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 (f2fs_readpage(sbi
, page
, index
, READ
)) {
104 f2fs_put_page(page
, 1);
107 page_cache_release(page
);
111 static struct nat_entry
*__lookup_nat_cache(struct f2fs_nm_info
*nm_i
, nid_t n
)
113 return radix_tree_lookup(&nm_i
->nat_root
, n
);
116 static unsigned int __gang_lookup_nat_cache(struct f2fs_nm_info
*nm_i
,
117 nid_t start
, unsigned int nr
, struct nat_entry
**ep
)
119 return radix_tree_gang_lookup(&nm_i
->nat_root
, (void **)ep
, start
, nr
);
122 static void __del_from_nat_cache(struct f2fs_nm_info
*nm_i
, struct nat_entry
*e
)
125 radix_tree_delete(&nm_i
->nat_root
, nat_get_nid(e
));
127 kmem_cache_free(nat_entry_slab
, e
);
130 int is_checkpointed_node(struct f2fs_sb_info
*sbi
, nid_t nid
)
132 struct f2fs_nm_info
*nm_i
= NM_I(sbi
);
136 read_lock(&nm_i
->nat_tree_lock
);
137 e
= __lookup_nat_cache(nm_i
, nid
);
138 if (e
&& !e
->checkpointed
)
140 read_unlock(&nm_i
->nat_tree_lock
);
144 static struct nat_entry
*grab_nat_entry(struct f2fs_nm_info
*nm_i
, nid_t nid
)
146 struct nat_entry
*new;
148 new = kmem_cache_alloc(nat_entry_slab
, GFP_ATOMIC
);
151 if (radix_tree_insert(&nm_i
->nat_root
, nid
, new)) {
152 kmem_cache_free(nat_entry_slab
, new);
155 memset(new, 0, sizeof(struct nat_entry
));
156 nat_set_nid(new, nid
);
157 list_add_tail(&new->list
, &nm_i
->nat_entries
);
162 static void cache_nat_entry(struct f2fs_nm_info
*nm_i
, nid_t nid
,
163 struct f2fs_nat_entry
*ne
)
167 write_lock(&nm_i
->nat_tree_lock
);
168 e
= __lookup_nat_cache(nm_i
, nid
);
170 e
= grab_nat_entry(nm_i
, nid
);
172 write_unlock(&nm_i
->nat_tree_lock
);
175 nat_set_blkaddr(e
, le32_to_cpu(ne
->block_addr
));
176 nat_set_ino(e
, le32_to_cpu(ne
->ino
));
177 nat_set_version(e
, ne
->version
);
178 e
->checkpointed
= true;
180 write_unlock(&nm_i
->nat_tree_lock
);
183 static void set_node_addr(struct f2fs_sb_info
*sbi
, struct node_info
*ni
,
186 struct f2fs_nm_info
*nm_i
= NM_I(sbi
);
189 write_lock(&nm_i
->nat_tree_lock
);
190 e
= __lookup_nat_cache(nm_i
, ni
->nid
);
192 e
= grab_nat_entry(nm_i
, ni
->nid
);
194 write_unlock(&nm_i
->nat_tree_lock
);
198 e
->checkpointed
= true;
199 BUG_ON(ni
->blk_addr
== NEW_ADDR
);
200 } else if (new_blkaddr
== NEW_ADDR
) {
202 * when nid is reallocated,
203 * previous nat entry can be remained in nat cache.
204 * So, reinitialize it with new information.
207 BUG_ON(ni
->blk_addr
!= NULL_ADDR
);
210 if (new_blkaddr
== NEW_ADDR
)
211 e
->checkpointed
= false;
214 BUG_ON(nat_get_blkaddr(e
) != ni
->blk_addr
);
215 BUG_ON(nat_get_blkaddr(e
) == NULL_ADDR
&&
216 new_blkaddr
== NULL_ADDR
);
217 BUG_ON(nat_get_blkaddr(e
) == NEW_ADDR
&&
218 new_blkaddr
== NEW_ADDR
);
219 BUG_ON(nat_get_blkaddr(e
) != NEW_ADDR
&&
220 nat_get_blkaddr(e
) != NULL_ADDR
&&
221 new_blkaddr
== NEW_ADDR
);
223 /* increament version no as node is removed */
224 if (nat_get_blkaddr(e
) != NEW_ADDR
&& new_blkaddr
== NULL_ADDR
) {
225 unsigned char version
= nat_get_version(e
);
226 nat_set_version(e
, inc_node_version(version
));
230 nat_set_blkaddr(e
, new_blkaddr
);
231 __set_nat_cache_dirty(nm_i
, e
);
232 write_unlock(&nm_i
->nat_tree_lock
);
235 static int try_to_free_nats(struct f2fs_sb_info
*sbi
, int nr_shrink
)
237 struct f2fs_nm_info
*nm_i
= NM_I(sbi
);
239 if (nm_i
->nat_cnt
< 2 * NM_WOUT_THRESHOLD
)
242 write_lock(&nm_i
->nat_tree_lock
);
243 while (nr_shrink
&& !list_empty(&nm_i
->nat_entries
)) {
244 struct nat_entry
*ne
;
245 ne
= list_first_entry(&nm_i
->nat_entries
,
246 struct nat_entry
, list
);
247 __del_from_nat_cache(nm_i
, ne
);
250 write_unlock(&nm_i
->nat_tree_lock
);
255 * This function returns always success
257 void get_node_info(struct f2fs_sb_info
*sbi
, nid_t nid
, struct node_info
*ni
)
259 struct f2fs_nm_info
*nm_i
= NM_I(sbi
);
260 struct curseg_info
*curseg
= CURSEG_I(sbi
, CURSEG_HOT_DATA
);
261 struct f2fs_summary_block
*sum
= curseg
->sum_blk
;
262 nid_t start_nid
= START_NID(nid
);
263 struct f2fs_nat_block
*nat_blk
;
264 struct page
*page
= NULL
;
265 struct f2fs_nat_entry ne
;
269 memset(&ne
, 0, sizeof(struct f2fs_nat_entry
));
272 /* Check nat cache */
273 read_lock(&nm_i
->nat_tree_lock
);
274 e
= __lookup_nat_cache(nm_i
, nid
);
276 ni
->ino
= nat_get_ino(e
);
277 ni
->blk_addr
= nat_get_blkaddr(e
);
278 ni
->version
= nat_get_version(e
);
280 read_unlock(&nm_i
->nat_tree_lock
);
284 /* Check current segment summary */
285 mutex_lock(&curseg
->curseg_mutex
);
286 i
= lookup_journal_in_cursum(sum
, NAT_JOURNAL
, nid
, 0);
288 ne
= nat_in_journal(sum
, i
);
289 node_info_from_raw_nat(ni
, &ne
);
291 mutex_unlock(&curseg
->curseg_mutex
);
295 /* Fill node_info from nat page */
296 page
= get_current_nat_page(sbi
, start_nid
);
297 nat_blk
= (struct f2fs_nat_block
*)page_address(page
);
298 ne
= nat_blk
->entries
[nid
- start_nid
];
299 node_info_from_raw_nat(ni
, &ne
);
300 f2fs_put_page(page
, 1);
302 /* cache nat entry */
303 cache_nat_entry(NM_I(sbi
), nid
, &ne
);
307 * The maximum depth is four.
