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 #define on_build_free_nids(nmi) mutex_is_locked(&nm_i->build_lock)
26 static struct kmem_cache
*nat_entry_slab
;
27 static struct kmem_cache
*free_nid_slab
;
28 static struct kmem_cache
*nat_entry_set_slab
;
30 bool available_free_memory(struct f2fs_sb_info
*sbi
, int type
)
32 struct f2fs_nm_info
*nm_i
= NM_I(sbi
);
34 unsigned long mem_size
= 0;
38 /* give 25%, 25%, 50% memory for each components respectively */
39 if (type
== FREE_NIDS
) {
40 mem_size
= (nm_i
->fcnt
* sizeof(struct free_nid
)) >> 12;
41 res
= mem_size
< ((val
.totalram
* nm_i
->ram_thresh
/ 100) >> 2);
42 } else if (type
== NAT_ENTRIES
) {
43 mem_size
= (nm_i
->nat_cnt
* sizeof(struct nat_entry
)) >> 12;
44 res
= mem_size
< ((val
.totalram
* nm_i
->ram_thresh
/ 100) >> 2);
45 } else if (type
== DIRTY_DENTS
) {
46 if (sbi
->sb
->s_bdi
->dirty_exceeded
)
48 mem_size
= get_pages(sbi
, F2FS_DIRTY_DENTS
);
49 res
= mem_size
< ((val
.totalram
* nm_i
->ram_thresh
/ 100) >> 1);
54 static void clear_node_page_dirty(struct page
*page
)
56 struct address_space
*mapping
= page
->mapping
;
57 unsigned int long flags
;
59 if (PageDirty(page
)) {
60 spin_lock_irqsave(&mapping
->tree_lock
, flags
);
61 radix_tree_tag_clear(&mapping
->page_tree
,
64 spin_unlock_irqrestore(&mapping
->tree_lock
, flags
);
66 clear_page_dirty_for_io(page
);
67 dec_page_count(F2FS_M_SB(mapping
), F2FS_DIRTY_NODES
);
69 ClearPageUptodate(page
);
72 static struct page
*get_current_nat_page(struct f2fs_sb_info
*sbi
, nid_t nid
)
74 pgoff_t index
= current_nat_addr(sbi
, nid
);
75 return get_meta_page(sbi
, index
);
78 static struct page
*get_next_nat_page(struct f2fs_sb_info
*sbi
, nid_t nid
)
80 struct page
*src_page
;
81 struct page
*dst_page
;
86 struct f2fs_nm_info
*nm_i
= NM_I(sbi
);
88 src_off
= current_nat_addr(sbi
, nid
);
89 dst_off
= next_nat_addr(sbi
, src_off
);
91 /* get current nat block page with lock */
92 src_page
= get_meta_page(sbi
, src_off
);
93 dst_page
= grab_meta_page(sbi
, dst_off
);
94 f2fs_bug_on(sbi
, PageDirty(src_page
));
96 src_addr
= page_address(src_page
);
97 dst_addr
= page_address(dst_page
);
98 memcpy(dst_addr
, src_addr
, PAGE_CACHE_SIZE
);
99 set_page_dirty(dst_page
);
100 f2fs_put_page(src_page
, 1);
102 set_to_next_nat(nm_i
, nid
);
107 static struct nat_entry
*__lookup_nat_cache(struct f2fs_nm_info
*nm_i
, nid_t n
)
109 return radix_tree_lookup(&nm_i
->nat_root
, n
);
112 static unsigned int __gang_lookup_nat_cache(struct f2fs_nm_info
*nm_i
,
113 nid_t start
, unsigned int nr
, struct nat_entry
**ep
)
115 return radix_tree_gang_lookup(&nm_i
->nat_root
, (void **)ep
, start
, nr
);
118 static void __del_from_nat_cache(struct f2fs_nm_info
*nm_i
, struct nat_entry
*e
)
121 radix_tree_delete(&nm_i
->nat_root
, nat_get_nid(e
));
123 kmem_cache_free(nat_entry_slab
, e
);
126 int is_checkpointed_node(struct f2fs_sb_info
*sbi
, nid_t nid
)
128 struct f2fs_nm_info
*nm_i
= NM_I(sbi
);
132 read_lock(&nm_i
->nat_tree_lock
);
133 e
= __lookup_nat_cache(nm_i
, nid
);
134 if (e
&& !e
->checkpointed
)
136 read_unlock(&nm_i
->nat_tree_lock
);
140 bool fsync_mark_done(struct f2fs_sb_info
*sbi
, nid_t nid
)
142 struct f2fs_nm_info
*nm_i
= NM_I(sbi
);
144 bool fsync_done
= false;
146 read_lock(&nm_i
->nat_tree_lock
);
147 e
= __lookup_nat_cache(nm_i
, nid
);
149 fsync_done
= e
->fsync_done
;
150 read_unlock(&nm_i
->nat_tree_lock
);
154 void fsync_mark_clear(struct f2fs_sb_info
*sbi
, nid_t nid
)
156 struct f2fs_nm_info
*nm_i
= NM_I(sbi
);
159 write_lock(&nm_i
->nat_tree_lock
);
160 e
= __lookup_nat_cache(nm_i
, nid
);
162 e
->fsync_done
= false;
163 write_unlock(&nm_i
->nat_tree_lock
);
166 static struct nat_entry
*grab_nat_entry(struct f2fs_nm_info
*nm_i
, nid_t nid
)
168 struct nat_entry
*new;
170 new = kmem_cache_alloc(nat_entry_slab
, GFP_ATOMIC
);
173 if (radix_tree_insert(&nm_i
->nat_root
, nid
, new)) {
174 kmem_cache_free(nat_entry_slab
, new);
177 memset(new, 0, sizeof(struct nat_entry
));
178 nat_set_nid(new, nid
);
179 new->checkpointed
= true;
180 list_add_tail(&new->list
, &nm_i
->nat_entries
);
185 static void cache_nat_entry(struct f2fs_nm_info
*nm_i
, nid_t nid
,
186 struct f2fs_nat_entry
*ne
)
190 write_lock(&nm_i
->nat_tree_lock
);
191 e
= __lookup_nat_cache(nm_i
, nid
);
193 e
= grab_nat_entry(nm_i
, nid
);
195 write_unlock(&nm_i
->nat_tree_lock
);
198 node_info_from_raw_nat(&e
->ni
, ne
);
200 write_unlock(&nm_i
->nat_tree_lock
);
203 static void set_node_addr(struct f2fs_sb_info
*sbi
, struct node_info
*ni
,
204 block_t new_blkaddr
, bool fsync_done
)
206 struct f2fs_nm_info
*nm_i
= NM_I(sbi
);
209 write_lock(&nm_i
->nat_tree_lock
);
210 e
= __lookup_nat_cache(nm_i
, ni
->nid
);
212 e
= grab_nat_entry(nm_i
, ni
->nid
);
214 write_unlock(&nm_i
->nat_tree_lock
);
218 f2fs_bug_on(sbi
, ni
->blk_addr
== NEW_ADDR
);
219 } else if (new_blkaddr
== NEW_ADDR
) {
221 * when nid is reallocated,
222 * previous nat entry can be remained in nat cache.
223 * So, reinitialize it with new information.
