4 * Copyright (c) 2012 Samsung Electronics Co., Ltd.
5 * http://www.samsung.com/
7 * This program is free software; you can redistribute it and/or modify
8 * it under the terms of the GNU General Public License version 2 as
9 * published by the Free Software Foundation.
12 #include <linux/f2fs_fs.h>
13 #include <linux/mpage.h>
14 #include <linux/backing-dev.h>
15 #include <linux/blkdev.h>
16 #include <linux/pagevec.h>
17 #include <linux/swap.h>
23 #include <trace/events/f2fs.h>
25 #define on_build_free_nids(nmi) mutex_is_locked(&nm_i->build_lock)
27 static struct kmem_cache
*nat_entry_slab
;
28 static struct kmem_cache
*free_nid_slab
;
29 static struct kmem_cache
*nat_entry_set_slab
;
31 bool available_free_memory(struct f2fs_sb_info
*sbi
, int type
)
33 struct f2fs_nm_info
*nm_i
= NM_I(sbi
);
35 unsigned long avail_ram
;
36 unsigned long mem_size
= 0;
41 /* only uses low memory */
42 avail_ram
= val
.totalram
- val
.totalhigh
;
45 * give 25%, 25%, 50%, 50%, 50% memory for each components respectively
47 if (type
== FREE_NIDS
) {
48 mem_size
= (nm_i
->fcnt
* sizeof(struct free_nid
)) >>
50 res
= mem_size
< ((avail_ram
* nm_i
->ram_thresh
/ 100) >> 2);
51 } else if (type
== NAT_ENTRIES
) {
52 mem_size
= (nm_i
->nat_cnt
* sizeof(struct nat_entry
)) >>
54 res
= mem_size
< ((avail_ram
* nm_i
->ram_thresh
/ 100) >> 2);
55 } else if (type
== DIRTY_DENTS
) {
56 if (sbi
->sb
->s_bdi
->wb
.dirty_exceeded
)
58 mem_size
= get_pages(sbi
, F2FS_DIRTY_DENTS
);
59 res
= mem_size
< ((avail_ram
* nm_i
->ram_thresh
/ 100) >> 1);
60 } else if (type
== INO_ENTRIES
) {
63 for (i
= 0; i
<= UPDATE_INO
; i
++)
64 mem_size
+= (sbi
->im
[i
].ino_num
*
65 sizeof(struct ino_entry
)) >> PAGE_SHIFT
;
66 res
= mem_size
< ((avail_ram
* nm_i
->ram_thresh
/ 100) >> 1);
67 } else if (type
== EXTENT_CACHE
) {
68 mem_size
= (atomic_read(&sbi
->total_ext_tree
) *
69 sizeof(struct extent_tree
) +
70 atomic_read(&sbi
->total_ext_node
) *
71 sizeof(struct extent_node
)) >> PAGE_SHIFT
;
72 res
= mem_size
< ((avail_ram
* nm_i
->ram_thresh
/ 100) >> 1);
74 if (!sbi
->sb
->s_bdi
->wb
.dirty_exceeded
)
80 static void clear_node_page_dirty(struct page
*page
)
82 struct address_space
*mapping
= page
->mapping
;
83 unsigned int long flags
;
85 if (PageDirty(page
)) {
86 spin_lock_irqsave(&mapping
->tree_lock
, flags
);
87 radix_tree_tag_clear(&mapping
->page_tree
,
90 spin_unlock_irqrestore(&mapping
->tree_lock
, flags
);
92 clear_page_dirty_for_io(page
);
93 dec_page_count(F2FS_M_SB(mapping
), F2FS_DIRTY_NODES
);
95 ClearPageUptodate(page
);
98 static struct page
*get_current_nat_page(struct f2fs_sb_info
*sbi
, nid_t nid
)
100 pgoff_t index
= current_nat_addr(sbi
, nid
);
101 return get_meta_page(sbi
, index
);
104 static struct page
*get_next_nat_page(struct f2fs_sb_info
*sbi
, nid_t nid
)
106 struct page
*src_page
;
107 struct page
*dst_page
;
112 struct f2fs_nm_info
*nm_i
= NM_I(sbi
);
114 src_off
= current_nat_addr(sbi
, nid
);
115 dst_off
= next_nat_addr(sbi
, src_off
);
117 /* get current nat block page with lock */
118 src_page
= get_meta_page(sbi
, src_off
);
119 dst_page
= grab_meta_page(sbi
, dst_off
);
120 f2fs_bug_on(sbi
, PageDirty(src_page
));
122 src_addr
= page_address(src_page
);
123 dst_addr
= page_address(dst_page
);
124 memcpy(dst_addr
, src_addr
, PAGE_SIZE
);
125 set_page_dirty(dst_page
);
126 f2fs_put_page(src_page
, 1);
128 set_to_next_nat(nm_i
, nid
);
133 static struct nat_entry
*__lookup_nat_cache(struct f2fs_nm_info
*nm_i
, nid_t n
)
135 return radix_tree_lookup(&nm_i
->nat_root
, n
);
138 static unsigned int __gang_lookup_nat_cache(struct f2fs_nm_info
*nm_i
,
139 nid_t start
, unsigned int nr
, struct nat_entry
**ep
)
141 return radix_tree_gang_lookup(&nm_i
->nat_root
, (void **)ep
, start
, nr
);
144 static void __del_from_nat_cache(struct f2fs_nm_info
*nm_i
, struct nat_entry
*e
)
147 radix_tree_delete(&nm_i
->nat_root
, nat_get_nid(e
));
149 kmem_cache_free(nat_entry_slab
, e
);
152 static void __set_nat_cache_dirty(struct f2fs_nm_info
*nm_i
,
153 struct nat_entry
*ne
)
155 nid_t set
= NAT_BLOCK_OFFSET(ne
->ni
.nid
);
156 struct nat_entry_set
*head
;
158 if (get_nat_flag(ne
, IS_DIRTY
))
161 head
= radix_tree_lookup(&nm_i
->nat_set_root
, set
);
163 head
= f2fs_kmem_cache_alloc(nat_entry_set_slab
, GFP_NOFS
);
165 INIT_LIST_HEAD(&head
->entry_list
);
166 INIT_LIST_HEAD(&head
->set_list
);
169 f2fs_radix_tree_insert(&nm_i
->nat_set_root
, set
, head
);
171 list_move_tail(&ne
->list
, &head
->entry_list
);
172 nm_i
->dirty_nat_cnt
++;
174 set_nat_flag(ne
, IS_DIRTY
, true);
177 static void __clear_nat_cache_dirty(struct f2fs_nm_info
*nm_i
,
178 struct nat_entry
*ne
)
180 nid_t set
= NAT_BLOCK_OFFSET(ne
->ni
.nid
);
181 struct nat_entry_set
*head
;
183 head
= radix_tree_lookup(&nm_i
->nat_set_root
, set
);
185 list_move_tail(&ne
->list
, &nm_i
->nat_entries
);
186 set_nat_flag(ne
, IS_DIRTY
, false);
188 nm_i
->dirty_nat_cnt
--;
192 static unsigned int __gang_lookup_nat_set(struct f2fs_nm_info
*nm_i
,
193 nid_t start
, unsigned int nr
, struct nat_entry_set
**ep
)
195 return radix_tree_gang_lookup(&nm_i
->nat_set_root
, (void **)ep
,
199 int need_dentry_mark(struct f2fs_sb_info
*sbi
, nid_t nid
)
201 struct f2fs_nm_info
*nm_i
= NM_I(sbi
);
205 down_read(&nm_i
->nat_tree_lock
);
206 e
= __lookup_nat_cache(nm_i
, nid
);
208 if (!get_nat_flag(e
, IS_CHECKPOINTED
) &&
209 !get_nat_flag(e
, HAS_FSYNCED_INODE
))
212 up_read(&nm_i
->nat_tree_lock
);
216 bool is_checkpointed_node(struct f2fs_sb_info
*sbi
, nid_t nid
)
218 struct f2fs_nm_info
*nm_i
= NM_I(sbi
);
222 down_read(&nm_i
->nat_tree_lock
);
223 e
= __lookup_nat_cache(nm_i
, nid
);
224 if (e
&& !get_nat_flag(e
, IS_CHECKPOINTED
))
226 up_read(&nm_i
->nat_tree_lock
);
230 bool need_inode_block_update(struct f2fs_sb_info
*sbi
, nid_t ino
)
232 struct f2fs_nm_info
*nm_i
= NM_I(sbi
);
234 bool need_update
= true;
236 down_read(&nm_i
->nat_tree_lock
);
237 e
= __lookup_nat_cache(nm_i
, ino
);
238 if (e
&& get_nat_flag(e
, HAS_LAST_FSYNC
) &&
239 (get_nat_flag(e
, IS_CHECKPOINTED
) ||
240 get_nat_flag(e
, HAS_FSYNCED_INODE
)))
242 up_read(&nm_i
->nat_tree_lock
);
246 static struct nat_entry
*grab_nat_entry(struct f2fs_nm_info
*nm_i
, nid_t nid
)
248 struct nat_entry
*new;
250 new = f2fs_kmem_cache_alloc(nat_entry_slab
, GFP_NOFS
);
251 f2fs_radix_tree_insert(&nm_i
->nat_root
, nid
, new);
252 memset(new, 0, sizeof(struct nat_entry
));
253 nat_set_nid(new, nid
);
255 list_add_tail(&new->list
, &nm_i
->nat_entries
);
260 static void cache_nat_entry(struct f2fs_sb_info
*sbi
, nid_t nid
,
261 struct f2fs_nat_entry
*ne
)
263 struct f2fs_nm_info
*nm_i
= NM_I(sbi
);
266 e
= __lookup_nat_cache(nm_i
, nid
);
268 e
= grab_nat_entry(nm_i
, nid
);
269 node_info_from_raw_nat(&e
->ni
, ne
);
271 f2fs_bug_on(sbi
, nat_get_ino(e
) != ne
->ino
||
272 nat_get_blkaddr(e
) != ne
->block_addr
||
273 nat_get_version(e
) != ne
->version
);
277 static void set_node_addr(struct f2fs_sb_info
*sbi
, struct node_info
*ni
,
278 block_t new_blkaddr
, bool fsync_done
)
280 struct f2fs_nm_info
*nm_i
= NM_I(sbi
);
283 down_write(&nm_i
->nat_tree_lock
);
284 e
= __lookup_nat_cache(nm_i
, ni
->nid
);
286 e
= grab_nat_entry(nm_i
, ni
->nid
);
287 copy_node_info(&e
->ni
, ni
);
288 f2fs_bug_on(sbi
, ni
->blk_addr
== NEW_ADDR
);
289 } else if (new_blkaddr
== NEW_ADDR
) {
291 * when nid is reallocated,
292 * previous nat entry can be remained in nat cache.
