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
);
95 struct f2fs_io_info fio
= {
97 .rw
= READ_SYNC
| REQ_META
| REQ_PRIO
101 for (i
= 0; i
< FREE_NID_PAGES
; i
++, nid
+= NAT_ENTRY_PER_BLOCK
) {
102 if (unlikely(nid
>= nm_i
->max_nid
))
104 index
= current_nat_addr(sbi
, nid
);
106 page
= grab_cache_page(mapping
, index
);
109 if (PageUptodate(page
)) {
110 mark_page_accessed(page
);
111 f2fs_put_page(page
, 1);
114 f2fs_submit_page_mbio(sbi
, page
, index
, &fio
);
115 mark_page_accessed(page
);
116 f2fs_put_page(page
, 0);
118 f2fs_submit_merged_bio(sbi
, META
, READ
);
121 static struct nat_entry
*__lookup_nat_cache(struct f2fs_nm_info
*nm_i
, nid_t n
)
123 return radix_tree_lookup(&nm_i
->nat_root
, n
);
126 static unsigned int __gang_lookup_nat_cache(struct f2fs_nm_info
*nm_i
,
127 nid_t start
, unsigned int nr
, struct nat_entry
**ep
)
129 return radix_tree_gang_lookup(&nm_i
->nat_root
, (void **)ep
, start
, nr
);
132 static void __del_from_nat_cache(struct f2fs_nm_info
*nm_i
, struct nat_entry
*e
)
135 radix_tree_delete(&nm_i
->nat_root
, nat_get_nid(e
));
137 kmem_cache_free(nat_entry_slab
, e
);
140 int is_checkpointed_node(struct f2fs_sb_info
*sbi
, nid_t nid
)
142 struct f2fs_nm_info
*nm_i
= NM_I(sbi
);
146 read_lock(&nm_i
->nat_tree_lock
);
147 e
= __lookup_nat_cache(nm_i
, nid
);
148 if (e
&& !e
->checkpointed
)
150 read_unlock(&nm_i
->nat_tree_lock
);
154 static struct nat_entry
*grab_nat_entry(struct f2fs_nm_info
*nm_i
, nid_t nid
)
156 struct nat_entry
*new;
158 new = kmem_cache_alloc(nat_entry_slab
, GFP_ATOMIC
);
161 if (radix_tree_insert(&nm_i
->nat_root
, nid
, new)) {
162 kmem_cache_free(nat_entry_slab
, new);
165 memset(new, 0, sizeof(struct nat_entry
));
166 nat_set_nid(new, nid
);
167 list_add_tail(&new->list
, &nm_i
->nat_entries
);
172 static void cache_nat_entry(struct f2fs_nm_info
*nm_i
, nid_t nid
,
173 struct f2fs_nat_entry
*ne
)
177 write_lock(&nm_i
->nat_tree_lock
);
178 e
= __lookup_nat_cache(nm_i
, nid
);
180 e
= grab_nat_entry(nm_i
, nid
);
182 write_unlock(&nm_i
->nat_tree_lock
);
185 nat_set_blkaddr(e
, le32_to_cpu(ne
->block_addr
));
186 nat_set_ino(e
, le32_to_cpu(ne
->ino
));
187 nat_set_version(e
, ne
->version
);
188 e
->checkpointed
= true;
190 write_unlock(&nm_i
->nat_tree_lock
);
193 static void set_node_addr(struct f2fs_sb_info
*sbi
, struct node_info
*ni
,
196 struct f2fs_nm_info
*nm_i
= NM_I(sbi
);
199 write_lock(&nm_i
->nat_tree_lock
);
200 e
= __lookup_nat_cache(nm_i
, ni
->nid
);
202 e
= grab_nat_entry(nm_i
, ni
->nid
);
204 write_unlock(&nm_i
->nat_tree_lock
);
208 e
->checkpointed
= true;
209 f2fs_bug_on(ni
->blk_addr
== NEW_ADDR
);
210 } else if (new_blkaddr
== NEW_ADDR
) {
212 * when nid is reallocated,
213 * previous nat entry can be remained in nat cache.
214 * So, reinitialize it with new information.
217 f2fs_bug_on(ni
->blk_addr
!= NULL_ADDR
);
220 if (new_blkaddr
== NEW_ADDR
)
221 e
->checkpointed
= false;
224 f2fs_bug_on(nat_get_blkaddr(e
) != ni
->blk_addr
);
225 f2fs_bug_on(nat_get_blkaddr(e
) == NULL_ADDR
&&
226 new_blkaddr
== NULL_ADDR
);
227 f2fs_bug_on(nat_get_blkaddr(e
) == NEW_ADDR
&&
228 new_blkaddr
== NEW_ADDR
);
229 f2fs_bug_on(nat_get_blkaddr(e
) != NEW_ADDR
&&
230 nat_get_blkaddr(e
) != NULL_ADDR
&&
231 new_blkaddr
== NEW_ADDR
);
233 /* increament version no as node is removed */
234 if (nat_get_blkaddr(e
) != NEW_ADDR
&& new_blkaddr
== NULL_ADDR
) {
235 unsigned char version
= nat_get_version(e
);
236 nat_set_version(e
, inc_node_version(version
));
240 nat_set_blkaddr(e
, new_blkaddr
);
241 __set_nat_cache_dirty(nm_i
, e
);
242 write_unlock(&nm_i
->nat_tree_lock
);
245 int try_to_free_nats(struct f2fs_sb_info
*sbi
, int nr_shrink
)
247 struct f2fs_nm_info
*nm_i
= NM_I(sbi
);
249 if (nm_i
->nat_cnt
<= NM_WOUT_THRESHOLD
)
252 write_lock(&nm_i
->nat_tree_lock
);
253 while (nr_shrink
&& !list_empty(&nm_i
->nat_entries
)) {
254 struct nat_entry
*ne
;
255 ne
= list_first_entry(&nm_i
->nat_entries
,
256 struct nat_entry
, list
);
257 __del_from_nat_cache(nm_i
, ne
);
260 write_unlock(&nm_i
->nat_tree_lock
);
265 * This function returns always success
267 void get_node_info(struct f2fs_sb_info
*sbi
, nid_t nid
, struct node_info
*ni
)
269 struct f2fs_nm_info
*nm_i
= NM_I(sbi
);
270 struct curseg_info
*curseg
= CURSEG_I(sbi
, CURSEG_HOT_DATA
);
271 struct f2fs_summary_block
*sum
= curseg
->sum_blk
;
272 nid_t start_nid
= START_NID(nid
);
273 struct f2fs_nat_block
*nat_blk
;
274 struct page
*page
= NULL
;
275 struct f2fs_nat_entry ne
;
279 memset(&ne
, 0, sizeof(struct f2fs_nat_entry
));
282 /* Check nat cache */
283 read_lock(&nm_i
->nat_tree_lock
);
284 e
= __lookup_nat_cache(nm_i
, nid
);
286 ni
->ino
= nat_get_ino(e
);
287 ni
->blk_addr
= nat_get_blkaddr(e
);
288 ni
->version
= nat_get_version(e
);
290 read_unlock(&nm_i
->nat_tree_lock
);
294 /* Check current segment summary */
295 mutex_lock(&curseg
->curseg_mutex
);
296 i
= lookup_journal_in_cursum(sum
, NAT_JOURNAL
, nid
, 0);
298 ne
= nat_in_journal(sum
, i
);
299 node_info_from_raw_nat(ni
, &ne
);
301 mutex_unlock(&curseg
->curseg_mutex
);
305 /* Fill node_info from nat page */
306 page
= get_current_nat_page(sbi
, start_nid
);
307 nat_blk
= (struct f2fs_nat_block
*)page_address(page
);
308 ne
= nat_blk
->entries
[nid
- start_nid
];
309 node_info_from_raw_nat(ni
, &ne
);
310 f2fs_put_page(page
, 1);
312 /* cache nat entry */
313 cache_nat_entry(NM_I(sbi
), nid
, &ne
);
317 * The maximum depth is four.