226 f2fs_bug_on(sbi
, ni
->blk_addr
!= NULL_ADDR
);
230 f2fs_bug_on(sbi
, nat_get_blkaddr(e
) != ni
->blk_addr
);
231 f2fs_bug_on(sbi
, nat_get_blkaddr(e
) == NULL_ADDR
&&
232 new_blkaddr
== NULL_ADDR
);
233 f2fs_bug_on(sbi
, nat_get_blkaddr(e
) == NEW_ADDR
&&
234 new_blkaddr
== NEW_ADDR
);
235 f2fs_bug_on(sbi
, nat_get_blkaddr(e
) != NEW_ADDR
&&
236 nat_get_blkaddr(e
) != NULL_ADDR
&&
237 new_blkaddr
== NEW_ADDR
);
239 /* increment version no as node is removed */
240 if (nat_get_blkaddr(e
) != NEW_ADDR
&& new_blkaddr
== NULL_ADDR
) {
241 unsigned char version
= nat_get_version(e
);
242 nat_set_version(e
, inc_node_version(version
));
246 nat_set_blkaddr(e
, new_blkaddr
);
247 __set_nat_cache_dirty(nm_i
, e
);
249 /* update fsync_mark if its inode nat entry is still alive */
250 e
= __lookup_nat_cache(nm_i
, ni
->ino
);
252 e
->fsync_done
= fsync_done
;
253 write_unlock(&nm_i
->nat_tree_lock
);
256 int try_to_free_nats(struct f2fs_sb_info
*sbi
, int nr_shrink
)
258 struct f2fs_nm_info
*nm_i
= NM_I(sbi
);
260 if (available_free_memory(sbi
, NAT_ENTRIES
))
263 write_lock(&nm_i
->nat_tree_lock
);
264 while (nr_shrink
&& !list_empty(&nm_i
->nat_entries
)) {
265 struct nat_entry
*ne
;
266 ne
= list_first_entry(&nm_i
->nat_entries
,
267 struct nat_entry
, list
);
268 __del_from_nat_cache(nm_i
, ne
);
271 write_unlock(&nm_i
->nat_tree_lock
);
276 * This function always returns success
278 void get_node_info(struct f2fs_sb_info
*sbi
, nid_t nid
, struct node_info
*ni
)
280 struct f2fs_nm_info
*nm_i
= NM_I(sbi
);
281 struct curseg_info
*curseg
= CURSEG_I(sbi
, CURSEG_HOT_DATA
);
282 struct f2fs_summary_block
*sum
= curseg
->sum_blk
;
283 nid_t start_nid
= START_NID(nid
);
284 struct f2fs_nat_block
*nat_blk
;
285 struct page
*page
= NULL
;
286 struct f2fs_nat_entry ne
;
290 memset(&ne
, 0, sizeof(struct f2fs_nat_entry
));
293 /* Check nat cache */
294 read_lock(&nm_i
->nat_tree_lock
);
295 e
= __lookup_nat_cache(nm_i
, nid
);
297 ni
->ino
= nat_get_ino(e
);
298 ni
->blk_addr
= nat_get_blkaddr(e
);
299 ni
->version
= nat_get_version(e
);
301 read_unlock(&nm_i
->nat_tree_lock
);
305 /* Check current segment summary */
306 mutex_lock(&curseg
->curseg_mutex
);
307 i
= lookup_journal_in_cursum(sum
, NAT_JOURNAL
, nid
, 0);
309 ne
= nat_in_journal(sum
, i
);
310 node_info_from_raw_nat(ni
, &ne
);
312 mutex_unlock(&curseg
->curseg_mutex
);
316 /* Fill node_info from nat page */
317 page
= get_current_nat_page(sbi
, start_nid
);
318 nat_blk
= (struct f2fs_nat_block
*)page_address(page
);
319 ne
= nat_blk
->entries
[nid
- start_nid
];
320 node_info_from_raw_nat(ni
, &ne
);
321 f2fs_put_page(page
, 1);
323 /* cache nat entry */
324 cache_nat_entry(NM_I(sbi
), nid
, &ne
);
328 * The maximum depth is four.
329 * Offset[0] will have raw inode offset.
331 static int get_node_path(struct f2fs_inode_info
*fi
, long block
,
332 int offset
[4], unsigned int noffset
[4])
334 const long direct_index
= ADDRS_PER_INODE(fi
);
335 const long direct_blks
= ADDRS_PER_BLOCK
;
336 const long dptrs_per_blk
= NIDS_PER_BLOCK
;
337 const long indirect_blks
= ADDRS_PER_BLOCK
* NIDS_PER_BLOCK
;
338 const long dindirect_blks
= indirect_blks
* NIDS_PER_BLOCK
;
344 if (block
< direct_index
) {
348 block
-= direct_index
;
349 if (block
< direct_blks
) {
350 offset
[n
++] = NODE_DIR1_BLOCK
;
356 block
-= direct_blks
;
357 if (block
< direct_blks
) {
358 offset
[n
++] = NODE_DIR2_BLOCK
;
364 block
-= direct_blks
;
365 if (block
< indirect_blks
) {
366 offset
[n
++] = NODE_IND1_BLOCK
;
368 offset
[n
++] = block
/ direct_blks
;
369 noffset
[n
] = 4 + offset
[n
- 1];
370 offset
[n
] = block
% direct_blks
;
374 block
-= indirect_blks
;
375 if (block
< indirect_blks
) {
376 offset
[n
++] = NODE_IND2_BLOCK
;
377 noffset
[n
] = 4 + dptrs_per_blk
;
378 offset
[n
++] = block
/ direct_blks
;
379 noffset
[n
] = 5 + dptrs_per_blk
+ offset
[n
- 1];
380 offset
[n
] = block
% direct_blks
;
384 block
-= indirect_blks
;
385 if (block
< dindirect_blks
) {
386 offset
[n
++] = NODE_DIND_BLOCK
;
387 noffset
[n
] = 5 + (dptrs_per_blk
* 2);
388 offset
[n
++] = block
/ indirect_blks
;
389 noffset
[n
] = 6 + (dptrs_per_blk
* 2) +
390 offset
[n
- 1] * (dptrs_per_blk
+ 1);
391 offset
[n
++] = (block
/ direct_blks
) % dptrs_per_blk
;
392 noffset
[n
] = 7 + (dptrs_per_blk
* 2) +
393 offset
[n
- 2] * (dptrs_per_blk
+ 1) +
395 offset
[n
] = block
% direct_blks
;
406 * Caller should call f2fs_put_dnode(dn).
407 * Also, it should grab and release a rwsem by calling f2fs_lock_op() and
408 * f2fs_unlock_op() only if ro is not set RDONLY_NODE.
409 * In the case of RDONLY_NODE, we don't need to care about mutex.
411 int get_dnode_of_data(struct dnode_of_data
*dn
, pgoff_t index
, int mode
)
413 struct f2fs_sb_info
*sbi
= F2FS_I_SB(dn
->inode
);
414 struct page
*npage
[4];
417 unsigned int noffset
[4];
422 level
= get_node_path(F2FS_I(dn
->inode
), index
, offset
, noffset
);
424 nids
[0] = dn
->inode
->i_ino
;
425 npage
[0] = dn
->inode_page
;
428 npage
[0] = get_node_page(sbi
, nids
[0]);
429 if (IS_ERR(npage
[0]))
430 return PTR_ERR(npage
[0]);
434 nids
[1] = get_nid(parent
, offset
[0], true);
435 dn
->inode_page
= npage
[0];
436 dn
->inode_page_locked
= true;
438 /* get indirect or direct nodes */
439 for (i
= 1; i
<= level
; i
++) {
442 if (!nids
[i
] && mode
== ALLOC_NODE
) {
444 if (!alloc_nid(sbi
, &(nids
[i
]))) {
450 npage
[i
] = new_node_page(dn
, noffset
[i
], NULL
);
451 if (IS_ERR(npage
[i
])) {
452 alloc_nid_failed(sbi
, nids
[i
]);
453 err
= PTR_ERR(npage
[i
]);
457 set_nid(parent
, offset
[i
- 1], nids
[i
], i
== 1);
458 alloc_nid_done(sbi
, nids
[i
]);
460 } else if (mode
== LOOKUP_NODE_RA
&& i
== level
&& level
> 1) {
461 npage
[i
] = get_node_page_ra(parent
, offset
[i
- 1]);
462 if (IS_ERR(npage
[i
])) {
463 err
= PTR_ERR(npage
[i
]);
469 dn
->inode_page_locked
= false;
472 f2fs_put_page(parent
, 1);
476 npage
[i
] = get_node_page(sbi
, nids
[i
]);
477 if (IS_ERR(npage
[i
])) {
478 err
= PTR_ERR(npage
[i
]);
479 f2fs_put_page(npage
