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>
23 static struct kmem_cache
*nat_entry_slab
;
24 static struct kmem_cache
*free_nid_slab
;
26 static void clear_node_page_dirty(struct page
*page
)
28 struct address_space
*mapping
= page
->mapping
;
29 struct f2fs_sb_info
*sbi
= F2FS_SB(mapping
->host
->i_sb
);
30 unsigned int long flags
;
32 if (PageDirty(page
)) {
33 spin_lock_irqsave(&mapping
->tree_lock
, flags
);
34 radix_tree_tag_clear(&mapping
->page_tree
,
37 spin_unlock_irqrestore(&mapping
->tree_lock
, flags
);
39 clear_page_dirty_for_io(page
);
40 dec_page_count(sbi
, F2FS_DIRTY_NODES
);
42 ClearPageUptodate(page
);
45 static struct page
*get_current_nat_page(struct f2fs_sb_info
*sbi
, nid_t nid
)
47 pgoff_t index
= current_nat_addr(sbi
, nid
);
48 return get_meta_page(sbi
, index
);
51 static struct page
*get_next_nat_page(struct f2fs_sb_info
*sbi
, nid_t nid
)
53 struct page
*src_page
;
54 struct page
*dst_page
;
59 struct f2fs_nm_info
*nm_i
= NM_I(sbi
);
61 src_off
= current_nat_addr(sbi
, nid
);
62 dst_off
= next_nat_addr(sbi
, src_off
);
64 /* get current nat block page with lock */
65 src_page
= get_meta_page(sbi
, src_off
);
67 /* Dirty src_page means that it is already the new target NAT page. */
68 if (PageDirty(src_page
))
71 dst_page
= grab_meta_page(sbi
, dst_off
);
73 src_addr
= page_address(src_page
);
74 dst_addr
= page_address(dst_page
);
75 memcpy(dst_addr
, src_addr
, PAGE_CACHE_SIZE
);
76 set_page_dirty(dst_page
);
77 f2fs_put_page(src_page
, 1);
79 set_to_next_nat(nm_i
, nid
);
87 static void ra_nat_pages(struct f2fs_sb_info
*sbi
, int nid
)
89 struct address_space
*mapping
= sbi
->meta_inode
->i_mapping
;
90 struct f2fs_nm_info
*nm_i
= NM_I(sbi
);
95 for (i
= 0; i
< FREE_NID_PAGES
; i
++, nid
+= NAT_ENTRY_PER_BLOCK
) {
96 if (nid
>= nm_i
->max_nid
)
98 index
= current_nat_addr(sbi
, nid
);
100 page
= grab_cache_page(mapping
, index
);
103 if (PageUptodate(page
)) {
104 f2fs_put_page(page
, 1);
107 if (f2fs_readpage(sbi
, page
, index
, READ
))
110 f2fs_put_page(page
, 0);
114 static struct nat_entry
*__lookup_nat_cache(struct f2fs_nm_info
*nm_i
, nid_t n
)
116 return radix_tree_lookup(&nm_i
->nat_root
, n
);
119 static unsigned int __gang_lookup_nat_cache(struct f2fs_nm_info
*nm_i
,
120 nid_t start
, unsigned int nr
, struct nat_entry
**ep
)
122 return radix_tree_gang_lookup(&nm_i
->nat_root
, (void **)ep
, start
, nr
);
125 static void __del_from_nat_cache(struct f2fs_nm_info
*nm_i
, struct nat_entry
*e
)
128 radix_tree_delete(&nm_i
->nat_root
, nat_get_nid(e
));
130 kmem_cache_free(nat_entry_slab
, e
);
133 int is_checkpointed_node(struct f2fs_sb_info
*sbi
, nid_t nid
)
135 struct f2fs_nm_info
*nm_i
= NM_I(sbi
);
139 read_lock(&nm_i
->nat_tree_lock
);
140 e
= __lookup_nat_cache(nm_i
, nid
);
141 if (e
&& !e
->checkpointed
)
143 read_unlock(&nm_i
->nat_tree_lock
);
147 static struct nat_entry
*grab_nat_entry(struct f2fs_nm_info
*nm_i
, nid_t nid
)
149 struct nat_entry
*new;
151 new = kmem_cache_alloc(nat_entry_slab
, GFP_ATOMIC
);
154 if (radix_tree_insert(&nm_i
->nat_root
, nid
, new)) {
155 kmem_cache_free(nat_entry_slab
, new);
158 memset(new, 0, sizeof(struct nat_entry
));
159 nat_set_nid(new, nid
);
160 list_add_tail(&new->list
, &nm_i
->nat_entries
);
165 static void cache_nat_entry(struct f2fs_nm_info
*nm_i
, nid_t nid
,
166 struct f2fs_nat_entry
*ne
)
170 write_lock(&nm_i
->nat_tree_lock
);
171 e
= __lookup_nat_cache(nm_i
, nid
);
173 e
= grab_nat_entry(nm_i
, nid
);
175 write_unlock(&nm_i
->nat_tree_lock
);
178 nat_set_blkaddr(e
, le32_to_cpu(ne
->block_addr
));
179 nat_set_ino(e
, le32_to_cpu(ne
->ino
));
180 nat_set_version(e
, ne
->version
);
181 e
->checkpointed
= true;
183 write_unlock(&nm_i
->nat_tree_lock
);
186 static void set_node_addr(struct f2fs_sb_info
*sbi
, struct node_info
*ni
,
189 struct f2fs_nm_info
*nm_i
= NM_I(sbi
);
192 write_lock(&nm_i
->nat_tree_lock
);
193 e
= __lookup_nat_cache(nm_i
, ni
->nid
);
195 e
= grab_nat_entry(nm_i
, ni
->nid
);
197 write_unlock(&nm_i
->nat_tree_lock
);
201 e
->checkpointed
= true;
202 BUG_ON(ni
->blk_addr
== NEW_ADDR
);
203 } else if (new_blkaddr
== NEW_ADDR
) {
205 * when nid is reallocated,
206 * previous nat entry can be remained in nat cache.
207 * So, reinitialize it with new information.
