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>
24 #include <trace/events/f2fs.h>
26 #define on_build_free_nids(nmi) mutex_is_locked(&(nm_i)->build_lock)
28 static struct kmem_cache
*nat_entry_slab
;
29 static struct kmem_cache
*free_nid_slab
;
30 static struct kmem_cache
*nat_entry_set_slab
;
32 bool available_free_memory(struct f2fs_sb_info
*sbi
, int type
)
34 struct f2fs_nm_info
*nm_i
= NM_I(sbi
);
36 unsigned long avail_ram
;
37 unsigned long mem_size
= 0;
42 /* only uses low memory */
43 avail_ram
= val
.totalram
- val
.totalhigh
;
46 * give 25%, 25%, 50%, 50%, 50% memory for each components respectively
48 if (type
== FREE_NIDS
) {
49 mem_size
= (nm_i
->nid_cnt
[FREE_NID
] *
50 sizeof(struct free_nid
)) >> PAGE_SHIFT
;
51 res
= mem_size
< ((avail_ram
* nm_i
->ram_thresh
/ 100) >> 2);
52 } else if (type
== NAT_ENTRIES
) {
53 mem_size
= (nm_i
->nat_cnt
* sizeof(struct nat_entry
)) >>
55 res
= mem_size
< ((avail_ram
* nm_i
->ram_thresh
/ 100) >> 2);
56 if (excess_cached_nats(sbi
))
58 } else if (type
== DIRTY_DENTS
) {
59 if (sbi
->sb
->s_bdi
->wb
.dirty_exceeded
)
61 mem_size
= get_pages(sbi
, F2FS_DIRTY_DENTS
);
62 res
= mem_size
< ((avail_ram
* nm_i
->ram_thresh
/ 100) >> 1);
63 } else if (type
== INO_ENTRIES
) {
66 for (i
= 0; i
< MAX_INO_ENTRY
; i
++)
67 mem_size
+= sbi
->im
[i
].ino_num
*
68 sizeof(struct ino_entry
);
69 mem_size
>>= PAGE_SHIFT
;
70 res
= mem_size
< ((avail_ram
* nm_i
->ram_thresh
/ 100) >> 1);
71 } else if (type
== EXTENT_CACHE
) {
72 mem_size
= (atomic_read(&sbi
->total_ext_tree
) *
73 sizeof(struct extent_tree
) +
74 atomic_read(&sbi
->total_ext_node
) *
75 sizeof(struct extent_node
)) >> PAGE_SHIFT
;
76 res
= mem_size
< ((avail_ram
* nm_i
->ram_thresh
/ 100) >> 1);
77 } else if (type
== INMEM_PAGES
) {
78 /* it allows 20% / total_ram for inmemory pages */
79 mem_size
= get_pages(sbi
, F2FS_INMEM_PAGES
);
80 res
= mem_size
< (val
.totalram
/ 5);
82 if (!sbi
->sb
->s_bdi
->wb
.dirty_exceeded
)
88 static void clear_node_page_dirty(struct page
*page
)
90 struct address_space
*mapping
= page
->mapping
;
91 unsigned int long flags
;
93 if (PageDirty(page
)) {
94 spin_lock_irqsave(&mapping
->tree_lock
, flags
);
95 radix_tree_tag_clear(&mapping
->page_tree
,
98 spin_unlock_irqrestore(&mapping
->tree_lock
, flags
);
100 clear_page_dirty_for_io(page
);
101 dec_page_count(F2FS_M_SB(mapping
), F2FS_DIRTY_NODES
);
103 ClearPageUptodate(page
);
106 static struct page
*get_current_nat_page(struct f2fs_sb_info
*sbi
, nid_t nid
)
108 pgoff_t index
= current_nat_addr(sbi
, nid
);
109 return get_meta_page(sbi
, index
);
112 static struct page
*get_next_nat_page(struct f2fs_sb_info
*sbi
, nid_t nid
)
114 struct page
*src_page
;
115 struct page
*dst_page
;
120 struct f2fs_nm_info
*nm_i
= NM_I(sbi
);
122 src_off
= current_nat_addr(sbi
, nid
);
123 dst_off
= next_nat_addr(sbi
, src_off
);
125 /* get current nat block page with lock */
126 src_page
= get_meta_page(sbi
, src_off
);
127 dst_page
= grab_meta_page(sbi
, dst_off
);
128 f2fs_bug_on(sbi
, PageDirty(src_page
));
130 src_addr
= page_address(src_page
);
131 dst_addr
= page_address(dst_page
);
132 memcpy(dst_addr
, src_addr
, PAGE_SIZE
);
133 set_page_dirty(dst_page
);
134 f2fs_put_page(src_page
, 1);
136 set_to_next_nat(nm_i
, nid
);
141 static struct nat_entry
*__lookup_nat_cache(struct f2fs_nm_info
*nm_i
, nid_t n
)
143 return radix_tree_lookup(&nm_i
->nat_root
, n
);
146 static unsigned int __gang_lookup_nat_cache(struct f2fs_nm_info
*nm_i
,
147 nid_t start
, unsigned int nr
, struct nat_entry
**ep
)
149 return radix_tree_gang_lookup(&nm_i
->nat_root
, (void **)ep
, start
, nr
);
152 static void __del_from_nat_cache(struct f2fs_nm_info
*nm_i
, struct nat_entry
*e
)
155 radix_tree_delete(&nm_i
->nat_root
, nat_get_nid(e
));
157 kmem_cache_free(nat_entry_slab
, e
);
160 static void __set_nat_cache_dirty(struct f2fs_nm_info
*nm_i
,
161 struct nat_entry
*ne
)
163 nid_t set
= NAT_BLOCK_OFFSET(ne
->ni
.nid
);
164 struct nat_entry_set
*head
;
166 head
= radix_tree_lookup(&nm_i
->nat_set_root
, set
);
168 head
= f2fs_kmem_cache_alloc(nat_entry_set_slab
, GFP_NOFS
);
170 INIT_LIST_HEAD(&head
->entry_list
);
171 INIT_LIST_HEAD(&head
->set_list
);
174 f2fs_radix_tree_insert(&nm_i
->nat_set_root
, set
, head
);
177 if (get_nat_flag(ne
, IS_DIRTY
))
180 nm_i
->dirty_nat_cnt
++;
182 set_nat_flag(ne
, IS_DIRTY
, true);
184 if (nat_get_blkaddr(ne
) == NEW_ADDR
)
185 list_del_init(&ne
->list
);
187 list_move_tail(&ne
->list
, &head
->entry_list
);
190 static void __clear_nat_cache_dirty(struct f2fs_nm_info
*nm_i
,
191 struct nat_entry_set
*set
, struct nat_entry
*ne
)
193 list_move_tail(&ne
->list
, &nm_i
->nat_entries
);
194 set_nat_flag(ne
, IS_DIRTY
, false);
196 nm_i
->dirty_nat_cnt
--;
199 static unsigned int __gang_lookup_nat_set(struct f2fs_nm_info
*nm_i
,
200 nid_t start
, unsigned int nr
, struct nat_entry_set
**ep
)
202 return radix_tree_gang_lookup(&nm_i
->nat_set_root
, (void **)ep
,
206 int need_dentry_mark(struct f2fs_sb_info
*sbi
, nid_t nid
)
208 struct f2fs_nm_info
*nm_i
= NM_I(sbi
);
212 down_read(&nm_i
->nat_tree_lock
);
213 e
= __lookup_nat_cache(nm_i
, nid
);
215 if (!get_nat_flag(e
, IS_CHECKPOINTED
) &&
216 !get_nat_flag(e
, HAS_FSYNCED_INODE
))
219 up_read(&nm_i
->nat_tree_lock
);
223 bool is_checkpointed_node(struct f2fs_sb_info
*sbi
, nid_t nid
)
225 struct f2fs_nm_info
*nm_i
= NM_I(sbi
);
229 down_read(&nm_i
->nat_tree_lock
);
230 e
= __lookup_nat_cache(nm_i
, nid
);
231 if (e
&& !get_nat_flag(e
, IS_CHECKPOINTED
))
233 up_read(&nm_i
->nat_tree_lock
);
237 bool need_inode_block_update(struct f2fs_sb_info
*sbi
, nid_t ino
)
239 struct f2fs_nm_info
*nm_i
= NM_I(sbi
);
241 bool need_update
= true;
243 down_read(&nm_i
->nat_tree_lock
);
244 e
= __lookup_nat_cache(nm_i
, ino
);
245 if (e
&& get_nat_flag(e
, HAS_LAST_FSYNC
) &&
246 (get_nat_flag(e
, IS_CHECKPOINTED
) ||
247 get_nat_flag(e
, HAS_FSYNCED_INODE
)))
249 up_read(&nm_i
->nat_tree_lock
);
253 static struct nat_entry
*grab_nat_entry(struct f2fs_nm_info
*nm_i
, nid_t nid
,
256 struct nat_entry
*new;
259 new = f2fs_kmem_cache_alloc(nat_entry_slab
, GFP_NOFS
);
260 f2fs_radix_tree_insert(&nm_i
->nat_root
, nid
, new);
262 new = kmem_cache_alloc(nat_entry_slab
, GFP_NOFS
);
265 if (radix_tree_insert(&nm_i
->nat_root
, nid
, new)) {
266 kmem_cache_free(nat_entry_slab
, new);
271 memset(new, 0, sizeof(struct nat_entry
));
272 nat_set_nid(new, nid
);
274 list_add_tail(&new->list
, &nm_i
->nat_entries
);
279 static void cache_nat_entry(struct f2fs_sb_info
*sbi
, nid_t nid
,
280 struct f2fs_nat_entry
*ne
)
282 struct f2fs_nm_info
*nm_i
= NM_I(sbi
);
285 e
= __lookup_nat_cache(nm_i
, nid
);
287 e
= grab_nat_entry(nm_i
, nid
, false);
289 node_info_from_raw_nat(&e
->ni
, ne
);
291 f2fs_bug_on(sbi
, nat_get_ino(e
) != le32_to_cpu(ne
->ino
) ||
292 nat_get_blkaddr(e
) !=
293 le32_to_cpu(ne
->block_addr
) ||
294 nat_get_version(e
) != ne
->version
);
298 static void set_node_addr(struct f2fs_sb_info
*sbi
, struct node_info
*ni
,
299 block_t new_blkaddr
, bool fsync_done
)
301 struct f2fs_nm_info
*nm_i
= NM_I(sbi
);
304 down_write(&nm_i
->nat_tree_lock
);
305 e
= __lookup_nat_cache(nm_i
, ni
->nid
);
307 e
= grab_nat_entry(nm_i
, ni
->nid
, true);
308 copy_node_info(&e
->ni
, ni
);
309 f2fs_bug_on(sbi
, ni
->blk_addr
== NEW_ADDR
);
310 } else if (new_blkaddr
== NEW_ADDR
) {
312 * when nid is reallocated,
313 * previous nat entry can be remained in nat cache.
314 * So, reinitialize it with new information.
