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
->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_CACHE_SHIFT
;
66 res
= mem_size
< ((avail_ram
* nm_i
->ram_thresh
/ 100) >> 1);
67 } else if (type
== EXTENT_CACHE
) {
68 mem_size
= (sbi
->total_ext_tree
* sizeof(struct extent_tree
) +
69 atomic_read(&sbi
->total_ext_node
) *
70 sizeof(struct extent_node
)) >> PAGE_CACHE_SHIFT
;
71 res
= mem_size
< ((avail_ram
* nm_i
->ram_thresh
/ 100) >> 1);
73 if (sbi
->sb
->s_bdi
->dirty_exceeded
)
79 static void clear_node_page_dirty(struct page
*page
)
81 struct address_space
*mapping
= page
->mapping
;
82 unsigned int long flags
;
84 if (PageDirty(page
)) {
85 spin_lock_irqsave(&mapping
->tree_lock
, flags
);
86 radix_tree_tag_clear(&mapping
->page_tree
,
89 spin_unlock_irqrestore(&mapping
->tree_lock
, flags
);
91 clear_page_dirty_for_io(page
);
92 dec_page_count(F2FS_M_SB(mapping
), F2FS_DIRTY_NODES
);
94 ClearPageUptodate(page
);
97 static struct page
*get_current_nat_page(struct f2fs_sb_info
*sbi
, nid_t nid
)
99 pgoff_t index
= current_nat_addr(sbi
, nid
);
100 return get_meta_page(sbi
, index
);
103 static struct page
*get_next_nat_page(struct f2fs_sb_info
*sbi
, nid_t nid
)
105 struct page
*src_page
;
106 struct page
*dst_page
;
111 struct f2fs_nm_info
*nm_i
= NM_I(sbi
);
113 src_off
= current_nat_addr(sbi
, nid
);
114 dst_off
= next_nat_addr(sbi
, src_off
);
116 /* get current nat block page with lock */
117 src_page
= get_meta_page(sbi
, src_off
);
118 dst_page
= grab_meta_page(sbi
, dst_off
);
119 f2fs_bug_on(sbi
, PageDirty(src_page
));
121 src_addr
= page_address(src_page
);
122 dst_addr
= page_address(dst_page
);
123 memcpy(dst_addr
, src_addr
, PAGE_CACHE_SIZE
);
124 set_page_dirty(dst_page
);
125 f2fs_put_page(src_page
, 1);
127 set_to_next_nat(nm_i
, nid
);
132 static struct nat_entry
*__lookup_nat_cache(struct f2fs_nm_info
*nm_i
, nid_t n
)
134 return radix_tree_lookup(&nm_i
->nat_root
, n
);
137 static unsigned int __gang_lookup_nat_cache(struct f2fs_nm_info
*nm_i
,
138 nid_t start
, unsigned int nr
, struct nat_entry
**ep
)
140 return radix_tree_gang_lookup(&nm_i
->nat_root
, (void **)ep
, start
, nr
);
143 static void __del_from_nat_cache(struct f2fs_nm_info
*nm_i
, struct nat_entry
*e
)
146 radix_tree_delete(&nm_i
->nat_root
, nat_get_nid(e
));
148 kmem_cache_free(nat_entry_slab
, e
);
151 static void __set_nat_cache_dirty(struct f2fs_nm_info
*nm_i
,
152 struct nat_entry
*ne
)
154 nid_t set
= NAT_BLOCK_OFFSET(ne
->ni
.nid
);
155 struct nat_entry_set
*head
;
157 if (get_nat_flag(ne
, IS_DIRTY
))
160 head
= radix_tree_lookup(&nm_i
->nat_set_root
, set
);
162 head
= f2fs_kmem_cache_alloc(nat_entry_set_slab
, GFP_ATOMIC
);
164 INIT_LIST_HEAD(&head
->entry_list
);
165 INIT_LIST_HEAD(&head
->set_list
);
168 f2fs_radix_tree_insert(&nm_i
->nat_set_root
, set
, head
);
170 list_move_tail(&ne
->list
, &head
->entry_list
);
171 nm_i
->dirty_nat_cnt
++;
173 set_nat_flag(ne
, IS_DIRTY
, true);
176 static void __clear_nat_cache_dirty(struct f2fs_nm_info
*nm_i
,
177 struct nat_entry
*ne
)
179 nid_t set
= NAT_BLOCK_OFFSET(ne
->ni
.nid
);
180 struct nat_entry_set
*head
;
182 head
= radix_tree_lookup(&nm_i
->nat_set_root
, set
);
184 list_move_tail(&ne
->list
, &nm_i
->nat_entries
);
185 set_nat_flag(ne
, IS_DIRTY
, false);
187 nm_i
->dirty_nat_cnt
--;
191 static unsigned int __gang_lookup_nat_set(struct f2fs_nm_info
*nm_i
,
192 nid_t start
, unsigned int nr
, struct nat_entry_set
**ep
)
194 return radix_tree_gang_lookup(&nm_i
->nat_set_root
, (void **)ep
,
198 int need_dentry_mark(struct f2fs_sb_info
*sbi
, nid_t nid
)
200 struct f2fs_nm_info
*nm_i
= NM_I(sbi
);
204 down_read(&nm_i
->nat_tree_lock
);
205 e
= __lookup_nat_cache(nm_i
, nid
);
207 if (!get_nat_flag(e
, IS_CHECKPOINTED
) &&
208 !get_nat_flag(e
, HAS_FSYNCED_INODE
))
211 up_read(&nm_i
->nat_tree_lock
);
215 bool is_checkpointed_node(struct f2fs_sb_info
*sbi
, nid_t nid
)
217 struct f2fs_nm_info
*nm_i
= NM_I(sbi
);
221 down_read(&nm_i
->nat_tree_lock
);
222 e
= __lookup_nat_cache(nm_i
, nid
);
223 if (e
&& !get_nat_flag(e
, IS_CHECKPOINTED
))
225 up_read(&nm_i
->nat_tree_lock
);
229 bool need_inode_block_update(struct f2fs_sb_info
*sbi
, nid_t ino
)
231 struct f2fs_nm_info
*nm_i
= NM_I(sbi
);
233 bool need_update
= true;
235 down_read(&nm_i
->nat_tree_lock
);
236 e
= __lookup_nat_cache(nm_i
, ino
);
237 if (e
&& get_nat_flag(e
, HAS_LAST_FSYNC
) &&
238 (get_nat_flag(e
, IS_CHECKPOINTED
) ||
239 get_nat_flag(e
, HAS_FSYNCED_INODE
)))
241 up_read(&nm_i
->nat_tree_lock
);
245 static struct nat_entry
*grab_nat_entry(struct f2fs_nm_info
*nm_i
, nid_t nid
)
247 struct nat_entry
*new;
249 new = f2fs_kmem_cache_alloc(nat_entry_slab
, GFP_ATOMIC
);
250 f2fs_radix_tree_insert(&nm_i
->nat_root
, nid
, new);
251 memset(new, 0, sizeof(struct nat_entry
));
252 nat_set_nid(new, nid
);
254 list_add_tail(&new->list
, &nm_i
->nat_entries
);
259 static void cache_nat_entry(struct f2fs_nm_info
*nm_i
, nid_t nid
,
260 struct f2fs_nat_entry
*ne
)
264 down_write(&nm_i
->nat_tree_lock
);
265 e
= __lookup_nat_cache(nm_i
, nid
);
267 e
= grab_nat_entry(nm_i
, nid
);
268 node_info_from_raw_nat(&e
->ni
, ne
);
270 up_write(&nm_i
->nat_tree_lock
);
273 static void set_node_addr(struct f2fs_sb_info
*sbi
, struct node_info
*ni
,
274 block_t new_blkaddr
, bool fsync_done
)
276 struct f2fs_nm_info
*nm_i
= NM_I(sbi
);
279 down_write(&nm_i
->nat_tree_lock
);
280 e
= __lookup_nat_cache(nm_i
, ni
->nid
);
282 e
= grab_nat_entry(nm_i
, ni
->nid
);
283 copy_node_info(&e
->ni
, ni
);
284 f2fs_bug_on(sbi
, ni
->blk_addr
== NEW_ADDR
);
285 } else if (new_blkaddr
== NEW_ADDR
) {
287 * when nid is reallocated,
288 * previous nat entry can be remained in nat cache.
289 * So, reinitialize it with new information.