308 * Offset[0] will have raw inode offset.
310 static int get_node_path(long block
, int offset
[4], unsigned int noffset
[4])
312 const long direct_index
= ADDRS_PER_INODE
;
313 const long direct_blks
= ADDRS_PER_BLOCK
;
314 const long dptrs_per_blk
= NIDS_PER_BLOCK
;
315 const long indirect_blks
= ADDRS_PER_BLOCK
* NIDS_PER_BLOCK
;
316 const long dindirect_blks
= indirect_blks
* NIDS_PER_BLOCK
;
322 if (block
< direct_index
) {
327 block
-= direct_index
;
328 if (block
< direct_blks
) {
329 offset
[n
++] = NODE_DIR1_BLOCK
;
335 block
-= direct_blks
;
336 if (block
< direct_blks
) {
337 offset
[n
++] = NODE_DIR2_BLOCK
;
343 block
-= direct_blks
;
344 if (block
< indirect_blks
) {
345 offset
[n
++] = NODE_IND1_BLOCK
;
347 offset
[n
++] = block
/ direct_blks
;
348 noffset
[n
] = 4 + offset
[n
- 1];
349 offset
[n
++] = block
% direct_blks
;
353 block
-= indirect_blks
;
354 if (block
< indirect_blks
) {
355 offset
[n
++] = NODE_IND2_BLOCK
;
356 noffset
[n
] = 4 + dptrs_per_blk
;
357 offset
[n
++] = block
/ direct_blks
;
358 noffset
[n
] = 5 + dptrs_per_blk
+ offset
[n
- 1];
359 offset
[n
++] = block
% direct_blks
;
363 block
-= indirect_blks
;
364 if (block
< dindirect_blks
) {
365 offset
[n
++] = NODE_DIND_BLOCK
;
366 noffset
[n
] = 5 + (dptrs_per_blk
* 2);
367 offset
[n
++] = block
/ indirect_blks
;
368 noffset
[n
] = 6 + (dptrs_per_blk
* 2) +
369 offset
[n
- 1] * (dptrs_per_blk
+ 1);
370 offset
[n
++] = (block
/ direct_blks
) % dptrs_per_blk
;
371 noffset
[n
] = 7 + (dptrs_per_blk
* 2) +
372 offset
[n
- 2] * (dptrs_per_blk
+ 1) +
374 offset
[n
++] = block
% direct_blks
;
385 * Caller should call f2fs_put_dnode(dn).
387 int get_dnode_of_data(struct dnode_of_data
*dn
, pgoff_t index
, int ro
)
389 struct f2fs_sb_info
*sbi
= F2FS_SB(dn
->inode
->i_sb
);
390 struct page
*npage
[4];
393 unsigned int noffset
[4];
398 level
= get_node_path(index
, offset
, noffset
);
400 nids
[0] = dn
->inode
->i_ino
;
401 npage
[0] = get_node_page(sbi
, nids
[0]);
402 if (IS_ERR(npage
[0]))
403 return PTR_ERR(npage
[0]);
406 nids
[1] = get_nid(parent
, offset
[0], true);
407 dn
->inode_page
= npage
[0];
408 dn
->inode_page_locked
= true;
410 /* get indirect or direct nodes */
411 for (i
= 1; i
<= level
; i
++) {
414 if (!nids
[i
] && !ro
) {
415 mutex_lock_op(sbi
, NODE_NEW
);
418 if (!alloc_nid(sbi
, &(nids
[i
]))) {
419 mutex_unlock_op(sbi
, NODE_NEW
);
425 npage
[i
] = new_node_page(dn
, noffset
[i
]);
426 if (IS_ERR(npage
[i
])) {
427 alloc_nid_failed(sbi
, nids
[i
]);
428 mutex_unlock_op(sbi
, NODE_NEW
);
429 err
= PTR_ERR(npage
[i
]);
433 set_nid(parent
, offset
[i
- 1], nids
[i
], i
== 1);
434 alloc_nid_done(sbi
, nids
[i
]);
435 mutex_unlock_op(sbi
, NODE_NEW
);
437 } else if (ro
&& i
== level
&& level
> 1) {
438 npage
[i
] = get_node_page_ra(parent
, offset
[i
- 1]);
439 if (IS_ERR(npage
[i
])) {
440 err
= PTR_ERR(npage
[i
]);
446 dn
->inode_page_locked
= false;
449 f2fs_put_page(parent
, 1);
453 npage
[i
] = get_node_page(sbi
, nids
[i
]);
454 if (IS_ERR(npage
[i
])) {
455 err
= PTR_ERR(npage
[i
]);
456 f2fs_put_page(npage
[0], 0);
462 nids
[i
+ 1] = get_nid(parent
, offset
[i
], false);
465 dn
->nid
= nids
[level
];
466 dn
->ofs_in_node
= offset
[level
];
467 dn
->node_page
= npage
[level
];
468 dn
->data_blkaddr
= datablock_addr(dn
->node_page
, dn
->ofs_in_node
);
472 f2fs_put_page(parent
, 1);
474 f2fs_put_page(npage
[0], 0);
476 dn
->inode_page
= NULL
;
477 dn
->node_page
= NULL
;
481 static void truncate_node(struct dnode_of_data
*dn
)
483 struct f2fs_sb_info
*sbi
= F2FS_SB(dn
->inode
->i_sb
);
486 get_node_info(sbi
, dn
->nid
, &ni
);
487 if (dn
->inode
->i_blocks
== 0) {
488 BUG_ON(ni
.blk_addr
!= NULL_ADDR
);
491 BUG_ON(ni
.blk_addr
== NULL_ADDR
);
493 /* Deallocate node address */
494 invalidate_blocks(sbi
, ni
.blk_addr
);
495 dec_valid_node_count(sbi
, dn
->inode
, 1);
496 set_node_addr(sbi
, &ni
, NULL_ADDR
);
498 if (dn
->nid
== dn
->inode
->i_ino
) {
499 remove_orphan_inode(sbi
, dn
->nid
);
500 dec_valid_inode_count(sbi
);
505 clear_node_page_dirty(dn
->node_page
);
506 F2FS_SET_SB_DIRT(sbi
);
508 f2fs_put_page(dn
->node_page
, 1);
509 dn
->node_page
= NULL
;
512 static int truncate_dnode(struct dnode_of_data
*dn
)
514 struct f2fs_sb_info
*sbi
= F2FS_SB(dn
->inode
->i_sb
);
520 /* get direct node */
521 page
= get_node_page(sbi
, dn
->nid
);
522 if (IS_ERR(page
) && PTR_ERR(page
) == -ENOENT
)
524 else if (IS_ERR(page
))
525 return PTR_ERR(page
);
527 /* Make dnode_of_data for parameter */
528 dn
->node_page
= page
;
