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/module.h>
13 #include <linux/backing-dev.h>
14 #include <linux/init.h>
15 #include <linux/f2fs_fs.h>
16 #include <linux/kthread.h>
17 #include <linux/delay.h>
18 #include <linux/freezer.h>
24 #include <trace/events/f2fs.h>
26 static int gc_thread_func(void *data
)
28 struct f2fs_sb_info
*sbi
= data
;
29 struct f2fs_gc_kthread
*gc_th
= sbi
->gc_thread
;
30 wait_queue_head_t
*wq
= &sbi
->gc_thread
->gc_wait_queue_head
;
33 wait_ms
= gc_th
->min_sleep_time
;
39 wait_event_interruptible_timeout(*wq
,
40 kthread_should_stop(),
41 msecs_to_jiffies(wait_ms
));
42 if (kthread_should_stop())
45 if (sbi
->sb
->s_writers
.frozen
>= SB_FREEZE_WRITE
) {
46 increase_sleep_time(gc_th
, &wait_ms
);
51 * [GC triggering condition]
52 * 0. GC is not conducted currently.
53 * 1. There are enough dirty segments.
54 * 2. IO subsystem is idle by checking the # of writeback pages.
55 * 3. IO subsystem is idle by checking the # of requests in
56 * bdev's request list.
58 * Note) We have to avoid triggering GCs frequently.
59 * Because it is possible that some segments can be
60 * invalidated soon after by user update or deletion.
61 * So, I'd like to wait some time to collect dirty segments.
63 if (!mutex_trylock(&sbi
->gc_mutex
))
67 increase_sleep_time(gc_th
, &wait_ms
);
68 mutex_unlock(&sbi
->gc_mutex
);
72 if (has_enough_invalid_blocks(sbi
))
73 decrease_sleep_time(gc_th
, &wait_ms
);
75 increase_sleep_time(gc_th
, &wait_ms
);
77 stat_inc_bggc_count(sbi
);
79 /* if return value is not zero, no victim was selected */
80 if (f2fs_gc(sbi
, test_opt(sbi
, FORCE_FG_GC
)))
81 wait_ms
= gc_th
->no_gc_sleep_time
;
83 trace_f2fs_background_gc(sbi
->sb
, wait_ms
,
84 prefree_segments(sbi
), free_segments(sbi
));
86 /* balancing f2fs's metadata periodically */
87 f2fs_balance_fs_bg(sbi
);
89 } while (!kthread_should_stop());
93 int start_gc_thread(struct f2fs_sb_info
*sbi
)
95 struct f2fs_gc_kthread
*gc_th
;
96 dev_t dev
= sbi
->sb
->s_bdev
->bd_dev
;
99 gc_th
= f2fs_kmalloc(sizeof(struct f2fs_gc_kthread
), GFP_KERNEL
);
105 gc_th
->min_sleep_time
= DEF_GC_THREAD_MIN_SLEEP_TIME
;
106 gc_th
->max_sleep_time
= DEF_GC_THREAD_MAX_SLEEP_TIME
;
107 gc_th
->no_gc_sleep_time
= DEF_GC_THREAD_NOGC_SLEEP_TIME
;
111 sbi
->gc_thread
= gc_th
;
112 init_waitqueue_head(&sbi
->gc_thread
->gc_wait_queue_head
);
113 sbi
->gc_thread
->f2fs_gc_task
= kthread_run(gc_thread_func
, sbi
,
114 "f2fs_gc-%u:%u", MAJOR(dev
), MINOR(dev
));
115 if (IS_ERR(gc_th
->f2fs_gc_task
)) {
116 err
= PTR_ERR(gc_th
->f2fs_gc_task
);
118 sbi
->gc_thread
= NULL
;
124 void stop_gc_thread(struct f2fs_sb_info
*sbi
)
126 struct f2fs_gc_kthread
*gc_th
= sbi
->gc_thread
;
129 kthread_stop(gc_th
->f2fs_gc_task
);
131 sbi
->gc_thread
= NULL
;
134 static int select_gc_type(struct f2fs_gc_kthread
*gc_th
, int gc_type
)
136 int gc_mode
= (gc_type
== BG_GC
) ? GC_CB
: GC_GREEDY
;
138 if (gc_th
&& gc_th
->gc_idle
) {
139 if (gc_th
->gc_idle
== 1)
141 else if (gc_th
->gc_idle
== 2)
147 static void select_policy(struct f2fs_sb_info
*sbi
, int gc_type
,
148 int type
, struct victim_sel_policy
*p
)