210 BUG_ON(ni
->blk_addr
!= NULL_ADDR
);
213 if (new_blkaddr
== NEW_ADDR
)
214 e
->checkpointed
= false;
217 BUG_ON(nat_get_blkaddr(e
) != ni
->blk_addr
);
218 BUG_ON(nat_get_blkaddr(e
) == NULL_ADDR
&&
219 new_blkaddr
== NULL_ADDR
);
220 BUG_ON(nat_get_blkaddr(e
) == NEW_ADDR
&&
221 new_blkaddr
== NEW_ADDR
);
222 BUG_ON(nat_get_blkaddr(e
) != NEW_ADDR
&&
223 nat_get_blkaddr(e
) != NULL_ADDR
&&
224 new_blkaddr
== NEW_ADDR
);
226 /* increament version no as node is removed */
227 if (nat_get_blkaddr(e
) != NEW_ADDR
&& new_blkaddr
== NULL_ADDR
) {
228 unsigned char version
= nat_get_version(e
);
229 nat_set_version(e
, inc_node_version(version
));
233 nat_set_blkaddr(e
, new_blkaddr
);
234 __set_nat_cache_dirty(nm_i
, e
);
235 write_unlock(&nm_i
->nat_tree_lock
);
238 static int try_to_free_nats(struct f2fs_sb_info
*sbi
, int nr_shrink
)
240 struct f2fs_nm_info
*nm_i
= NM_I(sbi
);
242 if (nm_i
->nat_cnt
< 2 * NM_WOUT_THRESHOLD
)
245 write_lock(&nm_i
->nat_tree_lock
);
246 while (nr_shrink
&& !list_empty(&nm_i
->nat_entries
)) {
247 struct nat_entry
*ne
;
248 ne
= list_first_entry(&nm_i
->nat_entries
,
249 struct nat_entry
, list
);
250 __del_from_nat_cache(nm_i
, ne
);
253 write_unlock(&nm_i
->nat_tree_lock
);
258 * This function returns always success
260 void get_node_info(struct f2fs_sb_info
*sbi
, nid_t nid
, struct node_info
*ni
)
262 struct f2fs_nm_info
*nm_i
= NM_I(sbi
);
263 struct curseg_info
*curseg
= CURSEG_I(sbi
, CURSEG_HOT_DATA
);
264 struct f2fs_summary_block
*sum
= curseg
->sum_blk
;
265 nid_t start_nid
= START_NID(nid
);
266 struct f2fs_nat_block
*nat_blk
;
267 struct page
*page
= NULL
;
268 struct f2fs_nat_entry ne
;
272 memset(&ne
, 0, sizeof(struct f2fs_nat_entry
));
275 /* Check nat cache */
276 read_lock(&nm_i
->nat_tree_lock
);
277 e
= __lookup_nat_cache(nm_i
, nid
);
279 ni
->ino
= nat_get_ino(e
);
280 ni
->blk_addr
= nat_get_blkaddr(e
);
281 ni
->version
= nat_get_version(e
);
283 read_unlock(&nm_i
->nat_tree_lock
);
287 /* Check current segment summary */
288 mutex_lock(&curseg
->curseg_mutex
);
289 i
= lookup_journal_in_cursum(sum
, NAT_JOURNAL
, nid
, 0);
291 ne
= nat_in_journal(sum
, i
);
292 node_info_from_raw_nat(ni
, &ne
);
294 mutex_unlock(&curseg
->curseg_mutex
);
298 /* Fill node_info from nat page */
299 page
= get_current_nat_page(sbi
, start_nid
);
300 nat_blk
= (struct f2fs_nat_block
*)page_address(page
);
301 ne
= nat_blk
->entries
[nid
- start_nid
];
302 node_info_from_raw_nat(ni
, &ne
);
303 f2fs_put_page(page
, 1);
305 /* cache nat entry */
306 cache_nat_entry(NM_I(sbi
), nid
, &ne
);
310 * The maximum depth is four.
311 * Offset[0] will have raw inode offset.
313 static int get_node_path(long block
, int offset
[4], unsigned int noffset
[4])
315 const long direct_index
= ADDRS_PER_INODE
;
316 const long direct_blks
= ADDRS_PER_BLOCK
;
317 const long dptrs_per_blk
= NIDS_PER_BLOCK
;
318 const long indirect_blks
= ADDRS_PER_BLOCK
* NIDS_PER_BLOCK
;
319 const long dindirect_blks
= indirect_blks
* NIDS_PER_BLOCK
;
325 if (block
< direct_index
) {
329 block
-= direct_index
;
330 if (block
< direct_blks
) {
331 offset
[n
++] = NODE_DIR1_BLOCK
;
337 block
-= direct_blks
;
338 if (block
< direct_blks
) {
339 offset
[n
++] = NODE_DIR2_BLOCK
;
345 block
-= direct_blks
;
346 if (block
< indirect_blks
) {
347 offset
[n
++] = NODE_IND1_BLOCK
;
349 offset
[n
++] = block
/ direct_blks
;
350 noffset
[n
] = 4 + offset
[n
- 1];
351 offset
[n
] = block
% direct_blks
;
355 block
-= indirect_blks
;
356 if (block
< indirect_blks
) {
357 offset
[n
++] = NODE_IND2_BLOCK
;
358 noffset
[n
] = 4 + dptrs_per_blk
;
359 offset
[n
++] = block
/ direct_blks
;
360 noffset
[n
] = 5 + dptrs_per_blk
+ offset
[n
- 1];
361 offset
[n
] = block
% direct_blks
;
365 block
-= indirect_blks
;
366 if (block
< dindirect_blks
) {
367 offset
[n
++] = NODE_DIND_BLOCK
;
368 noffset
[n
] = 5 + (dptrs_per_blk
* 2);
369 offset
[n
++] = block
/ indirect_blks
;
370 noffset
[n
] = 6 + (dptrs_per_blk
* 2) +
371 offset
[n
- 1] * (dptrs_per_blk
+ 1);
372 offset
[n
++] = (block
/ direct_blks
) % dptrs_per_blk
;
373 noffset
[n
] = 7 + (dptrs_per_blk
* 2) +
374 offset
[n
- 2] * (dptrs_per_blk
+ 1) +
376 offset
[n
] = block
% direct_blks
;
387 * Caller should call f2fs_put_dnode(dn).
388 * Also, it should grab and release a mutex by calling mutex_lock_op() and
389 * mutex_unlock_op() only if ro is not set RDONLY_NODE.
390 * In the case of RDONLY_NODE, we don't need to care about mutex.