530 truncate_data_blocks(dn
);
535 static int truncate_nodes(struct dnode_of_data
*dn
, unsigned int nofs
,
538 struct f2fs_sb_info
*sbi
= F2FS_SB(dn
->inode
->i_sb
);
539 struct dnode_of_data rdn
= *dn
;
541 struct f2fs_node
*rn
;
543 unsigned int child_nofs
;
548 return NIDS_PER_BLOCK
+ 1;
550 page
= get_node_page(sbi
, dn
->nid
);
552 return PTR_ERR(page
);
554 rn
= (struct f2fs_node
*)page_address(page
);
556 for (i
= ofs
; i
< NIDS_PER_BLOCK
; i
++, freed
++) {
557 child_nid
= le32_to_cpu(rn
->in
.nid
[i
]);
561 ret
= truncate_dnode(&rdn
);
564 set_nid(page
, i
, 0, false);
567 child_nofs
= nofs
+ ofs
* (NIDS_PER_BLOCK
+ 1) + 1;
568 for (i
= ofs
; i
< NIDS_PER_BLOCK
; i
++) {
569 child_nid
= le32_to_cpu(rn
->in
.nid
[i
]);
570 if (child_nid
== 0) {
571 child_nofs
+= NIDS_PER_BLOCK
+ 1;
575 ret
= truncate_nodes(&rdn
, child_nofs
, 0, depth
- 1);
576 if (ret
== (NIDS_PER_BLOCK
+ 1)) {
577 set_nid(page
, i
, 0, false);
579 } else if (ret
< 0 && ret
!= -ENOENT
) {
587 /* remove current indirect node */
588 dn
->node_page
= page
;
592 f2fs_put_page(page
, 1);
597 f2fs_put_page(page
, 1);
601 static int truncate_partial_nodes(struct dnode_of_data
*dn
,
602 struct f2fs_inode
*ri
, int *offset
, int depth
)
604 struct f2fs_sb_info
*sbi
= F2FS_SB(dn
->inode
->i_sb
);
605 struct page
*pages
[2];
612 nid
[0] = le32_to_cpu(ri
->i_nid
[offset
[0] - NODE_DIR1_BLOCK
]);
616 /* get indirect nodes in the path */
617 for (i
= 0; i
< depth
- 1; i
++) {
618 /* refernece count'll be increased */
619 pages
[i
] = get_node_page(sbi
, nid
[i
]);
620 if (IS_ERR(pages
[i
])) {
622 err
= PTR_ERR(pages
[i
]);
625 nid
[i
+ 1] = get_nid(pages
[i
], offset
[i
+ 1], false);
628 /* free direct nodes linked to a partial indirect node */
629 for (i
= offset
[depth
- 1]; i
< NIDS_PER_BLOCK
; i
++) {
630 child_nid
= get_nid(pages
[idx
], i
, false);
634 err
= truncate_dnode(dn
);
637 set_nid(pages
[idx
], i
, 0, false);
640 if (offset
[depth
- 1] == 0) {
641 dn
->node_page
= pages
[idx
];
645 f2fs_put_page(pages
[idx
], 1);
648 offset
[depth
- 1] = 0;
650 for (i
= depth
- 3; i
>= 0; i
--)
651 f2fs_put_page(pages
[i
], 1);
656 * All the block addresses of data and nodes should be nullified.
658 int truncate_inode_blocks(struct inode
*inode
, pgoff_t from
)
660 struct f2fs_sb_info
*sbi
= F2FS_SB(inode
->i_sb
);
661 int err
= 0, cont
= 1;
662 int level
, offset
[4], noffset
[4];
664 struct f2fs_node
*rn
;
665 struct dnode_of_data dn
;
668 level
= get_node_path(from
, offset
, noffset
);
670 page
= get_node_page(sbi
, inode
->i_ino
);
672 return PTR_ERR(page
);
674 set_new_dnode(&dn
, inode
, page
, NULL
, 0);
677 rn
= page_address(page
);
685 if (!offset
[level
- 1])
687 err
= truncate_partial_nodes(&dn
, &rn
->i
, offset
, level
);
688 if (err
< 0 && err
!= -ENOENT
)
690 nofs
+= 1 + NIDS_PER_BLOCK
;
693 nofs
= 5 + 2 * NIDS_PER_BLOCK
;
694 if (!offset
[level
- 1])
696 err
= truncate_partial_nodes(&dn
, &rn
->i
, offset
, level
);
697 if (err
< 0 && err
!= -ENOENT
)
706 dn
.nid
= le32_to_cpu(rn
->i
.i_nid
[offset
[0] - NODE_DIR1_BLOCK
]);
708 case NODE_DIR1_BLOCK
:
709 case NODE_DIR2_BLOCK
:
710 err
= truncate_dnode(&dn
);
713 case NODE_IND1_BLOCK
:
714 case NODE_IND2_BLOCK
:
715 err
= truncate_nodes(&dn
, nofs
, offset
[1], 2);
718 case NODE_DIND_BLOCK
:
719 err
= truncate_nodes(&dn
, nofs
, offset
[1], 3);
726 if (err
< 0 && err
!= -ENOENT
)
728 if (offset
[1] == 0 &&
729 rn
->i
.i_nid
[offset
[0] - NODE_DIR1_BLOCK
]) {
731 wait_on_page_writeback(page
);
732 rn
->i
.i_nid
[offset
[0] - NODE_DIR1_BLOCK
] = 0;
733 set_page_dirty(page
);
741 f2fs_put_page(page
, 0);
742 return err
> 0 ? 0 : err
;
745 int remove_inode_page(struct inode
*inode
)
747 struct f2fs_sb_info
*sbi
= F2FS_SB(inode
->i_sb
);
749 nid_t ino
= inode
->i_ino
;
750 struct dnode_of_data dn
;
752 mutex_lock_op(sbi
, NODE_TRUNC
);
753 page
= get_node_page(sbi
, ino
);
755 mutex_unlock_op(sbi
, NODE_TRUNC
);
756 return PTR_ERR(page
);
759 if (F2FS_I(inode
)->i_xattr_nid
) {
760 nid_t nid
= F2FS_I(inode
)->i_xattr_nid
;
761 struct page
*npage
= get_node_page(sbi
, nid
);
764 mutex_unlock_op(sbi
, NODE_TRUNC
);
765 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
);
779 mutex_unlock_op(sbi
, NODE_TRUNC
);
783 int new_inode_page(struct inode
*inode
, struct dentry
*dentry
)
785 struct f2fs_sb_info
*sbi
= F2FS_SB(inode
->i_sb
);
787 struct dnode_of_data dn
;
789 /* allocate inode page for new inode */
790 set_new_dnode(&dn
, inode
, NULL
, NULL
, inode
->i_ino
);
791 mutex_lock_op(sbi
, NODE_NEW
);
792 page
= new_node_page(&dn
