi
= list_first_entry(head
, struct f2fs_inode_info
, inmem_ilist
);
282 inode
= igrab(&fi
->vfs_inode
);
283 spin_unlock(&sbi
->inode_lock
[ATOMIC_FILE
]);
286 drop_inmem_pages(inode
);
289 congestion_wait(BLK_RW_ASYNC
, HZ
/50);
294 void drop_inmem_pages(struct inode
*inode
)
296 struct f2fs_sb_info
*sbi
= F2FS_I_SB(inode
);
297 struct f2fs_inode_info
*fi
= F2FS_I(inode
);
299 mutex_lock(&fi
->inmem_lock
);
300 __revoke_inmem_pages(inode
, &fi
->inmem_pages
, true, false);
301 spin_lock(&sbi
->inode_lock
[ATOMIC_FILE
]);
302 if (!list_empty(&fi
->inmem_ilist
))
303 list_del_init(&fi
->inmem_ilist
);
304 spin_unlock(&sbi
->inode_lock
[ATOMIC_FILE
]);
305 mutex_unlock(&fi
->inmem_lock
);
307 clear_inode_flag(inode
, FI_ATOMIC_FILE
);
308 clear_inode_flag(inode
, FI_HOT_DATA
);
309 stat_dec_atomic_write(inode
);
312 void drop_inmem_page(struct inode
*inode
, struct page
*page
)
314 struct f2fs_inode_info
*fi
= F2FS_I(inode
);
315 struct f2fs_sb_info
*sbi
= F2FS_I_SB(inode
);
316 struct list_head
*head
= &fi
->inmem_pages
;
317 struct inmem_pages
*cur
= NULL
;
319 f2fs_bug_on(sbi
, !IS_ATOMIC_WRITTEN_PAGE(page
));
321 mutex_lock(&fi
->inmem_lock
);
322 list_for_each_entry(cur
, head
, list
) {
323 if (cur
->page
== page
)
327 f2fs_bug_on(sbi
, !cur
|| cur
->page
!= page
);
328 list_del(&cur
->list
);
329 mutex_unlock(&fi
->inmem_lock
);
331 dec_page_count(sbi
, F2FS_INMEM_PAGES
);
332 kmem_cache_free(inmem_entry_slab
, cur
);
334 ClearPageUptodate(page
);
335 set_page_private(page
, 0);
336 ClearPagePrivate(page
);
337 f2fs_put_page(page
, 0);
339 trace_f2fs_commit_inmem_page(page
, INMEM_INVALIDATE
);
342 static int __commit_inmem_pages(struct inode
*inode
,
343 struct list_head
*revoke_list
)
345 struct f2fs_sb_info
*sbi
= F2FS_I_SB(inode
);
346 struct f2fs_inode_info
*fi
= F2FS_I(inode
);
347 struct inmem_pages
*cur
, *tmp
;
348 struct f2fs_io_info fio
= {
353 .op_flags
= REQ_SYNC
| REQ_PRIO
,
354 .io_type
= FS_DATA_IO
,
356 pgoff_t last_idx
= ULONG_MAX
;
359 list_for_each_entry_safe(cur
, tmp
, &fi
->inmem_pages
, list
) {
360 struct page
*page
= cur
->page
;
363 if (page
->mapping
== inode
->i_mapping
) {
364 trace_f2fs_commit_inmem_page(page
, INMEM
);
366 set_page_dirty(page
);
367 f2fs_wait_on_page_writeback(page
, DATA
, true);
368 if (clear_page_dirty_for_io(page
)) {
369 inode_dec_dirty_pages(inode
);
370 remove_dirty_inode(inode
);
374 fio
.old_blkaddr
= NULL_ADDR
;
375 fio
.encrypted_page
= NULL
;
376 fio
.need_lock
= LOCK_DONE
;
377 err
= do_write_data_page(&fio
);
379 if (err
== -ENOMEM
) {
380 congestion_wait(BLK_RW_ASYNC
, HZ
/50);
387 /* record old blkaddr for revoking */
388 cur
->old_addr
= fio
.old_blkaddr
;
389 last_idx
= page
->index
;
392 list_move_tail(&cur
->list
, revoke_list
);
395 if (last_idx
!= ULONG_MAX
)
396 f2fs_submit_merged_write_cond(sbi
, inode
, 0, last_idx
, DATA
);
399 __revoke_inmem_pages(inode
, revoke_list
, false, false);
404 int commit_inmem_pages(struct inode
*inode
)
406 struct f2fs_sb_info
*sbi
= F2FS_I_SB(inode
);
407 struct f2fs_inode_info
*fi
= F2FS_I(inode
);
408 struct list_head revoke_list
;
411 INIT_LIST_HEAD(&revoke_list
);
412 f2fs_balance_fs(sbi
, true);
415 set_inode_flag(inode
, FI_ATOMIC_COMMIT
);
417 mutex_lock(&fi
->inmem_lock
);
418 err
= __commit_inmem_pages(inode
, &revoke_list
);
422 * try to revoke all committed pages, but still we could fail
423 * due to no memory or other reason, if that happened, EAGAIN
424 * will be returned, which means in such case, transaction is
425 * already not integrity, caller should use journal to do the
426 * recovery or rewrite & commit last transaction. For other
427 * error number, revoking was done by filesystem itself.
429 ret
= __revoke_inmem_pages(inode
, &revoke_list
, false, true);
433 /* drop all uncommitted pages */
434 __revoke_inmem_pages(inode
, &fi
->inmem_pages
, true, false);
436 spin_lock(&sbi
->inode_lock
[ATOMIC_FILE
]);
437 if (!list_empty(&fi
->inmem_ilist
))
438 list_del_init(&fi
->inmem_ilist
);
439 spin_unlock(&sbi
->inode_lock
[ATOMIC_FILE
]);
440 mutex_unlock(&fi
->inmem_lock
);
442 clear_inode_flag(inode
, FI_ATOMIC_COMMIT
);
449 * This function balances dirty node and dentry pages.
450 * In addition, it controls garbage collection.
452 void f2fs_balance_fs(struct f2fs_sb_info
*sbi
, bool need
)
454 #ifdef CONFIG_F2FS_FAULT_INJECTION
455 if (time_to_inject(sbi
, FAULT_CHECKPOINT
)) {
456 f2fs_show_injection_info(FAULT_CHECKPOINT
);
457 f2fs_stop_checkpoint(sbi
, false);
461 /* balance_fs_bg is able to be pending */
462 if (need
&& excess_cached_nats(sbi
))
463 f2fs_balance_fs_bg(sbi
);
466 * We should do GC or end up with checkpoint, if there are so many dirty
467 * dir/node pages without enough free segments.
469 if (has_not_enough_free_secs(sbi
, 0, 0)) {
470 mutex_lock(&sbi
->gc_mutex
);
471 f2fs_gc(sbi
, false, false, NULL_SEGNO
);
475 void f2fs_balance_fs_bg(struct f2fs_sb_info
*sbi
)
477 /* try to shrink extent cache when there is no enough memory */
478 if (!available_free_memory(sbi
, EXTENT_CACHE
))
479 f2fs_shrink_extent_tree(sbi
, EXTENT_CACHE_SHRINK_NUMBER
);
481 /* check the # of cached NAT entries */
482 if (!available_free_memory(sbi
, NAT_ENTRIES
))
483 try_to_free_nats(sbi
, NAT_ENTRY_PER_BLOCK
);
485 if (!available_free_memory(sbi
, FREE_NIDS
))
486 try_to_free_nids(sbi
, MAX_FREE_NIDS
);
488 build_free_nids(sbi
, false, false);
490 if (!is_idle(sbi
) && !excess_dirty_nats(sbi
))
493 /* checkpoint is the only way to shrink partial cached entries */
494 if (!available_free_memory(sbi
, NAT_ENTRIES
) ||
495 !available_free_memory(sbi
, INO_ENTRIES
) ||
496 excess_prefree_segs(sbi
) ||
497 excess_dirty_nats(sbi
) ||
498 f2fs_time_over(sbi
, CP_TIME
)) {
499 if (test_opt(sbi
, DATA_FLUSH
)) {
500 struct blk_plug plug
;
502 blk_start_plug(&plug
);
503 sync_dirty_inodes(sbi
, FILE_INODE
);
504 blk_finish_plug(&plug
);
506 f2fs_sync_fs(sbi
->sb
, true);
507 stat_inc_bg_cp_count(sbi
->stat_info
);
511 static int __submit_flush_wait(struct f2fs_sb_info
*sbi
,
512 struct block_device
*bdev
)
514 struct bio
*bio
= f2fs_bio_alloc(sbi
, 0, true);
517 bio
->bi_opf
= REQ_OP_WRITE
| REQ_SYNC
| REQ_PREFLUSH
;
518 bio_set_dev(bio
, bdev
);
519 ret
= submit_bio_wait(bio
);
522 trace_f2fs_issue_flush(bdev
, test_opt(sbi
, NOBARRIER
),
523 test_opt(sbi
, FLUSH_MERGE
), ret
);
527 static int submit_flush_wait(struct f2fs_sb_info
*sbi
, nid_t ino
)
533 return __submit_flush_wait(sbi
, sbi
->sb
->s_bdev
);
535 for (i
= 0; i
< sbi
->s_ndevs
; i
++) {
536 if (!is_dirty_device(sbi
, ino
, i
, FLUSH_INO
))
538 ret
= __submit_flush_wait(sbi
, FDEV(i
).bdev
);
545 static int issue_flush_thread(void *data
)
547 struct f2fs_sb_info
*sbi
= data
;
548 struct flush_cmd_control
*fcc
= SM_I(sbi
)->fcc_info
;
549 wait_queue_head_t
*q
= &fcc
->flush_wait_queue
;
551 if (kthread_should_stop())
554 sb_start_intwrite(sbi
->sb
);
556 if (!llist_empty(&fcc
->issue_list
)) {
557 struct flush_cmd
*cmd
, *next
;
560 fcc
->dispatch_list
= llist_del_all(&fcc
->issue_list
);
561 fcc
->dispatch_list
= llist_reverse_order(fcc
->dispatch_list
);
563 cmd
= llist_entry(fcc
->dispatch_list
, struct flush_cmd
, llnode
);
565 ret
= submit_flush_wait(sbi
, cmd
->ino
);
566 atomic_inc(&fcc
->issued_flush
);
568 llist_for_each_entry_safe(cmd
, next
,
569 fcc
->dispatch_list
, llnode
) {
571 complete(&cmd
->wait
);
573 fcc
->dispatch_list
= NULL
;
576 sb_end_intwrite(sbi
->sb
);
578 wait_event_interruptible(*q
,
579 kthread_should_stop() || !llist_empty(&fcc
->issue_list
));
583 int f2fs_issue_flush(struct f2fs_sb_info
*sbi
, nid_t ino
)
585 struct flush_cmd_control
*fcc
= SM_I(sbi
)->fcc_info
;
586 struct flush_cmd cmd
;
589 if (test_opt(sbi
, NOBARRIER
))
592 if (!test_opt(sbi
, FLUSH_MERGE
)) {
593 ret
= submit_flush_wait(sbi
, ino
);
594 atomic_inc(&fcc
->issued_flush
);
598 if (atomic_inc_return(&fcc
->issing_flush
) == 1 || sbi
->s_ndevs
> 1) {
599 ret
= submit_flush_wait(sbi
, ino
);
600 atomic_dec(&fcc
->issing_flush
);
602 atomic_inc(&fcc
->issued_flush
);
607 init_completion(&cmd
.wait
);
609 llist_add(&cmd
.llnode
, &fcc
->issue_list
);
611 /* update issue_list before we wake up issue_flush thread */
614 if (waitqueue_active(&fcc
->flush_wait_queue
))
615 wake_up(&fcc
->flush_wait_queue
);
617 if (fcc
->f2fs_issue_flush
) {
618 wait_for_completion(&cmd
.wait
);
619 atomic_dec(&fcc
->issing_flush
);
621 struct llist_node
*list
;
623 list
= llist_del_all(&fcc
->issue_list
);
625 wait_for_completion(&cmd
.wait
);
626 atomic_dec(&fcc
->issing_flush
);
628 struct flush_cmd
*tmp
, *next
;
630 ret
= submit_flush_wait(sbi
, ino
);
632 llist_for_each_entry_safe(tmp
, next
, list
, llnode
) {
635 atomic_dec(&fcc
->issing_flush
);
639 complete(&tmp
->wait
);
647 int create_flush_cmd_control(struct f2fs_sb_info
*sbi
)
649 dev_t dev
= sbi
->sb
->s_bdev
->bd_dev
;
650 struct flush_cmd_control
*fcc
;
653 if (SM_I(sbi
)->fcc_info
) {
654 fcc
= SM_I(sbi
)->fcc_info
;
655 if (fcc
->f2fs_issue_flush
)
660 fcc
= kzalloc(sizeof(struct flush_cmd_control
), GFP_KERNEL
);
663 atomic_set(&fcc
->issued_flush
, 0);
664 atomic_set(&fcc
->issing_flush
, 0);
665 init_waitqueue_head(&fcc
->flush_wait_queue
);
666 init_llist_head(&fcc
->issue_list
);
667 SM_I(sbi
)->fcc_info
= fcc
;
668 if (!test_opt(sbi
, FLUSH_MERGE
))
672 fcc
->f2fs_issue_flush
= kthread_run(issue_flush_thread
, sbi
,
673 "f2fs_flush-%u:%u", MAJOR(dev
), MINOR(dev
));
674 if (IS_ERR(fcc
->f2fs_issue_flush
)) {
675 err
= PTR_ERR(fcc
->f2fs_issue_flush
);
677 SM_I(sbi
)->fcc_info
= NULL
;
684 void destroy_flush_cmd_control(struct f2fs_sb_info
*sbi
, bool free
)
686 struct flush_cmd_control
*fcc
= SM_I(sbi
)->fcc_info
;
688 if (fcc
&& fcc
->f2fs_issue_flush
) {
689 struct task_struct
*flush_thread
= fcc
->f2fs_issue_flush
;
691 fcc
->f2fs_issue_flush
= NULL
;
692 kthread_stop(flush_thread
);
696 SM_I(sbi
)->fcc_info
= NULL
;
700 int f2fs_flush_device_cache(struct f2fs_sb_info
*sbi
)
707 for (i
= 1; i
< sbi
->s_ndevs
; i
++) {
708 if (!f2fs_test_bit(i
, (char *)&sbi
->dirty_device
))
710 ret
= __submit_flush_wait(sbi
, FDEV(i
).bdev
);
714 spin_lock(&sbi
->dev_lock
);
715 f2fs_clear_bit(i
, (char *)&sbi
->dirty_device
);
716 spin_unlock(&sbi
->dev_lock
);
722 static void __locate_dirty_segment(struct f2fs_sb_info
*sbi
, unsigned int segno
,
723 enum dirty_type dirty_type
)
725 struct dirty_seglist_info
*dirty_i
= DIRTY_I(sbi
);
727 /* need not be added */
728 if (IS_CURSEG(sbi
, segno
))
731 if (!test_and_set_bit(segno
, dirty_i
->dirty_segmap
[dirty_type
]))
732 dirty_i
->nr_dirty
[dirty_type
]++;
734 if (dirty_type
== DIRTY
) {
735 struct seg_entry
*sentry
= get_seg_entry(sbi
, segno
);
736 enum dirty_type t
= sentry
->type
;
738 if (unlikely(t
>= DIRTY
)) {
742 if (!test_and_set_bit(segno
, dirty_i
->dirty_segmap
[t
]))
743 dirty_i
->nr_dirty
[t
]++;
747 static void __remove_dirty_segment(struct f2fs_sb_info
*sbi
, unsigned int segno
,
748 enum dirty_type dirty_type
)
750 struct dirty_seglist_info
*dirty_i
= DIRTY_I(sbi
);
752 if (test_and_clear_bit(segno
, dirty_i
->dirty_segmap
[dirty_type
]))
753 dirty_i
->nr_dirty
[dirty_type
]--;
755 if (dirty_type
== DIRTY
) {
756 struct seg_entry
*sentry
= get_seg_entry(sbi
, segno
);
757 enum dirty_type t
= sentry
->type
;
759 if (test_and_clear_bit(segno
, dirty_i
->dirty_segmap
[t
]))
760 dirty_i
->nr_dirty
[t
]--;
762 if (get_valid_blocks(sbi
, segno
, true) == 0)
763 clear_bit(GET_SEC_FROM_SEG(sbi
, segno
),
764 dirty_i
->victim_secmap
);
769 * Should not occur error such as -ENOMEM.
770 * Adding dirty entry into seglist is not critical operation.
771 * If a given segment is one of current working segments, it won't be added.