392 int get_dnode_of_data(struct dnode_of_data
*dn
, pgoff_t index
, int mode
)
394 struct f2fs_sb_info
*sbi
= F2FS_SB(dn
->inode
->i_sb
);
395 struct page
*npage
[4];
398 unsigned int noffset
[4];
403 level
= get_node_path(index
, offset
, noffset
);
405 nids
[0] = dn
->inode
->i_ino
;
406 npage
[0] = get_node_page(sbi
, nids
[0]);
407 if (IS_ERR(npage
[0]))
408 return PTR_ERR(npage
[0]);
412 nids
[1] = get_nid(parent
, offset
[0], true);
413 dn
->inode_page
= npage
[0];
414 dn
->inode_page_locked
= true;
416 /* get indirect or direct nodes */
417 for (i
= 1; i
<= level
; i
++) {
420 if (!nids
[i
] && mode
== ALLOC_NODE
) {
422 if (!alloc_nid(sbi
, &(nids
[i
]))) {
428 npage
[i
] = new_node_page(dn
, noffset
[i
]);
429 if (IS_ERR(npage
[i
])) {
430 alloc_nid_failed(sbi
, nids
[i
]);
431 err
= PTR_ERR(npage
[i
]);
435 set_nid(parent
, offset
[i
- 1], nids
[i
], i
== 1);
436 alloc_nid_done(sbi
, nids
[i
]);
438 } else if (mode
== LOOKUP_NODE_RA
&& i
== level
&& level
> 1) {
439 npage
[i
] = get_node_page_ra(parent
, offset
[i
- 1]);
440 if (IS_ERR(npage
[i
])) {
441 err
= PTR_ERR(npage
[i
]);
447 dn
->inode_page_locked
= false;
450 f2fs_put_page(parent
, 1);
454 npage
[i
] = get_node_page(sbi
, nids
[i
]);
455 if (IS_ERR(npage
[i
])) {
456 err
= PTR_ERR(npage
[i
]);
457 f2fs_put_page(npage
[0], 0);
463 nids
[i
+ 1] = get_nid(parent
, offset
[i
], false);
466 dn
->nid
= nids
[level
];
467 dn
->ofs_in_node
= offset
[level
];
468 dn
->node_page
= npage
[level
];
469 dn
->data_blkaddr
= datablock_addr(dn
->node_page
, dn
->ofs_in_node
);
473 f2fs_put_page(parent
, 1);
475 f2fs_put_page(npage
[0], 0);
477 dn
->inode_page
= NULL
;
478 dn
->node_page
= NULL
;
482 static void truncate_node(struct dnode_of_data
*dn
)
484 struct f2fs_sb_info
*sbi
= F2FS_SB(dn
->inode
->i_sb
);
487 get_node_info(sbi
, dn
->nid
, &ni
);
488 if (dn
->inode
->i_blocks
== 0) {
489 BUG_ON(ni
.blk_addr
!= NULL_ADDR
);
492 BUG_ON(ni
.blk_addr
== NULL_ADDR
);
494 /* Deallocate node address */
495 invalidate_blocks(sbi
, ni
.blk_addr
);
496 dec_valid_node_count(sbi
, dn
->inode
, 1);
497 set_node_addr(sbi
, &ni
, NULL_ADDR
);
499 if (dn
->nid
== dn
->inode
->i_ino
) {
500 remove_orphan_inode(sbi
, dn
->nid
);
501 dec_valid_inode_count(sbi
);
506 clear_node_page_dirty(dn
->node_page
);
507 F2FS_SET_SB_DIRT(sbi
);
509 f2fs_put_page(dn
->node_page
, 1);
510 dn
->node_page
= NULL
;
513 static int truncate_dnode(struct dnode_of_data
*dn
)
515 struct f2fs_sb_info
*sbi
= F2FS_SB(dn
->inode
->i_sb
);
521 /* get direct node */
522 page
= get_node_page(sbi
, dn
->nid
);
523 if (IS_ERR(page
) && PTR_ERR(page
) == -ENOENT
)
525 else if (IS_ERR(page
))
526 return PTR_ERR(page
);
528 /* Make dnode_of_data for parameter */
529 dn
->node_page
= page
;
531 truncate_data_blocks(dn
);
536 static int truncate_nodes(struct dnode_of_data
*dn
, unsigned int nofs
,
539 struct f2fs_sb_info
*sbi
= F2FS_SB(dn
->inode
->i_sb
);
540 struct dnode_of_data rdn
= *dn
;
542 struct f2fs_node
*rn
;
544 unsigned int child_nofs
;
549 return NIDS_PER_BLOCK
+ 1;
551 page
= get_node_page(sbi
, dn
->nid
);
553 return PTR_ERR(page
);
555 rn
= (struct f2fs_node
*)page_address(page
);
557 for (i
= ofs
; i
< NIDS_PER_BLOCK
; i
++, freed
++) {
558 child_nid
= le32_to_cpu(rn
->in
.nid
[i
]);
562 ret
= truncate_dnode(&rdn
);
565 set_nid(page
, i
, 0, false);
568 child_nofs
= nofs
+ ofs
* (NIDS_PER_BLOCK
+ 1) + 1;
569 for (i
= ofs
; i
< NIDS_PER_BLOCK
; i
++) {
570 child_nid
= le32_to_cpu(rn
->in
.nid
[i
]);
571 if (child_nid
== 0) {
572 child_nofs
+= NIDS_PER_BLOCK
+ 1;
576 ret
= truncate_nodes(&rdn
, child_nofs
, 0, depth
- 1);
577 if (ret
== (NIDS_PER_BLOCK
+ 1)) {
578 set_nid(page
, i
, 0, false);
580 } else if (ret
< 0 && ret
!= -ENOENT
) {
588 /* remove current indirect node */
589 dn
->node_page
= page
;
593 f2fs_put_page(page
, 1);
598 f2fs_put_page(page
, 1);
602 static int truncate_partial_nodes(struct dnode_of_data
*dn
,
603 struct f2fs_inode
*ri
, int *offset
, int depth
)
605 struct f2fs_sb_info
*sbi
= F2FS_SB(dn
->inode
->i_sb
);
606 struct page
*pages
[2];
613 nid
[0] = le32_to_cpu(ri
->i_nid
[offset
[0] - NODE_DIR1_BLOCK
]);
617 /* get indirect nodes in the path */
618 for (i
= 0; i
< depth
- 1; i
++) {
619 /* refernece count'll be increased */
620 pages
[i
] = get_node_page(sbi
, nid
[i
]);
621 if (IS_ERR(pages
[i
])) {
623 err
= PTR_ERR(pages
[i
]);
626 nid
[i
+ 1] = get_nid(pages
[i
], offset
[i
+ 1], false);
629 /* free direct nodes linked to a partial indirect node */
630 for (i
= offset
[depth
- 1]; i
< NIDS_PER_BLOCK
; i
++) {
631 child_nid
= get_nid(pages
[idx
], i
, false);
635 err
= truncate_dnode(dn
);
638 set_nid(pages
[idx
], i
, 0, false);
641 if (offset
[depth
- 1] == 0) {
642 dn
->node_page
= pages
[idx
];
646 f2fs_put_page(pages
[idx
], 1);
649 offset
[depth
- 1] = 0;
651 for (i
= depth
- 3; i
>= 0; i
--)
652 f2fs_put_page(pages
[i
], 1);
657 * All the block addresses of data and nodes should be nullified.