, 0);
793 init_dent_inode(dentry
, page
);
794 mutex_unlock_op(sbi
, NODE_NEW
);
796 return PTR_ERR(page
);
797 f2fs_put_page(page
, 1);
801 struct page
*new_node_page(struct dnode_of_data
*dn
, unsigned int ofs
)
803 struct f2fs_sb_info
*sbi
= F2FS_SB(dn
->inode
->i_sb
);
804 struct address_space
*mapping
= sbi
->node_inode
->i_mapping
;
805 struct node_info old_ni
, new_ni
;
809 if (is_inode_flag_set(F2FS_I(dn
->inode
), FI_NO_ALLOC
))
810 return ERR_PTR(-EPERM
);
812 page
= grab_cache_page(mapping
, dn
->nid
);
814 return ERR_PTR(-ENOMEM
);
816 get_node_info(sbi
, dn
->nid
, &old_ni
);
818 SetPageUptodate(page
);
819 fill_node_footer(page
, dn
->nid
, dn
->inode
->i_ino
, ofs
, true);
821 /* Reinitialize old_ni with new node page */
822 BUG_ON(old_ni
.blk_addr
!= NULL_ADDR
);
824 new_ni
.ino
= dn
->inode
->i_ino
;
826 if (!inc_valid_node_count(sbi
, dn
->inode
, 1)) {
830 set_node_addr(sbi
, &new_ni
, NEW_ADDR
);
831 set_cold_node(dn
->inode
, page
);
833 dn
->node_page
= page
;
835 set_page_dirty(page
);
837 inc_valid_inode_count(sbi
);
842 clear_node_page_dirty(page
);
843 f2fs_put_page(page
, 1);
847 static int read_node_page(struct page
*page
, int type
)
849 struct f2fs_sb_info
*sbi
= F2FS_SB(page
->mapping
->host
->i_sb
);
852 get_node_info(sbi
, page
->index
, &ni
);
854 if (ni
.blk_addr
== NULL_ADDR
)
856 return f2fs_readpage(sbi
, page
, ni
.blk_addr
, type
);
860 * Readahead a node page
862 void ra_node_page(struct f2fs_sb_info
*sbi
, nid_t nid
)
864 struct address_space
*mapping
= sbi
->node_inode
->i_mapping
;
867 apage
= find_get_page(mapping
, nid
);
868 if (apage
&& PageUptodate(apage
))
870 f2fs_put_page(apage
, 0);
872 apage
= grab_cache_page(mapping
, nid
);
876 if (read_node_page(apage
, READA
))
879 page_cache_release(apage
);
885 page_cache_release(apage
);
888 struct page
*get_node_page(struct f2fs_sb_info
*sbi
, pgoff_t nid
)
892 struct address_space
*mapping
= sbi
->node_inode
->i_mapping
;
894 page
= grab_cache_page(mapping
, nid
);
896 return ERR_PTR(-ENOMEM
);
898 err
= read_node_page(page
, READ_SYNC
);
900 f2fs_put_page(page
, 1);
904 BUG_ON(nid
!= nid_of_node(page
));
905 mark_page_accessed(page
);
910 * Return a locked page for the desired node page.
911 * And, readahead MAX_RA_NODE number of node pages.
913 struct page
*get_node_page_ra(struct page
*parent
, int start
)
915 struct f2fs_sb_info
*sbi
= F2FS_SB(parent
->mapping
->host
->i_sb
);
916 struct address_space
*mapping
= sbi
->node_inode
->i_mapping
;
922 /* First, try getting the desired direct node. */
923 nid
= get_nid(parent
, start
, false);
925 return ERR_PTR(-ENOENT
);
927 page
= find_get_page(mapping
, nid
);
928 if (page
&& PageUptodate(page
))
930 f2fs_put_page(page
, 0);
933 page
= grab_cache_page(mapping
, nid
);
935 return ERR_PTR(-ENOMEM
);
937 err
= read_node_page(page
, READA
);
939 f2fs_put_page(page
, 1);
943 /* Then, try readahead for siblings of the desired node */
944 end
= start
+ MAX_RA_NODE
;
945 end
= min(end
, NIDS_PER_BLOCK
);
946 for (i
= start
+ 1; i
< end
; i
++) {
947 nid
= get_nid(parent
, i
, false);
950 ra_node_page(sbi
, nid
);
955 if (PageError(page
)) {
956 f2fs_put_page(page
, 1);
957 return ERR_PTR(-EIO
);
960 /* Has the page been truncated? */
961 if (page
->mapping
!= mapping
) {
962 f2fs_put_page(page
, 1);
968 void sync_inode_page(struct dnode_of_data
*dn
)
970 if (IS_INODE(dn
->node_page
) || dn
->inode_page
== dn
->node_page
) {
971 update_inode(dn
->inode
, dn
->node_page
);
972 } else if (dn
->inode_page
) {
973 if (!dn
->inode_page_locked
)
974 lock_page(dn
->inode_page
);
975 update_inode(dn
->inode
, dn
->inode_page
);
976 if (!dn
->inode_page_locked
)
977 unlock_page(dn
->inode_page
);
979 f2fs_write_inode(dn
->inode
, NULL
);
983 int sync_node_pages(struct f2fs_sb_info
*sbi
, nid_t ino
,
984 struct writeback_control
*wbc
)
986 struct address_space
*mapping
= sbi
->node_inode
->i_mapping
;
989 int step
= ino
? 2 : 0;
990 int nwritten
= 0, wrote
= 0;
992 pagevec_init(&pvec
, 0);
998 while (index
<= end
) {
1000 nr_pages
= pagevec_lookup_tag(&pvec
, mapping
, &index
,
1001 PAGECACHE_TAG_DIRTY
,
1002 min(end
- index
, (pgoff_t
)PAGEVEC_SIZE
-1) + 1);
1006 for (i
= 0; i
< nr_pages
; i
++) {
1007 struct page
*page
= pvec
.pages
[i
];
1010 * flushing sequence with step:
1015 if (step
== 0 && IS_DNODE(page
))
1017 if (step
== 1 && (!IS_DNODE(page
) ||
1018 is_cold_node(page
)))
1020 if (step
== 2 && (!IS_DNODE(page
) ||
1021 !is_cold_node(page
)))
1026 * we should not skip writing node pages.
