4 * Copyright (c) 2012 Samsung Electronics Co., Ltd.
5 * http://www.samsung.com/
7 * This program is free software; you can redistribute it and/or modify
8 * it under the terms of the GNU General Public License version 2 as
9 * published by the Free Software Foundation.
12 #include <linux/f2fs_fs.h>
13 #include <linux/mpage.h>
14 #include <linux/backing-dev.h>
15 #include <linux/blkdev.h>
16 #include <linux/pagevec.h>
17 #include <linux/swap.h>
24 #include <trace/events/f2fs.h>
26 #define on_build_free_nids(nmi) mutex_is_locked(&(nm_i)->build_lock)
28 static struct kmem_cache *nat_entry_slab;
29 static struct kmem_cache *free_nid_slab;
30 static struct kmem_cache *nat_entry_set_slab;
32 bool available_free_memory(struct f2fs_sb_info *sbi, int type)
34 struct f2fs_nm_info *nm_i = NM_I(sbi);
36 unsigned long avail_ram;
37 unsigned long mem_size = 0;
42 /* only uses low memory */
43 avail_ram = val.totalram - val.totalhigh;
46 * give 25%, 25%, 50%, 50%, 50% memory for each components respectively
48 if (type == FREE_NIDS) {
49 mem_size = (nm_i->nid_cnt[FREE_NID] *
50 sizeof(struct free_nid)) >> PAGE_SHIFT;
51 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 2);
52 } else if (type == NAT_ENTRIES) {
53 mem_size = (nm_i->nat_cnt * sizeof(struct nat_entry)) >>
55 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 2);
56 if (excess_cached_nats(sbi))
58 } else if (type == DIRTY_DENTS) {
59 if (sbi->sb->s_bdi->wb.dirty_exceeded)
61 mem_size = get_pages(sbi, F2FS_DIRTY_DENTS);
62 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 1);
63 } else if (type == INO_ENTRIES) {
66 for (i = 0; i < MAX_INO_ENTRY; i++)
67 mem_size += sbi->im[i].ino_num *
68 sizeof(struct ino_entry);
69 mem_size >>= PAGE_SHIFT;
70 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 1);
71 } else if (type == EXTENT_CACHE) {
72 mem_size = (atomic_read(&sbi->total_ext_tree) *
73 sizeof(struct extent_tree) +
74 atomic_read(&sbi->total_ext_node) *
75 sizeof(struct extent_node)) >> PAGE_SHIFT;
76 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 1);
77 } else if (type == INMEM_PAGES) {
78 /* it allows 20% / total_ram for inmemory pages */
79 mem_size = get_pages(sbi, F2FS_INMEM_PAGES);
80 res = mem_size < (val.totalram / 5);
82 if (!sbi->sb->s_bdi->wb.dirty_exceeded)
88 static void clear_node_page_dirty(struct page *page)
90 struct address_space *mapping = page->mapping;
91 unsigned int long flags;
93 if (PageDirty(page)) {
94 spin_lock_irqsave(&mapping->tree_lock, flags);
95 radix_tree_tag_clear(&mapping->page_tree,
98 spin_unlock_irqrestore(&mapping->tree_lock, flags);
100 clear_page_dirty_for_io(page);
101 dec_page_count(F2FS_M_SB(mapping), F2FS_DIRTY_NODES);
103 ClearPageUptodate(page);
106 static struct page *get_current_nat_page(struct f2fs_sb_info *sbi, nid_t nid)
108 pgoff_t index = current_nat_addr(sbi, nid);
109 return get_meta_page(sbi, index);
112 static struct page *get_next_nat_page(struct f2fs_sb_info *sbi, nid_t nid)
114 struct page *src_page;
115 struct page *dst_page;
120 struct f2fs_nm_info *nm_i = NM_I(sbi);
122 src_off = current_nat_addr(sbi, nid);
123 dst_off = next_nat_addr(sbi, src_off);
125 /* get current nat block page with lock */
126 src_page = get_meta_page(sbi, src_off);
127 dst_page = grab_meta_page(sbi, dst_off);
128 f2fs_bug_on(sbi, PageDirty(src_page));
130 src_addr = page_address(src_page);
131 dst_addr = page_address(dst_page);
132 memcpy(dst_addr, src_addr, PAGE_SIZE);
133 set_page_dirty(dst_page);
134 f2fs_put_page(src_page, 1);
136 set_to_next_nat(nm_i, nid);
141 static struct nat_entry *__lookup_nat_cache(struct f2fs_nm_info *nm_i, nid_t n)
143 return radix_tree_lookup(&nm_i->nat_root, n);
146 static unsigned int __gang_lookup_nat_cache(struct f2fs_nm_info *nm_i,