150 struct dirty_seglist_info
*dirty_i
= DIRTY_I(sbi
);
152 if (p
->alloc_mode
== SSR
) {
153 p
->gc_mode
= GC_GREEDY
;
154 p
->dirty_segmap
= dirty_i
->dirty_segmap
[type
];
155 p
->max_search
= dirty_i
->nr_dirty
[type
];
158 p
->gc_mode
= select_gc_type(sbi
->gc_thread
, gc_type
);
159 p
->dirty_segmap
= dirty_i
->dirty_segmap
[DIRTY
];
160 p
->max_search
= dirty_i
->nr_dirty
[DIRTY
];
161 p
->ofs_unit
= sbi
->segs_per_sec
;
164 if (p
->max_search
> sbi
->max_victim_search
)
165 p
->max_search
= sbi
->max_victim_search
;
167 p
->offset
= sbi
->last_victim
[p
->gc_mode
];
170 static unsigned int get_max_cost(struct f2fs_sb_info
*sbi
,
171 struct victim_sel_policy
*p
)
173 /* SSR allocates in a segment unit */
174 if (p
->alloc_mode
== SSR
)
175 return sbi
->blocks_per_seg
;
176 if (p
->gc_mode
== GC_GREEDY
)
177 return sbi
->blocks_per_seg
* p
->ofs_unit
;
178 else if (p
->gc_mode
== GC_CB
)
180 else /* No other gc_mode */
184 static unsigned int check_bg_victims(struct f2fs_sb_info
*sbi
)
186 struct dirty_seglist_info
*dirty_i
= DIRTY_I(sbi
);
190 * If the gc_type is FG_GC, we can select victim segments
191 * selected by background GC before.
192 * Those segments guarantee they have small valid blocks.
194 for_each_set_bit(secno
, dirty_i
->victim_secmap
, MAIN_SECS(sbi
)) {
195 if (sec_usage_check(sbi
, secno
))
197 clear_bit(secno
, dirty_i
->victim_secmap
);
198 return secno
* sbi
->segs_per_sec
;
203 static unsigned int get_cb_cost(struct f2fs_sb_info
*sbi
, unsigned int segno
)
205 struct sit_info
*sit_i
= SIT_I(sbi
);
206 unsigned int secno
= GET_SECNO(sbi
, segno
);
207 unsigned int start
= secno
* sbi
->segs_per_sec
;
208 unsigned long long mtime
= 0;
209 unsigned int vblocks
;
210 unsigned char age
= 0;
214 for (i
= 0; i
< sbi
->segs_per_sec
; i
++)
215 mtime
+= get_seg_entry(sbi
, start
+ i
)->mtime
;
216 vblocks
= get_valid_blocks(sbi
, segno
, sbi
->segs_per_sec
);
218 mtime
= div_u64(mtime
, sbi
->segs_per_sec
);
219 vblocks
= div_u64(vblocks
, sbi
->segs_per_sec
);
221 u
= (vblocks
* 100) >> sbi
->log_blocks_per_seg
;
223 /* Handle if the system time has changed by the user */
224 if (mtime
< sit_i
->min_mtime
)
225 sit_i
->min_mtime
= mtime
;
226 if (mtime
> sit_i
->max_mtime
)
227 sit_i
->max_mtime
= mtime
;
228 if (sit_i
->max_mtime
!= sit_i
->min_mtime
)
229 age
= 100 - div64_u64(100 * (mtime
- sit_i
->min_mtime
),
230 sit_i
->max_mtime
- sit_i
->min_mtime
);
232 return UINT_MAX
- ((100 * (100 - u
) * age
) / (100 + u
));
235 static inline unsigned int get_gc_cost(struct f2fs_sb_info
*sbi
,
236 unsigned int segno
, struct victim_sel_policy
*p
)
238 if (p
->alloc_mode
== SSR
)
239 return get_seg_entry(sbi
, segno
)->ckpt_valid_blocks
;
241 /* alloc_mode == LFS */
242 if (p
->gc_mode
== GC_GREEDY
)
243 return get_valid_blocks(sbi
, segno
, sbi
->segs_per_sec
);
245 return get_cb_cost(sbi
, segno
);
248 static unsigned int count_bits(const unsigned long *addr
,
249 unsigned int offset
, unsigned int len
)
251 unsigned int end
= offset
+ len
, sum
= 0;
253 while (offset
< end
) {
254 if (test_bit(offset
++, addr
))
261 * This function is called from two paths.