773 static void locate_dirty_segment(struct f2fs_sb_info
*sbi
, unsigned int segno
)
775 struct dirty_seglist_info
*dirty_i
= DIRTY_I(sbi
);
776 unsigned short valid_blocks
;
778 if (segno
== NULL_SEGNO
|| IS_CURSEG(sbi
, segno
))
781 mutex_lock(&dirty_i
->seglist_lock
);
783 valid_blocks
= get_valid_blocks(sbi
, segno
, false);
785 if (valid_blocks
== 0) {
786 __locate_dirty_segment(sbi
, segno
, PRE
);
787 __remove_dirty_segment(sbi
, segno
, DIRTY
);
788 } else if (valid_blocks
< sbi
->blocks_per_seg
) {
789 __locate_dirty_segment(sbi
, segno
, DIRTY
);
791 /* Recovery routine with SSR needs this */
792 __remove_dirty_segment(sbi
, segno
, DIRTY
);
795 mutex_unlock(&dirty_i
->seglist_lock
);
798 static struct discard_cmd
*__create_discard_cmd(struct f2fs_sb_info
*sbi
,
799 struct block_device
*bdev
, block_t lstart
,
800 block_t start
, block_t len
)
802 struct discard_cmd_control
*dcc
= SM_I(sbi
)->dcc_info
;
803 struct list_head
*pend_list
;
804 struct discard_cmd
*dc
;
806 f2fs_bug_on(sbi
, !len
);
808 pend_list
= &dcc
->pend_list
[plist_idx(len
)];
810 dc
= f2fs_kmem_cache_alloc(discard_cmd_slab
, GFP_NOFS
);
811 INIT_LIST_HEAD(&dc
->list
);
819 init_completion(&dc
->wait
);
820 list_add_tail(&dc
->list
, pend_list
);
821 atomic_inc(&dcc
->discard_cmd_cnt
);
822 dcc
->undiscard_blks
+= len
;
827 static struct discard_cmd
*__attach_discard_cmd(struct f2fs_sb_info
*sbi
,
828 struct block_device
*bdev
, block_t lstart
,
829 block_t start
, block_t len
,
830 struct rb_node
*parent
, struct rb_node
**p
)
832 struct discard_cmd_control
*dcc
= SM_I(sbi
)->dcc_info
;
833 struct discard_cmd
*dc
;
835 dc
= __create_discard_cmd(sbi
, bdev
, lstart
, start
, len
);
837 rb_link_node(&dc
->rb_node
, parent
, p
);
838 rb_insert_color(&dc
->rb_node
, &dcc
->root
);
843 static void __detach_discard_cmd(struct discard_cmd_control
*dcc
,
844 struct discard_cmd
*dc
)
846 if (dc
->state
== D_DONE
)
847 atomic_dec(&dcc
->issing_discard
);
850 rb_erase(&dc
->rb_node
, &dcc
->root
);
851 dcc
->undiscard_blks
-= dc
->len
;
853 kmem_cache_free(discard_cmd_slab
, dc
);
855 atomic_dec(&dcc
->discard_cmd_cnt
);
858 static void __remove_discard_cmd(struct f2fs_sb_info
*sbi
,
859 struct discard_cmd
*dc
)
861 struct discard_cmd_control
*dcc
= SM_I(sbi
)->dcc_info
;
863 trace_f2fs_remove_discard(dc
->bdev
, dc
->start
, dc
->len
);
865 f2fs_bug_on(sbi
, dc
->ref
);
867 if (dc
->error
== -EOPNOTSUPP
)
871 f2fs_msg(sbi
->sb
, KERN_INFO
,
872 "Issue discard(%u, %u, %u) failed, ret: %d",
873 dc
->lstart
, dc
->start
, dc
->len
, dc
->error
);
874 __detach_discard_cmd(dcc
, dc
);
877 static void f2fs_submit_discard_endio(struct bio
*bio
)
879 struct discard_cmd
*dc
= (struct discard_cmd
*)bio
->bi_private
;
881 dc
->error
= blk_status_to_errno(bio
->bi_status
);
883 complete_all(&dc
->wait
);
887 void __check_sit_bitmap(struct f2fs_sb_info
*sbi
,
888 block_t start
, block_t end
)
890 #ifdef CONFIG_F2FS_CHECK_FS
891 struct seg_entry
*sentry
;
894 unsigned long offset
, size
, max_blocks
= sbi
->blocks_per_seg
;
898 segno
= GET_SEGNO(sbi
, blk
);
899 sentry
= get_seg_entry(sbi
, segno
);
900 offset
= GET_BLKOFF_FROM_SEG0(sbi
, blk
);
902 if (end
< START_BLOCK(sbi
, segno
+ 1))
903 size
= GET_BLKOFF_FROM_SEG0(sbi
, end
);
906 map
= (unsigned long *)(sentry
->cur_valid_map
);
907 offset
= __find_rev_next_bit(map
, size
, offset
);
908 f2fs_bug_on(sbi
, offset
!= size
);
909 blk
= START_BLOCK(sbi
, segno
+ 1);
914 /* this function is copied from blkdev_issue_discard from block/blk-lib.c */
915 static void __submit_discard_cmd(struct f2fs_sb_info
*sbi
,
916 struct discard_policy
*dpolicy
,
917 struct discard_cmd
*dc
)
919 struct discard_cmd_control
*dcc
= SM_I(sbi
)->dcc_info
;
920 struct list_head
*wait_list
= (dpolicy
->type
== DPOLICY_FSTRIM
) ?
921 &(dcc
->fstrim_list
) : &(dcc
->wait_list
);
922 struct bio
*bio
= NULL
;
923 int flag
= dpolicy
->sync
? REQ_SYNC
: 0;
925 if (dc
->state
!= D_PREP
)
928 trace_f2fs_issue_discard(dc
->bdev
, dc
->start
, dc
->len
);
930 dc
->error
= __blkdev_issue_discard(dc
->bdev
,
931 SECTOR_FROM_BLOCK(dc
->start
),
932 SECTOR_FROM_BLOCK(dc
->len
),
935 /* should keep before submission to avoid D_DONE right away */
936 dc
->state
= D_SUBMIT
;
937 atomic_inc(&dcc
->issued_discard
);
938 atomic_inc(&dcc
->issing_discard
);
940 bio
->bi_private
= dc
;
941 bio
->bi_end_io
= f2fs_submit_discard_endio
;
944 list_move_tail(&dc
->list
, wait_list
);
945 __check_sit_bitmap(sbi
, dc
->start
, dc
->start
+ dc
->len
);
947 f2fs_update_iostat(sbi
, FS_DISCARD
, 1);
950 __remove_discard_cmd(sbi
, dc
);
954 static struct discard_cmd
*__insert_discard_tree(struct f2fs_sb_info
*sbi
,
955 struct block_device
*bdev
, block_t lstart
,
956 block_t start
, block_t len
,
957 struct rb_node
**insert_p
,
958 struct rb_node
*insert_parent
)
960 struct discard_cmd_control
*dcc
= SM_I(sbi
)->dcc_info
;
962 struct rb_node
*parent
= NULL
;
963 struct discard_cmd
*dc
= NULL
;
965 if (insert_p
&& insert_parent
) {
966 parent
= insert_parent
;
971 p
= __lookup_rb_tree_for_insert(sbi
, &dcc
->root
, &parent
, lstart
);
973 dc
= __attach_discard_cmd(sbi
, bdev
, lstart
, start
, len
, parent
, p
);
980 static void __relocate_discard_cmd(struct discard_cmd_control
*dcc
,
981 struct discard_cmd
*dc
)
983 list_move_tail(&dc
->list
, &dcc
->pend_list
[plist_idx(dc
->len
)]);
986 static void __punch_discard_cmd(struct f2fs_sb_info
*sbi
,
987 struct discard_cmd
*dc
, block_t blkaddr
)
989 struct discard_cmd_control
*dcc
= SM_I(sbi
)->dcc_info
;
990 struct discard_info di
= dc
->di
;
991 bool modified
= false;
993 if (dc
->state
== D_DONE
|| dc
->len
== 1) {
994 __remove_discard_cmd(sbi
, dc
);
998 dcc
->undiscard_blks
-= di
.len
;
1000 if (blkaddr
> di
.lstart
) {
1001 dc
->len
= blkaddr
- dc
->lstart
;
1002 dcc
->undiscard_blks
+= dc
->len
;
1003 __relocate_discard_cmd(dcc
, dc
);
1007 if (blkaddr
< di
.lstart
+ di
.len
- 1) {
1009 __insert_discard_tree(sbi
, dc
->bdev
, blkaddr
+ 1,
1010 di
.start
+ blkaddr
+ 1 - di
.lstart
,
1011 di
.lstart
+ di
.len
- 1 - blkaddr
,
1017 dcc
->undiscard_blks
+= dc
->len
;
1018 __relocate_discard_cmd(dcc
, dc
);
1023 static void __update_discard_tree_range(struct f2fs_sb_info
*sbi
,
1024 struct block_device
*bdev
, block_t lstart
,
1025 block_t start
, block_t len
)
1027 struct discard_cmd_control
*dcc
= SM_I(sbi
)->dcc_info
;
1028 struct discard_cmd
*prev_dc
= NULL
, *next_dc
= NULL
;
1029 struct discard_cmd
*dc
;
1030 struct discard_info di
= {0};
1031 struct rb_node
**insert_p
= NULL
, *insert_parent
= NULL
;
1032 block_t end
= lstart
+ len
;
1034 mutex_lock(&dcc
->cmd_lock
);
1036 dc
= (struct discard_cmd
*)__lookup_rb_tree_ret(&dcc
->root
,
1038 (struct rb_entry
**)&prev_dc
,
1039 (struct rb_entry
**)&next_dc
,
1040 &insert_p
, &insert_parent
, true);
1046 di
.len
= next_dc
? next_dc
->lstart
- lstart
: len
;
1047 di
.len
= min(di
.len
, len
);
1052 struct rb_node
*node
;
1053 bool merged
= false;
1054 struct discard_cmd
*tdc
= NULL
;
1057 di
.lstart
= prev_dc
->lstart
+ prev_dc
->len
;
1058 if (di
.lstart
< lstart
)
1060 if (di
.lstart
>= end
)
1063 if (!next_dc
|| next_dc
->lstart
> end
)
1064 di
.len
= end
- di
.lstart
;
1066 di
.len
= next_dc
->lstart
- di
.lstart
;
1067 di
.start
= start
+ di
.lstart
- lstart
;
1073 if (prev_dc
&& prev_dc
->state
== D_PREP
&&
1074 prev_dc
->bdev
== bdev
&&
1075 __is_discard_back_mergeable(&di
, &prev_dc
->di
)) {
1076 prev_dc
->di
.len
+= di
.len
;
1077 dcc
->undiscard_blks
+= di
.len
;
1078 __relocate_discard_cmd(dcc
, prev_dc
);
1084 if (next_dc
&& next_dc
->state
== D_PREP
&&
1085 next_dc
->bdev
== bdev
&&
1086 __is_discard_front_mergeable(&di
, &next_dc
->di
)) {
1087 next_dc
->di
.lstart
= di
.lstart
;
1088 next_dc
->di
.len
+= di
.len
;
1089 next_dc
->di
.start
= di
.start
;
1090 dcc
->undiscard_blks
+= di
.len
;
1091 __relocate_discard_cmd(dcc
, next_dc
);
1093 __remove_discard_cmd(sbi
, tdc
);
1098 __insert_discard_tree(sbi
, bdev
, di
.lstart
, di
.start
,
1099 di
.len
, NULL
, NULL
);
1106 node
= rb_next(&prev_dc
->rb_node
);
1107 next_dc
= rb_entry_safe(node
, struct discard_cmd
, rb_node
);
1110 mutex_unlock(&dcc
->cmd_lock
);
1113 static int __queue_discard_cmd(struct f2fs_sb_info
*sbi
,
1114 struct block_device
*bdev
, block_t blkstart
, block_t blklen
)
1116 block_t lblkstart
= blkstart
;
1118 trace_f2fs_queue_discard(bdev
, blkstart
, blklen
);
1121 int devi
= f2fs_target_device_index(sbi
, blkstart
);
1123 blkstart
-= FDEV(devi
).start_blk
;
1125 __update_discard_tree_range(sbi
, bdev
, lblkstart
, blkstart
, blklen
);
1129 static void __issue_discard_cmd_range(struct f2fs_sb_info
*sbi
,
1130 struct discard_policy
*dpolicy
,
1131 unsigned int start
, unsigned int end
)
1133 struct discard_cmd_control
*dcc
= SM_I(sbi
)->dcc_info
;
1134 struct discard_cmd
*prev_dc
= NULL
, *next_dc
= NULL
;
1135 struct rb_node
**insert_p
= NULL
, *insert_parent
= NULL
;
1136 struct discard_cmd
*dc
;
1137 struct blk_plug plug
;
1143 mutex_lock(&dcc
->cmd_lock
);
1144 f2fs_bug_on(sbi
, !__check_rb_tree_consistence(sbi
, &dcc
->root
));
1146 dc
= (struct discard_cmd
*)__lookup_rb_tree_ret(&dcc
->root
,
1148 (struct rb_entry
**)&prev_dc
,
1149 (struct rb_entry
**)&next_dc
,
1150 &insert_p
, &insert_parent
, true);
1154 blk_start_plug(&plug
);
1156 while (dc
&& dc
->lstart
<= end
) {
1157 struct rb_node
*node
;
1159 if (dc
->len
< dpolicy
->granularity
)
1162 if (dc
->state
!= D_PREP
) {
1163 list_move_tail(&dc
->list
, &dcc
->fstrim_list
);
1167 __submit_discard_cmd(sbi
, dpolicy
, dc
);
1169 if (++issued
>= dpolicy
->max_requests
) {
1170 start
= dc
->lstart
+ dc
->len
;
1172 blk_finish_plug(&plug
);
1173 mutex_unlock(&dcc
->cmd_lock
);
1180 node
= rb_next(&dc
->rb_node
);
1181 dc
= rb_entry_safe(node
, struct discard_cmd
, rb_node
);
1183 if (fatal_signal_pending(current
))
1187 blk_finish_plug(&plug
);
1188 mutex_unlock(&dcc
->cmd_lock
);
1191 static int __issue_discard_cmd(struct f2fs_sb_info
*sbi
,
1192 struct discard_policy
*dpolicy
)
1194 struct discard_cmd_control
*dcc
= SM_I(sbi
)->dcc_info
;
1195 struct list_head
*pend_list
;
1196 struct discard_cmd
*dc
, *tmp
;
1197 struct blk_plug plug
;
1198 int i
, iter
= 0, issued
= 0;
1199 bool io_interrupted
= false;
1201 for (i
= MAX_PLIST_NUM
- 1; i
>= 0; i
--) {
1202 if (i
+ 1 < dpolicy
->granularity
)
1204 pend_list
= &dcc
->pend_list
[i
];
1206 mutex_lock(&dcc
->cmd_lock
);
1207 f2fs_bug_on(sbi
, !__check_rb_tree_consistence(sbi
, &dcc
->root
));
1208 blk_start_plug(&plug
);
1209 list_for_each_entry_safe(dc
, tmp
, pend_list
, list
) {
1210 f2fs_bug_on(sbi
, dc
->state
!= D_PREP
);
1212 if (dpolicy
->io_aware
&& i
< dpolicy
->io_aware_gran
&&
1214 io_interrupted
= true;
1218 __submit_discard_cmd(sbi
, dpolicy
, dc
);
1221 if (++iter
>= dpolicy
->max_requests
)
1224 blk_finish_plug(&plug
);
1225 mutex_unlock(&dcc
->cmd_lock
);
1227 if (iter
>= dpolicy
->max_requests
)
1231 if (!issued
&& io_interrupted
)
1237 static bool __drop_discard_cmd(struct f2fs_sb_info
*sbi
)
1239 struct discard_cmd_control
*dcc
= SM_I(sbi
)->dcc_info
;
1240 struct list_head
*pend_list
;
1241 struct discard_cmd
*dc
, *tmp
;
1243 bool dropped
= false;
1245 mutex_lock(&dcc
->cmd_lock
);
1246 for (i
= MAX_PLIST_NUM
- 1; i
>= 0; i
--) {
1247 pend_list
= &dcc
->pend_list
[i
];
1248 list_for_each_entry_safe(dc
, tmp
, pend_list
, list
) {
1249 f2fs_bug_on(sbi
, dc
->state
!= D_PREP
);
1250 __remove_discard_cmd(sbi
, dc
);
1254 mutex_unlock(&dcc
->cmd_lock
);
1259 static unsigned int __wait_one_discard_bio(struct f2fs_sb_info
*sbi
,
1260 struct discard_cmd
*dc
)
1262 struct discard_cmd_control
*dcc
= SM_I(sbi
)->dcc_info
;
1263 unsigned int len
= 0;
1265 wait_for_completion_io(&dc
->wait
);
1266 mutex_lock(&dcc
->cmd_lock
);
1267 f2fs_bug_on(sbi
, dc
->state
!= D_DONE
);
1272 __remove_discard_cmd(sbi
, dc
);
1274 mutex_unlock(&dcc
->cmd_lock
);
1279 static unsigned int __wait_discard_cmd_range(struct f2fs_sb_info
*sbi
,
1280 struct discard_policy
*dpolicy
,
1281 block_t start
, block_t end
)
1283 struct discard_cmd_control
*dcc
= SM_I(sbi
)->dcc_info
;
1284 struct list_head
*wait_list
= (dpolicy
->type
== DPOLICY_FSTRIM
) ?