291 copy_node_info(&e
->ni
, ni
);
292 f2fs_bug_on(sbi
, ni
->blk_addr
!= NULL_ADDR
);
296 f2fs_bug_on(sbi
, nat_get_blkaddr(e
) != ni
->blk_addr
);
297 f2fs_bug_on(sbi
, nat_get_blkaddr(e
) == NULL_ADDR
&&
298 new_blkaddr
== NULL_ADDR
);
299 f2fs_bug_on(sbi
, nat_get_blkaddr(e
) == NEW_ADDR
&&
300 new_blkaddr
== NEW_ADDR
);
301 f2fs_bug_on(sbi
, nat_get_blkaddr(e
) != NEW_ADDR
&&
302 nat_get_blkaddr(e
) != NULL_ADDR
&&
303 new_blkaddr
== NEW_ADDR
);
305 /* increment version no as node is removed */
306 if (nat_get_blkaddr(e
) != NEW_ADDR
&& new_blkaddr
== NULL_ADDR
) {
307 unsigned char version
= nat_get_version(e
);
308 nat_set_version(e
, inc_node_version(version
));
312 nat_set_blkaddr(e
, new_blkaddr
);
313 if (new_blkaddr
== NEW_ADDR
|| new_blkaddr
== NULL_ADDR
)
314 set_nat_flag(e
, IS_CHECKPOINTED
, false);
315 __set_nat_cache_dirty(nm_i
, e
);
317 /* update fsync_mark if its inode nat entry is still alive */
318 e
= __lookup_nat_cache(nm_i
, ni
->ino
);
320 if (fsync_done
&& ni
->nid
== ni
->ino
)
321 set_nat_flag(e
, HAS_FSYNCED_INODE
, true);
322 set_nat_flag(e
, HAS_LAST_FSYNC
, fsync_done
);
324 up_write(&nm_i
->nat_tree_lock
);
327 int try_to_free_nats(struct f2fs_sb_info
*sbi
, int nr_shrink
)
329 struct f2fs_nm_info
*nm_i
= NM_I(sbi
);
331 if (available_free_memory(sbi
, NAT_ENTRIES
))
334 down_write(&nm_i
->nat_tree_lock
);
335 while (nr_shrink
&& !list_empty(&nm_i
->nat_entries
)) {
336 struct nat_entry
*ne
;
337 ne
= list_first_entry(&nm_i
->nat_entries
,
338 struct nat_entry
, list
);
339 __del_from_nat_cache(nm_i
, ne
);
342 up_write(&nm_i
->nat_tree_lock
);
347 * This function always returns success
349 void get_node_info(struct f2fs_sb_info
*sbi
, nid_t nid
, struct node_info
*ni
)
351 struct f2fs_nm_info
*nm_i
= NM_I(sbi
);
352 struct curseg_info
*curseg
= CURSEG_I(sbi
, CURSEG_HOT_DATA
);
353 struct f2fs_summary_block
*sum
= curseg
->sum_blk
;
354 nid_t start_nid
= START_NID(nid
);
355 struct f2fs_nat_block
*nat_blk
;
356 struct page
*page
= NULL
;
357 struct f2fs_nat_entry ne
;
363 /* Check nat cache */
364 down_read(&nm_i
->nat_tree_lock
);
365 e
= __lookup_nat_cache(nm_i
, nid
);
367 ni
->ino
= nat_get_ino(e
);
368 ni
->blk_addr
= nat_get_blkaddr(e
);
369 ni
->version
= nat_get_version(e
);
371 up_read(&nm_i
->nat_tree_lock
);
375 memset(&ne
, 0, sizeof(struct f2fs_nat_entry
));
377 /* Check current segment summary */
378 mutex_lock(&curseg
->curseg_mutex
);
379 i
= lookup_journal_in_cursum(sum
, NAT_JOURNAL
, nid
, 0);
381 ne
= nat_in_journal(sum
, i
);
382 node_info_from_raw_nat(ni
, &ne
);
384 mutex_unlock(&curseg
->curseg_mutex
);
388 /* Fill node_info from nat page */
389 page
= get_current_nat_page(sbi
, start_nid
);
390 nat_blk
= (struct f2fs_nat_block
*)page_address(page
);
391 ne
= nat_blk
->entries
[nid
- start_nid
];
392 node_info_from_raw_nat(ni
, &ne
);
393 f2fs_put_page(page
, 1);
395 /* cache nat entry */
396 cache_nat_entry(NM_I(sbi
), nid
, &ne
);
400 * The maximum depth is four.
401 * Offset[0] will have raw inode offset.
403 static int get_node_path(struct f2fs_inode_info
*fi
, long block
,
404 int offset
[4], unsigned int noffset
[4])
406 const long direct_index
= ADDRS_PER_INODE(fi
);
407 const long direct_blks
= ADDRS_PER_BLOCK
;
408 const long dptrs_per_blk
= NIDS_PER_BLOCK
;
409 const long indirect_blks
= ADDRS_PER_BLOCK
* NIDS_PER_BLOCK
;
410 const long dindirect_blks
= indirect_blks
* NIDS_PER_BLOCK
;
416 if (block
< direct_index
) {
420 block
-= direct_index
;
421 if (block
< direct_blks
) {
422 offset
[n
++] = NODE_DIR1_BLOCK
;
428 block
-= direct_blks
;
429 if (block
< direct_blks
) {
430 offset
[n
++] = NODE_DIR2_BLOCK
;
436 block
-= direct_blks
;
437 if (block
< indirect_blks
) {
438 offset
[n
++] = NODE_IND1_BLOCK
;
440 offset
[n
++] = block
/ direct_blks
;
441 noffset
[n
] = 4 + offset
[n
- 1];
442 offset
[n
] = block
% direct_blks
;
446 block
-= indirect_blks
;
447 if (block
< indirect_blks
) {
448 offset
[n
++] = NODE_IND2_BLOCK
;
449 noffset
[n
] = 4 + dptrs_per_blk
;
450 offset
[n
++] = block
/ direct_blks
;
451 noffset
[n
] = 5 + dptrs_per_blk
+ offset
[n
- 1];
452 offset
[n
] = block
% direct_blks
;
456 block
-= indirect_blks
;
457 if (block
< dindirect_blks
) {
458 offset
[n
++] = NODE_DIND_BLOCK
;
459 noffset
[n
] = 5 + (dptrs_per_blk
* 2);
460 offset
[n
++] = block
/ indirect_blks
;
461 noffset
[n
] = 6 + (dptrs_per_blk
* 2) +
462 offset
[n
- 1] * (dptrs_per_blk
+ 1);
463 offset
[n
++] = (block
/ direct_blks
) % dptrs_per_blk
;
464 noffset
[n
] = 7 + (dptrs_per_blk
* 2) +
465 offset
[n
- 2] * (dptrs_per_blk
+ 1) +
467 offset
[n
] = block
% direct_blks
;
478 * Caller should call f2fs_put_dnode(dn).
479 * Also, it should grab and release a rwsem by calling f2fs_lock_op() and
480 * f2fs_unlock_op() only if ro is not set RDONLY_NODE.
481 * In the case of RDONLY_NODE, we don't need to care about mutex.