[0], 0);
485 nids
[i
+ 1] = get_nid(parent
, offset
[i
], false);
488 dn
->nid
= nids
[level
];
489 dn
->ofs_in_node
= offset
[level
];
490 dn
->node_page
= npage
[level
];
491 dn
->data_blkaddr
= datablock_addr(dn
->node_page
, dn
->ofs_in_node
);
495 f2fs_put_page(parent
, 1);
497 f2fs_put_page(npage
[0], 0);
499 dn
->inode_page
= NULL
;
500 dn
->node_page
= NULL
;
504 static void truncate_node(struct dnode_of_data
*dn
)
506 struct f2fs_sb_info
*sbi
= F2FS_I_SB(dn
->inode
);
509 get_node_info(sbi
, dn
->nid
, &ni
);
510 if (dn
->inode
->i_blocks
== 0) {
511 f2fs_bug_on(sbi
, ni
.blk_addr
!= NULL_ADDR
);
514 f2fs_bug_on(sbi
, ni
.blk_addr
== NULL_ADDR
);
516 /* Deallocate node address */
517 invalidate_blocks(sbi
, ni
.blk_addr
);
518 dec_valid_node_count(sbi
, dn
->inode
);
519 set_node_addr(sbi
, &ni
, NULL_ADDR
, false);
521 if (dn
->nid
== dn
->inode
->i_ino
) {
522 remove_orphan_inode(sbi
, dn
->nid
);
523 dec_valid_inode_count(sbi
);
528 clear_node_page_dirty(dn
->node_page
);
529 F2FS_SET_SB_DIRT(sbi
);
531 f2fs_put_page(dn
->node_page
, 1);
533 invalidate_mapping_pages(NODE_MAPPING(sbi
),
534 dn
->node_page
->index
, dn
->node_page
->index
);
536 dn
->node_page
= NULL
;
537 trace_f2fs_truncate_node(dn
->inode
, dn
->nid
, ni
.blk_addr
);
540 static int truncate_dnode(struct dnode_of_data
*dn
)
547 /* get direct node */
548 page
= get_node_page(F2FS_I_SB(dn
->inode
), dn
->nid
);
549 if (IS_ERR(page
) && PTR_ERR(page
) == -ENOENT
)
551 else if (IS_ERR(page
))
552 return PTR_ERR(page
);
554 /* Make dnode_of_data for parameter */
555 dn
->node_page
= page
;
557 truncate_data_blocks(dn
);
562 static int truncate_nodes(struct dnode_of_data
*dn
, unsigned int nofs
,
565 struct dnode_of_data rdn
= *dn
;
567 struct f2fs_node
*rn
;
569 unsigned int child_nofs
;
574 return NIDS_PER_BLOCK
+ 1;
576 trace_f2fs_truncate_nodes_enter(dn
->inode
, dn
->nid
, dn
->data_blkaddr
);
578 page
= get_node_page(F2FS_I_SB(dn
->inode
), dn
->nid
);
580 trace_f2fs_truncate_nodes_exit(dn
->inode
, PTR_ERR(page
));
581 return PTR_ERR(page
);
584 rn
= F2FS_NODE(page
);
586 for (i
= ofs
; i
< NIDS_PER_BLOCK
; i
++, freed
++) {
587 child_nid
= le32_to_cpu(rn
->in
.nid
[i
]);
591 ret
= truncate_dnode(&rdn
);
594 set_nid(page
, i
, 0, false);
597 child_nofs
= nofs
+ ofs
* (NIDS_PER_BLOCK
+ 1) + 1;
598 for (i
= ofs
; i
< NIDS_PER_BLOCK
; i
++) {
599 child_nid
= le32_to_cpu(rn
->in
.nid
[i
]);
600 if (child_nid
== 0) {
601 child_nofs
+= NIDS_PER_BLOCK
+ 1;
605 ret
= truncate_nodes(&rdn
, child_nofs
, 0, depth
- 1);
606 if (ret
== (NIDS_PER_BLOCK
+ 1)) {
607 set_nid(page
, i
, 0, false);
609 } else if (ret
< 0 && ret
!= -ENOENT
) {
617 /* remove current indirect node */
618 dn
->node_page
= page
;
622 f2fs_put_page(page
, 1);
624 trace_f2fs_truncate_nodes_exit(dn
->inode
, freed
);
628 f2fs_put_page(page
, 1);
629 trace_f2fs_truncate_nodes_exit(dn
->inode
, ret
);
633 static int truncate_partial_nodes(struct dnode_of_data
*dn
,
634 struct f2fs_inode
*ri
, int *offset
, int depth
)
636 struct page
*pages
[2];
643 nid
[0] = le32_to_cpu(ri
->i_nid
[offset
[0] - NODE_DIR1_BLOCK
]);
647 /* get indirect nodes in the path */
648 for (i
= 0; i
< idx
+ 1; i
++) {
649 /* reference count'll be increased */
650 pages
[i
] = get_node_page(F2FS_I_SB(dn
->inode
), nid
[i
]);
651 if (IS_ERR(pages
[i
])) {
652 err
= PTR_ERR(pages
[i
]);
656 nid
[i
+ 1] = get_nid(pages
[i
], offset
[i
+ 1], false);
659 /* free direct nodes linked to a partial indirect node */
660 for (i
= offset
[idx
+ 1]; i
< NIDS_PER_BLOCK
; i
++) {
661 child_nid
= get_nid(pages
[idx
], i
, false);
665 err
= truncate_dnode(dn
);
668 set_nid(pages
[idx
], i
, 0, false);
671 if (offset
[idx
+ 1] == 0) {
672 dn
->node_page
= pages
[idx
];
676 f2fs_put_page(pages
[idx
], 1);
682 for (i
= idx
; i
>= 0; i
--)
683 f2fs_put_page(pages
[i
], 1);
685 trace_f2fs_truncate_partial_nodes(dn
->inode
, nid
, depth
, err
);
691 * All the block addresses of data and nodes should be nullified.
693 int truncate_inode_blocks(struct inode
*inode
, pgoff_t from
)
695 struct f2fs_sb_info
*sbi
= F2FS_I_SB(inode
);
696 int err
= 0, cont
= 1;
697 int level
, offset
[4], noffset
[4];
698 unsigned int nofs
= 0;
699 struct f2fs_inode
*ri
;
700 struct dnode_of_data dn
;
703 trace_f2fs_truncate_inode_blocks_enter(inode
, from
);
705 level
= get_node_path(F2FS_I(inode
), from
, offset
, noffset
);
707 page
= get_node_page(sbi
, inode
->i_ino
);
709 trace_f2fs_truncate_inode_blocks_exit(inode
, PTR_ERR(page
));
710 return PTR_ERR(page
);
713 set_new_dnode(&dn
, inode
, page
, NULL
, 0);
716 ri
= F2FS_INODE(page
);
724 if (!offset
[level
- 1])
726 err
= truncate_partial_nodes(&dn
, ri
, offset
, level
);
727 if (err
< 0 && err
!= -ENOENT
)
729 nofs
+= 1 + NIDS_PER_BLOCK
;
732 nofs
= 5 + 2 * NIDS_PER_BLOCK
;
733 if (!offset
[level
- 1])
735 err
= truncate_partial_nodes(&dn
, ri
, offset
, level
);
736 if (err
< 0 && err
!= -ENOENT
)
745 dn
.nid
= le32_to_cpu(ri
->i_nid
[offset
[0] - NODE_DIR1_BLOCK
]);
747 case NODE_DIR1_BLOCK
:
748 case NODE_DIR2_BLOCK
:
749 err
= truncate_dnode(&dn
);
752 case NODE_IND1_BLOCK
:
753 case NODE_IND2_BLOCK
:
754 err
= truncate_nodes(&dn
, nofs
, offset
[1], 2);
757 case NODE_DIND_BLOCK
:
758 err
= truncate_nodes(&dn
, nofs
, offset
[1], 3);
765 if (err
< 0 && err
!= -ENOENT
)
767 if (offset
[1] == 0 &&
768 ri
->i_nid
[offset
[0] - NODE_DIR1_BLOCK
]) {
770 if (unlikely(page
->mapping
!= NODE_MAPPING(sbi
))) {
771 f2fs_put_page(page
, 1);
774 f2fs_wait_on_page_writeback(page
, NODE
);
775 ri
->i_nid
[offset
[0] - NODE_DIR1_BLOCK
] = 0;
776 set_page_dirty(page
);
784 f2fs_put_page(page
, 0);
785 trace_f2fs_truncate_inode_blocks_exit(inode
, err
);
786 return err
> 0 ? 0 : err
;
789 int truncate_xattr_node(struct inode
*inode
, struct page
*page
)
791 struct f2fs_sb_info
*sbi
= F2FS_I_SB(inode
);
792 nid_t nid
= F2FS_I(inode
)->i_xattr_nid
;
793 struct dnode_of_data dn
;
799 npage
= get_node_page(sbi
, nid
);
801 return PTR_ERR(npage