1285 &(dcc
->fstrim_list
) : &(dcc
->wait_list
);
1286 struct discard_cmd
*dc
, *tmp
;
1288 unsigned int trimmed
= 0;
1293 mutex_lock(&dcc
->cmd_lock
);
1294 list_for_each_entry_safe(dc
, tmp
, wait_list
, list
) {
1295 if (dc
->lstart
+ dc
->len
<= start
|| end
<= dc
->lstart
)
1297 if (dc
->len
< dpolicy
->granularity
)
1299 if (dc
->state
== D_DONE
&& !dc
->ref
) {
1300 wait_for_completion_io(&dc
->wait
);
1303 __remove_discard_cmd(sbi
, dc
);
1310 mutex_unlock(&dcc
->cmd_lock
);
1313 trimmed
+= __wait_one_discard_bio(sbi
, dc
);
1320 static void __wait_all_discard_cmd(struct f2fs_sb_info
*sbi
,
1321 struct discard_policy
*dpolicy
)
1323 __wait_discard_cmd_range(sbi
, dpolicy
, 0, UINT_MAX
);
1326 /* This should be covered by global mutex, &sit_i->sentry_lock */
1327 void f2fs_wait_discard_bio(struct f2fs_sb_info
*sbi
, block_t blkaddr
)
1329 struct discard_cmd_control
*dcc
= SM_I(sbi
)->dcc_info
;
1330 struct discard_cmd
*dc
;
1331 bool need_wait
= false;
1333 mutex_lock(&dcc
->cmd_lock
);
1334 dc
= (struct discard_cmd
*)__lookup_rb_tree(&dcc
->root
, NULL
, blkaddr
);
1336 if (dc
->state
== D_PREP
) {
1337 __punch_discard_cmd(sbi
, dc
, blkaddr
);
1343 mutex_unlock(&dcc
->cmd_lock
);
1346 __wait_one_discard_bio(sbi
, dc
);
1349 void stop_discard_thread(struct f2fs_sb_info
*sbi
)
1351 struct discard_cmd_control
*dcc
= SM_I(sbi
)->dcc_info
;
1353 if (dcc
&& dcc
->f2fs_issue_discard
) {
1354 struct task_struct
*discard_thread
= dcc
->f2fs_issue_discard
;
1356 dcc
->f2fs_issue_discard
= NULL
;
1357 kthread_stop(discard_thread
);
1361 /* This comes from f2fs_put_super */
1362 bool f2fs_wait_discard_bios(struct f2fs_sb_info
*sbi
)
1364 struct discard_cmd_control
*dcc
= SM_I(sbi
)->dcc_info
;
1365 struct discard_policy dpolicy
;
1368 init_discard_policy(&dpolicy
, DPOLICY_UMOUNT
, dcc
->discard_granularity
);
1369 __issue_discard_cmd(sbi
, &dpolicy
);
1370 dropped
= __drop_discard_cmd(sbi
);
1371 __wait_all_discard_cmd(sbi
, &dpolicy
);
1376 static int issue_discard_thread(void *data
)
1378 struct f2fs_sb_info
*sbi
= data
;
1379 struct discard_cmd_control
*dcc
= SM_I(sbi
)->dcc_info
;
1380 wait_queue_head_t
*q
= &dcc
->discard_wait_queue
;
1381 struct discard_policy dpolicy
;
1382 unsigned int wait_ms
= DEF_MIN_DISCARD_ISSUE_TIME
;
1388 init_discard_policy(&dpolicy
, DPOLICY_BG
,
1389 dcc
->discard_granularity
);
1391 wait_event_interruptible_timeout(*q
,
1392 kthread_should_stop() || freezing(current
) ||
1394 msecs_to_jiffies(wait_ms
));
1395 if (try_to_freeze())
1397 if (kthread_should_stop())
1400 if (dcc
->discard_wake
) {
1401 dcc
->discard_wake
= 0;
1402 if (sbi
->gc_thread
&& sbi
->gc_thread
->gc_urgent
)
1403 init_discard_policy(&dpolicy
,
1407 sb_start_intwrite(sbi
->sb
);
1409 issued
= __issue_discard_cmd(sbi
, &dpolicy
);
1411 __wait_all_discard_cmd(sbi
, &dpolicy
);
1412 wait_ms
= dpolicy
.min_interval
;
1414 wait_ms
= dpolicy
.max_interval
;
1417 sb_end_intwrite(sbi
->sb
);
1419 } while (!kthread_should_stop());
1423 #ifdef CONFIG_BLK_DEV_ZONED
1424 static int __f2fs_issue_discard_zone(struct f2fs_sb_info
*sbi
,
1425 struct block_device
*bdev
, block_t blkstart
, block_t blklen
)
1427 sector_t sector
, nr_sects
;
1428 block_t lblkstart
= blkstart
;
1432 devi
= f2fs_target_device_index(sbi
, blkstart
);
1433 blkstart
-= FDEV(devi
).start_blk
;
1437 * We need to know the type of the zone: for conventional zones,
1438 * use regular discard if the drive supports it. For sequential
1439 * zones, reset the zone write pointer.
1441 switch (get_blkz_type(sbi
, bdev
, blkstart
)) {
1443 case BLK_ZONE_TYPE_CONVENTIONAL
:
1444 if (!blk_queue_discard(bdev_get_queue(bdev
)))
1446 return __queue_discard_cmd(sbi
, bdev
, lblkstart
, blklen
);
1447 case BLK_ZONE_TYPE_SEQWRITE_REQ
:
1448 case BLK_ZONE_TYPE_SEQWRITE_PREF
:
1449 sector
= SECTOR_FROM_BLOCK(blkstart
);
1450 nr_sects
= SECTOR_FROM_BLOCK(blklen
);
1452 if (sector
& (bdev_zone_sectors(bdev
) - 1) ||
1453 nr_sects
!= bdev_zone_sectors(bdev
)) {
1454 f2fs_msg(sbi
->sb
, KERN_INFO
,
1455 "(%d) %s: Unaligned discard attempted (block %x + %x)",
1456 devi
, sbi
->s_ndevs
? FDEV(devi
).path
: "",
1460 trace_f2fs_issue_reset_zone(bdev
, blkstart
);
1461 return blkdev_reset_zones(bdev
, sector
,
1462 nr_sects
, GFP_NOFS
);
1464 /* Unknown zone type: broken device ? */
1470 static int __issue_discard_async(struct f2fs_sb_info
*sbi
,
1471 struct block_device
*bdev
, block_t blkstart
, block_t blklen
)
1473 #ifdef CONFIG_BLK_DEV_ZONED
1474 if (f2fs_sb_mounted_blkzoned(sbi
->sb
) &&
1475 bdev_zoned_model(bdev
) != BLK_ZONED_NONE
)
1476 return __f2fs_issue_discard_zone(sbi
, bdev
, blkstart
, blklen
);
1478 return __queue_discard_cmd(sbi
, bdev
, blkstart
, blklen
);
1481 static int f2fs_issue_discard(struct f2fs_sb_info
*sbi
,
1482 block_t blkstart
, block_t blklen
)
1484 sector_t start
= blkstart
, len
= 0;
1485 struct block_device
*bdev
;
1486 struct seg_entry
*se
;
1487 unsigned int offset
;
1491 bdev
= f2fs_target_device(sbi
, blkstart
, NULL
);
1493 for (i
= blkstart
; i
< blkstart
+ blklen
; i
++, len
++) {
1495 struct block_device
*bdev2
=
1496 f2fs_target_device(sbi
, i
, NULL
);
1498 if (bdev2
!= bdev
) {
1499 err
= __issue_discard_async(sbi
, bdev
,
1509 se
= get_seg_entry(sbi
, GET_SEGNO(sbi
, i
));
1510 offset
= GET_BLKOFF_FROM_SEG0(sbi
, i
);
1512 if (!f2fs_test_and_set_bit(offset
, se
->discard_map
))
1513 sbi
->discard_blks
--;
1517 err
= __issue_discard_async(sbi
, bdev
, start
, len
);
1521 static bool add_discard_addrs(struct f2fs_sb_info
*sbi
, struct cp_control
*cpc
,
1524 int entries
= SIT_VBLOCK_MAP_SIZE
/ sizeof(unsigned long);
1525 int max_blocks
= sbi
->blocks_per_seg
;
1526 struct seg_entry
*se
= get_seg_entry(sbi
, cpc
->trim_start
);
1527 unsigned long *cur_map
= (unsigned long *)se
->cur_valid_map
;
1528 unsigned long *ckpt_map
= (unsigned long *)se
->ckpt_valid_map
;
1529 unsigned long *discard_map
= (unsigned long *)se
->discard_map
;
1530 unsigned long *dmap
= SIT_I(sbi
)->tmp_map
;
1531 unsigned int start
= 0, end
= -1;
1532 bool force
= (cpc
->reason
& CP_DISCARD
);
1533 struct discard_entry
*de
= NULL
;
1534 struct list_head
*head
= &SM_I(sbi
)->dcc_info
->entry_list
;
1537 if (se
->valid_blocks
== max_blocks
|| !f2fs_discard_en(sbi
))
1541 if (!test_opt(sbi
, DISCARD
) || !se
->valid_blocks
||
1542 SM_I(sbi
)->dcc_info
->nr_discards
>=
1543 SM_I(sbi
)->dcc_info
->max_discards
)
1547 /* SIT_VBLOCK_MAP_SIZE should be multiple of sizeof(unsigned long) */
1548 for (i
= 0; i
< entries
; i
++)
1549 dmap
[i
] = force
? ~ckpt_map
[i
] & ~discard_map
[i
] :
1550 (cur_map
[i
] ^ ckpt_map
[i
]) & ckpt_map
[i
];
1552 while (force
|| SM_I(sbi
)->dcc_info
->nr_discards
<=
1553 SM_I(sbi
)->dcc_info
->max_discards
) {
1554 start
= __find_rev_next_bit(dmap
, max_blocks
, end
+ 1);
1555 if (start
>= max_blocks
)
1558 end
= __find_rev_next_zero_bit(dmap
, max_blocks
, start
+ 1);
1559 if (force
&& start
&& end
!= max_blocks
1560 && (end
- start
) < cpc
->trim_minlen
)
1567 de
= f2fs_kmem_cache_alloc(discard_entry_slab
,
1569 de
->start_blkaddr
= START_BLOCK(sbi
, cpc
->trim_start
);
1570 list_add_tail(&de
->list
, head
);
1573 for (i
= start
; i
< end
; i
++)
1574 __set_bit_le(i
, (void *)de
->discard_map
);
1576 SM_I(sbi
)->dcc_info
->nr_discards
+= end
- start
;
1581 void release_discard_addrs(struct f2fs_sb_info
*sbi
)
1583 struct list_head
*head
= &(SM_I(sbi
)->dcc_info
->entry_list
);
1584 struct discard_entry
*entry
, *this;
1587 list_for_each_entry_safe(entry
, this, head
, list
) {
1588 list_del(&entry
->list
);
1589 kmem_cache_free(discard_entry_slab
, entry
);
1594 * Should call clear_prefree_segments after checkpoint is done.