262 * One is garbage collection and the other is SSR segment selection.
263 * When it is called during GC, it just gets a victim segment
264 * and it does not remove it from dirty seglist.
265 * When it is called from SSR segment selection, it finds a segment
266 * which has minimum valid blocks and removes it from dirty seglist.
268 static int get_victim_by_default(struct f2fs_sb_info
*sbi
,
269 unsigned int *result
, int gc_type
, int type
, char alloc_mode
)
271 struct dirty_seglist_info
*dirty_i
= DIRTY_I(sbi
);
272 struct victim_sel_policy p
;
273 unsigned int secno
, max_cost
, last_victim
;
274 unsigned int last_segment
= MAIN_SEGS(sbi
);
275 unsigned int nsearched
= 0;
277 mutex_lock(&dirty_i
->seglist_lock
);
279 p
.alloc_mode
= alloc_mode
;
280 select_policy(sbi
, gc_type
, type
, &p
);
282 p
.min_segno
= NULL_SEGNO
;
283 p
.min_cost
= max_cost
= get_max_cost(sbi
, &p
);
285 if (p
.max_search
== 0)
288 last_victim
= sbi
->last_victim
[p
.gc_mode
];
289 if (p
.alloc_mode
== LFS
&& gc_type
== FG_GC
) {
290 p
.min_segno
= check_bg_victims(sbi
);
291 if (p
.min_segno
!= NULL_SEGNO
)
299 segno
= find_next_bit(p
.dirty_segmap
, last_segment
, p
.offset
);
300 if (segno
>= last_segment
) {
301 if (sbi
->last_victim
[p
.gc_mode
]) {
302 last_segment
= sbi
->last_victim
[p
.gc_mode
];
303 sbi
->last_victim
[p
.gc_mode
] = 0;
310 p
.offset
= segno
+ p
.ofs_unit
;
311 if (p
.ofs_unit
> 1) {
312 p
.offset
-= segno
% p
.ofs_unit
;
313 nsearched
+= count_bits(p
.dirty_segmap
,
314 p
.offset
- p
.ofs_unit
,
321 secno
= GET_SECNO(sbi
, segno
);
323 if (sec_usage_check(sbi
, secno
))
325 if (gc_type
== BG_GC
&& test_bit(secno
, dirty_i
->victim_secmap
))
328 cost
= get_gc_cost(sbi
, segno
, &p
);
330 if (p
.min_cost
> cost
) {
335 if (nsearched
>= p
.max_search
) {
336 if (!sbi
->last_victim
[p
.gc_mode
] && segno
<= last_victim
)
337 sbi
->last_victim
[p
.gc_mode
] = last_victim
+ 1;
339 sbi
->last_victim
[p
.gc_mode
] = segno
+ 1;
343 if (p
.min_segno
!= NULL_SEGNO
) {
345 if (p
.alloc_mode
== LFS
) {
346 secno
= GET_SECNO(sbi
, p
.min_segno
);
347 if (gc_type
== FG_GC
)
348 sbi
->cur_victim_sec
= secno
;
350 set_bit(secno
, dirty_i
->victim_secmap
);
352 *result
= (p
.min_segno
/ p
.ofs_unit
) * p
.ofs_unit
;
354 trace_f2fs_get_victim(sbi
->sb
, type
, gc_type
, &p
,
356 prefree_segments(sbi
), free_segments(sbi
));
359 mutex_unlock(&dirty_i
->seglist_lock
);
361 return (p
.min_segno
== NULL_SEGNO
) ? 0 : 1;
364 static const struct victim_selection default_v_ops
= {
365 .get_victim
= get_victim_by_default
,
368 static struct inode
*find_gc_inode(struct gc_inode_list
*gc_list
, nid_t ino
)
370 struct inode_entry
*ie
;
372 ie
= radix_tree_lookup(&gc_list
->iroot
, ino
);
378 static void add_gc_inode(struct gc_inode_list
*gc_list
, struct inode
*inode
)
380 struct inode_entry
*new_ie
;
382 if (inode
== find_gc_inode(gc_list
, inode
->i_ino
)) {
386 new_ie
= f2fs_kmem_cache_alloc(inode_entry_slab
, GFP_NOFS
);
387 new_ie
->inode
= inode
;
389 f2fs_radix_tree_insert(&gc_list
->iroot
, inode
->i_ino
, new_ie
);
390 list_add_tail(&new_ie
->list
, &gc_list
->ilist
);
393 static void put_gc_inode(struct gc_inode_list
*gc_list
)
395 struct inode_entry
*ie
, *next_ie
;
396 list_for_each_entry_safe(ie
, next_ie
, &gc_list
->ilist
, list
) {
397 radix_tree_delete(&gc_list
->iroot
, ie
->inode
->i_ino
);
400 kmem_cache_free(inode_entry_slab
, ie
);
404 static int check_valid_map(struct f2fs_sb_info
*sbi
,
405 unsigned int segno
, int offset
)
407 struct sit_info
*sit_i
= SIT_I(sbi
);
408 struct seg_entry
*sentry
;
411 mutex_lock(&sit_i
->sentry_lock
);
412 sentry
= get_seg_entry(sbi
, segno
);
413 ret
= f2fs_test_bit(offset
, sentry
->cur_valid_map
);
414 mutex_unlock(&sit_i
->sentry_lock
);
419 * This function compares node address got in summary with that in NAT.