);
803 F2FS_I(inode
)->i_xattr_nid
= 0;
805 /* need to do checkpoint during fsync */
806 F2FS_I(inode
)->xattr_ver
= cur_cp_version(F2FS_CKPT(sbi
));
808 set_new_dnode(&dn
, inode
, page
, npage
, nid
);
811 dn
.inode_page_locked
= true;
817 * Caller should grab and release a rwsem by calling f2fs_lock_op() and
820 void remove_inode_page(struct inode
*inode
)
822 struct dnode_of_data dn
;
824 set_new_dnode(&dn
, inode
, NULL
, NULL
, inode
->i_ino
);
825 if (get_dnode_of_data(&dn
, 0, LOOKUP_NODE
))
828 if (truncate_xattr_node(inode
, dn
.inode_page
)) {
833 /* remove potential inline_data blocks */
834 if (S_ISREG(inode
->i_mode
) || S_ISDIR(inode
->i_mode
) ||
835 S_ISLNK(inode
->i_mode
))
836 truncate_data_blocks_range(&dn
, 1);
838 /* 0 is possible, after f2fs_new_inode() has failed */
839 f2fs_bug_on(F2FS_I_SB(inode
),
840 inode
->i_blocks
!= 0 && inode
->i_blocks
!= 1);
842 /* will put inode & node pages */
846 struct page
*new_inode_page(struct inode
*inode
)
848 struct dnode_of_data dn
;
850 /* allocate inode page for new inode */
851 set_new_dnode(&dn
, inode
, NULL
, NULL
, inode
->i_ino
);
853 /* caller should f2fs_put_page(page, 1); */
854 return new_node_page(&dn
, 0, NULL
);
857 struct page
*new_node_page(struct dnode_of_data
*dn
,
858 unsigned int ofs
, struct page
*ipage
)
860 struct f2fs_sb_info
*sbi
= F2FS_I_SB(dn
->inode
);
861 struct node_info old_ni
, new_ni
;
865 if (unlikely(is_inode_flag_set(F2FS_I(dn
->inode
), FI_NO_ALLOC
)))
866 return ERR_PTR(-EPERM
);
868 page
= grab_cache_page(NODE_MAPPING(sbi
), dn
->nid
);
870 return ERR_PTR(-ENOMEM
);
872 if (unlikely(!inc_valid_node_count(sbi
, dn
->inode
))) {
877 get_node_info(sbi
, dn
->nid
, &old_ni
);
879 /* Reinitialize old_ni with new node page */
880 f2fs_bug_on(sbi
, old_ni
.blk_addr
!= NULL_ADDR
);
882 new_ni
.ino
= dn
->inode
->i_ino
;
883 set_node_addr(sbi
, &new_ni
, NEW_ADDR
, false);
885 f2fs_wait_on_page_writeback(page
, NODE
);
886 fill_node_footer(page
, dn
->nid
, dn
->inode
->i_ino
, ofs
, true);
887 set_cold_node(dn
->inode
, page
);
888 SetPageUptodate(page
);
889 set_page_dirty(page
);
891 if (f2fs_has_xattr_block(ofs
))
892 F2FS_I(dn
->inode
)->i_xattr_nid
= dn
->nid
;
894 dn
->node_page
= page
;
896 update_inode(dn
->inode
, ipage
);
900 inc_valid_inode_count(sbi
);
905 clear_node_page_dirty(page
);
906 f2fs_put_page(page
, 1);
911 * Caller should do after getting the following values.
912 * 0: f2fs_put_page(page, 0)
913 * LOCKED_PAGE: f2fs_put_page(page, 1)
916 static int read_node_page(struct page
*page
, int rw
)
918 struct f2fs_sb_info
*sbi
= F2FS_P_SB(page
);
921 get_node_info(sbi
, page
->index
, &ni
);
923 if (unlikely(ni
.blk_addr
== NULL_ADDR
)) {
924 f2fs_put_page(page
, 1);
928 if (PageUptodate(page
))
931 return f2fs_submit_page_bio(sbi
, page
, ni
.blk_addr
, rw
);
935 * Readahead a node page
937 void ra_node_page(struct f2fs_sb_info
*sbi
, nid_t nid
)
942 apage
= find_get_page(NODE_MAPPING(sbi
), nid
);
943 if (apage
&& PageUptodate(apage
)) {
944 f2fs_put_page(apage
, 0);
947 f2fs_put_page(apage
, 0);
949 apage
= grab_cache_page(NODE_MAPPING(sbi
), nid
);
953 err
= read_node_page(apage
, READA
);
955 f2fs_put_page(apage
, 0);
956 else if (err
== LOCKED_PAGE
)
957 f2fs_put_page(apage
, 1);
960 struct page
*get_node_page(struct f2fs_sb_info
*sbi
, pgoff_t nid
)
965 page
= grab_cache_page(NODE_MAPPING(sbi
), nid
);
967 return ERR_PTR(-ENOMEM
);
969 err
= read_node_page(page
, READ_SYNC
);
972 else if (err
== LOCKED_PAGE
)
976 if (unlikely(!PageUptodate(page
) || nid
!= nid_of_node(page
))) {
977 f2fs_put_page(page
, 1);
978 return ERR_PTR(-EIO
);
980 if (unlikely(page
->mapping
!= NODE_MAPPING(sbi
))) {
981 f2fs_put_page(page
, 1);
989 * Return a locked page for the desired node page.
990 * And, readahead MAX_RA_NODE number of node pages.
992 struct page
*get_node_page_ra(struct page
*parent
, int start
)
994 struct f2fs_sb_info
*sbi
= F2FS_P_SB(parent
);
995 struct blk_plug plug
;
1000 /* First, try getting the desired direct node. */
1001 nid
= get_nid(parent
, start
, false);
1003 return ERR_PTR(-ENOENT
);
1005 page
= grab_cache_page(NODE_MAPPING(sbi
), nid
);
1007 return ERR_PTR(-ENOMEM
);
1009 err
= read_node_page(page
, READ_SYNC
);
1011 return ERR_PTR(err
);
1012 else if (err
== LOCKED_PAGE
)
1015 blk_start_plug(&plug
);
1017 /* Then, try readahead for siblings of the desired node */
1018 end
= start
+ MAX_RA_NODE
;
1019 end
= min(end
, NIDS_PER_BLOCK
);
1020 for (i
= start
+ 1; i
< end
; i
++) {
1021 nid
= get_nid(parent
, i
, false);
1024 ra_node_page(sbi
, nid
);
1027 blk_finish_plug(&plug
);
1030 if (unlikely(page
->mapping
!= NODE_MAPPING(sbi
))) {
1031 f2fs_put_page(page
, 1);
1035 if (unlikely(!PageUptodate(page
))) {
1036 f2fs_put_page(page
, 1);
1037 return ERR_PTR(-EIO
);
1042 void sync_inode_page(struct dnode_of_data
*dn
)
1044 if (IS_INODE(dn
->node_page
) || dn
->inode_page
== dn
->node_page
) {
1045 update_inode(dn
->inode
, dn
->node_page
);
1046 } else if (dn
->inode_page
) {
1047 if (!dn
->inode_page_locked
)
1048 lock_page(dn
->inode_page
);
1049 update_inode(dn
->inode
, dn
->inode_page
);
1050 if (!dn
->inode_page_locked
)
1051 unlock_page(dn
->inode_page
);
1053 update_inode_page(dn
->inode
);
1057 int sync_node_pages(struct f2fs_sb_info
*sbi
, nid_t ino
,
1058 struct writeback_control
*wbc
)
1061 struct pagevec pvec
;
1062 int step
= ino
? 2 : 0;
1063 int nwritten
= 0, wrote
= 0;
1065 pagevec_init(&pvec
, 0);
1071 while (index
<= end
) {
1073 nr_pages
= pagevec_lookup_tag(&pvec
, NODE_MAPPING(sbi
), &index
,
1074 PAGECACHE_TAG_DIRTY
,
1075 min(end
- index
, (pgoff_t
)PAGEVEC_SIZE
-1) + 1);
1079 for (i
= 0; i
< nr_pages
; i
++) {
1080 struct page
*page
= pvec
.pages
[i
];
1083 * flushing sequence with step:
1088 if (step
== 0 && IS_DNODE(page
))
1090 if (step
== 1 && (!IS_DNODE(page
) ||
1091 is_cold_node(page
)))
1093 if (step
== 2 && (!IS_DNODE(page
) ||
1094 !is_cold_node(page
)))
1099 * we should not skip writing node pages.