318 * Offset[0] will have raw inode offset.
320 static int get_node_path(struct f2fs_inode_info
*fi
, long block
,
321 int offset
[4], unsigned int noffset
[4])
323 const long direct_index
= ADDRS_PER_INODE(fi
);
324 const long direct_blks
= ADDRS_PER_BLOCK
;
325 const long dptrs_per_blk
= NIDS_PER_BLOCK
;
326 const long indirect_blks
= ADDRS_PER_BLOCK
* NIDS_PER_BLOCK
;
327 const long dindirect_blks
= indirect_blks
* NIDS_PER_BLOCK
;
333 if (block
< direct_index
) {
337 block
-= direct_index
;
338 if (block
< direct_blks
) {
339 offset
[n
++] = NODE_DIR1_BLOCK
;
345 block
-= direct_blks
;
346 if (block
< direct_blks
) {
347 offset
[n
++] = NODE_DIR2_BLOCK
;
353 block
-= direct_blks
;
354 if (block
< indirect_blks
) {
355 offset
[n
++] = NODE_IND1_BLOCK
;
357 offset
[n
++] = block
/ direct_blks
;
358 noffset
[n
] = 4 + offset
[n
- 1];
359 offset
[n
] = block
% direct_blks
;
363 block
-= indirect_blks
;
364 if (block
< indirect_blks
) {
365 offset
[n
++] = NODE_IND2_BLOCK
;
366 noffset
[n
] = 4 + dptrs_per_blk
;
367 offset
[n
++] = block
/ direct_blks
;
368 noffset
[n
] = 5 + dptrs_per_blk
+ offset
[n
- 1];
369 offset
[n
] = block
% direct_blks
;
373 block
-= indirect_blks
;
374 if (block
< dindirect_blks
) {
375 offset
[n
++] = NODE_DIND_BLOCK
;
376 noffset
[n
] = 5 + (dptrs_per_blk
* 2);
377 offset
[n
++] = block
/ indirect_blks
;
378 noffset
[n
] = 6 + (dptrs_per_blk
* 2) +
379 offset
[n
- 1] * (dptrs_per_blk
+ 1);
380 offset
[n
++] = (block
/ direct_blks
) % dptrs_per_blk
;
381 noffset
[n
] = 7 + (dptrs_per_blk
* 2) +
382 offset
[n
- 2] * (dptrs_per_blk
+ 1) +
384 offset
[n
] = block
% direct_blks
;
395 * Caller should call f2fs_put_dnode(dn).
396 * Also, it should grab and release a mutex by calling mutex_lock_op() and
397 * mutex_unlock_op() only if ro is not set RDONLY_NODE.
398 * In the case of RDONLY_NODE, we don't need to care about mutex.
400 int get_dnode_of_data(struct dnode_of_data
*dn
, pgoff_t index
, int mode
)
402 struct f2fs_sb_info
*sbi
= F2FS_SB(dn
->inode
->i_sb
);
403 struct page
*npage
[4];
406 unsigned int noffset
[4];
411 level
= get_node_path(F2FS_I(dn
->inode
), index
, offset
, noffset
);
413 nids
[0] = dn
->inode
->i_ino
;
414 npage
[0] = dn
->inode_page
;
417 npage
[0] = get_node_page(sbi
, nids
[0]);
418 if (IS_ERR(npage
[0]))
419 return PTR_ERR(npage
[0]);
423 nids
[1] = get_nid(parent
, offset
[0], true);
424 dn
->inode_page
= npage
[0];
425 dn
->inode_page_locked
= true;
427 /* get indirect or direct nodes */
428 for (i
= 1; i
<= level
; i
++) {
431 if (!nids
[i
] && mode
== ALLOC_NODE
) {
433 if (!alloc_nid(sbi
, &(nids
[i
]))) {
439 npage
[i
] = new_node_page(dn
, noffset
[i
], NULL
);
440 if (IS_ERR(npage
[i
])) {
441 alloc_nid_failed(sbi
, nids
[i
]);
442 err
= PTR_ERR(npage
[i
]);
446 set_nid(parent
, offset
[i
- 1], nids
[i
], i
== 1);
447 alloc_nid_done(sbi
, nids
[i
]);
449 } else if (mode
== LOOKUP_NODE_RA
&& i
== level
&& level
> 1) {
450 npage
[i
] = get_node_page_ra(parent
, offset
[i
- 1]);
451 if (IS_ERR(npage
[i
])) {
452 err
= PTR_ERR(npage
[i
]);
458 dn
->inode_page_locked
= false;
461 f2fs_put_page(parent
, 1);
465 npage
[i
] = get_node_page(sbi
, nids
[i
]);
466 if (IS_ERR(npage
[i
])) {
467 err
= PTR_ERR(npage
[i
]);
468 f2fs_put_page(npage
[0], 0);
474 nids
[i
+ 1] = get_nid(parent
, offset
[i
], false);
477 dn
->nid
= nids
[level
];
478 dn
->ofs_in_node
= offset
[level
];
479 dn
->node_page
= npage
[level
];
480 dn
->data_blkaddr
= datablock_addr(dn
->node_page
, dn
->ofs_in_node
);
484 f2fs_put_page(parent
, 1);
486 f2fs_put_page(npage
[0], 0);
488 dn
->inode_page
= NULL
;
489 dn
->node_page
= NULL
;
493 static void truncate_node(struct dnode_of_data
*dn
)
495 struct f2fs_sb_info
*sbi
= F2FS_SB(dn
->inode
->i_sb
);
498 get_node_info(sbi
, dn
->nid
, &ni
);
499 if (dn
->inode
->i_blocks
== 0) {
500 f2fs_bug_on(ni
.blk_addr
!= NULL_ADDR
);
503 f2fs_bug_on(ni
.blk_addr
== NULL_ADDR
);
505 /* Deallocate node address */
506 invalidate_blocks(sbi
, ni
.blk_addr
);
507 dec_valid_node_count(sbi
, dn
->inode
);
508 set_node_addr(sbi
, &ni
, NULL_ADDR
);
510 if (dn
->nid
== dn
->inode
->i_ino
) {
511 remove_orphan_inode(sbi
, dn
->nid
);
512 dec_valid_inode_count(sbi
);
517 clear_node_page_dirty(dn
->node_page
);
518 F2FS_SET_SB_DIRT(sbi
);
520 f2fs_put_page(dn
->node_page
, 1);
521 dn
->node_page
= NULL
;
522 trace_f2fs_truncate_node(dn
->inode
, dn
->nid
, ni
.blk_addr
);
525 static int truncate_dnode(struct dnode_of_data
*dn
)
527 struct f2fs_sb_info
*sbi
= F2FS_SB(dn
->inode
->i_sb
);
533 /* get direct node */
534 page
= get_node_page(sbi
, dn
->nid
);
535 if (IS_ERR(page
) && PTR_ERR(page
) == -ENOENT
)
537 else if (IS_ERR(page
))
538 return PTR_ERR(page
);
540 /* Make dnode_of_data for parameter */
541 dn
->node_page
= page
;
543 truncate_data_blocks(dn
);
548 static int truncate_nodes(struct dnode_of_data
*dn
, unsigned int nofs
,
551 struct f2fs_sb_info
*sbi
= F2FS_SB(dn
->inode
->i_sb
);
552 struct dnode_of_data rdn
= *dn
;
554 struct f2fs_node
*rn
;
556 unsigned int child_nofs
;
561 return NIDS_PER_BLOCK
+ 1;
563 trace_f2fs_truncate_nodes_enter(dn
->inode
, dn
->nid
, dn
->data_blkaddr
);
565 page
