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/mpage.h>
14 #include <linux/backing-dev.h>
15 #include <linux/blkdev.h>
16 #include <linux/pagevec.h>
17 #include <linux/swap.h>
22 #include <trace/events/f2fs.h>
24 static struct kmem_cache
*nat_entry_slab
;
25 static struct kmem_cache
*free_nid_slab
;
27 static void clear_node_page_dirty(struct page
*page
)
29 struct address_space
*mapping
= page
->mapping
;
30 struct f2fs_sb_info
*sbi
= F2FS_SB(mapping
->host
->i_sb
);
31 unsigned int long flags
;
33 if (PageDirty(page
)) {
34 spin_lock_irqsave(&mapping
->tree_lock
, flags
);
35 radix_tree_tag_clear(&mapping
->page_tree
,
38 spin_unlock_irqrestore(&mapping
->tree_lock
, flags
);
40 clear_page_dirty_for_io(page
);
41 dec_page_count(sbi
, F2FS_DIRTY_NODES
);
43 ClearPageUptodate(page
);
46 static struct page
*get_current_nat_page(struct f2fs_sb_info
*sbi
, nid_t nid
)
48 pgoff_t index
= current_nat_addr(sbi
, nid
);
49 return get_meta_page(sbi
, index
);
52 static struct page
*get_next_nat_page(struct f2fs_sb_info
*sbi
, nid_t nid
)
54 struct page
*src_page
;
55 struct page
*dst_page
;
60 struct f2fs_nm_info
*nm_i
= NM_I(sbi
);
62 src_off
= current_nat_addr(sbi
, nid
);
63 dst_off
= next_nat_addr(sbi
, src_off
);
65 /* get current nat block page with lock */
66 src_page
= get_meta_page(sbi
, src_off
);
68 /* Dirty src_page means that it is already the new target NAT page. */
69 if (PageDirty(src_page
))
72 dst_page
= grab_meta_page(sbi
, dst_off
);
74 src_addr
= page_address(src_page
);
75 dst_addr
= page_address(dst_page
);
76 memcpy(dst_addr
, src_addr
, PAGE_CACHE_SIZE
);
77 set_page_dirty(dst_page
);
78 f2fs_put_page(src_page
, 1);
80 set_to_next_nat(nm_i
, nid
);
88 static void ra_nat_pages(struct f2fs_sb_info
*sbi
, int nid
)
90 struct address_space
*mapping
= sbi
->meta_inode
->i_mapping
;
91 struct f2fs_nm_info
*nm_i
= NM_I(sbi
);
97 blk_start_plug(&plug
);
99 for (i
= 0; i
< FREE_NID_PAGES
; i
++, nid
+= NAT_ENTRY_PER_BLOCK
) {
100 if (nid
>= nm_i
->max_nid
)
102 index
= current_nat_addr(sbi
, nid
);
104 page
= grab_cache_page(mapping
, index
);
107 if (PageUptodate(page
)) {
108 f2fs_put_page(page
, 1);
111 if (f2fs_readpage(sbi
, page
, index
, READ
))
114 f2fs_put_page(page
, 0);
116 blk_finish_plug(&plug
);
119 static struct nat_entry
*__lookup_nat_cache(struct f2fs_nm_info
*nm_i
, nid_t n
)
121 return radix_tree_lookup(&nm_i
->nat_root
, n
);
124 static unsigned int __gang_lookup_nat_cache(struct f2fs_nm_info
*nm_i
,
125 nid_t start
, unsigned int nr
, struct nat_entry
**ep
)
127 return radix_tree_gang_lookup(&nm_i
->nat_root
, (void **)ep
, start
, nr
);
130 static void __del_from_nat_cache(struct f2fs_nm_info
*nm_i
, struct nat_entry
*e
)
133 radix_tree_delete(&nm_i
->nat_root
, nat_get_nid(e
));
135 kmem_cache_free(nat_entry_slab
, e
);
138 int is_checkpointed_node(struct f2fs_sb_info
*sbi
, nid_t nid
)
140 struct f2fs_nm_info
*nm_i
= NM_I(sbi
);
144 read_lock(&nm_i
->nat_tree_lock
);
145 e
= __lookup_nat_cache(nm_i
, nid
);
146 if (e
&& !e
->checkpointed
)
148 read_unlock(&nm_i
->nat_tree_lock
);
152 static struct nat_entry
*grab_nat_entry(struct f2fs_nm_info
*nm_i
, nid_t nid
)
154 struct nat_entry
*new;
156 new = kmem_cache_alloc(nat_entry_slab
, GFP_ATOMIC
);
159 if (radix_tree_insert(&nm_i
->nat_root
, nid
, new)) {
160 kmem_cache_free(nat_entry_slab
, new);
163 memset(new, 0, sizeof(struct nat_entry
));
164 nat_set_nid(new, nid
);
165 list_add_tail(&new->list
, &nm_i
->nat_entries
);
170 static void cache_nat_entry(struct f2fs_nm_info
*nm_i
, nid_t nid
,
171 struct f2fs_nat_entry
*ne
)
175 write_lock(&nm_i
->nat_tree_lock
);
176 e
= __lookup_nat_cache(nm_i
, nid
);
178 e
= grab_nat_entry(nm_i
, nid
);
180 write_unlock(&nm_i
->nat_tree_lock
);
183 nat_set_blkaddr(e
, le32_to_cpu(ne
->block_addr
));
184 nat_set_ino(e
, le32_to_cpu(ne
->ino
));
185 nat_set_version(e
, ne
->version
);
186 e
->checkpointed
= true;
188 write_unlock(&nm_i
->nat_tree_lock
);
191 static void set_node_addr(struct f2fs_sb_info
*sbi
, struct node_info
*ni
,
194 struct f2fs_nm_info
*nm_i
= NM_I(sbi
);
197 write_lock(&nm_i
->nat_tree_lock
);
198 e
= __lookup_nat_cache(nm_i
, ni
->nid
);
200 e
= grab_nat_entry(nm_i
, ni
->nid
);
202 write_unlock(&nm_i
->nat_tree_lock
);
206 e
->checkpointed
= true;
207 BUG_ON(ni
->blk_addr
== NEW_ADDR
);
208 } else if (new_blkaddr
== NEW_ADDR
) {
210 * when nid is reallocated,
211 * previous nat entry can be remained in nat cache.
212 * So, reinitialize it with new information.