659 int truncate_inode_blocks(struct inode
*inode
, pgoff_t from
)
661 struct f2fs_sb_info
*sbi
= F2FS_SB(inode
->i_sb
);
662 int err
= 0, cont
= 1;
663 int level
, offset
[4], noffset
[4];
664 unsigned int nofs
= 0;
665 struct f2fs_node
*rn
;
666 struct dnode_of_data dn
;
669 level
= get_node_path(from
, offset
, noffset
);
671 page
= get_node_page(sbi
, inode
->i_ino
);
673 return PTR_ERR(page
);
675 set_new_dnode(&dn
, inode
, page
, NULL
, 0);
678 rn
= page_address(page
);
686 if (!offset
[level
- 1])
688 err
= truncate_partial_nodes(&dn
, &rn
->i
, offset
, level
);
689 if (err
< 0 && err
!= -ENOENT
)
691 nofs
+= 1 + NIDS_PER_BLOCK
;
694 nofs
= 5 + 2 * NIDS_PER_BLOCK
;
695 if (!offset
[level
- 1])
697 err
= truncate_partial_nodes(&dn
, &rn
->i
, offset
, level
);
698 if (err
< 0 && err
!= -ENOENT
)
707 dn
.nid
= le32_to_cpu(rn
->i
.i_nid
[offset
[0] - NODE_DIR1_BLOCK
]);
709 case NODE_DIR1_BLOCK
:
710 case NODE_DIR2_BLOCK
:
711 err
= truncate_dnode(&dn
);
714 case NODE_IND1_BLOCK
:
715 case NODE_IND2_BLOCK
:
716 err
= truncate_nodes(&dn
, nofs
, offset
[1], 2);
719 case NODE_DIND_BLOCK
:
720 err
= truncate_nodes(&dn
, nofs
, offset
[1], 3);
727 if (err
< 0 && err
!= -ENOENT
)
729 if (offset
[1] == 0 &&
730 rn
->i
.i_nid
[offset
[0] - NODE_DIR1_BLOCK
]) {
732 wait_on_page_writeback(page
);
733 rn
->i
.i_nid
[offset
[0] - NODE_DIR1_BLOCK
] = 0;
734 set_page_dirty(page
);
742 f2fs_put_page(page
, 0);
743 return err
> 0 ? 0 : err
;
747 * Caller should grab and release a mutex by calling mutex_lock_op() and
750 int remove_inode_page(struct inode
*inode
)
752 struct f2fs_sb_info
*sbi
= F2FS_SB(inode
->i_sb
);
754 nid_t ino
= inode
->i_ino
;
755 struct dnode_of_data dn
;
757 page
= get_node_page(sbi
, ino
);
759 return PTR_ERR(page
);
761 if (F2FS_I(inode
)->i_xattr_nid
) {
762 nid_t nid
= F2FS_I(inode
)->i_xattr_nid
;
763 struct page
*npage
= get_node_page(sbi
, nid
);
766 return PTR_ERR(npage
);
768 F2FS_I(inode
)->i_xattr_nid
= 0;
769 set_new_dnode(&dn
, inode
, page
, npage
, nid
);
770 dn
.inode_page_locked
= 1;
774 /* 0 is possible, after f2fs_new_inode() is failed */
775 BUG_ON(inode
->i_blocks
!= 0 && inode
->i_blocks
!= 1);
776 set_new_dnode(&dn
, inode
, page
, page
, ino
);
781 int new_inode_page(struct inode
*inode
, const struct qstr
*name
)
784 struct dnode_of_data dn
;
786 /* allocate inode page for new inode */
787 set_new_dnode(&dn
, inode
, NULL
, NULL
, inode
->i_ino
);
788 page
= new_node_page(&dn
, 0);
789 init_dent_inode(name
, page
);
791 return PTR_ERR(page
);
792 f2fs_put_page(page
, 1);
796 struct page
*new_node_page(struct dnode_of_data
*dn
, unsigned int ofs
)
798 struct f2fs_sb_info
*sbi
= F2FS_SB(dn
->inode
->i_sb
);
799 struct address_space
*mapping
= sbi
->node_inode
->i_mapping
;
800 struct node_info old_ni
, new_ni
;
804 if (is_inode_flag_set(F2FS_I(dn
->inode
), FI_NO_ALLOC
))
805 return ERR_PTR(-EPERM
);
807 page
= grab_cache_page(mapping
, dn
->nid
);
809 return ERR_PTR(-ENOMEM
);
811 get_node_info(sbi
, dn
->nid
, &old_ni
);
813 SetPageUptodate(page
);
814 fill_node_footer(page
, dn
->nid
, dn
->inode
->i_ino
, ofs
, true);
816 /* Reinitialize old_ni with new node page */
817 BUG_ON(old_ni
.blk_addr
!= NULL_ADDR
);
819 new_ni
.ino
= dn
->inode
->i_ino
;
821 if (!inc_valid_node_count(sbi
, dn
->inode
, 1)) {
825 set_node_addr(sbi
, &new_ni
, NEW_ADDR
);
826 set_cold_node(dn
->inode
, page
);
828 dn
->node_page
= page
;
830 set_page_dirty(page
);
832 inc_valid_inode_count(sbi
);
837 clear_node_page_dirty(page
);
838 f2fs_put_page(page
, 1);
843 * Caller should do after getting the following values.
844 * 0: f2fs_put_page(page, 0)
845 * LOCKED_PAGE: f2fs_put_page(page, 1)
848 static int read_node_page(struct page
*page
, int type
)
850 struct f2fs_sb_info
*sbi
= F2FS_SB(page
->mapping
->host
->i_sb
);
853 get_node_info(sbi
, page
->index
, &ni
);
855 if (ni
.blk_addr
== NULL_ADDR
) {
856 f2fs_put_page(page
, 1);
860 if (PageUptodate(page
))
863 return f2fs_readpage(sbi
, page
, ni
.blk_addr
, type
);
867 * Readahead a node page
869 void ra_node_page(struct f2fs_sb_info
*sbi
, nid_t nid
)
871 struct address_space
*mapping
= sbi
->node_inode
->i_mapping
;
875 apage
= find_get_page(mapping
, nid
);
876 if (apage
&& PageUptodate(apage
)) {
877 f2fs_put_page(apage
, 0);
880 f2fs_put_page(apage
, 0);
882 apage
= grab_cache_page(mapping
, nid
);
886 err
= read_node_page(apage
, READA
);
888 f2fs_put_page(apage
, 0);
889 else if (err
== LOCKED_PAGE
)
890 f2fs_put_page(apage
, 1);
894 struct page
*get_node_page(struct f2fs_sb_info
*sbi
, pgoff_t nid
)
896 struct address_space
*mapping
= sbi
->node_inode
->i_mapping
;
900 page
= grab_cache_page(mapping
, nid
);
902 return ERR_PTR(-ENOMEM
);
904 err
= read_node_page(page
, READ_SYNC
);
907 else if (err
== LOCKED_PAGE
)
911 if (!PageUptodate(page
)) {
912 f2fs_put_page(page
, 1);
913 return ERR_PTR(-EIO
);
916 BUG_ON(nid
!= nid_of_node(page
));
917 mark_page_accessed(page
);
922 * Return a locked page for the desired node page.
923 * And, readahead MAX_RA_NODE number of node pages.