420 * On validity, copy that node with cold status, otherwise (invalid node)
423 static void gc_node_segment(struct f2fs_sb_info
*sbi
,
424 struct f2fs_summary
*sum
, unsigned int segno
, int gc_type
)
426 struct f2fs_summary
*entry
;
431 start_addr
= START_BLOCK(sbi
, segno
);
436 for (off
= 0; off
< sbi
->blocks_per_seg
; off
++, entry
++) {
437 nid_t nid
= le32_to_cpu(entry
->nid
);
438 struct page
*node_page
;
441 /* stop BG_GC if there is not enough free sections. */
442 if (gc_type
== BG_GC
&& has_not_enough_free_secs(sbi
, 0, 0))
445 if (check_valid_map(sbi
, segno
, off
) == 0)
449 ra_meta_pages(sbi
, NAT_BLOCK_OFFSET(nid
), 1,
455 ra_node_page(sbi
, nid
);
460 node_page
= get_node_page(sbi
, nid
);
461 if (IS_ERR(node_page
))
464 /* block may become invalid during get_node_page */
465 if (check_valid_map(sbi
, segno
, off
) == 0) {
466 f2fs_put_page(node_page
, 1);
470 get_node_info(sbi
, nid
, &ni
);
471 if (ni
.blk_addr
!= start_addr
+ off
) {
472 f2fs_put_page(node_page
, 1);
476 move_node_page(node_page
, gc_type
);
477 stat_inc_node_blk_count(sbi
, 1, gc_type
);
485 * Calculate start block index indicating the given node offset.
486 * Be careful, caller should give this node offset only indicating direct node
487 * blocks. If any node offsets, which point the other types of node blocks such
488 * as indirect or double indirect node blocks, are given, it must be a caller's
491 block_t
start_bidx_of_node(unsigned int node_ofs
, struct inode
*inode
)
493 unsigned int indirect_blks
= 2 * NIDS_PER_BLOCK
+ 4;
501 } else if (node_ofs
<= indirect_blks
) {
502 int dec
= (node_ofs
- 4) / (NIDS_PER_BLOCK
+ 1);
503 bidx
= node_ofs
- 2 - dec
;
505 int dec
= (node_ofs
- indirect_blks
- 3) / (NIDS_PER_BLOCK
+ 1);
506 bidx
= node_ofs
- 5 - dec
;
508 return bidx
* ADDRS_PER_BLOCK
+ ADDRS_PER_INODE(inode
);
511 static bool is_alive(struct f2fs_sb_info
*sbi
, struct f2fs_summary
*sum
,
512 struct node_info
*dni
, block_t blkaddr
, unsigned int *nofs
)
514 struct page
*node_page
;
516 unsigned int ofs_in_node
;
517 block_t source_blkaddr
;
519 nid
= le32_to_cpu(sum
->nid
);
520 ofs_in_node
= le16_to_cpu(sum
->ofs_in_node
);
522 node_page
= get_node_page(sbi
, nid
);
523 if (IS_ERR(node_page
))
526 get_node_info(sbi
, nid
, dni
);
528 if (sum
->version
!= dni
->version
) {
529 f2fs_put_page(node_page
, 1);
533 *nofs
= ofs_of_node(node_page
);
534 source_blkaddr
= datablock_addr(node_page
, ofs_in_node
);
535 f2fs_put_page(node_page
, 1);
537 if (source_blkaddr
!= blkaddr
)
542 static void move_encrypted_block(struct inode
*inode
, block_t bidx
)
544 struct f2fs_io_info fio
= {
545 .sbi
= F2FS_I_SB(inode
),
548 .op_flags
= READ_SYNC
,
549 .encrypted_page
= NULL
,
551 struct dnode_of_data dn
;
552 struct f2fs_summary sum
;
558 /* do not read out */
559 page
= f2fs_grab_cache_page(inode
->i_mapping
, bidx
, false);
563 set_new_dnode(&dn
, inode
, NULL
, NULL
, 0);
564 err
= get_dnode_of_data(&dn
, bidx
, LOOKUP_NODE
);
568 if (unlikely(dn
.data_blkaddr
== NULL_ADDR
)) {
569 ClearPageUptodate(page
);
574 * don't cache encrypted data into meta inode until previous dirty
575 * data were writebacked to avoid racing between GC and flush.