215 BUG_ON(ni
->blk_addr
!= NULL_ADDR
);
218 if (new_blkaddr
== NEW_ADDR
)
219 e
->checkpointed
= false;
222 BUG_ON(nat_get_blkaddr(e
) != ni
->blk_addr
);
223 BUG_ON(nat_get_blkaddr(e
) == NULL_ADDR
&&
224 new_blkaddr
== NULL_ADDR
);
225 BUG_ON(nat_get_blkaddr(e
) == NEW_ADDR
&&
226 new_blkaddr
== NEW_ADDR
);
227 BUG_ON(nat_get_blkaddr(e
) != NEW_ADDR
&&
228 nat_get_blkaddr(e
) != NULL_ADDR
&&
229 new_blkaddr
== NEW_ADDR
);
231 /* increament version no as node is removed */
232 if (nat_get_blkaddr(e
) != NEW_ADDR
&& new_blkaddr
== NULL_ADDR
) {
233 unsigned char version
= nat_get_version(e
);
234 nat_set_version(e
, inc_node_version(version
));
238 nat_set_blkaddr(e
, new_blkaddr
);
239 __set_nat_cache_dirty(nm_i
, e
);
240 write_unlock(&nm_i
->nat_tree_lock
);
243 static int try_to_free_nats(struct f2fs_sb_info
*sbi
, int nr_shrink
)
245 struct f2fs_nm_info
*nm_i
= NM_I(sbi
);
247 if (nm_i
->nat_cnt
<= NM_WOUT_THRESHOLD
)
250 write_lock(&nm_i
->nat_tree_lock
);
251 while (nr_shrink
&& !list_empty(&nm_i
->nat_entries
)) {
252 struct nat_entry
*ne
;
253 ne
= list_first_entry(&nm_i
->nat_entries
,
254 struct nat_entry
, list
);
255 __del_from_nat_cache(nm_i
, ne
);
258 write_unlock(&nm_i
->nat_tree_lock
);
263 * This function returns always success
265 void get_node_info(struct f2fs_sb_info
*sbi
, nid_t nid
, struct node_info
*ni
)
267 struct f2fs_nm_info
*nm_i
= NM_I(sbi
);
268 struct curseg_info
*curseg
= CURSEG_I(sbi
, CURSEG_HOT_DATA
);
269 struct f2fs_summary_block
*sum
= curseg
->sum_blk
;
270 nid_t start_nid
= START_NID(nid
);
271 struct f2fs_nat_block
*nat_blk
;
272 struct page
*page
= NULL
;
273 struct f2fs_nat_entry ne
;
277 memset(&ne
, 0, sizeof(struct f2fs_nat_entry
));
280 /* Check nat cache */
281 read_lock(&nm_i
->nat_tree_lock
);
282 e
= __lookup_nat_cache(nm_i
, nid
);
284 ni
->ino
= nat_get_ino(e
);
285 ni
->blk_addr
= nat_get_blkaddr(e
);
286 ni
->version
= nat_get_version(e
);
288 read_unlock(&nm_i
->nat_tree_lock
);
292 /* Check current segment summary */
293 mutex_lock(&curseg
->curseg_mutex
);
294 i
= lookup_journal_in_cursum(sum
, NAT_JOURNAL
, nid
, 0);
296 ne
= nat_in_journal(sum
, i
);
297 node_info_from_raw_nat(ni
, &ne
);
299 mutex_unlock(&curseg
->curseg_mutex
);
303 /* Fill node_info from nat page */
304 page
= get_current_nat_page(sbi
, start_nid
);
305 nat_blk
= (struct f2fs_nat_block
*)page_address(page
);
306 ne
= nat_blk
->entries
[nid
- start_nid
];
307 node_info_from_raw_nat(ni
, &ne
);
308 f2fs_put_page(page
, 1);
310 /* cache nat entry */
311 cache_nat_entry(NM_I(sbi
), nid
, &ne
);
315 * The maximum depth is four.
316 * Offset[0] will have raw inode offset.
318 static int get_node_path(struct f2fs_inode_info
*fi
, long block
,
319 int offset
[4], unsigned int noffset
[4])
321 const long direct_index
= ADDRS_PER_INODE(fi
);
322 const long direct_blks
= ADDRS_PER_BLOCK
;
323 const long dptrs_per_blk
= NIDS_PER_BLOCK
;
324 const long indirect_blks
= ADDRS_PER_BLOCK
* NIDS_PER_BLOCK
;
325 const long dindirect_blks
= indirect_blks
* NIDS_PER_BLOCK
;
331 if (block
< direct_index
) {
335 block
-= direct_index
;
336 if (block
< direct_blks
) {
337 offset
[n
++] = NODE_DIR1_BLOCK
;
343 block
-= direct_blks
;
344 if (block
< direct_blks
) {
345 offset
[n
++] = NODE_DIR2_BLOCK
;
351 block
-= direct_blks
;
352 if (block
< indirect_blks
) {
353 offset
[n
++] = NODE_IND1_BLOCK
;
355 offset
[n
++] = block
/ direct_blks
;
356 noffset
[n
] = 4 + offset
[n
- 1];
357 offset
[n
] = block
% direct_blks
;
361 block
-= indirect_blks
;
362 if (block
< indirect_blks
) {
363 offset
[n
++] = NODE_IND2_BLOCK
;
364 noffset
[n
] = 4 + dptrs_per_blk
;
365 offset
[n
++] = block
/ direct_blks
;
366 noffset
[n
] = 5 + dptrs_per_blk
+ offset
[n
- 1];
367 offset
[n
] = block
% direct_blks
;
371 block
-= indirect_blks
;
372 if (block
< dindirect_blks
) {
373 offset
[n
++] = NODE_DIND_BLOCK
;
374 noffset
[n
] = 5 + (dptrs_per_blk
* 2);
375 offset
[n
++] = block
/ indirect_blks
;
376 noffset
[n
] = 6 + (dptrs_per_blk
* 2) +
377 offset
[n
- 1] * (dptrs_per_blk
+ 1);
378 offset
[n
++] = (block
/ direct_blks
) % dptrs_per_blk
;
379 noffset
[n
] = 7 + (dptrs_per_blk
* 2) +
380 offset
[n
- 2] * (dptrs_per_blk
+ 1) +
382 offset
[n
] = block
% direct_blks
;
393 * Caller should call f2fs_put_dnode(dn).
394 * Also, it should grab and release a mutex by calling mutex_lock_op() and
395 * mutex_unlock_op() only if ro is not set RDONLY_NODE.
396 * In the case of RDONLY_NODE, we don't need to care about mutex.