925 struct page
*get_node_page_ra(struct page
*parent
, int start
)
927 struct f2fs_sb_info
*sbi
= F2FS_SB(parent
->mapping
->host
->i_sb
);
928 struct address_space
*mapping
= sbi
->node_inode
->i_mapping
;
933 /* First, try getting the desired direct node. */
934 nid
= get_nid(parent
, start
, false);
936 return ERR_PTR(-ENOENT
);
938 page
= grab_cache_page(mapping
, nid
);
940 return ERR_PTR(-ENOMEM
);
942 err
= read_node_page(page
, READ_SYNC
);
945 else if (err
== LOCKED_PAGE
)
948 /* Then, try readahead for siblings of the desired node */
949 end
= start
+ MAX_RA_NODE
;
950 end
= min(end
, NIDS_PER_BLOCK
);
951 for (i
= start
+ 1; i
< end
; i
++) {
952 nid
= get_nid(parent
, i
, false);
955 ra_node_page(sbi
, nid
);
961 if (!PageUptodate(page
)) {
962 f2fs_put_page(page
, 1);
963 return ERR_PTR(-EIO
);
965 mark_page_accessed(page
);
969 void sync_inode_page(struct dnode_of_data
*dn
)
971 if (IS_INODE(dn
->node_page
) || dn
->inode_page
== dn
->node_page
) {
972 update_inode(dn
->inode
, dn
->node_page
);
973 } else if (dn
->inode_page
) {
974 if (!dn
->inode_page_locked
)
975 lock_page(dn
->inode_page
);
976 update_inode(dn
->inode
, dn
->inode_page
);
977 if (!dn
->inode_page_locked
)
978 unlock_page(dn
->inode_page
);
980 update_inode_page(dn
->inode
);
984 int sync_node_pages(struct f2fs_sb_info
*sbi
, nid_t ino
,
985 struct writeback_control
*wbc
)
987 struct address_space
*mapping
= sbi
->node_inode
->i_mapping
;
990 int step
= ino
? 2 : 0;
991 int nwritten
= 0, wrote
= 0;
993 pagevec_init(&pvec
, 0);
999 while (index
<= end
) {
1001 nr_pages
= pagevec_lookup_tag(&pvec
, mapping
, &index
,
1002 PAGECACHE_TAG_DIRTY
,
1003 min(end
- index
, (pgoff_t
)PAGEVEC_SIZE
-1) + 1);
1007 for (i
= 0; i
< nr_pages
; i
++) {
1008 struct page
*page
= pvec
.pages
[i
];
1011 * flushing sequence with step:
1016 if (step
== 0 && IS_DNODE(page
))
1018 if (step
== 1 && (!IS_DNODE(page
) ||
1019 is_cold_node(page
)))
1021 if (step
== 2 && (!IS_DNODE(page
) ||
1022 !is_cold_node(page
)))
1027 * we should not skip writing node pages.
1029 if (ino
&& ino_of_node(page
) == ino
)
1031 else if (!trylock_page(page
))
1034 if (unlikely(page
->mapping
!= mapping
)) {
1039 if (ino
&& ino_of_node(page
) != ino
)
1040 goto continue_unlock
;
1042 if (!PageDirty(page
)) {
1043 /* someone wrote it for us */
1044 goto continue_unlock
;
1047 if (!clear_page_dirty_for_io(page
))
1048 goto continue_unlock
;
1050 /* called by fsync() */
1051 if (ino
&& IS_DNODE(page
)) {
1052 int mark
= !is_checkpointed_node(sbi
, ino
);
1053 set_fsync_mark(page
, 1);
1055 set_dentry_mark(page
, mark
);
1058 set_fsync_mark(page
, 0);
1059 set_dentry_mark(page
, 0);
1061 mapping
->a_ops
->writepage(page
, wbc
);
1064 if (--wbc
->nr_to_write
== 0)
1067 pagevec_release(&pvec
);
1070 if (wbc
->nr_to_write
== 0) {
1082 f2fs_submit_bio(sbi
, NODE
, wbc
->sync_mode
== WB_SYNC_ALL
);
1087 static int f2fs_write_node_page(struct page
*page
,
1088 struct writeback_control
*wbc
)
1090 struct f2fs_sb_info
*sbi
= F2FS_SB(page
->mapping
->host
->i_sb
);
1093 struct node_info ni
;
1095 wait_on_page_writeback(page
);
1097 /* get old block addr of this node page */
1098 nid
= nid_of_node(page
);
1099 BUG_ON(page
->index
!= nid
);
1101 get_node_info(sbi
, nid
, &ni
);
1103 /* This page is already truncated */
1104 if (ni
.blk_addr
== NULL_ADDR
) {
1105 dec_page_count(sbi
, F2FS_DIRTY_NODES
);
1110 if (wbc
->for_reclaim
) {
1111 dec_page_count(sbi
, F2FS_DIRTY_NODES
);
1112 wbc
->pages_skipped
++;
1113 set_page_dirty(page
);
1114 return AOP_WRITEPAGE_ACTIVATE
;
1117 mutex_lock(&sbi
->node_write
);
1118 set_page_writeback(page
);
1119 write_node_page(sbi
, page
, nid
, ni
.blk_addr
, &new_addr
);
1120 set_node_addr(sbi
, &ni
, new_addr
);
1121 dec_page_count(sbi
, F2FS_DIRTY_NODES
);
1122 mutex_unlock(&sbi
->node_write
);
1128 * It is very important to gather dirty pages and write at once, so that we can
1129 * submit a big bio without interfering other data writes.
1130 * Be default, 512 pages (2MB), a segment size, is quite reasonable.