1596 static void set_prefree_as_free_segments(struct f2fs_sb_info
*sbi
)
1598 struct dirty_seglist_info
*dirty_i
= DIRTY_I(sbi
);
1601 mutex_lock(&dirty_i
->seglist_lock
);
1602 for_each_set_bit(segno
, dirty_i
->dirty_segmap
[PRE
], MAIN_SEGS(sbi
))
1603 __set_test_and_free(sbi
, segno
);
1604 mutex_unlock(&dirty_i
->seglist_lock
);
1607 void clear_prefree_segments(struct f2fs_sb_info
*sbi
, struct cp_control
*cpc
)
1609 struct discard_cmd_control
*dcc
= SM_I(sbi
)->dcc_info
;
1610 struct list_head
*head
= &dcc
->entry_list
;
1611 struct discard_entry
*entry
, *this;
1612 struct dirty_seglist_info
*dirty_i
= DIRTY_I(sbi
);
1613 unsigned long *prefree_map
= dirty_i
->dirty_segmap
[PRE
];
1614 unsigned int start
= 0, end
= -1;
1615 unsigned int secno
, start_segno
;
1616 bool force
= (cpc
->reason
& CP_DISCARD
);
1618 mutex_lock(&dirty_i
->seglist_lock
);
1622 start
= find_next_bit(prefree_map
, MAIN_SEGS(sbi
), end
+ 1);
1623 if (start
>= MAIN_SEGS(sbi
))
1625 end
= find_next_zero_bit(prefree_map
, MAIN_SEGS(sbi
),
1628 for (i
= start
; i
< end
; i
++)
1629 clear_bit(i
, prefree_map
);
1631 dirty_i
->nr_dirty
[PRE
] -= end
- start
;
1633 if (!test_opt(sbi
, DISCARD
))
1636 if (force
&& start
>= cpc
->trim_start
&&
1637 (end
- 1) <= cpc
->trim_end
)
1640 if (!test_opt(sbi
, LFS
) || sbi
->segs_per_sec
== 1) {
1641 f2fs_issue_discard(sbi
, START_BLOCK(sbi
, start
),
1642 (end
- start
) << sbi
->log_blocks_per_seg
);
1646 secno
= GET_SEC_FROM_SEG(sbi
, start
);
1647 start_segno
= GET_SEG_FROM_SEC(sbi
, secno
);
1648 if (!IS_CURSEC(sbi
, secno
) &&
1649 !get_valid_blocks(sbi
, start
, true))
1650 f2fs_issue_discard(sbi
, START_BLOCK(sbi
, start_segno
),
1651 sbi
->segs_per_sec
<< sbi
->log_blocks_per_seg
);
1653 start
= start_segno
+ sbi
->segs_per_sec
;
1659 mutex_unlock(&dirty_i
->seglist_lock
);
1661 /* send small discards */
1662 list_for_each_entry_safe(entry
, this, head
, list
) {
1663 unsigned int cur_pos
= 0, next_pos
, len
, total_len
= 0;
1664 bool is_valid
= test_bit_le(0, entry
->discard_map
);
1668 next_pos
= find_next_zero_bit_le(entry
->discard_map
,
1669 sbi
->blocks_per_seg
, cur_pos
);
1670 len
= next_pos
- cur_pos
;
1672 if (f2fs_sb_mounted_blkzoned(sbi
->sb
) ||
1673 (force
&& len
< cpc
->trim_minlen
))
1676 f2fs_issue_discard(sbi
, entry
->start_blkaddr
+ cur_pos
,
1680 next_pos
= find_next_bit_le(entry
->discard_map
,
1681 sbi
->blocks_per_seg
, cur_pos
);
1685 is_valid
= !is_valid
;
1687 if (cur_pos
< sbi
->blocks_per_seg
)
1690 list_del(&entry
->list
);
1691 dcc
->nr_discards
-= total_len
;
1692 kmem_cache_free(discard_entry_slab
, entry
);
1695 wake_up_discard_thread(sbi
, false);
1698 void init_discard_policy(struct discard_policy
*dpolicy
,
1699 int discard_type
, unsigned int granularity
)
1702 dpolicy
->type
= discard_type
;
1703 dpolicy
->sync
= true;
1704 dpolicy
->granularity
= granularity
;
1706 if (discard_type
== DPOLICY_BG
) {
1707 dpolicy
->min_interval
= DEF_MIN_DISCARD_ISSUE_TIME
;
1708 dpolicy
->max_interval
= DEF_MAX_DISCARD_ISSUE_TIME
;
1709 dpolicy
->max_requests
= DEF_MAX_DISCARD_REQUEST
;
1710 dpolicy
->io_aware_gran
= MAX_PLIST_NUM
;
1711 dpolicy
->io_aware
= true;
1712 } else if (discard_type
== DPOLICY_FORCE
) {
1713 dpolicy
->min_interval
= DEF_MIN_DISCARD_ISSUE_TIME
;
1714 dpolicy
->max_interval
= DEF_MAX_DISCARD_ISSUE_TIME
;
1715 dpolicy
->max_requests
= DEF_MAX_DISCARD_REQUEST
;
1716 dpolicy
->io_aware_gran
= MAX_PLIST_NUM
;
1717 dpolicy
->io_aware
= true;
1718 } else if (discard_type
== DPOLICY_FSTRIM
) {
1719 dpolicy
->max_requests
= DEF_MAX_DISCARD_REQUEST
;
1720 dpolicy
->io_aware_gran
= MAX_PLIST_NUM
;
1721 dpolicy
->io_aware
= false;
1722 } else if (discard_type
== DPOLICY_UMOUNT
) {
1723 dpolicy
->max_requests
= DEF_MAX_DISCARD_REQUEST
;
1724 dpolicy
->io_aware_gran
= MAX_PLIST_NUM
;
1725 dpolicy
->io_aware
= false;
1729 static int create_discard_cmd_control(struct f2fs_sb_info
*sbi
)
1731 dev_t dev
= sbi
->sb
->s_bdev
->bd_dev
;
1732 struct discard_cmd_control
*dcc
;
1735 if (SM_I(sbi
)->dcc_info
) {
1736 dcc
= SM_I(sbi
)->dcc_info
;
1740 dcc
= kzalloc(sizeof(struct discard_cmd_control
), GFP_KERNEL
);
1744 dcc
->discard_granularity
= DEFAULT_DISCARD_GRANULARITY
;
1745 INIT_LIST_HEAD(&dcc
->entry_list
);
1746 for (i
= 0; i
< MAX_PLIST_NUM
; i
++)
1747 INIT_LIST_HEAD(&dcc
->pend_list
[i
]);
1748 INIT_LIST_HEAD(&dcc
->wait_list
);
1749 INIT_LIST_HEAD(&dcc
->fstrim_list
);
1750 mutex_init(&dcc
->cmd_lock
);
1751 atomic_set(&dcc
->issued_discard
, 0);
1752 atomic_set(&dcc
->issing_discard
, 0);
1753 atomic_set(&dcc
->discard_cmd_cnt
, 0);
1754 dcc
->nr_discards
= 0;
1755 dcc
->max_discards
= MAIN_SEGS(sbi
) << sbi
->log_blocks_per_seg
;
1756 dcc
->undiscard_blks
= 0;
1757 dcc
->root
= RB_ROOT
;
1759 init_waitqueue_head(&dcc
->discard_wait_queue
);
1760 SM_I(sbi
)->dcc_info
= dcc
;
1762 dcc
->f2fs_issue_discard
= kthread_run(issue_discard_thread
, sbi
,
1763 "f2fs_discard-%u:%u", MAJOR(dev
), MINOR(dev
));
1764 if (IS_ERR(dcc
->f2fs_issue_discard
)) {
1765 err
= PTR_ERR(dcc
->f2fs_issue_discard
);
1767 SM_I(sbi
)->dcc_info
= NULL
;
1774 static void destroy_discard_cmd_control(struct f2fs_sb_info
*sbi
)
1776 struct discard_cmd_control
*dcc
= SM_I(sbi
)->dcc_info
;
1781 stop_discard_thread(sbi
);
1784 SM_I(sbi
)->dcc_info
= NULL
;
1787 static bool __mark_sit_entry_dirty(struct f2fs_sb_info
*sbi
, unsigned int segno
)
1789 struct sit_info
*sit_i
= SIT_I(sbi
);
1791 if (!__test_and_set_bit(segno
, sit_i
->dirty_sentries_bitmap
)) {
1792 sit_i
->dirty_sentries
++;
1799 static void __set_sit_entry_type(struct f2fs_sb_info
*sbi
, int type
,
1800 unsigned int segno
, int modified
)
1802 struct seg_entry
*se
= get_seg_entry(sbi
, segno
);
1805 __mark_sit_entry_dirty(sbi
, segno
);
1808 static void update_sit_entry(struct f2fs_sb_info
*sbi
, block_t blkaddr
, int del
)
1810 struct seg_entry
*se
;
1811 unsigned int segno
, offset
;
1812 long int new_vblocks
;
1814 #ifdef CONFIG_F2FS_CHECK_FS
1818 segno
= GET_SEGNO(sbi
, blkaddr
);
1820 se
= get_seg_entry(sbi
, segno
);
1821 new_vblocks
= se
->valid_blocks
+ del
;
1822 offset
= GET_BLKOFF_FROM_SEG0(sbi
, blkaddr
);
1824 f2fs_bug_on(sbi
, (new_vblocks
>> (sizeof(unsigned short) << 3) ||
1825 (new_vblocks
> sbi
->blocks_per_seg
)));
1827 se
->valid_blocks
= new_vblocks
;
1828 se
->mtime
= get_mtime(sbi
);
1829 SIT_I(sbi
)->max_mtime
= se
->mtime
;
1831 /* Update valid block bitmap */
1833 exist
= f2fs_test_and_set_bit(offset
, se
->cur_valid_map
);
1834 #ifdef CONFIG_F2FS_CHECK_FS
1835 mir_exist
= f2fs_test_and_set_bit(offset
,
1836 se
->cur_valid_map_mir
);
1837 if (unlikely(exist
!= mir_exist
)) {
1838 f2fs_msg(sbi
->sb
, KERN_ERR
, "Inconsistent error "
1839 "when setting bitmap, blk:%u, old bit:%d",
1841 f2fs_bug_on(sbi
, 1);
1844 if (unlikely(exist
)) {
1845 f2fs_msg(sbi
->sb
, KERN_ERR
,
1846 "Bitmap was wrongly set, blk:%u", blkaddr
);
1847 f2fs_bug_on(sbi
, 1);
1852 if (f2fs_discard_en(sbi
) &&
1853 !f2fs_test_and_set_bit(offset
, se
->discard_map
))
1854 sbi
->discard_blks
--;
1856 /* don't overwrite by SSR to keep node chain */
1857 if (se
->type
== CURSEG_WARM_NODE
) {
1858 if (!f2fs_test_and_set_bit(offset
, se
->ckpt_valid_map
))
1859 se
->ckpt_valid_blocks
++;
1862 exist
= f2fs_test_and_clear_bit(offset
, se
->cur_valid_map
);
1863 #ifdef CONFIG_F2FS_CHECK_FS
1864 mir_exist
= f2fs_test_and_clear_bit(offset
,
1865 se
->cur_valid_map_mir
);
1866 if (unlikely(exist
!= mir_exist
)) {
1867 f2fs_msg(sbi
->sb
, KERN_ERR
, "Inconsistent error "
1868 "when clearing bitmap, blk:%u, old bit:%d",
1870 f2fs_bug_on(sbi
, 1);
1873 if (unlikely(!exist
)) {
1874 f2fs_msg(sbi
->sb
, KERN_ERR
,
1875 "Bitmap was wrongly cleared, blk:%u", blkaddr
);
1876 f2fs_bug_on(sbi
, 1);
1881 if (f2fs_discard_en(sbi
) &&
1882 f2fs_test_and_clear_bit(offset
, se
->discard_map
))
1883 sbi
->discard_blks
++;
1885 if (!f2fs_test_bit(offset
, se
->ckpt_valid_map
))
1886 se
->ckpt_valid_blocks
+= del
;
1888 __mark_sit_entry_dirty(sbi
, segno
);
1890 /* update total number of valid blocks to be written in ckpt area */
1891 SIT_I(sbi
)->written_valid_blocks
+= del
;
1893 if (sbi
->segs_per_sec
> 1)
1894 get_sec_entry(sbi
, segno
)->valid_blocks
+= del
;
1897 void invalidate_blocks(struct f2fs_sb_info
*sbi
, block_t addr
)
1899 unsigned int segno
= GET_SEGNO(sbi
, addr
);
1900 struct sit_info
*sit_i
= SIT_I(sbi
);
1902 f2fs_bug_on(sbi
, addr
== NULL_ADDR
);
1903 if (addr
== NEW_ADDR
)
1906 /* add it into sit main buffer */
1907 down_write(&sit_i
->sentry_lock
);
1909 update_sit_entry(sbi
, addr
, -1);
1911 /* add it into dirty seglist */
1912 locate_dirty_segment(sbi
, segno
);
1914 up_write(&sit_i
->sentry_lock
);
1917 bool is_checkpointed_data(struct f2fs_sb_info
*sbi
, block_t blkaddr
)
1919 struct sit_info
*sit_i
= SIT_I(sbi
);
1920 unsigned int segno
, offset
;
1921 struct seg_entry
*se
;
1924 if (blkaddr
== NEW_ADDR
|| blkaddr
== NULL_ADDR
)
1927 down_read(&sit_i
->sentry_lock
);
1929 segno
= GET_SEGNO(sbi
, blkaddr
);
1930 se
= get_seg_entry(sbi
, segno
);
1931 offset
= GET_BLKOFF_FROM_SEG0(sbi
, blkaddr
);
1933 if (f2fs_test_bit(offset
, se
->ckpt_valid_map
))
1936 up_read(&sit_i
->sentry_lock
);
1942 * This function should be resided under the curseg_mutex lock
1944 static void __add_sum_entry(struct f2fs_sb_info
*sbi
, int type
,
1945 struct f2fs_summary
*sum
)
1947 struct curseg_info
*curseg
= CURSEG_I(sbi
, type
);
1948 void *addr
= curseg
->sum_blk
;
1949 addr
+= curseg
->next_blkoff
* sizeof(struct f2fs_summary
);
1950 memcpy(addr
, sum
, sizeof(struct f2fs_summary
));
1954 * Calculate the number of current summary pages for writing
1956 int npages_for_summary_flush(struct f2fs_sb_info
*sbi
, bool for_ra
)
1958 int valid_sum_count
= 0;
1961 for (i
= CURSEG_HOT_DATA
; i
<= CURSEG_COLD_DATA
; i
++) {
1962 if (sbi
->ckpt
->alloc_type
[i
] == SSR
)
1963 valid_sum_count
+= sbi
->blocks_per_seg
;
1966 valid_sum_count
+= le16_to_cpu(
1967 F2FS_CKPT(sbi
)->cur_data_blkoff
[i
]);
1969 valid_sum_count
+= curseg_blkoff(sbi
, i
);
1973 sum_in_page
= (PAGE_SIZE
- 2 * SUM_JOURNAL_SIZE
-
1974 SUM_FOOTER_SIZE
) / SUMMARY_SIZE
;
1975 if (valid_sum_count
<= sum_in_page
)
1977 else if ((valid_sum_count
- sum_in_page
) <=
1978 (PAGE_SIZE
- SUM_FOOTER_SIZE
) / SUMMARY_SIZE
)
1984 * Caller should put this summary page
1986 struct page
*get_sum_page(struct f2fs_sb_info
*sbi
, unsigned int segno
)
1988 return get_meta_page(sbi
, GET_SUM_BLOCK(sbi
, segno
));
1991 void update_meta_page(struct f2fs_sb_info
*sbi
, void *src
, block_t blk_addr
)
1993 struct page
*page
= grab_meta_page(sbi
, blk_addr
);
1995 memcpy(page_address(page
), src
, PAGE_SIZE
);
1996 set_page_dirty(page
);
1997 f2fs_put_page(page
, 1);
2000 static void write_sum_page(struct f2fs_sb_info
*sbi
,
2001 struct f2fs_summary_block
*sum_blk
, block_t blk_addr
)
2003 update_meta_page(sbi
, (void *)sum_blk
, blk_addr
);
2006 static void write_current_sum_page(struct f2fs_sb_info
*sbi
,
2007 int type
, block_t blk_addr
)
2009 struct curseg_info
*curseg
= CURSEG_I(sbi
, type
);
2010 struct page
*page
= grab_meta_page(sbi
, blk_addr
);
2011 struct f2fs_summary_block
*src
= curseg
->sum_blk
;
2012 struct f2fs_summary_block
*dst
;
2014 dst
= (struct f2fs_summary_block
*)page_address(page
);
2015 memset(dst
, 0, PAGE_SIZE
);
2017 mutex_lock(&curseg
->curseg_mutex
);
2019 down_read(&curseg
->journal_rwsem
);
2020 memcpy(&dst
->journal
, curseg
->journal
, SUM_JOURNAL_SIZE
);
2021 up_read(&curseg
->journal_rwsem
);
2023 memcpy(dst
->entries
, src
->entries
, SUM_ENTRY_SIZE
);
2024 memcpy(&dst
->footer
, &src
->footer
, SUM_FOOTER_SIZE
);
2026 mutex_unlock(&curseg
->curseg_mutex
);
2028 set_page_dirty(page
);
2029 f2fs_put_page(page
, 1);
2032 static int is_next_segment_free(struct f2fs_sb_info
*sbi
, int type
)
2034 struct curseg_info
*curseg
= CURSEG_I(sbi
, type
);
2035 unsigned int segno
= curseg
->segno
+ 1;
2036 struct free_segmap_info
*free_i
= FREE_I(sbi
);
2038 if (segno
< MAIN_SEGS(sbi
) && segno
% sbi
->segs_per_sec
)
2039 return !test_bit(segno
, free_i
->free_segmap
);
2044 * Find a new segment from the free segments bitmap to right order
2045 * This function should be returned with success, otherwise BUG
2047 static void get_new_segment(struct f2fs_sb_info
*sbi
,
2048 unsigned int *newseg
, bool new_sec
, int dir
)
2050 struct free_segmap_info
*free_i
= FREE_I(sbi
);
2051 unsigned int segno
, secno
, zoneno
;
2052 unsigned int total_zones
= MAIN_SECS(sbi
) / sbi
->secs_per_zone
;
2053 unsigned int hint
= GET_SEC_FROM_SEG(sbi
, *newseg
);
2054 unsigned int old_zoneno
= GET_ZONE_FROM_SEG(sbi
, *newseg
);
2055 unsigned int left_start
= hint
;
2060 spin_lock(&free_i
->segmap_lock
);
2062 if (!new_sec
&& ((*newseg
+ 1) % sbi
->segs_per_sec
)) {
2063 segno
= find_next_zero_bit(free_i
->free_segmap
,
2064 GET_SEG_FROM_SEC(sbi
, hint
+ 1), *newseg
+ 1);
2065 if (segno
< GET_SEG_FROM_SEC(sbi
, hint
+ 1))
2069 secno
= find_next_zero_bit(free_i
->free_secmap
, MAIN_SECS(sbi
), hint
);
2070 if (secno
>= MAIN_SECS(sbi
)) {
2071 if (dir
== ALLOC_RIGHT
) {
2072 secno
= find_next_zero_bit(free_i
->free_secmap
,
2074 f2fs_bug_on(sbi
, secno
>= MAIN_SECS(sbi
));
2077 left_start
= hint
- 1;
2083 while (test_bit(left_start
, free_i
->free_secmap
)) {
2084 if (left_start
> 0) {
2088 left_start
= find_next_zero_bit(free_i
->free_secmap
,
2090 f2fs_bug_on(sbi
, left_start
>= MAIN_SECS(sbi
));
2095 segno
= GET_SEG_FROM_SEC(sbi
, secno
);
2096 zoneno
= GET_ZONE_FROM_SEC(sbi
, secno
);
2098 /* give up on finding another zone */
2101 if (sbi
->secs_per_zone
== 1)
2103 if (zoneno
== old_zoneno
)
2105 if (dir
== ALLOC_LEFT
) {
2106 if (!go_left
&& zoneno
+ 1 >= total_zones
)
2108 if (go_left
&& zoneno
== 0)
2111 for (i
= 0; i
< NR_CURSEG_TYPE
; i
++)
2112 if (CURSEG_I(sbi
, i
)->zone
== zoneno
)
2115 if (i
< NR_CURSEG_TYPE
) {
2116 /* zone is in user, try another */
2118 hint
= zoneno
* sbi
->secs_per_zone
- 1;
2119 else if (zoneno
+ 1 >= total_zones
)
2122 hint
= (zoneno
+ 1) * sbi
->secs_per_zone
;
2124 goto find_other_zone
;
2127 /* set it as dirty segment in free segmap */
2128 f2fs_bug_on(sbi
, test_bit(segno
, free_i
->free_segmap
));
2129 __set_inuse(sbi
, segno
);
2131 spin_unlock(&free_i
->segmap_lock
);
2134 static void reset_curseg(struct f2fs_sb_info
*sbi
, int type
, int modified
)
2136 struct curseg_info
*curseg
= CURSEG_I(sbi
, type
);
2137 struct summary_footer
*sum_footer
;
2139 curseg
->segno
= curseg
->next_segno
;
2140 curseg
->zone
= GET_ZONE_FROM_SEG(sbi
, curseg
->segno
);
2141 curseg
->next_blkoff
= 0;
2142 curseg
->next_segno
= NULL_SEGNO
;
2144 sum_footer
= &(curseg
->sum_blk
->footer
);
2145 memset(sum_footer
, 0, sizeof(struct summary_footer
));
2146 if (IS_DATASEG(type
))
2147 SET_SUM_TYPE(sum_footer
, SUM_TYPE_DATA
);
2148 if (IS_NODESEG(type
))
2149 SET_SUM_TYPE(sum_footer
, SUM_TYPE_NODE
);
2150 __set_sit_entry_type(sbi
, type
, curseg
->segno
, modified
);
2153 static unsigned int __get_next_segno(struct f2fs_sb_info
*sbi
, int type
)
2155 /* if segs_per_sec is large than 1, we need to keep original policy. */
2156 if (sbi
->segs_per_sec
!= 1)
2157 return CURSEG_I(sbi
, type
)->segno
;
2159 if (test_opt(sbi
, NOHEAP
) &&
2160 (type
== CURSEG_HOT_DATA
|| IS_NODESEG(type
)))
2163 if (SIT_I(sbi
)->last_victim
[ALLOC_NEXT
])
2164 return SIT_I(sbi
)->last_victim
[ALLOC_NEXT
];
2165 return CURSEG_I(sbi
, type
)->segno
;
2169 * Allocate a current working segment.