398 int get_dnode_of_data(struct dnode_of_data
*dn
, pgoff_t index
, int mode
)
400 struct f2fs_sb_info
*sbi
= F2FS_SB(dn
->inode
->i_sb
);
401 struct page
*npage
[4];
404 unsigned int noffset
[4];
409 level
= get_node_path(F2FS_I(dn
->inode
), index
, offset
, noffset
);
411 nids
[0] = dn
->inode
->i_ino
;
412 npage
[0] = dn
->inode_page
;
415 npage
[0] = get_node_page(sbi
, nids
[0]);
416 if (IS_ERR(npage
[0]))
417 return PTR_ERR(npage
[0]);
421 nids
[1] = get_nid(parent
, offset
[0], true);
422 dn
->inode_page
= npage
[0];
423 dn
->inode_page_locked
= true;
425 /* get indirect or direct nodes */
426 for (i
= 1; i
<= level
; i
++) {
429 if (!nids
[i
] && mode
== ALLOC_NODE
) {
431 if (!alloc_nid(sbi
, &(nids
[i
]))) {
437 npage
[i
] = new_node_page(dn
, noffset
[i
], NULL
);
438 if (IS_ERR(npage
[i
])) {
439 alloc_nid_failed(sbi
, nids
[i
]);
440 err
= PTR_ERR(npage
[i
]);
444 set_nid(parent
, offset
[i
- 1], nids
[i
], i
== 1);
445 alloc_nid_done(sbi
, nids
[i
]);
447 } else if (mode
== LOOKUP_NODE_RA
&& i
== level
&& level
> 1) {
448 npage
[i
] = get_node_page_ra(parent
, offset
[i
- 1]);
449 if (IS_ERR(npage
[i
])) {
450 err
= PTR_ERR(npage
[i
]);
456 dn
->inode_page_locked
= false;
459 f2fs_put_page(parent
, 1);
463 npage
[i
] = get_node_page(sbi
, nids
[i
]);
464 if (IS_ERR(npage
[i
])) {
465 err
= PTR_ERR(npage
[i
]);
466 f2fs_put_page(npage
[0], 0);
472 nids
[i
+ 1] = get_nid(parent
, offset
[i
], false);
475 dn
->nid
= nids
[level
];
476 dn
->ofs_in_node
= offset
[level
];
477 dn
->node_page
= npage
[level
];
478 dn
->data_blkaddr
= datablock_addr(dn
->node_page
, dn
->ofs_in_node
);
482 f2fs_put_page(parent
, 1);
484 f2fs_put_page(npage
[0], 0);
486 dn
->inode_page
= NULL
;
487 dn
->node_page
= NULL
;
491 static void truncate_node(struct dnode_of_data
*dn
)
493 struct f2fs_sb_info
*sbi
= F2FS_SB(dn
->inode
->i_sb
);
496 get_node_info(sbi
, dn
->nid
, &ni
);
497 if (dn
->inode
->i_blocks
== 0) {
498 BUG_ON(ni
.blk_addr
!= NULL_ADDR
);
501 BUG_ON(ni
.blk_addr
== NULL_ADDR
);
503 /* Deallocate node address */
504 invalidate_blocks(sbi
, ni
.blk_addr
);
505 dec_valid_node_count(sbi
, dn
->inode
, 1);
506 set_node_addr(sbi
, &ni
, NULL_ADDR
);
508 if (dn
->nid
== dn
->inode
->i_ino
) {
509 remove_orphan_inode(sbi
, dn
->nid
);
510 dec_valid_inode_count(sbi
);
515 clear_node_page_dirty(dn
->node_page
);
516 F2FS_SET_SB_DIRT(sbi
);
518 f2fs_put_page(dn
->node_page
, 1);
519 dn
->node_page
= NULL
;
520 trace_f2fs_truncate_node(dn
->inode
, dn
->nid
, ni
.blk_addr
);
523 static int truncate_dnode(struct dnode_of_data
*dn
)
525 struct f2fs_sb_info
*sbi
= F2FS_SB(dn
->inode
->i_sb
);
531 /* get direct node */
532 page
= get_node_page(sbi
, dn
->nid
);
533 if (IS_ERR(page
) && PTR_ERR(page
) == -ENOENT
)
535 else if (IS_ERR(page
))
536 return PTR_ERR(page
);
538 /* Make dnode_of_data for parameter */
539 dn
->node_page
= page
;
541 truncate_data_blocks(dn
);
546 static int truncate_nodes(struct dnode_of_data
*dn
, unsigned int nofs
,
549 struct f2fs_sb_info
*sbi
= F2FS_SB(dn
->inode
->i_sb
);
550 struct dnode_of_data rdn
= *dn
;
552 struct f2fs_node
*rn
;
554 unsigned int child_nofs
;
559 return NIDS_PER_BLOCK
+ 1;
561 trace_f2fs_truncate_nodes_enter(dn
->inode
, dn
->nid
, dn
->data_blkaddr
);
563 page
= get_node_page(sbi
, dn
->nid
);
565 trace_f2fs_truncate_nodes_exit(dn
->inode
, PTR_ERR(page
));
566 return PTR_ERR(page
);
569 rn
= F2FS_NODE(page
);
571 for (i
= ofs
; i
< NIDS_PER_BLOCK
; i
++, freed
++) {
572 child_nid
= le32_to_cpu(rn
->in
.nid
[i
]);
576 ret
= truncate_dnode(&rdn
);
579 set_nid(page
, i
, 0, false);
582 child_nofs
= nofs
+ ofs
* (NIDS_PER_BLOCK
+ 1) + 1;
583 for (i
= ofs
; i
< NIDS_PER_BLOCK
; i
++) {
584 child_nid
= le32_to_cpu(rn
->in
.nid
[i
]);
585 if (child_nid
== 0) {
586 child_nofs
+= NIDS_PER_BLOCK
+ 1;
590 ret
= truncate_nodes(&rdn
, child_nofs
, 0, depth
- 1);
591 if (ret
== (NIDS_PER_BLOCK
+ 1)) {
592 set_nid(page
, i
, 0, false);
594 } else if (ret
< 0 && ret
!= -ENOENT
) {
602 /* remove current indirect node */
603 dn
->node_page
= page
;
607 f2fs_put_page(page
, 1);
609 trace_f2fs_truncate_nodes_exit(dn
->inode
, freed
);
613 f2fs_put_page(page
, 1);
614 trace_f2fs_truncate_nodes_exit(dn
->inode
, ret
);
618 static int truncate_partial_nodes(struct dnode_of_data
*dn
,
619 struct f2fs_inode
*ri
, int *offset
, int depth
)
621 struct f2fs_sb_info
*sbi
= F2FS_SB(dn
->inode
->i_sb
);
622 struct page
*pages
[2];
629 nid
[0] = le32_to_cpu(ri
->i_nid
[offset
[0] - NODE_DIR1_BLOCK
]);
633 /* get indirect nodes in the path */
634 for (i
= 0; i
< depth
- 1; i
++) {
635 /* refernece count'll be increased */
636 pages
[i
] = get_node_page(sbi
, nid
[i
]);
637 if (IS_ERR(pages
[i
])) {
639 err
= PTR_ERR(pages
[i
]);
642 nid
[i
+ 1] = get_nid(pages
[i
], offset
[i
+ 1], false);
645 /* free direct nodes linked to a partial indirect node */
646 for (i
= offset
[depth
- 1]; i
< NIDS_PER_BLOCK
; i
++) {
647 child_nid
= get_nid(pages
[idx
], i
, false);
651 err
= truncate_dnode(dn
);
654 set_nid(pages
[idx
], i
, 0, false);
657 if (offset
[depth
- 1] == 0) {
658 dn
->node_page
= pages
[idx
];
662 f2fs_put_page(pages
[idx
], 1);
665 offset
[depth
- 1] = 0;
667 for (i
= depth
- 3; i
>= 0; i
--)
668 f2fs_put_page(pages
[i
], 1);
670 trace_f2fs_truncate_partial_nodes(dn
->inode
, nid
, depth
, err
);
676 * All the block addresses of data and nodes should be nullified.