316 copy_node_info(&e
->ni
, ni
);
317 f2fs_bug_on(sbi
, ni
->blk_addr
!= NULL_ADDR
);
321 f2fs_bug_on(sbi
, nat_get_blkaddr(e
) != ni
->blk_addr
);
322 f2fs_bug_on(sbi
, nat_get_blkaddr(e
) == NULL_ADDR
&&
323 new_blkaddr
== NULL_ADDR
);
324 f2fs_bug_on(sbi
, nat_get_blkaddr(e
) == NEW_ADDR
&&
325 new_blkaddr
== NEW_ADDR
);
326 f2fs_bug_on(sbi
, nat_get_blkaddr(e
) != NEW_ADDR
&&
327 nat_get_blkaddr(e
) != NULL_ADDR
&&
328 new_blkaddr
== NEW_ADDR
);
330 /* increment version no as node is removed */
331 if (nat_get_blkaddr(e
) != NEW_ADDR
&& new_blkaddr
== NULL_ADDR
) {
332 unsigned char version
= nat_get_version(e
);
333 nat_set_version(e
, inc_node_version(version
));
337 nat_set_blkaddr(e
, new_blkaddr
);
338 if (new_blkaddr
== NEW_ADDR
|| new_blkaddr
== NULL_ADDR
)
339 set_nat_flag(e
, IS_CHECKPOINTED
, false);
340 __set_nat_cache_dirty(nm_i
, e
);
342 /* update fsync_mark if its inode nat entry is still alive */
343 if (ni
->nid
!= ni
->ino
)
344 e
= __lookup_nat_cache(nm_i
, ni
->ino
);
346 if (fsync_done
&& ni
->nid
== ni
->ino
)
347 set_nat_flag(e
, HAS_FSYNCED_INODE
, true);
348 set_nat_flag(e
, HAS_LAST_FSYNC
, fsync_done
);
350 up_write(&nm_i
->nat_tree_lock
);
353 int try_to_free_nats(struct f2fs_sb_info
*sbi
, int nr_shrink
)
355 struct f2fs_nm_info
*nm_i
= NM_I(sbi
);
358 if (!down_write_trylock(&nm_i
->nat_tree_lock
))
361 while (nr_shrink
&& !list_empty(&nm_i
->nat_entries
)) {
362 struct nat_entry
*ne
;
363 ne
= list_first_entry(&nm_i
->nat_entries
,
364 struct nat_entry
, list
);
365 __del_from_nat_cache(nm_i
, ne
);
368 up_write(&nm_i
->nat_tree_lock
);
369 return nr
- nr_shrink
;
373 * This function always returns success
375 void get_node_info(struct f2fs_sb_info
*sbi
, nid_t nid
, struct node_info
*ni
)
377 struct f2fs_nm_info
*nm_i
= NM_I(sbi
);
378 struct curseg_info
*curseg
= CURSEG_I(sbi
, CURSEG_HOT_DATA
);
379 struct f2fs_journal
*journal
= curseg
->journal
;
380 nid_t start_nid
= START_NID(nid
);
381 struct f2fs_nat_block
*nat_blk
;
382 struct page
*page
= NULL
;
383 struct f2fs_nat_entry ne
;
390 /* Check nat cache */
391 down_read(&nm_i
->nat_tree_lock
);
392 e
= __lookup_nat_cache(nm_i
, nid
);
394 ni
->ino
= nat_get_ino(e
);
395 ni
->blk_addr
= nat_get_blkaddr(e
);
396 ni
->version
= nat_get_version(e
);
397 up_read(&nm_i
->nat_tree_lock
);
401 memset(&ne
, 0, sizeof(struct f2fs_nat_entry
));
403 /* Check current segment summary */
404 down_read(&curseg
->journal_rwsem
);
405 i
= lookup_journal_in_cursum(journal
, NAT_JOURNAL
, nid
, 0);
407 ne
= nat_in_journal(journal
, i
);
408 node_info_from_raw_nat(ni
, &ne
);
410 up_read(&curseg
->journal_rwsem
);
412 up_read(&nm_i
->nat_tree_lock
);
416 /* Fill node_info from nat page */
417 index
= current_nat_addr(sbi
, nid
);
418 up_read(&nm_i
->nat_tree_lock
);
420 page
= get_meta_page(sbi
, index
);
421 nat_blk
= (struct f2fs_nat_block
*)page_address(page
);
422 ne
= nat_blk
->entries
[nid
- start_nid
];
423 node_info_from_raw_nat(ni
, &ne
);
424 f2fs_put_page(page
, 1);
426 /* cache nat entry */
427 down_write(&nm_i
->nat_tree_lock
);
428 cache_nat_entry(sbi
, nid
, &ne
);
429 up_write(&nm_i
->nat_tree_lock
);
433 * readahead MAX_RA_NODE number of node pages.
435 static void ra_node_pages(struct page
*parent
, int start
, int n
)
437 struct f2fs_sb_info
*sbi
= F2FS_P_SB(parent
);
438 struct blk_plug plug
;
442 blk_start_plug(&plug
);
444 /* Then, try readahead for siblings of the desired node */
446 end
= min(end
, NIDS_PER_BLOCK
);
447 for (i
= start
; i
< end
; i
++) {
448 nid
= get_nid(parent
, i
, false);
449 ra_node_page(sbi
, nid
);
452 blk_finish_plug(&plug
);
455 pgoff_t
get_next_page_offset(struct dnode_of_data
*dn
, pgoff_t pgofs
)
457 const long direct_index
= ADDRS_PER_INODE(dn
->inode
);
458 const long direct_blks
= ADDRS_PER_BLOCK
;
459 const long indirect_blks
= ADDRS_PER_BLOCK
* NIDS_PER_BLOCK
;
460 unsigned int skipped_unit
= ADDRS_PER_BLOCK
;
461 int cur_level
= dn
->cur_level
;
462 int max_level
= dn
->max_level
;
468 while (max_level
-- > cur_level
)
469 skipped_unit
*= NIDS_PER_BLOCK
;
471 switch (dn
->max_level
) {
473 base
+= 2 * indirect_blks
;
475 base
+= 2 * direct_blks
;
477 base
+= direct_index
;
480 f2fs_bug_on(F2FS_I_SB(dn
->inode
), 1);
483 return ((pgofs
- base
) / skipped_unit
+ 1) * skipped_unit
+ base
;
487 * The maximum depth is four.
488 * Offset[0] will have raw inode offset.
490 static int get_node_path(struct inode
*inode
, long block
,
491 int offset
[4], unsigned int noffset
[4])
493 const long direct_index
= ADDRS_PER_INODE(inode
);
494 const long direct_blks
= ADDRS_PER_BLOCK
;
495 const long dptrs_per_blk
= NIDS_PER_BLOCK
;
496 const long indirect_blks
= ADDRS_PER_BLOCK
* NIDS_PER_BLOCK
;
497 const long dindirect_blks
= indirect_blks
* NIDS_PER_BLOCK
;
503 if (block
< direct_index
) {
507 block
-= direct_index
;
508 if (block
< direct_blks
) {
509 offset
[n
++] = NODE_DIR1_BLOCK
;
515 block
-= direct_blks
;
516 if (block
< direct_blks
) {
517 offset
[n
++] = NODE_DIR2_BLOCK
;
523 block
-= direct_blks
;
524 if (block
< indirect_blks
) {
525 offset
[n
++] = NODE_IND1_BLOCK
;
527 offset
[n
++] = block
/ direct_blks
;
528 noffset
[n
] = 4 + offset
[n
- 1];
529 offset
[n
] = block
% direct_blks
;
533 block
-= indirect_blks
;
534 if (block
< indirect_blks
) {
535 offset
[n
++] = NODE_IND2_BLOCK
;
536 noffset
[n
] = 4 + dptrs_per_blk
;
537 offset
[n
++] = block
/ direct_blks
;
538 noffset
[n
] = 5 + dptrs_per_blk
+ offset
[n
- 1];
539 offset
[n
] = block
% direct_blks
;
543 block
-= indirect_blks
;
544 if (block
< dindirect_blks
) {
545 offset
[n
++] = NODE_DIND_BLOCK
;
546 noffset
[n
] = 5 + (dptrs_per_blk
* 2);
547 offset
[n
++] = block
/ indirect_blks
;
548 noffset
[n
] = 6 + (dptrs_per_blk
* 2) +
549 offset
[n
- 1] * (dptrs_per_blk
+ 1);
550 offset
[n
++] = (block
/ direct_blks
) % dptrs_per_blk
;
551 noffset
[n
] = 7 + (dptrs_per_blk
* 2) +
552 offset
[n
- 2] * (dptrs_per_blk
+ 1) +
554 offset
[n
] = block
% direct_blks
;
565 * Caller should call f2fs_put_dnode(dn).
566 * Also, it should grab and release a rwsem by calling f2fs_lock_op() and
567 * f2fs_unlock_op() only if ro is not set RDONLY_NODE.
568 * In the case of RDONLY_NODE, we don't need to care about mutex.
570 int get_dnode_of_data(struct dnode_of_data
*dn
, pgoff_t index
, int mode
)
572 struct f2fs_sb_info
*sbi
= F2FS_I_SB(dn
->inode
);
573 struct page
*npage
[4];
574 struct page
*parent
= NULL
;
576 unsigned int noffset
[4];
581 level
= get_node_path(dn
->inode
, index
, offset
, noffset
);
585 nids
[0] = dn
->inode
->i_ino
;
586 npage
[0] = dn
->inode_page
;
589 npage
[0] = get_node_page(sbi
, nids
[0]);
590 if (IS_ERR(npage
[0]))
591 return PTR_ERR(npage
[0]);
594 /* if inline_data is set, should not report any block indices */
595 if (f2fs_has_inline_data(dn
->inode
) && index
) {
597 f2fs_put_page(npage
[0], 1);
603 nids
[1] = get_nid(parent
, offset
[0], true);
604 dn
->inode_page
= npage
[0];
605 dn
->inode_page_locked
= true;
607 /* get indirect or direct nodes */
608 for (i
= 1; i
<= level
; i
++) {
611 if (!nids
[i
] && mode
== ALLOC_NODE
) {
613 if (!alloc_nid(sbi
, &(nids
[i
]))) {
619 npage
[i
] = new_node_page(dn
, noffset
[i
]);
620 if (IS_ERR(npage
[i
])) {
621 alloc_nid_failed(sbi
, nids
[i
]);
622 err
= PTR_ERR(npage
[i
]);
626 set_nid(parent
, offset
[i
- 1], nids
[i
], i
== 1);
627 alloc_nid_done(sbi
, nids
[i
]);
629 } else if (mode
== LOOKUP_NODE_RA
&& i
== level
&& level
> 1) {
630 npage
[i
] = get_node_page_ra(parent
, offset
[i
- 1]);
631 if (IS_ERR(npage
[i
])) {
632 err
= PTR_ERR(npage
[i
]);
638 dn
->inode_page_locked
= false;
641 f2fs_put_page(parent
, 1);
645 npage
[i
] = get_node_page(sbi
, nids
[i
]);
646 if (IS_ERR(npage
[i
])) {
647 err
= PTR_ERR(npage
[i
]);
648 f2fs_put_page(npage
[0], 0);
654 nids
[i
+ 1] = get_nid(parent
, offset
[i
], false);
657 dn
->nid
= nids
[level
];
658 dn
->ofs_in_node
= offset
[level
];
659 dn
->node_page
= npage
[level
];
660 dn
->data_blkaddr
= datablock_addr(dn
->inode
,
661 dn
->node_page
, dn
->ofs_in_node
);
665 f2fs_put_page(parent
, 1);
667 f2fs_put_page(npage
[0], 0);
669 dn
->inode_page
= NULL
;
670 dn
->node_page
= NULL
;
671 if (err
== -ENOENT
) {
673 dn
->max_level
= level
;
674 dn
->ofs_in_node