293 * So, reinitialize it with new information.
295 copy_node_info(&e
->ni
, ni
);
296 f2fs_bug_on(sbi
, ni
->blk_addr
!= NULL_ADDR
);
300 f2fs_bug_on(sbi
, nat_get_blkaddr(e
) != ni
->blk_addr
);
301 f2fs_bug_on(sbi
, nat_get_blkaddr(e
) == NULL_ADDR
&&
302 new_blkaddr
== NULL_ADDR
);
303 f2fs_bug_on(sbi
, nat_get_blkaddr(e
) == NEW_ADDR
&&
304 new_blkaddr
== NEW_ADDR
);
305 f2fs_bug_on(sbi
, nat_get_blkaddr(e
) != NEW_ADDR
&&
306 nat_get_blkaddr(e
) != NULL_ADDR
&&
307 new_blkaddr
== NEW_ADDR
);
309 /* increment version no as node is removed */
310 if (nat_get_blkaddr(e
) != NEW_ADDR
&& new_blkaddr
== NULL_ADDR
) {
311 unsigned char version
= nat_get_version(e
);
312 nat_set_version(e
, inc_node_version(version
));
314 /* in order to reuse the nid */
315 if (nm_i
->next_scan_nid
> ni
->nid
)
316 nm_i
->next_scan_nid
= ni
->nid
;
320 nat_set_blkaddr(e
, new_blkaddr
);
321 if (new_blkaddr
== NEW_ADDR
|| new_blkaddr
== NULL_ADDR
)
322 set_nat_flag(e
, IS_CHECKPOINTED
, false);
323 __set_nat_cache_dirty(nm_i
, e
);
325 /* update fsync_mark if its inode nat entry is still alive */
326 if (ni
->nid
!= ni
->ino
)
327 e
= __lookup_nat_cache(nm_i
, ni
->ino
);
329 if (fsync_done
&& ni
->nid
== ni
->ino
)
330 set_nat_flag(e
, HAS_FSYNCED_INODE
, true);
331 set_nat_flag(e
, HAS_LAST_FSYNC
, fsync_done
);
333 up_write(&nm_i
->nat_tree_lock
);
336 int try_to_free_nats(struct f2fs_sb_info
*sbi
, int nr_shrink
)
338 struct f2fs_nm_info
*nm_i
= NM_I(sbi
);
341 if (!down_write_trylock(&nm_i
->nat_tree_lock
))
344 while (nr_shrink
&& !list_empty(&nm_i
->nat_entries
)) {
345 struct nat_entry
*ne
;
346 ne
= list_first_entry(&nm_i
->nat_entries
,
347 struct nat_entry
, list
);
348 __del_from_nat_cache(nm_i
, ne
);
351 up_write(&nm_i
->nat_tree_lock
);
352 return nr
- nr_shrink
;
356 * This function always returns success
358 void get_node_info(struct f2fs_sb_info
*sbi
, nid_t nid
, struct node_info
*ni
)
360 struct f2fs_nm_info
*nm_i
= NM_I(sbi
);
361 struct curseg_info
*curseg
= CURSEG_I(sbi
, CURSEG_HOT_DATA
);
362 struct f2fs_journal
*journal
= curseg
->journal
;
363 nid_t start_nid
= START_NID(nid
);
364 struct f2fs_nat_block
*nat_blk
;
365 struct page
*page
= NULL
;
366 struct f2fs_nat_entry ne
;
372 /* Check nat cache */
373 down_read(&nm_i
->nat_tree_lock
);
374 e
= __lookup_nat_cache(nm_i
, nid
);
376 ni
->ino
= nat_get_ino(e
);
377 ni
->blk_addr
= nat_get_blkaddr(e
);
378 ni
->version
= nat_get_version(e
);
379 up_read(&nm_i
->nat_tree_lock
);
383 memset(&ne
, 0, sizeof(struct f2fs_nat_entry
));
385 /* Check current segment summary */
386 down_read(&curseg
->journal_rwsem
);
387 i
= lookup_journal_in_cursum(journal
, NAT_JOURNAL
, nid
, 0);
389 ne
= nat_in_journal(journal
, i
);
390 node_info_from_raw_nat(ni
, &ne
);
392 up_read(&curseg
->journal_rwsem
);
396 /* Fill node_info from nat page */
397 page
= get_current_nat_page(sbi
, start_nid
);
398 nat_blk
= (struct f2fs_nat_block
*)page_address(page
);
399 ne
= nat_blk
->entries
[nid
- start_nid
];
400 node_info_from_raw_nat(ni
, &ne
);
401 f2fs_put_page(page
, 1);
403 up_read(&nm_i
->nat_tree_lock
);
404 /* cache nat entry */
405 down_write(&nm_i
->nat_tree_lock
);
406 cache_nat_entry(sbi
, nid
, &ne
);
407 up_write(&nm_i
->nat_tree_lock
);
411 * readahead MAX_RA_NODE number of node pages.
413 static void ra_node_pages(struct page
*parent
, int start
, int n
)
415 struct f2fs_sb_info
*sbi
= F2FS_P_SB(parent
);
416 struct blk_plug plug
;
420 blk_start_plug(&plug
);
422 /* Then, try readahead for siblings of the desired node */
424 end
= min(end
, NIDS_PER_BLOCK
);
425 for (i
= start
; i
< end
; i
++) {
426 nid
= get_nid(parent
, i
, false);
427 ra_node_page(sbi
, nid
);
430 blk_finish_plug(&plug
);
433 pgoff_t
get_next_page_offset(struct dnode_of_data
*dn
, pgoff_t pgofs
)
435 const long direct_index
= ADDRS_PER_INODE(dn
->inode
);
436 const long direct_blks
= ADDRS_PER_BLOCK
;
437 const long indirect_blks
= ADDRS_PER_BLOCK
* NIDS_PER_BLOCK
;
438 unsigned int skipped_unit
= ADDRS_PER_BLOCK
;
439 int cur_level
= dn
->cur_level
;
440 int max_level
= dn
->max_level
;
446 while (max_level
-- > cur_level
)
447 skipped_unit
*= NIDS_PER_BLOCK
;
449 switch (dn
->max_level
) {
451 base
+= 2 * indirect_blks
;
453 base
+= 2 * direct_blks
;
455 base
+= direct_index
;
458 f2fs_bug_on(F2FS_I_SB(dn
->inode
), 1);
461 return ((pgofs
- base
) / skipped_unit
+ 1) * skipped_unit
+ base
;
465 * The maximum depth is four.
466 * Offset[0] will have raw inode offset.
468 static int get_node_path(struct inode
*inode
, long block
,
469 int offset
[4], unsigned int noffset
[4])
471 const long direct_index
= ADDRS_PER_INODE(inode
);
472 const long direct_blks
= ADDRS_PER_BLOCK
;
473 const long dptrs_per_blk
= NIDS_PER_BLOCK
;
474 const long indirect_blks
= ADDRS_PER_BLOCK
* NIDS_PER_BLOCK
;
475 const long dindirect_blks
= indirect_blks
* NIDS_PER_BLOCK
;
481 if (block
< direct_index
) {
485 block
-= direct_index
;
486 if (block
< direct_blks
) {
487 offset
[n
++] = NODE_DIR1_BLOCK
;
493 block
-= direct_blks
;
494 if (block
< direct_blks
) {
495 offset
[n
++] = NODE_DIR2_BLOCK
;
501 block
-= direct_blks
;
502 if (block
< indirect_blks
) {
503 offset
[n
++] = NODE_IND1_BLOCK
;
505 offset
[n
++] = block
/ direct_blks
;
506 noffset
[n
] = 4 + offset
[n
- 1];
507 offset
[n
] = block
% direct_blks
;
511 block
-= indirect_blks
;
512 if (block
< indirect_blks
) {
513 offset
[n
++] = NODE_IND2_BLOCK
;
514 noffset
[n
] = 4 + dptrs_per_blk
;
515 offset
[n
++] = block
/ direct_blks
;
516 noffset
[n
] = 5 + dptrs_per_blk
+ offset
[n
- 1];
517 offset
[n
] = block
% direct_blks
;
521 block
-= indirect_blks
;
522 if (block
< dindirect_blks
) {
523 offset
[n
++] = NODE_DIND_BLOCK
;
524 noffset
[n
] = 5 + (dptrs_per_blk
* 2);
525 offset
[n
++] = block
/ indirect_blks
;
526 noffset
[n
] = 6 + (dptrs_per_blk
* 2) +
527 offset
[n
- 1] * (dptrs_per_blk
+ 1);
528 offset
[n
++] = (block
/ direct_blks
) % dptrs_per_blk
;
529 noffset
[n
] = 7 + (dptrs_per_blk
* 2) +
530 offset
[n
- 2] * (dptrs_per_blk
+ 1) +
532 offset
[n
] = block
% direct_blks
;
543 * Caller should call f2fs_put_dnode(dn).
544 * Also, it should grab and release a rwsem by calling f2fs_lock_op() and
545 * f2fs_unlock_op() only if ro is not set RDONLY_NODE.