= get_node_page(sbi
, dn
->nid
);
567 trace_f2fs_truncate_nodes_exit(dn
->inode
, PTR_ERR(page
));
568 return PTR_ERR(page
);
571 rn
= F2FS_NODE(page
);
573 for (i
= ofs
; i
< NIDS_PER_BLOCK
; i
++, freed
++) {
574 child_nid
= le32_to_cpu(rn
->in
.nid
[i
]);
578 ret
= truncate_dnode(&rdn
);
581 set_nid(page
, i
, 0, false);
584 child_nofs
= nofs
+ ofs
* (NIDS_PER_BLOCK
+ 1) + 1;
585 for (i
= ofs
; i
< NIDS_PER_BLOCK
; i
++) {
586 child_nid
= le32_to_cpu(rn
->in
.nid
[i
]);
587 if (child_nid
== 0) {
588 child_nofs
+= NIDS_PER_BLOCK
+ 1;
592 ret
= truncate_nodes(&rdn
, child_nofs
, 0, depth
- 1);
593 if (ret
== (NIDS_PER_BLOCK
+ 1)) {
594 set_nid(page
, i
, 0, false);
596 } else if (ret
< 0 && ret
!= -ENOENT
) {
604 /* remove current indirect node */
605 dn
->node_page
= page
;
609 f2fs_put_page(page
, 1);
611 trace_f2fs_truncate_nodes_exit(dn
->inode
, freed
);
615 f2fs_put_page(page
, 1);
616 trace_f2fs_truncate_nodes_exit(dn
->inode
, ret
);
620 static int truncate_partial_nodes(struct dnode_of_data
*dn
,
621 struct f2fs_inode
*ri
, int *offset
, int depth
)
623 struct f2fs_sb_info
*sbi
= F2FS_SB(dn
->inode
->i_sb
);
624 struct page
*pages
[2];
631 nid
[0] = le32_to_cpu(ri
->i_nid
[offset
[0] - NODE_DIR1_BLOCK
]);
635 /* get indirect nodes in the path */
636 for (i
= 0; i
< depth
- 1; i
++) {
637 /* refernece count'll be increased */
638 pages
[i
] = get_node_page(sbi
, nid
[i
]);
639 if (IS_ERR(pages
[i
])) {
641 err
= PTR_ERR(pages
[i
]);
644 nid
[i
+ 1] = get_nid(pages
[i
], offset
[i
+ 1], false);
647 /* free direct nodes linked to a partial indirect node */
648 for (i
= offset
[depth
- 1]; i
< NIDS_PER_BLOCK
; i
++) {
649 child_nid
= get_nid(pages
[idx
], i
, false);
653 err
= truncate_dnode(dn
);
656 set_nid(pages
[idx
], i
, 0, false);
659 if (offset
[depth
- 1] == 0) {
660 dn
->node_page
= pages
[idx
];
664 f2fs_put_page(pages
[idx
], 1);
667 offset
[depth
- 1] = 0;
669 for (i
= depth
- 3; i
>= 0; i
--)
670 f2fs_put_page(pages
[i
], 1);
672 trace_f2fs_truncate_partial_nodes(dn
->inode
, nid
, depth
, err
);
678 * All the block addresses of data and nodes should be nullified.
680 int truncate_inode_blocks(struct inode
*inode
, pgoff_t from
)
682 struct f2fs_sb_info
*sbi
= F2FS_SB(inode
->i_sb
);
683 struct address_space
*node_mapping
= sbi
->node_inode
->i_mapping
;
684 int err
= 0, cont
= 1;
685 int level
, offset
[4], noffset
[4];
686 unsigned int nofs
= 0;
687 struct f2fs_node
*rn
;
688 struct dnode_of_data dn
;
691 trace_f2fs_truncate_inode_blocks_enter(inode
, from
);
693 level
= get_node_path(F2FS_I(inode
), from
, offset
, noffset
);
695 page
= get_node_page(sbi
, inode
->i_ino
);
697 trace_f2fs_truncate_inode_blocks_exit(inode
, PTR_ERR(page
));
698 return PTR_ERR(page
);
701 set_new_dnode(&dn
, inode
, page
, NULL
, 0);
704 rn
= F2FS_NODE(page
);
712 if (!offset
[level
- 1])
714 err
= truncate_partial_nodes(&dn
, &rn
->i
, offset
, level
);
715 if (err
< 0 && err
!= -ENOENT
)
717 nofs
+= 1 + NIDS_PER_BLOCK
;
720 nofs
= 5 + 2 * NIDS_PER_BLOCK
;
721 if (!offset
[level
- 1])
723 err
= truncate_partial_nodes(&dn
, &rn
->i
, offset
, level
);
724 if (err
< 0 && err
!= -ENOENT
)
733 dn
.nid
= le32_to_cpu(rn
->i
.i_nid
[offset
[0] - NODE_DIR1_BLOCK
]);
735 case NODE_DIR1_BLOCK
:
736 case NODE_DIR2_BLOCK
:
737 err
= truncate_dnode(&dn
);
740 case NODE_IND1_BLOCK
:
741 case NODE_IND2_BLOCK
:
742 err
= truncate_nodes(&dn
, nofs
, offset
[1], 2);
745 case NODE_DIND_BLOCK
:
746 err
= truncate_nodes(&dn
, nofs
, offset
[1], 3);
753 if (err
< 0 && err
!= -ENOENT
)
755 if (offset
[1] == 0 &&
756 rn
->i
.i_nid
[offset
[0] - NODE_DIR1_BLOCK
]) {
758 if (unlikely(page
->mapping
!= node_mapping
)) {
759 f2fs_put_page(page
, 1);
762 wait_on_page_writeback(page
);
763 rn
->i
.i_nid
[offset
[0] - NODE_DIR1_BLOCK
] = 0;
764 set_page_dirty(page
);
772 f2fs_put_page(page
, 0);
773 trace_f2fs_truncate_inode_blocks_exit(inode
, err
);
774 return err
> 0 ? 0 : err
;
777 int truncate_xattr_node(struct inode
*inode
, struct page
*page
)
779 struct f2fs_sb_info
*sbi
= F2FS_SB(inode
->i_sb
);
780 nid_t nid
= F2FS_I(inode
)->i_xattr_nid
;
781 struct dnode_of_data dn
;
787 npage
= get_node_page(sbi
, nid
);
789 return PTR_ERR(npage
);
791 F2FS_I(inode
)->i_xattr_nid
= 0;
793 /* need to do checkpoint during fsync */
794 F2FS_I(inode
)->xattr_ver
= cur_cp_version(F2FS_CKPT(sbi
));
796 set_new_dnode(&dn
, inode
, page
, npage
, nid
);
799 dn
.inode_page_locked
= true;
805 * Caller should grab and release a mutex by calling mutex_lock_op() and
808 void remove_inode_page(struct inode
*inode
)
810 struct f2fs_sb_info
*sbi
= F2FS_SB(inode
->i_sb
);
812 nid_t ino
= inode
->i_ino
;
813 struct dnode_of_data dn
;
815 page
= get_node_page(sbi
, ino
);
819 if (truncate_xattr_node(inode
, page
)) {
820 f2fs_put_page(page
, 1);
823 /* 0 is possible, after f2fs_new_inode() is failed */
824 f2fs_bug_on(inode
->i_blocks
!= 0 && inode
->i_blocks
!= 1);
825 set_new_dnode(&dn
, inode
, page
, page
, ino
);
829 struct page
*new_inode_page(struct inode
*inode
, const struct qstr
*name
)
831 struct dnode_of_data dn
;
833 /* allocate inode page for new inode */
834 set_new_dnode(&dn
, inode
, NULL
, NULL
, inode