2170 * This function always allocates a free segment in LFS manner.
2172 static void new_curseg(struct f2fs_sb_info
*sbi
, int type
, bool new_sec
)
2174 struct curseg_info
*curseg
= CURSEG_I(sbi
, type
);
2175 unsigned int segno
= curseg
->segno
;
2176 int dir
= ALLOC_LEFT
;
2178 write_sum_page(sbi
, curseg
->sum_blk
,
2179 GET_SUM_BLOCK(sbi
, segno
));
2180 if (type
== CURSEG_WARM_DATA
|| type
== CURSEG_COLD_DATA
)
2183 if (test_opt(sbi
, NOHEAP
))
2186 segno
= __get_next_segno(sbi
, type
);
2187 get_new_segment(sbi
, &segno
, new_sec
, dir
);
2188 curseg
->next_segno
= segno
;
2189 reset_curseg(sbi
, type
, 1);
2190 curseg
->alloc_type
= LFS
;
2193 static void __next_free_blkoff(struct f2fs_sb_info
*sbi
,
2194 struct curseg_info
*seg
, block_t start
)
2196 struct seg_entry
*se
= get_seg_entry(sbi
, seg
->segno
);
2197 int entries
= SIT_VBLOCK_MAP_SIZE
/ sizeof(unsigned long);
2198 unsigned long *target_map
= SIT_I(sbi
)->tmp_map
;
2199 unsigned long *ckpt_map
= (unsigned long *)se
->ckpt_valid_map
;
2200 unsigned long *cur_map
= (unsigned long *)se
->cur_valid_map
;
2203 for (i
= 0; i
< entries
; i
++)
2204 target_map
[i
] = ckpt_map
[i
] | cur_map
[i
];
2206 pos
= __find_rev_next_zero_bit(target_map
, sbi
->blocks_per_seg
, start
);
2208 seg
->next_blkoff
= pos
;
2212 * If a segment is written by LFS manner, next block offset is just obtained
2213 * by increasing the current block offset. However, if a segment is written by
2214 * SSR manner, next block offset obtained by calling __next_free_blkoff
2216 static void __refresh_next_blkoff(struct f2fs_sb_info
*sbi
,
2217 struct curseg_info
*seg
)
2219 if (seg
->alloc_type
== SSR
)
2220 __next_free_blkoff(sbi
, seg
, seg
->next_blkoff
+ 1);
2226 * This function always allocates a used segment(from dirty seglist) by SSR
2227 * manner, so it should recover the existing segment information of valid blocks
2229 static void change_curseg(struct f2fs_sb_info
*sbi
, int type
)
2231 struct dirty_seglist_info
*dirty_i
= DIRTY_I(sbi
);
2232 struct curseg_info
*curseg
= CURSEG_I(sbi
, type
);
2233 unsigned int new_segno
= curseg
->next_segno
;
2234 struct f2fs_summary_block
*sum_node
;
2235 struct page
*sum_page
;
2237 write_sum_page(sbi
, curseg
->sum_blk
,
2238 GET_SUM_BLOCK(sbi
, curseg
->segno
));
2239 __set_test_and_inuse(sbi
, new_segno
);
2241 mutex_lock(&dirty_i
->seglist_lock
);
2242 __remove_dirty_segment(sbi
, new_segno
, PRE
);
2243 __remove_dirty_segment(sbi
, new_segno
, DIRTY
);
2244 mutex_unlock(&dirty_i
->seglist_lock
);
2246 reset_curseg(sbi
, type
, 1);
2247 curseg
->alloc_type
= SSR
;
2248 __next_free_blkoff(sbi
, curseg
, 0);
2250 sum_page
= get_sum_page(sbi
, new_segno
);
2251 sum_node
= (struct f2fs_summary_block
*)page_address(sum_page
);
2252 memcpy(curseg
->sum_blk
, sum_node
, SUM_ENTRY_SIZE
);
2253 f2fs_put_page(sum_page
, 1);
2256 static int get_ssr_segment(struct f2fs_sb_info
*sbi
, int type
)
2258 struct curseg_info
*curseg
= CURSEG_I(sbi
, type
);
2259 const struct victim_selection
*v_ops
= DIRTY_I(sbi
)->v_ops
;
2260 unsigned segno
= NULL_SEGNO
;
2262 bool reversed
= false;
2264 /* need_SSR() already forces to do this */
2265 if (v_ops
->get_victim(sbi
, &segno
, BG_GC
, type
, SSR
)) {
2266 curseg
->next_segno
= segno
;
2270 /* For node segments, let's do SSR more intensively */
2271 if (IS_NODESEG(type
)) {
2272 if (type
>= CURSEG_WARM_NODE
) {
2274 i
= CURSEG_COLD_NODE
;
2276 i
= CURSEG_HOT_NODE
;
2278 cnt
= NR_CURSEG_NODE_TYPE
;
2280 if (type
>= CURSEG_WARM_DATA
) {
2282 i
= CURSEG_COLD_DATA
;
2284 i
= CURSEG_HOT_DATA
;
2286 cnt
= NR_CURSEG_DATA_TYPE
;
2289 for (; cnt
-- > 0; reversed
? i
-- : i
++) {
2292 if (v_ops
->get_victim(sbi
, &segno
, BG_GC
, i
, SSR
)) {
2293 curseg
->next_segno
= segno
;
2301 * flush out current segment and replace it with new segment
2302 * This function should be returned with success, otherwise BUG
2304 static void allocate_segment_by_default(struct f2fs_sb_info
*sbi
,
2305 int type
, bool force
)
2307 struct curseg_info
*curseg
= CURSEG_I(sbi
, type
);
2310 new_curseg(sbi
, type
, true);
2311 else if (!is_set_ckpt_flags(sbi
, CP_CRC_RECOVERY_FLAG
) &&
2312 type
== CURSEG_WARM_NODE
)
2313 new_curseg(sbi
, type
, false);
2314 else if (curseg
->alloc_type
== LFS
&& is_next_segment_free(sbi
, type
))
2315 new_curseg(sbi
, type
, false);
2316 else if (need_SSR(sbi
) && get_ssr_segment(sbi
, type
))
2317 change_curseg(sbi
, type
);
2319 new_curseg(sbi
, type
, false);
2321 stat_inc_seg_type(sbi
, curseg
);
2324 void allocate_new_segments(struct f2fs_sb_info
*sbi
)
2326 struct curseg_info
*curseg
;
2327 unsigned int old_segno
;
2330 down_write(&SIT_I(sbi
)->sentry_lock
);
2332 for (i
= CURSEG_HOT_DATA
; i
<= CURSEG_COLD_DATA
; i
++) {
2333 curseg
= CURSEG_I(sbi
, i
);
2334 old_segno
= curseg
->segno
;
2335 SIT_I(sbi
)->s_ops
->allocate_segment(sbi
, i
, true);
2336 locate_dirty_segment(sbi
, old_segno
);
2339 up_write(&SIT_I(sbi
)->sentry_lock
);
2342 static const struct segment_allocation default_salloc_ops
= {
2343 .allocate_segment
= allocate_segment_by_default
,
2346 bool exist_trim_candidates(struct f2fs_sb_info
*sbi
, struct cp_control
*cpc
)
2348 __u64 trim_start
= cpc
->trim_start
;
2349 bool has_candidate
= false;
2351 down_write(&SIT_I(sbi
)->sentry_lock
);
2352 for (; cpc
->trim_start
<= cpc
->trim_end
; cpc
->trim_start
++) {
2353 if (add_discard_addrs(sbi
, cpc
, true)) {
2354 has_candidate
= true;
2358 up_write(&SIT_I(sbi
)->sentry_lock
);
2360 cpc
->trim_start
= trim_start
;
2361 return has_candidate
;
2364 int f2fs_trim_fs(struct f2fs_sb_info
*sbi
, struct fstrim_range
*range
)
2366 __u64 start
= F2FS_BYTES_TO_BLK(range
->start
);
2367 __u64 end
= start
+ F2FS_BYTES_TO_BLK(range
->len
) - 1;
2368 unsigned int start_segno
, end_segno
, cur_segno
;
2369 block_t start_block
, end_block
;
2370 struct cp_control cpc
;
2371 struct discard_policy dpolicy
;
2372 unsigned long long trimmed
= 0;
2375 if (start
>= MAX_BLKADDR(sbi
) || range
->len
< sbi
->blocksize
)
2378 if (end
<= MAIN_BLKADDR(sbi
))
2381 if (is_sbi_flag_set(sbi
, SBI_NEED_FSCK
)) {
2382 f2fs_msg(sbi
->sb
, KERN_WARNING
,
2383 "Found FS corruption, run fsck to fix.");
2387 /* start/end segment number in main_area */
2388 start_segno
= (start
<= MAIN_BLKADDR(sbi
)) ? 0 : GET_SEGNO(sbi
, start
);
2389 end_segno
= (end
>= MAX_BLKADDR(sbi
)) ? MAIN_SEGS(sbi
) - 1 :
2390 GET_SEGNO(sbi
, end
);
2392 cpc
.reason
= CP_DISCARD
;
2393 cpc
.trim_minlen
= max_t(__u64
, 1, F2FS_BYTES_TO_BLK(range
->minlen
));
2395 /* do checkpoint to issue discard commands safely */
2396 for (cur_segno
= start_segno
; cur_segno
<= end_segno
;
2397 cur_segno
= cpc
.trim_end
+ 1) {
2398 cpc
.trim_start
= cur_segno
;
2400 if (sbi
->discard_blks
== 0)
2402 else if (sbi
->discard_blks
< BATCHED_TRIM_BLOCKS(sbi
))
2403 cpc
.trim_end
= end_segno
;
2405 cpc
.trim_end
= min_t(unsigned int,
2406 rounddown(cur_segno
+
2407 BATCHED_TRIM_SEGMENTS(sbi
),
2408 sbi
->segs_per_sec
) - 1, end_segno
);
2410 mutex_lock(&sbi
->gc_mutex
);
2411 err
= write_checkpoint(sbi
, &cpc
);
2412 mutex_unlock(&sbi
->gc_mutex
);
2419 start_block
= START_BLOCK(sbi
, start_segno
);
2420 end_block
= START_BLOCK(sbi
, min(cur_segno
, end_segno
) + 1);
2422 init_discard_policy(&dpolicy
, DPOLICY_FSTRIM
, cpc
.trim_minlen
);
2423 __issue_discard_cmd_range(sbi
, &dpolicy
, start_block
, end_block
);
2424 trimmed
= __wait_discard_cmd_range(sbi
, &dpolicy
,
2425 start_block
, end_block
);
2427 range
->len
= F2FS_BLK_TO_BYTES(trimmed
);
2431 static bool __has_curseg_space(struct f2fs_sb_info
*sbi
, int type
)
2433 struct curseg_info
*curseg
= CURSEG_I(sbi
, type
);
2434 if (curseg
->next_blkoff
< sbi
->blocks_per_seg
)
2439 int rw_hint_to_seg_type(enum rw_hint hint
)
2442 case WRITE_LIFE_SHORT
:
2443 return CURSEG_HOT_DATA
;
2444 case WRITE_LIFE_EXTREME
:
2445 return CURSEG_COLD_DATA
;
2447 return CURSEG_WARM_DATA
;
2451 static int __get_segment_type_2(struct f2fs_io_info
*fio
)
2453 if (fio
->type
== DATA
)
2454 return CURSEG_HOT_DATA
;
2456 return CURSEG_HOT_NODE
;
2459 static int __get_segment_type_4(struct f2fs_io_info
*fio
)
2461 if (fio
->type
== DATA
) {
2462 struct inode
*inode
= fio
->page
->mapping
->host
;
2464 if (S_ISDIR(inode
->i_mode
))
2465 return CURSEG_HOT_DATA
;
2467 return CURSEG_COLD_DATA
;
2469 if (IS_DNODE(fio
->page
) && is_cold_node(fio
->page
))
2470 return CURSEG_WARM_NODE
;
2472 return CURSEG_COLD_NODE
;
2476 static int __get_segment_type_6(struct f2fs_io_info
*fio
)
2478 if (fio
->type
== DATA
) {
2479 struct inode
*inode
= fio
->page
->mapping
->host
;
2481 if (is_cold_data(fio
->page
) || file_is_cold(inode
))
2482 return CURSEG_COLD_DATA
;
2483 if (is_inode_flag_set(inode
, FI_HOT_DATA
))
2484 return CURSEG_HOT_DATA
;
2485 return rw_hint_to_seg_type(inode
->i_write_hint
);
2487 if (IS_DNODE(fio
->page
))
2488 return is_cold_node(fio
->page
) ? CURSEG_WARM_NODE
:
2490 return CURSEG_COLD_NODE
;
2494 static int __get_segment_type(struct f2fs_io_info
*fio
)
2498 switch (fio
->sbi
->active_logs
) {
2500 type
= __get_segment_type_2(fio
);
2503 type
= __get_segment_type_4(fio
);
2506 type
= __get_segment_type_6(fio
);
2509 f2fs_bug_on(fio
->sbi
, true);
2514 else if (IS_WARM(type
))
2521 void allocate_data_block(struct f2fs_sb_info
*sbi
, struct page
*page
,
2522 block_t old_blkaddr
, block_t
*new_blkaddr
,
2523 struct f2fs_summary
*sum
, int type
,
2524 struct f2fs_io_info
*fio
, bool add_list
)
2526 struct sit_info
*sit_i
= SIT_I(sbi
);
2527 struct curseg_info
*curseg
= CURSEG_I(sbi
, type
);
2529 down_read(&SM_I(sbi
)->curseg_lock
);
2531 mutex_lock(&curseg
->curseg_mutex
);
2532 down_write(&sit_i
->sentry_lock
);
2534 *new_blkaddr
= NEXT_FREE_BLKADDR(sbi
, curseg
);
2536 f2fs_wait_discard_bio(sbi
, *new_blkaddr
);
2539 * __add_sum_entry should be resided under the curseg_mutex
2540 * because, this function updates a summary entry in the
2541 * current summary block.
2543 __add_sum_entry(sbi
, type
, sum
);
2545 __refresh_next_blkoff(sbi
, curseg
);
2547 stat_inc_block_count(sbi
, curseg
);
2550 * SIT information should be updated before segment allocation,
2551 * since SSR needs latest valid block information.
2553 update_sit_entry(sbi
, *new_blkaddr
, 1);
2554 if (GET_SEGNO(sbi
, old_blkaddr
) != NULL_SEGNO
)
2555 update_sit_entry(sbi
, old_blkaddr
, -1);
2557 if (!__has_curseg_space(sbi
, type
))
2558 sit_i
->s_ops
->allocate_segment(sbi
, type
, false);
2561 * segment dirty status should be updated after segment allocation,
2562 * so we just need to update status only one time after previous
2563 * segment being closed.