577 f2fs_wait_on_page_writeback(page
, DATA
, true);
579 get_node_info(fio
.sbi
, dn
.nid
, &ni
);
580 set_summary(&sum
, dn
.nid
, dn
.ofs_in_node
, ni
.version
);
584 fio
.new_blkaddr
= fio
.old_blkaddr
= dn
.data_blkaddr
;
586 allocate_data_block(fio
.sbi
, NULL
, fio
.old_blkaddr
, &newaddr
,
587 &sum
, CURSEG_COLD_DATA
);
589 fio
.encrypted_page
= pagecache_get_page(META_MAPPING(fio
.sbi
), newaddr
,
590 FGP_LOCK
| FGP_CREAT
, GFP_NOFS
);
591 if (!fio
.encrypted_page
) {
596 err
= f2fs_submit_page_bio(&fio
);
601 lock_page(fio
.encrypted_page
);
603 if (unlikely(fio
.encrypted_page
->mapping
!= META_MAPPING(fio
.sbi
))) {
607 if (unlikely(!PageUptodate(fio
.encrypted_page
))) {
612 set_page_dirty(fio
.encrypted_page
);
613 f2fs_wait_on_page_writeback(fio
.encrypted_page
, DATA
, true);
614 if (clear_page_dirty_for_io(fio
.encrypted_page
))
615 dec_page_count(fio
.sbi
, F2FS_DIRTY_META
);
617 set_page_writeback(fio
.encrypted_page
);
619 /* allocate block address */
620 f2fs_wait_on_page_writeback(dn
.node_page
, NODE
, true);
622 fio
.op
= REQ_OP_WRITE
;
623 fio
.op_flags
= WRITE_SYNC
;
624 fio
.new_blkaddr
= newaddr
;
625 f2fs_submit_page_mbio(&fio
);
627 f2fs_update_data_blkaddr(&dn
, newaddr
);
628 set_inode_flag(inode
, FI_APPEND_WRITE
);
629 if (page
->index
== 0)
630 set_inode_flag(inode
, FI_FIRST_BLOCK_WRITTEN
);
632 f2fs_put_page(fio
.encrypted_page
, 1);
635 __f2fs_replace_block(fio
.sbi
, &sum
, newaddr
, fio
.old_blkaddr
,
640 f2fs_put_page(page
, 1);
643 static void move_data_page(struct inode
*inode
, block_t bidx
, int gc_type
)
647 page
= get_lock_data_page(inode
, bidx
, true);
651 if (gc_type
== BG_GC
) {
652 if (PageWriteback(page
))
654 set_page_dirty(page
);
657 struct f2fs_io_info fio
= {
658 .sbi
= F2FS_I_SB(inode
),
661 .op_flags
= WRITE_SYNC
,
663 .encrypted_page
= NULL
,
665 bool is_dirty
= PageDirty(page
);
669 set_page_dirty(page
);
670 f2fs_wait_on_page_writeback(page
, DATA
, true);
671 if (clear_page_dirty_for_io(page
))
672 inode_dec_dirty_pages(inode
);
676 err
= do_write_data_page(&fio
);
677 if (err
== -ENOMEM
&& is_dirty
) {
678 congestion_wait(BLK_RW_ASYNC
, HZ
/50);
682 clear_cold_data(page
);
685 f2fs_put_page(page
, 1);
689 * This function tries to get parent node of victim data block, and identifies
690 * data block validity. If the block is valid, copy that with cold status and
691 * modify parent node.