678 int truncate_inode_blocks(struct inode
*inode
, pgoff_t from
)
680 struct f2fs_sb_info
*sbi
= F2FS_SB(inode
->i_sb
);
681 struct address_space
*node_mapping
= sbi
->node_inode
->i_mapping
;
682 int err
= 0, cont
= 1;
683 int level
, offset
[4], noffset
[4];
684 unsigned int nofs
= 0;
685 struct f2fs_node
*rn
;
686 struct dnode_of_data dn
;
689 trace_f2fs_truncate_inode_blocks_enter(inode
, from
);
691 level
= get_node_path(F2FS_I(inode
), from
, offset
, noffset
);
693 page
= get_node_page(sbi
, inode
->i_ino
);
695 trace_f2fs_truncate_inode_blocks_exit(inode
, PTR_ERR(page
));
696 return PTR_ERR(page
);
699 set_new_dnode(&dn
, inode
, page
, NULL
, 0);
702 rn
= F2FS_NODE(page
);
710 if (!offset
[level
- 1])
712 err
= truncate_partial_nodes(&dn
, &rn
->i
, offset
, level
);
713 if (err
< 0 && err
!= -ENOENT
)
715 nofs
+= 1 + NIDS_PER_BLOCK
;
718 nofs
= 5 + 2 * NIDS_PER_BLOCK
;
719 if (!offset
[level
- 1])
721 err
= truncate_partial_nodes(&dn
, &rn
->i
, offset
, level
);
722 if (err
< 0 && err
!= -ENOENT
)
731 dn
.nid
= le32_to_cpu(rn
->i
.i_nid
[offset
[0] - NODE_DIR1_BLOCK
]);
733 case NODE_DIR1_BLOCK
:
734 case NODE_DIR2_BLOCK
:
735 err
= truncate_dnode(&dn
);
738 case NODE_IND1_BLOCK
:
739 case NODE_IND2_BLOCK
:
740 err
= truncate_nodes(&dn
, nofs
, offset
[1], 2);
743 case NODE_DIND_BLOCK
:
744 err
= truncate_nodes(&dn
, nofs
, offset
[1], 3);
751 if (err
< 0 && err
!= -ENOENT
)
753 if (offset
[1] == 0 &&
754 rn
->i
.i_nid
[offset
[0] - NODE_DIR1_BLOCK
]) {
756 if (page
->mapping
!= node_mapping
) {
757 f2fs_put_page(page
, 1);
760 wait_on_page_writeback(page
);
761 rn
->i
.i_nid
[offset
[0] - NODE_DIR1_BLOCK
] = 0;
762 set_page_dirty(page
);
770 f2fs_put_page(page
, 0);
771 trace_f2fs_truncate_inode_blocks_exit(inode
, err
);
772 return err
> 0 ? 0 : err
;
775 int truncate_xattr_node(struct inode
*inode
, struct page
*page
)
777 struct f2fs_sb_info
*sbi
= F2FS_SB(inode
->i_sb
);
778 nid_t nid
= F2FS_I(inode
)->i_xattr_nid
;
779 struct dnode_of_data dn
;
785 npage
= get_node_page(sbi
, nid
);
787 return PTR_ERR(npage
);
789 F2FS_I(inode
)->i_xattr_nid
= 0;
791 /* need to do checkpoint during fsync */
792 F2FS_I(inode
)->xattr_ver
= cur_cp_version(F2FS_CKPT(sbi
));
794 set_new_dnode(&dn
, inode
, page
, npage
, nid
);
797 dn
.inode_page_locked
= 1;
803 * Caller should grab and release a mutex by calling mutex_lock_op() and
806 int remove_inode_page(struct inode
*inode
)
808 struct f2fs_sb_info
*sbi
= F2FS_SB(inode
->i_sb
);
810 nid_t ino
= inode
->i_ino
;
811 struct dnode_of_data dn
;
814 page
= get_node_page(sbi
, ino
);
816 return PTR_ERR(page
);
818 err
= truncate_xattr_node(inode
, page
);
820 f2fs_put_page(page
, 1);
824 /* 0 is possible, after f2fs_new_inode() is failed */
825 BUG_ON(inode
->i_blocks
!= 0 && inode
->i_blocks
!= 1);
826 set_new_dnode(&dn
, inode
, page
, page
, ino
);
831 struct page
*new_inode_page(struct inode
*inode
, const struct qstr
*name
)
833 struct dnode_of_data dn
;
835 /* allocate inode page for new inode */
836 set_new_dnode(&dn
, inode
, NULL
, NULL
, inode
->i_ino
);
838 /* caller should f2fs_put_page(page, 1); */
839 return new_node_page(&dn
, 0, NULL
);
842 struct page
*new_node_page(struct dnode_of_data
*dn
,
843 unsigned int ofs
, struct page
*ipage
)
845 struct f2fs_sb_info
*sbi
= F2FS_SB(dn
->inode
->i_sb
);
846 struct address_space
*mapping
= sbi
->node_inode
->i_mapping
;
847 struct node_info old_ni
, new_ni
;
851 if (is_inode_flag_set(F2FS_I(dn
->inode
), FI_NO_ALLOC
))
852 return ERR_PTR(-EPERM
);
854 page
= grab_cache_page(mapping
, dn
->nid
);
856 return ERR_PTR(-ENOMEM
);
858 if (!inc_valid_node_count(sbi
, dn
->inode
, 1)) {
863 get_node_info(sbi
, dn
->nid
, &old_ni
);
865 /* Reinitialize old_ni with new node page */
866 BUG_ON(old_ni
.blk_addr
!= NULL_ADDR
);
868 new_ni
.ino
= dn
->inode
->i_ino
;
869 set_node_addr(sbi
, &new_ni
, NEW_ADDR
);
871 fill_node_footer(page
, dn
->nid
, dn
->inode
->i_ino
, ofs
, true);
872 set_cold_node(dn
->inode
, page
);
873 SetPageUptodate(page
);
874 set_page_dirty(page
);
876 if (ofs
== XATTR_NODE_OFFSET
)
877 F2FS_I(dn
->inode
)->i_xattr_nid
= dn
->nid
;
879 dn
->node_page
= page
;
881 update_inode(dn
->inode
, ipage
);
885 inc_valid_inode_count(sbi
);
890 clear_node_page_dirty(page
);
891 f2fs_put_page(page
, 1);
896 * Caller should do after getting the following values.
897 * 0: f2fs_put_page(page, 0)
898 * LOCKED_PAGE: f2fs_put_page(page, 1)
901 static int read_node_page(struct page
*page
, int type
)
903 struct f2fs_sb_info
*sbi
= F2FS_SB(page
->mapping
->host
->i_sb
);
906 get_node_info(sbi
, page
->index
, &ni
);
908 if (ni
.blk_addr
== NULL_ADDR
) {
909 f2fs_put_page(page
, 1);
913 if (PageUptodate(page
))
916 return f2fs_readpage(sbi
, page
, ni
.blk_addr
, type
);
920 * Readahead a node page
922 void ra_node_page(struct f2fs_sb_info
*sbi
, nid_t nid
)
924 struct address_space
*mapping
= sbi
->node_inode
->i_mapping
;
928 apage
= find_get_page(mapping
, nid
);
929 if (apage
&& PageUptodate(apage
)) {
930 f2fs_put_page(apage
, 0);
933 f2fs_put_page(apage
, 0);
935 apage
= grab_cache_page(mapping
, nid
);
939 err
= read_node_page(apage
, READA
);
941 f2fs_put_page(apage
, 0);
942 else if (err
== LOCKED_PAGE
)
943 f2fs_put_page(apage
, 1);
946 struct page
*get_node_page(struct f2fs_sb_info
*sbi
, pgoff_t nid
)
948 struct address_space
*mapping
= sbi
->node_inode
->i_mapping
;
952 page
= grab_cache_page(mapping
, nid
);
954 return ERR_PTR(-ENOMEM
);
956 err
= read_node_page(page
, READ_SYNC
);
959 else if (err
== LOCKED_PAGE
)
963 if (!PageUptodate(page
)) {
964 f2fs_put_page(page
, 1);
965 return ERR_PTR(-EIO
);
967 if (page
->mapping
!= mapping
) {
968 f2fs_put_page(page
, 1);
972 BUG_ON(nid
!= nid_of_node(page
));
973 mark_page_accessed(page
);
978 * Return a locked page for the desired node page.