= offset
[level
];
679 static void truncate_node(struct dnode_of_data
*dn
)
681 struct f2fs_sb_info
*sbi
= F2FS_I_SB(dn
->inode
);
684 get_node_info(sbi
, dn
->nid
, &ni
);
685 f2fs_bug_on(sbi
, ni
.blk_addr
== NULL_ADDR
);
687 /* Deallocate node address */
688 invalidate_blocks(sbi
, ni
.blk_addr
);
689 dec_valid_node_count(sbi
, dn
->inode
, dn
->nid
== dn
->inode
->i_ino
);
690 set_node_addr(sbi
, &ni
, NULL_ADDR
, false);
692 if (dn
->nid
== dn
->inode
->i_ino
) {
693 remove_orphan_inode(sbi
, dn
->nid
);
694 dec_valid_inode_count(sbi
);
695 f2fs_inode_synced(dn
->inode
);
698 clear_node_page_dirty(dn
->node_page
);
699 set_sbi_flag(sbi
, SBI_IS_DIRTY
);
701 f2fs_put_page(dn
->node_page
, 1);
703 invalidate_mapping_pages(NODE_MAPPING(sbi
),
704 dn
->node_page
->index
, dn
->node_page
->index
);
706 dn
->node_page
= NULL
;
707 trace_f2fs_truncate_node(dn
->inode
, dn
->nid
, ni
.blk_addr
);
710 static int truncate_dnode(struct dnode_of_data
*dn
)
717 /* get direct node */
718 page
= get_node_page(F2FS_I_SB(dn
->inode
), dn
->nid
);
719 if (IS_ERR(page
) && PTR_ERR(page
) == -ENOENT
)
721 else if (IS_ERR(page
))
722 return PTR_ERR(page
);
724 /* Make dnode_of_data for parameter */
725 dn
->node_page
= page
;
727 truncate_data_blocks(dn
);
732 static int truncate_nodes(struct dnode_of_data
*dn
, unsigned int nofs
,
735 struct dnode_of_data rdn
= *dn
;
737 struct f2fs_node
*rn
;
739 unsigned int child_nofs
;
744 return NIDS_PER_BLOCK
+ 1;
746 trace_f2fs_truncate_nodes_enter(dn
->inode
, dn
->nid
, dn
->data_blkaddr
);
748 page
= get_node_page(F2FS_I_SB(dn
->inode
), dn
->nid
);
750 trace_f2fs_truncate_nodes_exit(dn
->inode
, PTR_ERR(page
));
751 return PTR_ERR(page
);
754 ra_node_pages(page
, ofs
, NIDS_PER_BLOCK
);
756 rn
= F2FS_NODE(page
);
758 for (i
= ofs
; i
< NIDS_PER_BLOCK
; i
++, freed
++) {
759 child_nid
= le32_to_cpu(rn
->in
.nid
[i
]);
763 ret
= truncate_dnode(&rdn
);
766 if (set_nid(page
, i
, 0, false))
767 dn
->node_changed
= true;
770 child_nofs
= nofs
+ ofs
* (NIDS_PER_BLOCK
+ 1) + 1;
771 for (i
= ofs
; i
< NIDS_PER_BLOCK
; i
++) {
772 child_nid
= le32_to_cpu(rn
->in
.nid
[i
]);
773 if (child_nid
== 0) {
774 child_nofs
+= NIDS_PER_BLOCK
+ 1;
778 ret
= truncate_nodes(&rdn
, child_nofs
, 0, depth
- 1);
779 if (ret
== (NIDS_PER_BLOCK
+ 1)) {
780 if (set_nid(page
, i
, 0, false))
781 dn
->node_changed
= true;
783 } else if (ret
< 0 && ret
!= -ENOENT
) {
791 /* remove current indirect node */
792 dn
->node_page
= page
;
796 f2fs_put_page(page
, 1);
798 trace_f2fs_truncate_nodes_exit(dn
->inode
, freed
);
802 f2fs_put_page(page
, 1);
803 trace_f2fs_truncate_nodes_exit(dn
->inode
, ret
);
807 static int truncate_partial_nodes(struct dnode_of_data
*dn
,
808 struct f2fs_inode
*ri
, int *offset
, int depth
)
810 struct page
*pages
[2];
817 nid
[0] = le32_to_cpu(ri
->i_nid
[offset
[0] - NODE_DIR1_BLOCK
]);
821 /* get indirect nodes in the path */
822 for (i
= 0; i
< idx
+ 1; i
++) {
823 /* reference count'll be increased */
824 pages
[i
] = get_node_page(F2FS_I_SB(dn
->inode
), nid
[i
]);
825 if (IS_ERR(pages
[i
])) {
826 err
= PTR_ERR(pages
[i
]);
830 nid
[i
+ 1] = get_nid(pages
[i
], offset
[i
+ 1], false);
833 ra_node_pages(pages
[idx
], offset
[idx
+ 1], NIDS_PER_BLOCK
);
835 /* free direct nodes linked to a partial indirect node */
836 for (i
= offset
[idx
+ 1]; i
< NIDS_PER_BLOCK
; i
++) {
837 child_nid
= get_nid(pages
[idx
], i
, false);
841 err
= truncate_dnode(dn
);
844 if (set_nid(pages
[idx
], i
, 0, false))
845 dn
->node_changed
= true;
848 if (offset
[idx
+ 1] == 0) {
849 dn
->node_page
= pages
[idx
];
853 f2fs_put_page(pages
[idx
], 1);
859 for (i
= idx
; i
>= 0; i
--)
860 f2fs_put_page(pages
[i
], 1);
862 trace_f2fs_truncate_partial_nodes(dn
->inode
, nid
, depth
, err
);
868 * All the block addresses of data and nodes should be nullified.
870 int truncate_inode_blocks(struct inode
*inode
, pgoff_t from
)
872 struct f2fs_sb_info
*sbi
= F2FS_I_SB(inode
);
873 int err
= 0, cont
= 1;
874 int level
, offset
[4], noffset
[4];
875 unsigned int nofs
= 0;
876 struct f2fs_inode
*ri
;
877 struct dnode_of_data dn
;
880 trace_f2fs_truncate_inode_blocks_enter(inode
, from
);
882 level
= get_node_path(inode
, from
, offset
, noffset
);
886 page
= get_node_page(sbi
, inode
->i_ino
);
888 trace_f2fs_truncate_inode_blocks_exit(inode
, PTR_ERR(page
));
889 return PTR_ERR(page
);
892 set_new_dnode(&dn
, inode
, page
, NULL
, 0);
895 ri
= F2FS_INODE(page
);
903 if (!offset
[level
- 1])
905 err
= truncate_partial_nodes(&dn
, ri
, offset
, level
);
906 if (err
< 0 && err
!= -ENOENT
)
908 nofs
+= 1 + NIDS_PER_BLOCK
;
911 nofs
= 5 + 2 * NIDS_PER_BLOCK
;
912 if (!offset
[level
- 1])
914 err
= truncate_partial_nodes(&dn
, ri
, offset
, level
);
915 if (err
< 0 && err
!= -ENOENT
)
924 dn
.nid
= le32_to_cpu(ri
->i_nid
[offset
[0] - NODE_DIR1_BLOCK
]);
926 case NODE_DIR1_BLOCK
:
927 case NODE_DIR2_BLOCK
:
928 err
= truncate_dnode(&dn
);
931 case NODE_IND1_BLOCK
:
932 case NODE_IND2_BLOCK
:
933 err
= truncate_nodes(&dn
, nofs
, offset
[1], 2);
936 case NODE_DIND_BLOCK
:
937 err
= truncate_nodes(&dn
, nofs
, offset
[1], 3);
944 if (err
< 0 && err
!= -ENOENT
)
946 if (offset
[1] == 0 &&
947 ri
->i_nid
[offset
[0] - NODE_DIR1_BLOCK
]) {
949 BUG_ON(page
->mapping
!= NODE_MAPPING(sbi
));
950 f2fs_wait_on_page_writeback(page
, NODE
, true);
951 ri
->i_nid
[offset
[0] - NODE_DIR1_BLOCK
] = 0;
952 set_page_dirty(page
);
960 f2fs_put_page(page
, 0);
961 trace_f2fs_truncate_inode_blocks_exit(inode
, err
);
962 return err
> 0 ? 0 : err
;
965 int truncate_xattr_node(struct inode
*inode
, struct page
*page
)
967 struct f2fs_sb_info
*sbi
= F2FS_I_SB(inode
);
968 nid_t nid
= F2FS_I(inode
)->i_xattr_nid
;
969 struct dnode_of_data dn
;
975 npage
= get_node_page(sbi
, nid
);
977 return PTR_ERR(npage
);
979 f2fs_i_xnid_write(inode
, 0);
981 set_new_dnode(&dn
, inode
, page
, npage
, nid
);
984 dn
.inode_page_locked
= true;
990 * Caller should grab and release a rwsem by calling f2fs_lock_op() and
993 int remove_inode_page(struct inode
*inode
)
995 struct dnode_of_data dn
;
998 set_new_dnode(&dn
, inode
, NULL
, NULL
, inode
->i_ino
);
999 err
= get_dnode_of_data(&dn
, 0, LOOKUP_NODE
);
1003 err
= truncate_xattr_node(inode
, dn
.inode_page
);
1005 f2fs_put_dnode(&dn
);
1009 /* remove potential inline_data blocks */
1010 if (S_ISREG(inode
->i_mode
) || S_ISDIR(inode
->i_mode
) ||
1011 S_ISLNK(inode
->i_mode
))
1012 truncate_data_blocks_range(&dn
, 1);
1014 /* 0 is possible, after f2fs_new_inode() has failed */
1015 f2fs_bug_on(F2FS_I_SB(inode
),
1016 inode
->i_blocks
!= 0 && inode
->i_blocks
!= 8);
1018 /* will put inode & node pages */
1023 struct page
*new_inode_page(struct inode
*inode
)
1025 struct dnode_of_data dn
;
1027 /* allocate inode page for new inode */
1028 set_new_dnode(&dn
, inode
, NULL
, NULL
, inode
->i_ino
);
1030 /* caller should f2fs_put_page(page, 1); */
1031 return new_node_page(&dn
, 0);
1034 struct page
*new_node_page(struct dnode_of_data
*dn
, unsigned int ofs
)
1036 struct f2fs_sb_info
*sbi
= F2FS_I_SB(dn
->inode
);
1037 struct node_info new_ni
;
1041 if (unlikely(is_inode_flag_set(dn
->inode
, FI_NO_ALLOC
)))
1042 return ERR_PTR(-EPERM
);
1044 page
= f2fs_grab_cache_page(NODE_MAPPING(sbi
), dn
->nid
, false);
1046 return ERR_PTR(-ENOMEM
);
1048 if (unlikely((err
= inc_valid_node_count(sbi
, dn
->inode
, !ofs
))))
1051 #ifdef CONFIG_F2FS_CHECK_FS
1052 get_node_info(sbi
, dn
->nid
, &new_ni
);
1053 f2fs_bug_on(sbi
, new_ni
.blk_addr
!= NULL_ADDR
);
1055 new_ni
.nid
= dn
->nid
;
1056 new_ni
.ino
= dn
->inode
->i_ino
;
1057 new_ni
.blk_addr
= NULL_ADDR
;
1060 set_node_addr(sbi
, &new_ni
, NEW_ADDR
, false);
1062 f2fs_wait_on_page_writeback(page
, NODE
, true);
1063 fill_node_footer(page
, dn
->nid
, dn
->inode
->i_ino
, ofs
, true);
1064 set_cold_node(dn
->inode
, page
);
1065 if (!PageUptodate(page
))
1066 SetPageUptodate(page
);
1067 if (set_page_dirty(page
))
1068 dn
->node_changed
= true;
1070 if (f2fs_has_xattr_block(ofs
))
1071 f2fs_i_xnid_write(dn
->inode
, dn
->nid
);
1074 inc_valid_inode_count(sbi
);
1078 clear_node_page_dirty(page
);
1079 f2fs_put_page(page
, 1);
1080 return ERR_PTR(err
);
1084 * Caller should do after getting the following values.