483 int get_dnode_of_data(struct dnode_of_data
*dn
, pgoff_t index
, int mode
)
485 struct f2fs_sb_info
*sbi
= F2FS_I_SB(dn
->inode
);
486 struct page
*npage
[4];
487 struct page
*parent
= NULL
;
489 unsigned int noffset
[4];
494 level
= get_node_path(F2FS_I(dn
->inode
), index
, offset
, noffset
);
496 nids
[0] = dn
->inode
->i_ino
;
497 npage
[0] = dn
->inode_page
;
500 npage
[0] = get_node_page(sbi
, nids
[0]);
501 if (IS_ERR(npage
[0]))
502 return PTR_ERR(npage
[0]);
505 /* if inline_data is set, should not report any block indices */
506 if (f2fs_has_inline_data(dn
->inode
) && index
) {
508 f2fs_put_page(npage
[0], 1);
514 nids
[1] = get_nid(parent
, offset
[0], true);
515 dn
->inode_page
= npage
[0];
516 dn
->inode_page_locked
= true;
518 /* get indirect or direct nodes */
519 for (i
= 1; i
<= level
; i
++) {
522 if (!nids
[i
] && mode
== ALLOC_NODE
) {
524 if (!alloc_nid(sbi
, &(nids
[i
]))) {
530 npage
[i
] = new_node_page(dn
, noffset
[i
], NULL
);
531 if (IS_ERR(npage
[i
])) {
532 alloc_nid_failed(sbi
, nids
[i
]);
533 err
= PTR_ERR(npage
[i
]);
537 set_nid(parent
, offset
[i
- 1], nids
[i
], i
== 1);
538 alloc_nid_done(sbi
, nids
[i
]);
540 } else if (mode
== LOOKUP_NODE_RA
&& i
== level
&& level
> 1) {
541 npage
[i
] = get_node_page_ra(parent
, offset
[i
- 1]);
542 if (IS_ERR(npage
[i
])) {
543 err
= PTR_ERR(npage
[i
]);
549 dn
->inode_page_locked
= false;
552 f2fs_put_page(parent
, 1);
556 npage
[i
] = get_node_page(sbi
, nids
[i
]);
557 if (IS_ERR(npage
[i
])) {
558 err
= PTR_ERR(npage
[i
]);
559 f2fs_put_page(npage
[0], 0);
565 nids
[i
+ 1] = get_nid(parent
, offset
[i
], false);
568 dn
->nid
= nids
[level
];
569 dn
->ofs_in_node
= offset
[level
];
570 dn
->node_page
= npage
[level
];
571 dn
->data_blkaddr
= datablock_addr(dn
->node_page
, dn
->ofs_in_node
);
575 f2fs_put_page(parent
, 1);
577 f2fs_put_page(npage
[0], 0);
579 dn
->inode_page
= NULL
;
580 dn
->node_page
= NULL
;
584 static void truncate_node(struct dnode_of_data
*dn
)
586 struct f2fs_sb_info
*sbi
= F2FS_I_SB(dn
->inode
);
589 get_node_info(sbi
, dn
->nid
, &ni
);
590 if (dn
->inode
->i_blocks
== 0) {
591 f2fs_bug_on(sbi
, ni
.blk_addr
!= NULL_ADDR
);
594 f2fs_bug_on(sbi
, ni
.blk_addr
== NULL_ADDR
);
596 /* Deallocate node address */
597 invalidate_blocks(sbi
, ni
.blk_addr
);
598 dec_valid_node_count(sbi
, dn
->inode
);
599 set_node_addr(sbi
, &ni
, NULL_ADDR
, false);
601 if (dn
->nid
== dn
->inode
->i_ino
) {
602 remove_orphan_inode(sbi
, dn
->nid
);
603 dec_valid_inode_count(sbi
);
608 clear_node_page_dirty(dn
->node_page
);
609 set_sbi_flag(sbi
, SBI_IS_DIRTY
);
611 f2fs_put_page(dn
->node_page
, 1);
613 invalidate_mapping_pages(NODE_MAPPING(sbi
),
614 dn
->node_page
->index
, dn
->node_page
->index
);
616 dn
->node_page
= NULL
;
617 trace_f2fs_truncate_node(dn
->inode
, dn
->nid
, ni
.blk_addr
);
620 static int truncate_dnode(struct dnode_of_data
*dn
)
627 /* get direct node */
628 page
= get_node_page(F2FS_I_SB(dn
->inode
), dn
->nid
);
629 if (IS_ERR(page
) && PTR_ERR(page
) == -ENOENT
)
631 else if (IS_ERR(page
))
632 return PTR_ERR(page
);
634 /* Make dnode_of_data for parameter */
635 dn
->node_page
= page
;
637 truncate_data_blocks(dn
);
642 static int truncate_nodes(struct dnode_of_data
*dn
, unsigned int nofs
,
645 struct dnode_of_data rdn
= *dn
;
647 struct f2fs_node
*rn
;
649 unsigned int child_nofs
;
654 return NIDS_PER_BLOCK
+ 1;
656 trace_f2fs_truncate_nodes_enter(dn
->inode
, dn
->nid
, dn
->data_blkaddr
);
658 page
= get_node_page(F2FS_I_SB(dn
->inode
), dn
->nid
);
660 trace_f2fs_truncate_nodes_exit(dn
->inode
, PTR_ERR(page
));
661 return PTR_ERR(page
);
664 rn
= F2FS_NODE(page
);
666 for (i
= ofs
; i
< NIDS_PER_BLOCK
; i
++, freed
++) {
667 child_nid
= le32_to_cpu(rn
->in
.nid
[i
]);
671 ret
= truncate_dnode(&rdn
);
674 set_nid(page
, i
, 0, false);
677 child_nofs
= nofs
+ ofs
* (NIDS_PER_BLOCK
+ 1) + 1;
678 for (i
= ofs
; i
< NIDS_PER_BLOCK
; i
++) {
679 child_nid
= le32_to_cpu(rn
->in
.nid
[i
]);
680 if (child_nid
== 0) {
681 child_nofs
+= NIDS_PER_BLOCK
+ 1;
685 ret
= truncate_nodes(&rdn
, child_nofs
, 0, depth
- 1);
686 if (ret
== (NIDS_PER_BLOCK
+ 1)) {
687 set_nid(page
, i
, 0, false);
689 } else if (ret
< 0 && ret
!= -ENOENT
) {
697 /* remove current indirect node */
698 dn
->node_page
= page
;
702 f2fs_put_page(page
, 1);
704 trace_f2fs_truncate_nodes_exit(dn
->inode
, freed
);
708 f2fs_put_page(page
, 1);
709 trace_f2fs_truncate_nodes_exit(dn
->inode
, ret
);
713 static int truncate_partial_nodes(struct dnode_of_data
*dn
,
714 struct f2fs_inode
*ri
, int *offset
, int depth
)
716 struct page
*pages
[2];
723 nid
[0] = le32_to_cpu(ri
->i_nid
[offset
[0] - NODE_DIR1_BLOCK
]);
727 /* get indirect nodes in the path */
728 for (i
= 0; i
< idx
+ 1; i
++) {
729 /* reference count'll be increased */
730 pages
[i
] = get_node_page(F2FS_I_SB(dn
->inode
), nid
[i
]);
731 if (IS_ERR(pages
[i
])) {
732 err
= PTR_ERR(pages
[i
]);
736 nid
[i
+ 1] = get_nid(pages
[i
], offset
[i
+ 1], false);
739 /* free direct nodes linked to a partial indirect node */
740 for (i
= offset
[idx
+ 1]; i
< NIDS_PER_BLOCK
; i
++) {
741 child_nid
= get_nid(pages
[idx
], i
, false);
745 err
= truncate_dnode(dn
);
748 set_nid(pages
[idx
], i
, 0, false);
751 if (offset
[idx
+ 1] == 0) {
752 dn
->node_page
= pages
[idx
];
756 f2fs_put_page(pages
[idx
], 1);
762 for (i
= idx
; i
>= 0; i
--)
763 f2fs_put_page(pages
[i
], 1);
765 trace_f2fs_truncate_partial_nodes(dn
->inode
, nid
, depth
, err
);
771 * All the block addresses of data and nodes should be nullified.
