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.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_rw
= REQ_OP_WRITE
;
519 ret
= submit_bio_wait(WRITE_FLUSH
, 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
= bio
->bi_error
;
883 complete_all(&dc
->wait
);
887 /* copied from block/blk-lib.c in 4.10-rc1 */
888 static int __blkdev_issue_discard(struct block_device
*bdev
, sector_t sector
,
889 sector_t nr_sects
, gfp_t gfp_mask
, int flags
,
892 struct request_queue
*q
= bdev_get_queue(bdev
);
893 struct bio
*bio
= *biop
;
894 unsigned int granularity
;
895 int op
= REQ_WRITE
| REQ_DISCARD
;
902 if (!blk_queue_discard(q
))
905 if (flags
& BLKDEV_DISCARD_SECURE
) {
906 if (!blk_queue_secdiscard(q
))
911 bs_mask
= (bdev_logical_block_size(bdev
) >> 9) - 1;
912 if ((sector
| nr_sects
) & bs_mask
)
915 /* Zero-sector (unknown) and one-sector granularities are the same. */
916 granularity
= max(q
->limits
.discard_granularity
>> 9, 1U);
917 alignment
= (bdev_discard_alignment(bdev
) >> 9) % granularity
;
920 unsigned int req_sects
;
921 sector_t end_sect
, tmp
;
923 /* Make sure bi_size doesn't overflow */
924 req_sects
= min_t(sector_t
, nr_sects
, UINT_MAX
>> 9);
927 * If splitting a request, and the next starting sector would be
928 * misaligned, stop the discard at the previous aligned sector.
930 end_sect
= sector
+ req_sects
;
932 if (req_sects
< nr_sects
&&
933 sector_div(tmp
, granularity
) != alignment
) {
934 end_sect
= end_sect
- alignment
;
935 sector_div(end_sect
, granularity
);
936 end_sect
= end_sect
* granularity
+ alignment
;
937 req_sects
= end_sect
- sector
;
941 int ret
= submit_bio_wait(op
, bio
);
946 bio
= bio_alloc(GFP_NOIO
| __GFP_NOFAIL
, 1);
947 bio
->bi_iter
.bi_sector
= sector
;
949 bio_set_op_attrs(bio
, op
, 0);
951 bio
->bi_iter
.bi_size
= req_sects
<< 9;
952 nr_sects
-= req_sects
;
956 * We can loop for a long time in here, if someone does
957 * full device discards (like mkfs). Be nice and allow
958 * us to schedule out to avoid softlocking if preempt
968 void __check_sit_bitmap(struct f2fs_sb_info
*sbi
,
969 block_t start
, block_t end
)
971 #ifdef CONFIG_F2FS_CHECK_FS
972 struct seg_entry
*sentry
;
975 unsigned long offset
, size
, max_blocks
= sbi
->blocks_per_seg
;
979 segno
= GET_SEGNO(sbi
, blk
);
980 sentry
= get_seg_entry(sbi
, segno
);
981 offset
= GET_BLKOFF_FROM_SEG0(sbi
, blk
);
983 if (end
< START_BLOCK(sbi
, segno
+ 1))
984 size
= GET_BLKOFF_FROM_SEG0(sbi
, end
);
987 map
= (unsigned long *)(sentry
->cur_valid_map
);
988 offset
= __find_rev_next_bit(map
, size
, offset
);
989 f2fs_bug_on(sbi
, offset
!= size
);
990 blk
= START_BLOCK(sbi
, segno
+ 1);
995 /* this function is copied from blkdev_issue_discard from block/blk-lib.c */
996 static void __submit_discard_cmd(struct f2fs_sb_info
*sbi
,
997 struct discard_policy
*dpolicy
,
998 struct discard_cmd
*dc
)
1000 struct discard_cmd_control
*dcc
= SM_I(sbi
)->dcc_info
;
1001 struct list_head
*wait_list
= (dpolicy
->type
== DPOLICY_FSTRIM
) ?
1002 &(dcc
->fstrim_list
) : &(dcc
->wait_list
);
1003 struct bio
*bio
= NULL
;
1004 int flag
= dpolicy
->sync
? REQ_SYNC
: 0;
1006 if (dc
->state
!= D_PREP
)
1009 trace_f2fs_issue_discard(dc
->bdev
, dc
->start
, dc
->len
);
1011 dc
->error
= __blkdev_issue_discard(dc
->bdev
,
1012 SECTOR_FROM_BLOCK(dc
->start
),
1013 SECTOR_FROM_BLOCK(dc
->len
),
1016 /* should keep before submission to avoid D_DONE right away */
1017 dc
->state
= D_SUBMIT
;
1018 atomic_inc(&dcc
->issued_discard
);
1019 atomic_inc(&dcc
->issing_discard
);
1021 bio
->bi_private
= dc
;
1022 bio
->bi_end_io
= f2fs_submit_discard_endio
;
1023 submit_bio(flag
, bio
);
1024 list_move_tail(&dc
->list
, wait_list
);
1025 __check_sit_bitmap(sbi
, dc
->start
, dc
->start
+ dc
->len
);
1027 f2fs_update_iostat(sbi
, FS_DISCARD
, 1);
1030 __remove_discard_cmd(sbi
, dc
);
1034 static struct discard_cmd
*__insert_discard_tree(struct f2fs_sb_info
*sbi
,
1035 struct block_device
*bdev
, block_t lstart
,
1036 block_t start
, block_t len
,
1037 struct rb_node
**insert_p
,
1038 struct rb_node
*insert_parent
)
1040 struct discard_cmd_control
*dcc
= SM_I(sbi
)->dcc_info
;
1042 struct rb_node
*parent
= NULL
;
1043 struct discard_cmd
*dc
= NULL
;
1045 if (insert_p
&& insert_parent
) {
1046 parent
= insert_parent
;
1051 p
= __lookup_rb_tree_for_insert(sbi
, &dcc
->root
, &parent
, lstart
);
1053 dc
= __attach_discard_cmd(sbi
, bdev
, lstart
, start
, len
, parent
, p
);
1060 static void __relocate_discard_cmd(struct discard_cmd_control
*dcc
,
1061 struct discard_cmd
*dc
)
1063 list_move_tail(&dc
->list
, &dcc
->pend_list
[plist_idx(dc
->len
)]);
1066 static void __punch_discard_cmd(struct f2fs_sb_info
*sbi
,
1067 struct discard_cmd
*dc
, block_t blkaddr
)
1069 struct discard_cmd_control
*dcc
= SM_I(sbi
)->dcc_info
;
1070 struct discard_info di
= dc
->di
;
1071 bool modified
= false;
1073 if (dc
->state
== D_DONE
|| dc
->len
== 1) {
1074 __remove_discard_cmd(sbi
, dc
);
1078 dcc
->undiscard_blks
-= di
.len
;
1080 if (blkaddr
> di
.lstart
) {
1081 dc
->len
= blkaddr
- dc
->lstart
;
1082 dcc
->undiscard_blks
+= dc
->len
;
1083 __relocate_discard_cmd(dcc
, dc
);
1087 if (blkaddr
< di
.lstart
+ di
.len
- 1) {
1089 __insert_discard_tree(sbi
, dc
->bdev
, blkaddr
+ 1,
1090 di
.start
+ blkaddr
+ 1 - di
.lstart
,
1091 di
.lstart
+ di
.len
- 1 - blkaddr
,
1097 dcc
->undiscard_blks
+= dc
->len
;
1098 __relocate_discard_cmd(dcc
, dc
);
1103 static void __update_discard_tree_range(struct f2fs_sb_info
*sbi
,
1104 struct block_device
*bdev
, block_t lstart
,
1105 block_t start
, block_t len
)
1107 struct discard_cmd_control
*dcc
= SM_I(sbi
)->dcc_info
;
1108 struct discard_cmd
*prev_dc
= NULL
, *next_dc
= NULL
;
1109 struct discard_cmd
*dc
;
1110 struct discard_info di
= {0};
1111 struct rb_node
**insert_p
= NULL
, *insert_parent
= NULL
;
1112 block_t end
= lstart
+ len
;
1114 mutex_lock(&dcc
->cmd_lock
);
1116 dc
= (struct discard_cmd
*)__lookup_rb_tree_ret(&dcc
->root
,
1118 (struct rb_entry
**)&prev_dc
,
1119 (struct rb_entry
**)&next_dc
,
1120 &insert_p
, &insert_parent
, true);
1126 di
.len
= next_dc
? next_dc
->lstart
- lstart
: len
;
1127 di
.len
= min(di
.len
, len
);
1132 struct rb_node
*node
;
1133 bool merged
= false;
1134 struct discard_cmd
*tdc
= NULL
;
1137 di
.lstart
= prev_dc
->lstart
+ prev_dc
->len
;
1138 if (di
.lstart
< lstart
)
1140 if (di
.lstart
>= end
)
1143 if (!next_dc
|| next_dc
->lstart
> end
)
1144 di
.len
= end
- di
.lstart
;
1146 di
.len
= next_dc
->lstart
- di
.lstart
;
1147 di
.start
= start
+ di
.lstart
- lstart
;
1153 if (prev_dc
&& prev_dc
->state
== D_PREP
&&
1154 prev_dc
->bdev
== bdev
&&
1155 __is_discard_back_mergeable(&di
, &prev_dc
->di
)) {
1156 prev_dc
->di
.len
+= di
.len
;
1157 dcc
->undiscard_blks
+= di
.len
;
1158 __relocate_discard_cmd(dcc
, prev_dc
);
1164 if (next_dc
&& next_dc
->state
== D_PREP
&&
1165 next_dc
->bdev
== bdev
&&
1166 __is_discard_front_mergeable(&di
, &next_dc
->di
)) {
1167 next_dc
->di
.lstart
= di
.lstart
;
1168 next_dc
->di
.len
+= di
.len
;
1169 next_dc
->di
.start
= di
.start
;
1170 dcc
->undiscard_blks
+= di
.len
;
1171 __relocate_discard_cmd(dcc
, next_dc
);
1173 __remove_discard_cmd(sbi
, tdc
);
1178 __insert_discard_tree(sbi
, bdev
, di
.lstart
, di
.start
,
1179 di
.len
, NULL
, NULL
);
1186 node
= rb_next(&prev_dc
->rb_node
);
1187 next_dc
= rb_entry_safe(node
, struct discard_cmd
, rb_node
);
1190 mutex_unlock(&dcc
->cmd_lock
);
1193 static int __queue_discard_cmd(struct f2fs_sb_info
*sbi
,
1194 struct block_device
*bdev
, block_t blkstart
, block_t blklen
)
1196 block_t lblkstart
= blkstart
;
1198 trace_f2fs_queue_discard(bdev
, blkstart
, blklen
);
1201 int devi
= f2fs_target_device_index(sbi
, blkstart
);
1203 blkstart
-= FDEV(devi
).start_blk
;
1205 __update_discard_tree_range(sbi
, bdev
, lblkstart
, blkstart
, blklen
);
1209 static void __issue_discard_cmd_range(struct f2fs_sb_info
*sbi
,
1210 struct discard_policy
*dpolicy
,
1211 unsigned int start
, unsigned int end
)
1213 struct discard_cmd_control
*dcc
= SM_I(sbi
)->dcc_info
;
1214 struct discard_cmd
*prev_dc
= NULL
, *next_dc
= NULL
;
1215 struct rb_node
**insert_p
= NULL
, *insert_parent
= NULL
;
1216 struct discard_cmd
*dc
;
1217 struct blk_plug plug
;
1223 mutex_lock(&dcc
->cmd_lock
);
1224 f2fs_bug_on(sbi
, !__check_rb_tree_consistence(sbi
, &dcc
->root
));
1226 dc
= (struct discard_cmd
*)__lookup_rb_tree_ret(&dcc
->root
,
1228 (struct rb_entry
**)&prev_dc
,
1229 (struct rb_entry
**)&next_dc
,
1230 &insert_p
, &insert_parent
, true);
1234 blk_start_plug(&plug
);
1236 while (dc
&& dc
->lstart
<= end
) {
1237 struct rb_node
*node
;
1239 if (dc
->len
< dpolicy
->granularity
)
1242 if (dc
->state
!= D_PREP
) {
1243 list_move_tail(&dc
->list
, &dcc
->fstrim_list
);
1247 __submit_discard_cmd(sbi
, dpolicy
, dc
);
1249 if (++issued
>= dpolicy
->max_requests
) {
1250 start
= dc
->lstart
+ dc
->len
;
1252 blk_finish_plug(&plug
);
1253 mutex_unlock(&dcc
->cmd_lock
);
1260 node
= rb_next(&dc
->rb_node
);
1261 dc
= rb_entry_safe(node
, struct discard_cmd
, rb_node
);
1263 if (fatal_signal_pending(current
))
1267 blk_finish_plug(&plug
);
1268 mutex_unlock(&dcc
->cmd_lock
);
1271 static int __issue_discard_cmd(struct f2fs_sb_info
*sbi
,
1272 struct discard_policy
*dpolicy
)
1274 struct discard_cmd_control
*dcc
= SM_I(sbi
)->dcc_info
;
1275 struct list_head
*pend_list
;
1276 struct discard_cmd
*dc
, *tmp
;
1277 struct blk_plug plug
;
1278 int i
, iter
= 0, issued
= 0;
1279 bool io_interrupted
= false;
1281 for (i
= MAX_PLIST_NUM
- 1; i
>= 0; i
--) {
1282 if (i
+ 1 < dpolicy
->granularity
)
1284 pend_list
= &dcc
->pend_list
[i
];
1286 mutex_lock(&dcc
->cmd_lock
);
1287 f2fs_bug_on(sbi
, !__check_rb_tree_consistence(sbi
, &dcc
->root
));
1288 blk_start_plug(&plug
);
1289 list_for_each_entry_safe(dc
, tmp
, pend_list
, list
) {
1290 f2fs_bug_on(sbi
, dc
->state
!= D_PREP
);
1292 if (dpolicy
->io_aware
&& i
< dpolicy
->io_aware_gran
&&
1294 io_interrupted
= true;
1298 __submit_discard_cmd(sbi
, dpolicy
, dc
);
1301 if (++iter
>= dpolicy
->max_requests
)
1304 blk_finish_plug(&plug
);
1305 mutex_unlock(&dcc
->cmd_lock
);
1307 if (iter
>= dpolicy
->max_requests
)
1311 if (!issued
&& io_interrupted
)
1317 static bool __drop_discard_cmd(struct f2fs_sb_info
*sbi
)
1319 struct discard_cmd_control
*dcc
= SM_I(sbi
)->dcc_info
;
1320 struct list_head
*pend_list
;
1321 struct discard_cmd
*dc
, *tmp
;
1323 bool dropped
= false;
1325 mutex_lock(&dcc
->cmd_lock
);
1326 for (i
= MAX_PLIST_NUM
- 1; i
>= 0; i
--) {
1327 pend_list
= &dcc
->pend_list
[i
];
1328 list_for_each_entry_safe(dc
, tmp
, pend_list
, list
) {
1329 f2fs_bug_on(sbi
, dc
->state
!= D_PREP
);
1330 __remove_discard_cmd(sbi
, dc
);
1334 mutex_unlock(&dcc
->cmd_lock
);
1339 static unsigned int __wait_one_discard_bio(struct f2fs_sb_info
*sbi
,
1340 struct discard_cmd
*dc
)
1342 struct discard_cmd_control
*dcc
= SM_I(sbi
)->dcc_info
;
1343 unsigned int len
= 0;
1345 wait_for_completion_io(&dc
->wait
);
1346 mutex_lock(&dcc
->cmd_lock
);
1347 f2fs_bug_on(sbi
, dc
->state
!= D_DONE
);
1352 __remove_discard_cmd(sbi
, dc
);
1354 mutex_unlock(&dcc
->cmd_lock
);
1359 static unsigned int __wait_discard_cmd_range(struct f2fs_sb_info
*sbi
,
1360 struct discard_policy
*dpolicy
,
1361 block_t start
, block_t end
)
1363 struct discard_cmd_control
*dcc
= SM_I(sbi
)->dcc_info
;
1364 struct list_head
*wait_list
= (dpolicy
->type
== DPOLICY_FSTRIM
) ?