1028 if (ino
&& ino_of_node(page
) == ino
)
1030 else if (!trylock_page(page
))
1033 if (unlikely(page
->mapping
!= mapping
)) {
1038 if (ino
&& ino_of_node(page
) != ino
)
1039 goto continue_unlock
;
1041 if (!PageDirty(page
)) {
1042 /* someone wrote it for us */
1043 goto continue_unlock
;
1046 if (!clear_page_dirty_for_io(page
))
1047 goto continue_unlock
;
1049 /* called by fsync() */
1050 if (ino
&& IS_DNODE(page
)) {
1051 int mark
= !is_checkpointed_node(sbi
, ino
);
1052 set_fsync_mark(page
, 1);
1054 set_dentry_mark(page
, mark
);
1057 set_fsync_mark(page
, 0);
1058 set_dentry_mark(page
, 0);
1060 mapping
->a_ops
->writepage(page
, wbc
);
1063 if (--wbc
->nr_to_write
== 0)
1066 pagevec_release(&pvec
);
1069 if (wbc
->nr_to_write
== 0) {
1081 f2fs_submit_bio(sbi
, NODE
, wbc
->sync_mode
== WB_SYNC_ALL
);
1086 static int f2fs_write_node_page(struct page
*page
,
1087 struct writeback_control
*wbc
)
1089 struct f2fs_sb_info
*sbi
= F2FS_SB(page
->mapping
->host
->i_sb
);
1092 struct node_info ni
;
1094 if (wbc
->for_reclaim
) {
1095 dec_page_count(sbi
, F2FS_DIRTY_NODES
);
1096 wbc
->pages_skipped
++;
1097 set_page_dirty(page
);
1098 return AOP_WRITEPAGE_ACTIVATE
;
1101 wait_on_page_writeback(page
);
1103 mutex_lock_op(sbi
, NODE_WRITE
);
1105 /* get old block addr of this node page */
1106 nid
= nid_of_node(page
);
1107 BUG_ON(page
->index
!= nid
);
1109 get_node_info(sbi
, nid
, &ni
);
1111 /* This page is already truncated */
1112 if (ni
.blk_addr
== NULL_ADDR
)
1115 set_page_writeback(page
);
1117 /* insert node offset */
1118 write_node_page(sbi
, page
, nid
, ni
.blk_addr
, &new_addr
);
1119 set_node_addr(sbi
, &ni
, new_addr
);
1120 dec_page_count(sbi
, F2FS_DIRTY_NODES
);
1122 mutex_unlock_op(sbi
, NODE_WRITE
);
1127 static int f2fs_write_node_pages(struct address_space
*mapping
,
1128 struct writeback_control
*wbc
)
1130 struct f2fs_sb_info
*sbi
= F2FS_SB(mapping
->host
->i_sb
);
1131 struct block_device
*bdev
= sbi
->sb
->s_bdev
;
1132 long nr_to_write
= wbc
->nr_to_write
;
1134 if (wbc
->for_kupdate
)
1137 if (get_pages(sbi
, F2FS_DIRTY_NODES
) == 0)
1140 if (try_to_free_nats(sbi
, NAT_ENTRY_PER_BLOCK
)) {
1141 write_checkpoint(sbi
, false, false);
1145 /* if mounting is failed, skip writing node pages */
1146 wbc
->nr_to_write
= bio_get_nr_vecs(bdev
);
1147 sync_node_pages(sbi
, 0, wbc
);
1148 wbc
->nr_to_write
= nr_to_write
-
1149 (bio_get_nr_vecs(bdev
) - wbc
->nr_to_write
);
1153 static int f2fs_set_node_page_dirty(struct page
*page
)
1155 struct address_space
*mapping
= page
->mapping
;
1156 struct f2fs_sb_info
*sbi
= F2FS_SB(mapping
->host
->i_sb
);
1158 SetPageUptodate(page
);
1159 if (!PageDirty(page
)) {
1160 __set_page_dirty_nobuffers(page
);
1161 inc_page_count(sbi
, F2FS_DIRTY_NODES
);
1162 SetPagePrivate(page
);
1168 static void f2fs_invalidate_node_page(struct page
*page
, unsigned long offset
)
1170 struct inode
*inode
= page
->mapping
->host
;
1171 struct f2fs_sb_info
*sbi
= F2FS_SB(inode
->i_sb
);
1172 if (PageDirty(page
))
1173 dec_page_count(sbi
, F2FS_DIRTY_NODES
);
1174 ClearPagePrivate(page
);
1177 static int f2fs_release_node_page(struct page
*page
, gfp_t wait
)
1179 ClearPagePrivate(page
);
1184 * Structure of the f2fs node operations
1186 const struct address_space_operations f2fs_node_aops
= {
1187 .writepage
= f2fs_write_node_page
,
1188 .writepages
= f2fs_write_node_pages
,
1189 .set_page_dirty
= f2fs_set_node_page_dirty
,
1190 .invalidatepage
= f2fs_invalidate_node_page
,
1191 .releasepage
= f2fs_release_node_page
,
1194 static struct free_nid
*__lookup_free_nid_list(nid_t n
, struct list_head
*head
)
1196 struct list_head
*this;
1197 struct free_nid
*i
= NULL
;
1198 list_for_each(this, head
) {
1199 i
= list_entry(this, struct free_nid
, list
);
1207 static void __del_from_free_nid_list(struct free_nid
*i
)
1210 kmem_cache_free(free_nid_slab
, i
);
1213 static int add_free_nid(struct f2fs_nm_info
*nm_i
, nid_t nid
)
1217 if (nm_i
->fcnt
> 2 * MAX_FREE_NIDS
)
1220 i
= kmem_cache_alloc(free_nid_slab
, GFP_NOFS
);
1228 spin_lock(&nm_i
->free_nid_list_lock
);
1229 if (__lookup_free_nid_list(nid
, &nm_i