->i_ino
);
836 /* caller should f2fs_put_page(page, 1); */
837 return new_node_page(&dn
, 0, NULL
);
840 struct page
*new_node_page(struct dnode_of_data
*dn
,
841 unsigned int ofs
, struct page
*ipage
)
843 struct f2fs_sb_info
*sbi
= F2FS_SB(dn
->inode
->i_sb
);
844 struct address_space
*mapping
= sbi
->node_inode
->i_mapping
;
845 struct node_info old_ni
, new_ni
;
849 if (unlikely(is_inode_flag_set(F2FS_I(dn
->inode
), FI_NO_ALLOC
)))
850 return ERR_PTR(-EPERM
);
852 page
= grab_cache_page(mapping
, dn
->nid
);
854 return ERR_PTR(-ENOMEM
);
856 if (unlikely(!inc_valid_node_count(sbi
, dn
->inode
))) {
861 get_node_info(sbi
, dn
->nid
, &old_ni
);
863 /* Reinitialize old_ni with new node page */
864 f2fs_bug_on(old_ni
.blk_addr
!= NULL_ADDR
);
866 new_ni
.ino
= dn
->inode
->i_ino
;
867 set_node_addr(sbi
, &new_ni
, NEW_ADDR
);
869 fill_node_footer(page
, dn
->nid
, dn
->inode
->i_ino
, ofs
, true);
870 set_cold_node(dn
->inode
, page
);
871 SetPageUptodate(page
);
872 set_page_dirty(page
);
874 if (ofs
== XATTR_NODE_OFFSET
)
875 F2FS_I(dn
->inode
)->i_xattr_nid
= dn
->nid
;
877 dn
->node_page
= page
;
879 update_inode(dn
->inode
, ipage
);
883 inc_valid_inode_count(sbi
);
888 clear_node_page_dirty(page
);
889 f2fs_put_page(page
, 1);
894 * Caller should do after getting the following values.
895 * 0: f2fs_put_page(page, 0)
896 * LOCKED_PAGE: f2fs_put_page(page, 1)
899 static int read_node_page(struct page
*page
, int rw
)
901 struct f2fs_sb_info
*sbi
= F2FS_SB(page
->mapping
->host
->i_sb
);
904 get_node_info(sbi
, page
->index
, &ni
);
906 if (unlikely(ni
.blk_addr
== NULL_ADDR
)) {
907 f2fs_put_page(page
, 1);
911 if (PageUptodate(page
))
914 return f2fs_submit_page_bio(sbi
, page
, ni
.blk_addr
, rw
);
918 * Readahead a node page
920 void ra_node_page(struct f2fs_sb_info
*sbi
, nid_t nid
)
922 struct address_space
*mapping
= sbi
->node_inode
->i_mapping
;
926 apage
= find_get_page(mapping
, nid
);
927 if (apage
&& PageUptodate(apage
)) {
928 f2fs_put_page(apage
, 0);
931 f2fs_put_page(apage
, 0);
933 apage
= grab_cache_page(mapping
, nid
);
937 err
= read_node_page(apage
, READA
);
939 f2fs_put_page(apage
, 0);
940 else if (err
== LOCKED_PAGE
)
941 f2fs_put_page(apage
, 1);
944 struct page
*get_node_page(struct f2fs_sb_info
*sbi
, pgoff_t nid
)
946 struct address_space
*mapping
= sbi
->node_inode
->i_mapping
;
950 page
= grab_cache_page(mapping
, nid
);
952 return ERR_PTR(-ENOMEM
);
954 err
= read_node_page(page
, READ_SYNC
);
957 else if (err
== LOCKED_PAGE
)
961 if (unlikely(!PageUptodate(page
))) {
962 f2fs_put_page(page
, 1);
963 return ERR_PTR(-EIO
);
965 if (unlikely(page
->mapping
!= mapping
)) {
966 f2fs_put_page(page
, 1);
970 f2fs_bug_on(nid
!= nid_of_node(page
));
971 mark_page_accessed(page
);
976 * Return a locked page for the desired node page.
977 * And, readahead MAX_RA_NODE number of node pages.
979 struct page
*get_node_page_ra(struct page
*parent
, int start
)
981 struct f2fs_sb_info
*sbi
= F2FS_SB(parent
->mapping
->host
->i_sb
);
982 struct address_space
*mapping
= sbi
->node_inode
->i_mapping
;
983 struct blk_plug plug
;
988 /* First, try getting the desired direct node. */
989 nid
= get_nid(parent
, start
, false);
991 return ERR_PTR(-ENOENT
);
993 page
= grab_cache_page(mapping
, nid
);
995 return ERR_PTR(-ENOMEM
);
997 err
= read_node_page(page
, READ_SYNC
);
1000 else if (err
== LOCKED_PAGE
)
1003 blk_start_plug(&plug
);
1005 /* Then, try readahead for siblings of the desired node */
1006 end
= start
+ MAX_RA_NODE
;
1007 end
= min(end
, NIDS_PER_BLOCK
);
1008 for (i
= start
+ 1; i
< end
; i
++) {
1009 nid
= get_nid(parent
, i
, false);
1012 ra_node_page(sbi
, nid
);
1015 blk_finish_plug(&plug
);
1018 if (unlikely(page
->mapping
!= mapping
)) {
1019 f2fs_put_page(page
, 1);
1023 if (unlikely(!PageUptodate(page
))) {
1024 f2fs_put_page(page
, 1);
1025 return ERR_PTR(-EIO
);
1027 mark_page_accessed(page
);
1031 void sync_inode_page(struct dnode_of_data
*dn
)
1033 if (IS_INODE(dn
->node_page
) || dn
->inode_page
== dn
->node_page
) {
1034 update_inode(dn
->inode
, dn
->node_page
);
1035 } else if (dn
->inode_page
) {
1036 if (!dn
->inode_page_locked
)
1037 lock_page(dn
->inode_page
);
1038 update_inode(dn
->inode
, dn
->inode_page
);
1039 if (!dn
->inode_page_locked
)
1040 unlock_page(dn
->inode_page
);
1042 update_inode_page(dn
->inode
);
1046 int sync_node_pages(struct f2fs_sb_info
*sbi
, nid_t ino
,
1047 struct writeback_control
*wbc
)
1049 struct address_space
*mapping
= sbi
->node_inode
->i_mapping
;
1051 struct pagevec pvec
;
1052 int step
= ino
? 2 : 0;
1053 int nwritten
= 0, wrote
= 0;
1055 pagevec_init(&pvec
, 0);
1061 while (index
<= end
) {
1063 nr_pages
= pagevec_lookup_tag(&pvec
, mapping
, &index
,
1064 PAGECACHE_TAG_DIRTY
,
1065 min(end
- index
, (pgoff_t
)PAGEVEC_SIZE
-1) + 1);
1069 for (i
= 0; i
< nr_pages
; i
++) {
1070 struct page
*page
= pvec
.pages
[i
];
1073 * flushing sequence with step:
1078 if (step
== 0 && IS_DNODE(page
))
1080 if (step
== 1 && (!IS_DNODE(page
) ||
1081 is_cold_node(page
)))
1083 if (step
== 2 && (!IS_DNODE(page
) ||
1084 !is_cold_node(page
)))
1089 * we should not skip writing node pages.