1365 &(dcc
->fstrim_list
) : &(dcc
->wait_list
);
1366 struct discard_cmd
*dc
, *tmp
;
1368 unsigned int trimmed
= 0;
1373 mutex_lock(&dcc
->cmd_lock
);
1374 list_for_each_entry_safe(dc
, tmp
, wait_list
, list
) {
1375 if (dc
->lstart
+ dc
->len
<= start
|| end
<= dc
->lstart
)
1377 if (dc
->len
< dpolicy
->granularity
)
1379 if (dc
->state
== D_DONE
&& !dc
->ref
) {
1380 wait_for_completion_io(&dc
->wait
);
1383 __remove_discard_cmd(sbi
, dc
);
1390 mutex_unlock(&dcc
->cmd_lock
);
1393 trimmed
+= __wait_one_discard_bio(sbi
, dc
);
1400 static void __wait_all_discard_cmd(struct f2fs_sb_info
*sbi
,
1401 struct discard_policy
*dpolicy
)
1403 __wait_discard_cmd_range(sbi
, dpolicy
, 0, UINT_MAX
);
1406 /* This should be covered by global mutex, &sit_i->sentry_lock */
1407 void f2fs_wait_discard_bio(struct f2fs_sb_info
*sbi
, block_t blkaddr
)
1409 struct discard_cmd_control
*dcc
= SM_I(sbi
)->dcc_info
;
1410 struct discard_cmd
*dc
;
1411 bool need_wait
= false;
1413 mutex_lock(&dcc
->cmd_lock
);
1414 dc
= (struct discard_cmd
*)__lookup_rb_tree(&dcc
->root
, NULL
, blkaddr
);
1416 if (dc
->state
== D_PREP
) {
1417 __punch_discard_cmd(sbi
, dc
, blkaddr
);
1423 mutex_unlock(&dcc
->cmd_lock
);
1426 __wait_one_discard_bio(sbi
, dc
);
1429 void stop_discard_thread(struct f2fs_sb_info
*sbi
)
1431 struct discard_cmd_control
*dcc
= SM_I(sbi
)->dcc_info
;
1433 if (dcc
&& dcc
->f2fs_issue_discard
) {
1434 struct task_struct
*discard_thread
= dcc
->f2fs_issue_discard
;
1436 dcc
->f2fs_issue_discard
= NULL
;
1437 kthread_stop(discard_thread
);
1441 /* This comes from f2fs_put_super */
1442 bool f2fs_wait_discard_bios(struct f2fs_sb_info
*sbi
)
1444 struct discard_cmd_control
*dcc
= SM_I(sbi
)->dcc_info
;
1445 struct discard_policy dpolicy
;
1448 init_discard_policy(&dpolicy
, DPOLICY_UMOUNT
, dcc
->discard_granularity
);
1449 __issue_discard_cmd(sbi
, &dpolicy
);
1450 dropped
= __drop_discard_cmd(sbi
);
1451 __wait_all_discard_cmd(sbi
, &dpolicy
);
1456 static int issue_discard_thread(void *data
)
1458 struct f2fs_sb_info
*sbi
= data
;
1459 struct discard_cmd_control
*dcc
= SM_I(sbi
)->dcc_info
;
1460 wait_queue_head_t
*q
= &dcc
->discard_wait_queue
;
1461 struct discard_policy dpolicy
;
1462 unsigned int wait_ms
= DEF_MIN_DISCARD_ISSUE_TIME
;
1468 init_discard_policy(&dpolicy
, DPOLICY_BG
,
1469 dcc
->discard_granularity
);
1471 wait_event_interruptible_timeout(*q
,
1472 kthread_should_stop() || freezing(current
) ||
1474 msecs_to_jiffies(wait_ms
));
1475 if (try_to_freeze())
1477 if (kthread_should_stop())
1480 if (dcc
->discard_wake
) {
1481 dcc
->discard_wake
= 0;
1482 if (sbi
->gc_thread
&& sbi
->gc_thread
->gc_urgent
)
1483 init_discard_policy(&dpolicy
,
1487 sb_start_intwrite(sbi
->sb
);
1489 issued
= __issue_discard_cmd(sbi
, &dpolicy
);
1491 __wait_all_discard_cmd(sbi
, &dpolicy
);
1492 wait_ms
= dpolicy
.min_interval
;
1494 wait_ms
= dpolicy
.max_interval
;
1497 sb_end_intwrite(sbi
->sb
);
1499 } while (!kthread_should_stop());
1503 #ifdef CONFIG_BLK_DEV_ZONED
1504 static int __f2fs_issue_discard_zone(struct f2fs_sb_info
*sbi
,
1505 struct block_device
*bdev
, block_t blkstart
, block_t blklen
)
1507 sector_t sector
, nr_sects
;
1508 block_t lblkstart
= blkstart
;
1512 devi
= f2fs_target_device_index(sbi
, blkstart
);
1513 blkstart
-= FDEV(devi
).start_blk
;
1517 * We need to know the type of the zone: for conventional zones,
1518 * use regular discard if the drive supports it. For sequential
1519 * zones, reset the zone write pointer.
1521 switch (get_blkz_type(sbi
, bdev
, blkstart
)) {
1523 case BLK_ZONE_TYPE_CONVENTIONAL
:
1524 if (!blk_queue_discard(bdev_get_queue(bdev
)))
1526 return __queue_discard_cmd(sbi
, bdev
, lblkstart
, blklen
);
1527 case BLK_ZONE_TYPE_SEQWRITE_REQ
:
1528 case BLK_ZONE_TYPE_SEQWRITE_PREF
:
1529 sector
= SECTOR_FROM_BLOCK(blkstart
);
1530 nr_sects
= SECTOR_FROM_BLOCK(blklen
);
1532 if (sector
& (bdev_zone_sectors(bdev
) - 1) ||
1533 nr_sects
!= bdev_zone_sectors(bdev
)) {
1534 f2fs_msg(sbi
->sb
, KERN_INFO
,
1535 "(%d) %s: Unaligned discard attempted (block %x + %x)",
1536 devi
, sbi
->s_ndevs
? FDEV(devi
).path
: "",
1540 trace_f2fs_issue_reset_zone(bdev
, blkstart
);
1541 return blkdev_reset_zones(bdev
, sector
,
1542 nr_sects
, GFP_NOFS
);
1544 /* Unknown zone type: broken device ? */
1550 static int __issue_discard_async(struct f2fs_sb_info
*sbi
,
1551 struct block_device
*bdev
, block_t blkstart
, block_t blklen
)
1553 #ifdef CONFIG_BLK_DEV_ZONED
1554 if (f2fs_sb_mounted_blkzoned(sbi
->sb
) &&
1555 bdev_zoned_model(bdev
) != BLK_ZONED_NONE
)
1556 return __f2fs_issue_discard_zone(sbi
, bdev
, blkstart
, blklen
);
1558 return __queue_discard_cmd(sbi
, bdev
, blkstart
, blklen
);
1561 static int f2fs_issue_discard(struct f2fs_sb_info
*sbi
,
1562 block_t blkstart
, block_t blklen
)
1564 sector_t start
= blkstart
, len
= 0;
1565 struct block_device
*bdev
;
1566 struct seg_entry
*se
;
1567 unsigned int offset
;
1571 bdev
= f2fs_target_device(sbi
, blkstart
, NULL
);
1573 for (i
= blkstart
; i
< blkstart
+ blklen
; i
++, len
++) {
1575 struct block_device
*bdev2
=
1576 f2fs_target_device(sbi
, i
, NULL
);
1578 if (bdev2
!= bdev
) {
1579 err
= __issue_discard_async(sbi
, bdev
,
1589 se
= get_seg_entry(sbi
, GET_SEGNO(sbi
, i
));
1590 offset
= GET_BLKOFF_FROM_SEG0(sbi
, i
);
1592 if (!f2fs_test_and_set_bit(offset
, se
->discard_map
))
1593 sbi
->discard_blks
--;
1597 err
= __issue_discard_async(sbi
, bdev
, start
, len
);
1601 static bool add_discard_addrs(struct f2fs_sb_info
*sbi
, struct cp_control
*cpc
,
1604 int entries
= SIT_VBLOCK_MAP_SIZE
/ sizeof(unsigned long);
1605 int max_blocks
= sbi
->blocks_per_seg
;
1606 struct seg_entry
*se
= get_seg_entry(sbi
, cpc
->trim_start
);
1607 unsigned long *cur_map
= (unsigned long *)se
->cur_valid_map
;
1608 unsigned long *ckpt_map
= (unsigned long *)se
->ckpt_valid_map
;
1609 unsigned long *discard_map
= (unsigned long *)se
->discard_map
;
1610 unsigned long *dmap
= SIT_I(sbi
)->tmp_map
;
1611 unsigned int start
= 0, end
= -1;
1612 bool force
= (cpc
->reason
& CP_DISCARD
);
1613 struct discard_entry
*de
= NULL
;
1614 struct list_head
*head
= &SM_I(sbi
)->dcc_info
->entry_list
;
1617 if (se
->valid_blocks
== max_blocks
|| !f2fs_discard_en(sbi
))
1621 if (!test_opt(sbi
, DISCARD
) || !se
->valid_blocks
||
1622 SM_I(sbi
)->dcc_info
->nr_discards
>=
1623 SM_I(sbi
)->dcc_info
->max_discards
)
1627 /* SIT_VBLOCK_MAP_SIZE should be multiple of sizeof(unsigned long) */
1628 for (i
= 0; i
< entries
; i
++)
1629 dmap
[i
] = force
? ~ckpt_map
[i
] & ~discard_map
[i
] :
1630 (cur_map
[i
] ^ ckpt_map
[i
]) & ckpt_map
[i
];
1632 while (force
|| SM_I(sbi
)->dcc_info
->nr_discards
<=
1633 SM_I(sbi
)->dcc_info
->max_discards
) {
1634 start
= __find_rev_next_bit(dmap
, max_blocks
, end
+ 1);
1635 if (start
>= max_blocks
)
1638 end
= __find_rev_next_zero_bit(dmap
, max_blocks
, start
+ 1);
1639 if (force
&& start
&& end
!= max_blocks
1640 && (end
- start
) < cpc
->trim_minlen
)
1647 de
= f2fs_kmem_cache_alloc(discard_entry_slab
,
1649 de
->start_blkaddr
= START_BLOCK(sbi
, cpc
->trim_start
);
1650 list_add_tail(&de
->list
, head
);
1653 for (i
= start
; i
< end
; i
++)
1654 __set_bit_le(i
, (void *)de
->discard_map
);
1656 SM_I(sbi
)->dcc_info
->nr_discards
+= end
- start
;
1661 void release_discard_addrs(struct f2fs_sb_info
*sbi
)
1663 struct list_head
*head
= &(SM_I(sbi
)->dcc_info
->entry_list
);
1664 struct discard_entry
*entry
, *this;
1667 list_for_each_entry_safe(entry
, this, head
, list
) {
1668 list_del(&entry
->list
);
1669 kmem_cache_free(discard_entry_slab
, entry
);
1674 * Should call clear_prefree_segments after checkpoint is done.