2565 locate_dirty_segment(sbi
, GET_SEGNO(sbi
, old_blkaddr
));
2566 locate_dirty_segment(sbi
, GET_SEGNO(sbi
, *new_blkaddr
));
2568 up_write(&sit_i
->sentry_lock
);
2570 if (page
&& IS_NODESEG(type
)) {
2571 fill_node_footer_blkaddr(page
, NEXT_FREE_BLKADDR(sbi
, curseg
));
2573 f2fs_inode_chksum_set(sbi
, page
);
2577 struct f2fs_bio_info
*io
;
2579 INIT_LIST_HEAD(&fio
->list
);
2580 fio
->in_list
= true;
2581 io
= sbi
->write_io
[fio
->type
] + fio
->temp
;
2582 spin_lock(&io
->io_lock
);
2583 list_add_tail(&fio
->list
, &io
->io_list
);
2584 spin_unlock(&io
->io_lock
);
2587 mutex_unlock(&curseg
->curseg_mutex
);
2589 up_read(&SM_I(sbi
)->curseg_lock
);
2592 static void update_device_state(struct f2fs_io_info
*fio
)
2594 struct f2fs_sb_info
*sbi
= fio
->sbi
;
2595 unsigned int devidx
;
2600 devidx
= f2fs_target_device_index(sbi
, fio
->new_blkaddr
);
2602 /* update device state for fsync */
2603 set_dirty_device(sbi
, fio
->ino
, devidx
, FLUSH_INO
);
2605 /* update device state for checkpoint */
2606 if (!f2fs_test_bit(devidx
, (char *)&sbi
->dirty_device
)) {
2607 spin_lock(&sbi
->dev_lock
);
2608 f2fs_set_bit(devidx
, (char *)&sbi
->dirty_device
);
2609 spin_unlock(&sbi
->dev_lock
);
2613 static void do_write_page(struct f2fs_summary
*sum
, struct f2fs_io_info
*fio
)
2615 int type
= __get_segment_type(fio
);
2619 allocate_data_block(fio
->sbi
, fio
->page
, fio
->old_blkaddr
,
2620 &fio
->new_blkaddr
, sum
, type
, fio
, true);
2622 /* writeout dirty page into bdev */
2623 err
= f2fs_submit_page_write(fio
);
2624 if (err
== -EAGAIN
) {
2625 fio
->old_blkaddr
= fio
->new_blkaddr
;
2628 update_device_state(fio
);
2632 void write_meta_page(struct f2fs_sb_info
*sbi
, struct page
*page
,
2633 enum iostat_type io_type
)
2635 struct f2fs_io_info fio
= {
2639 .op_flags
= REQ_SYNC
| REQ_META
| REQ_PRIO
,
2640 .old_blkaddr
= page
->index
,
2641 .new_blkaddr
= page
->index
,
2643 .encrypted_page
= NULL
,
2647 if (unlikely(page
->index
>= MAIN_BLKADDR(sbi
)))
2648 fio
.op_flags
&= ~REQ_META
;
2650 set_page_writeback(page
);
2651 f2fs_submit_page_write(&fio
);
2653 f2fs_update_iostat(sbi
, io_type
, F2FS_BLKSIZE
);
2656 void write_node_page(unsigned int nid
, struct f2fs_io_info
*fio
)
2658 struct f2fs_summary sum
;
2660 set_summary(&sum
, nid
, 0, 0);
2661 do_write_page(&sum
, fio
);
2663 f2fs_update_iostat(fio
->sbi
, fio
->io_type
, F2FS_BLKSIZE
);
2666 void write_data_page(struct dnode_of_data
*dn
, struct f2fs_io_info
*fio
)
2668 struct f2fs_sb_info
*sbi
= fio
->sbi
;
2669 struct f2fs_summary sum
;
2670 struct node_info ni
;
2672 f2fs_bug_on(sbi
, dn
->data_blkaddr
== NULL_ADDR
);
2673 get_node_info(sbi
, dn
->nid
, &ni
);
2674 set_summary(&sum
, dn
->nid
, dn
->ofs_in_node
, ni
.version
);
2675 do_write_page(&sum
, fio
);
2676 f2fs_update_data_blkaddr(dn
, fio
->new_blkaddr
);
2678 f2fs_update_iostat(sbi
, fio
->io_type
, F2FS_BLKSIZE
);
2681 int rewrite_data_page(struct f2fs_io_info
*fio
)
2685 fio
->new_blkaddr
= fio
->old_blkaddr
;
2686 stat_inc_inplace_blocks(fio
->sbi
);
2688 err
= f2fs_submit_page_bio(fio
);
2690 update_device_state(fio
);
2692 f2fs_update_iostat(fio
->sbi
, fio
->io_type
, F2FS_BLKSIZE
);
2697 static inline int __f2fs_get_curseg(struct f2fs_sb_info
*sbi
,
2702 for (i
= CURSEG_HOT_DATA
; i
< NO_CHECK_TYPE
; i
++) {
2703 if (CURSEG_I(sbi
, i
)->segno
== segno
)
2709 void __f2fs_replace_block(struct f2fs_sb_info
*sbi
, struct f2fs_summary
*sum
,
2710 block_t old_blkaddr
, block_t new_blkaddr
,
2711 bool recover_curseg
, bool recover_newaddr
)
2713 struct sit_info
*sit_i
= SIT_I(sbi
);
2714 struct curseg_info
*curseg
;
2715 unsigned int segno
, old_cursegno
;
2716 struct seg_entry
*se
;
2718 unsigned short old_blkoff
;
2720 segno
= GET_SEGNO(sbi
, new_blkaddr
);
2721 se
= get_seg_entry(sbi
, segno
);
2724 down_write(&SM_I(sbi
)->curseg_lock
);
2726 if (!recover_curseg
) {
2727 /* for recovery flow */
2728 if (se
->valid_blocks
== 0 && !IS_CURSEG(sbi
, segno
)) {
2729 if (old_blkaddr
== NULL_ADDR
)
2730 type
= CURSEG_COLD_DATA
;
2732 type
= CURSEG_WARM_DATA
;
2735 if (IS_CURSEG(sbi
, segno
)) {
2736 /* se->type is volatile as SSR allocation */
2737 type
= __f2fs_get_curseg(sbi
, segno
);
2738 f2fs_bug_on(sbi
, type
== NO_CHECK_TYPE
);
2740 type
= CURSEG_WARM_DATA
;
2744 curseg
= CURSEG_I(sbi
, type
);
2746 mutex_lock(&curseg
->curseg_mutex
);
2747 down_write(&sit_i
->sentry_lock
);
2749 old_cursegno
= curseg
->segno
;
2750 old_blkoff
= curseg
->next_blkoff
;
2752 /* change the current segment */
2753 if (segno
!= curseg
->segno
) {
2754 curseg
->next_segno
= segno
;
2755 change_curseg(sbi
, type
);
2758 curseg
->next_blkoff
= GET_BLKOFF_FROM_SEG0(sbi
, new_blkaddr
);
2759 __add_sum_entry(sbi
, type
, sum
);
2761 if (!recover_curseg
|| recover_newaddr
)
2762 update_sit_entry(sbi
, new_blkaddr
, 1);
2763 if (GET_SEGNO(sbi
, old_blkaddr
) != NULL_SEGNO
)
2764 update_sit_entry(sbi
, old_blkaddr
, -1);
2766 locate_dirty_segment(sbi
, GET_SEGNO(sbi
, old_blkaddr
));
2767 locate_dirty_segment(sbi
, GET_SEGNO(sbi
, new_blkaddr
));
2769 locate_dirty_segment(sbi
, old_cursegno
);
2771 if (recover_curseg
) {
2772 if (old_cursegno
!= curseg
->segno
) {
2773 curseg
->next_segno
= old_cursegno
;
2774 change_curseg(sbi
, type
);
2776 curseg
->next_blkoff
= old_blkoff
;
2779 up_write(&sit_i
->sentry_lock
);
2780 mutex_unlock(&curseg
->curseg_mutex
);
2781 up_write(&SM_I(sbi
)->curseg_lock
);
2784 void f2fs_replace_block(struct f2fs_sb_info
*sbi
, struct dnode_of_data
*dn
,
2785 block_t old_addr
, block_t new_addr
,
2786 unsigned char version
, bool recover_curseg
,
2787 bool recover_newaddr
)
2789 struct f2fs_summary sum
;
2791 set_summary(&sum
, dn
->nid
, dn
->ofs_in_node
, version
);
2793 __f2fs_replace_block(sbi
, &sum
, old_addr
, new_addr
,
2794 recover_curseg
, recover_newaddr
);
2796 f2fs_update_data_blkaddr(dn
, new_addr
);
2799 void f2fs_wait_on_page_writeback(struct page
*page
,
2800 enum page_type type
, bool ordered
)
2802 if (PageWriteback(page
)) {
2803 struct f2fs_sb_info
*sbi
= F2FS_P_SB(page
);
2805 f2fs_submit_merged_write_cond(sbi
, page
->mapping
->host
,
2806 0, page
->index
, type
);
2808 wait_on_page_writeback(page
);
2810 wait_for_stable_page(page
);
2814 void f2fs_wait_on_block_writeback(struct f2fs_sb_info
*sbi
, block_t blkaddr
)
2818 if (blkaddr
== NEW_ADDR
|| blkaddr
== NULL_ADDR
)
2821 cpage
= find_lock_page(META_MAPPING(sbi
), blkaddr
);
2823 f2fs_wait_on_page_writeback(cpage
, DATA
, true);
2824 f2fs_put_page(cpage
, 1);
2828 static int read_compacted_summaries(struct f2fs_sb_info
*sbi
)
2830 struct f2fs_checkpoint
*ckpt
= F2FS_CKPT(sbi
);
2831 struct curseg_info
*seg_i
;
2832 unsigned char *kaddr
;
2837 start
= start_sum_block(sbi
);
2839 page
= get_meta_page(sbi
, start
++);
2840 kaddr
= (unsigned char *)page_address(page
);
2842 /* Step 1: restore nat cache */
2843 seg_i
= CURSEG_I(sbi
, CURSEG_HOT_DATA
);
2844 memcpy(seg_i
->journal
, kaddr
, SUM_JOURNAL_SIZE
);
2846 /* Step 2: restore sit cache */
2847 seg_i
= CURSEG_I(sbi
, CURSEG_COLD_DATA
);
2848 memcpy(seg_i
->journal
, kaddr
+ SUM_JOURNAL_SIZE
, SUM_JOURNAL_SIZE
);
2849 offset
= 2 * SUM_JOURNAL_SIZE
;
2851 /* Step 3: restore summary entries */
2852 for (i
= CURSEG_HOT_DATA
; i
<= CURSEG_COLD_DATA
; i
++) {
2853 unsigned short blk_off
;
2856 seg_i
= CURSEG_I(sbi
, i
);
2857 segno
= le32_to_cpu(ckpt
->cur_data_segno
[i
]);
2858 blk_off
= le16_to_cpu(ckpt
->cur_data_blkoff
[i
]);
2859 seg_i
->next_segno
= segno
;
2860 reset_curseg(sbi
, i
, 0);
2861 seg_i
->alloc_type
= ckpt
->alloc_type
[i
];
2862 seg_i
->next_blkoff
= blk_off
;
2864 if (seg_i
->alloc_type
== SSR
)
2865 blk_off
= sbi
->blocks_per_seg
;
2867 for (j
= 0; j
< blk_off
; j
++) {
2868 struct f2fs_summary
*s
;
2869 s
= (struct f2fs_summary
*)(kaddr
+ offset
);
2870 seg_i
->sum_blk
->entries
[j
] = *s
;
2871 offset
+= SUMMARY_SIZE
;
2872 if (offset
+ SUMMARY_SIZE
<= PAGE_SIZE
-
2876 f2fs_put_page(page
, 1);
2879 page
= get_meta_page(sbi
, start
++);
2880 kaddr
= (unsigned char *)page_address(page
);
2884 f2fs_put_page(page
, 1);
2888 static int read_normal_summaries(struct f2fs_sb_info
*sbi
, int type
)
2890 struct f2fs_checkpoint
*ckpt
= F2FS_CKPT(sbi
);
2891 struct f2fs_summary_block
*sum
;
2892 struct curseg_info
*curseg
;
2894 unsigned short blk_off
;
2895 unsigned int segno
= 0;
2896 block_t blk_addr
= 0;
2898 /* get segment number and block addr */
2899 if (IS_DATASEG(type
)) {
2900 segno
= le32_to_cpu(ckpt
->cur_data_segno
[type
]);
2901 blk_off
= le16_to_cpu(ckpt
->cur_data_blkoff
[type
-
2903 if (__exist_node_summaries(sbi
))
2904 blk_addr
= sum_blk_addr(sbi
, NR_CURSEG_TYPE
, type
);
2906 blk_addr
= sum_blk_addr(sbi
, NR_CURSEG_DATA_TYPE
, type
);
2908 segno
= le32_to_cpu(ckpt
->cur_node_segno
[type
-
2910 blk_off
= le16_to_cpu(ckpt
->cur_node_blkoff
[type
-
2912 if (__exist_node_summaries(sbi
))
2913 blk_addr
= sum_blk_addr(sbi
, NR_CURSEG_NODE_TYPE
,
2914 type
- CURSEG_HOT_NODE
);
2916 blk_addr
= GET_SUM_BLOCK(sbi
, segno
);
2919 new = get_meta_page(sbi
, blk_addr
);
2920 sum
= (struct f2fs_summary_block
*)page_address(new);
2922 if (IS_NODESEG(type
)) {
2923 if (__exist_node_summaries(sbi
)) {
2924 struct f2fs_summary
*ns
= &sum
->entries
[0];
2926 for (i
= 0; i
< sbi
->blocks_per_seg
; i
++, ns
++) {
2928 ns
->ofs_in_node
= 0;
2933 err
= restore_node_summary(sbi
, segno
, sum
);
2935 f2fs_put_page(new, 1);
2941 /* set uncompleted segment to curseg */
2942 curseg
= CURSEG_I(sbi
, type
);
2943 mutex_lock(&curseg
->curseg_mutex
);
2945 /* update journal info */
2946 down_write(&curseg
->journal_rwsem
);
2947 memcpy(curseg
->journal
, &sum
->journal
, SUM_JOURNAL_SIZE
);
2948 up_write(&curseg
->journal_rwsem
);
2950 memcpy(curseg
->sum_blk
->entries
, sum
->entries
, SUM_ENTRY_SIZE
);
2951 memcpy(&curseg
->sum_blk
->footer
, &sum
->footer
, SUM_FOOTER_SIZE
);
2952 curseg
->next_segno
= segno
;
2953 reset_curseg(sbi
, type
, 0);
2954 curseg
->alloc_type
= ckpt
->alloc_type
[type
];
2955 curseg
->next_blkoff
= blk_off
;
2956 mutex_unlock(&curseg
->curseg_mutex
);
2957 f2fs_put_page(new, 1);
2961 static int restore_curseg_summaries(struct f2fs_sb_info
*sbi
)
2963 struct f2fs_journal
*sit_j
= CURSEG_I(sbi
, CURSEG_COLD_DATA
)->journal
;
2964 struct f2fs_journal
*nat_j
= CURSEG_I(sbi
, CURSEG_HOT_DATA
)->journal
;
2965 int type
= CURSEG_HOT_DATA
;
2968 if (is_set_ckpt_flags(sbi
, CP_COMPACT_SUM_FLAG
)) {
2969 int npages
= npages_for_summary_flush(sbi
, true);
2972 ra_meta_pages(sbi
, start_sum_block(sbi
), npages
,
2975 /* restore for compacted data summary */
2976 if (read_compacted_summaries(sbi
))
2978 type
= CURSEG_HOT_NODE
;
2981 if (__exist_node_summaries(sbi
))
2982 ra_meta_pages(sbi
, sum_blk_addr(sbi
, NR_CURSEG_TYPE
, type
),
2983 NR_CURSEG_TYPE
- type
, META_CP
, true);
2985 for (; type
<= CURSEG_COLD_NODE
; type
++) {
2986 err
= read_normal_summaries(sbi
, type
);
2991 /* sanity check for summary blocks */
2992 if (nats_in_cursum(nat_j
) > NAT_JOURNAL_ENTRIES
||
2993 sits_in_cursum(sit_j
) > SIT_JOURNAL_ENTRIES
)
2999 static void write_compacted_summaries(struct f2fs_sb_info
*sbi
, block_t blkaddr
)
3002 unsigned char *kaddr
;
3003 struct f2fs_summary
*summary
;
3004 struct curseg_info
*seg_i
;
3005 int written_size
= 0;
3008 page
= grab_meta_page(sbi
, blkaddr
++);
3009 kaddr
= (unsigned char *)page_address(page
);
3010 memset(kaddr
, 0, PAGE_SIZE
);
3012 /* Step 1: write nat cache */
3013 seg_i
= CURSEG_I(sbi
, CURSEG_HOT_DATA
);
3014 memcpy(kaddr
, seg_i
->journal
, SUM_JOURNAL_SIZE
);
3015 written_size
+= SUM_JOURNAL_SIZE
;
3017 /* Step 2: write sit cache */
3018 seg_i