773 int truncate_inode_blocks(struct inode
*inode
, pgoff_t from
)
775 struct f2fs_sb_info
*sbi
= F2FS_I_SB(inode
);
776 int err
= 0, cont
= 1;
777 int level
, offset
[4], noffset
[4];
778 unsigned int nofs
= 0;
779 struct f2fs_inode
*ri
;
780 struct dnode_of_data dn
;
783 trace_f2fs_truncate_inode_blocks_enter(inode
, from
);
785 level
= get_node_path(F2FS_I(inode
), from
, offset
, noffset
);
787 page
= get_node_page(sbi
, inode
->i_ino
);
789 trace_f2fs_truncate_inode_blocks_exit(inode
, PTR_ERR(page
));
790 return PTR_ERR(page
);
793 set_new_dnode(&dn
, inode
, page
, NULL
, 0);
796 ri
= F2FS_INODE(page
);
804 if (!offset
[level
- 1])
806 err
= truncate_partial_nodes(&dn
, ri
, offset
, level
);
807 if (err
< 0 && err
!= -ENOENT
)
809 nofs
+= 1 + NIDS_PER_BLOCK
;
812 nofs
= 5 + 2 * NIDS_PER_BLOCK
;
813 if (!offset
[level
- 1])
815 err
= truncate_partial_nodes(&dn
, ri
, offset
, level
);
816 if (err
< 0 && err
!= -ENOENT
)
825 dn
.nid
= le32_to_cpu(ri
->i_nid
[offset
[0] - NODE_DIR1_BLOCK
]);
827 case NODE_DIR1_BLOCK
:
828 case NODE_DIR2_BLOCK
:
829 err
= truncate_dnode(&dn
);
832 case NODE_IND1_BLOCK
:
833 case NODE_IND2_BLOCK
:
834 err
= truncate_nodes(&dn
, nofs
, offset
[1], 2);
837 case NODE_DIND_BLOCK
:
838 err
= truncate_nodes(&dn
, nofs
, offset
[1], 3);
845 if (err
< 0 && err
!= -ENOENT
)
847 if (offset
[1] == 0 &&
848 ri
->i_nid
[offset
[0] - NODE_DIR1_BLOCK
]) {
850 if (unlikely(page
->mapping
!= NODE_MAPPING(sbi
))) {
851 f2fs_put_page(page
, 1);
854 f2fs_wait_on_page_writeback(page
, NODE
);
855 ri
->i_nid
[offset
[0] - NODE_DIR1_BLOCK
] = 0;
856 set_page_dirty(page
);
864 f2fs_put_page(page
, 0);
865 trace_f2fs_truncate_inode_blocks_exit(inode
, err
);
866 return err
> 0 ? 0 : err
;
869 int truncate_xattr_node(struct inode
*inode
, struct page
*page
)
871 struct f2fs_sb_info
*sbi
= F2FS_I_SB(inode
);
872 nid_t nid
= F2FS_I(inode
)->i_xattr_nid
;
873 struct dnode_of_data dn
;
879 npage
= get_node_page(sbi
, nid
);
881 return PTR_ERR(npage
);
883 F2FS_I(inode
)->i_xattr_nid
= 0;
885 /* need to do checkpoint during fsync */
886 F2FS_I(inode
)->xattr_ver
= cur_cp_version(F2FS_CKPT(sbi
));
888 set_new_dnode(&dn
, inode
, page
, npage
, nid
);
891 dn
.inode_page_locked
= true;
897 * Caller should grab and release a rwsem by calling f2fs_lock_op() and
900 void remove_inode_page(struct inode
*inode
)
902 struct dnode_of_data dn
;
904 set_new_dnode(&dn
, inode
, NULL
, NULL
, inode
->i_ino
);
905 if (get_dnode_of_data(&dn
, 0, LOOKUP_NODE
))
908 if (truncate_xattr_node(inode
, dn
.inode_page
)) {
913 /* remove potential inline_data blocks */
914 if (S_ISREG(inode
->i_mode
) || S_ISDIR(inode
->i_mode
) ||
915 S_ISLNK(inode
->i_mode
))
916 truncate_data_blocks_range(&dn
, 1);
918 /* 0 is possible, after f2fs_new_inode() has failed */
919 f2fs_bug_on(F2FS_I_SB(inode
),
920 inode
->i_blocks
!= 0 && inode
->i_blocks
!= 1);
922 /* will put inode & node pages */
926 struct page
*new_inode_page(struct inode
*inode
)
928 struct dnode_of_data dn
;
930 /* allocate inode page for new inode */
931 set_new_dnode(&dn
, inode
, NULL
, NULL
, inode
->i_ino
);
933 /* caller should f2fs_put_page(page, 1); */
934 return new_node_page(&dn
, 0, NULL
);
937 struct page
*new_node_page(struct dnode_of_data
*dn
,
938 unsigned int ofs
, struct page
*ipage
)
940 struct f2fs_sb_info
*sbi
= F2FS_I_SB(dn
->inode
);
941 struct node_info old_ni
, new_ni
;
945 if (unlikely(is_inode_flag_set(F2FS_I(dn
->inode
), FI_NO_ALLOC
)))
946 return ERR_PTR(-EPERM
);
948 page
= grab_cache_page(NODE_MAPPING(sbi
), dn
->nid
);
950 return ERR_PTR(-ENOMEM
);
952 if (unlikely(!inc_valid_node_count(sbi
, dn
->inode
))) {
957 get_node_info(sbi
, dn
->nid
, &old_ni
);
959 /* Reinitialize old_ni with new node page */
960 f2fs_bug_on(sbi
, old_ni
.blk_addr
!= NULL_ADDR
);
962 new_ni
.ino
= dn
->inode
->i_ino
;
963 set_node_addr(sbi
, &new_ni
, NEW_ADDR
, false);
965 f2fs_wait_on_page_writeback(page
, NODE
);
966 fill_node_footer(page
, dn
->nid
, dn
->inode
->i_ino
, ofs
, true);
967 set_cold_node(dn
->inode
, page
);
968 SetPageUptodate(page
);
969 set_page_dirty(page
);
971 if (f2fs_has_xattr_block(ofs
))
972 F2FS_I(dn
->inode
)->i_xattr_nid
= dn
->nid
;
974 dn
->node_page
= page
;
976 update_inode(dn
->inode
, ipage
);
980 inc_valid_inode_count(sbi
);
985 clear_node_page_dirty(page
);
986 f2fs_put_page(page
, 1);
991 * Caller should do after getting the following values.
992 * 0: f2fs_put_page(page, 0)
993 * LOCKED_PAGE: f2fs_put_page(page, 1)
996 static int read_node_page(struct page
*page
, int rw
)
998 struct f2fs_sb_info
*sbi
= F2FS_P_SB(page
);
1000 struct f2fs_io_info fio
= {
1007 get_node_info(sbi
, page
->index
, &ni
);
1009 if (unlikely(ni
.blk_addr
== NULL_ADDR
)) {
1010 ClearPageUptodate(page
);
1011 f2fs_put_page(page
, 1);
1015 if (PageUptodate(page
))
1018 fio
.blk_addr
= ni
.blk_addr
;
1019 return f2fs_submit_page_bio(&fio
);
1023 * Readahead a node page
1025 void ra_node_page(struct f2fs_sb_info
*sbi
, nid_t nid
)
1030 apage
= find_get_page(NODE_MAPPING(sbi
), nid
);
1031 if (apage
&& PageUptodate(apage
)) {
1032 f2fs_put_page(apage
, 0);
1035 f2fs_put_page(apage
, 0);
1037 apage
= grab_cache_page(NODE_MAPPING(sbi
), nid
);
1041 err
= read_node_page(apage
, READA
);
1043 f2fs_put_page(apage
, 0);
1044 else if (err
== LOCKED_PAGE
)
1045 f2fs_put_page(apage
, 1);
1048 struct page
*get_node_page(struct f2fs_sb_info
*sbi
, pgoff_t nid
)
1053 page
= grab_cache_page(NODE_MAPPING(sbi
), nid
);
1055 return ERR_PTR(-ENOMEM
);
1057 err
= read_node_page(page
, READ_SYNC
);
1059 return ERR_PTR(err
);
1060 else if (err
!= LOCKED_PAGE
)
1063 if (unlikely(!PageUptodate(page
) || nid
!= nid_of_node(page
))) {
1064 ClearPageUptodate(page
);
1065 f2fs_put_page(page
, 1);
1066 return ERR_PTR(-EIO
);
1068 if (unlikely(page
->mapping
!= NODE_MAPPING(sbi
))) {
1069 f2fs_put_page(page
, 1);
1076 * Return a locked page for the desired node page.
1077 * And, readahead MAX_RA_NODE number of node pages.
1079 struct page
*get_node_page_ra(struct page
*parent
, int start
)
1081 struct f2fs_sb_info
*sbi
= F2FS_P_SB(parent
);
1082 struct blk_plug plug
;
1087 /* First, try getting the desired direct node. */
1088 nid
= get_nid(parent
, start
, false);
1090 return ERR_PTR(-ENOENT
);
1092 page
= grab_cache_page(NODE_MAPPING(sbi
), nid
);
1094 return ERR_PTR(-ENOMEM
);
1096 err
= read_node_page(page
, READ_SYNC
);
1098 return ERR_PTR(err
);
1099 else if (err
== LOCKED_PAGE
)
1102 blk_start_plug(&plug
);
1104 /* Then, try readahead for siblings of the desired node */
1105 end
= start
+ MAX_RA_NODE
;
1106 end
= min(end
, NIDS_PER_BLOCK
);
1107 for (i
= start
+ 1; i
< end
; i
++) {
1108 nid
= get_nid(parent
, i
, false);
1111 ra_node_page(sbi
, nid
);
1114 blk_finish_plug(&plug
);
1117 if (unlikely(page
->mapping
!= NODE_MAPPING(sbi
))) {
1118 f2fs_put_page(page
, 1);
1122 if (unlikely(!PageUptodate(page
))) {
1123 f2fs_put_page(page
, 1);
1124 return ERR_PTR(-EIO
);
1129 void sync_inode_page(struct dnode_of_data
*dn
)
1131 if (IS_INODE(dn
->node_page
) || dn
->inode_page
== dn
->node_page
) {
1132 update_inode(dn
->inode
, dn
->node_page
);
1133 } else if (dn
->inode_page
) {
1134 if (!dn
->inode_page_locked
)
1135 lock_page(dn
->inode_page
);
1136 update_inode(dn
->inode
, dn
->inode_page
);
1137 if (!dn
->inode_page_locked
)
1138 unlock_page(dn
->inode_page
);
1140 update_inode_page(dn
->inode
);
1144 int sync_node_pages(struct f2fs_sb_info
*sbi
, nid_t ino
,
1145 struct writeback_control
*wbc
)
1148 struct pagevec pvec
;
1149 int step
= ino
? 2 : 0;
1150 int nwritten
= 0, wrote
= 0;
1152 pagevec_init(&pvec
, 0);
1158 while (index
<= end
) {
1160 nr_pages
= pagevec_lookup_tag(&pvec
, NODE_MAPPING(sbi
), &index
,
1161 PAGECACHE_TAG_DIRTY
,
1162 min(end
- index
, (pgoff_t
)PAGEVEC_SIZE
-1) + 1);
1166 for (i
= 0; i
< nr_pages
; i
++) {
1167 struct page
*page
= pvec
.pages
[i
];
1170 * flushing sequence with step:
1175 if (step
== 0 && IS_DNODE(page
))
1177 if (step
== 1 && (!IS_DNODE(page
) ||
1178 is_cold_node(page
)))
1180 if (step
== 2 && (!IS_DNODE(page
) ||
1181 !is_cold_node(page
)))
1186 * we should not skip writing node pages.