1676 static void set_prefree_as_free_segments(struct f2fs_sb_info
*sbi
)
1678 struct dirty_seglist_info
*dirty_i
= DIRTY_I(sbi
);
1681 mutex_lock(&dirty_i
->seglist_lock
);
1682 for_each_set_bit(segno
, dirty_i
->dirty_segmap
[PRE
], MAIN_SEGS(sbi
))
1683 __set_test_and_free(sbi
, segno
);
1684 mutex_unlock(&dirty_i
->seglist_lock
);
1687 void clear_prefree_segments(struct f2fs_sb_info
*sbi
, struct cp_control
*cpc
)
1689 struct discard_cmd_control
*dcc
= SM_I(sbi
)->dcc_info
;
1690 struct list_head
*head
= &dcc
->entry_list
;
1691 struct discard_entry
*entry
, *this;
1692 struct dirty_seglist_info
*dirty_i
= DIRTY_I(sbi
);
1693 unsigned long *prefree_map
= dirty_i
->dirty_segmap
[PRE
];
1694 unsigned int start
= 0, end
= -1;
1695 unsigned int secno
, start_segno
;
1696 bool force
= (cpc
->reason
& CP_DISCARD
);
1698 mutex_lock(&dirty_i
->seglist_lock
);
1702 start
= find_next_bit(prefree_map
, MAIN_SEGS(sbi
), end
+ 1);
1703 if (start
>= MAIN_SEGS(sbi
))
1705 end
= find_next_zero_bit(prefree_map
, MAIN_SEGS(sbi
),
1708 for (i
= start
; i
< end
; i
++)
1709 clear_bit(i
, prefree_map
);
1711 dirty_i
->nr_dirty
[PRE
] -= end
- start
;
1713 if (!test_opt(sbi
, DISCARD
))
1716 if (force
&& start
>= cpc
->trim_start
&&
1717 (end
- 1) <= cpc
->trim_end
)
1720 if (!test_opt(sbi
, LFS
) || sbi
->segs_per_sec
== 1) {
1721 f2fs_issue_discard(sbi
, START_BLOCK(sbi
, start
),
1722 (end
- start
) << sbi
->log_blocks_per_seg
);
1726 secno
= GET_SEC_FROM_SEG(sbi
, start
);
1727 start_segno
= GET_SEG_FROM_SEC(sbi
, secno
);
1728 if (!IS_CURSEC(sbi
, secno
) &&
1729 !get_valid_blocks(sbi
, start
, true))
1730 f2fs_issue_discard(sbi
, START_BLOCK(sbi
, start_segno
),
1731 sbi
->segs_per_sec
<< sbi
->log_blocks_per_seg
);
1733 start
= start_segno
+ sbi
->segs_per_sec
;
1739 mutex_unlock(&dirty_i
->seglist_lock
);
1741 /* send small discards */
1742 list_for_each_entry_safe(entry
, this, head
, list
) {
1743 unsigned int cur_pos
= 0, next_pos
, len
, total_len
= 0;
1744 bool is_valid
= test_bit_le(0, entry
->discard_map
);
1748 next_pos
= find_next_zero_bit_le(entry
->discard_map
,
1749 sbi
->blocks_per_seg
, cur_pos
);
1750 len
= next_pos
- cur_pos
;
1752 if (f2fs_sb_mounted_blkzoned(sbi
->sb
) ||
1753 (force
&& len
< cpc
->trim_minlen
))
1756 f2fs_issue_discard(sbi
, entry
->start_blkaddr
+ cur_pos
,
1760 next_pos
= find_next_bit_le(entry
->discard_map
,
1761 sbi
->blocks_per_seg
, cur_pos
);
1765 is_valid
= !is_valid
;
1767 if (cur_pos
< sbi
->blocks_per_seg
)
1770 list_del(&entry
->list
);
1771 dcc
->nr_discards
-= total_len
;
1772 kmem_cache_free(discard_entry_slab
, entry
);
1775 wake_up_discard_thread(sbi
, false);
1778 void init_discard_policy(struct discard_policy
*dpolicy
,
1779 int discard_type
, unsigned int granularity
)
1782 dpolicy
->type
= discard_type
;
1783 dpolicy
->sync
= true;
1784 dpolicy
->granularity
= granularity
;
1786 if (discard_type
== DPOLICY_BG
) {
1787 dpolicy
->min_interval
= DEF_MIN_DISCARD_ISSUE_TIME
;
1788 dpolicy
->max_interval
= DEF_MAX_DISCARD_ISSUE_TIME
;
1789 dpolicy
->max_requests
= DEF_MAX_DISCARD_REQUEST
;
1790 dpolicy
->io_aware_gran
= MAX_PLIST_NUM
;
1791 dpolicy
->io_aware
= true;
1792 } else if (discard_type
== DPOLICY_FORCE
) {
1793 dpolicy
->min_interval
= DEF_MIN_DISCARD_ISSUE_TIME
;
1794 dpolicy
->max_interval
= DEF_MAX_DISCARD_ISSUE_TIME
;
1795 dpolicy
->max_requests
= DEF_MAX_DISCARD_REQUEST
;
1796 dpolicy
->io_aware_gran
= MAX_PLIST_NUM
;
1797 dpolicy
->io_aware
= true;
1798 } else if (discard_type
== DPOLICY_FSTRIM
) {
1799 dpolicy
->max_requests
= DEF_MAX_DISCARD_REQUEST
;
1800 dpolicy
->io_aware_gran
= MAX_PLIST_NUM
;
1801 dpolicy
->io_aware
= false;
1802 } else if (discard_type
== DPOLICY_UMOUNT
) {
1803 dpolicy
->max_requests
= DEF_MAX_DISCARD_REQUEST
;
1804 dpolicy
->io_aware_gran
= MAX_PLIST_NUM
;
1805 dpolicy
->io_aware
= false;
1809 static int create_discard_cmd_control(struct f2fs_sb_info
*sbi
)
1811 dev_t dev
= sbi
->sb
->s_bdev
->bd_dev
;
1812 struct discard_cmd_control
*dcc
;
1815 if (SM_I(sbi
)->dcc_info
) {
1816 dcc
= SM_I(sbi
)->dcc_info
;
1820 dcc
= kzalloc(sizeof(struct discard_cmd_control
), GFP_KERNEL
);
1824 dcc
->discard_granularity
= DEFAULT_DISCARD_GRANULARITY
;
1825 INIT_LIST_HEAD(&dcc
->entry_list
);
1826 for (i
= 0; i
< MAX_PLIST_NUM
; i
++)
1827 INIT_LIST_HEAD(&dcc
->pend_list
[i
]);
1828 INIT_LIST_HEAD(&dcc
->wait_list
);
1829 INIT_LIST_HEAD(&dcc
->fstrim_list
);
1830 mutex_init(&dcc
->cmd_lock
);
1831 atomic_set(&dcc
->issued_discard
, 0);
1832 atomic_set(&dcc
->issing_discard
, 0);
1833 atomic_set(&dcc
->discard_cmd_cnt
, 0);
1834 dcc
->nr_discards
= 0;
1835 dcc
->max_discards
= MAIN_SEGS(sbi
) << sbi
->log_blocks_per_seg
;
1836 dcc
->undiscard_blks
= 0;
1837 dcc
->root
= RB_ROOT
;
1839 init_waitqueue_head(&dcc
->discard_wait_queue
);
1840 SM_I(sbi
)->dcc_info
= dcc
;
1842 dcc
->f2fs_issue_discard
= kthread_run(issue_discard_thread
, sbi
,
1843 "f2fs_discard-%u:%u", MAJOR(dev
), MINOR(dev
));
1844 if (IS_ERR(dcc
->f2fs_issue_discard
)) {
1845 err
= PTR_ERR(dcc
->f2fs_issue_discard
);
1847 SM_I(sbi
)->dcc_info
= NULL
;
1854 static void destroy_discard_cmd_control(struct f2fs_sb_info
*sbi
)
1856 struct discard_cmd_control
*dcc
= SM_I(sbi
)->dcc_info
;
1861 stop_discard_thread(sbi
);
1864 SM_I(sbi
)->dcc_info
= NULL
;
1867 static bool __mark_sit_entry_dirty(struct f2fs_sb_info
*sbi
, unsigned int segno
)
1869 struct sit_info
*sit_i
= SIT_I(sbi
);
1871 if (!__test_and_set_bit(segno
, sit_i
->dirty_sentries_bitmap
)) {
1872 sit_i
->dirty_sentries
++;
1879 static void __set_sit_entry_type(struct f2fs_sb_info
*sbi
, int type
,
1880 unsigned int segno
, int modified
)
1882 struct seg_entry
*se
= get_seg_entry(sbi
, segno
);
1885 __mark_sit_entry_dirty(sbi
, segno
);
1888 static void update_sit_entry(struct f2fs_sb_info
*sbi
, block_t blkaddr
, int del
)
1890 struct seg_entry
*se
;
1891 unsigned int segno
, offset
;
1892 long int new_vblocks
;
1894 #ifdef CONFIG_F2FS_CHECK_FS
1898 segno
= GET_SEGNO(sbi
, blkaddr
);
1900 se
= get_seg_entry(sbi
, segno
);
1901 new_vblocks
= se
->valid_blocks
+ del
;
1902 offset
= GET_BLKOFF_FROM_SEG0(sbi
, blkaddr
);
1904 f2fs_bug_on(sbi
, (new_vblocks
>> (sizeof(unsigned short) << 3) ||
1905 (new_vblocks
> sbi
->blocks_per_seg
)));
1907 se
->valid_blocks
= new_vblocks
;
1908 se
->mtime
= get_mtime(sbi
);
1909 SIT_I(sbi
)->max_mtime
= se
->mtime
;
1911 /* Update valid block bitmap */
1913 exist
= f2fs_test_and_set_bit(offset
, se
->cur_valid_map
);
1914 #ifdef CONFIG_F2FS_CHECK_FS
1915 mir_exist
= f2fs_test_and_set_bit(offset
,
1916 se
->cur_valid_map_mir
);
1917 if (unlikely(exist
!= mir_exist
)) {
1918 f2fs_msg(sbi
->sb
, KERN_ERR
, "Inconsistent error "
1919 "when setting bitmap, blk:%u, old bit:%d",
1921 f2fs_bug_on(sbi
, 1);
1924 if (unlikely(exist
)) {
1925 f2fs_msg(sbi
->sb
, KERN_ERR
,
1926 "Bitmap was wrongly set, blk:%u", blkaddr
);
1927 f2fs_bug_on(sbi
, 1);
1932 if (f2fs_discard_en(sbi
) &&
1933 !f2fs_test_and_set_bit(offset
, se
->discard_map
))
1934 sbi
->discard_blks
--;
1936 /* don't overwrite by SSR to keep node chain */
1937 if (se
->type
== CURSEG_WARM_NODE
) {
1938 if (!f2fs_test_and_set_bit(offset
, se
->ckpt_valid_map
))
1939 se
->ckpt_valid_blocks
++;
1942 exist
= f2fs_test_and_clear_bit(offset
, se
->cur_valid_map
);
1943 #ifdef CONFIG_F2FS_CHECK_FS
1944 mir_exist
= f2fs_test_and_clear_bit(offset
,
1945 se
->cur_valid_map_mir
);
1946 if (unlikely(exist
!= mir_exist
)) {
1947 f2fs_msg(sbi
->sb
, KERN_ERR
, "Inconsistent error "
1948 "when clearing bitmap, blk:%u, old bit:%d",
1950 f2fs_bug_on(sbi
, 1);
1953 if (unlikely(!exist
)) {
1954 f2fs_msg(sbi
->sb
, KERN_ERR
,
1955 "Bitmap was wrongly cleared, blk:%u", blkaddr
);
1956 f2fs_bug_on(sbi
, 1);
1961 if (f2fs_discard_en(sbi
) &&
1962 f2fs_test_and_clear_bit(offset
, se
->discard_map
))
1963 sbi
->discard_blks
++;
1965 if (!f2fs_test_bit(offset
, se
->ckpt_valid_map
))
1966 se
->ckpt_valid_blocks
+= del
;
1968 __mark_sit_entry_dirty(sbi
, segno
);
1970 /* update total number of valid blocks to be written in ckpt area */
1971 SIT_I(sbi
)->written_valid_blocks
+= del
;
1973 if (sbi
->segs_per_sec
> 1)
1974 get_sec_entry(sbi
, segno
)->valid_blocks
+= del
;
1977 void invalidate_blocks(struct f2fs_sb_info
*sbi
, block_t addr
)
1979 unsigned int segno
= GET_SEGNO(sbi
, addr
);
1980 struct sit_info
*sit_i
= SIT_I(sbi
);
1982 f2fs_bug_on(sbi
, addr
== NULL_ADDR
);
1983 if (addr
== NEW_ADDR
)
1986 /* add it into sit main buffer */
1987 down_write(&sit_i
->sentry_lock
);
1989 update_sit_entry(sbi
, addr
, -1);
1991 /* add it into dirty seglist */
1992 locate_dirty_segment(sbi
, segno
);
1994 up_write(&sit_i
->sentry_lock
);
1997 bool is_checkpointed_data(struct f2fs_sb_info
*sbi
, block_t blkaddr
)
1999 struct sit_info
*sit_i
= SIT_I(sbi
);
2000 unsigned int segno
, offset
;
2001 struct seg_entry
*se
;
2004 if (blkaddr
== NEW_ADDR
|| blkaddr
== NULL_ADDR
)
2007 down_read(&sit_i
->sentry_lock
);
2009 segno
= GET_SEGNO(sbi
, blkaddr
);
2010 se
= get_seg_entry(sbi
, segno
);
2011 offset
= GET_BLKOFF_FROM_SEG0(sbi
, blkaddr
);
2013 if (f2fs_test_bit(offset
, se
->ckpt_valid_map
))
2016 up_read(&sit_i
->sentry_lock
);
2022 * This function should be resided under the curseg_mutex lock
2024 static void __add_sum_entry(struct f2fs_sb_info
*sbi
, int type
,
2025 struct f2fs_summary
*sum
)
2027 struct curseg_info
*curseg
= CURSEG_I(sbi
, type
);
2028 void *addr
= curseg
->sum_blk
;
2029 addr
+= curseg
->next_blkoff
* sizeof(struct f2fs_summary
);
2030 memcpy(addr
, sum
, sizeof(struct f2fs_summary
));
2034 * Calculate the number of current summary pages for writing
2036 int npages_for_summary_flush(struct f2fs_sb_info
*sbi
, bool for_ra
)
2038 int valid_sum_count
= 0;
2041 for (i
= CURSEG_HOT_DATA
; i
<= CURSEG_COLD_DATA
; i
++) {
2042 if (sbi
->ckpt
->alloc_type
[i
] == SSR
)
2043 valid_sum_count
+= sbi
->blocks_per_seg
;
2046 valid_sum_count
+= le16_to_cpu(
2047 F2FS_CKPT(sbi
)->cur_data_blkoff
[i
]);
2049 valid_sum_count
+= curseg_blkoff(sbi
, i
);
2053 sum_in_page
= (PAGE_SIZE
- 2 * SUM_JOURNAL_SIZE
-
2054 SUM_FOOTER_SIZE
) / SUMMARY_SIZE
;
2055 if (valid_sum_count
<= sum_in_page
)
2057 else if ((valid_sum_count
- sum_in_page
) <=
2058 (PAGE_SIZE
- SUM_FOOTER_SIZE
) / SUMMARY_SIZE
)
2064 * Caller should put this summary page
2066 struct page
*get_sum_page(struct f2fs_sb_info
*sbi
, unsigned int segno
)
2068 return get_meta_page(sbi
, GET_SUM_BLOCK(sbi
, segno
));
2071 void update_meta_page(struct f2fs_sb_info
*sbi
, void *src
, block_t blk_addr
)
2073 struct page
*page
= grab_meta_page(sbi
, blk_addr
);
2075 memcpy(page_address(page
), src
, PAGE_SIZE
);
2076 set_page_dirty(page
);
2077 f2fs_put_page(page