1132 #define COLLECT_DIRTY_NODES 512
1133 static int f2fs_write_node_pages(struct address_space
*mapping
,
1134 struct writeback_control
*wbc
)
1136 struct f2fs_sb_info
*sbi
= F2FS_SB(mapping
->host
->i_sb
);
1137 struct block_device
*bdev
= sbi
->sb
->s_bdev
;
1138 long nr_to_write
= wbc
->nr_to_write
;
1140 /* First check balancing cached NAT entries */
1141 if (try_to_free_nats(sbi
, NAT_ENTRY_PER_BLOCK
)) {
1142 f2fs_sync_fs(sbi
->sb
, true);
1146 /* collect a number of dirty node pages and write together */
1147 if (get_pages(sbi
, F2FS_DIRTY_NODES
) < COLLECT_DIRTY_NODES
)
1150 /* if mounting is failed, skip writing node pages */
1151 wbc
->nr_to_write
= bio_get_nr_vecs(bdev
);
1152 sync_node_pages(sbi
, 0, wbc
);
1153 wbc
->nr_to_write
= nr_to_write
-
1154 (bio_get_nr_vecs(bdev
) - wbc
->nr_to_write
);
1158 static int f2fs_set_node_page_dirty(struct page
*page
)
1160 struct address_space
*mapping
= page
->mapping
;
1161 struct f2fs_sb_info
*sbi
= F2FS_SB(mapping
->host
->i_sb
);
1163 SetPageUptodate(page
);
1164 if (!PageDirty(page
)) {
1165 __set_page_dirty_nobuffers(page
);
1166 inc_page_count(sbi
, F2FS_DIRTY_NODES
);
1167 SetPagePrivate(page
);
1173 static void f2fs_invalidate_node_page(struct page
*page
, unsigned long offset
)
1175 struct inode
*inode
= page
->mapping
->host
;
1176 struct f2fs_sb_info
*sbi
= F2FS_SB(inode
->i_sb
);
1177 if (PageDirty(page
))
1178 dec_page_count(sbi
, F2FS_DIRTY_NODES
);
1179 ClearPagePrivate(page
);
1182 static int f2fs_release_node_page(struct page
*page
, gfp_t wait
)
1184 ClearPagePrivate(page
);
1189 * Structure of the f2fs node operations
1191 const struct address_space_operations f2fs_node_aops
= {
1192 .writepage
= f2fs_write_node_page
,
1193 .writepages
= f2fs_write_node_pages
,
1194 .set_page_dirty
= f2fs_set_node_page_dirty
,
1195 .invalidatepage
= f2fs_invalidate_node_page
,
1196 .releasepage
= f2fs_release_node_page
,
1199 static struct free_nid
*__lookup_free_nid_list(nid_t n
, struct list_head
*head
)
1201 struct list_head
*this;
1203 list_for_each(this, head
) {
1204 i
= list_entry(this, struct free_nid
, list
);
1211 static void __del_from_free_nid_list(struct free_nid
*i
)
1214 kmem_cache_free(free_nid_slab
, i
);
1217 static int add_free_nid(struct f2fs_nm_info
*nm_i
, nid_t nid
)
1221 if (nm_i
->fcnt
> 2 * MAX_FREE_NIDS
)
1224 i
= kmem_cache_alloc(free_nid_slab
, GFP_NOFS
);
1232 spin_lock(&nm_i
->free_nid_list_lock
);
1233 if (__lookup_free_nid_list(nid
, &nm_i
->free_nid_list
)) {
1234 spin_unlock(&nm_i
->free_nid_list_lock
);
1235 kmem_cache_free(free_nid_slab
, i
);
1238 list_add_tail(&i
->list
, &nm_i
->free_nid_list
);
1240 spin_unlock(&nm_i
->free_nid_list_lock
);
1244 static void remove_free_nid(struct f2fs_nm_info
*nm_i
, nid_t nid
)
1247 spin_lock(&nm_i
->free_nid_list_lock
);
1248 i
= __lookup_free_nid_list(nid
, &nm_i
->free_nid_list
);
1249 if (i
&& i
->state
== NID_NEW
) {
1250 __del_from_free_nid_list(i
);
1253 spin_unlock(&nm_i
->free_nid_list_lock
);
1256 static int scan_nat_page(struct f2fs_nm_info
*nm_i
,
1257 struct page
*nat_page
, nid_t start_nid
)
1259 struct f2fs_nat_block
*nat_blk
= page_address(nat_page
);
1264 /* 0 nid should not be used */
1268 i
= start_nid
% NAT_ENTRY_PER_BLOCK
;
1270 for (; i
< NAT_ENTRY_PER_BLOCK
; i
++, start_nid
++) {
1271 if (start_nid
>= nm_i
->max_nid
)
1273 blk_addr
= le32_to_cpu(nat_blk
->entries
[i
].block_addr
);
1274 BUG_ON(blk_addr
== NEW_ADDR
);
1275 if (blk_addr
== NULL_ADDR
)
1276 fcnt
+= add_free_nid(nm_i
, start_nid
);
1281 static void build_free_nids(struct f2fs_sb_info
*sbi
)
1283 struct free_nid
*fnid
, *next_fnid
;
1284 struct f2fs_nm_info
*nm_i
= NM_I(sbi
);
1285 struct curseg_info
*curseg
= CURSEG_I(sbi
, CURSEG_HOT_DATA
);
1286 struct f2fs_summary_block
*sum
= curseg
->sum_blk
;
1288 bool is_cycled
= false;
1292 nid
= nm_i
->next_scan_nid
;
1293 nm_i
->init_scan_nid
= nid
;
1295 ra_nat_pages(sbi
, nid
);
1298 struct page
*page
= get_current_nat_page(sbi
, nid
);
1300 fcnt
+= scan_nat_page(nm_i
, page
, nid
);
1301 f2fs_put_page(page
, 1);
1303 nid
+= (NAT_ENTRY_PER_BLOCK
- (nid
% NAT_ENTRY_PER_BLOCK
));
1305 if (nid
>= nm_i
->max_nid
) {
1309 if (fcnt
> MAX_FREE_NIDS
)
1311 if (is_cycled
&& nm_i
->init_scan_nid
<= nid
)
1315 /* go to the next nat page in order to reuse free nids first */
1316 nm_i
->next_scan_nid
= nm_i
->init_scan_nid
+ NAT_ENTRY_PER_BLOCK
;
1318 /* find free nids from current sum_pages */
1319 mutex_lock(&curseg
->curseg_mutex
);
1320 for (i
= 0; i
< nats_in_cursum(sum
); i
++) {
1321 block_t addr
= le32_to_cpu(nat_in_journal(sum
, i
).block_addr
);
1322 nid
= le32_to_cpu(nid_in_journal(sum
, i
));
1323 if (addr
== NULL_ADDR
)
1324 add_free_nid(nm_i
, nid
);
1326 remove_free_nid(nm_i
, nid
);
1328 mutex_unlock(&curseg
->curseg_mutex
);
1330 /* remove the free nids from current allocated nids */
1331 list_for_each_entry_safe(fnid
, next_fnid
, &nm_i
->free_nid_list
, list
) {
1332 struct nat_entry
*ne
;
1334 read_lock(&nm_i
->nat_tree_lock
);
1335 ne
= __lookup_nat_cache(nm_i
, fnid
->nid
);
1336 if (ne
&& nat_get_blkaddr(ne
) != NULL_ADDR
)
1337 remove_free_nid(nm_i
, fnid
->nid
);
1338 read_unlock(&nm_i
->nat_tree_lock
);
1343 * If this function returns success, caller can obtain a new nid
1344 * from second parameter of this function.
1345 * The returned nid could be used ino as well as nid when inode is created.
1347 bool alloc_nid(struct f2fs_sb_info
*sbi
, nid_t
*nid
)
1349 struct f2fs_nm_info
*nm_i
= NM_I(sbi
);
1350 struct free_nid
*i
= NULL
;
1351 struct list_head
*this;
1353 mutex_lock(&nm_i
->build_lock
);
1355 /* scan NAT in order to build free nid list */
1356 build_free_nids(sbi
);
1358 mutex_unlock(&nm_i
->build_lock
);
1362 mutex_unlock(&nm_i
->build_lock
);
1365 * We check fcnt again since previous check is racy as
1366 * we didn't hold free_nid_list_lock. So other thread
1367 * could consume all of free nids.
1369 spin_lock(&nm_i
->free_nid_list_lock
);
1371 spin_unlock(&nm_i
->free_nid_list_lock
);
1375 BUG_ON(list_empty(&nm_i
->free_nid_list
));
1376 list_for_each(this, &nm_i
->free_nid_list
) {
1377 i
= list_entry(this, struct free_nid
, list
);
1378 if (i
->state
== NID_NEW
)
1382 BUG_ON(i
->state
!= NID_NEW
);
1384 i
->state
= NID_ALLOC
;
1386 spin_unlock(&nm_i
->free_nid_list_lock
);
1391 * alloc_nid() should be called prior to this function.