546 * In the case of RDONLY_NODE, we don't need to care about mutex.
548 int get_dnode_of_data(struct dnode_of_data
*dn
, pgoff_t index
, int mode
)
550 struct f2fs_sb_info
*sbi
= F2FS_I_SB(dn
->inode
);
551 struct page
*npage
[4];
552 struct page
*parent
= NULL
;
554 unsigned int noffset
[4];
559 level
= get_node_path(dn
->inode
, index
, offset
, noffset
);
561 nids
[0] = dn
->inode
->i_ino
;
562 npage
[0] = dn
->inode_page
;
565 npage
[0] = get_node_page(sbi
, nids
[0]);
566 if (IS_ERR(npage
[0]))
567 return PTR_ERR(npage
[0]);
570 /* if inline_data is set, should not report any block indices */
571 if (f2fs_has_inline_data(dn
->inode
) && index
) {
573 f2fs_put_page(npage
[0], 1);
579 nids
[1] = get_nid(parent
, offset
[0], true);
580 dn
->inode_page
= npage
[0];
581 dn
->inode_page_locked
= true;
583 /* get indirect or direct nodes */
584 for (i
= 1; i
<= level
; i
++) {
587 if (!nids
[i
] && mode
== ALLOC_NODE
) {
589 if (!alloc_nid(sbi
, &(nids
[i
]))) {
595 npage
[i
] = new_node_page(dn
, noffset
[i
], NULL
);
596 if (IS_ERR(npage
[i
])) {
597 alloc_nid_failed(sbi
, nids
[i
]);
598 err
= PTR_ERR(npage
[i
]);
602 set_nid(parent
, offset
[i
- 1], nids
[i
], i
== 1);
603 alloc_nid_done(sbi
, nids
[i
]);
605 } else if (mode
== LOOKUP_NODE_RA
&& i
== level
&& level
> 1) {
606 npage
[i
] = get_node_page_ra(parent
, offset
[i
- 1]);
607 if (IS_ERR(npage
[i
])) {
608 err
= PTR_ERR(npage
[i
]);
614 dn
->inode_page_locked
= false;
617 f2fs_put_page(parent
, 1);
621 npage
[i
] = get_node_page(sbi
, nids
[i
]);
622 if (IS_ERR(npage
[i
])) {
623 err
= PTR_ERR(npage
[i
]);
624 f2fs_put_page(npage
[0], 0);
630 nids
[i
+ 1] = get_nid(parent
, offset
[i
], false);
633 dn
->nid
= nids
[level
];
634 dn
->ofs_in_node
= offset
[level
];
635 dn
->node_page
= npage
[level
];
636 dn
->data_blkaddr
= datablock_addr(dn
->node_page
, dn
->ofs_in_node
);
640 f2fs_put_page(parent
, 1);
642 f2fs_put_page(npage
[0], 0);
644 dn
->inode_page
= NULL
;
645 dn
->node_page
= NULL
;
646 if (err
== -ENOENT
) {
648 dn
->max_level
= level
;
653 static void truncate_node(struct dnode_of_data
*dn
)
655 struct f2fs_sb_info
*sbi
= F2FS_I_SB(dn
->inode
);
658 get_node_info(sbi
, dn
->nid
, &ni
);
659 if (dn
->inode
->i_blocks
== 0) {
660 f2fs_bug_on(sbi
, ni
.blk_addr
!= NULL_ADDR
);
663 f2fs_bug_on(sbi
, ni
.blk_addr
== NULL_ADDR
);
665 /* Deallocate node address */
666 invalidate_blocks(sbi
, ni
.blk_addr
);
667 dec_valid_node_count(sbi
, dn
->inode
);
668 set_node_addr(sbi
, &ni
, NULL_ADDR
, false);
670 if (dn
->nid
== dn
->inode
->i_ino
) {
671 remove_orphan_inode(sbi
, dn
->nid
);
672 dec_valid_inode_count(sbi
);
677 clear_node_page_dirty(dn
->node_page
);
678 set_sbi_flag(sbi
, SBI_IS_DIRTY
);
680 f2fs_put_page(dn
->node_page
, 1);
682 invalidate_mapping_pages(NODE_MAPPING(sbi
),
683 dn
->node_page
->index
, dn
->node_page
->index
);
685 dn
->node_page
= NULL
;
686 trace_f2fs_truncate_node(dn
->inode
, dn
->nid
, ni
.blk_addr
);
689 static int truncate_dnode(struct dnode_of_data
*dn
)
696 /* get direct node */
697 page
= get_node_page(F2FS_I_SB(dn
->inode
), dn
->nid
);
698 if (IS_ERR(page
) && PTR_ERR(page
) == -ENOENT
)
700 else if (IS_ERR(page
))
701 return PTR_ERR(page
);
703 /* Make dnode_of_data for parameter */
704 dn
->node_page
= page
;
706 truncate_data_blocks(dn
);
711 static int truncate_nodes(struct dnode_of_data
*dn
, unsigned int nofs
,
714 struct dnode_of_data rdn
= *dn
;
716 struct f2fs_node
*rn
;
718 unsigned int child_nofs
;
723 return NIDS_PER_BLOCK
+ 1;
725 trace_f2fs_truncate_nodes_enter(dn
->inode
, dn
->nid
, dn
->data_blkaddr
);
727 page
= get_node_page(F2FS_I_SB(dn
->inode
), dn
->nid
);
729 trace_f2fs_truncate_nodes_exit(dn
->inode
, PTR_ERR(page
));
730 return PTR_ERR(page
);
733 ra_node_pages(page
, ofs
, NIDS_PER_BLOCK
);
735 rn
= F2FS_NODE(page
);
737 for (i
= ofs
; i
< NIDS_PER_BLOCK
; i
++, freed
++) {
738 child_nid
= le32_to_cpu(rn
->in
.nid
[i
]);
742 ret
= truncate_dnode(&rdn
);
745 if (set_nid(page
, i
, 0, false))
746 dn
->node_changed
= true;
749 child_nofs
= nofs
+ ofs
* (NIDS_PER_BLOCK
+ 1) + 1;
750 for (i
= ofs
; i
< NIDS_PER_BLOCK
; i
++) {
751 child_nid
= le32_to_cpu(rn
->in
.nid
[i
]);
752 if (child_nid
== 0) {
753 child_nofs
+= NIDS_PER_BLOCK
+ 1;
757 ret
= truncate_nodes(&rdn
, child_nofs
, 0, depth
- 1);
758 if (ret
== (NIDS_PER_BLOCK
+ 1)) {
759 if (set_nid(page
, i
, 0, false))
760 dn
->node_changed
= true;
762 } else if (ret
< 0 && ret
!= -ENOENT
) {
770 /* remove current indirect node */
771 dn
->node_page
= page
;
775 f2fs_put_page(page
, 1);
777 trace_f2fs_truncate_nodes_exit(dn
->inode
, freed
);
781 f2fs_put_page(page
, 1);
782 trace_f2fs_truncate_nodes_exit(dn
->inode
, ret
);
786 static int truncate_partial_nodes(struct dnode_of_data
*dn
,
787 struct f2fs_inode
*ri
, int *offset
, int depth
)
789 struct page
*pages
[2];
796 nid
[0] = le32_to_cpu(ri
->i_nid
[offset
[0] - NODE_DIR1_BLOCK
]);
800 /* get indirect nodes in the path */
801 for (i
= 0; i
< idx
+ 1; i
++) {
802 /* reference count'll be increased */
803 pages
[i
] = get_node_page(F2FS_I_SB(dn
->inode
), nid
[i
]);
804 if (IS_ERR(pages
[i
])) {
805 err
= PTR_ERR(pages
[i
]);
809 nid
[i
+ 1] = get_nid(pages
[i
], offset
[i
+ 1], false);
812 ra_node_pages(pages
[idx
], offset
[idx
+ 1], NIDS_PER_BLOCK
);
814 /* free direct nodes linked to a partial indirect node */
815 for (i
= offset
[idx
+ 1]; i
< NIDS_PER_BLOCK
; i
++) {
816 child_nid
= get_nid(pages
[idx
], i
, false);
820 err
= truncate_dnode(dn
);
823 if (set_nid(pages
[idx
], i
, 0, false))
824 dn
->node_changed
= true;
827 if (offset
[idx
+ 1] == 0) {
828 dn
->node_page
= pages
[idx
];
832 f2fs_put_page(pages
[idx
], 1);
838 for (i
= idx
; i
>= 0; i
--)
839 f2fs_put_page(pages
[i
], 1);
841 trace_f2fs_truncate_partial_nodes(dn
->inode
, nid
, depth
, err
);
847 * All the block addresses of data and nodes should be nullified.