1085 * 0: f2fs_put_page(page, 0)
1086 * LOCKED_PAGE or error: f2fs_put_page(page, 1)
1088 static int read_node_page(struct page
*page
, int op_flags
)
1090 struct f2fs_sb_info
*sbi
= F2FS_P_SB(page
);
1091 struct node_info ni
;
1092 struct f2fs_io_info fio
= {
1096 .op_flags
= op_flags
,
1098 .encrypted_page
= NULL
,
1101 if (PageUptodate(page
))
1104 get_node_info(sbi
, page
->index
, &ni
);
1106 if (unlikely(ni
.blk_addr
== NULL_ADDR
)) {
1107 ClearPageUptodate(page
);
1111 fio
.new_blkaddr
= fio
.old_blkaddr
= ni
.blk_addr
;
1112 return f2fs_submit_page_bio(&fio
);
1116 * Readahead a node page
1118 void ra_node_page(struct f2fs_sb_info
*sbi
, nid_t nid
)
1125 f2fs_bug_on(sbi
, check_nid_range(sbi
, nid
));
1128 apage
= radix_tree_lookup(&NODE_MAPPING(sbi
)->page_tree
, nid
);
1133 apage
= f2fs_grab_cache_page(NODE_MAPPING(sbi
), nid
, false);
1137 err
= read_node_page(apage
, REQ_RAHEAD
);
1138 f2fs_put_page(apage
, err
? 1 : 0);
1141 static struct page
*__get_node_page(struct f2fs_sb_info
*sbi
, pgoff_t nid
,
1142 struct page
*parent
, int start
)
1148 return ERR_PTR(-ENOENT
);
1149 f2fs_bug_on(sbi
, check_nid_range(sbi
, nid
));
1151 page
= f2fs_grab_cache_page(NODE_MAPPING(sbi
), nid
, false);
1153 return ERR_PTR(-ENOMEM
);
1155 err
= read_node_page(page
, 0);
1157 f2fs_put_page(page
, 1);
1158 return ERR_PTR(err
);
1159 } else if (err
== LOCKED_PAGE
) {
1165 ra_node_pages(parent
, start
+ 1, MAX_RA_NODE
);
1169 if (unlikely(page
->mapping
!= NODE_MAPPING(sbi
))) {
1170 f2fs_put_page(page
, 1);
1174 if (unlikely(!PageUptodate(page
))) {
1179 if (!f2fs_inode_chksum_verify(sbi
, page
)) {
1184 if(unlikely(nid
!= nid_of_node(page
))) {
1185 f2fs_msg(sbi
->sb
, KERN_WARNING
, "inconsistent node block, "
1186 "nid:%lu, node_footer[nid:%u,ino:%u,ofs:%u,cpver:%llu,blkaddr:%u]",
1187 nid
, nid_of_node(page
), ino_of_node(page
),
1188 ofs_of_node(page
), cpver_of_node(page
),
1189 next_blkaddr_of_node(page
));
1192 ClearPageUptodate(page
);
1193 f2fs_put_page(page
, 1);
1194 return ERR_PTR(err
);
1199 struct page
*get_node_page(struct f2fs_sb_info
*sbi
, pgoff_t nid
)
1201 return __get_node_page(sbi
, nid
, NULL
, 0);
1204 struct page
*get_node_page_ra(struct page
*parent
, int start
)
1206 struct f2fs_sb_info
*sbi
= F2FS_P_SB(parent
);
1207 nid_t nid
= get_nid(parent
, start
, false);
1209 return __get_node_page(sbi
, nid
, parent
, start
);
1212 static void flush_inline_data(struct f2fs_sb_info
*sbi
, nid_t ino
)
1214 struct inode
*inode
;
1218 /* should flush inline_data before evict_inode */
1219 inode
= ilookup(sbi
->sb
, ino
);
1223 page
= pagecache_get_page(inode
->i_mapping
, 0, FGP_LOCK
|FGP_NOWAIT
, 0);
1227 if (!PageUptodate(page
))
1230 if (!PageDirty(page
))
1233 if (!clear_page_dirty_for_io(page
))
1236 ret
= f2fs_write_inline_data(inode
, page
);
1237 inode_dec_dirty_pages(inode
);
1238 remove_dirty_inode(inode
);
1240 set_page_dirty(page
);
1242 f2fs_put_page(page
, 1);
1247 void move_node_page(struct page
*node_page
, int gc_type
)
1249 if (gc_type
== FG_GC
) {
1250 struct f2fs_sb_info
*sbi
= F2FS_P_SB(node_page
);
1251 struct writeback_control wbc
= {
1252 .sync_mode
= WB_SYNC_ALL
,
1257 set_page_dirty(node_page
);
1258 f2fs_wait_on_page_writeback(node_page
, NODE
, true);
1260 f2fs_bug_on(sbi
, PageWriteback(node_page
));
1261 if (!clear_page_dirty_for_io(node_page
))
1264 if (NODE_MAPPING(sbi
)->a_ops
->writepage(node_page
, &wbc
))
1265 unlock_page(node_page
);
1268 /* set page dirty and write it */
1269 if (!PageWriteback(node_page
))
1270 set_page_dirty(node_page
);
1273 unlock_page(node_page
);
1275 f2fs_put_page(node_page
, 0);
1278 static struct page
*last_fsync_dnode(struct f2fs_sb_info
*sbi
, nid_t ino
)
1281 struct pagevec pvec
;
1282 struct page
*last_page
= NULL
;
1284 pagevec_init(&pvec
, 0);
1288 while (index
<= end
) {
1290 nr_pages
= pagevec_lookup_tag(&pvec
, NODE_MAPPING(sbi
), &index
,
1291 PAGECACHE_TAG_DIRTY
,
1292 min(end
- index
, (pgoff_t
)PAGEVEC_SIZE
-1) + 1);
1296 for (i
= 0; i
< nr_pages
; i
++) {
1297 struct page
*page
= pvec
.pages
[i
];
1299 if (unlikely(f2fs_cp_error(sbi
))) {
1300 f2fs_put_page(last_page
, 0);
1301 pagevec_release(&pvec
);
1302 return ERR_PTR(-EIO
);
1305 if (!IS_DNODE(page
) || !is_cold_node(page
))
1307 if (ino_of_node(page
) != ino
)
1312 if (unlikely(page
->mapping
!= NODE_MAPPING(sbi
))) {
1317 if (ino_of_node(page
) != ino
)
1318 goto continue_unlock
;
1320 if (!PageDirty(page
)) {
1321 /* someone wrote it for us */
1322 goto continue_unlock
;
1326 f2fs_put_page(last_page
, 0);
1332 pagevec_release(&pvec
);
1338 static int __write_node_page(struct page
*page
, bool atomic
, bool *submitted
,
1339 struct writeback_control
*wbc
, bool do_balance
,
1340 enum iostat_type io_type
)
1342 struct f2fs_sb_info
*sbi
= F2FS_P_SB(page
);
1344 struct node_info ni
;
1345 struct f2fs_io_info fio
= {
1347 .ino
= ino_of_node(page
),
1350 .op_flags
= wbc_to_write_flags(wbc
),
1352 .encrypted_page
= NULL
,
1357 trace_f2fs_writepage(page
, NODE
);
1359 if (unlikely(is_sbi_flag_set(sbi
, SBI_POR_DOING
)))
1361 if (unlikely(f2fs_cp_error(sbi
)))
1364 /* get old block addr of this node page */
1365 nid
= nid_of_node(page
);
1366 f2fs_bug_on(sbi
, page
->index
!= nid
);
1368 if (wbc
->for_reclaim
) {
1369 if (!down_read_trylock(&sbi
->node_write
))
1372 down_read(&sbi
->node_write
);
1375 get_node_info(sbi
, nid
, &ni
);
1377 /* This page is already truncated */
1378 if (unlikely(ni
.blk_addr
== NULL_ADDR
)) {
1379 ClearPageUptodate(page
);
1380 dec_page_count(sbi
, F2FS_DIRTY_NODES
);
1381 up_read(&sbi
->node_write
);
1386 if (atomic
&& !test_opt(sbi
, NOBARRIER
))
1387 fio
.op_flags
|= REQ_PREFLUSH
| REQ_FUA
;
1389 set_page_writeback(page
);
1390 fio
.old_blkaddr
= ni
.blk_addr
;
1391 write_node_page(nid
, &fio
);
1392 set_node_addr(sbi
, &ni
, fio
.new_blkaddr
, is_fsync_dnode(page
));
1393 dec_page_count(sbi
, F2FS_DIRTY_NODES
);
1394 up_read(&sbi
->node_write
);
1396 if (wbc
->for_reclaim
) {
1397 f2fs_submit_merged_write_cond(sbi
, page
->mapping
->host
, 0,
1404 if (unlikely(f2fs_cp_error(sbi
))) {
1405 f2fs_submit_merged_write(sbi
, NODE
);
1409 *submitted
= fio
.submitted
;
1412 f2fs_balance_fs(sbi
, false);
1416 redirty_page_for_writepage(wbc
, page
);
1417 return AOP_WRITEPAGE_ACTIVATE
;
1420 static int f2fs_write_node_page(struct page
*page
,
1421 struct writeback_control
*wbc
)
1423 return __write_node_page(page
, false, NULL
, wbc
, false, FS_NODE_IO
);
1426 int fsync_node_pages(struct f2fs_sb_info
*sbi
, struct inode
*inode
,
1427 struct writeback_control
*wbc
, bool atomic
)
1430 pgoff_t last_idx
= ULONG_MAX
;
1431 struct pagevec pvec
;
1433 struct page
*last_page
= NULL
;
1434 bool marked
= false;
1435 nid_t ino
= inode
->i_ino
;
1438 last_page
= last_fsync_dnode(sbi
, ino
);
1439 if (IS_ERR_OR_NULL(last_page
))
1440 return PTR_ERR_OR_ZERO(last_page
);
1443 pagevec_init(&pvec
, 0);
1447 while (index
<= end
) {
1449 nr_pages
= pagevec_lookup_tag(&pvec
, NODE_MAPPING(sbi
), &index
,
1450 PAGECACHE_TAG_DIRTY
,
1451 min(end
- index
, (pgoff_t
)PAGEVEC_SIZE
-1) + 1);
1455 for (i
= 0; i
< nr_pages
; i
++) {
1456 struct page
*page
= pvec
.pages
[i
];
1457 bool submitted
= false;
1459 if (unlikely(f2fs_cp_error(sbi
))) {
1460 f2fs_put_page(last_page
, 0);
1461 pagevec_release(&pvec
);
1466 if (!IS_DNODE(page
) || !is_cold_node(page
))
1468 if (ino_of_node(page
) != ino
)
1473 if (unlikely(page
->mapping
!= NODE_MAPPING(sbi
))) {
1478 if (ino_of_node(page
) != ino
)
1479 goto continue_unlock
;
1481 if (!PageDirty(page
) && page
!= last_page
) {
1482 /* someone wrote it for us */
1483 goto continue_unlock
;
1486 f2fs_wait_on_page_writeback(page
, NODE
, true);
1487 BUG_ON(PageWriteback(page
));
1489 set_fsync_mark(page
, 0);
1490 set_dentry_mark(page
, 0);
1492 if (!atomic
|| page
== last_page
) {
1493 set_fsync_mark(page
, 1);
1494 if (IS_INODE(page
)) {
1495 if (is_inode_flag_set(inode
,
1497 update_inode(inode
, page
);
1498 set_dentry_mark(page
,
1499 need_dentry_mark(sbi