->free_nid_list
)) {
1230 spin_unlock(&nm_i
->free_nid_list_lock
);
1231 kmem_cache_free(free_nid_slab
, i
);
1234 list_add_tail(&i
->list
, &nm_i
->free_nid_list
);
1236 spin_unlock(&nm_i
->free_nid_list_lock
);
1240 static void remove_free_nid(struct f2fs_nm_info
*nm_i
, nid_t nid
)
1243 spin_lock(&nm_i
->free_nid_list_lock
);
1244 i
= __lookup_free_nid_list(nid
, &nm_i
->free_nid_list
);
1245 if (i
&& i
->state
== NID_NEW
) {
1246 __del_from_free_nid_list(i
);
1249 spin_unlock(&nm_i
->free_nid_list_lock
);
1252 static int scan_nat_page(struct f2fs_nm_info
*nm_i
,
1253 struct page
*nat_page
, nid_t start_nid
)
1255 struct f2fs_nat_block
*nat_blk
= page_address(nat_page
);
1260 /* 0 nid should not be used */
1264 i
= start_nid
% NAT_ENTRY_PER_BLOCK
;
1266 for (; i
< NAT_ENTRY_PER_BLOCK
; i
++, start_nid
++) {
1267 blk_addr
= le32_to_cpu(nat_blk
->entries
[i
].block_addr
);
1268 BUG_ON(blk_addr
== NEW_ADDR
);
1269 if (blk_addr
== NULL_ADDR
)
1270 fcnt
+= add_free_nid(nm_i
, start_nid
);
1275 static void build_free_nids(struct f2fs_sb_info
*sbi
)
1277 struct free_nid
*fnid
, *next_fnid
;
1278 struct f2fs_nm_info
*nm_i
= NM_I(sbi
);
1279 struct curseg_info
*curseg
= CURSEG_I(sbi
, CURSEG_HOT_DATA
);
1280 struct f2fs_summary_block
*sum
= curseg
->sum_blk
;
1282 bool is_cycled
= false;
1286 nid
= nm_i
->next_scan_nid
;
1287 nm_i
->init_scan_nid
= nid
;
1289 ra_nat_pages(sbi
, nid
);
1292 struct page
*page
= get_current_nat_page(sbi
, nid
);
1294 fcnt
+= scan_nat_page(nm_i
, page
, nid
);
1295 f2fs_put_page(page
, 1);
1297 nid
+= (NAT_ENTRY_PER_BLOCK
- (nid
% NAT_ENTRY_PER_BLOCK
));
1299 if (nid
>= nm_i
->max_nid
) {
1303 if (fcnt
> MAX_FREE_NIDS
)
1305 if (is_cycled
&& nm_i
->init_scan_nid
<= nid
)
1309 nm_i
->next_scan_nid
= nid
;
1311 /* find free nids from current sum_pages */
1312 mutex_lock(&curseg
->curseg_mutex
);
1313 for (i
= 0; i
< nats_in_cursum(sum
); i
++) {
1314 block_t addr
= le32_to_cpu(nat_in_journal(sum
, i
).block_addr
);
1315 nid
= le32_to_cpu(nid_in_journal(sum
, i
));
1316 if (addr
== NULL_ADDR
)
1317 add_free_nid(nm_i
, nid
);
1319 remove_free_nid(nm_i
, nid
);
1321 mutex_unlock(&curseg
->curseg_mutex
);
1323 /* remove the free nids from current allocated nids */
1324 list_for_each_entry_safe(fnid
, next_fnid
, &nm_i
->free_nid_list
, list
) {
1325 struct nat_entry
*ne
;
1327 read_lock(&nm_i
->nat_tree_lock
);
1328 ne
= __lookup_nat_cache(nm_i
, fnid
->nid
);
1329 if (ne
&& nat_get_blkaddr(ne
) != NULL_ADDR
)
1330 remove_free_nid(nm_i
, fnid
->nid
);
1331 read_unlock(&nm_i
->nat_tree_lock
);
1336 * If this function returns success, caller can obtain a new nid
1337 * from second parameter of this function.
1338 * The returned nid could be used ino as well as nid when inode is created.
1340 bool alloc_nid(struct f2fs_sb_info
*sbi
, nid_t
*nid
)
1342 struct f2fs_nm_info
*nm_i
= NM_I(sbi
);
1343 struct free_nid
*i
= NULL
;
1344 struct list_head
*this;
1346 mutex_lock(&nm_i
->build_lock
);
1348 /* scan NAT in order to build free nid list */
1349 build_free_nids(sbi
);
1351 mutex_unlock(&nm_i
->build_lock
);
1355 mutex_unlock(&nm_i
->build_lock
);
1358 * We check fcnt again since previous check is racy as
1359 * we didn't hold free_nid_list_lock. So other thread
1360 * could consume all of free nids.
1362 spin_lock(&nm_i
->free_nid_list_lock
);
1364 spin_unlock(&nm_i
->free_nid_list_lock
);
1368 BUG_ON(list_empty(&nm_i
->free_nid_list
));
1369 list_for_each(this, &nm_i
->free_nid_list
) {
1370 i
= list_entry(this, struct free_nid
, list
);
1371 if (i
->state
== NID_NEW
)
1375 BUG_ON(i
->state
!= NID_NEW
);
1377 i
->state
= NID_ALLOC
;
1379 spin_unlock(&nm_i
->free_nid_list_lock
);
1384 * alloc_nid() should be called prior to this function.
1386 void alloc_nid_done(struct f2fs_sb_info
*sbi
, nid_t nid
)
1388 struct f2fs_nm_info
*nm_i
= NM_I(sbi
);
1391 spin_lock(&nm_i
->free_nid_list_lock
);
1392 i
= __lookup_free_nid_list(nid
, &nm_i
->free_nid_list
);
1394 BUG_ON(i
->state
!= NID_ALLOC
);
1395 __del_from_free_nid_list(i
);
1397 spin_unlock(&nm_i
->free_nid_list_lock
);
1401 * alloc_nid() should be called prior to this function.