1091 if (ino
&& ino_of_node(page
) == ino
)
1093 else if (!trylock_page(page
))
1096 if (unlikely(page
->mapping
!= mapping
)) {
1101 if (ino
&& ino_of_node(page
) != ino
)
1102 goto continue_unlock
;
1104 if (!PageDirty(page
)) {
1105 /* someone wrote it for us */
1106 goto continue_unlock
;
1109 if (!clear_page_dirty_for_io(page
))
1110 goto continue_unlock
;
1112 /* called by fsync() */
1113 if (ino
&& IS_DNODE(page
)) {
1114 int mark
= !is_checkpointed_node(sbi
, ino
);
1115 set_fsync_mark(page
, 1);
1117 set_dentry_mark(page
, mark
);
1120 set_fsync_mark(page
, 0);
1121 set_dentry_mark(page
, 0);
1123 mapping
->a_ops
->writepage(page
, wbc
);
1126 if (--wbc
->nr_to_write
== 0)
1129 pagevec_release(&pvec
);
1132 if (wbc
->nr_to_write
== 0) {
1144 f2fs_submit_merged_bio(sbi
, NODE
, WRITE
);
1148 int wait_on_node_pages_writeback(struct f2fs_sb_info
*sbi
, nid_t ino
)
1150 struct address_space
*mapping
= sbi
->node_inode
->i_mapping
;
1151 pgoff_t index
= 0, end
= LONG_MAX
;
1152 struct pagevec pvec
;
1154 int ret2
= 0, ret
= 0;
1156 pagevec_init(&pvec
, 0);
1157 while ((index
<= end
) &&
1158 (nr_pages
= pagevec_lookup_tag(&pvec
, mapping
, &index
,
1159 PAGECACHE_TAG_WRITEBACK
,
1160 min(end
- index
, (pgoff_t
)PAGEVEC_SIZE
-1) + 1)) != 0) {
1163 for (i
= 0; i
< nr_pages
; i
++) {
1164 struct page
*page
= pvec
.pages
[i
];
1166 /* until radix tree lookup accepts end_index */
1167 if (unlikely(page
->index
> end
))
1170 if (ino
&& ino_of_node(page
) == ino
) {
1171 wait_on_page_writeback(page
);
1172 if (TestClearPageError(page
))
1176 pagevec_release(&pvec
);
1180 if (unlikely(test_and_clear_bit(AS_ENOSPC
, &mapping
->flags
)))
1182 if (unlikely(test_and_clear_bit(AS_EIO
, &mapping
->flags
)))
1189 static int f2fs_write_node_page(struct page
*page
,
1190 struct writeback_control
*wbc
)
1192 struct f2fs_sb_info
*sbi
= F2FS_SB(page
->mapping
->host
->i_sb
);
1195 struct node_info ni
;
1197 if (unlikely(sbi
->por_doing
))
1200 wait_on_page_writeback(page
);
1202 /* get old block addr of this node page */
1203 nid
= nid_of_node(page
);
1204 f2fs_bug_on(page
->index
!= nid
);
1206 get_node_info(sbi
, nid
, &ni
);
1208 /* This page is already truncated */
1209 if (unlikely(ni
.blk_addr
== NULL_ADDR
)) {
1210 dec_page_count(sbi
, F2FS_DIRTY_NODES
);
1215 if (wbc
->for_reclaim
)
1218 mutex_lock(&sbi
->node_write
);
1219 set_page_writeback(page
);
1220 write_node_page(sbi
, page
, nid
, ni
.blk_addr
, &new_addr
);
1221 set_node_addr(sbi
, &ni
, new_addr
);
1222 dec_page_count(sbi
, F2FS_DIRTY_NODES
);
1223 mutex_unlock(&sbi
->node_write
);
1228 dec_page_count(sbi
, F2FS_DIRTY_NODES
);
1229 wbc
->pages_skipped
++;
1230 set_page_dirty(page
);
1231 return AOP_WRITEPAGE_ACTIVATE
;
1235 * It is very important to gather dirty pages and write at once, so that we can
1236 * submit a big bio without interfering other data writes.
1237 * Be default, 512 pages (2MB) * 3 node types, is more reasonable.