, ino
));
1501 /* may be written by other thread */
1502 if (!PageDirty(page
))
1503 set_page_dirty(page
);
1506 if (!clear_page_dirty_for_io(page
))
1507 goto continue_unlock
;
1509 ret
= __write_node_page(page
, atomic
&&
1511 &submitted
, wbc
, true,
1515 f2fs_put_page(last_page
, 0);
1517 } else if (submitted
) {
1518 last_idx
= page
->index
;
1521 if (page
== last_page
) {
1522 f2fs_put_page(page
, 0);
1527 pagevec_release(&pvec
);
1533 if (!ret
&& atomic
&& !marked
) {
1534 f2fs_msg(sbi
->sb
, KERN_DEBUG
,
1535 "Retry to write fsync mark: ino=%u, idx=%lx",
1536 ino
, last_page
->index
);
1537 lock_page(last_page
);
1538 f2fs_wait_on_page_writeback(last_page
, NODE
, true);
1539 set_page_dirty(last_page
);
1540 unlock_page(last_page
);
1544 if (last_idx
!= ULONG_MAX
)
1545 f2fs_submit_merged_write_cond(sbi
, NULL
, ino
, last_idx
, NODE
);
1546 return ret
? -EIO
: 0;
1549 int sync_node_pages(struct f2fs_sb_info
*sbi
, struct writeback_control
*wbc
,
1550 bool do_balance
, enum iostat_type io_type
)
1553 struct pagevec pvec
;
1558 pagevec_init(&pvec
, 0);
1564 while (index
<= end
) {
1566 nr_pages
= pagevec_lookup_tag(&pvec
, NODE_MAPPING(sbi
), &index
,
1567 PAGECACHE_TAG_DIRTY
,
1568 min(end
- index
, (pgoff_t
)PAGEVEC_SIZE
-1) + 1);
1572 for (i
= 0; i
< nr_pages
; i
++) {
1573 struct page
*page
= pvec
.pages
[i
];
1574 bool submitted
= false;
1576 if (unlikely(f2fs_cp_error(sbi
))) {
1577 pagevec_release(&pvec
);
1583 * flushing sequence with step:
1588 if (step
== 0 && IS_DNODE(page
))
1590 if (step
== 1 && (!IS_DNODE(page
) ||
1591 is_cold_node(page
)))
1593 if (step
== 2 && (!IS_DNODE(page
) ||
1594 !is_cold_node(page
)))
1597 if (!trylock_page(page
))
1600 if (unlikely(page
->mapping
!= NODE_MAPPING(sbi
))) {
1606 if (!PageDirty(page
)) {
1607 /* someone wrote it for us */
1608 goto continue_unlock
;
1611 /* flush inline_data */
1612 if (is_inline_node(page
)) {
1613 clear_inline_node(page
);
1615 flush_inline_data(sbi
, ino_of_node(page
));
1619 f2fs_wait_on_page_writeback(page
, NODE
, true);
1621 BUG_ON(PageWriteback(page
));
1622 if (!clear_page_dirty_for_io(page
))
1623 goto continue_unlock
;
1625 set_fsync_mark(page
, 0);
1626 set_dentry_mark(page
, 0);
1628 ret
= __write_node_page(page
, false, &submitted
,
1629 wbc
, do_balance
, io_type
);
1635 if (--wbc
->nr_to_write
== 0)
1638 pagevec_release(&pvec
);
1641 if (wbc
->nr_to_write
== 0) {
1653 f2fs_submit_merged_write(sbi
, NODE
);
1657 int wait_on_node_pages_writeback(struct f2fs_sb_info
*sbi
, nid_t ino
)
1659 pgoff_t index
= 0, end
= ULONG_MAX
;
1660 struct pagevec pvec
;
1663 pagevec_init(&pvec
, 0);
1665 while (index
<= end
) {
1667 nr_pages
= pagevec_lookup_tag(&pvec
, NODE_MAPPING(sbi
), &index
,
1668 PAGECACHE_TAG_WRITEBACK
,
1669 min(end
- index
, (pgoff_t
)PAGEVEC_SIZE
-1) + 1);
1673 for (i
= 0; i
< nr_pages
; i
++) {
1674 struct page
*page
= pvec
.pages
[i
];
1676 /* until radix tree lookup accepts end_index */
1677 if (unlikely(page
->index
> end
))
1680 if (ino
&& ino_of_node(page
) == ino
) {
1681 f2fs_wait_on_page_writeback(page
, NODE
, true);
1682 if (TestClearPageError(page
))
1686 pagevec_release(&pvec
);
1690 ret2
= filemap_check_errors(NODE_MAPPING(sbi
));
1696 static int f2fs_write_node_pages(struct address_space
*mapping
,
1697 struct writeback_control
*wbc
)
1699 struct f2fs_sb_info
*sbi
= F2FS_M_SB(mapping
);
1700 struct blk_plug plug
;
1703 if (unlikely(is_sbi_flag_set(sbi
, SBI_POR_DOING
)))
1706 /* balancing f2fs's metadata in background */
1707 f2fs_balance_fs_bg(sbi
);
1709 /* collect a number of dirty node pages and write together */
1710 if (get_pages(sbi
, F2FS_DIRTY_NODES
) < nr_pages_to_skip(sbi
, NODE
))
1713 trace_f2fs_writepages(mapping
->host
, wbc
, NODE
);
1715 diff
= nr_pages_to_write(sbi
, NODE
, wbc
);
1716 wbc
->sync_mode
= WB_SYNC_NONE
;
1717 blk_start_plug(&plug
);
1718 sync_node_pages(sbi
, wbc
, true, FS_NODE_IO
);
1719 blk_finish_plug(&plug
);
1720 wbc
->nr_to_write
= max((long)0, wbc
->nr_to_write
- diff
);
1724 wbc
->pages_skipped
+= get_pages(sbi
, F2FS_DIRTY_NODES
);
1725 trace_f2fs_writepages(mapping
->host
, wbc
, NODE
);
1729 static int f2fs_set_node_page_dirty(struct page
*page
)
1731 trace_f2fs_set_page_dirty(page
, NODE
);
1733 if (!PageUptodate(page
))
1734 SetPageUptodate(page
);
1735 if (!PageDirty(page
)) {
1736 f2fs_set_page_dirty_nobuffers(page
);
1737 inc_page_count(F2FS_P_SB(page
), F2FS_DIRTY_NODES
);
1738 SetPagePrivate(page
);
1739 f2fs_trace_pid(page
);
1746 * Structure of the f2fs node operations
1748 const struct address_space_operations f2fs_node_aops
= {
1749 .writepage
= f2fs_write_node_page
,
1750 .writepages
= f2fs_write_node_pages
,
1751 .set_page_dirty
= f2fs_set_node_page_dirty
,
1752 .invalidatepage
= f2fs_invalidate_page
,
1753 .releasepage
= f2fs_release_page
,
1754 #ifdef CONFIG_MIGRATION
1755 .migratepage
= f2fs_migrate_page
,
1759 static struct free_nid
*__lookup_free_nid_list(struct f2fs_nm_info
*nm_i
,
1762 return radix_tree_lookup(&nm_i
->free_nid_root
, n
);
1765 static int __insert_free_nid(struct f2fs_sb_info
*sbi
,
1766 struct free_nid
*i
, enum nid_state state
, bool new)
1768 struct f2fs_nm_info
*nm_i
= NM_I(sbi
);
1771 int err
= radix_tree_insert(&nm_i
->free_nid_root
, i
->nid
, i
);
1776 f2fs_bug_on(sbi
, state
!= i
->state
);
1777 nm_i
->nid_cnt
[state
]++;
1778 if (state
== FREE_NID
)
1779 list_add_tail(&i
->list
, &nm_i
->free_nid_list
);
1783 static void __remove_free_nid(struct f2fs_sb_info
*sbi
,
1784 struct free_nid
*i
, enum nid_state state
, bool reuse
)
1786 struct f2fs_nm_info
*nm_i
= NM_I(sbi
);
1788 f2fs_bug_on(sbi
, state
!= i
->state
);
1789 nm_i
->nid_cnt
[state
]--;
1790 if (state
== FREE_NID
)
1793 radix_tree_delete(&nm_i
->free_nid_root
, i
->nid
);
1796 /* return if the nid is recognized as free */
1797 static bool add_free_nid(struct f2fs_sb_info
*sbi
, nid_t nid
, bool build
)
1799 struct f2fs_nm_info
*nm_i
= NM_I(sbi
);
1800 struct free_nid
*i
, *e
;
1801 struct nat_entry
*ne
;
1805 /* 0 nid should not be used */
1806 if (unlikely(nid
== 0))
1809 i
= f2fs_kmem_cache_alloc(free_nid_slab
, GFP_NOFS
);
1811 i
->state
= FREE_NID
;
1813 if (radix_tree_preload(GFP_NOFS
))
1816 spin_lock(&nm_i
->nid_list_lock
);
1824 * - __insert_nid_to_list(PREALLOC_NID)
1825 * - f2fs_balance_fs_bg
1827 * - __build_free_nids
1830 * - __lookup_nat_cache
1832 * - init_inode_metadata
1837 * - __remove_nid_from_list(PREALLOC_NID)
1838 * - __insert_nid_to_list(FREE_NID)
1840 ne
= __lookup_nat_cache(nm_i
, nid
);
1841 if (ne
&& (!get_nat_flag(ne
, IS_CHECKPOINTED
) ||
1842 nat_get_blkaddr(ne
) != NULL_ADDR
))
1845 e
= __lookup_free_nid_list(nm_i
, nid
);
1847 if (e
->state
== FREE_NID
)
1853 err
= __insert_free_nid(sbi
, i
, FREE_NID
, true);
1855 spin_unlock(&nm_i
->nid_list_lock
);
1856 radix_tree_preload_end();
1859 kmem_cache_free(free_nid_slab
, i
);
1863 static void remove_free_nid(struct f2fs_sb_info
*sbi
, nid_t nid
)
1865 struct f2fs_nm_info
*nm_i
= NM_I(sbi
);
1867 bool need_free
= false;
1869 spin_lock(&nm_i
->nid_list_lock
);
1870 i
= __lookup_free_nid_list(nm_i
, nid
);
1871 if (i
&& i
->state
== FREE_NID
) {
1872 __remove_free_nid(sbi
, i
, FREE_NID
, false);
1875 spin_unlock(&nm_i
->nid_list_lock
);
1878 kmem_cache_free(free_nid_slab
, i
);
1881 static void update_free_nid_bitmap(struct f2fs_sb_info
*sbi
, nid_t nid
,
1882 bool set
, bool build
)
1884 struct f2fs_nm_info
*nm_i
= NM_I(sbi
);
1885 unsigned int nat_ofs
= NAT_BLOCK_OFFSET(nid
);
1886 unsigned int nid_ofs
= nid
- START_NID(nid
);
1888 if (!test_bit_le(nat_ofs
, nm_i
->nat_block_bitmap
))
1892 __set_bit_le(nid_ofs
, nm_i
->free_nid_bitmap
[nat_ofs
]);
1894 __clear_bit_le(nid_ofs
, nm_i
->free_nid_bitmap
[nat_ofs
]);
1897 nm_i
->free_nid_count
[nat_ofs
]++;
1899 nm_i
->free_nid_count
[nat_ofs
]--;
1902 static void scan_nat_page(struct f2fs_sb_info
*sbi
,
1903 struct page
*nat_page
, nid_t start_nid
)
1905 struct f2fs_nm_info
*nm_i
= NM_I(sbi
);
1906 struct f2fs_nat_block
*nat_blk