= CURSEG_I(sbi
, CURSEG_COLD_DATA
);
3019 memcpy(kaddr
+ written_size
, seg_i
->journal
, SUM_JOURNAL_SIZE
);
3020 written_size
+= SUM_JOURNAL_SIZE
;
3022 /* Step 3: write summary entries */
3023 for (i
= CURSEG_HOT_DATA
; i
<= CURSEG_COLD_DATA
; i
++) {
3024 unsigned short blkoff
;
3025 seg_i
= CURSEG_I(sbi
, i
);
3026 if (sbi
->ckpt
->alloc_type
[i
] == SSR
)
3027 blkoff
= sbi
->blocks_per_seg
;
3029 blkoff
= curseg_blkoff(sbi
, i
);
3031 for (j
= 0; j
< blkoff
; j
++) {
3033 page
= grab_meta_page(sbi
, blkaddr
++);
3034 kaddr
= (unsigned char *)page_address(page
);
3035 memset(kaddr
, 0, PAGE_SIZE
);
3038 summary
= (struct f2fs_summary
*)(kaddr
+ written_size
);
3039 *summary
= seg_i
->sum_blk
->entries
[j
];
3040 written_size
+= SUMMARY_SIZE
;
3042 if (written_size
+ SUMMARY_SIZE
<= PAGE_SIZE
-
3046 set_page_dirty(page
);
3047 f2fs_put_page(page
, 1);
3052 set_page_dirty(page
);
3053 f2fs_put_page(page
, 1);
3057 static void write_normal_summaries(struct f2fs_sb_info
*sbi
,
3058 block_t blkaddr
, int type
)
3061 if (IS_DATASEG(type
))
3062 end
= type
+ NR_CURSEG_DATA_TYPE
;
3064 end
= type
+ NR_CURSEG_NODE_TYPE
;
3066 for (i
= type
; i
< end
; i
++)
3067 write_current_sum_page(sbi
, i
, blkaddr
+ (i
- type
));
3070 void write_data_summaries(struct f2fs_sb_info
*sbi
, block_t start_blk
)
3072 if (is_set_ckpt_flags(sbi
, CP_COMPACT_SUM_FLAG
))
3073 write_compacted_summaries(sbi
, start_blk
);
3075 write_normal_summaries(sbi
, start_blk
, CURSEG_HOT_DATA
);
3078 void write_node_summaries(struct f2fs_sb_info
*sbi
, block_t start_blk
)
3080 write_normal_summaries(sbi
, start_blk
, CURSEG_HOT_NODE
);
3083 int lookup_journal_in_cursum(struct f2fs_journal
*journal
, int type
,
3084 unsigned int val
, int alloc
)
3088 if (type
== NAT_JOURNAL
) {
3089 for (i
= 0; i
< nats_in_cursum(journal
); i
++) {
3090 if (le32_to_cpu(nid_in_journal(journal
, i
)) == val
)
3093 if (alloc
&& __has_cursum_space(journal
, 1, NAT_JOURNAL
))
3094 return update_nats_in_cursum(journal
, 1);
3095 } else if (type
== SIT_JOURNAL
) {
3096 for (i
= 0; i
< sits_in_cursum(journal
); i
++)
3097 if (le32_to_cpu(segno_in_journal(journal
, i
)) == val
)
3099 if (alloc
&& __has_cursum_space(journal
, 1, SIT_JOURNAL
))
3100 return update_sits_in_cursum(journal
, 1);
3105 static struct page
*get_current_sit_page(struct f2fs_sb_info
*sbi
,
3108 return get_meta_page(sbi
, current_sit_addr(sbi
, segno
));
3111 static struct page
*get_next_sit_page(struct f2fs_sb_info
*sbi
,
3114 struct sit_info
*sit_i
= SIT_I(sbi
);
3115 struct page
*src_page
, *dst_page
;
3116 pgoff_t src_off
, dst_off
;
3117 void *src_addr
, *dst_addr
;
3119 src_off
= current_sit_addr(sbi
, start
);
3120 dst_off
= next_sit_addr(sbi
, src_off
);
3122 /* get current sit block page without lock */
3123 src_page
= get_meta_page(sbi
, src_off
);
3124 dst_page
= grab_meta_page(sbi
, dst_off
);
3125 f2fs_bug_on(sbi
, PageDirty(src_page
));
3127 src_addr
= page_address(src_page
);
3128 dst_addr
= page_address(dst_page
);
3129 memcpy(dst_addr
, src_addr
, PAGE_SIZE
);
3131 set_page_dirty(dst_page
);
3132 f2fs_put_page(src_page
, 1);
3134 set_to_next_sit(sit_i
, start
);
3139 static struct sit_entry_set
*grab_sit_entry_set(void)
3141 struct sit_entry_set
*ses
=
3142 f2fs_kmem_cache_alloc(sit_entry_set_slab
, GFP_NOFS
);
3145 INIT_LIST_HEAD(&ses
->set_list
);
3149 static void release_sit_entry_set(struct sit_entry_set
*ses
)
3151 list_del(&ses
->set_list
);
3152 kmem_cache_free(sit_entry_set_slab
, ses
);
3155 static void adjust_sit_entry_set(struct sit_entry_set
*ses
,
3156 struct list_head
*head
)
3158 struct sit_entry_set
*next
= ses
;
3160 if (list_is_last(&ses
->set_list
, head
))
3163 list_for_each_entry_continue(next
, head
, set_list
)
3164 if (ses
->entry_cnt
<= next
->entry_cnt
)
3167 list_move_tail(&ses
->set_list
, &next
->set_list
);
3170 static void add_sit_entry(unsigned int segno
, struct list_head
*head
)
3172 struct sit_entry_set
*ses
;
3173 unsigned int start_segno
= START_SEGNO(segno
);
3175 list_for_each_entry(ses
, head
, set_list
) {
3176 if (ses
->start_segno
== start_segno
) {
3178 adjust_sit_entry_set(ses
, head
);
3183 ses
= grab_sit_entry_set();
3185 ses
->start_segno
= start_segno
;
3187 list_add(&ses
->set_list
, head
);
3190 static void add_sits_in_set(struct f2fs_sb_info
*sbi
)
3192 struct f2fs_sm_info
*sm_info
= SM_I(sbi
);
3193 struct list_head
*set_list
= &sm_info
->sit_entry_set
;
3194 unsigned long *bitmap
= SIT_I(sbi
)->dirty_sentries_bitmap
;
3197 for_each_set_bit(segno
, bitmap
, MAIN_SEGS(sbi
))
3198 add_sit_entry(segno
, set_list
);
3201 static void remove_sits_in_journal(struct f2fs_sb_info
*sbi
)
3203 struct curseg_info
*curseg
= CURSEG_I(sbi
, CURSEG_COLD_DATA
);
3204 struct f2fs_journal
*journal
= curseg
->journal
;
3207 down_write(&curseg
->journal_rwsem
);
3208 for (i
= 0; i
< sits_in_cursum(journal
); i
++) {
3212 segno
= le32_to_cpu(segno_in_journal(journal
, i
));
3213 dirtied
= __mark_sit_entry_dirty(sbi
, segno
);
3216 add_sit_entry(segno
, &SM_I(sbi
)->sit_entry_set
);
3218 update_sits_in_cursum(journal
, -i
);
3219 up_write(&curseg
->journal_rwsem
);
3223 * CP calls this function, which flushes SIT entries including sit_journal,
3224 * and moves prefree segs to free segs.
3226 void flush_sit_entries(struct f2fs_sb_info
*sbi
, struct cp_control
*cpc
)
3228 struct sit_info
*sit_i
= SIT_I(sbi
);
3229 unsigned long *bitmap
= sit_i
->dirty_sentries_bitmap
;
3230 struct curseg_info
*curseg
= CURSEG_I(sbi
, CURSEG_COLD_DATA
);
3231 struct f2fs_journal
*journal
= curseg
->journal
;
3232 struct sit_entry_set
*ses
, *tmp
;
3233 struct list_head
*head
= &SM_I(sbi
)->sit_entry_set
;
3234 bool to_journal
= true;
3235 struct seg_entry
*se
;
3237 down_write(&sit_i
->sentry_lock
);
3239 if (!sit_i
->dirty_sentries
)
3243 * add and account sit entries of dirty bitmap in sit entry
3246 add_sits_in_set(sbi
);
3249 * if there are no enough space in journal to store dirty sit
3250 * entries, remove all entries from journal and add and account
3251 * them in sit entry set.
3253 if (!__has_cursum_space(journal
, sit_i
->dirty_sentries
, SIT_JOURNAL
))
3254 remove_sits_in_journal(sbi
);
3257 * there are two steps to flush sit entries:
3258 * #1, flush sit entries to journal in current cold data summary block.
3259 * #2, flush sit entries to sit page.
3261 list_for_each_entry_safe(ses
, tmp
, head
, set_list
) {
3262 struct page
*page
= NULL
;
3263 struct f2fs_sit_block
*raw_sit
= NULL
;
3264 unsigned int start_segno
= ses
->start_segno
;
3265 unsigned int end
= min(start_segno
+ SIT_ENTRY_PER_BLOCK
,
3266 (unsigned long)MAIN_SEGS(sbi
));
3267 unsigned int segno
= start_segno
;
3270 !__has_cursum_space(journal
, ses
->entry_cnt
, SIT_JOURNAL
))
3274 down_write(&curseg
->journal_rwsem
);
3276 page
= get_next_sit_page(sbi
, start_segno
);
3277 raw_sit
= page_address(page
);
3280 /* flush dirty sit entries in region of current sit set */
3281 for_each_set_bit_from(segno
, bitmap
, end
) {
3282 int offset
, sit_offset
;
3284 se
= get_seg_entry(sbi
, segno
);
3286 /* add discard candidates */
3287 if (!(cpc
->reason
& CP_DISCARD
)) {
3288 cpc
->trim_start
= segno
;
3289 add_discard_addrs(sbi
, cpc
, false);
3293 offset
= lookup_journal_in_cursum(journal
,
3294 SIT_JOURNAL
, segno
, 1);
3295 f2fs_bug_on(sbi
, offset
< 0);
3296 segno_in_journal(journal
, offset
) =
3298 seg_info_to_raw_sit(se
,
3299 &sit_in_journal(journal
, offset
));
3301 sit_offset
= SIT_ENTRY_OFFSET(sit_i
, segno
);
3302 seg_info_to_raw_sit(se
,
3303 &raw_sit
->entries
[sit_offset
]);
3306 __clear_bit(segno
, bitmap
);
3307 sit_i
->dirty_sentries
--;
3312 up_write(&curseg
->journal_rwsem
);
3314 f2fs_put_page(page
, 1);
3316 f2fs_bug_on(sbi
, ses
->entry_cnt
);
3317 release_sit_entry_set(ses
);
3320 f2fs_bug_on(sbi
, !list_empty(head
));
3321 f2fs_bug_on(sbi
, sit_i
->dirty_sentries
);
3323 if (cpc
->reason
& CP_DISCARD
) {
3324 __u64 trim_start
= cpc
->trim_start
;
3326 for (; cpc
->trim_start
<= cpc
->trim_end
; cpc
->trim_start
++)
3327 add_discard_addrs(sbi
, cpc
, false);
3329 cpc
->trim_start
= trim_start
;
3331 up_write(&sit_i
->sentry_lock
);
3333 set_prefree_as_free_segments(sbi
);
3336 static int build_sit_info(struct f2fs_sb_info
*sbi
)
3338 struct f2fs_super_block
*raw_super
= F2FS_RAW_SUPER(sbi
);
3339 struct sit_info
*sit_i
;
3340 unsigned int sit_segs
, start
;
3342 unsigned int bitmap_size
;
3344 /* allocate memory for SIT information */
3345 sit_i
= kzalloc(sizeof(struct sit_info
), GFP_KERNEL
);
3349 SM_I(sbi
)->sit_info
= sit_i
;
3351 sit_i
->sentries
= kvzalloc(MAIN_SEGS(sbi
) *
3352 sizeof(struct seg_entry
), GFP_KERNEL
);
3353 if (!sit_i
->sentries
)
3356 bitmap_size
= f2fs_bitmap_size(MAIN_SEGS(sbi
));
3357 sit_i
->dirty_sentries_bitmap
= kvzalloc(bitmap_size
, GFP_KERNEL
);
3358 if (!sit_i
->dirty_sentries_bitmap
)
3361 for (start
= 0; start
< MAIN_SEGS(sbi
); start
++) {
3362 sit_i
->sentries
[start
].cur_valid_map
3363 = kzalloc(SIT_VBLOCK_MAP_SIZE
, GFP_KERNEL
);
3364 sit_i
->sentries
[start
].ckpt_valid_map
3365 = kzalloc(SIT_VBLOCK_MAP_SIZE
, GFP_KERNEL
);
3366 if (!sit_i
->sentries
[start
].cur_valid_map
||
3367 !sit_i
->sentries
[start
].ckpt_valid_map
)
3370 #ifdef CONFIG_F2FS_CHECK_FS
3371 sit_i
->sentries
[start
].cur_valid_map_mir
3372 = kzalloc(SIT_VBLOCK_MAP_SIZE
, GFP_KERNEL
);
3373 if (!sit_i
->sentries
[start
].cur_valid_map_mir
)
3377 if (f2fs_discard_en(sbi
)) {
3378 sit_i
->sentries
[start
].discard_map
3379 = kzalloc(SIT_VBLOCK_MAP_SIZE
, GFP_KERNEL
);
3380 if (!sit_i
->sentries
[start
].discard_map
)
3385 sit_i
->tmp_map
= kzalloc(SIT_VBLOCK_MAP_SIZE
, GFP_KERNEL
);
3386 if (!sit_i
->tmp_map
)
3389 if (sbi
->segs_per_sec
> 1) {
3390 sit_i
->sec_entries
= kvzalloc(MAIN_SECS(sbi
) *
3391 sizeof(struct sec_entry
), GFP_KERNEL
);
3392 if (!sit_i
->sec_entries
)
3396 /* get information related with SIT */
3397 sit_segs
= le32_to_cpu(raw_super
->segment_count_sit
) >> 1;
3399 /* setup SIT bitmap from ckeckpoint pack */
3400 bitmap_size
= __bitmap_size(sbi
, SIT_BITMAP
);
3401 src_bitmap
= __bitmap_ptr(sbi
, SIT_BITMAP
);
3403 sit_i
->sit_bitmap
= kmemdup(src_bitmap
, bitmap_size
, GFP_KERNEL
);
3404 if (!sit_i
->sit_bitmap
)
3407 #ifdef CONFIG_F2FS_CHECK_FS
3408 sit_i
->sit_bitmap_mir
= kmemdup(src_bitmap
, bitmap_size
, GFP_KERNEL
);
3409 if (!sit_i
->sit_bitmap_mir
)
3413 /* init SIT information */
3414 sit_i
->s_ops
= &default_salloc_ops
;
3416 sit_i
->sit_base_addr
= le32_to_cpu(raw_super
->sit_blkaddr
);
3417 sit_i
->sit_blocks
= sit_segs
<< sbi
->log_blocks_per_seg
;
3418 sit_i
->written_valid_blocks
= 0;
3419 sit_i
->bitmap_size
= bitmap_size
;
3420 sit_i
->dirty_sentries
= 0;
3421 sit_i
->sents_per_block
= SIT_ENTRY_PER_BLOCK
;
3422 sit_i
->elapsed_time
= le64_to_cpu(sbi
->ckpt
->elapsed_time
);
3423 sit_i
->mounted_time
= ktime_get_real_seconds();
3424 init_rwsem(&sit_i
->sentry_lock
);
3428 static int build_free_segmap(struct f2fs_sb_info
*sbi
)
3430 struct free_segmap_info
*free_i
;
3431 unsigned int bitmap_size
, sec_bitmap_size
;
3433 /* allocate memory for free segmap information */
3434 free_i
= kzalloc(sizeof(struct free_segmap_info
), GFP_KERNEL
);
3438 SM_I(sbi
)->free_info
= free_i
;
3440 bitmap_size
= f2fs_bitmap_size(MAIN_SEGS(sbi
));
3441 free_i
->free_segmap
= kvmalloc(bitmap_size
, GFP_KERNEL
);
3442 if (!free_i
->free_segmap
)
3445 sec_bitmap_size
= f2fs_bitmap_size(MAIN_SECS(sbi
));
3446 free_i
->free_secmap
= kvmalloc(sec_bitmap_size
, GFP_KERNEL
);
3447 if (!free_i
->free_secmap
)
3450 /* set all segments as dirty temporarily */
3451 memset(free_i
->free_segmap
, 0xff, bitmap_size