, 1);
2080 static void write_sum_page(struct f2fs_sb_info
*sbi
,
2081 struct f2fs_summary_block
*sum_blk
, block_t blk_addr
)
2083 update_meta_page(sbi
, (void *)sum_blk
, blk_addr
);
2086 static void write_current_sum_page(struct f2fs_sb_info
*sbi
,
2087 int type
, block_t blk_addr
)
2089 struct curseg_info
*curseg
= CURSEG_I(sbi
, type
);
2090 struct page
*page
= grab_meta_page(sbi
, blk_addr
);
2091 struct f2fs_summary_block
*src
= curseg
->sum_blk
;
2092 struct f2fs_summary_block
*dst
;
2094 dst
= (struct f2fs_summary_block
*)page_address(page
);
2096 mutex_lock(&curseg
->curseg_mutex
);
2098 down_read(&curseg
->journal_rwsem
);
2099 memcpy(&dst
->journal
, curseg
->journal
, SUM_JOURNAL_SIZE
);
2100 up_read(&curseg
->journal_rwsem
);
2102 memcpy(dst
->entries
, src
->entries
, SUM_ENTRY_SIZE
);
2103 memcpy(&dst
->footer
, &src
->footer
, SUM_FOOTER_SIZE
);
2105 mutex_unlock(&curseg
->curseg_mutex
);
2107 set_page_dirty(page
);
2108 f2fs_put_page(page
, 1);
2111 static int is_next_segment_free(struct f2fs_sb_info
*sbi
, int type
)
2113 struct curseg_info
*curseg
= CURSEG_I(sbi
, type
);
2114 unsigned int segno
= curseg
->segno
+ 1;
2115 struct free_segmap_info
*free_i
= FREE_I(sbi
);
2117 if (segno
< MAIN_SEGS(sbi
) && segno
% sbi
->segs_per_sec
)
2118 return !test_bit(segno
, free_i
->free_segmap
);
2123 * Find a new segment from the free segments bitmap to right order
2124 * This function should be returned with success, otherwise BUG
2126 static void get_new_segment(struct f2fs_sb_info
*sbi
,
2127 unsigned int *newseg
, bool new_sec
, int dir
)
2129 struct free_segmap_info
*free_i
= FREE_I(sbi
);
2130 unsigned int segno
, secno
, zoneno
;
2131 unsigned int total_zones
= MAIN_SECS(sbi
) / sbi
->secs_per_zone
;
2132 unsigned int hint
= GET_SEC_FROM_SEG(sbi
, *newseg
);
2133 unsigned int old_zoneno
= GET_ZONE_FROM_SEG(sbi
, *newseg
);
2134 unsigned int left_start
= hint
;
2139 spin_lock(&free_i
->segmap_lock
);
2141 if (!new_sec
&& ((*newseg
+ 1) % sbi
->segs_per_sec
)) {
2142 segno
= find_next_zero_bit(free_i
->free_segmap
,
2143 GET_SEG_FROM_SEC(sbi
, hint
+ 1), *newseg
+ 1);
2144 if (segno
< GET_SEG_FROM_SEC(sbi
, hint
+ 1))
2148 secno
= find_next_zero_bit(free_i
->free_secmap
, MAIN_SECS(sbi
), hint
);
2149 if (secno
>= MAIN_SECS(sbi
)) {
2150 if (dir
== ALLOC_RIGHT
) {
2151 secno
= find_next_zero_bit(free_i
->free_secmap
,
2153 f2fs_bug_on(sbi
, secno
>= MAIN_SECS(sbi
));
2156 left_start
= hint
- 1;
2162 while (test_bit(left_start
, free_i
->free_secmap
)) {
2163 if (left_start
> 0) {
2167 left_start
= find_next_zero_bit(free_i
->free_secmap
,
2169 f2fs_bug_on(sbi
, left_start
>= MAIN_SECS(sbi
));
2174 segno
= GET_SEG_FROM_SEC(sbi
, secno
);
2175 zoneno
= GET_ZONE_FROM_SEC(sbi
, secno
);
2177 /* give up on finding another zone */
2180 if (sbi
->secs_per_zone
== 1)
2182 if (zoneno
== old_zoneno
)
2184 if (dir
== ALLOC_LEFT
) {
2185 if (!go_left
&& zoneno
+ 1 >= total_zones
)
2187 if (go_left
&& zoneno
== 0)
2190 for (i
= 0; i
< NR_CURSEG_TYPE
; i
++)
2191 if (CURSEG_I(sbi
, i
)->zone
== zoneno
)
2194 if (i
< NR_CURSEG_TYPE
) {
2195 /* zone is in user, try another */
2197 hint
= zoneno
* sbi
->secs_per_zone
- 1;
2198 else if (zoneno
+ 1 >= total_zones
)
2201 hint
= (zoneno
+ 1) * sbi
->secs_per_zone
;
2203 goto find_other_zone
;
2206 /* set it as dirty segment in free segmap */
2207 f2fs_bug_on(sbi
, test_bit(segno
, free_i
->free_segmap
));
2208 __set_inuse(sbi
, segno
);
2210 spin_unlock(&free_i
->segmap_lock
);
2213 static void reset_curseg(struct f2fs_sb_info
*sbi
, int type
, int modified
)
2215 struct curseg_info
*curseg
= CURSEG_I(sbi
, type
);
2216 struct summary_footer
*sum_footer
;
2218 curseg
->segno
= curseg
->next_segno
;
2219 curseg
->zone
= GET_ZONE_FROM_SEG(sbi
, curseg
->segno
);
2220 curseg
->next_blkoff
= 0;
2221 curseg
->next_segno
= NULL_SEGNO
;
2223 sum_footer
= &(curseg
->sum_blk
->footer
);
2224 memset(sum_footer
, 0, sizeof(struct summary_footer
));
2225 if (IS_DATASEG(type
))
2226 SET_SUM_TYPE(sum_footer
, SUM_TYPE_DATA
);
2227 if (IS_NODESEG(type
))
2228 SET_SUM_TYPE(sum_footer
, SUM_TYPE_NODE
);
2229 __set_sit_entry_type(sbi
, type
, curseg
->segno
, modified
);
2232 static unsigned int __get_next_segno(struct f2fs_sb_info
*sbi
, int type
)
2234 /* if segs_per_sec is large than 1, we need to keep original policy. */
2235 if (sbi
->segs_per_sec
!= 1)
2236 return CURSEG_I(sbi
, type
)->segno
;
2238 if (type
== CURSEG_HOT_DATA
|| IS_NODESEG(type
))
2241 if (SIT_I(sbi
)->last_victim
[ALLOC_NEXT
])
2242 return SIT_I(sbi
)->last_victim
[ALLOC_NEXT
];
2243 return CURSEG_I(sbi
, type
)->segno
;
2247 * Allocate a current working segment.
2248 * This function always allocates a free segment in LFS manner.
2250 static void new_curseg(struct f2fs_sb_info
*sbi
, int type
, bool new_sec
)
2252 struct curseg_info
*curseg
= CURSEG_I(sbi
, type
);
2253 unsigned int segno
= curseg
->segno
;
2254 int dir
= ALLOC_LEFT
;
2256 write_sum_page(sbi
, curseg
->sum_blk
,
2257 GET_SUM_BLOCK(sbi
, segno
));
2258 if (type
== CURSEG_WARM_DATA
|| type
== CURSEG_COLD_DATA
)
2261 if (test_opt(sbi
, NOHEAP
))
2264 segno
= __get_next_segno(sbi
, type
);
2265 get_new_segment(sbi
, &segno
, new_sec
, dir
);
2266 curseg
->next_segno
= segno
;
2267 reset_curseg(sbi
, type
, 1);
2268 curseg
->alloc_type
= LFS
;
2271 static void __next_free_blkoff(struct f2fs_sb_info
*sbi
,
2272 struct curseg_info
*seg
, block_t start
)
2274 struct seg_entry
*se
= get_seg_entry(sbi
, seg
->segno
);
2275 int entries
= SIT_VBLOCK_MAP_SIZE
/ sizeof(unsigned long);
2276 unsigned long *target_map
= SIT_I(sbi
)->tmp_map
;
2277 unsigned long *ckpt_map
= (unsigned long *)se
->ckpt_valid_map
;
2278 unsigned long *cur_map
= (unsigned long *)se
->cur_valid_map
;
2281 for (i
= 0; i
< entries
; i
++)
2282 target_map
[i
] = ckpt_map
[i
] | cur_map
[i
];
2284 pos
= __find_rev_next_zero_bit(target_map
, sbi
->blocks_per_seg
, start
);
2286 seg
->next_blkoff
= pos
;
2290 * If a segment is written by LFS manner, next block offset is just obtained
2291 * by increasing the current block offset. However, if a segment is written by
2292 * SSR manner, next block offset obtained by calling __next_free_blkoff
2294 static void __refresh_next_blkoff(struct f2fs_sb_info
*sbi
,
2295 struct curseg_info
*seg
)
2297 if (seg
->alloc_type
== SSR
)
2298 __next_free_blkoff(sbi
, seg
, seg
->next_blkoff
+ 1);
2304 * This function always allocates a used segment(from dirty seglist) by SSR
2305 * manner, so it should recover the existing segment information of valid blocks
2307 static void change_curseg(struct f2fs_sb_info
*sbi
, int type
)
2309 struct dirty_seglist_info
*dirty_i
= DIRTY_I(sbi
);
2310 struct curseg_info
*curseg
= CURSEG_I(sbi
, type
);
2311 unsigned int new_segno
= curseg
->next_segno
;
2312 struct f2fs_summary_block
*sum_node
;
2313 struct page
*sum_page
;
2315 write_sum_page(sbi
, curseg
->sum_blk
,
2316 GET_SUM_BLOCK(sbi
, curseg
->segno
));
2317 __set_test_and_inuse(sbi
, new_segno
);
2319 mutex_lock(&dirty_i
->seglist_lock
);
2320 __remove_dirty_segment(sbi
, new_segno
, PRE
);
2321 __remove_dirty_segment(sbi
, new_segno
, DIRTY
);
2322 mutex_unlock(&dirty_i
->seglist_lock
);
2324 reset_curseg(sbi
, type
, 1);
2325 curseg
->alloc_type
= SSR
;
2326 __next_free_blkoff(sbi
, curseg
, 0);
2328 sum_page
= get_sum_page(sbi
, new_segno
);
2329 sum_node
= (struct f2fs_summary_block
*)page_address(sum_page
);
2330 memcpy(curseg
->sum_blk
, sum_node
, SUM_ENTRY_SIZE
);
2331 f2fs_put_page(sum_page
, 1);
2334 static int get_ssr_segment(struct f2fs_sb_info
*sbi
, int type
)
2336 struct curseg_info
*curseg
= CURSEG_I(sbi
, type
);
2337 const struct victim_selection
*v_ops
= DIRTY_I(sbi
)->v_ops
;
2338 unsigned segno
= NULL_SEGNO
;
2340 bool reversed
= false;
2342 /* need_SSR() already forces to do this */
2343 if (v_ops
->get_victim(sbi
, &segno
, BG_GC
, type
, SSR
)) {
2344 curseg
->next_segno
= segno
;
2348 /* For node segments, let's do SSR more intensively */
2349 if (IS_NODESEG(type
)) {
2350 if (type
>= CURSEG_WARM_NODE
) {
2352 i
= CURSEG_COLD_NODE
;
2354 i
= CURSEG_HOT_NODE
;
2356 cnt
= NR_CURSEG_NODE_TYPE
;
2358 if (type
>= CURSEG_WARM_DATA
) {
2360 i
= CURSEG_COLD_DATA
;
2362 i
= CURSEG_HOT_DATA
;
2364 cnt
= NR_CURSEG_DATA_TYPE
;
2367 for (; cnt
-- > 0; reversed
? i
-- : i
++) {
2370 if (v_ops
->get_victim(sbi
, &segno
, BG_GC
, i
, SSR
)) {
2371 curseg
->next_segno
= segno
;
2379 * flush out current segment and replace it with new segment
2380 * This function should be returned with success, otherwise BUG
2382 static void allocate_segment_by_default(struct f2fs_sb_info
*sbi
,
2383 int type
, bool force
)
2385 struct curseg_info
*curseg
= CURSEG_I(sbi
, type
);
2388 new_curseg(sbi
, type
, true);
2389 else if (!is_set_ckpt_flags(sbi
, CP_CRC_RECOVERY_FLAG
) &&
2390 type
== CURSEG_WARM_NODE
)
2391 new_curseg(sbi
, type
, false);
2392 else if (curseg
->alloc_type
== LFS
&& is_next_segment_free(sbi
, type
))
2393 new_curseg(sbi
, type
, false);
2394 else if (need_SSR(sbi
) && get_ssr_segment(sbi
, type
))
2395 change_curseg(sbi
, type
);
2397 new_curseg(sbi
, type
, false);
2399 stat_inc_seg_type(sbi
, curseg
);
2402 void allocate_new_segments(struct f2fs_sb_info
*sbi
)
2404 struct curseg_info
*curseg
;
2405 unsigned int old_segno
;
2408 down_write(&SIT_I(sbi
)->sentry_lock
);
2410 for (i
= CURSEG_HOT_DATA
; i
<= CURSEG_COLD_DATA
; i
++) {
2411 curseg
= CURSEG_I(sbi
, i
);
2412 old_segno
= curseg
->segno
;
2413 SIT_I(sbi
)->s_ops
->allocate_segment(sbi
, i
, true);
2414 locate_dirty_segment(sbi
, old_segno
);
2417 up_write(&SIT_I(sbi
)->sentry_lock
);
2420 static const struct segment_allocation default_salloc_ops
= {
2421 .allocate_segment
= allocate_segment_by_default
,
2424 bool exist_trim_candidates(struct f2fs_sb_info
*sbi
, struct cp_control
*cpc
)
2426 __u64 trim_start
= cpc
->trim_start
;
2427 bool has_candidate
= false;
2429 down_write(&SIT_I(sbi
)->sentry_lock
);
2430 for (; cpc
->trim_start
<= cpc
->trim_end
; cpc
->trim_start
++) {
2431 if (add_discard_addrs(sbi
, cpc
, true)) {
2432 has_candidate
= true;
2436 up_write(&SIT_I(sbi
)->sentry_lock
);
2438 cpc
->trim_start
= trim_start
;
2439 return has_candidate
;
2442 int f2fs_trim_fs(struct f2fs_sb_info
*sbi
, struct fstrim_range
*range
)
2444 __u64 start
= F2FS_BYTES_TO_BLK(range
->start
);
2445 __u64 end
= start
+ F2FS_BYTES_TO_BLK(range
->len
) - 1;
2446 unsigned int start_segno
, end_segno
, cur_segno
;
2447 block_t start_block
, end_block
;
2448 struct cp_control cpc
;
2449 struct discard_policy dpolicy
;
2450 unsigned long long trimmed
= 0;
2453 if (start
>= MAX_BLKADDR(sbi
) || range
->len
< sbi
->blocksize
)
2456 if (end
<= MAIN_BLKADDR(sbi