692 * If the parent node is not valid or the data block address is different,
693 * the victim data block is ignored.
695 static void gc_data_segment(struct f2fs_sb_info
*sbi
, struct f2fs_summary
*sum
,
696 struct gc_inode_list
*gc_list
, unsigned int segno
, int gc_type
)
698 struct super_block
*sb
= sbi
->sb
;
699 struct f2fs_summary
*entry
;
704 start_addr
= START_BLOCK(sbi
, segno
);
709 for (off
= 0; off
< sbi
->blocks_per_seg
; off
++, entry
++) {
710 struct page
*data_page
;
712 struct node_info dni
; /* dnode info for the data */
713 unsigned int ofs_in_node
, nofs
;
715 nid_t nid
= le32_to_cpu(entry
->nid
);
717 /* stop BG_GC if there is not enough free sections. */
718 if (gc_type
== BG_GC
&& has_not_enough_free_secs(sbi
, 0, 0))
721 if (check_valid_map(sbi
, segno
, off
) == 0)
725 ra_meta_pages(sbi
, NAT_BLOCK_OFFSET(nid
), 1,
731 ra_node_page(sbi
, nid
);
735 /* Get an inode by ino with checking validity */
736 if (!is_alive(sbi
, entry
, &dni
, start_addr
+ off
, &nofs
))
740 ra_node_page(sbi
, dni
.ino
);
744 ofs_in_node
= le16_to_cpu(entry
->ofs_in_node
);
747 inode
= f2fs_iget(sb
, dni
.ino
);
748 if (IS_ERR(inode
) || is_bad_inode(inode
))
751 /* if encrypted inode, let's go phase 3 */
752 if (f2fs_encrypted_inode(inode
) &&
753 S_ISREG(inode
->i_mode
)) {
754 add_gc_inode(gc_list
, inode
);
758 start_bidx
= start_bidx_of_node(nofs
, inode
);
759 data_page
= get_read_data_page(inode
,
760 start_bidx
+ ofs_in_node
, REQ_RAHEAD
,
762 if (IS_ERR(data_page
)) {
767 f2fs_put_page(data_page
, 0);
768 add_gc_inode(gc_list
, inode
);
773 inode
= find_gc_inode(gc_list
, dni
.ino
);
775 struct f2fs_inode_info
*fi
= F2FS_I(inode
);
778 if (S_ISREG(inode
->i_mode
)) {
779 if (!down_write_trylock(&fi
->dio_rwsem
[READ
]))
781 if (!down_write_trylock(
782 &fi
->dio_rwsem
[WRITE
])) {
783 up_write(&fi
->dio_rwsem
[READ
]);
789 start_bidx
= start_bidx_of_node(nofs
, inode
)
791 if (f2fs_encrypted_inode(inode
) && S_ISREG(inode
->i_mode
))
792 move_encrypted_block(inode
, start_bidx
);
794 move_data_page(inode
, start_bidx
, gc_type
);
797 up_write(&fi
->dio_rwsem
[WRITE
]);
798 up_write(&fi
->dio_rwsem
[READ
]);
801 stat_inc_data_blk_count(sbi
, 1, gc_type
);
809 static int __get_victim(struct f2fs_sb_info
*sbi
, unsigned int *victim
,
812 struct sit_info
*sit_i
= SIT_I(sbi
);
815 mutex_lock(&sit_i
->sentry_lock
);
816 ret
= DIRTY_I(sbi
)->v_ops
->get_victim(sbi
, victim
, gc_type
,
818 mutex_unlock(&sit_i
->sentry_lock
);
822 static int do_garbage_collect(struct f2fs_sb_info
*sbi
,
823 unsigned int start_segno
,
824 struct gc_inode_list
*gc_list
, int gc_type
)
826 struct page
*sum_page
;
827 struct f2fs_summary_block
*sum
;
828 struct blk_plug plug
;
829 unsigned int segno
= start_segno
;
830 unsigned int end_segno
= start_segno
+ sbi
->segs_per_sec
;
832 unsigned char type
= IS_DATASEG(get_seg_entry(sbi
, segno
)->type
) ?