1101 if (ino
&& ino_of_node(page
) == ino
)
1103 else if (!trylock_page(page
))
1106 if (unlikely(page
->mapping
!= NODE_MAPPING(sbi
))) {
1111 if (ino
&& ino_of_node(page
) != ino
)
1112 goto continue_unlock
;
1114 if (!PageDirty(page
)) {
1115 /* someone wrote it for us */
1116 goto continue_unlock
;
1119 if (!clear_page_dirty_for_io(page
))
1120 goto continue_unlock
;
1122 /* called by fsync() */
1123 if (ino
&& IS_DNODE(page
)) {
1124 int mark
= !is_checkpointed_node(sbi
, ino
);
1125 set_fsync_mark(page
, 1);
1127 set_dentry_mark(page
, mark
);
1130 set_fsync_mark(page
, 0);
1131 set_dentry_mark(page
, 0);
1134 if (NODE_MAPPING(sbi
)->a_ops
->writepage(page
, wbc
))
1139 if (--wbc
->nr_to_write
== 0)
1142 pagevec_release(&pvec
);
1145 if (wbc
->nr_to_write
== 0) {
1157 f2fs_submit_merged_bio(sbi
, NODE
, WRITE
);
1161 int wait_on_node_pages_writeback(struct f2fs_sb_info
*sbi
, nid_t ino
)
1163 pgoff_t index
= 0, end
= LONG_MAX
;
1164 struct pagevec pvec
;
1165 int ret2
= 0, ret
= 0;
1167 pagevec_init(&pvec
, 0);
1169 while (index
<= end
) {
1171 nr_pages
= pagevec_lookup_tag(&pvec
, NODE_MAPPING(sbi
), &index
,
1172 PAGECACHE_TAG_WRITEBACK
,
1173 min(end
- index
, (pgoff_t
)PAGEVEC_SIZE
-1) + 1);
1177 for (i
= 0; i
< nr_pages
; i
++) {
1178 struct page
*page
= pvec
.pages
[i
];
1180 /* until radix tree lookup accepts end_index */
1181 if (unlikely(page
->index
> end
))
1184 if (ino
&& ino_of_node(page
) == ino
) {
1185 f2fs_wait_on_page_writeback(page
, NODE
);
1186 if (TestClearPageError(page
))
1190 pagevec_release(&pvec
);
1194 if (unlikely(test_and_clear_bit(AS_ENOSPC
, &NODE_MAPPING(sbi
)->flags
)))
1196 if (unlikely(test_and_clear_bit(AS_EIO
, &NODE_MAPPING(sbi
)->flags
)))
1203 static int f2fs_write_node_page(struct page
*page
,
1204 struct writeback_control
*wbc
)
1206 struct f2fs_sb_info
*sbi
= F2FS_P_SB(page
);
1209 struct node_info ni
;
1210 struct f2fs_io_info fio
= {
1212 .rw
= (wbc
->sync_mode
== WB_SYNC_ALL
) ? WRITE_SYNC
: WRITE
,
1215 trace_f2fs_writepage(page
, NODE
);
1217 if (unlikely(sbi
->por_doing
))
1219 if (unlikely(f2fs_cp_error(sbi
)))
1222 f2fs_wait_on_page_writeback(page
, NODE
);
1224 /* get old block addr of this node page */
1225 nid
= nid_of_node(page
);
1226 f2fs_bug_on(sbi
, page
->index
!= nid
);
1228 get_node_info(sbi
, nid
, &ni
);
1230 /* This page is already truncated */
1231 if (unlikely(ni
.blk_addr
== NULL_ADDR
)) {
1232 dec_page_count(sbi
, F2FS_DIRTY_NODES
);
1237 if (wbc
->for_reclaim
)
1240 down_read(&sbi
->node_write
);
1241 set_page_writeback(page
);
1242 write_node_page(sbi
, page
, &fio
, nid
, ni
.blk_addr
, &new_addr
);
1243 set_node_addr(sbi
, &ni
, new_addr
, is_fsync_dnode(page
));
1244 dec_page_count(sbi
, F2FS_DIRTY_NODES
);
1245 up_read(&sbi
->node_write
);
1250 redirty_page_for_writepage(wbc
, page
);
1251 return AOP_WRITEPAGE_ACTIVATE
;
1254 static int f2fs_write_node_pages(struct address_space
*mapping
,
1255 struct writeback_control
*wbc
)
1257 struct f2fs_sb_info
*sbi
= F2FS_M_SB(mapping
);
1260 trace_f2fs_writepages(mapping
->host
, wbc
, NODE
);
1262 /* balancing f2fs's metadata in background */
1263 f2fs_balance_fs_bg(sbi
);
1265 /* collect a number of dirty node pages and write together */
1266 if (get_pages(sbi
, F2FS_DIRTY_NODES
) < nr_pages_to_skip(sbi
, NODE
))
1269 diff
= nr_pages_to_write(sbi
, NODE
, wbc
);
1270 wbc
->sync_mode
= WB_SYNC_NONE
;
1271 sync_node_pages(sbi
, 0, wbc
);
1272 wbc
->nr_to_write
= max((long)0, wbc
->nr_to_write
- diff
);
1276 wbc
->pages_skipped
+= get_pages(sbi
, F2FS_DIRTY_NODES
);
1280 static int f2fs_set_node_page_dirty(struct page
*page
)
1282 trace_f2fs_set_page_dirty(page
, NODE
);
1284 SetPageUptodate(page
);
1285 if (!PageDirty(page
)) {
1286 __set_page_dirty_nobuffers(page
);
1287 inc_page_count(F2FS_P_SB(page
), F2FS_DIRTY_NODES
);
1288 SetPagePrivate(page
);
1294 static void f2fs_invalidate_node_page(struct page
*page
, unsigned int offset
,
1295 unsigned int length
)
1297 struct inode
*inode
= page
->mapping
->host
;
1298 if (PageDirty(page
))
1299 dec_page_count(F2FS_I_SB(inode
), F2FS_DIRTY_NODES
);
1300 ClearPagePrivate(page
);
1303 static int f2fs_release_node_page(struct page
*page
, gfp_t wait
)
1305 ClearPagePrivate(page
);
1310 * Structure of the f2fs node operations
1312 const struct address_space_operations f2fs_node_aops
= {
1313 .writepage
= f2fs_write_node_page
,
1314 .writepages
= f2fs_write_node_pages
,
1315 .set_page_dirty
= f2fs_set_node_page_dirty
,
1316 .invalidatepage
= f2fs_invalidate_node_page
,
1317 .releasepage
= f2fs_release_node_page
,
1320 static struct free_nid
*__lookup_free_nid_list(struct f2fs_nm_info
*nm_i
,
1323 return radix_tree_lookup(&nm_i
->free_nid_root
, n
);
1326 static void __del_from_free_nid_list(struct f2fs_nm_info
*nm_i
,
1330 radix_tree_delete(&nm_i
->free_nid_root
, i
->nid
);
1333 static int add_free_nid(struct f2fs_sb_info
*sbi
, nid_t nid
, bool build
)
1335 struct f2fs_nm_info
*nm_i
= NM_I(sbi
);
1337 struct nat_entry
*ne
;
1338 bool allocated
= false;
1340 if (!available_free_memory(sbi
, FREE_NIDS
))
1343 /* 0 nid should not be used */
1344 if (unlikely(nid
== 0))
1348 /* do not add allocated nids */
1349 read_lock(&nm_i
->nat_tree_lock
);
1350 ne
= __lookup_nat_cache(nm_i
, nid
);
1352 (!ne
->checkpointed
|| nat_get_blkaddr(ne
) != NULL_ADDR
))
1354 read_unlock(&nm_i
->nat_tree_lock
);
1359 i
= f2fs_kmem_cache_alloc(free_nid_slab
, GFP_NOFS
);
1363 spin_lock(&nm_i
->free_nid_list_lock
);
1364 if (radix_tree_insert(&nm_i
->free_nid_root
, i
->nid
, i
)) {
1365 spin_unlock(&nm_i
->free_nid_list_lock
);
1366 kmem_cache_free(free_nid_slab
, i
);
1369 list_add_tail(&i
->list
, &nm_i
->free_nid_list
);
1371 spin_unlock(&nm_i
->free_nid_list_lock
);
1375 static void remove_free_nid(struct f2fs_nm_info
*nm_i
, nid_t nid
)
1378 bool need_free
= false;
1380 spin_lock(&nm_i
->free_nid_list_lock
);
1381 i
= __lookup_free_nid_list(nm_i
, nid
);
1382 if (i
&& i
->state
== NID_NEW
) {
1383 __del_from_free_nid_list(nm_i
, i
);
1387 spin_unlock(&nm_i
->free_nid_list_lock
);
1390 kmem_cache_free(free_nid_slab
, i
);
1393 static void scan_nat_page(struct f2fs_sb_info
*sbi
,
1394 struct page
*nat_page
, nid_t start_nid
)
1396 struct f2fs_nm_info
*nm_i
= NM_I(sbi
);
1397 struct f2fs_nat_block
*nat_blk
= page_address(nat_page
);
1401 i
= start_nid
% NAT_ENTRY_PER_BLOCK
;
1403 for (; i
< NAT_ENTRY_PER_BLOCK
; i
++, start_nid
++) {