= page_address(nat_page
);
1908 unsigned int nat_ofs
= NAT_BLOCK_OFFSET(start_nid
);
1911 if (test_bit_le(nat_ofs
, nm_i
->nat_block_bitmap
))
1914 __set_bit_le(nat_ofs
, nm_i
->nat_block_bitmap
);
1916 i
= start_nid
% NAT_ENTRY_PER_BLOCK
;
1918 for (; i
< NAT_ENTRY_PER_BLOCK
; i
++, start_nid
++) {
1921 if (unlikely(start_nid
>= nm_i
->max_nid
))
1924 blk_addr
= le32_to_cpu(nat_blk
->entries
[i
].block_addr
);
1925 f2fs_bug_on(sbi
, blk_addr
== NEW_ADDR
);
1926 if (blk_addr
== NULL_ADDR
)
1927 freed
= add_free_nid(sbi
, start_nid
, true);
1928 spin_lock(&NM_I(sbi
)->nid_list_lock
);
1929 update_free_nid_bitmap(sbi
, start_nid
, freed
, true);
1930 spin_unlock(&NM_I(sbi
)->nid_list_lock
);
1934 static void scan_free_nid_bits(struct f2fs_sb_info
*sbi
)
1936 struct f2fs_nm_info
*nm_i
= NM_I(sbi
);
1937 struct curseg_info
*curseg
= CURSEG_I(sbi
, CURSEG_HOT_DATA
);
1938 struct f2fs_journal
*journal
= curseg
->journal
;
1939 unsigned int i
, idx
;
1941 down_read(&nm_i
->nat_tree_lock
);
1943 for (i
= 0; i
< nm_i
->nat_blocks
; i
++) {
1944 if (!test_bit_le(i
, nm_i
->nat_block_bitmap
))
1946 if (!nm_i
->free_nid_count
[i
])
1948 for (idx
= 0; idx
< NAT_ENTRY_PER_BLOCK
; idx
++) {
1951 if (!test_bit_le(idx
, nm_i
->free_nid_bitmap
[i
]))
1954 nid
= i
* NAT_ENTRY_PER_BLOCK
+ idx
;
1955 add_free_nid(sbi
, nid
, true);
1957 if (nm_i
->nid_cnt
[FREE_NID
] >= MAX_FREE_NIDS
)
1962 down_read(&curseg
->journal_rwsem
);
1963 for (i
= 0; i
< nats_in_cursum(journal
); i
++) {
1967 addr
= le32_to_cpu(nat_in_journal(journal
, i
).block_addr
);
1968 nid
= le32_to_cpu(nid_in_journal(journal
, i
));
1969 if (addr
== NULL_ADDR
)
1970 add_free_nid(sbi
, nid
, true);
1972 remove_free_nid(sbi
, nid
);
1974 up_read(&curseg
->journal_rwsem
);
1975 up_read(&nm_i
->nat_tree_lock
);
1978 static void __build_free_nids(struct f2fs_sb_info
*sbi
, bool sync
, bool mount
)
1980 struct f2fs_nm_info
*nm_i
= NM_I(sbi
);
1981 struct curseg_info
*curseg
= CURSEG_I(sbi
, CURSEG_HOT_DATA
);
1982 struct f2fs_journal
*journal
= curseg
->journal
;
1984 nid_t nid
= nm_i
->next_scan_nid
;
1986 if (unlikely(nid
>= nm_i
->max_nid
))
1989 /* Enough entries */
1990 if (nm_i
->nid_cnt
[FREE_NID
] >= NAT_ENTRY_PER_BLOCK
)
1993 if (!sync
&& !available_free_memory(sbi
, FREE_NIDS
))
1997 /* try to find free nids in free_nid_bitmap */
1998 scan_free_nid_bits(sbi
);
2000 if (nm_i
->nid_cnt
[FREE_NID
])
2004 /* readahead nat pages to be scanned */
2005 ra_meta_pages(sbi
, NAT_BLOCK_OFFSET(nid
), FREE_NID_PAGES
,
2008 down_read(&nm_i
->nat_tree_lock
);
2011 struct page
*page
= get_current_nat_page(sbi
, nid
);
2013 scan_nat_page(sbi
, page
, nid
);
2014 f2fs_put_page(page
, 1);
2016 nid
+= (NAT_ENTRY_PER_BLOCK
- (nid
% NAT_ENTRY_PER_BLOCK
));
2017 if (unlikely(nid
>= nm_i
->max_nid
))
2020 if (++i
>= FREE_NID_PAGES
)
2024 /* go to the next free nat pages to find free nids abundantly */
2025 nm_i
->next_scan_nid
= nid
;
2027 /* find free nids from current sum_pages */
2028 down_read(&curseg
->journal_rwsem
);
2029 for (i
= 0; i
< nats_in_cursum(journal
); i
++) {
2032 addr
= le32_to_cpu(nat_in_journal(journal
, i
).block_addr
);
2033 nid
= le32_to_cpu(nid_in_journal(journal
, i
));
2034 if (addr
== NULL_ADDR
)
2035 add_free_nid(sbi
, nid
, true);
2037 remove_free_nid(sbi
, nid
);
2039 up_read(&curseg
->journal_rwsem
);
2040 up_read(&nm_i
->nat_tree_lock
);
2042 ra_meta_pages(sbi
, NAT_BLOCK_OFFSET(nm_i
->next_scan_nid
),
2043 nm_i
->ra_nid_pages
, META_NAT
, false);
2046 void build_free_nids(struct f2fs_sb_info
*sbi
, bool sync
, bool mount
)
2048 mutex_lock(&NM_I(sbi
)->build_lock
);
2049 __build_free_nids(sbi
, sync
, mount
);
2050 mutex_unlock(&NM_I(sbi
)->build_lock
);
2054 * If this function returns success, caller can obtain a new nid
2055 * from second parameter of this function.
2056 * The returned nid could be used ino as well as nid when inode is created.
2058 bool alloc_nid(struct f2fs_sb_info
*sbi
, nid_t
*nid
)
2060 struct f2fs_nm_info
*nm_i
= NM_I(sbi
);
2061 struct free_nid
*i
= NULL
;
2063 #ifdef CONFIG_F2FS_FAULT_INJECTION
2064 if (time_to_inject(sbi
, FAULT_ALLOC_NID
)) {
2065 f2fs_show_injection_info(FAULT_ALLOC_NID
);
2069 spin_lock(&nm_i
->nid_list_lock
);
2071 if (unlikely(nm_i
->available_nids
== 0)) {
2072 spin_unlock(&nm_i
->nid_list_lock
);
2076 /* We should not use stale free nids created by build_free_nids */
2077 if (nm_i
->nid_cnt
[FREE_NID
] && !on_build_free_nids(nm_i
)) {
2078 f2fs_bug_on(sbi
, list_empty(&nm_i
->free_nid_list
));
2079 i
= list_first_entry(&nm_i
->free_nid_list
,
2080 struct free_nid
, list
);
2083 __remove_free_nid(sbi
, i
, FREE_NID
, true);
2084 i
->state
= PREALLOC_NID
;
2085 __insert_free_nid(sbi
, i
, PREALLOC_NID
, false);
2086 nm_i
->available_nids
--;
2088 update_free_nid_bitmap(sbi
, *nid
, false, false);
2090 spin_unlock(&nm_i
->nid_list_lock
);
2093 spin_unlock(&nm_i
->nid_list_lock
);
2095 /* Let's scan nat pages and its caches to get free nids */
2096 build_free_nids(sbi
, true, false);
2101 * alloc_nid() should be called prior to this function.
2103 void alloc_nid_done(struct f2fs_sb_info
*sbi
, nid_t nid
)
2105 struct f2fs_nm_info
*nm_i
= NM_I(sbi
);
2108 spin_lock(&nm_i
->nid_list_lock
);
2109 i
= __lookup_free_nid_list(nm_i
, nid
);
2110 f2fs_bug_on(sbi
, !i
);
2111 __remove_free_nid(sbi
, i
, PREALLOC_NID
, false);
2112 spin_unlock(&nm_i
->nid_list_lock
);
2114 kmem_cache_free(free_nid_slab
, i
);
2118 * alloc_nid() should be called prior to this function.
2120 void alloc_nid_failed(struct f2fs_sb_info
*sbi
, nid_t nid
)
2122 struct f2fs_nm_info
*nm_i
= NM_I(sbi
);
2124 bool need_free
= false;
2129 spin_lock(&nm_i
->nid_list_lock
);
2130 i
= __lookup_free_nid_list(nm_i
, nid
);
2131 f2fs_bug_on(sbi
, !i
);
2133 if (!available_free_memory(sbi
, FREE_NIDS
)) {
2134 __remove_free_nid(sbi
, i
, PREALLOC_NID
, false);
2137 __remove_free_nid(sbi
, i
, PREALLOC_NID
, true);
2138 i
->state
= FREE_NID
;
2139 __insert_free_nid(sbi
, i
, FREE_NID
, false);
2142 nm_i
->available_nids
++;
2144 update_free_nid_bitmap(sbi
, nid
, true, false);
2146 spin_unlock(&nm_i
->nid_list_lock
);
2149 kmem_cache_free(free_nid_slab
, i
);
2152 int try_to_free_nids(struct f2fs_sb_info
*sbi
, int nr_shrink
)
2154 struct f2fs_nm_info
*nm_i
= NM_I(sbi
);
2155 struct free_nid
*i
, *next
;
2158 if (nm_i
->nid_cnt
[FREE_NID
] <= MAX_FREE_NIDS
)
2161 if (!mutex_trylock(&nm_i
->build_lock
))
2164 spin_lock(&nm_i
->nid_list_lock
);
2165 list_for_each_entry_safe(i
, next
, &nm_i
->free_nid_list
, list
) {
2166 if (nr_shrink
<= 0 ||
2167 nm_i
->nid_cnt
[FREE_NID
] <= MAX_FREE_NIDS
)
2170 __remove_free_nid(sbi
, i
, FREE_NID
, false);
2171 kmem_cache_free(free_nid_slab
, i
);
2174 spin_unlock(&nm_i
->nid_list_lock
);
2175 mutex_unlock(&nm_i
->build_lock
);
2177 return nr
- nr_shrink
;
2180 void recover_inline_xattr(struct inode
*inode
, struct page
*page
)
2182 void *src_addr
, *dst_addr
;
2185 struct f2fs_inode
*ri
;
2187 ipage
= get_node_page(F2FS_I_SB(inode
), inode
->i_ino
);
2188 f2fs_bug_on(F2FS_I_SB(inode
), IS_ERR(ipage
));
2190 ri
= F2FS_INODE(page
);
2191 if (!(ri
->i_inline
& F2FS_INLINE_XATTR
)) {
2192 clear_inode_flag(inode
, FI_INLINE_XATTR
);
2196 dst_addr
= inline_xattr_addr(ipage
);
2197 src_addr
= inline_xattr_addr(page
);
2198 inline_size
= inline_xattr_size(inode
);
2200 f2fs_wait_on_page_writeback(ipage
, NODE
, true);
2201 memcpy(dst_addr
, src_addr
, inline_size
);
2203 update_inode(inode
, ipage
);
2204 f2fs_put_page(ipage
, 1);
2207 int recover_xattr_data(struct inode
*inode
, struct page
*page
, block_t blkaddr
)
2209 struct f2fs_sb_info
*sbi
= F2FS_I_SB(inode
);
2210 nid_t prev_xnid
= F2FS_I(inode
)->i_xattr_nid
;
2212 struct dnode_of_data dn
;
2213 struct node_info ni
;
2219 /* 1: invalidate the previous xattr nid */
2220 get_node_info(sbi
, prev_xnid
, &ni
);
2221 f2fs_bug_on(sbi
, ni
.blk_addr
== NULL_ADDR
);
2222 invalidate_blocks(sbi
, ni
.blk_addr
);
2223 dec_valid_node_count(sbi
, inode
, false);
2224 set_node_addr(sbi
, &ni
, NULL_ADDR