147 nid_t start, unsigned int nr, struct nat_entry **ep)
149 return radix_tree_gang_lookup(&nm_i->nat_root, (void **)ep, start, nr);
152 static void __del_from_nat_cache(struct f2fs_nm_info *nm_i, struct nat_entry *e)
155 radix_tree_delete(&nm_i->nat_root, nat_get_nid(e));
157 kmem_cache_free(nat_entry_slab, e);
160 static void __set_nat_cache_dirty(struct f2fs_nm_info *nm_i,
161 struct nat_entry *ne)
163 nid_t set = NAT_BLOCK_OFFSET(ne->ni.nid);
164 struct nat_entry_set *head;
166 head = radix_tree_lookup(&nm_i->nat_set_root, set);
168 head = f2fs_kmem_cache_alloc(nat_entry_set_slab, GFP_NOFS);
170 INIT_LIST_HEAD(&head->entry_list);
171 INIT_LIST_HEAD(&head->set_list);
174 f2fs_radix_tree_insert(&nm_i->nat_set_root, set, head);
177 if (get_nat_flag(ne, IS_DIRTY))
180 nm_i->dirty_nat_cnt++;
182 set_nat_flag(ne, IS_DIRTY, true);
184 if (nat_get_blkaddr(ne) == NEW_ADDR)
185 list_del_init(&ne->list);
187 list_move_tail(&ne->list, &head->entry_list);
190 static void __clear_nat_cache_dirty(struct f2fs_nm_info *nm_i,
191 struct nat_entry_set *set, struct nat_entry *ne)
193 list_move_tail(&ne->list, &nm_i->nat_entries);
194 set_nat_flag(ne, IS_DIRTY, false);
196 nm_i->dirty_nat_cnt--;
199 static unsigned int __gang_lookup_nat_set(struct f2fs_nm_info *nm_i,
200 nid_t start, unsigned int nr, struct nat_entry_set **ep)
202 return radix_tree_gang_lookup(&nm_i->nat_set_root, (void **)ep,
206 int need_dentry_mark(struct f2fs_sb_info *sbi, nid_t nid)
208 struct f2fs_nm_info *nm_i = NM_I(sbi);
212 down_read(&nm_i->nat_tree_lock);
213 e = __lookup_nat_cache(nm_i, nid);
215 if (!get_nat_flag(e, IS_CHECKPOINTED) &&
216 !get_nat_flag(e, HAS_FSYNCED_INODE))
219 up_read(&nm_i->nat_tree_lock);
223 bool is_checkpointed_node(struct f2fs_sb_info *sbi, nid_t nid)
225 struct f2fs_nm_info *nm_i = NM_I(sbi);
229 down_read(&nm_i->nat_tree_lock);
230 e = __lookup_nat_cache(nm_i, nid);
231 if (e && !get_nat_flag(e, IS_CHECKPOINTED))
233 up_read(&nm_i->nat_tree_lock);
237 bool need_inode_block_update(struct f2fs_sb_info *sbi, nid_t ino)
239 struct f2fs_nm_info *nm_i = NM_I(sbi);
241 bool need_update = true;
243 down_read(&nm_i->nat_tree_lock);
244 e = __lookup_nat_cache(nm_i, ino);
245 if (e && get_nat_flag(e, HAS_LAST_FSYNC) &&
246 (get_nat_flag(e, IS_CHECKPOINTED) ||
247 get_nat_flag(e, HAS_FSYNCED_INODE)))
249 up_read(&nm_i->nat_tree_lock);
253 static struct nat_entry *grab_nat_entry(struct f2fs_nm_info *nm_i, nid_t nid,
256 struct nat_entry *new;
259 new = f2fs_kmem_cache_alloc(nat_entry_slab, GFP_NOFS);
260 f2fs_radix_tree_insert(&nm_i->nat_root, nid, new);
262 new = kmem_cache_alloc(nat_entry_slab, GFP_NOFS);
265 if (radix_tree_insert(&nm_i->nat_root, nid, new)) {
266 kmem_cache_free(nat_entry_slab, new);
271 memset(new, 0, sizeof(struct nat_entry));
272 nat_set_nid(new, nid);
274 list_add_tail(&new->list, &nm_i->nat_entries);
279 static void cache_nat_entry(struct f2fs_sb_info *sbi, nid_t nid,
280 struct f2fs_nat_entry *ne)
282 struct f2fs_nm_info *nm_i = NM_I(sbi);
285 e = __lookup_nat_cache(nm_i, nid);
287 e = grab_nat_entry(nm_i, nid, false);
289 node_info_from_raw_nat(&e->ni, ne);
291 f2fs_bug_on(sbi, nat_get_ino(e) != le32_to_cpu(ne->ino) ||
292 nat_get_blkaddr(e) !=
293 le32_to_cpu(ne->block_addr) ||
294 nat_get_version(e) != ne->version);
298 static void set_node_addr(struct f2fs_sb_info *sbi, struct node_info *ni,
299 block_t new_blkaddr, bool fsync_done)
301 struct f2fs_nm_info *nm_i = NM_I(sbi);
304 down_write(&nm_i->nat_tree_lock);
305 e = __lookup_nat_cache(nm_i, ni->nid);
307 e = grab_nat_entry(nm_i, ni->nid, true);
308 copy_node_info(&e->ni, ni);
309 f2fs_bug_on(sbi, ni->blk_addr == NEW_ADDR);
310 } else if (new_blkaddr == NEW_ADDR) {
312 * when nid is reallocated,
313 * previous nat entry can be remained in nat cache.