1393 void alloc_nid_done(struct f2fs_sb_info
*sbi
, nid_t nid
)
1395 struct f2fs_nm_info
*nm_i
= NM_I(sbi
);
1398 spin_lock(&nm_i
->free_nid_list_lock
);
1399 i
= __lookup_free_nid_list(nid
, &nm_i
->free_nid_list
);
1400 BUG_ON(!i
|| i
->state
!= NID_ALLOC
);
1401 __del_from_free_nid_list(i
);
1402 spin_unlock(&nm_i
->free_nid_list_lock
);
1406 * alloc_nid() should be called prior to this function.
1408 void alloc_nid_failed(struct f2fs_sb_info
*sbi
, nid_t nid
)
1410 struct f2fs_nm_info
*nm_i
= NM_I(sbi
);
1413 spin_lock(&nm_i
->free_nid_list_lock
);
1414 i
= __lookup_free_nid_list(nid
, &nm_i
->free_nid_list
);
1415 BUG_ON(!i
|| i
->state
!= NID_ALLOC
);
1418 spin_unlock(&nm_i
->free_nid_list_lock
);
1421 void recover_node_page(struct f2fs_sb_info
*sbi
, struct page
*page
,
1422 struct f2fs_summary
*sum
, struct node_info
*ni
,
1423 block_t new_blkaddr
)
1425 rewrite_node_page(sbi
, page
, sum
, ni
->blk_addr
, new_blkaddr
);
1426 set_node_addr(sbi
, ni
, new_blkaddr
);
1427 clear_node_page_dirty(page
);
1430 int recover_inode_page(struct f2fs_sb_info
*sbi
, struct page
*page
)
1432 struct address_space
*mapping
= sbi
->node_inode
->i_mapping
;
1433 struct f2fs_node
*src
, *dst
;
1434 nid_t ino
= ino_of_node(page
);
1435 struct node_info old_ni
, new_ni
;
1438 ipage
= grab_cache_page(mapping
, ino
);
1442 /* Should not use this inode from free nid list */
1443 remove_free_nid(NM_I(sbi
), ino
);
1445 get_node_info(sbi
, ino
, &old_ni
);
1446 SetPageUptodate(ipage
);
1447 fill_node_footer(ipage
, ino
, ino
, 0, true);
1449 src
= (struct f2fs_node
*)page_address(page
);
1450 dst
= (struct f2fs_node
*)page_address(ipage
);
1452 memcpy(dst
, src
, (unsigned long)&src
->i
.i_ext
- (unsigned long)&src
->i
);
1454 dst
->i
.i_blocks
= cpu_to_le64(1);
1455 dst
->i
.i_links
= cpu_to_le32(1);
1456 dst
->i
.i_xattr_nid
= 0;
1461 set_node_addr(sbi
, &new_ni
, NEW_ADDR
);
1462 inc_valid_inode_count(sbi
);
1464 f2fs_put_page(ipage
, 1);
1468 int restore_node_summary(struct f2fs_sb_info
*sbi
,
1469 unsigned int segno
, struct f2fs_summary_block
*sum
)
1471 struct f2fs_node
*rn
;
1472 struct f2fs_summary
*sum_entry
;
1477 /* alloc temporal page for read node */
1478 page
= alloc_page(GFP_NOFS
| __GFP_ZERO
);
1480 return PTR_ERR(page
);
1483 /* scan the node segment */
1484 last_offset
= sbi
->blocks_per_seg
;
1485 addr
= START_BLOCK(sbi
, segno
);
1486 sum_entry
= &sum
->entries
[0];
1488 for (i
= 0; i
< last_offset
; i
++, sum_entry
++) {
1490 * In order to read next node page,
1491 * we must clear PageUptodate flag.
1493 ClearPageUptodate(page
);
1495 if (f2fs_readpage(sbi
, page
, addr
, READ_SYNC
))
1499 rn
= (struct f2fs_node
*)page_address(page
);
1500 sum_entry
->nid
= rn
->footer
.nid
;
1501 sum_entry
->version
= 0;
1502 sum_entry
->ofs_in_node
= 0;
1507 __free_pages(page
, 0);
1511 static bool flush_nats_in_journal(struct f2fs_sb_info
*sbi
)
1513 struct f2fs_nm_info
*nm_i
= NM_I(sbi
);
1514 struct curseg_info
*curseg
= CURSEG_I(sbi
, CURSEG_HOT_DATA
);
1515 struct f2fs_summary_block
*sum
= curseg
->sum_blk
;
1518 mutex_lock(&curseg
->curseg_mutex
);
1520 if (nats_in_cursum(sum
) < NAT_JOURNAL_ENTRIES
) {
1521 mutex_unlock(&curseg
->curseg_mutex
);
1525 for (i
= 0; i
< nats_in_cursum(sum
); i
++) {
1526 struct nat_entry
*ne
;
1527 struct f2fs_nat_entry raw_ne
;
1528 nid_t nid
= le32_to_cpu(nid_in_journal(sum
, i
));
1530 raw_ne
= nat_in_journal(sum
, i
);
1532 write_lock(&nm_i
->nat_tree_lock
);
1533 ne
= __lookup_nat_cache(nm_i
, nid
);
1535 __set_nat_cache_dirty(nm_i
, ne
);
1536 write_unlock(&nm_i
->nat_tree_lock
);
1539 ne
= grab_nat_entry(nm_i
, nid
);
1541 write_unlock(&nm_i
->nat_tree_lock
);
1544 nat_set_blkaddr(ne
, le32_to_cpu(raw_ne
.block_addr
));
1545 nat_set_ino(ne
, le32_to_cpu(raw_ne
.ino
));
1546 nat_set_version(ne
, raw_ne
.version
);
1547 __set_nat_cache_dirty(nm_i
, ne
);
1548 write_unlock(&nm_i
->nat_tree_lock
);
1550 update_nats_in_cursum(sum
, -i
);
1551 mutex_unlock(&curseg
->curseg_mutex
);
1556 * This function is called during the checkpointing process.