833 SUM_TYPE_DATA
: SUM_TYPE_NODE
;
835 /* readahead multi ssa blocks those have contiguous address */
836 if (sbi
->segs_per_sec
> 1)
837 ra_meta_pages(sbi
, GET_SUM_BLOCK(sbi
, segno
),
838 sbi
->segs_per_sec
, META_SSA
, true);
840 /* reference all summary page */
841 while (segno
< end_segno
) {
842 sum_page
= get_sum_page(sbi
, segno
++);
843 unlock_page(sum_page
);
846 blk_start_plug(&plug
);
848 for (segno
= start_segno
; segno
< end_segno
; segno
++) {
850 if (get_valid_blocks(sbi
, segno
, 1) == 0)
853 /* find segment summary of victim */
854 sum_page
= find_get_page(META_MAPPING(sbi
),
855 GET_SUM_BLOCK(sbi
, segno
));
856 f2fs_bug_on(sbi
, !PageUptodate(sum_page
));
857 f2fs_put_page(sum_page
, 0);
859 sum
= page_address(sum_page
);
860 f2fs_bug_on(sbi
, type
!= GET_SUM_TYPE((&sum
->footer
)));
863 * this is to avoid deadlock:
864 * - lock_page(sum_page) - f2fs_replace_block
865 * - check_valid_map() - mutex_lock(sentry_lock)
866 * - mutex_lock(sentry_lock) - change_curseg()
867 * - lock_page(sum_page)
870 if (type
== SUM_TYPE_NODE
)
871 gc_node_segment(sbi
, sum
->entries
, segno
, gc_type
);
873 gc_data_segment(sbi
, sum
->entries
, gc_list
, segno
,
876 stat_inc_seg_count(sbi
, type
, gc_type
);
878 f2fs_put_page(sum_page
, 0);
881 if (gc_type
== FG_GC
)
882 f2fs_submit_merged_bio(sbi
,
883 (type
== SUM_TYPE_NODE
) ? NODE
: DATA
, WRITE
);
885 blk_finish_plug(&plug
);
887 if (gc_type
== FG_GC
&&
888 get_valid_blocks(sbi
, start_segno
, sbi
->segs_per_sec
) == 0)
891 stat_inc_call_count(sbi
->stat_info
);
896 int f2fs_gc(struct f2fs_sb_info
*sbi
, bool sync
)
899 int gc_type
= sync
? FG_GC
: BG_GC
;
902 struct cp_control cpc
;
903 struct gc_inode_list gc_list
= {
904 .ilist
= LIST_HEAD_INIT(gc_list
.ilist
),
905 .iroot
= RADIX_TREE_INIT(GFP_NOFS
),
908 cpc
.reason
= __get_cp_reason(sbi
);
912 if (unlikely(!(sbi
->sb
->s_flags
& MS_ACTIVE
)))
914 if (unlikely(f2fs_cp_error(sbi
))) {
919 if (gc_type
== BG_GC
&& has_not_enough_free_secs(sbi
, sec_freed
, 0)) {
922 * If there is no victim and no prefree segment but still not
923 * enough free sections, we should flush dent/node blocks and do
924 * garbage collections.
926 if (__get_victim(sbi
, &segno
, gc_type
) ||
927 prefree_segments(sbi
)) {
928 write_checkpoint(sbi
, &cpc
);
930 } else if (has_not_enough_free_secs(sbi
, 0, 0)) {
931 write_checkpoint(sbi
, &cpc
);
935 if (segno
== NULL_SEGNO
&& !__get_victim(sbi
, &segno
, gc_type
))
939 if (do_garbage_collect(sbi
, segno
, &gc_list
, gc_type
) &&
943 if (gc_type
== FG_GC
)
944 sbi
->cur_victim_sec
= NULL_SEGNO
;
947 if (has_not_enough_free_secs(sbi
, sec_freed
, 0))
950 if (gc_type
== FG_GC
)
951 write_checkpoint(sbi
, &cpc
);
954 mutex_unlock(&sbi
->gc_mutex
);
956 put_gc_inode(&gc_list
);
959 ret
= sec_freed
? 0 : -EAGAIN
;
963 void build_gc_manager(struct f2fs_sb_info
*sbi
)
965 DIRTY_I(sbi
)->v_ops
= &default_v_ops
;