1405 if (unlikely(start_nid
>= nm_i
->max_nid
))
1408 blk_addr
= le32_to_cpu(nat_blk
->entries
[i
].block_addr
);
1409 f2fs_bug_on(sbi
, blk_addr
== NEW_ADDR
);
1410 if (blk_addr
== NULL_ADDR
) {
1411 if (add_free_nid(sbi
, start_nid
, true) < 0)
1417 static void build_free_nids(struct f2fs_sb_info
*sbi
)
1419 struct f2fs_nm_info
*nm_i
= NM_I(sbi
);
1420 struct curseg_info
*curseg
= CURSEG_I(sbi
, CURSEG_HOT_DATA
);
1421 struct f2fs_summary_block
*sum
= curseg
->sum_blk
;
1423 nid_t nid
= nm_i
->next_scan_nid
;
1425 /* Enough entries */
1426 if (nm_i
->fcnt
> NAT_ENTRY_PER_BLOCK
)
1429 /* readahead nat pages to be scanned */
1430 ra_meta_pages(sbi
, NAT_BLOCK_OFFSET(nid
), FREE_NID_PAGES
, META_NAT
);
1433 struct page
*page
= get_current_nat_page(sbi
, nid
);
1435 scan_nat_page(sbi
, page
, nid
);
1436 f2fs_put_page(page
, 1);
1438 nid
+= (NAT_ENTRY_PER_BLOCK
- (nid
% NAT_ENTRY_PER_BLOCK
));
1439 if (unlikely(nid
>= nm_i
->max_nid
))
1442 if (i
++ == FREE_NID_PAGES
)
1446 /* go to the next free nat pages to find free nids abundantly */
1447 nm_i
->next_scan_nid
= nid
;
1449 /* find free nids from current sum_pages */
1450 mutex_lock(&curseg
->curseg_mutex
);
1451 for (i
= 0; i
< nats_in_cursum(sum
); i
++) {
1452 block_t addr
= le32_to_cpu(nat_in_journal(sum
, i
).block_addr
);
1453 nid
= le32_to_cpu(nid_in_journal(sum
, i
));
1454 if (addr
== NULL_ADDR
)
1455 add_free_nid(sbi
, nid
, true);
1457 remove_free_nid(nm_i
, nid
);
1459 mutex_unlock(&curseg
->curseg_mutex
);
1463 * If this function returns success, caller can obtain a new nid
1464 * from second parameter of this function.
1465 * The returned nid could be used ino as well as nid when inode is created.
1467 bool alloc_nid(struct f2fs_sb_info
*sbi
, nid_t
*nid
)
1469 struct f2fs_nm_info
*nm_i
= NM_I(sbi
);
1470 struct free_nid
*i
= NULL
;
1472 if (unlikely(sbi
->total_valid_node_count
+ 1 > nm_i
->available_nids
))
1475 spin_lock(&nm_i
->free_nid_list_lock
);
1477 /* We should not use stale free nids created by build_free_nids */
1478 if (nm_i
->fcnt
&& !on_build_free_nids(nm_i
)) {
1479 f2fs_bug_on(sbi
, list_empty(&nm_i
->free_nid_list
));
1480 list_for_each_entry(i
, &nm_i
->free_nid_list
, list
)
1481 if (i
->state
== NID_NEW
)
1484 f2fs_bug_on(sbi
, i
->state
!= NID_NEW
);
1486 i
->state
= NID_ALLOC
;
1488 spin_unlock(&nm_i
->free_nid_list_lock
);
1491 spin_unlock(&nm_i
->free_nid_list_lock
);
1493 /* Let's scan nat pages and its caches to get free nids */
1494 mutex_lock(&nm_i
->build_lock
);
1495 build_free_nids(sbi
);
1496 mutex_unlock(&nm_i
->build_lock
);
1501 * alloc_nid() should be called prior to this function.
1503 void alloc_nid_done(struct f2fs_sb_info
*sbi
, nid_t nid
)
1505 struct f2fs_nm_info
*nm_i
= NM_I(sbi
);
1508 spin_lock(&nm_i
->free_nid_list_lock
);
1509 i
= __lookup_free_nid_list(nm_i
, nid
);
1510 f2fs_bug_on(sbi
, !i
|| i
->state
!= NID_ALLOC
);
1511 __del_from_free_nid_list(nm_i
, i
);
1512 spin_unlock(&nm_i
->free_nid_list_lock
);
1514 kmem_cache_free(free_nid_slab
, i
);
1518 * alloc_nid() should be called prior to this function.
1520 void alloc_nid_failed(struct f2fs_sb_info
*sbi
, nid_t nid
)
1522 struct f2fs_nm_info
*nm_i
= NM_I(sbi
);
1524 bool need_free
= false;
1529 spin_lock(&nm_i
->free_nid_list_lock
);
1530 i
= __lookup_free_nid_list(nm_i
, nid
);
1531 f2fs_bug_on(sbi
, !i
|| i
->state
!= NID_ALLOC
);
1532 if (!available_free_memory(sbi
, FREE_NIDS
)) {
1533 __del_from_free_nid_list(nm_i
, i
);
1539 spin_unlock(&nm_i
->free_nid_list_lock
);
1542 kmem_cache_free(free_nid_slab
, i
);
1545 void recover_inline_xattr(struct inode
*inode
, struct page
*page
)
1547 void *src_addr
, *dst_addr
;
1550 struct f2fs_inode
*ri
;
1552 ipage
= get_node_page(F2FS_I_SB(inode
), inode
->i_ino
);
1553 f2fs_bug_on(F2FS_I_SB(inode
), IS_ERR(ipage
));
1555 ri
= F2FS_INODE(page
);
1556 if (!(ri
->i_inline
& F2FS_INLINE_XATTR
)) {
1557 clear_inode_flag(F2FS_I(inode
), FI_INLINE_XATTR
);
1561 dst_addr
= inline_xattr_addr(ipage
);
1562 src_addr
= inline_xattr_addr(page
);
1563 inline_size
= inline_xattr_size(inode
);
1565 f2fs_wait_on_page_writeback(ipage
, NODE
);
1566 memcpy(dst_addr
, src_addr
, inline_size
);
1568 update_inode(inode
, ipage
);
1569 f2fs_put_page(ipage
, 1);
1572 void recover_xattr_data(struct inode
*inode
, struct page
*page
, block_t blkaddr
)
1574 struct f2fs_sb_info
*sbi
= F2FS_I_SB(inode
);
1575 nid_t prev_xnid
= F2FS_I(inode
)->i_xattr_nid
;
1576 nid_t new_xnid
= nid_of_node(page
);
1577 struct node_info ni
;
1579 /* 1: invalidate the previous xattr nid */
1583 /* Deallocate node address */
1584 get_node_info(sbi
, prev_xnid
, &ni
);
1585 f2fs_bug_on(sbi
, ni
.blk_addr
== NULL_ADDR
);
1586 invalidate_blocks(sbi
, ni
.blk_addr
);
1587 dec_valid_node_count(sbi
, inode
);
1588 set_node_addr(sbi
, &ni
, NULL_ADDR
, false);
1591 /* 2: allocate new xattr nid */
1592 if (unlikely(!inc_valid_node_count(sbi
, inode
)))
1593 f2fs_bug_on(sbi
, 1);
1595 remove_free_nid(NM_I(sbi
), new_xnid
);
1596 get_node_info(sbi
, new_xnid
, &ni
);
1597 ni
.ino
= inode
->i_ino
;
1598 set_node_addr(sbi
, &ni
, NEW_ADDR
, false);
1599 F2FS_I(inode
)->i_xattr_nid
= new_xnid
;
1601 /* 3: update xattr blkaddr */
1602 refresh_sit_entry(sbi
, NEW_ADDR
, blkaddr
);
1603 set_node_addr(sbi
, &ni
, blkaddr
, false);
1605 update_inode_page(inode
);
1608 int recover_inode_page(struct f2fs_sb_info
*sbi
, struct page
*page
)
1610 struct f2fs_inode
*src
, *dst
;
1611 nid_t ino
= ino_of_node(page
);
1612 struct node_info old_ni
, new_ni
;
1615 get_node_info(sbi
, ino
, &old_ni
);
1617 if (unlikely(old_ni
.blk_addr
!= NULL_ADDR
))
1620 ipage
= grab_cache_page(NODE_MAPPING(sbi
), ino
);
1624 /* Should not use this inode from free nid list */
1625 remove_free_nid(NM_I(sbi
), ino
);
1627 SetPageUptodate(ipage
);
1628 fill_node_footer(ipage
, ino
, ino
, 0, true);
1630 src
= F2FS_INODE(page
);
1631 dst
= F2FS_INODE(ipage
);
1633 memcpy(dst
, src
, (unsigned long)&src
->i_ext
- (unsigned long)src
);
1635 dst
->i_blocks
= cpu_to_le64(1);
1636 dst
->i_links
= cpu_to_le32(1);
1637 dst
->i_xattr_nid
= 0;
1638 dst
->i_inline
= src
->i_inline
& F2FS_INLINE_XATTR
;
1643 if (unlikely(!inc_valid_node_count(sbi
, NULL
)))
1645 set_node_addr(sbi
, &new_ni
, NEW_ADDR
, false);
1646 inc_valid_inode_count(sbi
);
1647 set_page_dirty(ipage
);
1648 f2fs_put_page(ipage
, 1);
1653 * ra_sum_pages() merge contiguous pages into one bio and submit.