1558 void flush_nat_entries(struct f2fs_sb_info
*sbi
)
1560 struct f2fs_nm_info
*nm_i
= NM_I(sbi
);
1561 struct curseg_info
*curseg
= CURSEG_I(sbi
, CURSEG_HOT_DATA
);
1562 struct f2fs_summary_block
*sum
= curseg
->sum_blk
;
1563 struct list_head
*cur
, *n
;
1564 struct page
*page
= NULL
;
1565 struct f2fs_nat_block
*nat_blk
= NULL
;
1566 nid_t start_nid
= 0, end_nid
= 0;
1569 flushed
= flush_nats_in_journal(sbi
);
1572 mutex_lock(&curseg
->curseg_mutex
);
1574 /* 1) flush dirty nat caches */
1575 list_for_each_safe(cur
, n
, &nm_i
->dirty_nat_entries
) {
1576 struct nat_entry
*ne
;
1578 struct f2fs_nat_entry raw_ne
;
1580 block_t new_blkaddr
;
1582 ne
= list_entry(cur
, struct nat_entry
, list
);
1583 nid
= nat_get_nid(ne
);
1585 if (nat_get_blkaddr(ne
) == NEW_ADDR
)
1590 /* if there is room for nat enries in curseg->sumpage */
1591 offset
= lookup_journal_in_cursum(sum
, NAT_JOURNAL
, nid
, 1);
1593 raw_ne
= nat_in_journal(sum
, offset
);
1597 if (!page
|| (start_nid
> nid
|| nid
> end_nid
)) {
1599 f2fs_put_page(page
, 1);
1602 start_nid
= START_NID(nid
);
1603 end_nid
= start_nid
+ NAT_ENTRY_PER_BLOCK
- 1;
1606 * get nat block with dirty flag, increased reference
1607 * count, mapped and lock
1609 page
= get_next_nat_page(sbi
, start_nid
);
1610 nat_blk
= page_address(page
);
1614 raw_ne
= nat_blk
->entries
[nid
- start_nid
];
1616 new_blkaddr
= nat_get_blkaddr(ne
);
1618 raw_ne
.ino
= cpu_to_le32(nat_get_ino(ne
));
1619 raw_ne
.block_addr
= cpu_to_le32(new_blkaddr
);
1620 raw_ne
.version
= nat_get_version(ne
);
1623 nat_blk
->entries
[nid
- start_nid
] = raw_ne
;
1625 nat_in_journal(sum
, offset
) = raw_ne
;
1626 nid_in_journal(sum
, offset
) = cpu_to_le32(nid
);
1629 if (nat_get_blkaddr(ne
) == NULL_ADDR
&&
1630 !add_free_nid(NM_I(sbi
), nid
)) {
1631 write_lock(&nm_i
->nat_tree_lock
);
1632 __del_from_nat_cache(nm_i
, ne
);
1633 write_unlock(&nm_i
->nat_tree_lock
);
1635 write_lock(&nm_i
->nat_tree_lock
);
1636 __clear_nat_cache_dirty(nm_i
, ne
);
1637 ne
->checkpointed
= true;
1638 write_unlock(&nm_i
->nat_tree_lock
);
1642 mutex_unlock(&curseg
->curseg_mutex
);
1643 f2fs_put_page(page
, 1);
1645 /* 2) shrink nat caches if necessary */
1646 try_to_free_nats(sbi
, nm_i
->nat_cnt
- NM_WOUT_THRESHOLD
);
1649 static int init_node_manager(struct f2fs_sb_info
*sbi
)
1651 struct f2fs_super_block
*sb_raw
= F2FS_RAW_SUPER(sbi
);
1652 struct f2fs_nm_info
*nm_i
= NM_I(sbi
);
1653 unsigned char *version_bitmap
;
1654 unsigned int nat_segs
, nat_blocks
;
1656 nm_i
->nat_blkaddr
= le32_to_cpu(sb_raw
->nat_blkaddr
);
1658 /* segment_count_nat includes pair segment so divide to 2. */
1659 nat_segs
= le32_to_cpu(sb_raw
->segment_count_nat
) >> 1;
1660 nat_blocks
= nat_segs
<< le32_to_cpu(sb_raw
->log_blocks_per_seg
);
1661 nm_i
->max_nid
= NAT_ENTRY_PER_BLOCK
* nat_blocks
;
1665 INIT_LIST_HEAD(&nm_i
->free_nid_list
);
1666 INIT_RADIX_TREE(&nm_i
->nat_root
, GFP_ATOMIC
);
1667 INIT_LIST_HEAD(&nm_i
->nat_entries
);
1668 INIT_LIST_HEAD(&nm_i
->dirty_nat_entries
);
1670 mutex_init(&nm_i
->build_lock
);
1671 spin_lock_init(&nm_i
->free_nid_list_lock
);
1672 rwlock_init(&nm_i
->nat_tree_lock
);
1674 nm_i
->init_scan_nid
= le32_to_cpu(sbi
->ckpt
->next_free_nid
);
1675 nm_i
->next_scan_nid
= le32_to_cpu(sbi
->ckpt
->next_free_nid
);
1676 nm_i
->bitmap_size
= __bitmap_size(sbi
, NAT_BITMAP
);
1677 version_bitmap
= __bitmap_ptr(sbi
, NAT_BITMAP
);
1678 if (!version_bitmap
)
1681 nm_i
->nat_bitmap
= kmemdup(version_bitmap
, nm_i
->bitmap_size
,
1683 if (!nm_i
->nat_bitmap
)
1688 int build_node_manager(struct f2fs_sb_info
*sbi
)
1692 sbi
->nm_info
= kzalloc(sizeof(struct f2fs_nm_info
), GFP_KERNEL
);
1696 err
= init_node_manager(sbi
);
1700 build_free_nids(sbi
);
1704 void destroy_node_manager(struct f2fs_sb_info
*sbi
)
1706 struct f2fs_nm_info
*nm_i
= NM_I(sbi
);
1707 struct free_nid
*i
, *next_i
;
1708 struct nat_entry
*natvec
[NATVEC_SIZE
];
1715 /* destroy free nid list */
1716 spin_lock(&nm_i
->free_nid_list_lock
);
1717 list_for_each_entry_safe(i
, next_i
, &nm_i
->free_nid_list
, list
) {
1718 BUG_ON(i
->state
== NID_ALLOC
);
1719 __del_from_free_nid_list(i
);
1723 spin_unlock(&nm_i
->free_nid_list_lock
);
1725 /* destroy nat cache */
1726 write_lock(&nm_i
->nat_tree_lock
);
1727 while ((found
= __gang_lookup_nat_cache(nm_i
,
1728 nid
, NATVEC_SIZE
, natvec
))) {
1730 for (idx
= 0; idx
< found
; idx
++) {
1731 struct nat_entry
*e
= natvec
[idx
];
1732 nid
= nat_get_nid(e
) + 1;
1733 __del_from_nat_cache(nm_i
, e
);
1736 BUG_ON(nm_i
->nat_cnt
);
1737 write_unlock(&nm_i
->nat_tree_lock
);
1739 kfree(nm_i
->nat_bitmap
);
1740 sbi
->nm_info
= NULL
;
1744 int __init
create_node_manager_caches(void)
1746 nat_entry_slab
= f2fs_kmem_cache_create("nat_entry",
1747 sizeof(struct nat_entry
), NULL
);
1748 if (!nat_entry_slab
)
1751 free_nid_slab
= f2fs_kmem_cache_create("free_nid",
1752 sizeof(struct free_nid
), NULL
);
1753 if (!free_nid_slab
) {
1754 kmem_cache_destroy(nat_entry_slab
);
1760 void destroy_node_manager_caches(void)
1762 kmem_cache_destroy(free_nid_slab
);
1763 kmem_cache_destroy(nat_entry_slab
);