849 int truncate_inode_blocks(struct inode
*inode
, pgoff_t from
)
851 struct f2fs_sb_info
*sbi
= F2FS_I_SB(inode
);
852 int err
= 0, cont
= 1;
853 int level
, offset
[4], noffset
[4];
854 unsigned int nofs
= 0;
855 struct f2fs_inode
*ri
;
856 struct dnode_of_data dn
;
859 trace_f2fs_truncate_inode_blocks_enter(inode
, from
);
861 level
= get_node_path(inode
, from
, offset
, noffset
);
863 page
= get_node_page(sbi
, inode
->i_ino
);
865 trace_f2fs_truncate_inode_blocks_exit(inode
, PTR_ERR(page
));
866 return PTR_ERR(page
);
869 set_new_dnode(&dn
, inode
, page
, NULL
, 0);
872 ri
= F2FS_INODE(page
);
880 if (!offset
[level
- 1])
882 err
= truncate_partial_nodes(&dn
, ri
, offset
, level
);
883 if (err
< 0 && err
!= -ENOENT
)
885 nofs
+= 1 + NIDS_PER_BLOCK
;
888 nofs
= 5 + 2 * NIDS_PER_BLOCK
;
889 if (!offset
[level
- 1])
891 err
= truncate_partial_nodes(&dn
, ri
, offset
, level
);
892 if (err
< 0 && err
!= -ENOENT
)
901 dn
.nid
= le32_to_cpu(ri
->i_nid
[offset
[0] - NODE_DIR1_BLOCK
]);
903 case NODE_DIR1_BLOCK
:
904 case NODE_DIR2_BLOCK
:
905 err
= truncate_dnode(&dn
);
908 case NODE_IND1_BLOCK
:
909 case NODE_IND2_BLOCK
:
910 err
= truncate_nodes(&dn
, nofs
, offset
[1], 2);
913 case NODE_DIND_BLOCK
:
914 err
= truncate_nodes(&dn
, nofs
, offset
[1], 3);
921 if (err
< 0 && err
!= -ENOENT
)
923 if (offset
[1] == 0 &&
924 ri
->i_nid
[offset
[0] - NODE_DIR1_BLOCK
]) {
926 BUG_ON(page
->mapping
!= NODE_MAPPING(sbi
));
927 f2fs_wait_on_page_writeback(page
, NODE
, true);
928 ri
->i_nid
[offset
[0] - NODE_DIR1_BLOCK
] = 0;
929 set_page_dirty(page
);
937 f2fs_put_page(page
, 0);
938 trace_f2fs_truncate_inode_blocks_exit(inode
, err
);
939 return err
> 0 ? 0 : err
;
942 int truncate_xattr_node(struct inode
*inode
, struct page
*page
)
944 struct f2fs_sb_info
*sbi
= F2FS_I_SB(inode
);
945 nid_t nid
= F2FS_I(inode
)->i_xattr_nid
;
946 struct dnode_of_data dn
;
952 npage
= get_node_page(sbi
, nid
);
954 return PTR_ERR(npage
);
956 F2FS_I(inode
)->i_xattr_nid
= 0;
958 /* need to do checkpoint during fsync */
959 F2FS_I(inode
)->xattr_ver
= cur_cp_version(F2FS_CKPT(sbi
));
961 set_new_dnode(&dn
, inode
, page
, npage
, nid
);
964 dn
.inode_page_locked
= true;
970 * Caller should grab and release a rwsem by calling f2fs_lock_op() and
973 int remove_inode_page(struct inode
*inode
)
975 struct dnode_of_data dn
;
978 set_new_dnode(&dn
, inode
, NULL
, NULL
, inode
->i_ino
);
979 err
= get_dnode_of_data(&dn
, 0, LOOKUP_NODE
);
983 err
= truncate_xattr_node(inode
, dn
.inode_page
);
989 /* remove potential inline_data blocks */
990 if (S_ISREG(inode
->i_mode
) || S_ISDIR(inode
->i_mode
) ||
991 S_ISLNK(inode
->i_mode
))
992 truncate_data_blocks_range(&dn
, 1);
994 /* 0 is possible, after f2fs_new_inode() has failed */
995 f2fs_bug_on(F2FS_I_SB(inode
),
996 inode
->i_blocks
!= 0 && inode
->i_blocks
!= 1);
998 /* will put inode & node pages */
1003 struct page
*new_inode_page(struct inode
*inode
)
1005 struct dnode_of_data dn
;
1007 /* allocate inode page for new inode */
1008 set_new_dnode(&dn
, inode
, NULL
, NULL
, inode
->i_ino
);
1010 /* caller should f2fs_put_page(page, 1); */
1011 return new_node_page(&dn
, 0, NULL
);
1014 struct page
*new_node_page(struct dnode_of_data
*dn
,
1015 unsigned int ofs
, struct page
*ipage
)
1017 struct f2fs_sb_info
*sbi
= F2FS_I_SB(dn
->inode
);
1018 struct node_info old_ni
, new_ni
;
1022 if (unlikely(is_inode_flag_set(F2FS_I(dn
->inode
), FI_NO_ALLOC
)))
1023 return ERR_PTR(-EPERM
);
1025 page
= f2fs_grab_cache_page(NODE_MAPPING(sbi
), dn
->nid
, false);
1027 return ERR_PTR(-ENOMEM
);
1029 if (unlikely(!inc_valid_node_count(sbi
, dn
->inode
))) {
1034 get_node_info(sbi
, dn
->nid
, &old_ni
);
1036 /* Reinitialize old_ni with new node page */
1037 f2fs_bug_on(sbi
, old_ni
.blk_addr
!= NULL_ADDR
);
1039 new_ni
.ino
= dn
->inode
->i_ino
;
1040 set_node_addr(sbi
, &new_ni
, NEW_ADDR
, false);
1042 f2fs_wait_on_page_writeback(page
, NODE
, true);
1043 fill_node_footer(page
, dn
->nid
, dn
->inode
->i_ino
, ofs
, true);
1044 set_cold_node(dn
->inode
, page
);
1045 SetPageUptodate(page
);
1046 if (set_page_dirty(page
))
1047 dn
->node_changed
= true;
1049 if (f2fs_has_xattr_block(ofs
))
1050 F2FS_I(dn
->inode
)->i_xattr_nid
= dn
->nid
;
1052 dn
->node_page
= page
;
1054 update_inode(dn
->inode
, ipage
);
1056 sync_inode_page(dn
);
1058 inc_valid_inode_count(sbi
);
1063 clear_node_page_dirty(page
);
1064 f2fs_put_page(page
, 1);
1065 return ERR_PTR(err
);
1069 * Caller should do after getting the following values.
1070 * 0: f2fs_put_page(page, 0)
1071 * LOCKED_PAGE or error: f2fs_put_page(page, 1)
1073 static int read_node_page(struct page
*page
, int op_flags
)
1075 struct f2fs_sb_info
*sbi
= F2FS_P_SB(page
);
1076 struct node_info ni
;
1077 struct f2fs_io_info fio
= {
1081 .op_flags
= op_flags
,
1083 .encrypted_page
= NULL
,
1086 get_node_info(sbi
, page
->index
, &ni
);
1088 if (unlikely(ni
.blk_addr
== NULL_ADDR
)) {
1089 ClearPageUptodate(page
);
1093 if (PageUptodate(page
))
1096 fio
.new_blkaddr
= fio
.old_blkaddr
= ni
.blk_addr
;
1097 return f2fs_submit_page_bio(&fio
);
1101 * Readahead a node page
1103 void ra_node_page(struct f2fs_sb_info
*sbi
, nid_t nid
)
1110 f2fs_bug_on(sbi
, check_nid_range(sbi
, nid
));
1113 apage
= radix_tree_lookup(&NODE_MAPPING(sbi
)->page_tree
, nid
);
1118 apage
= f2fs_grab_cache_page(NODE_MAPPING(sbi
), nid
, false);
1122 err
= read_node_page(apage
, READA
);
1123 f2fs_put_page(apage
, err
? 1 : 0);
1126 static struct page
*__get_node_page(struct f2fs_sb_info
*sbi
, pgoff_t nid
,
1127 struct page
*parent
, int start
)
1133 return ERR_PTR(-ENOENT
);
1134 f2fs_bug_on(sbi
, check_nid_range(sbi
, nid
));
1136 page
= f2fs_grab_cache_page(NODE_MAPPING(sbi
), nid
, false);
1138 return ERR_PTR(-ENOMEM
);
1140 err
= read_node_page(page
, READ_SYNC
);
1142 f2fs_put_page(page
, 1);
1143 return ERR_PTR(err
);
1144 } else if (err
== LOCKED_PAGE
) {
1149 ra_node_pages(parent
, start
+ 1, MAX_RA_NODE
);
1153 if (unlikely(!PageUptodate(page
))) {
1154 f2fs_put_page(page
, 1);
1155 return ERR_PTR(-EIO
);
1157 if (unlikely(page
->mapping
!= NODE_MAPPING(sbi
))) {
1158 f2fs_put_page(page
, 1);
1162 f2fs_bug_on(sbi
, nid
!= nid_of_node(page
));
1166 struct page
*get_node_page(struct f2fs_sb_info
*sbi
, pgoff_t nid
)
1168 return __get_node_page(sbi
, nid
, NULL
, 0);
1171 struct page
*get_node_page_ra(struct page
*parent
, int start
)
1173 struct f2fs_sb_info
*sbi
= F2FS_P_SB(parent
);
1174 nid_t nid
= get_nid(parent
, start
, false);
1176 return __get_node_page(sbi
, nid
, parent
, start
);
1179 void sync_inode_page(struct dnode_of_data
*dn
)
1183 if (IS_INODE(dn
->node_page
) || dn
->inode_page
== dn
->node_page
) {
1184 ret
= update_inode(dn
->inode
, dn
->node_page
);
1185 } else if (dn
->inode_page
) {
1186 if (!dn
->inode_page_locked
)
1187 lock_page(dn
->inode_page
);
1188 ret
= update_inode(dn
->inode
, dn
->inode_page
);
1189 if (!dn
->inode_page_locked
)
1190 unlock_page(dn
->inode_page
);
1192 ret
= update_inode_page(dn
->inode
);
1194 dn
->node_changed