, false);
2227 /* 2: update xattr nid in inode */
2228 if (!alloc_nid(sbi
, &new_xnid
))
2231 set_new_dnode(&dn
, inode
, NULL
, NULL
, new_xnid
);
2232 xpage
= new_node_page(&dn
, XATTR_NODE_OFFSET
);
2233 if (IS_ERR(xpage
)) {
2234 alloc_nid_failed(sbi
, new_xnid
);
2235 return PTR_ERR(xpage
);
2238 alloc_nid_done(sbi
, new_xnid
);
2239 update_inode_page(inode
);
2241 /* 3: update and set xattr node page dirty */
2242 memcpy(F2FS_NODE(xpage
), F2FS_NODE(page
), VALID_XATTR_BLOCK_SIZE
);
2244 set_page_dirty(xpage
);
2245 f2fs_put_page(xpage
, 1);
2250 int recover_inode_page(struct f2fs_sb_info
*sbi
, struct page
*page
)
2252 struct f2fs_inode
*src
, *dst
;
2253 nid_t ino
= ino_of_node(page
);
2254 struct node_info old_ni
, new_ni
;
2257 get_node_info(sbi
, ino
, &old_ni
);
2259 if (unlikely(old_ni
.blk_addr
!= NULL_ADDR
))
2262 ipage
= f2fs_grab_cache_page(NODE_MAPPING(sbi
), ino
, false);
2264 congestion_wait(BLK_RW_ASYNC
, HZ
/50);
2268 /* Should not use this inode from free nid list */
2269 remove_free_nid(sbi
, ino
);
2271 if (!PageUptodate(ipage
))
2272 SetPageUptodate(ipage
);
2273 fill_node_footer(ipage
, ino
, ino
, 0, true);
2275 src
= F2FS_INODE(page
);
2276 dst
= F2FS_INODE(ipage
);
2278 memcpy(dst
, src
, (unsigned long)&src
->i_ext
- (unsigned long)src
);
2280 dst
->i_blocks
= cpu_to_le64(1);
2281 dst
->i_links
= cpu_to_le32(1);
2282 dst
->i_xattr_nid
= 0;
2283 dst
->i_inline
= src
->i_inline
& (F2FS_INLINE_XATTR
| F2FS_EXTRA_ATTR
);
2284 if (dst
->i_inline
& F2FS_EXTRA_ATTR
) {
2285 dst
->i_extra_isize
= src
->i_extra_isize
;
2286 if (f2fs_sb_has_project_quota(sbi
->sb
) &&
2287 F2FS_FITS_IN_INODE(src
, le16_to_cpu(src
->i_extra_isize
),
2289 dst
->i_projid
= src
->i_projid
;
2295 if (unlikely(inc_valid_node_count(sbi
, NULL
, true)))
2297 set_node_addr(sbi
, &new_ni
, NEW_ADDR
, false);
2298 inc_valid_inode_count(sbi
);
2299 set_page_dirty(ipage
);
2300 f2fs_put_page(ipage
, 1);
2304 int restore_node_summary(struct f2fs_sb_info
*sbi
,
2305 unsigned int segno
, struct f2fs_summary_block
*sum
)
2307 struct f2fs_node
*rn
;
2308 struct f2fs_summary
*sum_entry
;
2310 int i
, idx
, last_offset
, nrpages
;
2312 /* scan the node segment */
2313 last_offset
= sbi
->blocks_per_seg
;
2314 addr
= START_BLOCK(sbi
, segno
);
2315 sum_entry
= &sum
->entries
[0];
2317 for (i
= 0; i
< last_offset
; i
+= nrpages
, addr
+= nrpages
) {
2318 nrpages
= min(last_offset
- i
, BIO_MAX_PAGES
);
2320 /* readahead node pages */
2321 ra_meta_pages(sbi
, addr
, nrpages
, META_POR
, true);
2323 for (idx
= addr
; idx
< addr
+ nrpages
; idx
++) {
2324 struct page
*page
= get_tmp_page(sbi
, idx
);
2326 rn
= F2FS_NODE(page
);
2327 sum_entry
->nid
= rn
->footer
.nid
;
2328 sum_entry
->version
= 0;
2329 sum_entry
->ofs_in_node
= 0;
2331 f2fs_put_page(page
, 1);
2334 invalidate_mapping_pages(META_MAPPING(sbi
), addr
,
2340 static void remove_nats_in_journal(struct f2fs_sb_info
*sbi
)
2342 struct f2fs_nm_info
*nm_i
= NM_I(sbi
);
2343 struct curseg_info
*curseg
= CURSEG_I(sbi
, CURSEG_HOT_DATA
);
2344 struct f2fs_journal
*journal
= curseg
->journal
;
2347 down_write(&curseg
->journal_rwsem
);
2348 for (i
= 0; i
< nats_in_cursum(journal
); i
++) {
2349 struct nat_entry
*ne
;
2350 struct f2fs_nat_entry raw_ne
;
2351 nid_t nid
= le32_to_cpu(nid_in_journal(journal
, i
));
2353 raw_ne
= nat_in_journal(journal
, i
);
2355 ne
= __lookup_nat_cache(nm_i
, nid
);
2357 ne
= grab_nat_entry(nm_i
, nid
, true);
2358 node_info_from_raw_nat(&ne
->ni
, &raw_ne
);
2362 * if a free nat in journal has not been used after last
2363 * checkpoint, we should remove it from available nids,
2364 * since later we will add it again.
2366 if (!get_nat_flag(ne
, IS_DIRTY
) &&
2367 le32_to_cpu(raw_ne
.block_addr
) == NULL_ADDR
) {
2368 spin_lock(&nm_i
->nid_list_lock
);
2369 nm_i
->available_nids
--;
2370 spin_unlock(&nm_i
->nid_list_lock
);
2373 __set_nat_cache_dirty(nm_i
, ne
);
2375 update_nats_in_cursum(journal
, -i
);
2376 up_write(&curseg
->journal_rwsem
);
2379 static void __adjust_nat_entry_set(struct nat_entry_set
*nes
,
2380 struct list_head
*head
, int max
)
2382 struct nat_entry_set
*cur
;
2384 if (nes
->entry_cnt
>= max
)
2387 list_for_each_entry(cur
, head
, set_list
) {
2388 if (cur
->entry_cnt
>= nes
->entry_cnt
) {
2389 list_add(&nes
->set_list
, cur
->set_list
.prev
);
2394 list_add_tail(&nes
->set_list
, head
);
2397 static void __update_nat_bits(struct f2fs_sb_info
*sbi
, nid_t start_nid
,
2400 struct f2fs_nm_info
*nm_i
= NM_I(sbi
);
2401 unsigned int nat_index
= start_nid
/ NAT_ENTRY_PER_BLOCK
;
2402 struct f2fs_nat_block
*nat_blk
= page_address(page
);
2406 if (!enabled_nat_bits(sbi
, NULL
))
2409 for (i
= 0; i
< NAT_ENTRY_PER_BLOCK
; i
++) {
2410 if (start_nid
== 0 && i
== 0)
2412 if (nat_blk
->entries
[i
].block_addr
)
2416 __set_bit_le(nat_index
, nm_i
->empty_nat_bits
);
2417 __clear_bit_le(nat_index
, nm_i
->full_nat_bits
);
2421 __clear_bit_le(nat_index
, nm_i
->empty_nat_bits
);
2422 if (valid
== NAT_ENTRY_PER_BLOCK
)
2423 __set_bit_le(nat_index
, nm_i
->full_nat_bits
);
2425 __clear_bit_le(nat_index
, nm_i
->full_nat_bits
);
2428 static void __flush_nat_entry_set(struct f2fs_sb_info
*sbi
,
2429 struct nat_entry_set
*set
, struct cp_control
*cpc
)
2431 struct curseg_info
*curseg
= CURSEG_I(sbi
, CURSEG_HOT_DATA
);
2432 struct f2fs_journal
*journal
= curseg
->journal
;
2433 nid_t start_nid
= set
->set
* NAT_ENTRY_PER_BLOCK
;
2434 bool to_journal
= true;
2435 struct f2fs_nat_block
*nat_blk
;
2436 struct nat_entry
*ne
, *cur
;
2437 struct page
*page
= NULL
;
2440 * there are two steps to flush nat entries:
2441 * #1, flush nat entries to journal in current hot data summary block.
2442 * #2, flush nat entries to nat page.
2444 if (enabled_nat_bits(sbi
, cpc
) ||
2445 !__has_cursum_space(journal
, set
->entry_cnt
, NAT_JOURNAL
))
2449 down_write(&curseg
->journal_rwsem
);
2451 page
= get_next_nat_page(sbi
, start_nid
);
2452 nat_blk
= page_address(page
);
2453 f2fs_bug_on(sbi
, !nat_blk
);
2456 /* flush dirty nats in nat entry set */
2457 list_for_each_entry_safe(ne
, cur
, &set
->entry_list
, list
) {
2458 struct f2fs_nat_entry
*raw_ne
;
2459 nid_t nid
= nat_get_nid(ne
);
2462 f2fs_bug_on(sbi
, nat_get_blkaddr(ne
) == NEW_ADDR
);
2465 offset
= lookup_journal_in_cursum(journal
,
2466 NAT_JOURNAL
, nid
, 1);
2467 f2fs_bug_on(sbi
, offset
< 0);
2468 raw_ne
= &nat_in_journal(journal
, offset
);
2469 nid_in_journal(journal
, offset
) = cpu_to_le32(nid
);
2471 raw_ne
= &nat_blk
->entries
[nid
- start_nid
];
2473 raw_nat_from_node_info(raw_ne
, &ne
->ni
);
2475 __clear_nat_cache_dirty(NM_I(sbi
), set
, ne
);
2476 if (nat_get_blkaddr(ne
) == NULL_ADDR
) {
2477 add_free_nid(sbi
, nid
, false);
2478 spin_lock(&NM_I(sbi
)->nid_list_lock
);
2479 NM_I(sbi
)->available_nids
++;
2480 update_free_nid_bitmap(sbi
, nid
, true, false);
2481 spin_unlock(&NM_I(sbi
)->nid_list_lock
);
2483 spin_lock(&NM_I(sbi
)->nid_list_lock
);
2484 update_free_nid_bitmap(sbi
, nid
, false, false);
2485 spin_unlock(&NM_I(sbi
)->nid_list_lock
);
2490 up_write(&curseg
->journal_rwsem
);
2492 __update_nat_bits(sbi
, start_nid
, page
);
2493 f2fs_put_page(page
, 1);
2496 /* Allow dirty nats by node block allocation in write_begin */
2497 if (!set
->entry_cnt
) {
2498 radix_tree_delete(&NM_I(sbi
)->nat_set_root
, set
->set
);
2499 kmem_cache_free(nat_entry_set_slab
, set
);
2504 * This function is called during the checkpointing process.
2506 void flush_nat_entries(struct f2fs_sb_info
*sbi
, struct cp_control
*cpc
)
2508 struct f2fs_nm_info
*nm_i
= NM_I(sbi
);
2509 struct curseg_info
*curseg
= CURSEG_I(sbi
, CURSEG_HOT_DATA
);
2510 struct f2fs_journal
*journal
= curseg
->journal
;
2511 struct nat_entry_set
*setvec
[SETVEC_SIZE
];
2512 struct nat_entry_set
*set
, *tmp
;
2517 if (!nm_i
->dirty_nat_cnt
)
2520 down_write(&nm_i
->nat_tree_lock
);
2523 * if there are no enough space in journal to store dirty nat
2524 * entries, remove all entries from journal and merge them
2525 * into nat entry set.