1239 #define COLLECT_DIRTY_NODES 1536
1240 static int f2fs_write_node_pages(struct address_space
*mapping
,
1241 struct writeback_control
*wbc
)
1243 struct f2fs_sb_info
*sbi
= F2FS_SB(mapping
->host
->i_sb
);
1244 long nr_to_write
= wbc
->nr_to_write
;
1246 /* balancing f2fs's metadata in background */
1247 f2fs_balance_fs_bg(sbi
);
1249 /* collect a number of dirty node pages and write together */
1250 if (get_pages(sbi
, F2FS_DIRTY_NODES
) < COLLECT_DIRTY_NODES
)
1253 /* if mounting is failed, skip writing node pages */
1254 wbc
->nr_to_write
= 3 * max_hw_blocks(sbi
);
1255 sync_node_pages(sbi
, 0, wbc
);
1256 wbc
->nr_to_write
= nr_to_write
- (3 * max_hw_blocks(sbi
) -
1261 static int f2fs_set_node_page_dirty(struct page
*page
)
1263 struct address_space
*mapping
= page
->mapping
;
1264 struct f2fs_sb_info
*sbi
= F2FS_SB(mapping
->host
->i_sb
);
1266 trace_f2fs_set_page_dirty(page
, NODE
);
1268 SetPageUptodate(page
);
1269 if (!PageDirty(page
)) {
1270 __set_page_dirty_nobuffers(page
);
1271 inc_page_count(sbi
, F2FS_DIRTY_NODES
);
1272 SetPagePrivate(page
);
1278 static void f2fs_invalidate_node_page(struct page
*page
, unsigned int offset
,
1279 unsigned int length
)
1281 struct inode
*inode
= page
->mapping
->host
;
1282 struct f2fs_sb_info
*sbi
= F2FS_SB(inode
->i_sb
);
1283 if (PageDirty(page
))
1284 dec_page_count(sbi
, F2FS_DIRTY_NODES
);
1285 ClearPagePrivate(page
);
1288 static int f2fs_release_node_page(struct page
*page
, gfp_t wait
)
1290 ClearPagePrivate(page
);
1295 * Structure of the f2fs node operations
1297 const struct address_space_operations f2fs_node_aops
= {
1298 .writepage
= f2fs_write_node_page
,
1299 .writepages
= f2fs_write_node_pages
,
1300 .set_page_dirty
= f2fs_set_node_page_dirty
,
1301 .invalidatepage
= f2fs_invalidate_node_page
,
1302 .releasepage
= f2fs_release_node_page
,
1305 static struct free_nid
*__lookup_free_nid_list(nid_t n
, struct list_head
*head
)
1307 struct list_head
*this;
1309 list_for_each(this, head
) {
1310 i
= list_entry(this, struct free_nid
, list
);
1317 static void __del_from_free_nid_list(struct free_nid
*i
)
1320 kmem_cache_free(free_nid_slab
, i
);
1323 static int add_free_nid(struct f2fs_nm_info
*nm_i
, nid_t nid
, bool build
)
1326 struct nat_entry
*ne
;
1327 bool allocated
= false;
1329 if (nm_i
->fcnt
> 2 * MAX_FREE_NIDS
)
1332 /* 0 nid should not be used */
1333 if (unlikely(nid
== 0))
1337 /* do not add allocated nids */
1338 read_lock(&nm_i
->nat_tree_lock
);
1339 ne
= __lookup_nat_cache(nm_i
, nid
);
1340 if (ne
&& nat_get_blkaddr(ne
) != NULL_ADDR
)
1342 read_unlock(&nm_i
->nat_tree_lock
);
1347 i
= f2fs_kmem_cache_alloc(free_nid_slab
, GFP_NOFS
);
1351 spin_lock(&nm_i
->free_nid_list_lock
);
1352 if (__lookup_free_nid_list(nid
, &nm_i
->free_nid_list
)) {
1353 spin_unlock(&nm_i
->free_nid_list_lock
);
1354 kmem_cache_free(free_nid_slab
, i
);
1357 list_add_tail(&i
->list
, &nm_i
->free_nid_list
);
1359 spin_unlock(&nm_i
->free_nid_list_lock
);
1363 static void remove_free_nid(struct f2fs_nm_info
*nm_i
, nid_t nid
)
1366 spin_lock(&nm_i
->free_nid_list_lock
);
1367 i
= __lookup_free_nid_list(nid
, &nm_i
->free_nid_list
);
1368 if (i
&& i
->state
== NID_NEW
) {
1369 __del_from_free_nid_list(i
);
1372 spin_unlock(&nm_i
->free_nid_list_lock
);
1375 static void scan_nat_page(struct f2fs_nm_info
*nm_i
,
1376 struct page
*nat_page
, nid_t start_nid
)
1378 struct f2fs_nat_block
*nat_blk
= page_address(nat_page
);
1382 i
= start_nid
% NAT_ENTRY_PER_BLOCK
;
1384 for (; i
< NAT_ENTRY_PER_BLOCK
; i
++, start_nid
++) {
1386 if (unlikely(start_nid
>= nm_i
->max_nid
))
1389 blk_addr
= le32_to_cpu(nat_blk
->entries
[i
].block_addr
);
1390 f2fs_bug_on(blk_addr
== NEW_ADDR
);
1391 if (blk_addr
== NULL_ADDR
) {
1392 if (add_free_nid(nm_i
, start_nid
, true) < 0)
1398 static void build_free_nids(struct f2fs_sb_info
*sbi
)
1400 struct f2fs_nm_info
*nm_i
= NM_I(sbi
);
1401 struct curseg_info
*curseg
= CURSEG_I(sbi
, CURSEG_HOT_DATA
);
1402 struct f2fs_summary_block
*sum
= curseg
->sum_blk
;
1404 nid_t nid
= nm_i
->next_scan_nid
;
1406 /* Enough entries */
1407 if (nm_i
->fcnt
> NAT_ENTRY_PER_BLOCK
)
1410 /* readahead nat pages to be scanned */
1411 ra_nat_pages(sbi
, nid
);
1414 struct page
*page
= get_current_nat_page(sbi
, nid
);
1416 scan_nat_page(nm_i
, page
, nid
);
1417 f2fs_put_page(page
, 1);
1419 nid
+= (NAT_ENTRY_PER_BLOCK
- (nid
% NAT_ENTRY_PER_BLOCK
));
1420 if (unlikely(nid
>= nm_i
->max_nid
))
1423 if (i
++ == FREE_NID_PAGES
)
1427 /* go to the next free nat pages to find free nids abundantly */
1428 nm_i
->next_scan_nid
= nid
;
1430 /* find free nids from current sum_pages */
1431 mutex_lock(&curseg
->curseg_mutex
);
1432 for (i
= 0; i
< nats_in_cursum(sum
); i
++) {
1433 block_t addr
= le32_to_cpu(nat_in_journal(sum
, i
).block_addr
);
1434 nid
= le32_to_cpu(nid_in_journal(sum
, i
));
1435 if (addr
== NULL_ADDR
)
1436 add_free_nid(nm_i
, nid
, true);
1438 remove_free_nid(nm_i
, nid
);
1440 mutex_unlock(&curseg
->curseg_mutex
);
1444 * If this function returns success, caller can obtain a new nid
1445 * from second parameter of this function.
1446 * The returned nid could be used ino as well as nid when inode is created.
1448 bool alloc_nid(struct f2fs_sb_info
*sbi
, nid_t
*nid
)
1450 struct f2fs_nm_info
*nm_i
= NM_I(sbi
);
1451 struct free_nid
*i
= NULL
;
1452 struct list_head
*this;
1454 if (unlikely(sbi
->total_valid_node_count
+ 1 >= nm_i
->max_nid
))
1457 spin_lock(&nm_i
->free_nid_list_lock
);
1459 /* We should not use stale free nids created by build_free_nids */
1460 if (nm_i
->fcnt
&& !sbi
->on_build_free_nids
) {
1461 f2fs_bug_on(list_empty(&nm_i
->free_nid_list
));
1462 list_for_each(this, &nm_i
->free_nid_list
) {
1463 i
= list_entry(this, struct free_nid
, list
);
1464 if (i
->state
== NID_NEW
)
1468 f2fs_bug_on(i
->state
!= NID_NEW
);
1470 i
->state
= NID_ALLOC
;
1472 spin_unlock(&nm_i
->free_nid_list_lock
);
1475 spin_unlock(&nm_i
->free_nid_list_lock
);
1477 /* Let's scan nat pages and its caches to get free nids */
1478 mutex_lock(&nm_i
->build_lock
);
1479 sbi
->on_build_free_nids
= true;
1480 build_free_nids(sbi
);
1481 sbi
->on_build_free_nids
= false;
1482 mutex_unlock(&nm_i
->build_lock
);
1487 * alloc_nid() should be called prior to this function.
1489 void alloc_nid_done(struct f2fs_sb_info
*sbi
, nid_t nid
)
1491 struct f2fs_nm_info
*nm_i
= NM_I(sbi
);
1494 spin_lock(&nm_i
->free_nid_list_lock
);
1495 i
= __lookup_free_nid_list(nid
, &nm_i
->free_nid_list
);
1496 f2fs_bug_on(!i
|| i
->state
!= NID_ALLOC
);
1497 __del_from_free_nid_list(i
);
1498 spin_unlock(&nm_i
->free_nid_list_lock
);
1502 * alloc_nid() should be called prior to this function.