= ret
? true: false;
1197 static void flush_inline_data(struct f2fs_sb_info
*sbi
, nid_t ino
)
1199 struct inode
*inode
;
1203 /* should flush inline_data before evict_inode */
1204 inode
= ilookup(sbi
->sb
, ino
);
1208 page
= pagecache_get_page(inode
->i_mapping
, 0, FGP_LOCK
|FGP_NOWAIT
, 0);
1212 if (!PageUptodate(page
))
1215 if (!PageDirty(page
))
1218 if (!clear_page_dirty_for_io(page
))
1221 ret
= f2fs_write_inline_data(inode
, page
);
1222 inode_dec_dirty_pages(inode
);
1224 set_page_dirty(page
);
1226 f2fs_put_page(page
, 1);
1231 void move_node_page(struct page
*node_page
, int gc_type
)
1233 if (gc_type
== FG_GC
) {
1234 struct f2fs_sb_info
*sbi
= F2FS_P_SB(node_page
);
1235 struct writeback_control wbc
= {
1236 .sync_mode
= WB_SYNC_ALL
,
1241 set_page_dirty(node_page
);
1242 f2fs_wait_on_page_writeback(node_page
, NODE
, true);
1244 f2fs_bug_on(sbi
, PageWriteback(node_page
));
1245 if (!clear_page_dirty_for_io(node_page
))
1248 if (NODE_MAPPING(sbi
)->a_ops
->writepage(node_page
, &wbc
))
1249 unlock_page(node_page
);
1252 /* set page dirty and write it */
1253 if (!PageWriteback(node_page
))
1254 set_page_dirty(node_page
);
1257 unlock_page(node_page
);
1259 f2fs_put_page(node_page
, 0);
1262 static struct page
*last_fsync_dnode(struct f2fs_sb_info
*sbi
, nid_t ino
)
1265 struct pagevec pvec
;
1266 struct page
*last_page
= NULL
;
1268 pagevec_init(&pvec
, 0);
1272 while (index
<= end
) {
1274 nr_pages
= pagevec_lookup_tag(&pvec
, NODE_MAPPING(sbi
), &index
,
1275 PAGECACHE_TAG_DIRTY
,
1276 min(end
- index
, (pgoff_t
)PAGEVEC_SIZE
-1) + 1);
1280 for (i
= 0; i
< nr_pages
; i
++) {
1281 struct page
*page
= pvec
.pages
[i
];
1283 if (unlikely(f2fs_cp_error(sbi
))) {
1284 f2fs_put_page(last_page
, 0);
1285 pagevec_release(&pvec
);
1286 return ERR_PTR(-EIO
);
1289 if (!IS_DNODE(page
) || !is_cold_node(page
))
1291 if (ino_of_node(page
) != ino
)
1296 if (unlikely(page
->mapping
!= NODE_MAPPING(sbi
))) {
1301 if (ino_of_node(page
) != ino
)
1302 goto continue_unlock
;
1304 if (!PageDirty(page
)) {
1305 /* someone wrote it for us */
1306 goto continue_unlock
;
1310 f2fs_put_page(last_page
, 0);
1316 pagevec_release(&pvec
);
1322 int fsync_node_pages(struct f2fs_sb_info
*sbi
, nid_t ino
,
1323 struct writeback_control
*wbc
, bool atomic
)
1326 struct pagevec pvec
;
1328 struct page
*last_page
= NULL
;
1329 bool marked
= false;
1332 last_page
= last_fsync_dnode(sbi
, ino
);
1333 if (IS_ERR_OR_NULL(last_page
))
1334 return PTR_ERR_OR_ZERO(last_page
);
1337 pagevec_init(&pvec
, 0);
1341 while (index
<= end
) {
1343 nr_pages
= pagevec_lookup_tag(&pvec
, NODE_MAPPING(sbi
), &index
,
1344 PAGECACHE_TAG_DIRTY
,
1345 min(end
- index
, (pgoff_t
)PAGEVEC_SIZE
-1) + 1);
1349 for (i
= 0; i
< nr_pages
; i
++) {
1350 struct page
*page
= pvec
.pages
[i
];
1352 if (unlikely(f2fs_cp_error(sbi
))) {
1353 f2fs_put_page(last_page
, 0);
1354 pagevec_release(&pvec
);
1358 if (!IS_DNODE(page
) || !is_cold_node(page
))
1360 if (ino_of_node(page
) != ino
)
1365 if (unlikely(page
->mapping
!= NODE_MAPPING(sbi
))) {
1370 if (ino_of_node(page
) != ino
)
1371 goto continue_unlock
;
1373 if (!PageDirty(page
) && page
!= last_page
) {
1374 /* someone wrote it for us */
1375 goto continue_unlock
;
1378 f2fs_wait_on_page_writeback(page
, NODE
, true);
1379 BUG_ON(PageWriteback(page
));
1381 if (!atomic
|| page
== last_page
) {
1382 set_fsync_mark(page
, 1);
1384 set_dentry_mark(page
,
1385 need_dentry_mark(sbi
, ino
));
1386 /* may be written by other thread */
1387 if (!PageDirty(page
))
1388 set_page_dirty(page
);
1391 if (!clear_page_dirty_for_io(page
))
1392 goto continue_unlock
;
1394 ret
= NODE_MAPPING(sbi
)->a_ops
->writepage(page
, wbc
);
1397 f2fs_put_page(last_page
, 0);
1400 if (page
== last_page
) {
1401 f2fs_put_page(page
, 0);
1406 pagevec_release(&pvec
);
1412 if (!ret
&& atomic
&& !marked
) {
1413 f2fs_msg(sbi
->sb
, KERN_DEBUG
,
1414 "Retry to write fsync mark: ino=%u, idx=%lx",
1415 ino
, last_page
->index
);
1416 lock_page(last_page
);
1417 set_page_dirty(last_page
);
1418 unlock_page(last_page
);
1421 return ret
? -EIO
: 0;
1424 int sync_node_pages(struct f2fs_sb_info
*sbi
, struct writeback_control
*wbc
)
1427 struct pagevec pvec
;
1431 pagevec_init(&pvec
, 0);
1437 while (index
<= end
) {
1439 nr_pages
= pagevec_lookup_tag(&pvec
, NODE_MAPPING(sbi
), &index
,
1440 PAGECACHE_TAG_DIRTY
,
1441 min(end
- index
, (pgoff_t
)PAGEVEC_SIZE
-1) + 1);
1445 for (i
= 0; i
< nr_pages
; i
++) {
1446 struct page
*page
= pvec
.pages
[i
];
1448 if (unlikely(f2fs_cp_error(sbi
))) {
1449 pagevec_release(&pvec
);
1454 * flushing sequence with step:
1459 if (step
== 0 && IS_DNODE(page
))
1461 if (step
== 1 && (!IS_DNODE(page
) ||
1462 is_cold_node(page
)))
1464 if (step
== 2 && (!IS_DNODE(page
) ||
1465 !is_cold_node(page
)))
1468 if (!trylock_page(page
))
1471 if (unlikely(page
->mapping
!= NODE_MAPPING(sbi
))) {
1477 if (!PageDirty(page
)) {
1478 /* someone wrote it for us */
1479 goto continue_unlock
;
1482 /* flush inline_data */
1483 if (is_inline_node(page
)) {
1484 clear_inline_node(page
);
1486 flush_inline_data(sbi
, ino_of_node(page
));
1490 f2fs_wait_on_page_writeback(page
, NODE
, true);
1492 BUG_ON(PageWriteback(page
));
1493 if (!clear_page_dirty_for_io(page
))
1494 goto continue_unlock
;
1496 set_fsync_mark(page
, 0);
1497 set_dentry_mark(page
, 0);
1499 if (NODE_MAPPING(sbi
)->a_ops
->writepage(page
, wbc
))
1502 if (--wbc
->nr_to_write
== 0)
1505 pagevec_release(&pvec
);
1508 if (wbc
->nr_to_write
== 0) {
1521 int wait_on_node_pages_writeback(struct f2fs_sb_info
*sbi
, nid_t ino
)
1523 pgoff_t index
= 0, end
= ULONG_MAX
;
1524 struct pagevec pvec
;
1525 int ret2
= 0, ret
= 0;
1527 pagevec_init(&pvec
, 0);
1529 while (index
<= end
) {
1531 nr_pages
= pagevec_lookup_tag(&pvec
, NODE_MAPPING(sbi
), &index
,
1532 PAGECACHE_TAG_WRITEBACK
,
1533 min(end
- index
, (pgoff_t
)PAGEVEC_SIZE
-1) + 1);
1537 for (i
= 0; i
< nr_pages
; i
++) {
1538 struct page
*page
= pvec
.pages
[i
];
1540 /* until radix tree lookup accepts end_index */
1541 if (unlikely(page
->index
> end
))
1544 if (ino
&& ino_of_node(page
) == ino
) {
1545 f2fs_wait_on_page_writeback(page
, NODE
, true);
1546 if (TestClearPageError(page
))
1550 pagevec_release(&pvec
);
1554 if (unlikely(test_and_clear_bit(AS_ENOSPC
, &NODE_MAPPING(sbi
)->flags
)))
1556 if (unlikely(test_and_clear_bit(AS_EIO
, &NODE_MAPPING(sbi
)->flags
)))
1563 static int f2fs_write_node_page(struct page
*page
,
1564 struct writeback_control
*wbc
)
1566 struct f2fs_sb_info
*sbi
= F2FS_P_SB(page
);
1568 struct node_info ni
;
1569 struct f2fs_io_info fio
= {
1573 .op_flags
= (wbc
->sync_mode
== WB_SYNC_ALL
) ? WRITE_SYNC
: 0,
1575 .encrypted_page
= NULL
,
1578 trace_f2fs_writepage(page
, NODE
);
1580 if (unlikely(is_sbi_flag_set(sbi
, SBI_POR_DOING
)))
1582 if (unlikely(f2fs_cp_error(sbi
)))
1585 /* get old block addr of this node page */
1586 nid
= nid_of_node(page
);
1587 f2fs_bug_on(sbi
, page
->index
!= nid
);
1589 if (wbc
->for_reclaim
) {
1590 if (!down_read_trylock(&sbi