1403 void alloc_nid_failed(struct f2fs_sb_info
*sbi
, nid_t nid
)
1405 alloc_nid_done(sbi
, nid
);
1406 add_free_nid(NM_I(sbi
), nid
);
1409 void recover_node_page(struct f2fs_sb_info
*sbi
, struct page
*page
,
1410 struct f2fs_summary
*sum
, struct node_info
*ni
,
1411 block_t new_blkaddr
)
1413 rewrite_node_page(sbi
, page
, sum
, ni
->blk_addr
, new_blkaddr
);
1414 set_node_addr(sbi
, ni
, new_blkaddr
);
1415 clear_node_page_dirty(page
);
1418 int recover_inode_page(struct f2fs_sb_info
*sbi
, struct page
*page
)
1420 struct address_space
*mapping
= sbi
->node_inode
->i_mapping
;
1421 struct f2fs_node
*src
, *dst
;
1422 nid_t ino
= ino_of_node(page
);
1423 struct node_info old_ni
, new_ni
;
1426 ipage
= grab_cache_page(mapping
, ino
);
1430 /* Should not use this inode from free nid list */
1431 remove_free_nid(NM_I(sbi
), ino
);
1433 get_node_info(sbi
, ino
, &old_ni
);
1434 SetPageUptodate(ipage
);
1435 fill_node_footer(ipage
, ino
, ino
, 0, true);
1437 src
= (struct f2fs_node
*)page_address(page
);
1438 dst
= (struct f2fs_node
*)page_address(ipage
);
1440 memcpy(dst
, src
, (unsigned long)&src
->i
.i_ext
- (unsigned long)&src
->i
);
1442 dst
->i
.i_blocks
= cpu_to_le64(1);
1443 dst
->i
.i_links
= cpu_to_le32(1);
1444 dst
->i
.i_xattr_nid
= 0;
1449 set_node_addr(sbi
, &new_ni
, NEW_ADDR
);
1450 inc_valid_inode_count(sbi
);
1452 f2fs_put_page(ipage
, 1);
1456 int restore_node_summary(struct f2fs_sb_info
*sbi
,
1457 unsigned int segno
, struct f2fs_summary_block
*sum
)
1459 struct f2fs_node
*rn
;
1460 struct f2fs_summary
*sum_entry
;
1465 /* alloc temporal page for read node */
1466 page
= alloc_page(GFP_NOFS
| __GFP_ZERO
);
1468 return PTR_ERR(page
);
1471 /* scan the node segment */
1472 last_offset
= sbi
->blocks_per_seg
;
1473 addr
= START_BLOCK(sbi
, segno
);
1474 sum_entry
= &sum
->entries
[0];
1476 for (i
= 0; i
< last_offset
; i
++, sum_entry
++) {
1477 if (f2fs_readpage(sbi
, page
, addr
, READ_SYNC
))
1480 rn
= (struct f2fs_node
*)page_address(page
);
1481 sum_entry
->nid
= rn
->footer
.nid
;
1482 sum_entry
->version
= 0;
1483 sum_entry
->ofs_in_node
= 0;
1487 * In order to read next node page,
1488 * we must clear PageUptodate flag.
1490 ClearPageUptodate(page
);
1494 __free_pages(page
, 0);
1498 static bool flush_nats_in_journal(struct f2fs_sb_info
*sbi
)
1500 struct f2fs_nm_info
*nm_i
= NM_I(sbi
);
1501 struct curseg_info
*curseg
= CURSEG_I(sbi
, CURSEG_HOT_DATA
);
1502 struct f2fs_summary_block
*sum
= curseg
->sum_blk
;
1505 mutex_lock(&curseg
->curseg_mutex
);
1507 if (nats_in_cursum(sum
) < NAT_JOURNAL_ENTRIES
) {
1508 mutex_unlock(&curseg
->curseg_mutex
);
1512 for (i
= 0; i
< nats_in_cursum(sum
); i
++) {
1513 struct nat_entry
*ne
;
1514 struct f2fs_nat_entry raw_ne
;
1515 nid_t nid
= le32_to_cpu(nid_in_journal(sum
, i
));
1517 raw_ne
= nat_in_journal(sum
, i
);
1519 write_lock(&nm_i
->nat_tree_lock
);
1520 ne
= __lookup_nat_cache(nm_i
, nid
);
1522 __set_nat_cache_dirty(nm_i
, ne
);
1523 write_unlock(&nm_i
->nat_tree_lock
);
1526 ne
= grab_nat_entry(nm_i
, nid
);
1528 write_unlock(&nm_i
->nat_tree_lock
);
1531 nat_set_blkaddr(ne
, le32_to_cpu(raw_ne
.block_addr
));
1532 nat_set_ino(ne
, le32_to_cpu(raw_ne
.ino
));
1533 nat_set_version(ne
, raw_ne
.version
);
1534 __set_nat_cache_dirty(nm_i
, ne
);
1535 write_unlock(&nm_i
->nat_tree_lock
);
1537 update_nats_in_cursum(sum
, -i
);
1538 mutex_unlock(&curseg
->curseg_mutex
);
1543 * This function is called during the checkpointing process.