))
2459 if (is_sbi_flag_set(sbi
, SBI_NEED_FSCK
)) {
2460 f2fs_msg(sbi
->sb
, KERN_WARNING
,
2461 "Found FS corruption, run fsck to fix.");
2465 /* start/end segment number in main_area */
2466 start_segno
= (start
<= MAIN_BLKADDR(sbi
)) ? 0 : GET_SEGNO(sbi
, start
);
2467 end_segno
= (end
>= MAX_BLKADDR(sbi
)) ? MAIN_SEGS(sbi
) - 1 :
2468 GET_SEGNO(sbi
, end
);
2470 cpc
.reason
= CP_DISCARD
;
2471 cpc
.trim_minlen
= max_t(__u64
, 1, F2FS_BYTES_TO_BLK(range
->minlen
));
2473 /* do checkpoint to issue discard commands safely */
2474 for (cur_segno
= start_segno
; cur_segno
<= end_segno
;
2475 cur_segno
= cpc
.trim_end
+ 1) {
2476 cpc
.trim_start
= cur_segno
;
2478 if (sbi
->discard_blks
== 0)
2480 else if (sbi
->discard_blks
< BATCHED_TRIM_BLOCKS(sbi
))
2481 cpc
.trim_end
= end_segno
;
2483 cpc
.trim_end
= min_t(unsigned int,
2484 rounddown(cur_segno
+
2485 BATCHED_TRIM_SEGMENTS(sbi
),
2486 sbi
->segs_per_sec
) - 1, end_segno
);
2488 mutex_lock(&sbi
->gc_mutex
);
2489 err
= write_checkpoint(sbi
, &cpc
);
2490 mutex_unlock(&sbi
->gc_mutex
);
2497 start_block
= START_BLOCK(sbi
, start_segno
);
2498 end_block
= START_BLOCK(sbi
, min(cur_segno
, end_segno
) + 1);
2500 init_discard_policy(&dpolicy
, DPOLICY_FSTRIM
, cpc
.trim_minlen
);
2501 __issue_discard_cmd_range(sbi
, &dpolicy
, start_block
, end_block
);
2502 trimmed
= __wait_discard_cmd_range(sbi
, &dpolicy
,
2503 start_block
, end_block
);
2505 range
->len
= F2FS_BLK_TO_BYTES(trimmed
);
2509 static bool __has_curseg_space(struct f2fs_sb_info
*sbi
, int type
)
2511 struct curseg_info
*curseg
= CURSEG_I(sbi
, type
);
2512 if (curseg
->next_blkoff
< sbi
->blocks_per_seg
)
2518 int rw_hint_to_seg_type(enum rw_hint hint
)
2521 case WRITE_LIFE_SHORT
:
2522 return CURSEG_HOT_DATA
;
2523 case WRITE_LIFE_EXTREME
:
2524 return CURSEG_COLD_DATA
;
2526 return CURSEG_WARM_DATA
;
2531 static int __get_segment_type_2(struct f2fs_io_info
*fio
)
2533 if (fio
->type
== DATA
)
2534 return CURSEG_HOT_DATA
;
2536 return CURSEG_HOT_NODE
;
2539 static int __get_segment_type_4(struct f2fs_io_info
*fio
)
2541 if (fio
->type
== DATA
) {
2542 struct inode
*inode
= fio
->page
->mapping
->host
;
2544 if (S_ISDIR(inode
->i_mode
))
2545 return CURSEG_HOT_DATA
;
2547 return CURSEG_COLD_DATA
;
2549 if (IS_DNODE(fio
->page
) && is_cold_node(fio
->page
))
2550 return CURSEG_WARM_NODE
;
2552 return CURSEG_COLD_NODE
;
2556 static int __get_segment_type_6(struct f2fs_io_info
*fio
)
2558 if (fio
->type
== DATA
) {
2559 struct inode
*inode
= fio
->page
->mapping
->host
;
2561 if (is_cold_data(fio
->page
) || file_is_cold(inode
))
2562 return CURSEG_COLD_DATA
;
2563 if (is_inode_flag_set(inode
, FI_HOT_DATA
))
2564 return CURSEG_HOT_DATA
;
2565 /* rw_hint_to_seg_type(inode->i_write_hint); */
2566 return CURSEG_WARM_DATA
;
2568 if (IS_DNODE(fio
->page
))
2569 return is_cold_node(fio
->page
) ? CURSEG_WARM_NODE
:
2571 return CURSEG_COLD_NODE
;
2575 static int __get_segment_type(struct f2fs_io_info
*fio
)
2579 switch (fio
->sbi
->active_logs
) {
2581 type
= __get_segment_type_2(fio
);
2584 type
= __get_segment_type_4(fio
);
2587 type
= __get_segment_type_6(fio
);
2590 f2fs_bug_on(fio
->sbi
, true);
2595 else if (IS_WARM(type
))
2602 void allocate_data_block(struct f2fs_sb_info
*sbi
, struct page
*page
,
2603 block_t old_blkaddr
, block_t
*new_blkaddr
,
2604 struct f2fs_summary
*sum
, int type
,
2605 struct f2fs_io_info
*fio
, bool add_list
)
2607 struct sit_info
*sit_i
= SIT_I(sbi
);
2608 struct curseg_info
*curseg
= CURSEG_I(sbi
, type
);
2610 down_read(&SM_I(sbi
)->curseg_lock
);
2612 mutex_lock(&curseg
->curseg_mutex
);
2613 down_write(&sit_i
->sentry_lock
);
2615 *new_blkaddr
= NEXT_FREE_BLKADDR(sbi
, curseg
);
2617 f2fs_wait_discard_bio(sbi
, *new_blkaddr
);
2620 * __add_sum_entry should be resided under the curseg_mutex
2621 * because, this function updates a summary entry in the
2622 * current summary block.
2624 __add_sum_entry(sbi
, type
, sum
);
2626 __refresh_next_blkoff(sbi
, curseg
);
2628 stat_inc_block_count(sbi
, curseg
);
2631 * SIT information should be updated before segment allocation,
2632 * since SSR needs latest valid block information.
2634 update_sit_entry(sbi
, *new_blkaddr
, 1);
2635 if (GET_SEGNO(sbi
, old_blkaddr
) != NULL_SEGNO
)
2636 update_sit_entry(sbi
, old_blkaddr
, -1);
2638 if (!__has_curseg_space(sbi
, type
))
2639 sit_i
->s_ops
->allocate_segment(sbi
, type
, false);
2642 * segment dirty status should be updated after segment allocation,
2643 * so we just need to update status only one time after previous
2644 * segment being closed.
2646 locate_dirty_segment(sbi
, GET_SEGNO(sbi
, old_blkaddr
));
2647 locate_dirty_segment(sbi
, GET_SEGNO(sbi
, *new_blkaddr
));
2649 up_write(&sit_i
->sentry_lock
);
2651 if (page
&& IS_NODESEG(type
)) {
2652 fill_node_footer_blkaddr(page
, NEXT_FREE_BLKADDR(sbi
, curseg
));
2654 f2fs_inode_chksum_set(sbi
, page
);
2658 struct f2fs_bio_info
*io
;
2660 INIT_LIST_HEAD(&fio
->list
);
2661 fio
->in_list
= true;
2662 io
= sbi
->write_io
[fio
->type
] + fio
->temp
;
2663 spin_lock(&io
->io_lock
);
2664 list_add_tail(&fio
->list
, &io
->io_list
);
2665 spin_unlock(&io
->io_lock
);
2668 mutex_unlock(&curseg
->curseg_mutex
);
2670 up_read(&SM_I(sbi
)->curseg_lock
);
2673 static void update_device_state(struct f2fs_io_info
*fio
)
2675 struct f2fs_sb_info
*sbi
= fio
->sbi
;
2676 unsigned int devidx
;
2681 devidx
= f2fs_target_device_index(sbi
, fio
->new_blkaddr
);
2683 /* update device state for fsync */
2684 set_dirty_device(sbi
, fio
->ino
, devidx
, FLUSH_INO
);
2686 /* update device state for checkpoint */
2687 if (!f2fs_test_bit(devidx
, (char *)&sbi
->dirty_device
)) {
2688 spin_lock(&sbi
->dev_lock
);
2689 f2fs_set_bit(devidx
, (char *)&sbi
->dirty_device
);
2690 spin_unlock(&sbi
->dev_lock
);
2694 static void do_write_page(struct f2fs_summary
*sum
, struct f2fs_io_info
*fio
)
2696 int type
= __get_segment_type(fio
);
2700 allocate_data_block(fio
->sbi
, fio
->page
, fio
->old_blkaddr
,
2701 &fio
->new_blkaddr
, sum
, type
, fio
, true);
2703 /* writeout dirty page into bdev */
2704 err
= f2fs_submit_page_write(fio
);
2705 if (err
== -EAGAIN
) {
2706 fio
->old_blkaddr
= fio
->new_blkaddr
;
2709 update_device_state(fio
);
2713 void write_meta_page(struct f2fs_sb_info
*sbi
, struct page
*page
,
2714 enum iostat_type io_type
)
2716 struct f2fs_io_info fio
= {
2720 .op_flags
= REQ_SYNC
| REQ_NOIDLE
| REQ_META
| REQ_PRIO
,
2721 .old_blkaddr
= page
->index
,
2722 .new_blkaddr
= page
->index
,
2724 .encrypted_page
= NULL
,
2728 if (unlikely(page
->index
>= MAIN_BLKADDR(sbi
)))
2729 fio
.op_flags
&= ~REQ_META
;
2731 set_page_writeback(page
);
2732 f2fs_submit_page_write(&fio
);
2734 f2fs_update_iostat(sbi
, io_type
, F2FS_BLKSIZE
);
2737 void write_node_page(unsigned int nid
, struct f2fs_io_info
*fio
)
2739 struct f2fs_summary sum
;
2741 set_summary(&sum
, nid
, 0, 0);
2742 do_write_page(&sum
, fio
);
2744 f2fs_update_iostat(fio
->sbi
, fio
->io_type
, F2FS_BLKSIZE
);
2747 void write_data_page(struct dnode_of_data
*dn
, struct f2fs_io_info
*fio
)
2749 struct f2fs_sb_info
*sbi
= fio
->sbi
;
2750 struct f2fs_summary sum
;
2751 struct node_info ni
;
2753 f2fs_bug_on(sbi
, dn
->data_blkaddr
== NULL_ADDR
);
2754 get_node_info(sbi
, dn
->nid
, &ni
);
2755 set_summary(&sum
, dn
->nid
, dn
->ofs_in_node
, ni
.version
);
2756 do_write_page(&sum
, fio
);
2757 f2fs_update_data_blkaddr(dn
, fio
->new_blkaddr
);
2759 f2fs_update_iostat(sbi
, fio
->io_type
, F2FS_BLKSIZE
);
2762 int rewrite_data_page(struct f2fs_io_info
*fio
)
2766 fio
->new_blkaddr
= fio
->old_blkaddr
;
2767 stat_inc_inplace_blocks(fio
->sbi
);
2769 err
= f2fs_submit_page_bio(fio
);
2771 update_device_state(fio
);
2773 f2fs_update_iostat(fio
->sbi
, fio
->io_type
, F2FS_BLKSIZE
);
2778 static inline int __f2fs_get_curseg(struct f2fs_sb_info
*sbi
,
2783 for (i
= CURSEG_HOT_DATA
; i
< NO_CHECK_TYPE
; i
++) {
2784 if (CURSEG_I(sbi
, i
)->segno
== segno
)
2790 void __f2fs_replace_block(struct f2fs_sb_info
*sbi
, struct f2fs_summary
*sum
,
2791 block_t old_blkaddr
, block_t new_blkaddr
,
2792 bool recover_curseg
, bool recover_newaddr
)
2794 struct sit_info
*sit_i
= SIT_I(sbi
);
2795 struct curseg_info
*curseg
;
2796 unsigned int segno
, old_cursegno
;
2797 struct seg_entry
*se
;
2799 unsigned short old_blkoff
;
2801 segno
= GET_SEGNO(sbi
, new_blkaddr
);
2802 se
= get_seg_entry(sbi
, segno
);
2805 down_write(&SM_I(sbi
)->curseg_lock
);
2807 if (!recover_curseg
) {
2808 /* for recovery flow */
2809 if (se
->valid_blocks
== 0 && !IS_CURSEG(sbi
, segno
)) {
2810 if (old_blkaddr
== NULL_ADDR
)
2811 type
= CURSEG_COLD_DATA
;
2813 type
= CURSEG_WARM_DATA
;
2816 if (IS_CURSEG(sbi
, segno
)) {
2817 /* se->type is volatile as SSR allocation */
2818 type
= __f2fs_get_curseg(sbi
, segno
);
2819 f2fs_bug_on(sbi
, type
== NO_CHECK_TYPE
);
2821 type
= CURSEG_WARM_DATA
;
2825 curseg
= CURSEG_I(sbi
, type
);
2827 mutex_lock(&curseg
->curseg_mutex
);
2828 down_write(&sit_i
->sentry_lock
);
2830 old_cursegno
= curseg
->segno
;
2831 old_blkoff
= curseg
->next_blkoff
;
2833 /* change the current segment */
2834 if (segno
!= curseg
->segno
) {
2835 curseg
->next_segno
= segno
;
2836 change_curseg(sbi
, type
);
2839 curseg
->next_blkoff
= GET_BLKOFF_FROM_SEG0(sbi
, new_blkaddr
);
2840 __add_sum_entry(sbi
, type
, sum
);
2842 if (!recover_curseg
|| recover_newaddr
)
2843 update_sit_entry(sbi
, new_blkaddr
, 1);
2844 if (GET_SEGNO(sbi
, old_blkaddr
) != NULL_SEGNO
)
2845 update_sit_entry(sbi
, old_blkaddr
, -1);
2847 locate_dirty_segment(sbi
, GET_SEGNO(sbi
, old_blkaddr
));
2848 locate_dirty_segment(sbi
, GET_SEGNO(sbi
, new_blkaddr
));
2850 locate_dirty_segment(sbi
, old_cursegno
);
2852 if (recover_curseg
) {
2853 if (old_cursegno
!= curseg
->segno
) {
2854 curseg
->next_segno
= old_cursegno
;
2855 change_curseg(sbi
, type
);
2857 curseg
->next_blkoff
= old_blkoff
;
2860 up_write(&sit_i
->sentry_lock
);
2861 mutex_unlock(&curseg
->curseg_mutex
);
2862 up_write(&SM_I(sbi
)->curseg_lock
);
2865 void f2fs_replace_block(struct f2fs_sb_info
*sbi
, struct dnode_of_data
*dn
,
2866 block_t old_addr
, block_t new_addr
,
2867 unsigned char version
, bool recover_curseg
,
2868 bool recover_newaddr
)
2870 struct f2fs_summary sum
;
2872 set_summary(&sum
, dn
->nid
, dn
->ofs_in_node
, version
);
2874 __f2fs_replace_block(sbi
, &sum
, old_addr
, new_addr
,
2875 recover_curseg
, recover_newaddr
);
2877 f2fs_update_data_blkaddr(dn
, new_addr
);
2880 void f2fs_wait_on_page_writeback(struct page
*page
,
2881 enum page_type type
, bool ordered
)
2883 if (PageWriteback(page
)) {
2884 struct f2fs_sb_info
*sbi
= F2FS_P_SB(page
);
2886 f2fs_submit_merged_write_cond(sbi
, page