1504 void alloc_nid_failed(struct f2fs_sb_info
*sbi
, nid_t nid
)
1506 struct f2fs_nm_info
*nm_i
= NM_I(sbi
);
1512 spin_lock(&nm_i
->free_nid_list_lock
);
1513 i
= __lookup_free_nid_list(nid
, &nm_i
->free_nid_list
);
1514 f2fs_bug_on(!i
|| i
->state
!= NID_ALLOC
);
1515 if (nm_i
->fcnt
> 2 * MAX_FREE_NIDS
) {
1516 __del_from_free_nid_list(i
);
1521 spin_unlock(&nm_i
->free_nid_list_lock
);
1524 void recover_node_page(struct f2fs_sb_info
*sbi
, struct page
*page
,
1525 struct f2fs_summary
*sum
, struct node_info
*ni
,
1526 block_t new_blkaddr
)
1528 rewrite_node_page(sbi
, page
, sum
, ni
->blk_addr
, new_blkaddr
);
1529 set_node_addr(sbi
, ni
, new_blkaddr
);
1530 clear_node_page_dirty(page
);
1533 int recover_inode_page(struct f2fs_sb_info
*sbi
, struct page
*page
)
1535 struct address_space
*mapping
= sbi
->node_inode
->i_mapping
;
1536 struct f2fs_node
*src
, *dst
;
1537 nid_t ino
= ino_of_node(page
);
1538 struct node_info old_ni
, new_ni
;
1541 ipage
= grab_cache_page(mapping
, ino
);
1545 /* Should not use this inode from free nid list */
1546 remove_free_nid(NM_I(sbi
), ino
);
1548 get_node_info(sbi
, ino
, &old_ni
);
1549 SetPageUptodate(ipage
);
1550 fill_node_footer(ipage
, ino
, ino
, 0, true);
1552 src
= F2FS_NODE(page
);
1553 dst
= F2FS_NODE(ipage
);
1555 memcpy(dst
, src
, (unsigned long)&src
->i
.i_ext
- (unsigned long)&src
->i
);
1557 dst
->i
.i_blocks
= cpu_to_le64(1);
1558 dst
->i
.i_links
= cpu_to_le32(1);
1559 dst
->i
.i_xattr_nid
= 0;
1564 if (unlikely(!inc_valid_node_count(sbi
, NULL
)))
1566 set_node_addr(sbi
, &new_ni
, NEW_ADDR
);
1567 inc_valid_inode_count(sbi
);
1568 f2fs_put_page(ipage
, 1);
1573 * ra_sum_pages() merge contiguous pages into one bio and submit.
1574 * these pre-readed pages are linked in pages list.
1576 static int ra_sum_pages(struct f2fs_sb_info
*sbi
, struct list_head
*pages
,
1577 int start
, int nrpages
)
1580 int page_idx
= start
;
1581 struct f2fs_io_info fio
= {
1583 .rw
= READ_SYNC
| REQ_META
| REQ_PRIO
1586 for (; page_idx
< start
+ nrpages
; page_idx
++) {
1587 /* alloc temporal page for read node summary info*/
1588 page
= alloc_page(GFP_F2FS_ZERO
);
1591 list_for_each_entry_safe(page
, tmp
, pages
, lru
) {
1592 list_del(&page
->lru
);
1594 __free_pages(page
, 0);
1600 page
->index
= page_idx
;
1601 list_add_tail(&page
->lru
, pages
);
1604 list_for_each_entry(page
, pages
, lru
)
1605 f2fs_submit_page_mbio(sbi
, page
, page
->index
, &fio
);
1607 f2fs_submit_merged_bio(sbi
, META
, READ
);
1611 int restore_node_summary(struct f2fs_sb_info
*sbi
,
1612 unsigned int segno
, struct f2fs_summary_block
*sum
)
1614 struct f2fs_node
*rn
;
1615 struct f2fs_summary
*sum_entry
;
1616 struct page
*page
, *tmp
;
1618 int bio_blocks
= MAX_BIO_BLOCKS(max_hw_blocks(sbi
));
1619 int i
, last_offset
, nrpages
, err
= 0;
1620 LIST_HEAD(page_list
);
1622 /* scan the node segment */
1623 last_offset
= sbi
->blocks_per_seg
;
1624 addr
= START_BLOCK(sbi
, segno
);
1625 sum_entry
= &sum
->entries
[0];
1627 for (i
= 0; i
< last_offset
; i
+= nrpages
, addr
+= nrpages
) {
1628 nrpages
= min(last_offset
- i
, bio_blocks
);
1630 /* read ahead node pages */
1631 err
= ra_sum_pages(sbi
, &page_list
, addr
, nrpages
);
1635 list_for_each_entry_safe(page
, tmp
, &page_list
, lru
) {
1638 if (unlikely(!PageUptodate(page
))) {
1641 rn
= F2FS_NODE(page
);
1642 sum_entry
->nid
= rn
->footer
.nid
;
1643 sum_entry
->version
= 0;
1644 sum_entry
->ofs_in_node
= 0;
1648 list_del(&page
->lru
);
1650 __free_pages(page
, 0);
1656 static bool flush_nats_in_journal(struct f2fs_sb_info
*sbi
)
1658 struct f2fs_nm_info
*nm_i
= NM_I(sbi
);
1659 struct curseg_info
*curseg
= CURSEG_I(sbi
, CURSEG_HOT_DATA
);
1660 struct f2fs_summary_block
*sum
= curseg
->sum_blk
;
1663 mutex_lock(&curseg
->curseg_mutex
);
1665 if (nats_in_cursum(sum
) < NAT_JOURNAL_ENTRIES
) {
1666 mutex_unlock(&curseg
->curseg_mutex
);
1670 for (i
= 0; i
< nats_in_cursum(sum
); i
++) {
1671 struct nat_entry
*ne
;
1672 struct f2fs_nat_entry raw_ne
;
1673 nid_t nid
= le32_to_cpu(nid_in_journal(sum
, i
));
1675 raw_ne
= nat_in_journal(sum
, i
);
1677 write_lock(&nm_i
->nat_tree_lock
);
1678 ne
= __lookup_nat_cache(nm_i
, nid
);
1680 __set_nat_cache_dirty(nm_i
, ne
);
1681 write_unlock(&nm_i
->nat_tree_lock
);
1684 ne
= grab_nat_entry(nm_i
, nid
);
1686 write_unlock(&nm_i
->nat_tree_lock
);
1689 nat_set_blkaddr(ne
, le32_to_cpu(raw_ne
.block_addr
));
1690 nat_set_ino(ne
, le32_to_cpu(raw_ne
.ino
));
1691 nat_set_version(ne
, raw_ne
.version
);
1692 __set_nat_cache_dirty(nm_i
, ne
);
1693 write_unlock(&nm_i
->nat_tree_lock
);
1695 update_nats_in_cursum(sum
, -i
);
1696 mutex_unlock(&curseg
->curseg_mutex
);
1701 * This function is called during the checkpointing process.