->node_write
))
1593 down_read(&sbi
->node_write
);
1596 get_node_info(sbi
, nid
, &ni
);
1598 /* This page is already truncated */
1599 if (unlikely(ni
.blk_addr
== NULL_ADDR
)) {
1600 ClearPageUptodate(page
);
1601 dec_page_count(sbi
, F2FS_DIRTY_NODES
);
1602 up_read(&sbi
->node_write
);
1607 set_page_writeback(page
);
1608 fio
.old_blkaddr
= ni
.blk_addr
;
1609 write_node_page(nid
, &fio
);
1610 set_node_addr(sbi
, &ni
, fio
.new_blkaddr
, is_fsync_dnode(page
));
1611 dec_page_count(sbi
, F2FS_DIRTY_NODES
);
1612 up_read(&sbi
->node_write
);
1614 if (wbc
->for_reclaim
)
1615 f2fs_submit_merged_bio_cond(sbi
, NULL
, page
, 0, NODE
, WRITE
);
1619 if (unlikely(f2fs_cp_error(sbi
)))
1620 f2fs_submit_merged_bio(sbi
, NODE
, WRITE
);
1625 redirty_page_for_writepage(wbc
, page
);
1626 return AOP_WRITEPAGE_ACTIVATE
;
1629 static int f2fs_write_node_pages(struct address_space
*mapping
,
1630 struct writeback_control
*wbc
)
1632 struct f2fs_sb_info
*sbi
= F2FS_M_SB(mapping
);
1635 /* balancing f2fs's metadata in background */
1636 f2fs_balance_fs_bg(sbi
);
1638 /* collect a number of dirty node pages and write together */
1639 if (get_pages(sbi
, F2FS_DIRTY_NODES
) < nr_pages_to_skip(sbi
, NODE
))
1642 trace_f2fs_writepages(mapping
->host
, wbc
, NODE
);
1644 diff
= nr_pages_to_write(sbi
, NODE
, wbc
);
1645 wbc
->sync_mode
= WB_SYNC_NONE
;
1646 sync_node_pages(sbi
, wbc
);
1647 wbc
->nr_to_write
= max((long)0, wbc
->nr_to_write
- diff
);
1651 wbc
->pages_skipped
+= get_pages(sbi
, F2FS_DIRTY_NODES
);
1652 trace_f2fs_writepages(mapping
->host
, wbc
, NODE
);
1656 static int f2fs_set_node_page_dirty(struct page
*page
)
1658 trace_f2fs_set_page_dirty(page
, NODE
);
1660 SetPageUptodate(page
);
1661 if (!PageDirty(page
)) {
1662 __set_page_dirty_nobuffers(page
);
1663 inc_page_count(F2FS_P_SB(page
), F2FS_DIRTY_NODES
);
1664 SetPagePrivate(page
);
1665 f2fs_trace_pid(page
);
1672 * Structure of the f2fs node operations
1674 const struct address_space_operations f2fs_node_aops
= {
1675 .writepage
= f2fs_write_node_page
,
1676 .writepages
= f2fs_write_node_pages
,
1677 .set_page_dirty
= f2fs_set_node_page_dirty
,
1678 .invalidatepage
= f2fs_invalidate_page
,
1679 .releasepage
= f2fs_release_page
,
1682 static struct free_nid
*__lookup_free_nid_list(struct f2fs_nm_info
*nm_i
,
1685 return radix_tree_lookup(&nm_i
->free_nid_root
, n
);
1688 static void __del_from_free_nid_list(struct f2fs_nm_info
*nm_i
,
1692 radix_tree_delete(&nm_i
->free_nid_root
, i
->nid
);
1695 static int add_free_nid(struct f2fs_sb_info
*sbi
, nid_t nid
, bool build
)
1697 struct f2fs_nm_info
*nm_i
= NM_I(sbi
);
1699 struct nat_entry
*ne
;
1701 if (!available_free_memory(sbi
, FREE_NIDS
))
1704 /* 0 nid should not be used */
1705 if (unlikely(nid
== 0))
1709 /* do not add allocated nids */
1710 ne
= __lookup_nat_cache(nm_i
, nid
);
1711 if (ne
&& (!get_nat_flag(ne
, IS_CHECKPOINTED
) ||
1712 nat_get_blkaddr(ne
) != NULL_ADDR
))
1716 i
= f2fs_kmem_cache_alloc(free_nid_slab
, GFP_NOFS
);
1720 if (radix_tree_preload(GFP_NOFS
)) {
1721 kmem_cache_free(free_nid_slab
, i
);
1725 spin_lock(&nm_i
->free_nid_list_lock
);
1726 if (radix_tree_insert(&nm_i
->free_nid_root
, i
->nid
, i
)) {
1727 spin_unlock(&nm_i
->free_nid_list_lock
);
1728 radix_tree_preload_end();
1729 kmem_cache_free(free_nid_slab
, i
);
1732 list_add_tail(&i
->list
, &nm_i
->free_nid_list
);
1734 spin_unlock(&nm_i
->free_nid_list_lock
);
1735 radix_tree_preload_end();
1739 static void remove_free_nid(struct f2fs_nm_info
*nm_i
, nid_t nid
)
1742 bool need_free
= false;
1744 spin_lock(&nm_i
->free_nid_list_lock
);
1745 i
= __lookup_free_nid_list(nm_i
, nid
);
1746 if (i
&& i
->state
== NID_NEW
) {
1747 __del_from_free_nid_list(nm_i
, i
);
1751 spin_unlock(&nm_i
->free_nid_list_lock
);
1754 kmem_cache_free(free_nid_slab
, i
);
1757 static void scan_nat_page(struct f2fs_sb_info
*sbi
,
1758 struct page
*nat_page
, nid_t start_nid
)
1760 struct f2fs_nm_info
*nm_i
= NM_I(sbi
);
1761 struct f2fs_nat_block
*nat_blk
= page_address(nat_page
);
1765 i
= start_nid
% NAT_ENTRY_PER_BLOCK
;
1767 for (; i
< NAT_ENTRY_PER_BLOCK
; i
++, start_nid
++) {
1769 if (unlikely(start_nid
>= nm_i
->max_nid
))
1772 blk_addr
= le32_to_cpu(nat_blk
->entries
[i
].block_addr
);
1773 f2fs_bug_on(sbi
, blk_addr
== NEW_ADDR
);
1774 if (blk_addr
== NULL_ADDR
) {
1775 if (add_free_nid(sbi
, start_nid
, true) < 0)
1781 static void build_free_nids(struct f2fs_sb_info
*sbi
)
1783 struct f2fs_nm_info
*nm_i
= NM_I(sbi
);
1784 struct curseg_info
*curseg
= CURSEG_I(sbi
, CURSEG_HOT_DATA
);
1785 struct f2fs_journal
*journal
= curseg
->journal
;
1787 nid_t nid
= nm_i
->next_scan_nid
;
1789 /* Enough entries */
1790 if (nm_i
->fcnt
> NAT_ENTRY_PER_BLOCK
)
1793 /* readahead nat pages to be scanned */
1794 ra_meta_pages(sbi
, NAT_BLOCK_OFFSET(nid
), FREE_NID_PAGES
,
1797 down_read(&nm_i
->nat_tree_lock
);
1800 struct page
*page
= get_current_nat_page(sbi
, nid
);
1802 scan_nat_page(sbi
, page
, nid
);
1803 f2fs_put_page(page
, 1);
1805 nid
+= (NAT_ENTRY_PER_BLOCK
- (nid
% NAT_ENTRY_PER_BLOCK
));
1806 if (unlikely(nid
>= nm_i
->max_nid
))
1809 if (++i
>= FREE_NID_PAGES
)
1813 /* go to the next free nat pages to find free nids abundantly */
1814 nm_i
->next_scan_nid
= nid
;
1816 /* find free nids from current sum_pages */
1817 down_read(&curseg
->journal_rwsem
);
1818 for (i
= 0; i
< nats_in_cursum(journal
); i
++) {
1821 addr
= le32_to_cpu(nat_in_journal(journal
, i
).block_addr
);
1822 nid
= le32_to_cpu(nid_in_journal(journal
, i
));
1823 if (addr
== NULL_ADDR
)
1824 add_free_nid(sbi
, nid
, true);
1826 remove_free_nid(nm_i
, nid
);
1828 up_read(&curseg
->journal_rwsem
);
1829 up_read(&nm_i
->nat_tree_lock
);
1831 ra_meta_pages(sbi
, NAT_BLOCK_OFFSET(nm_i
->next_scan_nid
),
1832 nm_i
->ra_nid_pages
, META_NAT
, false);
1836 * If this function returns success, caller can obtain a new nid
1837 * from second parameter of this function.
1838 * The returned nid could be used ino as well as nid when inode is created.
1840 bool alloc_nid(struct f2fs_sb_info
*sbi
, nid_t
*nid
)
1842 struct f2fs_nm_info
*nm_i
= NM_I(sbi
);
1843 struct free_nid
*i
= NULL
;
1845 #ifdef CONFIG_F2FS_FAULT_INJECTION
1846 if (time_to_inject(FAULT_ALLOC_NID
))
1849 if (unlikely(sbi
->total_valid_node_count
+ 1 > nm_i
->available_nids
))
1852 spin_lock(&nm_i
->free_nid_list_lock
);
1854 /* We should not use stale free nids created by build_free_nids */
1855 if (nm_i
->fcnt
&& !on_build_free_nids(nm_i
)) {
1856 f2fs_bug_on(sbi
, list_empty(&nm_i
->free_nid_list
));
1857 list_for_each_entry(i
, &nm_i
->free_nid_list
, list
)
1858 if (i
->state
== NID_NEW
)
1861 f2fs_bug_on(sbi
, i
->state
!= NID_NEW
);
1863 i
->state
= NID_ALLOC
;
1865 spin_unlock(&nm_i
->free_nid_list_lock
);
1868 spin_unlock(&nm_i
->free_nid_list_lock
);
1870 /* Let's scan nat pages and its caches to get free nids */
1871 mutex_lock(&nm_i
->build_lock
);
1872 build_free_nids(sbi
);
1873 mutex_unlock(&nm_i
->build_lock
);
1878 * alloc_nid() should be called prior to this function.