1188 if (ino
&& ino_of_node(page
) == ino
)
1190 else if (!trylock_page(page
))
1193 if (unlikely(page
->mapping
!= NODE_MAPPING(sbi
))) {
1198 if (ino
&& ino_of_node(page
) != ino
)
1199 goto continue_unlock
;
1201 if (!PageDirty(page
)) {
1202 /* someone wrote it for us */
1203 goto continue_unlock
;
1206 if (!clear_page_dirty_for_io(page
))
1207 goto continue_unlock
;
1209 /* called by fsync() */
1210 if (ino
&& IS_DNODE(page
)) {
1211 set_fsync_mark(page
, 1);
1213 set_dentry_mark(page
,
1214 need_dentry_mark(sbi
, ino
));
1217 set_fsync_mark(page
, 0);
1218 set_dentry_mark(page
, 0);
1221 if (NODE_MAPPING(sbi
)->a_ops
->writepage(page
, wbc
))
1226 if (--wbc
->nr_to_write
== 0)
1229 pagevec_release(&pvec
);
1232 if (wbc
->nr_to_write
== 0) {
1244 f2fs_submit_merged_bio(sbi
, NODE
, WRITE
);
1248 int wait_on_node_pages_writeback(struct f2fs_sb_info
*sbi
, nid_t ino
)
1250 pgoff_t index
= 0, end
= LONG_MAX
;
1251 struct pagevec pvec
;
1252 int ret2
= 0, ret
= 0;
1254 pagevec_init(&pvec
, 0);
1256 while (index
<= end
) {
1258 nr_pages
= pagevec_lookup_tag(&pvec
, NODE_MAPPING(sbi
), &index
,
1259 PAGECACHE_TAG_WRITEBACK
,
1260 min(end
- index
, (pgoff_t
)PAGEVEC_SIZE
-1) + 1);
1264 for (i
= 0; i
< nr_pages
; i
++) {
1265 struct page
*page
= pvec
.pages
[i
];
1267 /* until radix tree lookup accepts end_index */
1268 if (unlikely(page
->index
> end
))
1271 if (ino
&& ino_of_node(page
) == ino
) {
1272 f2fs_wait_on_page_writeback(page
, NODE
);
1273 if (TestClearPageError(page
))
1277 pagevec_release(&pvec
);
1281 if (unlikely(test_and_clear_bit(AS_ENOSPC
, &NODE_MAPPING(sbi
)->flags
)))
1283 if (unlikely(test_and_clear_bit(AS_EIO
, &NODE_MAPPING(sbi
)->flags
)))
1290 static int f2fs_write_node_page(struct page
*page
,
1291 struct writeback_control
*wbc
)
1293 struct f2fs_sb_info
*sbi
= F2FS_P_SB(page
);
1295 struct node_info ni
;
1296 struct f2fs_io_info fio
= {
1299 .rw
= (wbc
->sync_mode
== WB_SYNC_ALL
) ? WRITE_SYNC
: WRITE
,
1303 trace_f2fs_writepage(page
, NODE
);
1305 if (unlikely(is_sbi_flag_set(sbi
, SBI_POR_DOING
)))
1307 if (unlikely(f2fs_cp_error(sbi
)))
1310 f2fs_wait_on_page_writeback(page
, NODE
);
1312 /* get old block addr of this node page */
1313 nid
= nid_of_node(page
);
1314 f2fs_bug_on(sbi
, page
->index
!= nid
);
1316 get_node_info(sbi
, nid
, &ni
);
1318 /* This page is already truncated */
1319 if (unlikely(ni
.blk_addr
== NULL_ADDR
)) {
1320 ClearPageUptodate(page
);
1321 dec_page_count(sbi
, F2FS_DIRTY_NODES
);
1326 if (wbc
->for_reclaim
) {
1327 if (!down_read_trylock(&sbi
->node_write
))
1330 down_read(&sbi
->node_write
);
1333 set_page_writeback(page
);
1334 fio
.blk_addr
= ni
.blk_addr
;
1335 write_node_page(nid
, &fio
);
1336 set_node_addr(sbi
, &ni
, fio
.blk_addr
, is_fsync_dnode(page
));
1337 dec_page_count(sbi
, F2FS_DIRTY_NODES
);
1338 up_read(&sbi
->node_write
);
1341 if (wbc
->for_reclaim
)
1342 f2fs_submit_merged_bio(sbi
, NODE
, WRITE
);
1347 redirty_page_for_writepage(wbc
, page
);
1348 return AOP_WRITEPAGE_ACTIVATE
;
1351 static int f2fs_write_node_pages(struct address_space
*mapping
,
1352 struct writeback_control
*wbc
)
1354 struct f2fs_sb_info
*sbi
= F2FS_M_SB(mapping
);
1357 trace_f2fs_writepages(mapping
->host
, wbc
, NODE
);
1359 /* balancing f2fs's metadata in background */
1360 f2fs_balance_fs_bg(sbi
);
1362 /* collect a number of dirty node pages and write together */
1363 if (get_pages(sbi
, F2FS_DIRTY_NODES
) < nr_pages_to_skip(sbi
, NODE
))
1366 diff
= nr_pages_to_write(sbi
, NODE
, wbc
);
1367 wbc
->sync_mode
= WB_SYNC_NONE
;
1368 sync_node_pages(sbi
, 0, wbc
);
1369 wbc
->nr_to_write
= max((long)0, wbc
->nr_to_write
- diff
);
1373 wbc
->pages_skipped
+= get_pages(sbi
, F2FS_DIRTY_NODES
);
1377 static int f2fs_set_node_page_dirty(struct page
*page
)
1379 trace_f2fs_set_page_dirty(page
, NODE
);
1381 SetPageUptodate(page
);
1382 if (!PageDirty(page
)) {
1383 __set_page_dirty_nobuffers(page
);
1384 inc_page_count(F2FS_P_SB(page
), F2FS_DIRTY_NODES
);
1385 SetPagePrivate(page
);
1386 f2fs_trace_pid(page
);
1393 * Structure of the f2fs node operations
1395 const struct address_space_operations f2fs_node_aops
= {
1396 .writepage
= f2fs_write_node_page
,
1397 .writepages
= f2fs_write_node_pages
,
1398 .set_page_dirty
= f2fs_set_node_page_dirty
,
1399 .invalidatepage
= f2fs_invalidate_page
,
1400 .releasepage
= f2fs_release_page
,
1403 static struct free_nid
*__lookup_free_nid_list(struct f2fs_nm_info
*nm_i
,
1406 return radix_tree_lookup(&nm_i
->free_nid_root
, n
);
1409 static void __del_from_free_nid_list(struct f2fs_nm_info
*nm_i
,
1413 radix_tree_delete(&nm_i
->free_nid_root
, i
->nid
);
1416 static int add_free_nid(struct f2fs_sb_info
*sbi
, nid_t nid
, bool build
)