979 * And, readahead MAX_RA_NODE number of node pages.
981 struct page
*get_node_page_ra(struct page
*parent
, int start
)
983 struct f2fs_sb_info
*sbi
= F2FS_SB(parent
->mapping
->host
->i_sb
);
984 struct address_space
*mapping
= sbi
->node_inode
->i_mapping
;
985 struct blk_plug plug
;
990 /* First, try getting the desired direct node. */
991 nid
= get_nid(parent
, start
, false);
993 return ERR_PTR(-ENOENT
);
995 page
= grab_cache_page(mapping
, nid
);
997 return ERR_PTR(-ENOMEM
);
999 err
= read_node_page(page
, READ_SYNC
);
1001 return ERR_PTR(err
);
1002 else if (err
== LOCKED_PAGE
)
1005 blk_start_plug(&plug
);
1007 /* Then, try readahead for siblings of the desired node */
1008 end
= start
+ MAX_RA_NODE
;
1009 end
= min(end
, NIDS_PER_BLOCK
);
1010 for (i
= start
+ 1; i
< end
; i
++) {
1011 nid
= get_nid(parent
, i
, false);
1014 ra_node_page(sbi
, nid
);
1017 blk_finish_plug(&plug
);
1020 if (page
->mapping
!= mapping
) {
1021 f2fs_put_page(page
, 1);
1025 if (!PageUptodate(page
)) {
1026 f2fs_put_page(page
, 1);
1027 return ERR_PTR(-EIO
);
1029 mark_page_accessed(page
);
1033 void sync_inode_page(struct dnode_of_data
*dn
)
1035 if (IS_INODE(dn
->node_page
) || dn
->inode_page
== dn
->node_page
) {
1036 update_inode(dn
->inode
, dn
->node_page
);
1037 } else if (dn
->inode_page
) {
1038 if (!dn
->inode_page_locked
)
1039 lock_page(dn
->inode_page
);
1040 update_inode(dn
->inode
, dn
->inode_page
);
1041 if (!dn
->inode_page_locked
)
1042 unlock_page(dn
->inode_page
);
1044 update_inode_page(dn
->inode
);
1048 int sync_node_pages(struct f2fs_sb_info
*sbi
, nid_t ino
,
1049 struct writeback_control
*wbc
)
1051 struct address_space
*mapping
= sbi
->node_inode
->i_mapping
;
1053 struct pagevec pvec
;
1054 int step
= ino
? 2 : 0;
1055 int nwritten
= 0, wrote
= 0;
1057 pagevec_init(&pvec
, 0);
1063 while (index
<= end
) {
1065 nr_pages
= pagevec_lookup_tag(&pvec
, mapping
, &index
,
1066 PAGECACHE_TAG_DIRTY
,
1067 min(end
- index
, (pgoff_t
)PAGEVEC_SIZE
-1) + 1);
1071 for (i
= 0; i
< nr_pages
; i
++) {
1072 struct page
*page
= pvec
.pages
[i
];
1075 * flushing sequence with step:
1080 if (step
== 0 && IS_DNODE(page
))
1082 if (step
== 1 && (!IS_DNODE(page
) ||
1083 is_cold_node(page
)))
1085 if (step
== 2 && (!IS_DNODE(page
) ||
1086 !is_cold_node(page
)))
1091 * we should not skip writing node pages.
1093 if (ino
&& ino_of_node(page
) == ino
)
1095 else if (!trylock_page(page
))
1098 if (unlikely(page
->mapping
!= mapping
)) {
1103 if (ino
&& ino_of_node(page
) != ino
)
1104 goto continue_unlock
;
1106 if (!PageDirty(page
)) {
1107 /* someone wrote it for us */
1108 goto continue_unlock
;
1111 if (!clear_page_dirty_for_io(page
))
1112 goto continue_unlock
;
1114 /* called by fsync() */
1115 if (ino
&& IS_DNODE(page
)) {
1116 int mark
= !is_checkpointed_node(sbi
, ino
);
1117 set_fsync_mark(page
, 1);
1119 set_dentry_mark(page
, mark
);
1122 set_fsync_mark(page
, 0);
1123 set_dentry_mark(page
, 0);
1125 mapping
->a_ops
->writepage(page
, wbc
);
1128 if (--wbc
->nr_to_write
== 0)
1131 pagevec_release(&pvec
);
1134 if (wbc
->nr_to_write
== 0) {
1146 f2fs_submit_bio(sbi
, NODE
, wbc
->sync_mode
== WB_SYNC_ALL
);
1151 static int f2fs_write_node_page(struct page
*page
,
1152 struct writeback_control
*wbc
)
1154 struct f2fs_sb_info
*sbi
= F2FS_SB(page
->mapping
->host
->i_sb
);
1157 struct node_info ni
;
1159 wait_on_page_writeback(page
);
1161 /* get old block addr of this node page */
1162 nid
= nid_of_node(page
);
1163 BUG_ON(page
->index
!= nid
);
1165 get_node_info(sbi
, nid
, &ni
);
1167 /* This page is already truncated */
1168 if (ni
.blk_addr
== NULL_ADDR
) {
1169 dec_page_count(sbi
, F2FS_DIRTY_NODES
);
1174 if (wbc
->for_reclaim
) {
1175 dec_page_count(sbi
, F2FS_DIRTY_NODES
);
1176 wbc
->pages_skipped
++;
1177 set_page_dirty(page
);
1178 return AOP_WRITEPAGE_ACTIVATE
;
1181 mutex_lock(&sbi
->node_write
);
1182 set_page_writeback(page
);
1183 write_node_page(sbi
, page
, nid
, ni
.blk_addr
, &new_addr
);
1184 set_node_addr(sbi
, &ni
, new_addr
);
1185 dec_page_count(sbi
, F2FS_DIRTY_NODES
);
1186 mutex_unlock(&sbi
->node_write
);
1192 * It is very important to gather dirty pages and write at once, so that we can
1193 * submit a big bio without interfering other data writes.
1194 * Be default, 512 pages (2MB) * 3 node types, is more reasonable.