1654 * these pre-read pages are allocated in bd_inode's mapping tree.
1656 static int ra_sum_pages(struct f2fs_sb_info
*sbi
, struct page
**pages
,
1657 int start
, int nrpages
)
1659 struct inode
*inode
= sbi
->sb
->s_bdev
->bd_inode
;
1660 struct address_space
*mapping
= inode
->i_mapping
;
1661 int i
, page_idx
= start
;
1662 struct f2fs_io_info fio
= {
1664 .rw
= READ_SYNC
| REQ_META
| REQ_PRIO
1667 for (i
= 0; page_idx
< start
+ nrpages
; page_idx
++, i
++) {
1668 /* alloc page in bd_inode for reading node summary info */
1669 pages
[i
] = grab_cache_page(mapping
, page_idx
);
1672 f2fs_submit_page_mbio(sbi
, pages
[i
], page_idx
, &fio
);
1675 f2fs_submit_merged_bio(sbi
, META
, READ
);
1679 int restore_node_summary(struct f2fs_sb_info
*sbi
,
1680 unsigned int segno
, struct f2fs_summary_block
*sum
)
1682 struct f2fs_node
*rn
;
1683 struct f2fs_summary
*sum_entry
;
1684 struct inode
*inode
= sbi
->sb
->s_bdev
->bd_inode
;
1686 int bio_blocks
= MAX_BIO_BLOCKS(max_hw_blocks(sbi
));
1687 struct page
*pages
[bio_blocks
];
1688 int i
, idx
, last_offset
, nrpages
, err
= 0;
1690 /* scan the node segment */
1691 last_offset
= sbi
->blocks_per_seg
;
1692 addr
= START_BLOCK(sbi
, segno
);
1693 sum_entry
= &sum
->entries
[0];
1695 for (i
= 0; !err
&& i
< last_offset
; i
+= nrpages
, addr
+= nrpages
) {
1696 nrpages
= min(last_offset
- i
, bio_blocks
);
1698 /* readahead node pages */
1699 nrpages
= ra_sum_pages(sbi
, pages
, addr
, nrpages
);
1703 for (idx
= 0; idx
< nrpages
; idx
++) {
1707 lock_page(pages
[idx
]);
1708 if (unlikely(!PageUptodate(pages
[idx
]))) {
1711 rn
= F2FS_NODE(pages
[idx
]);
1712 sum_entry
->nid
= rn
->footer
.nid
;
1713 sum_entry
->version
= 0;
1714 sum_entry
->ofs_in_node
= 0;
1717 unlock_page(pages
[idx
]);
1719 page_cache_release(pages
[idx
]);
1722 invalidate_mapping_pages(inode
->i_mapping
, addr
,
1728 static struct nat_entry_set
*grab_nat_entry_set(void)
1730 struct nat_entry_set
*nes
=
1731 f2fs_kmem_cache_alloc(nat_entry_set_slab
, GFP_ATOMIC
);
1734 INIT_LIST_HEAD(&nes
->set_list
);
1735 INIT_LIST_HEAD(&nes
->entry_list
);
1739 static void release_nat_entry_set(struct nat_entry_set
*nes
,
1740 struct f2fs_nm_info
*nm_i
)
1742 nm_i
->dirty_nat_cnt
-= nes
->entry_cnt
;
1743 list_del(&nes
->set_list
);
1744 kmem_cache_free(nat_entry_set_slab
, nes
);
1747 static void adjust_nat_entry_set(struct nat_entry_set
*nes
,
1748 struct list_head
*head
)
1750 struct nat_entry_set
*next
= nes
;
1752 if (list_is_last(&nes
->set_list
, head
))
1755 list_for_each_entry_continue(next
, head
, set_list
)
1756 if (nes
->entry_cnt
<= next
->entry_cnt
)
1759 list_move_tail(&nes
->set_list
, &next
->set_list
);
1762 static void add_nat_entry(struct nat_entry
*ne
, struct list_head
*head
)
1764 struct nat_entry_set
*nes
;
1765 nid_t start_nid
= START_NID(ne
->ni
.nid
);
1767 list_for_each_entry(nes
, head
, set_list
) {
1768 if (nes
->start_nid
== start_nid
) {
1769 list_move_tail(&ne
->list
, &nes
->entry_list
);
1771 adjust_nat_entry_set(nes
, head
);
1776 nes
= grab_nat_entry_set();
1778 nes
->start_nid
= start_nid
;
1779 list_move_tail(&ne
->list
, &nes
->entry_list
);
1781 list_add(&nes
->set_list
, head
);
1784 static void merge_nats_in_set(struct f2fs_sb_info
*sbi
)
1786 struct f2fs_nm_info
*nm_i
= NM_I(sbi
);
1787 struct list_head
*dirty_list
= &nm_i
->dirty_nat_entries
;
1788 struct list_head
*set_list
= &nm_i
->nat_entry_set
;
1789 struct nat_entry
*ne
, *tmp
;
1791 write_lock(&nm_i
->nat_tree_lock
);
1792 list_for_each_entry_safe(ne
, tmp
, dirty_list
, list
) {
1793 if (nat_get_blkaddr(ne
) == NEW_ADDR
)
1795 add_nat_entry(ne
, set_list
);
1796 nm_i
->dirty_nat_cnt
++;
1798 write_unlock(&nm_i
->nat_tree_lock
);
1801 static bool __has_cursum_space(struct f2fs_summary_block
*sum
, int size
)
1803 if (nats_in_cursum(sum
) + size
<= NAT_JOURNAL_ENTRIES
)
1809 static void remove_nats_in_journal(struct f2fs_sb_info
*sbi
)
1811 struct f2fs_nm_info
*nm_i
= NM_I(sbi
);
1812 struct curseg_info
*curseg
= CURSEG_I(sbi
, CURSEG_HOT_DATA
);
1813 struct f2fs_summary_block
*sum
= curseg
->sum_blk
;
1816 mutex_lock(&curseg
->curseg_mutex
);
1817 for (i
= 0; i
< nats_in_cursum(sum
); i
++) {
1818 struct nat_entry
*ne
;
1819 struct f2fs_nat_entry raw_ne
;
1820 nid_t nid
= le32_to_cpu(nid_in_journal(sum
, i
));
1822 raw_ne
= nat_in_journal(sum
, i
);
1824 write_lock(&nm_i
->nat_tree_lock
);
1825 ne
= __lookup_nat_cache(nm_i
, nid
);
1829 ne
= grab_nat_entry(nm_i
, nid
);
1831 write_unlock(&nm_i
->nat_tree_lock
);
1834 node_info_from_raw_nat(&ne
->ni
, &raw_ne
);
1836 __set_nat_cache_dirty(nm_i
, ne
);
1837 write_unlock(&nm_i
->nat_tree_lock
);
1839 update_nats_in_cursum(sum
, -i
);
1840 mutex_unlock(&curseg
->curseg_mutex
);
1844 * This function is called during the checkpointing process.
1846 void flush_nat_entries(struct f2fs_sb_info
*sbi
)
1848 struct f2fs_nm_info
*nm_i
= NM_I(sbi
);
1849 struct curseg_info
*curseg
= CURSEG_I(sbi
, CURSEG_HOT_DATA
);
1850 struct f2fs_summary_block
*sum
= curseg
->sum_blk
;
1851 struct nat_entry_set
*nes
, *tmp
;
1852 struct list_head
*head
= &nm_i
->nat_entry_set
;
1853 bool to_journal
= true;
1855 /* merge nat entries of dirty list to nat entry set temporarily */
1856 merge_nats_in_set(sbi
);
1859 * if there are no enough space in journal to store dirty nat
1860 * entries, remove all entries from journal and merge them
1861 * into nat entry set.