1418 struct f2fs_nm_info
*nm_i
= NM_I(sbi
);
1420 struct nat_entry
*ne
;
1421 bool allocated
= false;
1423 if (!available_free_memory(sbi
, FREE_NIDS
))
1426 /* 0 nid should not be used */
1427 if (unlikely(nid
== 0))
1431 /* do not add allocated nids */
1432 down_read(&nm_i
->nat_tree_lock
);
1433 ne
= __lookup_nat_cache(nm_i
, nid
);
1435 (!get_nat_flag(ne
, IS_CHECKPOINTED
) ||
1436 nat_get_blkaddr(ne
) != NULL_ADDR
))
1438 up_read(&nm_i
->nat_tree_lock
);
1443 i
= f2fs_kmem_cache_alloc(free_nid_slab
, GFP_NOFS
);
1447 if (radix_tree_preload(GFP_NOFS
)) {
1448 kmem_cache_free(free_nid_slab
, i
);
1452 spin_lock(&nm_i
->free_nid_list_lock
);
1453 if (radix_tree_insert(&nm_i
->free_nid_root
, i
->nid
, i
)) {
1454 spin_unlock(&nm_i
->free_nid_list_lock
);
1455 radix_tree_preload_end();
1456 kmem_cache_free(free_nid_slab
, i
);
1459 list_add_tail(&i
->list
, &nm_i
->free_nid_list
);
1461 spin_unlock(&nm_i
->free_nid_list_lock
);
1462 radix_tree_preload_end();
1466 static void remove_free_nid(struct f2fs_nm_info
*nm_i
, nid_t nid
)
1469 bool need_free
= false;
1471 spin_lock(&nm_i
->free_nid_list_lock
);
1472 i
= __lookup_free_nid_list(nm_i
, nid
);
1473 if (i
&& i
->state
== NID_NEW
) {
1474 __del_from_free_nid_list(nm_i
, i
);
1478 spin_unlock(&nm_i
->free_nid_list_lock
);
1481 kmem_cache_free(free_nid_slab
, i
);
1484 static void scan_nat_page(struct f2fs_sb_info
*sbi
,
1485 struct page
*nat_page
, nid_t start_nid
)
1487 struct f2fs_nm_info
*nm_i
= NM_I(sbi
);
1488 struct f2fs_nat_block
*nat_blk
= page_address(nat_page
);
1492 i
= start_nid
% NAT_ENTRY_PER_BLOCK
;
1494 for (; i
< NAT_ENTRY_PER_BLOCK
; i
++, start_nid
++) {
1496 if (unlikely(start_nid
>= nm_i
->max_nid
))
1499 blk_addr
= le32_to_cpu(nat_blk
->entries
[i
].block_addr
);
1500 f2fs_bug_on(sbi
, blk_addr
== NEW_ADDR
);
1501 if (blk_addr
== NULL_ADDR
) {
1502 if (add_free_nid(sbi
, start_nid
, true) < 0)
1508 static void build_free_nids(struct f2fs_sb_info
*sbi
)
1510 struct f2fs_nm_info
*nm_i
= NM_I(sbi
);
1511 struct curseg_info
*curseg
= CURSEG_I(sbi
, CURSEG_HOT_DATA
);
1512 struct f2fs_summary_block
*sum
= curseg
->sum_blk
;
1514 nid_t nid
= nm_i
->next_scan_nid
;
1516 /* Enough entries */
1517 if (nm_i
->fcnt
> NAT_ENTRY_PER_BLOCK
)
1520 /* readahead nat pages to be scanned */
1521 ra_meta_pages(sbi
, NAT_BLOCK_OFFSET(nid
), FREE_NID_PAGES
, META_NAT
);
1524 struct page
*page
= get_current_nat_page(sbi
, nid
);
1526 scan_nat_page(sbi
, page
, nid
);
1527 f2fs_put_page(page
, 1);
1529 nid
+= (NAT_ENTRY_PER_BLOCK
- (nid
% NAT_ENTRY_PER_BLOCK
));
1530 if (unlikely(nid
>= nm_i
->max_nid
))
1533 if (i
++ == FREE_NID_PAGES
)
1537 /* go to the next free nat pages to find free nids abundantly */
1538 nm_i
->next_scan_nid
= nid
;
1540 /* find free nids from current sum_pages */
1541 mutex_lock(&curseg
->curseg_mutex
);
1542 for (i
= 0; i
< nats_in_cursum(sum
); i
++) {
1543 block_t addr
= le32_to_cpu(nat_in_journal(sum
, i
).block_addr
);
1544 nid
= le32_to_cpu(nid_in_journal(sum
, i
));
1545 if (addr
== NULL_ADDR
)
1546 add_free_nid(sbi
, nid
, true);
1548 remove_free_nid(nm_i
, nid
);
1550 mutex_unlock(&curseg
->curseg_mutex
);
1554 * If this function returns success, caller can obtain a new nid
1555 * from second parameter of this function.
1556 * The returned nid could be used ino as well as nid when inode is created.
1558 bool alloc_nid(struct f2fs_sb_info
*sbi
, nid_t
*nid
)
1560 struct f2fs_nm_info
*nm_i
= NM_I(sbi
);
1561 struct free_nid
*i
= NULL
;
1563 if (unlikely(sbi
->total_valid_node_count
+ 1 > nm_i
->available_nids
))
1566 spin_lock(&nm_i
->free_nid_list_lock
);
1568 /* We should not use stale free nids created by build_free_nids */
1569 if (nm_i
->fcnt
&& !on_build_free_nids(nm_i
)) {
1570 f2fs_bug_on(sbi
, list_empty(&nm_i
->free_nid_list
));
1571 list_for_each_entry(i
, &nm_i
->free_nid_list
, list
)
1572 if (i
->state
== NID_NEW
)
1575 f2fs_bug_on(sbi
, i
->state
!= NID_NEW
);
1577 i
->state
= NID_ALLOC
;
1579 spin_unlock(&nm_i
->free_nid_list_lock
);
1582 spin_unlock(&nm_i
->free_nid_list_lock
);
1584 /* Let's scan nat pages and its caches to get free nids */
1585 mutex_lock(&nm_i
->build_lock
);
1586 build_free_nids(sbi
);
1587 mutex_unlock(&nm_i
->build_lock
);
1592 * alloc_nid() should be called prior to this function.
1594 void alloc_nid_done(struct f2fs_sb_info
*sbi
, nid_t nid
)
1596 struct f2fs_nm_info
*nm_i
= NM_I(sbi
);
1599 spin_lock(&nm_i
->free_nid_list_lock
);
1600 i
= __lookup_free_nid_list(nm_i
, nid
);
1601 f2fs_bug_on(sbi
, !i
|| i
->state
!= NID_ALLOC
);
1602 __del_from_free_nid_list(nm_i
, i
);
1603 spin_unlock(&nm_i
->free_nid_list_lock
);
1605 kmem_cache_free(free_nid_slab
, i
);
1609 * alloc_nid() should be called prior to this function.