->mapping
->host
,
2887 0, page
->index
, type
);
2889 wait_on_page_writeback(page
);
2891 wait_for_stable_page(page
);
2895 void f2fs_wait_on_block_writeback(struct f2fs_sb_info
*sbi
, block_t blkaddr
)
2899 if (blkaddr
== NEW_ADDR
|| blkaddr
== NULL_ADDR
)
2902 cpage
= find_lock_page(META_MAPPING(sbi
), blkaddr
);
2904 f2fs_wait_on_page_writeback(cpage
, DATA
, true);
2905 f2fs_put_page(cpage
, 1);
2909 static int read_compacted_summaries(struct f2fs_sb_info
*sbi
)
2911 struct f2fs_checkpoint
*ckpt
= F2FS_CKPT(sbi
);
2912 struct curseg_info
*seg_i
;
2913 unsigned char *kaddr
;
2918 start
= start_sum_block(sbi
);
2920 page
= get_meta_page(sbi
, start
++);
2921 kaddr
= (unsigned char *)page_address(page
);
2923 /* Step 1: restore nat cache */
2924 seg_i
= CURSEG_I(sbi
, CURSEG_HOT_DATA
);
2925 memcpy(seg_i
->journal
, kaddr
, SUM_JOURNAL_SIZE
);
2927 /* Step 2: restore sit cache */
2928 seg_i
= CURSEG_I(sbi
, CURSEG_COLD_DATA
);
2929 memcpy(seg_i
->journal
, kaddr
+ SUM_JOURNAL_SIZE
, SUM_JOURNAL_SIZE
);
2930 offset
= 2 * SUM_JOURNAL_SIZE
;
2932 /* Step 3: restore summary entries */
2933 for (i
= CURSEG_HOT_DATA
; i
<= CURSEG_COLD_DATA
; i
++) {
2934 unsigned short blk_off
;
2937 seg_i
= CURSEG_I(sbi
, i
);
2938 segno
= le32_to_cpu(ckpt
->cur_data_segno
[i
]);
2939 blk_off
= le16_to_cpu(ckpt
->cur_data_blkoff
[i
]);
2940 seg_i
->next_segno
= segno
;
2941 reset_curseg(sbi
, i
, 0);
2942 seg_i
->alloc_type
= ckpt
->alloc_type
[i
];
2943 seg_i
->next_blkoff
= blk_off
;
2945 if (seg_i
->alloc_type
== SSR
)
2946 blk_off
= sbi
->blocks_per_seg
;
2948 for (j
= 0; j
< blk_off
; j
++) {
2949 struct f2fs_summary
*s
;
2950 s
= (struct f2fs_summary
*)(kaddr
+ offset
);
2951 seg_i
->sum_blk
->entries
[j
] = *s
;
2952 offset
+= SUMMARY_SIZE
;
2953 if (offset
+ SUMMARY_SIZE
<= PAGE_SIZE
-
2957 f2fs_put_page(page
, 1);
2960 page
= get_meta_page(sbi
, start
++);
2961 kaddr
= (unsigned char *)page_address(page
);
2965 f2fs_put_page(page
, 1);
2969 static int read_normal_summaries(struct f2fs_sb_info
*sbi
, int type
)
2971 struct f2fs_checkpoint
*ckpt
= F2FS_CKPT(sbi
);
2972 struct f2fs_summary_block
*sum
;
2973 struct curseg_info
*curseg
;
2975 unsigned short blk_off
;
2976 unsigned int segno
= 0;
2977 block_t blk_addr
= 0;
2979 /* get segment number and block addr */
2980 if (IS_DATASEG(type
)) {
2981 segno
= le32_to_cpu(ckpt
->cur_data_segno
[type
]);
2982 blk_off
= le16_to_cpu(ckpt
->cur_data_blkoff
[type
-
2984 if (__exist_node_summaries(sbi
))
2985 blk_addr
= sum_blk_addr(sbi
, NR_CURSEG_TYPE
, type
);
2987 blk_addr
= sum_blk_addr(sbi
, NR_CURSEG_DATA_TYPE
, type
);
2989 segno
= le32_to_cpu(ckpt
->cur_node_segno
[type
-
2991 blk_off
= le16_to_cpu(ckpt
->cur_node_blkoff
[type
-
2993 if (__exist_node_summaries(sbi
))
2994 blk_addr
= sum_blk_addr(sbi
, NR_CURSEG_NODE_TYPE
,
2995 type
- CURSEG_HOT_NODE
);
2997 blk_addr
= GET_SUM_BLOCK(sbi
, segno
);
3000 new = get_meta_page(sbi
, blk_addr
);
3001 sum
= (struct f2fs_summary_block
*)page_address(new);
3003 if (IS_NODESEG(type
)) {
3004 if (__exist_node_summaries(sbi
)) {
3005 struct f2fs_summary
*ns
= &sum
->entries
[0];
3007 for (i
= 0; i
< sbi
->blocks_per_seg
; i
++, ns
++) {
3009 ns
->ofs_in_node
= 0;
3014 err
= restore_node_summary(sbi
, segno
, sum
);
3016 f2fs_put_page(new, 1);
3022 /* set uncompleted segment to curseg */
3023 curseg
= CURSEG_I(sbi
, type
);
3024 mutex_lock(&curseg
->curseg_mutex
);
3026 /* update journal info */
3027 down_write(&curseg
->journal_rwsem
);
3028 memcpy(curseg
->journal
, &sum
->journal
, SUM_JOURNAL_SIZE
);
3029 up_write(&curseg
->journal_rwsem
);
3031 memcpy(curseg
->sum_blk
->entries
, sum
->entries
, SUM_ENTRY_SIZE
);
3032 memcpy(&curseg
->sum_blk
->footer
, &sum
->footer
, SUM_FOOTER_SIZE
);
3033 curseg
->next_segno
= segno
;
3034 reset_curseg(sbi
, type
, 0);
3035 curseg
->alloc_type
= ckpt
->alloc_type
[type
];
3036 curseg
->next_blkoff
= blk_off
;
3037 mutex_unlock(&curseg
->curseg_mutex
);
3038 f2fs_put_page(new, 1);
3042 static int restore_curseg_summaries(struct f2fs_sb_info
*sbi
)
3044 struct f2fs_journal
*sit_j
= CURSEG_I(sbi
, CURSEG_COLD_DATA
)->journal
;
3045 struct f2fs_journal
*nat_j
= CURSEG_I(sbi
, CURSEG_HOT_DATA
)->journal
;
3046 int type
= CURSEG_HOT_DATA
;
3049 if (is_set_ckpt_flags(sbi
, CP_COMPACT_SUM_FLAG
)) {
3050 int npages
= npages_for_summary_flush(sbi
, true);
3053 ra_meta_pages(sbi
, start_sum_block(sbi
), npages
,
3056 /* restore for compacted data summary */
3057 if (read_compacted_summaries(sbi
))
3059 type
= CURSEG_HOT_NODE
;
3062 if (__exist_node_summaries(sbi
))
3063 ra_meta_pages(sbi
, sum_blk_addr(sbi
, NR_CURSEG_TYPE
, type
),
3064 NR_CURSEG_TYPE
- type
, META_CP
, true);
3066 for (; type
<= CURSEG_COLD_NODE
; type
++) {
3067 err
= read_normal_summaries(sbi
, type
);
3072 /* sanity check for summary blocks */
3073 if (nats_in_cursum(nat_j
) > NAT_JOURNAL_ENTRIES
||
3074 sits_in_cursum(sit_j
) > SIT_JOURNAL_ENTRIES
)
3080 static void write_compacted_summaries(struct f2fs_sb_info
*sbi
, block_t blkaddr
)
3083 unsigned char *kaddr
;
3084 struct f2fs_summary
*summary
;
3085 struct curseg_info
*seg_i
;
3086 int written_size
= 0;
3089 page
= grab_meta_page(sbi
, blkaddr
++);
3090 kaddr
= (unsigned char *)page_address(page
);
3092 /* Step 1: write nat cache */
3093 seg_i
= CURSEG_I(sbi
, CURSEG_HOT_DATA
);
3094 memcpy(kaddr
, seg_i
->journal
, SUM_JOURNAL_SIZE
);
3095 written_size
+= SUM_JOURNAL_SIZE
;
3097 /* Step 2: write sit cache */
3098 seg_i
= CURSEG_I(sbi
, CURSEG_COLD_DATA
);
3099 memcpy(kaddr
+ written_size
, seg_i
->journal
, SUM_JOURNAL_SIZE
);
3100 written_size
+= SUM_JOURNAL_SIZE
;
3102 /* Step 3: write summary entries */
3103 for (i
= CURSEG_HOT_DATA
; i
<= CURSEG_COLD_DATA
; i
++) {
3104 unsigned short blkoff
;
3105 seg_i
= CURSEG_I(sbi
, i
);
3106 if (sbi
->ckpt
->alloc_type
[i
] == SSR
)
3107 blkoff
= sbi
->blocks_per_seg
;
3109 blkoff
= curseg_blkoff(sbi
, i
);
3111 for (j
= 0; j
< blkoff
; j
++) {
3113 page
= grab_meta_page(sbi
, blkaddr
++);
3114 kaddr
= (unsigned char *)page_address(page
);
3117 summary
= (struct f2fs_summary
*)(kaddr
+ written_size
);
3118 *summary
= seg_i
->sum_blk
->entries
[j
];
3119 written_size
+= SUMMARY_SIZE
;
3121 if (written_size
+ SUMMARY_SIZE
<= PAGE_SIZE
-
3125 set_page_dirty(page
);
3126 f2fs_put_page(page
, 1);
3131 set_page_dirty(page
);
3132 f2fs_put_page(page
, 1);
3136 static void write_normal_summaries(struct f2fs_sb_info
*sbi
,
3137 block_t blkaddr
, int type
)
3140 if (IS_DATASEG(type
))
3141 end
= type
+ NR_CURSEG_DATA_TYPE
;
3143 end
= type
+ NR_CURSEG_NODE_TYPE
;
3145 for (i
= type
; i
< end
; i
++)
3146 write_current_sum_page(sbi
, i
, blkaddr
+ (i
- type
));
3149 void write_data_summaries(struct f2fs_sb_info
*sbi
, block_t start_blk
)
3151 if (is_set_ckpt_flags(sbi
, CP_COMPACT_SUM_FLAG
))
3152 write_compacted_summaries(sbi
, start_blk
);
3154 write_normal_summaries(sbi
, start_blk
, CURSEG_HOT_DATA
);
3157 void write_node_summaries(struct f2fs_sb_info
*sbi
, block_t start_blk
)
3159 write_normal_summaries(sbi
, start_blk
, CURSEG_HOT_NODE
);
3162 int lookup_journal_in_cursum(struct f2fs_journal
*journal
, int type
,
3163 unsigned int val
, int alloc
)
3167 if (type
== NAT_JOURNAL
) {
3168 for (i
= 0; i
< nats_in_cursum(journal
); i
++) {
3169 if (le32_to_cpu(nid_in_journal(journal
, i
)) == val
)
3172 if (alloc
&& __has_cursum_space(journal
, 1, NAT_JOURNAL
))
3173 return update_nats_in_cursum(journal
, 1);
3174 } else if (type
== SIT_JOURNAL
) {
3175 for (i
= 0; i
< sits_in_cursum(journal
); i
++)
3176 if (le32_to_cpu(segno_in_journal(journal
, i
)) == val
)
3178 if (alloc
&& __has_cursum_space(journal
, 1, SIT_JOURNAL
))
3179 return update_sits_in_cursum(journal
, 1);
3184 static struct page
*get_current_sit_page(struct f2fs_sb_info
*sbi
,
3187 return get_meta_page(sbi
, current_sit_addr(sbi
, segno
));
3190 static struct page
*get_next_sit_page(struct f2fs_sb_info
*sbi
,
3193 struct sit_info
*sit_i
= SIT_I(sbi
);
3194 struct page
*src_page
, *dst_page
;
3195 pgoff_t src_off
, dst_off
;
3196 void *src_addr
, *dst_addr
;
3198 src_off
= current_sit_addr(sbi
, start
);
3199 dst_off
= next_sit_addr(sbi
, src_off
);
3201 /* get current sit block page without lock */
3202 src_page
= get_meta_page(sbi
, src_off
);
3203 dst_page
= grab_meta_page(sbi
, dst_off
);
3204 f2fs_bug_on(sbi
, PageDirty(src_page
));
3206 src_addr
= page_address(src_page
);
3207 dst_addr
= page_address(dst_page
);
3208 memcpy(dst_addr
, src_addr
, PAGE_SIZE
);
3210 set_page_dirty(dst_page
);
3211 f2fs_put_page(src_page
, 1);
3213 set_to_next_sit(sit_i
, start
);
3218 static struct sit_entry_set
*grab_sit_entry_set(void)
3220 struct sit_entry_set
*ses
=
3221 f2fs_kmem_cache_alloc(sit_entry_set_slab
, GFP_NOFS
);
3224 INIT_LIST_HEAD(&ses
->set_list
);
3228 static void release_sit_entry_set(struct sit_entry_set
*ses
)
3230 list_del(&ses
->set_list
);
3231 kmem_cache_free(sit_entry_set_slab
, ses
);
3234 static void adjust_sit_entry_set(struct sit_entry_set
*ses
,
3235 struct list_head
*head
)
3237 struct sit_entry_set
*next
= ses
;
3239 if (list_is_last(&ses
->set_list
, head
))
3242 list_for_each_entry_continue(next
, head
, set_list
)
3243 if (ses
->entry_cnt
<= next
->entry_cnt
)
3246 list_move_tail(&ses
->set_list
, &next
->set_list
);
3249 static void add_sit_entry(unsigned int segno
, struct list_head
*head
)
3251 struct sit_entry_set
*ses
;
3252 unsigned int start_segno
= START_SEGNO(segno
);
3254 list_for_each_entry(ses
, head
, set_list
) {
3255 if (ses
->start_segno
== start_segno
) {
3257 adjust_sit_entry_set(ses
, head
);
3262 ses
= grab_sit_entry_set();
3264 ses
->start_segno
= start_segno
;
3266 list_add(&ses
->set_list
, head
);
3269 static void add_sits_in_set(struct f2fs_sb_info
*sbi
)
3271 struct f2fs_sm_info
*sm_info
= SM_I(sbi
);
3272 struct list_head
*set_list
= &sm_info
->sit_entry_set
;
3273 unsigned long *bitmap
= SIT_I(sbi
)->dirty_sentries_bitmap
;
3276 for_each_set_bit(segno
, bitmap
, MAIN_SEGS(sbi
))
3277 add_sit_entry(segno
, set_list
);
3280 static void remove_sits_in_journal(struct f2fs_sb_info
*sbi
)
3282 struct curseg_info
*curseg
= CURSEG_I(sbi
, CURSEG_COLD_DATA
);
3283 struct f2fs_journal
*journal
= curseg
->journal
;
3286 down_write(&curseg
->journal_rwsem
);
3287 for (i
= 0; i
< sits_in_cursum(journal
); i
++) {
3291 segno
= le32_to_cpu(segno_in_journal(journal
, i
));
3292 dirtied
= __mark_sit_entry_dirty(sbi
, segno
);
3295 add_sit_entry(segno
, &SM_I(sbi
)->sit_entry_set
);
3297 update_sits_in_cursum(journal
, -i
);
3298 up_write(&curseg
->journal_rwsem
);
3302 * CP calls this function, which flushes SIT entries including sit_journal,
3303 * and moves prefree segs to free segs.