1863 if (!__has_cursum_space(sum
, nm_i
->dirty_nat_cnt
)) {
1864 remove_nats_in_journal(sbi
);
1867 * merge nat entries of dirty list to nat entry set temporarily
1869 merge_nats_in_set(sbi
);
1872 if (!nm_i
->dirty_nat_cnt
)
1876 * there are two steps to flush nat entries:
1877 * #1, flush nat entries to journal in current hot data summary block.
1878 * #2, flush nat entries to nat page.
1880 list_for_each_entry_safe(nes
, tmp
, head
, set_list
) {
1881 struct f2fs_nat_block
*nat_blk
;
1882 struct nat_entry
*ne
, *cur
;
1884 nid_t start_nid
= nes
->start_nid
;
1886 if (to_journal
&& !__has_cursum_space(sum
, nes
->entry_cnt
))
1890 mutex_lock(&curseg
->curseg_mutex
);
1892 page
= get_next_nat_page(sbi
, start_nid
);
1893 nat_blk
= page_address(page
);
1894 f2fs_bug_on(sbi
, !nat_blk
);
1897 /* flush dirty nats in nat entry set */
1898 list_for_each_entry_safe(ne
, cur
, &nes
->entry_list
, list
) {
1899 struct f2fs_nat_entry
*raw_ne
;
1900 nid_t nid
= nat_get_nid(ne
);
1904 offset
= lookup_journal_in_cursum(sum
,
1905 NAT_JOURNAL
, nid
, 1);
1906 f2fs_bug_on(sbi
, offset
< 0);
1907 raw_ne
= &nat_in_journal(sum
, offset
);
1908 nid_in_journal(sum
, offset
) = cpu_to_le32(nid
);
1910 raw_ne
= &nat_blk
->entries
[nid
- start_nid
];
1912 raw_nat_from_node_info(raw_ne
, &ne
->ni
);
1914 if (nat_get_blkaddr(ne
) == NULL_ADDR
&&
1915 add_free_nid(sbi
, nid
, false) <= 0) {
1916 write_lock(&nm_i
->nat_tree_lock
);
1917 __del_from_nat_cache(nm_i
, ne
);
1918 write_unlock(&nm_i
->nat_tree_lock
);
1920 write_lock(&nm_i
->nat_tree_lock
);
1921 __clear_nat_cache_dirty(nm_i
, ne
);
1922 write_unlock(&nm_i
->nat_tree_lock
);
1927 mutex_unlock(&curseg
->curseg_mutex
);
1929 f2fs_put_page(page
, 1);
1931 f2fs_bug_on(sbi
, !list_empty(&nes
->entry_list
));
1932 release_nat_entry_set(nes
, nm_i
);
1935 f2fs_bug_on(sbi
, !list_empty(head
));
1936 f2fs_bug_on(sbi
, nm_i
->dirty_nat_cnt
);
1939 static int init_node_manager(struct f2fs_sb_info
*sbi
)
1941 struct f2fs_super_block
*sb_raw
= F2FS_RAW_SUPER(sbi
);
1942 struct f2fs_nm_info
*nm_i
= NM_I(sbi
);
1943 unsigned char *version_bitmap
;
1944 unsigned int nat_segs
, nat_blocks
;
1946 nm_i
->nat_blkaddr
= le32_to_cpu(sb_raw
->nat_blkaddr
);
1948 /* segment_count_nat includes pair segment so divide to 2. */
1949 nat_segs
= le32_to_cpu(sb_raw
->segment_count_nat
) >> 1;
1950 nat_blocks
= nat_segs
<< le32_to_cpu(sb_raw
->log_blocks_per_seg
);
1952 nm_i
->max_nid
= NAT_ENTRY_PER_BLOCK
* nat_blocks
;
1954 /* not used nids: 0, node, meta, (and root counted as valid node) */
1955 nm_i
->available_nids
= nm_i
->max_nid
- F2FS_RESERVED_NODE_NUM
;
1958 nm_i
->ram_thresh
= DEF_RAM_THRESHOLD
;
1960 INIT_RADIX_TREE(&nm_i
->free_nid_root
, GFP_ATOMIC
);
1961 INIT_LIST_HEAD(&nm_i
->free_nid_list
);
1962 INIT_RADIX_TREE(&nm_i
->nat_root
, GFP_ATOMIC
);
1963 INIT_LIST_HEAD(&nm_i
->nat_entries
);
1964 INIT_LIST_HEAD(&nm_i
->dirty_nat_entries
);
1965 INIT_LIST_HEAD(&nm_i
->nat_entry_set
);
1967 mutex_init(&nm_i
->build_lock
);
1968 spin_lock_init(&nm_i
->free_nid_list_lock
);
1969 rwlock_init(&nm_i
->nat_tree_lock
);
1971 nm_i
->next_scan_nid
= le32_to_cpu(sbi
->ckpt
->next_free_nid
);
1972 nm_i
->bitmap_size
= __bitmap_size(sbi
, NAT_BITMAP
);
1973 version_bitmap
= __bitmap_ptr(sbi
, NAT_BITMAP
);
1974 if (!version_bitmap
)
1977 nm_i
->nat_bitmap
= kmemdup(version_bitmap
, nm_i
->bitmap_size
,
1979 if (!nm_i
->nat_bitmap
)
1984 int build_node_manager(struct f2fs_sb_info
*sbi
)
1988 sbi
->nm_info
= kzalloc(sizeof(struct f2fs_nm_info
), GFP_KERNEL
);
1992 err
= init_node_manager(sbi
);
1996 build_free_nids(sbi
);
2000 void destroy_node_manager(struct f2fs_sb_info
*sbi
)
2002 struct f2fs_nm_info
*nm_i
= NM_I(sbi
);
2003 struct free_nid
*i
, *next_i
;
2004 struct nat_entry
*natvec
[NATVEC_SIZE
];
2011 /* destroy free nid list */
2012 spin_lock(&nm_i
->free_nid_list_lock
);
2013 list_for_each_entry_safe(i
, next_i
, &nm_i
->free_nid_list
, list
) {
2014 f2fs_bug_on(sbi
, i
->state
== NID_ALLOC
);
2015 __del_from_free_nid_list(nm_i
, i
);
2017 spin_unlock(&nm_i
->free_nid_list_lock
);
2018 kmem_cache_free(free_nid_slab
, i
);
2019 spin_lock(&nm_i
->free_nid_list_lock
);
2021 f2fs_bug_on(sbi
, nm_i
->fcnt
);
2022 spin_unlock(&nm_i
->free_nid_list_lock
);
2024 /* destroy nat cache */
2025 write_lock(&nm_i
->nat_tree_lock
);
2026 while ((found
= __gang_lookup_nat_cache(nm_i
,
2027 nid
, NATVEC_SIZE
, natvec
))) {
2029 nid
= nat_get_nid(natvec
[found
- 1]) + 1;
2030 for (idx
= 0; idx
< found
; idx
++)
2031 __del_from_nat_cache(nm_i
, natvec
[idx
]);
2033 f2fs_bug_on(sbi
, nm_i
->nat_cnt
);
2034 write_unlock(&nm_i
->nat_tree_lock
);
2036 kfree(nm_i
->nat_bitmap
);
2037 sbi
->nm_info
= NULL
;
2041 int __init
create_node_manager_caches(void)
2043 nat_entry_slab
= f2fs_kmem_cache_create("nat_entry",
2044 sizeof(struct nat_entry
));
2045 if (!nat_entry_slab
)
2048 free_nid_slab
= f2fs_kmem_cache_create("free_nid",
2049 sizeof(struct free_nid
));
2051 goto destory_nat_entry
;
2053 nat_entry_set_slab
= f2fs_kmem_cache_create("nat_entry_set",
2054 sizeof(struct nat_entry_set
));
2055 if (!nat_entry_set_slab
)
2056 goto destory_free_nid
;
2060 kmem_cache_destroy(free_nid_slab
);
2062 kmem_cache_destroy(nat_entry_slab
);
2067 void destroy_node_manager_caches(void)
2069 kmem_cache_destroy(nat_entry_set_slab
);
2070 kmem_cache_destroy(free_nid_slab
);
2071 kmem_cache_destroy(nat_entry_slab
);