1196 #define COLLECT_DIRTY_NODES 1536
1197 static int f2fs_write_node_pages(struct address_space
*mapping
,
1198 struct writeback_control
*wbc
)
1200 struct f2fs_sb_info
*sbi
= F2FS_SB(mapping
->host
->i_sb
);
1201 long nr_to_write
= wbc
->nr_to_write
;
1203 /* First check balancing cached NAT entries */
1204 if (try_to_free_nats(sbi
, NAT_ENTRY_PER_BLOCK
)) {
1205 f2fs_sync_fs(sbi
->sb
, true);
1209 /* collect a number of dirty node pages and write together */
1210 if (get_pages(sbi
, F2FS_DIRTY_NODES
) < COLLECT_DIRTY_NODES
)
1213 /* if mounting is failed, skip writing node pages */
1214 wbc
->nr_to_write
= 3 * max_hw_blocks(sbi
);
1215 sync_node_pages(sbi
, 0, wbc
);
1216 wbc
->nr_to_write
= nr_to_write
- (3 * max_hw_blocks(sbi
) -
1221 static int f2fs_set_node_page_dirty(struct page
*page
)
1223 struct address_space
*mapping
= page
->mapping
;
1224 struct f2fs_sb_info
*sbi
= F2FS_SB(mapping
->host
->i_sb
);
1226 SetPageUptodate(page
);
1227 if (!PageDirty(page
)) {
1228 __set_page_dirty_nobuffers(page
);
1229 inc_page_count(sbi
, F2FS_DIRTY_NODES
);
1230 SetPagePrivate(page
);
1236 static void f2fs_invalidate_node_page(struct page
*page
, unsigned int offset
,
1237 unsigned int length
)
1239 struct inode
*inode
= page
->mapping
->host
;
1240 struct f2fs_sb_info
*sbi
= F2FS_SB(inode
->i_sb
);
1241 if (PageDirty(page
))
1242 dec_page_count(sbi
, F2FS_DIRTY_NODES
);
1243 ClearPagePrivate(page
);
1246 static int f2fs_release_node_page(struct page
*page
, gfp_t wait
)
1248 ClearPagePrivate(page
);
1253 * Structure of the f2fs node operations
1255 const struct address_space_operations f2fs_node_aops
= {
1256 .writepage
= f2fs_write_node_page
,
1257 .writepages
= f2fs_write_node_pages
,
1258 .set_page_dirty
= f2fs_set_node_page_dirty
,
1259 .invalidatepage
= f2fs_invalidate_node_page
,
1260 .releasepage
= f2fs_release_node_page
,
1263 static struct free_nid
*__lookup_free_nid_list(nid_t n
, struct list_head
*head
)
1265 struct list_head
*this;
1267 list_for_each(this, head
) {
1268 i
= list_entry(this, struct free_nid
, list
);
1275 static void __del_from_free_nid_list(struct free_nid
*i
)
1278 kmem_cache_free(free_nid_slab
, i
);
1281 static int add_free_nid(struct f2fs_nm_info
*nm_i
, nid_t nid
, bool build
)
1284 struct nat_entry
*ne
;
1285 bool allocated
= false;
1287 if (nm_i
->fcnt
> 2 * MAX_FREE_NIDS
)
1290 /* 0 nid should not be used */
1297 /* do not add allocated nids */
1298 read_lock(&nm_i
->nat_tree_lock
);
1299 ne
= __lookup_nat_cache(nm_i
, nid
);
1300 if (ne
&& nat_get_blkaddr(ne
) != NULL_ADDR
)
1302 read_unlock(&nm_i
->nat_tree_lock
);
1306 i
= kmem_cache_alloc(free_nid_slab
, GFP_NOFS
);
1314 spin_lock(&nm_i
->free_nid_list_lock
);
1315 if (__lookup_free_nid_list(nid
, &nm_i
->free_nid_list
)) {
1316 spin_unlock(&nm_i
->free_nid_list_lock
);
1317 kmem_cache_free(free_nid_slab
, i
);
1320 list_add_tail(&i
->list
, &nm_i
->free_nid_list
);
1322 spin_unlock(&nm_i
->free_nid_list_lock
);
1326 static void remove_free_nid(struct f2fs_nm_info
*nm_i
, nid_t nid
)
1329 spin_lock(&nm_i
->free_nid_list_lock
);
1330 i
= __lookup_free_nid_list(nid
, &nm_i
->free_nid_list
);
1331 if (i
&& i
->state
== NID_NEW
) {
1332 __del_from_free_nid_list(i
);
1335 spin_unlock(&nm_i
->free_nid_list_lock
);
1338 static void scan_nat_page(struct f2fs_nm_info
*nm_i
,
1339 struct page
*nat_page
, nid_t start_nid
)
1341 struct f2fs_nat_block
*nat_blk
= page_address(nat_page
);
1345 i
= start_nid
% NAT_ENTRY_PER_BLOCK
;
1347 for (; i
< NAT_ENTRY_PER_BLOCK
; i
++, start_nid
++) {
1349 if (start_nid
>= nm_i
->max_nid
)
1352 blk_addr
= le32_to_cpu(nat_blk
->entries
[i
].block_addr
);
1353 BUG_ON(blk_addr
== NEW_ADDR
);
1354 if (blk_addr
== NULL_ADDR
) {
1355 if (add_free_nid(nm_i
, start_nid
, true) < 0)
1361 static void build_free_nids(struct f2fs_sb_info
*sbi
)
1363 struct f2fs_nm_info
*nm_i
= NM_I(sbi
);
1364 struct curseg_info
*curseg
= CURSEG_I(sbi
, CURSEG_HOT_DATA
);
1365 struct f2fs_summary_block
*sum
= curseg
->sum_blk
;
1367 nid_t nid
= nm_i
->next_scan_nid
;
1369 /* Enough entries */
1370 if (nm_i
->fcnt
> NAT_ENTRY_PER_BLOCK
)
1373 /* readahead nat pages to be scanned */
1374 ra_nat_pages(sbi
, nid
);
1377 struct page
*page
= get_current_nat_page(sbi
, nid
);
1379 scan_nat_page(nm_i
, page
, nid
);
1380 f2fs_put_page(page
, 1);
1382 nid
+= (NAT_ENTRY_PER_BLOCK
- (nid
% NAT_ENTRY_PER_BLOCK
));
1383 if (nid
>= nm_i
->max_nid
)
1386 if (i
++ == FREE_NID_PAGES
)
1390 /* go to the next free nat pages to find free nids abundantly */
1391 nm_i
->next_scan_nid
= nid
;
1393 /* find free nids from current sum_pages */
1394 mutex_lock(&curseg
->curseg_mutex
);
1395 for (i
= 0; i
< nats_in_cursum(sum
); i
++) {
1396 block_t addr
= le32_to_cpu(nat_in_journal(sum
, i
).block_addr
);
1397 nid
= le32_to_cpu(nid_in_journal(sum
, i
));
1398 if (addr
== NULL_ADDR
)
1399 add_free_nid(nm_i
, nid
, true);
1401 remove_free_nid(nm_i
, nid
);
1403 mutex_unlock(&curseg
->curseg_mutex
);
1407 * If this function returns success, caller can obtain a new nid
1408 * from second parameter of this function.
1409 * The returned nid could be used ino as well as nid when inode is created.
1411 bool alloc_nid(struct f2fs_sb_info
*sbi
, nid_t
*nid
)
1413 struct f2fs_nm_info
*nm_i
= NM_I(sbi
);
1414 struct free_nid
*i
= NULL
;
1415 struct list_head
*this;
1417 if (sbi
->total_valid_node_count
+ 1 >= nm_i
->max_nid
)
1420 spin_lock(&nm_i
->free_nid_list_lock
);
1422 /* We should not use stale free nids created by build_free_nids */
1423 if (nm_i
->fcnt
&& !sbi
->on_build_free_nids
) {
1424 BUG_ON(list_empty(&nm_i
->free_nid_list
));
1425 list_for_each(this, &nm_i
->free_nid_list
) {
1426 i
= list_entry(this, struct free_nid
, list
);
1427 if (i
->state
== NID_NEW
)
1431 BUG_ON(i
->state
!= NID_NEW
);
1433 i
->state
= NID_ALLOC
;
1435 spin_unlock(&nm_i
->free_nid_list_lock
);
1438 spin_unlock(&nm_i
->free_nid_list_lock
);
1440 /* Let's scan nat pages and its caches to get free nids */
1441 mutex_lock(&nm_i
->build_lock
);
1442 sbi
->on_build_free_nids
= 1;
1443 build_free_nids(sbi
);
1444 sbi
->on_build_free_nids
= 0;
1445 mutex_unlock(&nm_i
->build_lock
);
1450 * alloc_nid() should be called prior to this function.