1611 void alloc_nid_failed(struct f2fs_sb_info
*sbi
, nid_t nid
)
1613 struct f2fs_nm_info
*nm_i
= NM_I(sbi
);
1615 bool need_free
= false;
1620 spin_lock(&nm_i
->free_nid_list_lock
);
1621 i
= __lookup_free_nid_list(nm_i
, nid
);
1622 f2fs_bug_on(sbi
, !i
|| i
->state
!= NID_ALLOC
);
1623 if (!available_free_memory(sbi
, FREE_NIDS
)) {
1624 __del_from_free_nid_list(nm_i
, i
);
1630 spin_unlock(&nm_i
->free_nid_list_lock
);
1633 kmem_cache_free(free_nid_slab
, i
);
1636 void recover_inline_xattr(struct inode
*inode
, struct page
*page
)
1638 void *src_addr
, *dst_addr
;
1641 struct f2fs_inode
*ri
;
1643 ipage
= get_node_page(F2FS_I_SB(inode
), inode
->i_ino
);
1644 f2fs_bug_on(F2FS_I_SB(inode
), IS_ERR(ipage
));
1646 ri
= F2FS_INODE(page
);
1647 if (!(ri
->i_inline
& F2FS_INLINE_XATTR
)) {
1648 clear_inode_flag(F2FS_I(inode
), FI_INLINE_XATTR
);
1652 dst_addr
= inline_xattr_addr(ipage
);
1653 src_addr
= inline_xattr_addr(page
);
1654 inline_size
= inline_xattr_size(inode
);
1656 f2fs_wait_on_page_writeback(ipage
, NODE
);
1657 memcpy(dst_addr
, src_addr
, inline_size
);
1659 update_inode(inode
, ipage
);
1660 f2fs_put_page(ipage
, 1);
1663 void recover_xattr_data(struct inode
*inode
, struct page
*page
, block_t blkaddr
)
1665 struct f2fs_sb_info
*sbi
= F2FS_I_SB(inode
);
1666 nid_t prev_xnid
= F2FS_I(inode
)->i_xattr_nid
;
1667 nid_t new_xnid
= nid_of_node(page
);
1668 struct node_info ni
;
1670 /* 1: invalidate the previous xattr nid */
1674 /* Deallocate node address */
1675 get_node_info(sbi
, prev_xnid
, &ni
);
1676 f2fs_bug_on(sbi
, ni
.blk_addr
== NULL_ADDR
);
1677 invalidate_blocks(sbi
, ni
.blk_addr
);
1678 dec_valid_node_count(sbi
, inode
);
1679 set_node_addr(sbi
, &ni
, NULL_ADDR
, false);
1682 /* 2: allocate new xattr nid */
1683 if (unlikely(!inc_valid_node_count(sbi
, inode
)))
1684 f2fs_bug_on(sbi
, 1);
1686 remove_free_nid(NM_I(sbi
), new_xnid
);
1687 get_node_info(sbi
, new_xnid
, &ni
);
1688 ni
.ino
= inode
->i_ino
;
1689 set_node_addr(sbi
, &ni
, NEW_ADDR
, false);
1690 F2FS_I(inode
)->i_xattr_nid
= new_xnid
;
1692 /* 3: update xattr blkaddr */
1693 refresh_sit_entry(sbi
, NEW_ADDR
, blkaddr
);
1694 set_node_addr(sbi
, &ni
, blkaddr
, false);
1696 update_inode_page(inode
);
1699 int recover_inode_page(struct f2fs_sb_info
*sbi
, struct page
*page
)
1701 struct f2fs_inode
*src
, *dst
;
1702 nid_t ino
= ino_of_node(page
);
1703 struct node_info old_ni
, new_ni
;
1706 get_node_info(sbi
, ino
, &old_ni
);
1708 if (unlikely(old_ni
.blk_addr
!= NULL_ADDR
))
1711 ipage
= grab_cache_page(NODE_MAPPING(sbi
), ino
);
1715 /* Should not use this inode from free nid list */
1716 remove_free_nid(NM_I(sbi
), ino
);
1718 SetPageUptodate(ipage
);
1719 fill_node_footer(ipage
, ino
, ino
, 0, true);
1721 src
= F2FS_INODE(page
);
1722 dst
= F2FS_INODE(ipage
);
1724 memcpy(dst
, src
, (unsigned long)&src
->i_ext
- (unsigned long)src
);
1726 dst
->i_blocks
= cpu_to_le64(1);
1727 dst
->i_links
= cpu_to_le32(1);
1728 dst
->i_xattr_nid
= 0;
1729 dst
->i_inline
= src
->i_inline
& F2FS_INLINE_XATTR
;
1734 if (unlikely(!inc_valid_node_count(sbi
, NULL
)))
1736 set_node_addr(sbi
, &new_ni
, NEW_ADDR
, false);
1737 inc_valid_inode_count(sbi
);
1738 set_page_dirty(ipage
);
1739 f2fs_put_page(ipage
, 1);
1743 int restore_node_summary(struct f2fs_sb_info
*sbi
,
1744 unsigned int segno
, struct f2fs_summary_block
*sum
)
1746 struct f2fs_node
*rn
;
1747 struct f2fs_summary
*sum_entry
;
1749 int bio_blocks
= MAX_BIO_BLOCKS(sbi
);
1750 int i
, idx
, last_offset
, nrpages
;
1752 /* scan the node segment */
1753 last_offset
= sbi
->blocks_per_seg
;
1754 addr
= START_BLOCK(sbi
, segno
);
1755 sum_entry
= &sum
->entries
[0];
1757 for (i
= 0; i
< last_offset
; i
+= nrpages
, addr
+= nrpages
) {
1758 nrpages
= min(last_offset
- i
, bio_blocks
);
1760 /* readahead node pages */
1761 ra_meta_pages(sbi
, addr
, nrpages
, META_POR
);
1763 for (idx
= addr
; idx
< addr
+ nrpages
; idx
++) {
1764 struct page
*page
= get_meta_page(sbi
, idx
);
1766 rn
= F2FS_NODE(page
);
1767 sum_entry
->nid
= rn
->footer
.nid
;
1768 sum_entry
->version
= 0;
1769 sum_entry
->ofs_in_node
= 0;
1771 f2fs_put_page(page
, 1);
1774 invalidate_mapping_pages(META_MAPPING(sbi
), addr
,
1780 static void remove_nats_in_journal(struct f2fs_sb_info
*sbi
)
1782 struct f2fs_nm_info
*nm_i
= NM_I(sbi
);
1783 struct curseg_info
*curseg
= CURSEG_I(sbi
, CURSEG_HOT_DATA
);
1784 struct f2fs_summary_block
*sum
= curseg
->sum_blk
;
1787 mutex_lock(&curseg
->curseg_mutex
);
1788 for (i
= 0; i
< nats_in_cursum(sum
); i
++) {
1789 struct nat_entry
*ne
;
1790 struct f2fs_nat_entry raw_ne
;
1791 nid_t nid
= le32_to_cpu(nid_in_journal(sum
, i
));
1793 raw_ne
= nat_in_journal(sum
, i
);
1795 down_write(&nm_i
->nat_tree_lock
);
1796 ne
= __lookup_nat_cache(nm_i
, nid
);
1798 ne
= grab_nat_entry(nm_i
, nid
);
1799 node_info_from_raw_nat(&ne
->ni
, &raw_ne
);
1801 __set_nat_cache_dirty(nm_i
, ne
);
1802 up_write(&nm_i
->nat_tree_lock
);
1804 update_nats_in_cursum(sum
, -i
);
1805 mutex_unlock(&curseg
->curseg_mutex
);
1808 static void __adjust_nat_entry_set(struct nat_entry_set
*nes
,
1809 struct list_head
*head
, int max
)
1811 struct nat_entry_set
*cur
;
1813 if (nes
->entry_cnt
>= max
)
1816 list_for_each_entry(cur
, head
, set_list
) {
1817 if (cur
->entry_cnt
>= nes
->entry_cnt
) {
1818 list_add(&nes
->set_list
, cur
->set_list
.prev
);
1823 list_add_tail(&nes
->set_list
, head
);
1826 static void __flush_nat_entry_set(struct f2fs_sb_info
*sbi
,
1827 struct nat_entry_set
*set
)
1829 struct curseg_info
*curseg
= CURSEG_I(sbi
, CURSEG_HOT_DATA
);
1830 struct f2fs_summary_block
*sum
= curseg
->sum_blk
;
1831 nid_t start_nid
= set
->set
* NAT_ENTRY_PER_BLOCK
;
1832 bool to_journal
= true;
1833 struct f2fs_nat_block
*nat_blk
;
1834 struct nat_entry
*ne
, *cur
;
1835 struct page
*page
= NULL
;
1836 struct f2fs_nm_info
*nm_i
= NM_I(sbi
);
1839 * there are two steps to flush nat entries:
1840 * #1, flush nat entries to journal in current hot data summary block.
1841 * #2, flush nat entries to nat page.
1843 if (!__has_cursum_space(sum
, set
->entry_cnt
, NAT_JOURNAL
))
1847 mutex_lock(&curseg
->curseg_mutex
);
1849 page
= get_next_nat_page(sbi
, start_nid
);
1850 nat_blk
= page_address(page
);
1851 f2fs_bug_on(sbi
, !nat_blk
);
1854 /* flush dirty nats in nat entry set */
1855 list_for_each_entry_safe(ne
, cur
, &set
->entry_list
, list
) {
1856 struct f2fs_nat_entry
*raw_ne
;
1857 nid_t nid
= nat_get_nid(ne
);
1860 if (nat_get_blkaddr(ne
) == NEW_ADDR
)
1864 offset
= lookup_journal_in_cursum(sum
,
1865 NAT_JOURNAL
, nid
, 1);
1866 f2fs_bug_on(sbi
, offset
< 0);
1867 raw_ne
= &nat_in_journal(sum
, offset
);
1868 nid_in_journal(sum
, offset
) = cpu_to_le32(nid
);
1870 raw_ne
= &nat_blk
->entries
[nid
- start_nid
];
1872 raw_nat_from_node_info(raw_ne
, &ne
->ni
);
1874 down_write(&NM_I(sbi
)->nat_tree_lock
);
1876 __clear_nat_cache_dirty(NM_I(sbi
), ne
);
1877 up_write(&NM_I(sbi
)->nat_tree_lock
);
1879 if (nat_get_blkaddr(ne
) == NULL_ADDR
)
1880 add_free_nid(sbi
, nid
, false);
1884 mutex_unlock(&curseg
->curseg_mutex
);
1886 f2fs_put_page(page
, 1);
1888 f2fs_bug_on(sbi
, set
->entry_cnt
);
1890 down_write(&nm_i
->nat_tree_lock
);
1891 radix_tree_delete(&NM_I(sbi
)->nat_set_root
, set
->set
);
1892 up_write(&nm_i
->nat_tree_lock
);
1893 kmem_cache_free(nat_entry_set_slab
, set
);
1897 * This function is called during the checkpointing process.