314 * So, reinitialize it with new information.
316 copy_node_info(&e->ni, ni);
317 f2fs_bug_on(sbi, ni->blk_addr != NULL_ADDR);
321 f2fs_bug_on(sbi, nat_get_blkaddr(e) != ni->blk_addr);
322 f2fs_bug_on(sbi, nat_get_blkaddr(e) == NULL_ADDR &&
323 new_blkaddr == NULL_ADDR);
324 f2fs_bug_on(sbi, nat_get_blkaddr(e) == NEW_ADDR &&
325 new_blkaddr == NEW_ADDR);
326 f2fs_bug_on(sbi, nat_get_blkaddr(e) != NEW_ADDR &&
327 nat_get_blkaddr(e) != NULL_ADDR &&
328 new_blkaddr == NEW_ADDR);
330 /* increment version no as node is removed */
331 if (nat_get_blkaddr(e) != NEW_ADDR && new_blkaddr == NULL_ADDR) {
332 unsigned char version = nat_get_version(e);
333 nat_set_version(e, inc_node_version(version));
337 nat_set_blkaddr(e, new_blkaddr);
338 if (new_blkaddr == NEW_ADDR || new_blkaddr == NULL_ADDR)
339 set_nat_flag(e, IS_CHECKPOINTED, false);
340 __set_nat_cache_dirty(nm_i, e);
342 /* update fsync_mark if its inode nat entry is still alive */
343 if (ni->nid != ni->ino)
344 e = __lookup_nat_cache(nm_i, ni->ino);
346 if (fsync_done && ni->nid == ni->ino)
347 set_nat_flag(e, HAS_FSYNCED_INODE, true);
348 set_nat_flag(e, HAS_LAST_FSYNC, fsync_done);
350 up_write(&nm_i->nat_tree_lock);
353 int try_to_free_nats(struct f2fs_sb_info *sbi, int nr_shrink)
355 struct f2fs_nm_info *nm_i = NM_I(sbi);
358 if (!down_write_trylock(&nm_i->nat_tree_lock))
361 while (nr_shrink && !list_empty(&nm_i->nat_entries)) {
362 struct nat_entry *ne;
363 ne = list_first_entry(&nm_i->nat_entries,
364 struct nat_entry, list);
365 __del_from_nat_cache(nm_i, ne);
368 up_write(&nm_i->nat_tree_lock);
369 return nr - nr_shrink;
373 * This function always returns success
375 void get_node_info(struct f2fs_sb_info *sbi, nid_t nid, struct node_info *ni)
377 struct f2fs_nm_info *nm_i = NM_I(sbi);
378 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
379 struct f2fs_journal *journal = curseg->journal;
380 nid_t start_nid = START_NID(nid);
381 struct f2fs_nat_block *nat_blk;
382 struct page *page = NULL;
383 struct f2fs_nat_entry ne;
390 /* Check nat cache */
391 down_read(&nm_i->nat_tree_lock);
392 e = __lookup_nat_cache(nm_i, nid);
394 ni->ino = nat_get_ino(e);
395 ni->blk_addr = nat_get_blkaddr(e);
396 ni->version = nat_get_version(e);
397 up_read(&nm_i->nat_tree_lock);
401 memset(&ne, 0, sizeof(struct f2fs_nat_entry));
403 /* Check current segment summary */
404 down_read(&curseg->journal_rwsem);
405 i = lookup_journal_in_cursum(journal, NAT_JOURNAL, nid, 0);
407 ne = nat_in_journal(journal, i);
408 node_info_from_raw_nat(ni, &ne);
410 up_read(&curseg->journal_rwsem);
412 up_read(&nm_i->nat_tree_lock);
416 /* Fill node_info from nat page */
417 index = current_nat_addr(sbi, nid);
418 up_read(&nm_i->nat_tree_lock);
420 page = get_meta_page(sbi, index);
421 nat_blk = (struct f2fs_nat_block *)page_address(page);
422 ne = nat_blk->entries[nid - start_nid];
423 node_info_from_raw_nat(ni, &ne);
424 f2fs_put_page(page, 1);
426 /* cache nat entry */
427 down_write(&nm_i->nat_tree_lock);
428 cache_nat_entry(sbi, nid, &ne);
429 up_write(&nm_i->nat_tree_lock);
433 * readahead MAX_RA_NODE number of node pages.