1452 void alloc_nid_done(struct f2fs_sb_info
*sbi
, nid_t nid
)
1454 struct f2fs_nm_info
*nm_i
= NM_I(sbi
);
1457 spin_lock(&nm_i
->free_nid_list_lock
);
1458 i
= __lookup_free_nid_list(nid
, &nm_i
->free_nid_list
);
1459 BUG_ON(!i
|| i
->state
!= NID_ALLOC
);
1460 __del_from_free_nid_list(i
);
1461 spin_unlock(&nm_i
->free_nid_list_lock
);
1465 * alloc_nid() should be called prior to this function.
1467 void alloc_nid_failed(struct f2fs_sb_info
*sbi
, nid_t nid
)
1469 struct f2fs_nm_info
*nm_i
= NM_I(sbi
);
1475 spin_lock(&nm_i
->free_nid_list_lock
);
1476 i
= __lookup_free_nid_list(nid
, &nm_i
->free_nid_list
);
1477 BUG_ON(!i
|| i
->state
!= NID_ALLOC
);
1478 if (nm_i
->fcnt
> 2 * MAX_FREE_NIDS
) {
1479 __del_from_free_nid_list(i
);
1484 spin_unlock(&nm_i
->free_nid_list_lock
);
1487 void recover_node_page(struct f2fs_sb_info
*sbi
, struct page
*page
,
1488 struct f2fs_summary
*sum
, struct node_info
*ni
,
1489 block_t new_blkaddr
)
1491 rewrite_node_page(sbi
, page
, sum
, ni
->blk_addr
, new_blkaddr
);
1492 set_node_addr(sbi
, ni
, new_blkaddr
);
1493 clear_node_page_dirty(page
);
1496 int recover_inode_page(struct f2fs_sb_info
*sbi
, struct page
*page
)
1498 struct address_space
*mapping
= sbi
->node_inode
->i_mapping
;
1499 struct f2fs_node
*src
, *dst
;
1500 nid_t ino
= ino_of_node(page
);
1501 struct node_info old_ni
, new_ni
;
1504 ipage
= grab_cache_page(mapping
, ino
);
1508 /* Should not use this inode from free nid list */
1509 remove_free_nid(NM_I(sbi
), ino
);
1511 get_node_info(sbi
, ino
, &old_ni
);
1512 SetPageUptodate(ipage
);
1513 fill_node_footer(ipage
, ino
, ino
, 0, true);
1515 src
= F2FS_NODE(page
);
1516 dst
= F2FS_NODE(ipage
);
1518 memcpy(dst
, src
, (unsigned long)&src
->i
.i_ext
- (unsigned long)&src
->i
);
1520 dst
->i
.i_blocks
= cpu_to_le64(1);
1521 dst
->i
.i_links
= cpu_to_le32(1);
1522 dst
->i
.i_xattr_nid
= 0;
1527 if (!inc_valid_node_count(sbi
, NULL
, 1))
1529 set_node_addr(sbi
, &new_ni
, NEW_ADDR
);
1530 inc_valid_inode_count(sbi
);
1531 f2fs_put_page(ipage
, 1);
1535 int restore_node_summary(struct f2fs_sb_info
*sbi
,
1536 unsigned int segno
, struct f2fs_summary_block
*sum
)
1538 struct f2fs_node
*rn
;
1539 struct f2fs_summary
*sum_entry
;
1544 /* alloc temporal page for read node */
1545 page
= alloc_page(GFP_NOFS
| __GFP_ZERO
);
1550 /* scan the node segment */
1551 last_offset
= sbi
->blocks_per_seg
;
1552 addr
= START_BLOCK(sbi
, segno
);
1553 sum_entry
= &sum
->entries
[0];
1555 for (i
= 0; i
< last_offset
; i
++, sum_entry
++) {
1557 * In order to read next node page,
1558 * we must clear PageUptodate flag.
1560 ClearPageUptodate(page
);
1562 if (f2fs_readpage(sbi
, page
, addr
, READ_SYNC
))
1566 rn
= F2FS_NODE(page
);
1567 sum_entry
->nid
= rn
->footer
.nid
;
1568 sum_entry
->version
= 0;
1569 sum_entry
->ofs_in_node
= 0;
1574 __free_pages(page
, 0);
1578 static bool flush_nats_in_journal(struct f2fs_sb_info
*sbi
)
1580 struct f2fs_nm_info
*nm_i
= NM_I(sbi
);
1581 struct curseg_info
*curseg
= CURSEG_I(sbi
, CURSEG_HOT_DATA
);
1582 struct f2fs_summary_block
*sum
= curseg
->sum_blk
;
1585 mutex_lock(&curseg
->curseg_mutex
);
1587 if (nats_in_cursum(sum
) < NAT_JOURNAL_ENTRIES
) {
1588 mutex_unlock(&curseg
->curseg_mutex
);
1592 for (i
= 0; i
< nats_in_cursum(sum
); i
++) {
1593 struct nat_entry
*ne
;
1594 struct f2fs_nat_entry raw_ne
;
1595 nid_t nid
= le32_to_cpu(nid_in_journal(sum
, i
));
1597 raw_ne
= nat_in_journal(sum
, i
);
1599 write_lock(&nm_i
->nat_tree_lock
);
1600 ne
= __lookup_nat_cache(nm_i
, nid
);
1602 __set_nat_cache_dirty(nm_i
, ne
);
1603 write_unlock(&nm_i
->nat_tree_lock
);
1606 ne
= grab_nat_entry(nm_i
, nid
);
1608 write_unlock(&nm_i
->nat_tree_lock
);
1611 nat_set_blkaddr(ne
, le32_to_cpu(raw_ne
.block_addr
));
1612 nat_set_ino(ne
, le32_to_cpu(raw_ne
.ino
));
1613 nat_set_version(ne
, raw_ne
.version
);
1614 __set_nat_cache_dirty(nm_i
, ne
);
1615 write_unlock(&nm_i
->nat_tree_lock
);
1617 update_nats_in_cursum(sum
, -i
);
1618 mutex_unlock(&curseg
->curseg_mutex
);
1623 * This function is called during the checkpointing process.