3305 void flush_sit_entries(struct f2fs_sb_info
*sbi
, struct cp_control
*cpc
)
3307 struct sit_info
*sit_i
= SIT_I(sbi
);
3308 unsigned long *bitmap
= sit_i
->dirty_sentries_bitmap
;
3309 struct curseg_info
*curseg
= CURSEG_I(sbi
, CURSEG_COLD_DATA
);
3310 struct f2fs_journal
*journal
= curseg
->journal
;
3311 struct sit_entry_set
*ses
, *tmp
;
3312 struct list_head
*head
= &SM_I(sbi
)->sit_entry_set
;
3313 bool to_journal
= true;
3314 struct seg_entry
*se
;
3316 down_write(&sit_i
->sentry_lock
);
3318 if (!sit_i
->dirty_sentries
)
3322 * add and account sit entries of dirty bitmap in sit entry
3325 add_sits_in_set(sbi
);
3328 * if there are no enough space in journal to store dirty sit
3329 * entries, remove all entries from journal and add and account
3330 * them in sit entry set.
3332 if (!__has_cursum_space(journal
, sit_i
->dirty_sentries
, SIT_JOURNAL
))
3333 remove_sits_in_journal(sbi
);
3336 * there are two steps to flush sit entries:
3337 * #1, flush sit entries to journal in current cold data summary block.
3338 * #2, flush sit entries to sit page.
3340 list_for_each_entry_safe(ses
, tmp
, head
, set_list
) {
3341 struct page
*page
= NULL
;
3342 struct f2fs_sit_block
*raw_sit
= NULL
;
3343 unsigned int start_segno
= ses
->start_segno
;
3344 unsigned int end
= min(start_segno
+ SIT_ENTRY_PER_BLOCK
,
3345 (unsigned long)MAIN_SEGS(sbi
));
3346 unsigned int segno
= start_segno
;
3349 !__has_cursum_space(journal
, ses
->entry_cnt
, SIT_JOURNAL
))
3353 down_write(&curseg
->journal_rwsem
);
3355 page
= get_next_sit_page(sbi
, start_segno
);
3356 raw_sit
= page_address(page
);
3359 /* flush dirty sit entries in region of current sit set */
3360 for_each_set_bit_from(segno
, bitmap
, end
) {
3361 int offset
, sit_offset
;
3363 se
= get_seg_entry(sbi
, segno
);
3365 /* add discard candidates */
3366 if (!(cpc
->reason
& CP_DISCARD
)) {
3367 cpc
->trim_start
= segno
;
3368 add_discard_addrs(sbi
, cpc
, false);
3372 offset
= lookup_journal_in_cursum(journal
,
3373 SIT_JOURNAL
, segno
, 1);
3374 f2fs_bug_on(sbi
, offset
< 0);
3375 segno_in_journal(journal
, offset
) =
3377 seg_info_to_raw_sit(se
,
3378 &sit_in_journal(journal
, offset
));
3380 sit_offset
= SIT_ENTRY_OFFSET(sit_i
, segno
);
3381 seg_info_to_raw_sit(se
,
3382 &raw_sit
->entries
[sit_offset
]);
3385 __clear_bit(segno
, bitmap
);
3386 sit_i
->dirty_sentries
--;
3391 up_write(&curseg
->journal_rwsem
);
3393 f2fs_put_page(page
, 1);
3395 f2fs_bug_on(sbi
, ses
->entry_cnt
);
3396 release_sit_entry_set(ses
);
3399 f2fs_bug_on(sbi
, !list_empty(head
));
3400 f2fs_bug_on(sbi
, sit_i
->dirty_sentries
);
3402 if (cpc
->reason
& CP_DISCARD
) {
3403 __u64 trim_start
= cpc
->trim_start
;
3405 for (; cpc
->trim_start
<= cpc
->trim_end
; cpc
->trim_start
++)
3406 add_discard_addrs(sbi
, cpc
, false);
3408 cpc
->trim_start
= trim_start
;
3410 up_write(&sit_i
->sentry_lock
);
3412 set_prefree_as_free_segments(sbi
);
3415 static int build_sit_info(struct f2fs_sb_info
*sbi
)
3417 struct f2fs_super_block
*raw_super
= F2FS_RAW_SUPER(sbi
);
3418 struct sit_info
*sit_i
;
3419 unsigned int sit_segs
, start
;
3421 unsigned int bitmap_size
;
3423 /* allocate memory for SIT information */
3424 sit_i
= kzalloc(sizeof(struct sit_info
), GFP_KERNEL
);
3428 SM_I(sbi
)->sit_info
= sit_i
;
3430 sit_i
->sentries
= kvzalloc(MAIN_SEGS(sbi
) *
3431 sizeof(struct seg_entry
), GFP_KERNEL
);
3432 if (!sit_i
->sentries
)
3435 bitmap_size
= f2fs_bitmap_size(MAIN_SEGS(sbi
));
3436 sit_i
->dirty_sentries_bitmap
= kvzalloc(bitmap_size
, GFP_KERNEL
);
3437 if (!sit_i
->dirty_sentries_bitmap
)
3440 for (start
= 0; start
< MAIN_SEGS(sbi
); start
++) {
3441 sit_i
->sentries
[start
].cur_valid_map
3442 = kzalloc(SIT_VBLOCK_MAP_SIZE
, GFP_KERNEL
);
3443 sit_i
->sentries
[start
].ckpt_valid_map
3444 = kzalloc(SIT_VBLOCK_MAP_SIZE
, GFP_KERNEL
);
3445 if (!sit_i
->sentries
[start
].cur_valid_map
||
3446 !sit_i
->sentries
[start
].ckpt_valid_map
)
3449 #ifdef CONFIG_F2FS_CHECK_FS
3450 sit_i
->sentries
[start
].cur_valid_map_mir
3451 = kzalloc(SIT_VBLOCK_MAP_SIZE
, GFP_KERNEL
);
3452 if (!sit_i
->sentries
[start
].cur_valid_map_mir
)
3456 if (f2fs_discard_en(sbi
)) {
3457 sit_i
->sentries
[start
].discard_map
3458 = kzalloc(SIT_VBLOCK_MAP_SIZE
, GFP_KERNEL
);
3459 if (!sit_i
->sentries
[start
].discard_map
)
3464 sit_i
->tmp_map
= kzalloc(SIT_VBLOCK_MAP_SIZE
, GFP_KERNEL
);
3465 if (!sit_i
->tmp_map
)
3468 if (sbi
->segs_per_sec
> 1) {
3469 sit_i
->sec_entries
= kvzalloc(MAIN_SECS(sbi
) *
3470 sizeof(struct sec_entry
), GFP_KERNEL
);
3471 if (!sit_i
->sec_entries
)
3475 /* get information related with SIT */
3476 sit_segs
= le32_to_cpu(raw_super
->segment_count_sit
) >> 1;
3478 /* setup SIT bitmap from ckeckpoint pack */
3479 bitmap_size
= __bitmap_size(sbi
, SIT_BITMAP
);
3480 src_bitmap
= __bitmap_ptr(sbi
, SIT_BITMAP
);
3482 sit_i
->sit_bitmap
= kmemdup(src_bitmap
, bitmap_size
, GFP_KERNEL
);
3483 if (!sit_i
->sit_bitmap
)
3486 #ifdef CONFIG_F2FS_CHECK_FS
3487 sit_i
->sit_bitmap_mir
= kmemdup(src_bitmap
, bitmap_size
, GFP_KERNEL
);
3488 if (!sit_i
->sit_bitmap_mir
)
3492 /* init SIT information */
3493 sit_i
->s_ops
= &default_salloc_ops
;
3495 sit_i
->sit_base_addr
= le32_to_cpu(raw_super
->sit_blkaddr
);
3496 sit_i
->sit_blocks
= sit_segs
<< sbi
->log_blocks_per_seg
;
3497 sit_i
->written_valid_blocks
= 0;
3498 sit_i
->bitmap_size
= bitmap_size
;
3499 sit_i
->dirty_sentries
= 0;
3500 sit_i
->sents_per_block
= SIT_ENTRY_PER_BLOCK
;
3501 sit_i
->elapsed_time
= le64_to_cpu(sbi
->ckpt
->elapsed_time
);
3502 sit_i
->mounted_time
= CURRENT_TIME_SEC
.tv_sec
;
3503 init_rwsem(&sit_i
->sentry_lock
);
3507 static int build_free_segmap(struct f2fs_sb_info
*sbi
)
3509 struct free_segmap_info
*free_i
;
3510 unsigned int bitmap_size
, sec_bitmap_size
;
3512 /* allocate memory for free segmap information */
3513 free_i
= kzalloc(sizeof(struct free_segmap_info
), GFP_KERNEL
);
3517 SM_I(sbi
)->free_info
= free_i
;
3519 bitmap_size
= f2fs_bitmap_size(MAIN_SEGS(sbi
));
3520 free_i
->free_segmap
= kvmalloc(bitmap_size
, GFP_KERNEL
);
3521 if (!free_i
->free_segmap
)
3524 sec_bitmap_size
= f2fs_bitmap_size(MAIN_SECS(sbi
));
3525 free_i
->free_secmap
= kvmalloc(sec_bitmap_size
, GFP_KERNEL
);
3526 if (!free_i
->free_secmap
)
3529 /* set all segments as dirty temporarily */
3530 memset(free_i
->free_segmap
, 0xff, bitmap_size
);
3531 memset(free_i
->free_secmap
, 0xff, sec_bitmap_size
);
3533 /* init free segmap information */
3534 free_i
->start_segno
= GET_SEGNO_FROM_SEG0(sbi
, MAIN_BLKADDR(sbi
));
3535 free_i
->free_segments
= 0;
3536 free_i
->free_sections
= 0;
3537 spin_lock_init(&free_i
->segmap_lock
);
3541 static int build_curseg(struct f2fs_sb_info
*sbi
)
3543 struct curseg_info
*array
;
3546 array
= kcalloc(NR_CURSEG_TYPE
, sizeof(*array
), GFP_KERNEL
);
3550 SM_I(sbi
)->curseg_array
= array
;
3552 for (i
= 0; i
< NR_CURSEG_TYPE
; i
++) {
3553 mutex_init(&array
[i
].curseg_mutex
);
3554 array
[i
].sum_blk
= kzalloc(PAGE_SIZE
, GFP_KERNEL
);
3555 if (!array
[i
].sum_blk
)
3557 init_rwsem(&array
[i
].journal_rwsem
);
3558 array
[i
].journal
= kzalloc(sizeof(struct f2fs_journal
),
3560 if (!array
[i
].journal
)
3562 array
[i
].segno
= NULL_SEGNO
;
3563 array
[i
].next_blkoff
= 0;
3565 return restore_curseg_summaries(sbi
);
3568 static void build_sit_entries(struct f2fs_sb_info
*sbi
)
3570 struct sit_info
*sit_i
= SIT_I(sbi
);
3571 struct curseg_info
*curseg
= CURSEG_I(sbi
, CURSEG_COLD_DATA
);
3572 struct f2fs_journal
*journal
= curseg
->journal
;
3573 struct seg_entry
*se
;
3574 struct f2fs_sit_entry sit
;
3575 int sit_blk_cnt
= SIT_BLK_CNT(sbi
);
3576 unsigned int i
, start
, end
;
3577 unsigned int readed
, start_blk
= 0;
3580 readed
= ra_meta_pages(sbi
, start_blk
, BIO_MAX_PAGES
,
3583 start
= start_blk
* sit_i
->sents_per_block
;
3584 end
= (start_blk
+ readed
) * sit_i
->sents_per_block
;
3586 for (; start
< end
&& start
< MAIN_SEGS(sbi
); start
++) {
3587 struct f2fs_sit_block
*sit_blk
;
3590 se
= &sit_i
->sentries
[start
];
3591 page
= get_current_sit_page(sbi
, start
);
3592 sit_blk
= (struct f2fs_sit_block
*)page_address(page
);
3593 sit
= sit_blk
->entries
[SIT_ENTRY_OFFSET(sit_i
, start
)];
3594 f2fs_put_page(page
, 1);
3596 check_block_count(sbi
, start
, &sit
);
3597 seg_info_from_raw_sit(se
, &sit
);
3599 /* build discard map only one time */
3600 if (f2fs_discard_en(sbi
)) {
3601 if (is_set_ckpt_flags(sbi
, CP_TRIMMED_FLAG
)) {
3602 memset(se
->discard_map
, 0xff,
3603 SIT_VBLOCK_MAP_SIZE
);
3605 memcpy(se
->discard_map
,
3607 SIT_VBLOCK_MAP_SIZE
);
3608 sbi
->discard_blks
+=
3609 sbi
->blocks_per_seg
-
3614 if (sbi
->segs_per_sec
> 1)
3615 get_sec_entry(sbi
, start
)->valid_blocks
+=
3618 start_blk
+= readed
;
3619 } while (start_blk
< sit_blk_cnt
);
3621 down_read(&curseg
->journal_rwsem
);
3622 for (i
= 0; i
< sits_in_cursum(journal
); i
++) {
3623 unsigned int old_valid_blocks
;
3625 start
= le32_to_cpu(segno_in_journal(journal
, i
));
3626 se
= &sit_i