1545 void flush_nat_entries(struct f2fs_sb_info
*sbi
)
1547 struct f2fs_nm_info
*nm_i
= NM_I(sbi
);
1548 struct curseg_info
*curseg
= CURSEG_I(sbi
, CURSEG_HOT_DATA
);
1549 struct f2fs_summary_block
*sum
= curseg
->sum_blk
;
1550 struct list_head
*cur
, *n
;
1551 struct page
*page
= NULL
;
1552 struct f2fs_nat_block
*nat_blk
= NULL
;
1553 nid_t start_nid
= 0, end_nid
= 0;
1556 flushed
= flush_nats_in_journal(sbi
);
1559 mutex_lock(&curseg
->curseg_mutex
);
1561 /* 1) flush dirty nat caches */
1562 list_for_each_safe(cur
, n
, &nm_i
->dirty_nat_entries
) {
1563 struct nat_entry
*ne
;
1565 struct f2fs_nat_entry raw_ne
;
1567 block_t new_blkaddr
;
1569 ne
= list_entry(cur
, struct nat_entry
, list
);
1570 nid
= nat_get_nid(ne
);
1572 if (nat_get_blkaddr(ne
) == NEW_ADDR
)
1577 /* if there is room for nat enries in curseg->sumpage */
1578 offset
= lookup_journal_in_cursum(sum
, NAT_JOURNAL
, nid
, 1);
1580 raw_ne
= nat_in_journal(sum
, offset
);
1584 if (!page
|| (start_nid
> nid
|| nid
> end_nid
)) {
1586 f2fs_put_page(page
, 1);
1589 start_nid
= START_NID(nid
);
1590 end_nid
= start_nid
+ NAT_ENTRY_PER_BLOCK
- 1;
1593 * get nat block with dirty flag, increased reference
1594 * count, mapped and lock
1596 page
= get_next_nat_page(sbi
, start_nid
);
1597 nat_blk
= page_address(page
);
1601 raw_ne
= nat_blk
->entries
[nid
- start_nid
];
1603 new_blkaddr
= nat_get_blkaddr(ne
);
1605 raw_ne
.ino
= cpu_to_le32(nat_get_ino(ne
));
1606 raw_ne
.block_addr
= cpu_to_le32(new_blkaddr
);
1607 raw_ne
.version
= nat_get_version(ne
);
1610 nat_blk
->entries
[nid
- start_nid
] = raw_ne
;
1612 nat_in_journal(sum
, offset
) = raw_ne
;
1613 nid_in_journal(sum
, offset
) = cpu_to_le32(nid
);
1616 if (nat_get_blkaddr(ne
) == NULL_ADDR
) {
1617 write_lock(&nm_i
->nat_tree_lock
);
1618 __del_from_nat_cache(nm_i
, ne
);
1619 write_unlock(&nm_i
->nat_tree_lock
);
1621 /* We can reuse this freed nid at this point */
1622 add_free_nid(NM_I(sbi
), nid
);
1624 write_lock(&nm_i
->nat_tree_lock
);
1625 __clear_nat_cache_dirty(nm_i
, ne
);
1626 ne
->checkpointed
= true;
1627 write_unlock(&nm_i
->nat_tree_lock
);
1631 mutex_unlock(&curseg
->curseg_mutex
);
1632 f2fs_put_page(page
, 1);
1634 /* 2) shrink nat caches if necessary */
1635 try_to_free_nats(sbi
, nm_i
->nat_cnt
- NM_WOUT_THRESHOLD
);
1638 static int init_node_manager(struct f2fs_sb_info
*sbi
)
1640 struct f2fs_super_block
*sb_raw
= F2FS_RAW_SUPER(sbi
);
1641 struct f2fs_nm_info
*nm_i
= NM_I(sbi
);
1642 unsigned char *version_bitmap
;
1643 unsigned int nat_segs
, nat_blocks
;
1645 nm_i
->nat_blkaddr
= le32_to_cpu(sb_raw
->nat_blkaddr
);
1647 /* segment_count_nat includes pair segment so divide to 2. */
1648 nat_segs
= le32_to_cpu(sb_raw
->segment_count_nat
) >> 1;
1649 nat_blocks
= nat_segs
<< le32_to_cpu(sb_raw
->log_blocks_per_seg
);
1650 nm_i
->max_nid
= NAT_ENTRY_PER_BLOCK
* nat_blocks
;
1654 INIT_LIST_HEAD(&nm_i
->free_nid_list
);
1655 INIT_RADIX_TREE(&nm_i
->nat_root
, GFP_ATOMIC
);
1656 INIT_LIST_HEAD(&nm_i
->nat_entries
);
1657 INIT_LIST_HEAD(&nm_i
->dirty_nat_entries
);
1659 mutex_init(&nm_i
->build_lock
);
1660 spin_lock_init(&nm_i
->free_nid_list_lock
);
1661 rwlock_init(&nm_i
->nat_tree_lock
);
1663 nm_i
->bitmap_size
= __bitmap_size(sbi
, NAT_BITMAP
);
1664 nm_i
->init_scan_nid
= le32_to_cpu(sbi
->ckpt
->next_free_nid
);
1665 nm_i
->next_scan_nid
= le32_to_cpu(sbi
->ckpt
->next_free_nid
);
1667 nm_i
->nat_bitmap
= kzalloc(nm_i
->bitmap_size
, GFP_KERNEL
);
1668 if (!nm_i
->nat_bitmap
)
1670 version_bitmap
= __bitmap_ptr(sbi
, NAT_BITMAP
);
1671 if (!version_bitmap
)
1674 /* copy version bitmap */
1675 memcpy(nm_i
->nat_bitmap
, version_bitmap
, nm_i
->bitmap_size
);
1679 int build_node_manager(struct f2fs_sb_info
*sbi
)
1683 sbi
->nm_info
= kzalloc(sizeof(struct f2fs_nm_info
), GFP_KERNEL
);
1687 err
= init_node_manager(sbi
);
1691 build_free_nids(sbi
);
1695 void destroy_node_manager(struct f2fs_sb_info
*sbi
)
1697 struct f2fs_nm_info
*nm_i
= NM_I(sbi
);
1698 struct free_nid
*i
, *next_i
;
1699 struct nat_entry
*natvec
[NATVEC_SIZE
];
1706 /* destroy free nid list */
1707 spin_lock(&nm_i
->free_nid_list_lock
);
1708 list_for_each_entry_safe(i
, next_i
, &nm_i
->free_nid_list
, list
) {
1709 BUG_ON(i
->state
== NID_ALLOC
);
1710 __del_from_free_nid_list(i
);
1714 spin_unlock(&nm_i
->free_nid_list_lock
);
1716 /* destroy nat cache */
1717 write_lock(&nm_i
->nat_tree_lock
);
1718 while ((found
= __gang_lookup_nat_cache(nm_i
,
1719 nid
, NATVEC_SIZE
, natvec
))) {
1721 for (idx
= 0; idx
< found
; idx
++) {
1722 struct nat_entry
*e
= natvec
[idx
];
1723 nid
= nat_get_nid(e
) + 1;
1724 __del_from_nat_cache(nm_i
, e
);
1727 BUG_ON(nm_i
->nat_cnt
);
1728 write_unlock(&nm_i
->nat_tree_lock
);
1730 kfree(nm_i
->nat_bitmap
);
1731 sbi
->nm_info
= NULL
;
1735 int create_node_manager_caches(void)
1737 nat_entry_slab
= f2fs_kmem_cache_create("nat_entry",
1738 sizeof(struct nat_entry
), NULL
);
1739 if (!nat_entry_slab
)
1742 free_nid_slab
= f2fs_kmem_cache_create("free_nid",
1743 sizeof(struct free_nid
), NULL
);
1744 if (!free_nid_slab
) {
1745 kmem_cache_destroy(nat_entry_slab
);
1751 void destroy_node_manager_caches(void)
1753 kmem_cache_destroy(free_nid_slab
);
1754 kmem_cache_destroy(nat_entry_slab
);