1880 void alloc_nid_done(struct f2fs_sb_info
*sbi
, nid_t nid
)
1882 struct f2fs_nm_info
*nm_i
= NM_I(sbi
);
1885 spin_lock(&nm_i
->free_nid_list_lock
);
1886 i
= __lookup_free_nid_list(nm_i
, nid
);
1887 f2fs_bug_on(sbi
, !i
|| i
->state
!= NID_ALLOC
);
1888 __del_from_free_nid_list(nm_i
, i
);
1889 spin_unlock(&nm_i
->free_nid_list_lock
);
1891 kmem_cache_free(free_nid_slab
, i
);
1895 * alloc_nid() should be called prior to this function.
1897 void alloc_nid_failed(struct f2fs_sb_info
*sbi
, nid_t nid
)
1899 struct f2fs_nm_info
*nm_i
= NM_I(sbi
);
1901 bool need_free
= false;
1906 spin_lock(&nm_i
->free_nid_list_lock
);
1907 i
= __lookup_free_nid_list(nm_i
, nid
);
1908 f2fs_bug_on(sbi
, !i
|| i
->state
!= NID_ALLOC
);
1909 if (!available_free_memory(sbi
, FREE_NIDS
)) {
1910 __del_from_free_nid_list(nm_i
, i
);
1916 spin_unlock(&nm_i
->free_nid_list_lock
);
1919 kmem_cache_free(free_nid_slab
, i
);
1922 int try_to_free_nids(struct f2fs_sb_info
*sbi
, int nr_shrink
)
1924 struct f2fs_nm_info
*nm_i
= NM_I(sbi
);
1925 struct free_nid
*i
, *next
;
1928 if (!mutex_trylock(&nm_i
->build_lock
))
1931 spin_lock(&nm_i
->free_nid_list_lock
);
1932 list_for_each_entry_safe(i
, next
, &nm_i
->free_nid_list
, list
) {
1933 if (nr_shrink
<= 0 || nm_i
->fcnt
<= NAT_ENTRY_PER_BLOCK
)
1935 if (i
->state
== NID_ALLOC
)
1937 __del_from_free_nid_list(nm_i
, i
);
1938 kmem_cache_free(free_nid_slab
, i
);
1942 spin_unlock(&nm_i
->free_nid_list_lock
);
1943 mutex_unlock(&nm_i
->build_lock
);
1945 return nr
- nr_shrink
;
1948 void recover_inline_xattr(struct inode
*inode
, struct page
*page
)
1950 void *src_addr
, *dst_addr
;
1953 struct f2fs_inode
*ri
;
1955 ipage
= get_node_page(F2FS_I_SB(inode
), inode
->i_ino
);
1956 f2fs_bug_on(F2FS_I_SB(inode
), IS_ERR(ipage
));
1958 ri
= F2FS_INODE(page
);
1959 if (!(ri
->i_inline
& F2FS_INLINE_XATTR
)) {
1960 clear_inode_flag(F2FS_I(inode
), FI_INLINE_XATTR
);
1964 dst_addr
= inline_xattr_addr(ipage
);
1965 src_addr
= inline_xattr_addr(page
);
1966 inline_size
= inline_xattr_size(inode
);
1968 f2fs_wait_on_page_writeback(ipage
, NODE
, true);
1969 memcpy(dst_addr
, src_addr
, inline_size
);
1971 update_inode(inode
, ipage
);
1972 f2fs_put_page(ipage
, 1);
1975 void recover_xattr_data(struct inode
*inode
, struct page
*page
, block_t blkaddr
)
1977 struct f2fs_sb_info
*sbi
= F2FS_I_SB(inode
);
1978 nid_t prev_xnid
= F2FS_I(inode
)->i_xattr_nid
;
1979 nid_t new_xnid
= nid_of_node(page
);
1980 struct node_info ni
;
1982 /* 1: invalidate the previous xattr nid */
1986 /* Deallocate node address */
1987 get_node_info(sbi
, prev_xnid
, &ni
);
1988 f2fs_bug_on(sbi
, ni
.blk_addr
== NULL_ADDR
);
1989 invalidate_blocks(sbi
, ni
.blk_addr
);
1990 dec_valid_node_count(sbi
, inode
);
1991 set_node_addr(sbi
, &ni
, NULL_ADDR
, false);
1994 /* 2: allocate new xattr nid */
1995 if (unlikely(!inc_valid_node_count(sbi
, inode
)))
1996 f2fs_bug_on(sbi
, 1);
1998 remove_free_nid(NM_I(sbi
), new_xnid
);
1999 get_node_info(sbi
, new_xnid
, &ni
);
2000 ni
.ino
= inode
->i_ino
;
2001 set_node_addr(sbi
, &ni
, NEW_ADDR
, false);
2002 F2FS_I(inode
)->i_xattr_nid
= new_xnid
;
2004 /* 3: update xattr blkaddr */
2005 refresh_sit_entry(sbi
, NEW_ADDR
, blkaddr
);
2006 set_node_addr(sbi
, &ni
, blkaddr
, false);
2008 update_inode_page(inode
);
2011 int recover_inode_page(struct f2fs_sb_info
*sbi
, struct page
*page
)
2013 struct f2fs_inode
*src
, *dst
;
2014 nid_t ino
= ino_of_node(page
);
2015 struct node_info old_ni
, new_ni
;
2018 get_node_info(sbi
, ino
, &old_ni
);
2020 if (unlikely(old_ni
.blk_addr
!= NULL_ADDR
))
2023 ipage
= f2fs_grab_cache_page(NODE_MAPPING(sbi
), ino
, false);
2027 /* Should not use this inode from free nid list */
2028 remove_free_nid(NM_I(sbi
), ino
);
2030 SetPageUptodate(ipage
);
2031 fill_node_footer(ipage
, ino
, ino
, 0, true);
2033 src
= F2FS_INODE(page
);
2034 dst
= F2FS_INODE(ipage
);
2036 memcpy(dst
, src
, (unsigned long)&src
->i_ext
- (unsigned long)src
);
2038 dst
->i_blocks
= cpu_to_le64(1);
2039 dst
->i_links
= cpu_to_le32(1);
2040 dst
->i_xattr_nid
= 0;
2041 dst
->i_inline
= src
->i_inline
& F2FS_INLINE_XATTR
;
2046 if (unlikely(!inc_valid_node_count(sbi
, NULL
)))
2048 set_node_addr(sbi
, &new_ni
, NEW_ADDR
, false);
2049 inc_valid_inode_count(sbi
);
2050 set_page_dirty(ipage
);
2051 f2fs_put_page(ipage
, 1);
2055 int restore_node_summary(struct f2fs_sb_info
*sbi
,
2056 unsigned int segno
, struct f2fs_summary_block
*sum
)
2058 struct f2fs_node
*rn
;
2059 struct f2fs_summary
*sum_entry
;
2061 int bio_blocks
= MAX_BIO_BLOCKS(sbi
);
2062 int i
, idx
, last_offset
, nrpages
;
2064 /* scan the node segment */
2065 last_offset
= sbi
->blocks_per_seg
;
2066 addr
= START_BLOCK(sbi
, segno
);
2067 sum_entry
= &sum
->entries
[0];
2069 for (i
= 0; i
< last_offset
; i
+= nrpages
, addr
+= nrpages
) {
2070 nrpages
= min(last_offset
- i
, bio_blocks
);
2072 /* readahead node pages */
2073 ra_meta_pages(sbi
, addr
, nrpages
, META_POR
, true);
2075 for (idx
= addr
; idx
< addr
+ nrpages
; idx
++) {
2076 struct page
*page
= get_tmp_page(sbi
, idx
);
2078 rn
= F2FS_NODE(page
);
2079 sum_entry
->nid
= rn
->footer
.nid
;
2080 sum_entry
->version
= 0;
2081 sum_entry
->ofs_in_node
= 0;
2083 f2fs_put_page(page
, 1);
2086 invalidate_mapping_pages(META_MAPPING(sbi
), addr
,
2092 static void remove_nats_in_journal(struct f2fs_sb_info
*sbi
)
2094 struct f2fs_nm_info
*nm_i
= NM_I(sbi
);
2095 struct curseg_info
*curseg
= CURSEG_I(sbi
, CURSEG_HOT_DATA
);
2096 struct f2fs_journal
*journal
= curseg
->journal
;
2099 down_write(&curseg
->journal_rwsem
);
2100 for (i
= 0; i
< nats_in_cursum(journal
); i
++) {
2101 struct nat_entry
*ne
;
2102 struct f2fs_nat_entry raw_ne
;
2103 nid_t nid
= le32_to_cpu(nid_in_journal(journal
, i
));
2105 raw_ne
= nat_in_journal(journal
, i
);
2107 ne
= __lookup_nat_cache(nm_i
, nid
);
2109 ne
= grab_nat_entry(nm_i
, nid
);
2110 node_info_from_raw_nat(&ne
->ni
, &raw_ne
);
2112 __set_nat_cache_dirty(nm_i
, ne
);
2114 update_nats_in_cursum(journal
, -i
);
2115 up_write(&curseg
->journal_rwsem
);
2118 static void __adjust_nat_entry_set(struct nat_entry_set
*nes
,
2119 struct list_head
*head
, int max
)
2121 struct nat_entry_set
*cur
;
2123 if (nes
->entry_cnt
>= max
)
2126 list_for_each_entry(cur
, head
, set_list
) {
2127 if (cur
->entry_cnt
>= nes
->entry_cnt
) {
2128 list_add(&nes
->set_list
, cur
->set_list
.prev
);
2133 list_add_tail(&nes
->set_list
, head
);
2136 static void __flush_nat_entry_set(struct f2fs_sb_info
*sbi
,
2137 struct nat_entry_set
*set
)
2139 struct curseg_info
*curseg
= CURSEG_I(sbi
, CURSEG_HOT_DATA
);
2140 struct f2fs_journal
*journal
= curseg
->journal
;
2141 nid_t start_nid
= set
->set
* NAT_ENTRY_PER_BLOCK
;
2142 bool to_journal
= true;
2143 struct f2fs_nat_block
*nat_blk
;
2144 struct nat_entry
*ne
, *cur
;
2145 struct page
*page
= NULL
;
2148 * there are two steps to flush nat entries:
2149 * #1, flush nat entries to journal in current hot data summary block.