->sentries
[start
];
3627 sit
= sit_in_journal(journal
, i
);
3629 old_valid_blocks
= se
->valid_blocks
;
3631 check_block_count(sbi
, start
, &sit
);
3632 seg_info_from_raw_sit(se
, &sit
);
3634 if (f2fs_discard_en(sbi
)) {
3635 if (is_set_ckpt_flags(sbi
, CP_TRIMMED_FLAG
)) {
3636 memset(se
->discard_map
, 0xff,
3637 SIT_VBLOCK_MAP_SIZE
);
3639 memcpy(se
->discard_map
, se
->cur_valid_map
,
3640 SIT_VBLOCK_MAP_SIZE
);
3641 sbi
->discard_blks
+= old_valid_blocks
-
3646 if (sbi
->segs_per_sec
> 1)
3647 get_sec_entry(sbi
, start
)->valid_blocks
+=
3648 se
->valid_blocks
- old_valid_blocks
;
3650 up_read(&curseg
->journal_rwsem
);
3653 static void init_free_segmap(struct f2fs_sb_info
*sbi
)
3658 for (start
= 0; start
< MAIN_SEGS(sbi
); start
++) {
3659 struct seg_entry
*sentry
= get_seg_entry(sbi
, start
);
3660 if (!sentry
->valid_blocks
)
3661 __set_free(sbi
, start
);
3663 SIT_I(sbi
)->written_valid_blocks
+=
3664 sentry
->valid_blocks
;
3667 /* set use the current segments */
3668 for (type
= CURSEG_HOT_DATA
; type
<= CURSEG_COLD_NODE
; type
++) {
3669 struct curseg_info
*curseg_t
= CURSEG_I(sbi
, type
);
3670 __set_test_and_inuse(sbi
, curseg_t
->segno
);
3674 static void init_dirty_segmap(struct f2fs_sb_info
*sbi
)
3676 struct dirty_seglist_info
*dirty_i
= DIRTY_I(sbi
);
3677 struct free_segmap_info
*free_i
= FREE_I(sbi
);
3678 unsigned int segno
= 0, offset
= 0;
3679 unsigned short valid_blocks
;
3682 /* find dirty segment based on free segmap */
3683 segno
= find_next_inuse(free_i
, MAIN_SEGS(sbi
), offset
);
3684 if (segno
>= MAIN_SEGS(sbi
))
3687 valid_blocks
= get_valid_blocks(sbi
, segno
, false);
3688 if (valid_blocks
== sbi
->blocks_per_seg
|| !valid_blocks
)
3690 if (valid_blocks
> sbi
->blocks_per_seg
) {
3691 f2fs_bug_on(sbi
, 1);
3694 mutex_lock(&dirty_i
->seglist_lock
);
3695 __locate_dirty_segment(sbi
, segno
, DIRTY
);
3696 mutex_unlock(&dirty_i
->seglist_lock
);
3700 static int init_victim_secmap(struct f2fs_sb_info
*sbi
)
3702 struct dirty_seglist_info
*dirty_i
= DIRTY_I(sbi
);
3703 unsigned int bitmap_size
= f2fs_bitmap_size(MAIN_SECS(sbi
));
3705 dirty_i
->victim_secmap
= kvzalloc(bitmap_size
, GFP_KERNEL
);
3706 if (!dirty_i
->victim_secmap
)
3711 static int build_dirty_segmap(struct f2fs_sb_info
*sbi
)
3713 struct dirty_seglist_info
*dirty_i
;
3714 unsigned int bitmap_size
, i
;
3716 /* allocate memory for dirty segments list information */
3717 dirty_i
= kzalloc(sizeof(struct dirty_seglist_info
), GFP_KERNEL
);
3721 SM_I(sbi
)->dirty_info
= dirty_i
;
3722 mutex_init(&dirty_i
->seglist_lock
);
3724 bitmap_size
= f2fs_bitmap_size(MAIN_SEGS(sbi
));
3726 for (i
= 0; i
< NR_DIRTY_TYPE
; i
++) {
3727 dirty_i
->dirty_segmap
[i
] = kvzalloc(bitmap_size
, GFP_KERNEL
);
3728 if (!dirty_i
->dirty_segmap
[i
])
3732 init_dirty_segmap(sbi
);
3733 return init_victim_secmap(sbi
);
3737 * Update min, max modified time for cost-benefit GC algorithm
3739 static void init_min_max_mtime(struct f2fs_sb_info
*sbi
)
3741 struct sit_info
*sit_i
= SIT_I(sbi
);
3744 down_write(&sit_i
->sentry_lock
);
3746 sit_i
->min_mtime
= LLONG_MAX
;
3748 for (segno
= 0; segno
< MAIN_SEGS(sbi
); segno
+= sbi
->segs_per_sec
) {
3750 unsigned long long mtime
= 0;
3752 for (i
= 0; i
< sbi
->segs_per_sec
; i
++)
3753 mtime
+= get_seg_entry(sbi
, segno
+ i
)->mtime
;
3755 mtime
= div_u64(mtime
, sbi
->segs_per_sec
);
3757 if (sit_i
->min_mtime
> mtime
)
3758 sit_i
->min_mtime
= mtime
;
3760 sit_i
->max_mtime
= get_mtime(sbi
);
3761 up_write(&sit_i
->sentry_lock
);
3764 int build_segment_manager(struct f2fs_sb_info
*sbi
)
3766 struct f2fs_super_block
*raw_super
= F2FS_RAW_SUPER(sbi
);
3767 struct f2fs_checkpoint
*ckpt
= F2FS_CKPT(sbi
);
3768 struct f2fs_sm_info
*sm_info
;
3771 sm_info
= kzalloc(sizeof(struct f2fs_sm_info
), GFP_KERNEL
);
3776 sbi
->sm_info
= sm_info
;
3777 sm_info
->seg0_blkaddr
= le32_to_cpu(raw_super
->segment0_blkaddr
);
3778 sm_info
->main_blkaddr
= le32_to_cpu(raw_super
->main_blkaddr
);
3779 sm_info
->segment_count
= le32_to_cpu(raw_super
->segment_count
);
3780 sm_info
->reserved_segments
= le32_to_cpu(ckpt
->rsvd_segment_count
);
3781 sm_info
->ovp_segments
= le32_to_cpu(ckpt
->overprov_segment_count
);
3782 sm_info
->main_segments
= le32_to_cpu(raw_super
->segment_count_main
);
3783 sm_info
->ssa_blkaddr
= le32_to_cpu(raw_super
->ssa_blkaddr
);
3784 sm_info
->rec_prefree_segments
= sm_info
->main_segments
*
3785 DEF_RECLAIM_PREFREE_SEGMENTS
/ 100;
3786 if (sm_info
->rec_prefree_segments
> DEF_MAX_RECLAIM_PREFREE_SEGMENTS
)
3787 sm_info
->rec_prefree_segments
= DEF_MAX_RECLAIM_PREFREE_SEGMENTS
;
3789 if (!test_opt(sbi
, LFS
))
3790 sm_info
->ipu_policy
= 1 << F2FS_IPU_FSYNC
;
3791 sm_info
->min_ipu_util
= DEF_MIN_IPU_UTIL
;
3792 sm_info
->min_fsync_blocks
= DEF_MIN_FSYNC_BLOCKS
;
3793 sm_info
->min_hot_blocks
= DEF_MIN_HOT_BLOCKS
;
3794 sm_info
->min_ssr_sections
= reserved_sections(sbi
);
3796 sm_info
->trim_sections
= DEF_BATCHED_TRIM_SECTIONS
;
3798 INIT_LIST_HEAD(&sm_info
->sit_entry_set
);
3800 init_rwsem(&sm_info
->curseg_lock
);
3802 if (!f2fs_readonly(sbi
->sb
)) {
3803 err
= create_flush_cmd_control(sbi
);
3808 err
= create_discard_cmd_control(sbi
);
3812 err
= build_sit_info(sbi
);
3815 err
= build_free_segmap(sbi
);
3818 err
= build_curseg(sbi
);
3822 /* reinit free segmap based on SIT */
3823 build_sit_entries(sbi
);
3825 init_free_segmap(sbi
);
3826 err
= build_dirty_segmap(sbi
);
3830 init_min_max_mtime(sbi
);
3834 static void discard_dirty_segmap(struct f2fs_sb_info
*sbi
,
3835 enum dirty_type dirty_type
)
3837 struct dirty_seglist_info
*dirty_i
= DIRTY_I(sbi
);
3839 mutex_lock(&dirty_i
->seglist_lock
);
3840 kvfree(dirty_i
->dirty_segmap
[dirty_type
]);
3841 dirty_i
->nr_dirty
[dirty_type
] = 0;
3842 mutex_unlock(&dirty_i
->seglist_lock
);
3845 static void destroy_victim_secmap(struct f2fs_sb_info
*sbi
)
3847 struct dirty_seglist_info
*dirty_i
= DIRTY_I(sbi
);
3848 kvfree(dirty_i
->victim_secmap
);
3851 static void destroy_dirty_segmap(struct f2fs_sb_info
*sbi
)
3853 struct dirty_seglist_info
*dirty_i
= DIRTY_I(sbi
);
3859 /* discard pre-free/dirty segments list */
3860 for (i
= 0; i
< NR_DIRTY_TYPE
; i
++)
3861 discard_dirty_segmap(sbi
, i
);
3863 destroy_victim_secmap(sbi
);
3864 SM_I(sbi
)->dirty_info
= NULL
;
3868 static void destroy_curseg(struct f2fs_sb_info
*sbi
)
3870 struct curseg_info
*array
= SM_I(sbi
)->curseg_array
;
3875 SM_I(sbi
)->curseg_array
= NULL
;
3876 for (i
= 0; i
< NR_CURSEG_TYPE
; i
++) {
3877 kfree(array
[i
].sum_blk
);
3878 kfree(array
[i
].journal
);
3883 static void destroy_free_segmap(struct f2fs_sb_info
*sbi
)
3885 struct free_segmap_info
*free_i
= SM_I(sbi
)->free_info
;
3888 SM_I(sbi
)->free_info
= NULL
;
3889 kvfree(free_i
->free_segmap
);
3890 kvfree(free_i
->free_secmap
);
3894 static void destroy_sit_info(struct f2fs_sb_info
*sbi
)
3896 struct sit_info
*sit_i
= SIT_I(sbi
);
3902 if (sit_i
->sentries
) {
3903 for (start
= 0; start
< MAIN_SEGS(sbi
); start
++) {
3904 kfree(sit_i
->sentries
[start
].cur_valid_map
);
3905 #ifdef CONFIG_F2FS_CHECK_FS
3906 kfree(sit_i
->sentries
[start
].cur_valid_map_mir
);
3908 kfree(sit_i
->sentries
[start
].ckpt_valid_map
);
3909 kfree(sit_i
->sentries
[start
].discard_map
);
3912 kfree(sit_i
->tmp_map
);
3914 kvfree(sit_i
->sentries
);
3915 kvfree(sit_i
->sec_entries
);
3916 kvfree(sit_i
->dirty_sentries_bitmap
);
3918 SM_I(sbi
)->sit_info
= NULL
;
3919 kfree(sit_i
->sit_bitmap
);
3920 #ifdef CONFIG_F2FS_CHECK_FS
3921 kfree(sit_i
->sit_bitmap_mir
);
3926 void destroy_segment_manager(struct f2fs_sb_info
*sbi
)
3928 struct f2fs_sm_info
*sm_info
= SM_I(sbi
);
3932 destroy_flush_cmd_control(sbi
, true);
3933 destroy_discard_cmd_control(sbi
);
3934 destroy_dirty_segmap(sbi
);
3935 destroy_curseg(sbi
);
3936 destroy_free_segmap(sbi
);
3937 destroy_sit_info(sbi
);
3938 sbi
->sm_info
= NULL
;
3942 int __init
create_segment_manager_caches(void)
3944 discard_entry_slab
= f2fs_kmem_cache_create("discard_entry",
3945 sizeof(struct discard_entry
));
3946 if (!discard_entry_slab
)
3949 discard_cmd_slab
= f2fs_kmem_cache_create("discard_cmd",
3950 sizeof(struct discard_cmd
));
3951 if (!discard_cmd_slab
)
3952 goto destroy_discard_entry
;
3954 sit_entry_set_slab
= f2fs_kmem_cache_create("sit_entry_set",
3955 sizeof(struct sit_entry_set
));
3956 if (!sit_entry_set_slab
)
3957 goto destroy_discard_cmd
;
3959 inmem_entry_slab
= f2fs_kmem_cache_create("inmem_page_entry",
3960 sizeof(struct inmem_pages
));
3961 if (!inmem_entry_slab
)
3962 goto destroy_sit_entry_set
;
3965 destroy_sit_entry_set
:
3966 kmem_cache_destroy(sit_entry_set_slab
);
3967 destroy_discard_cmd
:
3968 kmem_cache_destroy(discard_cmd_slab
);
3969 destroy_discard_entry
:
3970 kmem_cache_destroy(discard_entry_slab
);
3975 void destroy_segment_manager_caches(void)
3977 kmem_cache_destroy(sit_entry_set_slab
);
3978 kmem_cache_destroy(discard_cmd_slab
);
3979 kmem_cache_destroy(discard_entry_slab
);
3980 kmem_cache_destroy(inmem_entry_slab
);