1625 void flush_nat_entries(struct f2fs_sb_info
*sbi
)
1627 struct f2fs_nm_info
*nm_i
= NM_I(sbi
);
1628 struct curseg_info
*curseg
= CURSEG_I(sbi
, CURSEG_HOT_DATA
);
1629 struct f2fs_summary_block
*sum
= curseg
->sum_blk
;
1630 struct list_head
*cur
, *n
;
1631 struct page
*page
= NULL
;
1632 struct f2fs_nat_block
*nat_blk
= NULL
;
1633 nid_t start_nid
= 0, end_nid
= 0;
1636 flushed
= flush_nats_in_journal(sbi
);
1639 mutex_lock(&curseg
->curseg_mutex
);
1641 /* 1) flush dirty nat caches */
1642 list_for_each_safe(cur
, n
, &nm_i
->dirty_nat_entries
) {
1643 struct nat_entry
*ne
;
1645 struct f2fs_nat_entry raw_ne
;
1647 block_t new_blkaddr
;
1649 ne
= list_entry(cur
, struct nat_entry
, list
);
1650 nid
= nat_get_nid(ne
);
1652 if (nat_get_blkaddr(ne
) == NEW_ADDR
)
1657 /* if there is room for nat enries in curseg->sumpage */
1658 offset
= lookup_journal_in_cursum(sum
, NAT_JOURNAL
, nid
, 1);
1660 raw_ne
= nat_in_journal(sum
, offset
);
1664 if (!page
|| (start_nid
> nid
|| nid
> end_nid
)) {
1666 f2fs_put_page(page
, 1);
1669 start_nid
= START_NID(nid
);
1670 end_nid
= start_nid
+ NAT_ENTRY_PER_BLOCK
- 1;
1673 * get nat block with dirty flag, increased reference
1674 * count, mapped and lock
1676 page
= get_next_nat_page(sbi
, start_nid
);
1677 nat_blk
= page_address(page
);
1681 raw_ne
= nat_blk
->entries
[nid
- start_nid
];
1683 new_blkaddr
= nat_get_blkaddr(ne
);
1685 raw_ne
.ino
= cpu_to_le32(nat_get_ino(ne
));
1686 raw_ne
.block_addr
= cpu_to_le32(new_blkaddr
);
1687 raw_ne
.version
= nat_get_version(ne
);
1690 nat_blk
->entries
[nid
- start_nid
] = raw_ne
;
1692 nat_in_journal(sum
, offset
) = raw_ne
;
1693 nid_in_journal(sum
, offset
) = cpu_to_le32(nid
);
1696 if (nat_get_blkaddr(ne
) == NULL_ADDR
&&
1697 add_free_nid(NM_I(sbi
), nid
, false) <= 0) {
1698 write_lock(&nm_i
->nat_tree_lock
);
1699 __del_from_nat_cache(nm_i
, ne
);
1700 write_unlock(&nm_i
->nat_tree_lock
);
1702 write_lock(&nm_i
->nat_tree_lock
);
1703 __clear_nat_cache_dirty(nm_i
, ne
);
1704 ne
->checkpointed
= true;
1705 write_unlock(&nm_i
->nat_tree_lock
);
1709 mutex_unlock(&curseg
->curseg_mutex
);
1710 f2fs_put_page(page
, 1);
1712 /* 2) shrink nat caches if necessary */
1713 try_to_free_nats(sbi
, nm_i
->nat_cnt
- NM_WOUT_THRESHOLD
);
1716 static int init_node_manager(struct f2fs_sb_info
*sbi
)
1718 struct f2fs_super_block
*sb_raw
= F2FS_RAW_SUPER(sbi
);
1719 struct f2fs_nm_info
*nm_i
= NM_I(sbi
);
1720 unsigned char *version_bitmap
;
1721 unsigned int nat_segs
, nat_blocks
;
1723 nm_i
->nat_blkaddr
= le32_to_cpu(sb_raw
->nat_blkaddr
);
1725 /* segment_count_nat includes pair segment so divide to 2. */
1726 nat_segs
= le32_to_cpu(sb_raw
->segment_count_nat
) >> 1;
1727 nat_blocks
= nat_segs
<< le32_to_cpu(sb_raw
->log_blocks_per_seg
);
1728 nm_i
->max_nid
= NAT_ENTRY_PER_BLOCK
* nat_blocks
;
1732 INIT_LIST_HEAD(&nm_i
->free_nid_list
);
1733 INIT_RADIX_TREE(&nm_i
->nat_root
, GFP_ATOMIC
);
1734 INIT_LIST_HEAD(&nm_i
->nat_entries
);
1735 INIT_LIST_HEAD(&nm_i
->dirty_nat_entries
);
1737 mutex_init(&nm_i
->build_lock
);
1738 spin_lock_init(&nm_i
->free_nid_list_lock
);
1739 rwlock_init(&nm_i
->nat_tree_lock
);
1741 nm_i
->next_scan_nid
= le32_to_cpu(sbi
->ckpt
->next_free_nid
);
1742 nm_i
->bitmap_size
= __bitmap_size(sbi
, NAT_BITMAP
);
1743 version_bitmap
= __bitmap_ptr(sbi
, NAT_BITMAP
);
1744 if (!version_bitmap
)
1747 nm_i
->nat_bitmap
= kmemdup(version_bitmap
, nm_i
->bitmap_size
,
1749 if (!nm_i
->nat_bitmap
)
1754 int build_node_manager(struct f2fs_sb_info
*sbi
)
1758 sbi
->nm_info
= kzalloc(sizeof(struct f2fs_nm_info
), GFP_KERNEL
);
1762 err
= init_node_manager(sbi
);
1766 build_free_nids(sbi
);
1770 void destroy_node_manager(struct f2fs_sb_info
*sbi
)
1772 struct f2fs_nm_info
*nm_i
= NM_I(sbi
);
1773 struct free_nid
*i
, *next_i
;
1774 struct nat_entry
*natvec
[NATVEC_SIZE
];
1781 /* destroy free nid list */
1782 spin_lock(&nm_i
->free_nid_list_lock
);
1783 list_for_each_entry_safe(i
, next_i
, &nm_i
->free_nid_list
, list
) {
1784 BUG_ON(i
->state
== NID_ALLOC
);
1785 __del_from_free_nid_list(i
);
1789 spin_unlock(&nm_i
->free_nid_list_lock
);
1791 /* destroy nat cache */
1792 write_lock(&nm_i
->nat_tree_lock
);
1793 while ((found
= __gang_lookup_nat_cache(nm_i
,
1794 nid
, NATVEC_SIZE
, natvec
))) {
1796 for (idx
= 0; idx
< found
; idx
++) {
1797 struct nat_entry
*e
= natvec
[idx
];
1798 nid
= nat_get_nid(e
) + 1;
1799 __del_from_nat_cache(nm_i
, e
);
1802 BUG_ON(nm_i
->nat_cnt
);
1803 write_unlock(&nm_i
->nat_tree_lock
);
1805 kfree(nm_i
->nat_bitmap
);
1806 sbi
->nm_info
= NULL
;
1810 int __init
create_node_manager_caches(void)
1812 nat_entry_slab
= f2fs_kmem_cache_create("nat_entry",
1813 sizeof(struct nat_entry
), NULL
);
1814 if (!nat_entry_slab
)
1817 free_nid_slab
= f2fs_kmem_cache_create("free_nid",
1818 sizeof(struct free_nid
), NULL
);
1819 if (!free_nid_slab
) {
1820 kmem_cache_destroy(nat_entry_slab
);
1826 void destroy_node_manager_caches(void)
1828 kmem_cache_destroy(free_nid_slab
);
1829 kmem_cache_destroy(nat_entry_slab
);