);
3452 memset(free_i
->free_secmap
, 0xff, sec_bitmap_size
);
3454 /* init free segmap information */
3455 free_i
->start_segno
= GET_SEGNO_FROM_SEG0(sbi
, MAIN_BLKADDR(sbi
));
3456 free_i
->free_segments
= 0;
3457 free_i
->free_sections
= 0;
3458 spin_lock_init(&free_i
->segmap_lock
);
3462 static int build_curseg(struct f2fs_sb_info
*sbi
)
3464 struct curseg_info
*array
;
3467 array
= kcalloc(NR_CURSEG_TYPE
, sizeof(*array
), GFP_KERNEL
);
3471 SM_I(sbi
)->curseg_array
= array
;
3473 for (i
= 0; i
< NR_CURSEG_TYPE
; i
++) {
3474 mutex_init(&array
[i
].curseg_mutex
);
3475 array
[i
].sum_blk
= kzalloc(PAGE_SIZE
, GFP_KERNEL
);
3476 if (!array
[i
].sum_blk
)
3478 init_rwsem(&array
[i
].journal_rwsem
);
3479 array
[i
].journal
= kzalloc(sizeof(struct f2fs_journal
),
3481 if (!array
[i
].journal
)
3483 array
[i
].segno
= NULL_SEGNO
;
3484 array
[i
].next_blkoff
= 0;
3486 return restore_curseg_summaries(sbi
);
3489 static int build_sit_entries(struct f2fs_sb_info
*sbi
)
3491 struct sit_info
*sit_i
= SIT_I(sbi
);
3492 struct curseg_info
*curseg
= CURSEG_I(sbi
, CURSEG_COLD_DATA
);
3493 struct f2fs_journal
*journal
= curseg
->journal
;
3494 struct seg_entry
*se
;
3495 struct f2fs_sit_entry sit
;
3496 int sit_blk_cnt
= SIT_BLK_CNT(sbi
);
3497 unsigned int i
, start
, end
;
3498 unsigned int readed
, start_blk
= 0;
3502 readed
= ra_meta_pages(sbi
, start_blk
, BIO_MAX_PAGES
,
3505 start
= start_blk
* sit_i
->sents_per_block
;
3506 end
= (start_blk
+ readed
) * sit_i
->sents_per_block
;
3508 for (; start
< end
&& start
< MAIN_SEGS(sbi
); start
++) {
3509 struct f2fs_sit_block
*sit_blk
;
3512 se
= &sit_i
->sentries
[start
];
3513 page
= get_current_sit_page(sbi
, start
);
3514 sit_blk
= (struct f2fs_sit_block
*)page_address(page
);
3515 sit
= sit_blk
->entries
[SIT_ENTRY_OFFSET(sit_i
, start
)];
3516 f2fs_put_page(page
, 1);
3518 err
= check_block_count(sbi
, start
, &sit
);
3521 seg_info_from_raw_sit(se
, &sit
);
3523 /* build discard map only one time */
3524 if (f2fs_discard_en(sbi
)) {
3525 if (is_set_ckpt_flags(sbi
, CP_TRIMMED_FLAG
)) {
3526 memset(se
->discard_map
, 0xff,
3527 SIT_VBLOCK_MAP_SIZE
);
3529 memcpy(se
->discard_map
,
3531 SIT_VBLOCK_MAP_SIZE
);
3532 sbi
->discard_blks
+=
3533 sbi
->blocks_per_seg
-
3538 if (sbi
->segs_per_sec
> 1)
3539 get_sec_entry(sbi
, start
)->valid_blocks
+=
3542 start_blk
+= readed
;
3543 } while (start_blk
< sit_blk_cnt
);
3545 down_read(&curseg
->journal_rwsem
);
3546 for (i
= 0; i
< sits_in_cursum(journal
); i
++) {
3547 unsigned int old_valid_blocks
;
3549 start
= le32_to_cpu(segno_in_journal(journal
, i
));
3550 se
= &sit_i
->sentries
[start
];
3551 sit
= sit_in_journal(journal
, i
);
3553 old_valid_blocks
= se
->valid_blocks
;
3555 err
= check_block_count(sbi
, start
, &sit
);
3558 seg_info_from_raw_sit(se
, &sit
);
3560 if (f2fs_discard_en(sbi
)) {
3561 if (is_set_ckpt_flags(sbi
, CP_TRIMMED_FLAG
)) {
3562 memset(se
->discard_map
, 0xff,
3563 SIT_VBLOCK_MAP_SIZE
);
3565 memcpy(se
->discard_map
, se
->cur_valid_map
,
3566 SIT_VBLOCK_MAP_SIZE
);
3567 sbi
->discard_blks
+= old_valid_blocks
-
3572 if (sbi
->segs_per_sec
> 1)
3573 get_sec_entry(sbi
, start
)->valid_blocks
+=
3574 se
->valid_blocks
- old_valid_blocks
;
3576 up_read(&curseg
->journal_rwsem
);
3580 static void init_free_segmap(struct f2fs_sb_info
*sbi
)
3585 for (start
= 0; start
< MAIN_SEGS(sbi
); start
++) {
3586 struct seg_entry
*sentry
= get_seg_entry(sbi
, start
);
3587 if (!sentry
->valid_blocks
)
3588 __set_free(sbi
, start
);
3590 SIT_I(sbi
)->written_valid_blocks
+=
3591 sentry
->valid_blocks
;
3594 /* set use the current segments */
3595 for (type
= CURSEG_HOT_DATA
; type
<= CURSEG_COLD_NODE
; type
++) {
3596 struct curseg_info
*curseg_t
= CURSEG_I(sbi
, type
);
3597 __set_test_and_inuse(sbi
, curseg_t
->segno
);
3601 static void init_dirty_segmap(struct f2fs_sb_info
*sbi
)
3603 struct dirty_seglist_info
*dirty_i
= DIRTY_I(sbi
);
3604 struct free_segmap_info
*free_i
= FREE_I(sbi
);
3605 unsigned int segno
= 0, offset
= 0;
3606 unsigned short valid_blocks
;
3609 /* find dirty segment based on free segmap */
3610 segno
= find_next_inuse(free_i
, MAIN_SEGS(sbi
), offset
);
3611 if (segno
>= MAIN_SEGS(sbi
))
3614 valid_blocks
= get_valid_blocks(sbi
, segno
, false);
3615 if (valid_blocks
== sbi
->blocks_per_seg
|| !valid_blocks
)
3617 if (valid_blocks
> sbi
->blocks_per_seg
) {
3618 f2fs_bug_on(sbi
, 1);
3621 mutex_lock(&dirty_i
->seglist_lock
);
3622 __locate_dirty_segment(sbi
, segno
, DIRTY
);
3623 mutex_unlock(&dirty_i
->seglist_lock
);
3627 static int init_victim_secmap(struct f2fs_sb_info
*sbi
)
3629 struct dirty_seglist_info
*dirty_i
= DIRTY_I(sbi
);
3630 unsigned int bitmap_size
= f2fs_bitmap_size(MAIN_SECS(sbi
));
3632 dirty_i
->victim_secmap
= kvzalloc(bitmap_size
, GFP_KERNEL
);
3633 if (!dirty_i
->victim_secmap
)
3638 static int build_dirty_segmap(struct f2fs_sb_info
*sbi
)
3640 struct dirty_seglist_info
*dirty_i
;
3641 unsigned int bitmap_size
, i
;
3643 /* allocate memory for dirty segments list information */
3644 dirty_i
= kzalloc(sizeof(struct dirty_seglist_info
), GFP_KERNEL
);
3648 SM_I(sbi
)->dirty_info
= dirty_i
;
3649 mutex_init(&dirty_i
->seglist_lock
);
3651 bitmap_size
= f2fs_bitmap_size(MAIN_SEGS(sbi
));
3653 for (i
= 0; i
< NR_DIRTY_TYPE
; i
++) {
3654 dirty_i
->dirty_segmap
[i
] = kvzalloc(bitmap_size
, GFP_KERNEL
);
3655 if (!dirty_i
->dirty_segmap
[i
])
3659 init_dirty_segmap(sbi
);
3660 return init_victim_secmap(sbi
);
3664 * Update min, max modified time for cost-benefit GC algorithm
3666 static void init_min_max_mtime(struct f2fs_sb_info
*sbi
)
3668 struct sit_info
*sit_i
= SIT_I(sbi
);
3671 down_write(&sit_i
->sentry_lock
);
3673 sit_i
->min_mtime
= LLONG_MAX
;
3675 for (segno
= 0; segno
< MAIN_SEGS(sbi
); segno
+= sbi
->segs_per_sec
) {
3677 unsigned long long mtime
= 0;
3679 for (i
= 0; i
< sbi
->segs_per_sec
; i
++)
3680 mtime
+= get_seg_entry(sbi
, segno
+ i
)->mtime
;
3682 mtime
= div_u64(mtime
, sbi
->segs_per_sec
);
3684 if (sit_i
->min_mtime
> mtime
)
3685 sit_i
->min_mtime
= mtime
;
3687 sit_i
->max_mtime
= get_mtime(sbi
);
3688 up_write(&sit_i
->sentry_lock
);
3691 int build_segment_manager(struct f2fs_sb_info
*sbi
)
3693 struct f2fs_super_block
*raw_super
= F2FS_RAW_SUPER(sbi
);
3694 struct f2fs_checkpoint
*ckpt
= F2FS_CKPT(sbi
);
3695 struct f2fs_sm_info
*sm_info
;
3698 sm_info
= kzalloc(sizeof(struct f2fs_sm_info
), GFP_KERNEL
);
3703 sbi
->sm_info
= sm_info
;
3704 sm_info
->seg0_blkaddr
= le32_to_cpu(raw_super
->segment0_blkaddr
);
3705 sm_info
->main_blkaddr
= le32_to_cpu(raw_super
->main_blkaddr
);
3706 sm_info
->segment_count
= le32_to_cpu(raw_super
->segment_count
);
3707 sm_info
->reserved_segments
= le32_to_cpu(ckpt
->rsvd_segment_count
);
3708 sm_info
->ovp_segments
= le32_to_cpu(ckpt
->overprov_segment_count
);
3709 sm_info
->main_segments
= le32_to_cpu(raw_super
->segment_count_main
);
3710 sm_info
->ssa_blkaddr
= le32_to_cpu(raw_super
->ssa_blkaddr
);
3711 sm_info
->rec_prefree_segments
= sm_info
->main_segments
*
3712 DEF_RECLAIM_PREFREE_SEGMENTS
/ 100;
3713 if (sm_info
->rec_prefree_segments
> DEF_MAX_RECLAIM_PREFREE_SEGMENTS
)
3714 sm_info
->rec_prefree_segments
= DEF_MAX_RECLAIM_PREFREE_SEGMENTS
;
3716 if (!test_opt(sbi
, LFS
))
3717 sm_info
->ipu_policy
= 1 << F2FS_IPU_FSYNC
;
3718 sm_info
->min_ipu_util
= DEF_MIN_IPU_UTIL
;
3719 sm_info
->min_fsync_blocks
= DEF_MIN_FSYNC_BLOCKS
;
3720 sm_info
->min_hot_blocks
= DEF_MIN_HOT_BLOCKS
;
3721 sm_info
->min_ssr_sections
= reserved_sections(sbi
);
3723 sm_info
->trim_sections
= DEF_BATCHED_TRIM_SECTIONS
;
3725 INIT_LIST_HEAD(&sm_info
->sit_entry_set
);
3727 init_rwsem(&sm_info
->curseg_lock
);
3729 if (!f2fs_readonly(sbi
->sb
)) {
3730 err
= create_flush_cmd_control(sbi
);
3735 err
= create_discard_cmd_control(sbi
);
3739 err
= build_sit_info(sbi
);
3742 err
= build_free_segmap(sbi
);
3745 err
= build_curseg(sbi
);
3749 /* reinit free segmap based on SIT */
3750 err
= build_sit_entries(sbi
);
3754 init_free_segmap(sbi
);
3755 err
= build_dirty_segmap(sbi
);
3759 init_min_max_mtime(sbi
);
3763 static void discard_dirty_segmap(struct f2fs_sb_info
*sbi
,
3764 enum dirty_type dirty_type
)
3766 struct dirty_seglist_info
*dirty_i
= DIRTY_I(sbi
);
3768 mutex_lock(&dirty_i
->seglist_lock
);
3769 kvfree(dirty_i
->dirty_segmap
[dirty_type
]);
3770 dirty_i
->nr_dirty
[dirty_type
] = 0;
3771 mutex_unlock(&dirty_i
->seglist_lock
);
3774 static void destroy_victim_secmap(struct f2fs_sb_info
*sbi
)
3776 struct dirty_seglist_info
*dirty_i
= DIRTY_I(sbi
);
3777 kvfree(dirty_i
->victim_secmap
);
3780 static void destroy_dirty_segmap(struct f2fs_sb_info
*sbi
)
3782 struct dirty_seglist_info
*dirty_i
= DIRTY_I(sbi
);
3788 /* discard pre-free/dirty segments list */
3789 for (i
= 0; i
< NR_DIRTY_TYPE
; i
++)
3790 discard_dirty_segmap(sbi
, i
);
3792 destroy_victim_secmap(sbi
);
3793 SM_I(sbi
)->dirty_info
= NULL
;
3797 static void destroy_curseg(struct f2fs_sb_info
*sbi
)
3799 struct curseg_info
*array
= SM_I(sbi
)->curseg_array
;
3804 SM_I(sbi
)->curseg_array
= NULL
;
3805 for (i
= 0; i
< NR_CURSEG_TYPE
; i
++) {
3806 kfree(array
[i
].sum_blk
);
3807 kfree(array
[i
].journal
);
3812 static void destroy_free_segmap(struct f2fs_sb_info
*sbi
)
3814 struct free_segmap_info
*free_i
= SM_I(sbi
)->free_info
;
3817 SM_I(sbi
)->free_info
= NULL
;
3818 kvfree(free_i
->free_segmap
);
3819 kvfree(free_i
->free_secmap
);
3823 static void destroy_sit_info(struct f2fs_sb_info
*sbi
)
3825 struct sit_info
*sit_i
= SIT_I(sbi
);
3831 if (sit_i
->sentries
) {
3832 for (start
= 0; start
< MAIN_SEGS(sbi
); start
++) {
3833 kfree(sit_i
->sentries
[start
].cur_valid_map
);
3834 #ifdef CONFIG_F2FS_CHECK_FS
3835 kfree(sit_i
->sentries
[start
].cur_valid_map_mir
);
3837 kfree(sit_i
->sentries
[start
].ckpt_valid_map
);
3838 kfree(sit_i
->sentries
[start
].discard_map
);
3841 kfree(sit_i
->tmp_map
);
3843 kvfree(sit_i
->sentries
);
3844 kvfree(sit_i
->sec_entries
);
3845 kvfree(sit_i
->dirty_sentries_bitmap
);
3847 SM_I(sbi
)->sit_info
= NULL
;
3848 kfree(sit_i
->sit_bitmap
);
3849 #ifdef CONFIG_F2FS_CHECK_FS
3850 kfree(sit_i
->sit_bitmap_mir
);
3855 void destroy_segment_manager(struct f2fs_sb_info
*sbi
)
3857 struct f2fs_sm_info
*sm_info
= SM_I(sbi
);
3861 destroy_flush_cmd_control(sbi
, true);
3862 destroy_discard_cmd_control(sbi
);
3863 destroy_dirty_segmap(sbi
);
3864 destroy_curseg(sbi
);
3865 destroy_free_segmap(sbi
);
3866 destroy_sit_info(sbi
);
3867 sbi
->sm_info
= NULL
;
3871 int __init
create_segment_manager_caches(void)
3873 discard_entry_slab
= f2fs_kmem_cache_create("discard_entry",
3874 sizeof(struct discard_entry
));
3875 if (!discard_entry_slab
)
3878 discard_cmd_slab
= f2fs_kmem_cache_create("discard_cmd",
3879 sizeof(struct discard_cmd
));
3880 if (!discard_cmd_slab
)
3881 goto destroy_discard_entry
;
3883 sit_entry_set_slab
= f2fs_kmem_cache_create("sit_entry_set",
3884 sizeof(struct sit_entry_set
));
3885 if (!sit_entry_set_slab
)
3886 goto destroy_discard_cmd
;
3888 inmem_entry_slab
= f2fs_kmem_cache_create("inmem_page_entry",
3889 sizeof(struct inmem_pages
));
3890 if (!inmem_entry_slab
)
3891 goto destroy_sit_entry_set
;
3894 destroy_sit_entry_set
:
3895 kmem_cache_destroy(sit_entry_set_slab
);
3896 destroy_discard_cmd
:
3897 kmem_cache_destroy(discard_cmd_slab
);
3898 destroy_discard_entry
:
3899 kmem_cache_destroy(discard_entry_slab
);
3904 void destroy_segment_manager_caches(void)
3906 kmem_cache_destroy(sit_entry_set_slab
);
3907 kmem_cache_destroy(discard_cmd_slab
);
3908 kmem_cache_destroy(discard_entry_slab
);
3909 kmem_cache_destroy(inmem_entry_slab
);