2527 if (enabled_nat_bits(sbi
, cpc
) ||
2528 !__has_cursum_space(journal
, nm_i
->dirty_nat_cnt
, NAT_JOURNAL
))
2529 remove_nats_in_journal(sbi
);
2531 while ((found
= __gang_lookup_nat_set(nm_i
,
2532 set_idx
, SETVEC_SIZE
, setvec
))) {
2534 set_idx
= setvec
[found
- 1]->set
+ 1;
2535 for (idx
= 0; idx
< found
; idx
++)
2536 __adjust_nat_entry_set(setvec
[idx
], &sets
,
2537 MAX_NAT_JENTRIES(journal
));
2540 /* flush dirty nats in nat entry set */
2541 list_for_each_entry_safe(set
, tmp
, &sets
, set_list
)
2542 __flush_nat_entry_set(sbi
, set
, cpc
);
2544 up_write(&nm_i
->nat_tree_lock
);
2545 /* Allow dirty nats by node block allocation in write_begin */
2548 static int __get_nat_bitmaps(struct f2fs_sb_info
*sbi
)
2550 struct f2fs_checkpoint
*ckpt
= F2FS_CKPT(sbi
);
2551 struct f2fs_nm_info
*nm_i
= NM_I(sbi
);
2552 unsigned int nat_bits_bytes
= nm_i
->nat_blocks
/ BITS_PER_BYTE
;
2554 __u64 cp_ver
= cur_cp_version(ckpt
);
2555 block_t nat_bits_addr
;
2557 if (!enabled_nat_bits(sbi
, NULL
))
2560 nm_i
->nat_bits_blocks
= F2FS_BYTES_TO_BLK((nat_bits_bytes
<< 1) + 8 +
2562 nm_i
->nat_bits
= kzalloc(nm_i
->nat_bits_blocks
<< F2FS_BLKSIZE_BITS
,
2564 if (!nm_i
->nat_bits
)
2567 nat_bits_addr
= __start_cp_addr(sbi
) + sbi
->blocks_per_seg
-
2568 nm_i
->nat_bits_blocks
;
2569 for (i
= 0; i
< nm_i
->nat_bits_blocks
; i
++) {
2570 struct page
*page
= get_meta_page(sbi
, nat_bits_addr
++);
2572 memcpy(nm_i
->nat_bits
+ (i
<< F2FS_BLKSIZE_BITS
),
2573 page_address(page
), F2FS_BLKSIZE
);
2574 f2fs_put_page(page
, 1);
2577 cp_ver
|= (cur_cp_crc(ckpt
) << 32);
2578 if (cpu_to_le64(cp_ver
) != *(__le64
*)nm_i
->nat_bits
) {
2579 disable_nat_bits(sbi
, true);
2583 nm_i
->full_nat_bits
= nm_i
->nat_bits
+ 8;
2584 nm_i
->empty_nat_bits
= nm_i
->full_nat_bits
+ nat_bits_bytes
;
2586 f2fs_msg(sbi
->sb
, KERN_NOTICE
, "Found nat_bits in checkpoint");
2590 static inline void load_free_nid_bitmap(struct f2fs_sb_info
*sbi
)
2592 struct f2fs_nm_info
*nm_i
= NM_I(sbi
);
2594 nid_t nid
, last_nid
;
2596 if (!enabled_nat_bits(sbi
, NULL
))
2599 for (i
= 0; i
< nm_i
->nat_blocks
; i
++) {
2600 i
= find_next_bit_le(nm_i
->empty_nat_bits
, nm_i
->nat_blocks
, i
);
2601 if (i
>= nm_i
->nat_blocks
)
2604 __set_bit_le(i
, nm_i
->nat_block_bitmap
);
2606 nid
= i
* NAT_ENTRY_PER_BLOCK
;
2607 last_nid
= (i
+ 1) * NAT_ENTRY_PER_BLOCK
;
2609 spin_lock(&NM_I(sbi
)->nid_list_lock
);
2610 for (; nid
< last_nid
; nid
++)
2611 update_free_nid_bitmap(sbi
, nid
, true, true);
2612 spin_unlock(&NM_I(sbi
)->nid_list_lock
);
2615 for (i
= 0; i
< nm_i
->nat_blocks
; i
++) {
2616 i
= find_next_bit_le(nm_i
->full_nat_bits
, nm_i
->nat_blocks
, i
);
2617 if (i
>= nm_i
->nat_blocks
)
2620 __set_bit_le(i
, nm_i
->nat_block_bitmap
);
2624 static int init_node_manager(struct f2fs_sb_info
*sbi
)
2626 struct f2fs_super_block
*sb_raw
= F2FS_RAW_SUPER(sbi
);
2627 struct f2fs_nm_info
*nm_i
= NM_I(sbi
);
2628 unsigned char *version_bitmap
;
2629 unsigned int nat_segs
;
2632 nm_i
->nat_blkaddr
= le32_to_cpu(sb_raw
->nat_blkaddr
);
2634 /* segment_count_nat includes pair segment so divide to 2. */
2635 nat_segs
= le32_to_cpu(sb_raw
->segment_count_nat
) >> 1;
2636 nm_i
->nat_blocks
= nat_segs
<< le32_to_cpu(sb_raw
->log_blocks_per_seg
);
2637 nm_i
->max_nid
= NAT_ENTRY_PER_BLOCK
* nm_i
->nat_blocks
;
2639 /* not used nids: 0, node, meta, (and root counted as valid node) */
2640 nm_i
->available_nids
= nm_i
->max_nid
- sbi
->total_valid_node_count
-
2641 F2FS_RESERVED_NODE_NUM
;
2642 nm_i
->nid_cnt
[FREE_NID
] = 0;
2643 nm_i
->nid_cnt
[PREALLOC_NID
] = 0;
2645 nm_i
->ram_thresh
= DEF_RAM_THRESHOLD
;
2646 nm_i
->ra_nid_pages
= DEF_RA_NID_PAGES
;
2647 nm_i
->dirty_nats_ratio
= DEF_DIRTY_NAT_RATIO_THRESHOLD
;
2649 INIT_RADIX_TREE(&nm_i
->free_nid_root
, GFP_ATOMIC
);
2650 INIT_LIST_HEAD(&nm_i
->free_nid_list
);
2651 INIT_RADIX_TREE(&nm_i
->nat_root
, GFP_NOIO
);
2652 INIT_RADIX_TREE(&nm_i
->nat_set_root
, GFP_NOIO
);
2653 INIT_LIST_HEAD(&nm_i
->nat_entries
);
2655 mutex_init(&nm_i
->build_lock
);
2656 spin_lock_init(&nm_i
->nid_list_lock
);
2657 init_rwsem(&nm_i
->nat_tree_lock
);
2659 nm_i
->next_scan_nid
= le32_to_cpu(sbi
->ckpt
->next_free_nid
);
2660 nm_i
->bitmap_size
= __bitmap_size(sbi
, NAT_BITMAP
);
2661 version_bitmap
= __bitmap_ptr(sbi
, NAT_BITMAP
);
2662 if (!version_bitmap
)
2665 nm_i
->nat_bitmap
= kmemdup(version_bitmap
, nm_i
->bitmap_size
,
2667 if (!nm_i
->nat_bitmap
)
2670 err
= __get_nat_bitmaps(sbi
);
2674 #ifdef CONFIG_F2FS_CHECK_FS
2675 nm_i
->nat_bitmap_mir
= kmemdup(version_bitmap
, nm_i
->bitmap_size
,
2677 if (!nm_i
->nat_bitmap_mir
)
2684 static int init_free_nid_cache(struct f2fs_sb_info
*sbi
)
2686 struct f2fs_nm_info
*nm_i
= NM_I(sbi
);
2688 nm_i
->free_nid_bitmap
= kvzalloc(nm_i
->nat_blocks
*
2689 NAT_ENTRY_BITMAP_SIZE
, GFP_KERNEL
);
2690 if (!nm_i
->free_nid_bitmap
)
2693 nm_i
->nat_block_bitmap
= kvzalloc(nm_i
->nat_blocks
/ 8,
2695 if (!nm_i
->nat_block_bitmap
)
2698 nm_i
->free_nid_count
= kvzalloc(nm_i
->nat_blocks
*
2699 sizeof(unsigned short), GFP_KERNEL
);
2700 if (!nm_i
->free_nid_count
)
2705 int build_node_manager(struct f2fs_sb_info
*sbi
)
2709 sbi
->nm_info
= kzalloc(sizeof(struct f2fs_nm_info
), GFP_KERNEL
);
2713 err
= init_node_manager(sbi
);
2717 err
= init_free_nid_cache(sbi
);
2721 /* load free nid status from nat_bits table */
2722 load_free_nid_bitmap(sbi
);
2724 build_free_nids(sbi
, true, true);
2728 void destroy_node_manager(struct f2fs_sb_info
*sbi
)
2730 struct f2fs_nm_info
*nm_i
= NM_I(sbi
);
2731 struct free_nid
*i
, *next_i
;
2732 struct nat_entry
*natvec
[NATVEC_SIZE
];
2733 struct nat_entry_set
*setvec
[SETVEC_SIZE
];
2740 /* destroy free nid list */
2741 spin_lock(&nm_i
->nid_list_lock
);
2742 list_for_each_entry_safe(i
, next_i
, &nm_i
->free_nid_list
, list
) {
2743 __remove_free_nid(sbi
, i
, FREE_NID
, false);
2744 spin_unlock(&nm_i
->nid_list_lock
);
2745 kmem_cache_free(free_nid_slab
, i
);
2746 spin_lock(&nm_i
->nid_list_lock
);
2748 f2fs_bug_on(sbi
, nm_i
->nid_cnt
[FREE_NID
]);
2749 f2fs_bug_on(sbi
, nm_i
->nid_cnt
[PREALLOC_NID
]);
2750 f2fs_bug_on(sbi
, !list_empty(&nm_i
->free_nid_list
));
2751 spin_unlock(&nm_i
->nid_list_lock
);
2753 /* destroy nat cache */
2754 down_write(&nm_i
->nat_tree_lock
);
2755 while ((found
= __gang_lookup_nat_cache(nm_i
,
2756 nid
, NATVEC_SIZE
, natvec
))) {
2759 nid
= nat_get_nid(natvec
[found
- 1]) + 1;
2760 for (idx
= 0; idx
< found
; idx
++)
2761 __del_from_nat_cache(nm_i
, natvec
[idx
]);
2763 f2fs_bug_on(sbi
, nm_i
->nat_cnt
);
2765 /* destroy nat set cache */
2767 while ((found
= __gang_lookup_nat_set(nm_i
,
2768 nid
, SETVEC_SIZE
, setvec
))) {
2771 nid
= setvec
[found
- 1]->set
+ 1;
2772 for (idx
= 0; idx
< found
; idx
++) {
2773 /* entry_cnt is not zero, when cp_error was occurred */
2774 f2fs_bug_on(sbi
, !list_empty(&setvec
[idx
]->entry_list
));
2775 radix_tree_delete(&nm_i
->nat_set_root
, setvec
[idx
]->set
);
2776 kmem_cache_free(nat_entry_set_slab
, setvec
[idx
]);
2779 up_write(&nm_i
->nat_tree_lock
);
2781 kvfree(nm_i
->nat_block_bitmap
);
2782 kvfree(nm_i
->free_nid_bitmap
);
2783 kvfree(nm_i
->free_nid_count
);
2785 kfree(nm_i
->nat_bitmap
);
2786 kfree(nm_i
->nat_bits
);
2787 #ifdef CONFIG_F2FS_CHECK_FS
2788 kfree(nm_i
->nat_bitmap_mir
);
2790 sbi
->nm_info
= NULL
;
2794 int __init
create_node_manager_caches(void)
2796 nat_entry_slab
= f2fs_kmem_cache_create("nat_entry",
2797 sizeof(struct nat_entry
));
2798 if (!nat_entry_slab
)
2801 free_nid_slab
= f2fs_kmem_cache_create("free_nid",
2802 sizeof(struct free_nid
));
2804 goto destroy_nat_entry
;
2806 nat_entry_set_slab
= f2fs_kmem_cache_create("nat_entry_set",
2807 sizeof(struct nat_entry_set
));
2808 if (!nat_entry_set_slab
)
2809 goto destroy_free_nid
;
2813 kmem_cache_destroy(free_nid_slab
);
2815 kmem_cache_destroy(nat_entry_slab
);
2820 void destroy_node_manager_caches(void)
2822 kmem_cache_destroy(nat_entry_set_slab
);
2823 kmem_cache_destroy(free_nid_slab
);
2824 kmem_cache_destroy(nat_entry_slab
);