1703 void flush_nat_entries(struct f2fs_sb_info
*sbi
)
1705 struct f2fs_nm_info
*nm_i
= NM_I(sbi
);
1706 struct curseg_info
*curseg
= CURSEG_I(sbi
, CURSEG_HOT_DATA
);
1707 struct f2fs_summary_block
*sum
= curseg
->sum_blk
;
1708 struct list_head
*cur
, *n
;
1709 struct page
*page
= NULL
;
1710 struct f2fs_nat_block
*nat_blk
= NULL
;
1711 nid_t start_nid
= 0, end_nid
= 0;
1714 flushed
= flush_nats_in_journal(sbi
);
1717 mutex_lock(&curseg
->curseg_mutex
);
1719 /* 1) flush dirty nat caches */
1720 list_for_each_safe(cur
, n
, &nm_i
->dirty_nat_entries
) {
1721 struct nat_entry
*ne
;
1723 struct f2fs_nat_entry raw_ne
;
1725 block_t new_blkaddr
;
1727 ne
= list_entry(cur
, struct nat_entry
, list
);
1728 nid
= nat_get_nid(ne
);
1730 if (nat_get_blkaddr(ne
) == NEW_ADDR
)
1735 /* if there is room for nat enries in curseg->sumpage */
1736 offset
= lookup_journal_in_cursum(sum
, NAT_JOURNAL
, nid
, 1);
1738 raw_ne
= nat_in_journal(sum
, offset
);
1742 if (!page
|| (start_nid
> nid
|| nid
> end_nid
)) {
1744 f2fs_put_page(page
, 1);
1747 start_nid
= START_NID(nid
);
1748 end_nid
= start_nid
+ NAT_ENTRY_PER_BLOCK
- 1;
1751 * get nat block with dirty flag, increased reference
1752 * count, mapped and lock
1754 page
= get_next_nat_page(sbi
, start_nid
);
1755 nat_blk
= page_address(page
);
1758 f2fs_bug_on(!nat_blk
);
1759 raw_ne
= nat_blk
->entries
[nid
- start_nid
];
1761 new_blkaddr
= nat_get_blkaddr(ne
);
1763 raw_ne
.ino
= cpu_to_le32(nat_get_ino(ne
));
1764 raw_ne
.block_addr
= cpu_to_le32(new_blkaddr
);
1765 raw_ne
.version
= nat_get_version(ne
);
1768 nat_blk
->entries
[nid
- start_nid
] = raw_ne
;
1770 nat_in_journal(sum
, offset
) = raw_ne
;
1771 nid_in_journal(sum
, offset
) = cpu_to_le32(nid
);
1774 if (nat_get_blkaddr(ne
) == NULL_ADDR
&&
1775 add_free_nid(NM_I(sbi
), nid
, false) <= 0) {
1776 write_lock(&nm_i
->nat_tree_lock
);
1777 __del_from_nat_cache(nm_i
, ne
);
1778 write_unlock(&nm_i
->nat_tree_lock
);
1780 write_lock(&nm_i
->nat_tree_lock
);
1781 __clear_nat_cache_dirty(nm_i
, ne
);
1782 ne
->checkpointed
= true;
1783 write_unlock(&nm_i
->nat_tree_lock
);
1787 mutex_unlock(&curseg
->curseg_mutex
);
1788 f2fs_put_page(page
, 1);
1790 /* 2) shrink nat caches if necessary */
1791 try_to_free_nats(sbi
, nm_i
->nat_cnt
- NM_WOUT_THRESHOLD
);
1794 static int init_node_manager(struct f2fs_sb_info
*sbi
)
1796 struct f2fs_super_block
*sb_raw
= F2FS_RAW_SUPER(sbi
);
1797 struct f2fs_nm_info
*nm_i
= NM_I(sbi
);
1798 unsigned char *version_bitmap
;
1799 unsigned int nat_segs
, nat_blocks
;
1801 nm_i
->nat_blkaddr
= le32_to_cpu(sb_raw
->nat_blkaddr
);
1803 /* segment_count_nat includes pair segment so divide to 2. */
1804 nat_segs
= le32_to_cpu(sb_raw
->segment_count_nat
) >> 1;
1805 nat_blocks
= nat_segs
<< le32_to_cpu(sb_raw
->log_blocks_per_seg
);
1806 nm_i
->max_nid
= NAT_ENTRY_PER_BLOCK
* nat_blocks
;
1810 INIT_LIST_HEAD(&nm_i
->free_nid_list
);
1811 INIT_RADIX_TREE(&nm_i
->nat_root
, GFP_ATOMIC
);
1812 INIT_LIST_HEAD(&nm_i
->nat_entries
);
1813 INIT_LIST_HEAD(&nm_i
->dirty_nat_entries
);
1815 mutex_init(&nm_i
->build_lock
);
1816 spin_lock_init(&nm_i
->free_nid_list_lock
);
1817 rwlock_init(&nm_i
->nat_tree_lock
);
1819 nm_i
->next_scan_nid
= le32_to_cpu(sbi
->ckpt
->next_free_nid
);
1820 nm_i
->bitmap_size
= __bitmap_size(sbi
, NAT_BITMAP
);
1821 version_bitmap
= __bitmap_ptr(sbi
, NAT_BITMAP
);
1822 if (!version_bitmap
)
1825 nm_i
->nat_bitmap
= kmemdup(version_bitmap
, nm_i
->bitmap_size
,
1827 if (!nm_i
->nat_bitmap
)
1832 int build_node_manager(struct f2fs_sb_info
*sbi
)
1836 sbi
->nm_info
= kzalloc(sizeof(struct f2fs_nm_info
), GFP_KERNEL
);
1840 err
= init_node_manager(sbi
);
1844 build_free_nids(sbi
);
1848 void destroy_node_manager(struct f2fs_sb_info
*sbi
)
1850 struct f2fs_nm_info
*nm_i
= NM_I(sbi
);
1851 struct free_nid
*i
, *next_i
;
1852 struct nat_entry
*natvec
[NATVEC_SIZE
];
1859 /* destroy free nid list */
1860 spin_lock(&nm_i
->free_nid_list_lock
);
1861 list_for_each_entry_safe(i
, next_i
, &nm_i
->free_nid_list
, list
) {
1862 f2fs_bug_on(i
->state
== NID_ALLOC
);
1863 __del_from_free_nid_list(i
);
1866 f2fs_bug_on(nm_i
->fcnt
);
1867 spin_unlock(&nm_i
->free_nid_list_lock
);
1869 /* destroy nat cache */
1870 write_lock(&nm_i
->nat_tree_lock
);
1871 while ((found
= __gang_lookup_nat_cache(nm_i
,
1872 nid
, NATVEC_SIZE
, natvec
))) {
1874 for (idx
= 0; idx
< found
; idx
++) {
1875 struct nat_entry
*e
= natvec
[idx
];
1876 nid
= nat_get_nid(e
) + 1;
1877 __del_from_nat_cache(nm_i
, e
);
1880 f2fs_bug_on(nm_i
->nat_cnt
);
1881 write_unlock(&nm_i
->nat_tree_lock
);
1883 kfree(nm_i
->nat_bitmap
);
1884 sbi
->nm_info
= NULL
;
1888 int __init
create_node_manager_caches(void)
1890 nat_entry_slab
= f2fs_kmem_cache_create("nat_entry",
1891 sizeof(struct nat_entry
), NULL
);
1892 if (!nat_entry_slab
)
1895 free_nid_slab
= f2fs_kmem_cache_create("free_nid",
1896 sizeof(struct free_nid
), NULL
);
1897 if (!free_nid_slab
) {
1898 kmem_cache_destroy(nat_entry_slab
);
1904 void destroy_node_manager_caches(void)
1906 kmem_cache_destroy(free_nid_slab
);
1907 kmem_cache_destroy(nat_entry_slab
);