2150 * #2, flush nat entries to nat page.
2152 if (!__has_cursum_space(journal
, set
->entry_cnt
, NAT_JOURNAL
))
2156 down_write(&curseg
->journal_rwsem
);
2158 page
= get_next_nat_page(sbi
, start_nid
);
2159 nat_blk
= page_address(page
);
2160 f2fs_bug_on(sbi
, !nat_blk
);
2163 /* flush dirty nats in nat entry set */
2164 list_for_each_entry_safe(ne
, cur
, &set
->entry_list
, list
) {
2165 struct f2fs_nat_entry
*raw_ne
;
2166 nid_t nid
= nat_get_nid(ne
);
2169 if (nat_get_blkaddr(ne
) == NEW_ADDR
)
2173 offset
= lookup_journal_in_cursum(journal
,
2174 NAT_JOURNAL
, nid
, 1);
2175 f2fs_bug_on(sbi
, offset
< 0);
2176 raw_ne
= &nat_in_journal(journal
, offset
);
2177 nid_in_journal(journal
, offset
) = cpu_to_le32(nid
);
2179 raw_ne
= &nat_blk
->entries
[nid
- start_nid
];
2181 raw_nat_from_node_info(raw_ne
, &ne
->ni
);
2183 __clear_nat_cache_dirty(NM_I(sbi
), ne
);
2184 if (nat_get_blkaddr(ne
) == NULL_ADDR
)
2185 add_free_nid(sbi
, nid
, false);
2189 up_write(&curseg
->journal_rwsem
);
2191 f2fs_put_page(page
, 1);
2193 f2fs_bug_on(sbi
, set
->entry_cnt
);
2195 radix_tree_delete(&NM_I(sbi
)->nat_set_root
, set
->set
);
2196 kmem_cache_free(nat_entry_set_slab
, set
);
2200 * This function is called during the checkpointing process.
2202 void flush_nat_entries(struct f2fs_sb_info
*sbi
)
2204 struct f2fs_nm_info
*nm_i
= NM_I(sbi
);
2205 struct curseg_info
*curseg
= CURSEG_I(sbi
, CURSEG_HOT_DATA
);
2206 struct f2fs_journal
*journal
= curseg
->journal
;
2207 struct nat_entry_set
*setvec
[SETVEC_SIZE
];
2208 struct nat_entry_set
*set
, *tmp
;
2213 if (!nm_i
->dirty_nat_cnt
)
2216 down_write(&nm_i
->nat_tree_lock
);
2219 * if there are no enough space in journal to store dirty nat
2220 * entries, remove all entries from journal and merge them
2221 * into nat entry set.
2223 if (!__has_cursum_space(journal
, nm_i
->dirty_nat_cnt
, NAT_JOURNAL
))
2224 remove_nats_in_journal(sbi
);
2226 while ((found
= __gang_lookup_nat_set(nm_i
,
2227 set_idx
, SETVEC_SIZE
, setvec
))) {
2229 set_idx
= setvec
[found
- 1]->set
+ 1;
2230 for (idx
= 0; idx
< found
; idx
++)
2231 __adjust_nat_entry_set(setvec
[idx
], &sets
,
2232 MAX_NAT_JENTRIES(journal
));
2235 /* flush dirty nats in nat entry set */
2236 list_for_each_entry_safe(set
, tmp
, &sets
, set_list
)
2237 __flush_nat_entry_set(sbi
, set
);
2239 up_write(&nm_i
->nat_tree_lock
);
2241 f2fs_bug_on(sbi
, nm_i
->dirty_nat_cnt
);
2244 static int init_node_manager(struct f2fs_sb_info
*sbi
)
2246 struct f2fs_super_block
*sb_raw
= F2FS_RAW_SUPER(sbi
);
2247 struct f2fs_nm_info
*nm_i
= NM_I(sbi
);
2248 unsigned char *version_bitmap
;
2249 unsigned int nat_segs
, nat_blocks
;
2251 nm_i
->nat_blkaddr
= le32_to_cpu(sb_raw
->nat_blkaddr
);
2253 /* segment_count_nat includes pair segment so divide to 2. */
2254 nat_segs
= le32_to_cpu(sb_raw
->segment_count_nat
) >> 1;
2255 nat_blocks
= nat_segs
<< le32_to_cpu(sb_raw
->log_blocks_per_seg
);
2257 nm_i
->max_nid
= NAT_ENTRY_PER_BLOCK
* nat_blocks
;
2259 /* not used nids: 0, node, meta, (and root counted as valid node) */
2260 nm_i
->available_nids
= nm_i
->max_nid
- F2FS_RESERVED_NODE_NUM
;
2263 nm_i
->ram_thresh
= DEF_RAM_THRESHOLD
;
2264 nm_i
->ra_nid_pages
= DEF_RA_NID_PAGES
;
2265 nm_i
->dirty_nats_ratio
= DEF_DIRTY_NAT_RATIO_THRESHOLD
;
2267 INIT_RADIX_TREE(&nm_i
->free_nid_root
, GFP_ATOMIC
);
2268 INIT_LIST_HEAD(&nm_i
->free_nid_list
);
2269 INIT_RADIX_TREE(&nm_i
->nat_root
, GFP_NOIO
);
2270 INIT_RADIX_TREE(&nm_i
->nat_set_root
, GFP_NOIO
);
2271 INIT_LIST_HEAD(&nm_i
->nat_entries
);
2273 mutex_init(&nm_i
->build_lock
);
2274 spin_lock_init(&nm_i
->free_nid_list_lock
);
2275 init_rwsem(&nm_i
->nat_tree_lock
);
2277 nm_i
->next_scan_nid
= le32_to_cpu(sbi
->ckpt
->next_free_nid
);
2278 nm_i
->bitmap_size
= __bitmap_size(sbi
, NAT_BITMAP
);
2279 version_bitmap
= __bitmap_ptr(sbi
, NAT_BITMAP
);
2280 if (!version_bitmap
)
2283 nm_i
->nat_bitmap
= kmemdup(version_bitmap
, nm_i
->bitmap_size
,
2285 if (!nm_i
->nat_bitmap
)
2290 int build_node_manager(struct f2fs_sb_info
*sbi
)
2294 sbi
->nm_info
= kzalloc(sizeof(struct f2fs_nm_info
), GFP_KERNEL
);
2298 err
= init_node_manager(sbi
);
2302 build_free_nids(sbi
);
2306 void destroy_node_manager(struct f2fs_sb_info
*sbi
)
2308 struct f2fs_nm_info
*nm_i
= NM_I(sbi
);
2309 struct free_nid
*i
, *next_i
;
2310 struct nat_entry
*natvec
[NATVEC_SIZE
];
2311 struct nat_entry_set
*setvec
[SETVEC_SIZE
];
2318 /* destroy free nid list */
2319 spin_lock(&nm_i
->free_nid_list_lock
);
2320 list_for_each_entry_safe(i
, next_i
, &nm_i
->free_nid_list
, list
) {
2321 f2fs_bug_on(sbi
, i
->state
== NID_ALLOC
);
2322 __del_from_free_nid_list(nm_i
, i
);
2324 spin_unlock(&nm_i
->free_nid_list_lock
);
2325 kmem_cache_free(free_nid_slab
, i
);
2326 spin_lock(&nm_i
->free_nid_list_lock
);
2328 f2fs_bug_on(sbi
, nm_i
->fcnt
);
2329 spin_unlock(&nm_i
->free_nid_list_lock
);
2331 /* destroy nat cache */
2332 down_write(&nm_i
->nat_tree_lock
);
2333 while ((found
= __gang_lookup_nat_cache(nm_i
,
2334 nid
, NATVEC_SIZE
, natvec
))) {
2337 nid
= nat_get_nid(natvec
[found
- 1]) + 1;
2338 for (idx
= 0; idx
< found
; idx
++)
2339 __del_from_nat_cache(nm_i
, natvec
[idx
]);
2341 f2fs_bug_on(sbi
, nm_i
->nat_cnt
);
2343 /* destroy nat set cache */
2345 while ((found
= __gang_lookup_nat_set(nm_i
,
2346 nid
, SETVEC_SIZE
, setvec
))) {
2349 nid
= setvec
[found
- 1]->set
+ 1;
2350 for (idx
= 0; idx
< found
; idx
++) {
2351 /* entry_cnt is not zero, when cp_error was occurred */
2352 f2fs_bug_on(sbi
, !list_empty(&setvec
[idx
]->entry_list
));
2353 radix_tree_delete(&nm_i
->nat_set_root
, setvec
[idx
]->set
);
2354 kmem_cache_free(nat_entry_set_slab
, setvec
[idx
]);
2357 up_write(&nm_i
->nat_tree_lock
);
2359 kfree(nm_i
->nat_bitmap
);
2360 sbi
->nm_info
= NULL
;
2364 int __init
create_node_manager_caches(void)
2366 nat_entry_slab
= f2fs_kmem_cache_create("nat_entry",
2367 sizeof(struct nat_entry
));
2368 if (!nat_entry_slab
)
2371 free_nid_slab
= f2fs_kmem_cache_create("free_nid",
2372 sizeof(struct free_nid
));
2374 goto destroy_nat_entry
;
2376 nat_entry_set_slab
= f2fs_kmem_cache_create("nat_entry_set",
2377 sizeof(struct nat_entry_set
));
2378 if (!nat_entry_set_slab
)
2379 goto destroy_free_nid
;
2383 kmem_cache_destroy(free_nid_slab
);
2385 kmem_cache_destroy(nat_entry_slab
);
2390 void destroy_node_manager_caches(void)
2392 kmem_cache_destroy(nat_entry_set_slab
);
2393 kmem_cache_destroy(free_nid_slab
);
2394 kmem_cache_destroy(nat_entry_slab
);