1899 void flush_nat_entries(struct f2fs_sb_info
*sbi
)
1901 struct f2fs_nm_info
*nm_i
= NM_I(sbi
);
1902 struct curseg_info
*curseg
= CURSEG_I(sbi
, CURSEG_HOT_DATA
);
1903 struct f2fs_summary_block
*sum
= curseg
->sum_blk
;
1904 struct nat_entry_set
*setvec
[SETVEC_SIZE
];
1905 struct nat_entry_set
*set
, *tmp
;
1910 if (!nm_i
->dirty_nat_cnt
)
1913 * if there are no enough space in journal to store dirty nat
1914 * entries, remove all entries from journal and merge them
1915 * into nat entry set.
1917 if (!__has_cursum_space(sum
, nm_i
->dirty_nat_cnt
, NAT_JOURNAL
))
1918 remove_nats_in_journal(sbi
);
1920 down_write(&nm_i
->nat_tree_lock
);
1921 while ((found
= __gang_lookup_nat_set(nm_i
,
1922 set_idx
, SETVEC_SIZE
, setvec
))) {
1924 set_idx
= setvec
[found
- 1]->set
+ 1;
1925 for (idx
= 0; idx
< found
; idx
++)
1926 __adjust_nat_entry_set(setvec
[idx
], &sets
,
1927 MAX_NAT_JENTRIES(sum
));
1929 up_write(&nm_i
->nat_tree_lock
);
1931 /* flush dirty nats in nat entry set */
1932 list_for_each_entry_safe(set
, tmp
, &sets
, set_list
)
1933 __flush_nat_entry_set(sbi
, set
);
1935 f2fs_bug_on(sbi
, nm_i
->dirty_nat_cnt
);
1938 static int init_node_manager(struct f2fs_sb_info
*sbi
)
1940 struct f2fs_super_block
*sb_raw
= F2FS_RAW_SUPER(sbi
);
1941 struct f2fs_nm_info
*nm_i
= NM_I(sbi
);
1942 unsigned char *version_bitmap
;
1943 unsigned int nat_segs
, nat_blocks
;
1945 nm_i
->nat_blkaddr
= le32_to_cpu(sb_raw
->nat_blkaddr
);
1947 /* segment_count_nat includes pair segment so divide to 2. */
1948 nat_segs
= le32_to_cpu(sb_raw
->segment_count_nat
) >> 1;
1949 nat_blocks
= nat_segs
<< le32_to_cpu(sb_raw
->log_blocks_per_seg
);
1951 nm_i
->max_nid
= NAT_ENTRY_PER_BLOCK
* nat_blocks
;
1953 /* not used nids: 0, node, meta, (and root counted as valid node) */
1954 nm_i
->available_nids
= nm_i
->max_nid
- F2FS_RESERVED_NODE_NUM
;
1957 nm_i
->ram_thresh
= DEF_RAM_THRESHOLD
;
1959 INIT_RADIX_TREE(&nm_i
->free_nid_root
, GFP_ATOMIC
);
1960 INIT_LIST_HEAD(&nm_i
->free_nid_list
);
1961 INIT_RADIX_TREE(&nm_i
->nat_root
, GFP_NOIO
);
1962 INIT_RADIX_TREE(&nm_i
->nat_set_root
, GFP_NOIO
);
1963 INIT_LIST_HEAD(&nm_i
->nat_entries
);
1965 mutex_init(&nm_i
->build_lock
);
1966 spin_lock_init(&nm_i
->free_nid_list_lock
);
1967 init_rwsem(&nm_i
->nat_tree_lock
);
1969 nm_i
->next_scan_nid
= le32_to_cpu(sbi
->ckpt
->next_free_nid
);
1970 nm_i
->bitmap_size
= __bitmap_size(sbi
, NAT_BITMAP
);
1971 version_bitmap
= __bitmap_ptr(sbi
, NAT_BITMAP
);
1972 if (!version_bitmap
)
1975 nm_i
->nat_bitmap
= kmemdup(version_bitmap
, nm_i
->bitmap_size
,
1977 if (!nm_i
->nat_bitmap
)
1982 int build_node_manager(struct f2fs_sb_info
*sbi
)
1986 sbi
->nm_info
= kzalloc(sizeof(struct f2fs_nm_info
), GFP_KERNEL
);
1990 err
= init_node_manager(sbi
);
1994 build_free_nids(sbi
);
1998 void destroy_node_manager(struct f2fs_sb_info
*sbi
)
2000 struct f2fs_nm_info
*nm_i
= NM_I(sbi
);
2001 struct free_nid
*i
, *next_i
;
2002 struct nat_entry
*natvec
[NATVEC_SIZE
];
2003 struct nat_entry_set
*setvec
[SETVEC_SIZE
];
2010 /* destroy free nid list */
2011 spin_lock(&nm_i
->free_nid_list_lock
);
2012 list_for_each_entry_safe(i
, next_i
, &nm_i
->free_nid_list
, list
) {
2013 f2fs_bug_on(sbi
, i
->state
== NID_ALLOC
);
2014 __del_from_free_nid_list(nm_i
, i
);
2016 spin_unlock(&nm_i
->free_nid_list_lock
);
2017 kmem_cache_free(free_nid_slab
, i
);
2018 spin_lock(&nm_i
->free_nid_list_lock
);
2020 f2fs_bug_on(sbi
, nm_i
->fcnt
);
2021 spin_unlock(&nm_i
->free_nid_list_lock
);
2023 /* destroy nat cache */
2024 down_write(&nm_i
->nat_tree_lock
);
2025 while ((found
= __gang_lookup_nat_cache(nm_i
,
2026 nid
, NATVEC_SIZE
, natvec
))) {
2029 nid
= nat_get_nid(natvec
[found
- 1]) + 1;
2030 for (idx
= 0; idx
< found
; idx
++)
2031 __del_from_nat_cache(nm_i
, natvec
[idx
]);
2033 f2fs_bug_on(sbi
, nm_i
->nat_cnt
);
2035 /* destroy nat set cache */
2037 while ((found
= __gang_lookup_nat_set(nm_i
,
2038 nid
, SETVEC_SIZE
, setvec
))) {
2041 nid
= setvec
[found
- 1]->set
+ 1;
2042 for (idx
= 0; idx
< found
; idx
++) {
2043 /* entry_cnt is not zero, when cp_error was occurred */
2044 f2fs_bug_on(sbi
, !list_empty(&setvec
[idx
]->entry_list
));
2045 radix_tree_delete(&nm_i
->nat_set_root
, setvec
[idx
]->set
);
2046 kmem_cache_free(nat_entry_set_slab
, setvec
[idx
]);
2049 up_write(&nm_i
->nat_tree_lock
);
2051 kfree(nm_i
->nat_bitmap
);
2052 sbi
->nm_info
= NULL
;
2056 int __init
create_node_manager_caches(void)
2058 nat_entry_slab
= f2fs_kmem_cache_create("nat_entry",
2059 sizeof(struct nat_entry
));
2060 if (!nat_entry_slab
)
2063 free_nid_slab
= f2fs_kmem_cache_create("free_nid",
2064 sizeof(struct free_nid
));
2066 goto destroy_nat_entry
;
2068 nat_entry_set_slab
= f2fs_kmem_cache_create("nat_entry_set",
2069 sizeof(struct nat_entry_set
));
2070 if (!nat_entry_set_slab
)
2071 goto destroy_free_nid
;
2075 kmem_cache_destroy(free_nid_slab
);
2077 kmem_cache_destroy(nat_entry_slab
);
2082 void destroy_node_manager_caches(void)
2084 kmem_cache_destroy(nat_entry_set_slab
);
2085 kmem_cache_destroy(free_nid_slab
);
2086 kmem_cache_destroy(nat_entry_slab
);