435 static void ra_node_pages(struct page *parent, int start, int n)
437 struct f2fs_sb_info *sbi = F2FS_P_SB(parent);
438 struct blk_plug plug;
442 blk_start_plug(&plug);
444 /* Then, try readahead for siblings of the desired node */
446 end = min(end, NIDS_PER_BLOCK);
447 for (i = start; i < end; i++) {
448 nid = get_nid(parent, i, false);
449 ra_node_page(sbi, nid);
452 blk_finish_plug(&plug);
455 pgoff_t get_next_page_offset(struct dnode_of_data *dn, pgoff_t pgofs)
457 const long direct_index = ADDRS_PER_INODE(dn->inode);
458 const long direct_blks = ADDRS_PER_BLOCK;
459 const long indirect_blks = ADDRS_PER_BLOCK * NIDS_PER_BLOCK;
460 unsigned int skipped_unit = ADDRS_PER_BLOCK;
461 int cur_level = dn->cur_level;
462 int max_level = dn->max_level;
468 while (max_level-- > cur_level)
469 skipped_unit *= NIDS_PER_BLOCK;
471 switch (dn->max_level) {
473 base += 2 * indirect_blks;
475 base += 2 * direct_blks;
477 base += direct_index;
480 f2fs_bug_on(F2FS_I_SB(dn->inode), 1);
483 return ((pgofs - base) / skipped_unit + 1) * skipped_unit + base;
487 * The maximum depth is four.
488 * Offset[0] will have raw inode offset.
490 static int get_node_path(struct inode *inode, long block,
491 int offset[4], unsigned int noffset[4])
493 const long direct_index = ADDRS_PER_INODE(inode);
494 const long direct_blks = ADDRS_PER_BLOCK;
495 const long dptrs_per_blk = NIDS_PER_BLOCK;
496 const long indirect_blks = ADDRS_PER_BLOCK * NIDS_PER_BLOCK;
497 const long dindirect_blks = indirect_blks * NIDS_PER_BLOCK;
503 if (block < direct_index) {
507 block -= direct_index;
508 if (block < direct_blks) {
509 offset[n++] = NODE_DIR1_BLOCK;
515 block -= direct_blks;
516 if (block < direct_blks) {
517 offset[n++] = NODE_DIR2_BLOCK;
523 block -= direct_blks;
524 if (block < indirect_blks) {
525 offset[n++] = NODE_IND1_BLOCK;
527 offset[n++] = block / direct_blks;
528 noffset[n] = 4 + offset[n - 1];
529 offset[n] = block % direct_blks;
533 block -= indirect_blks;
534 if (block < indirect_blks) {
535 offset[n++] = NODE_IND2_BLOCK;
536 noffset[n] = 4 + dptrs_per_blk;
537 offset[n++] = block / direct_blks;
538 noffset[n] = 5 + dptrs_per_blk + offset[n - 1];
539 offset[n] = block % direct_blks;
543 block -= indirect_blks;
544 if (block < dindirect_blks) {
545 offset[n++] = NODE_DIND_BLOCK;
546 noffset[n] = 5 + (dptrs_per_blk * 2);
547 offset[n++] = block / indirect_blks;
548 noffset[n] = 6 + (dptrs_per_blk * 2) +
549 offset[n - 1] * (dptrs_per_blk + 1);
550 offset[n++] = (block / direct_blks) % dptrs_per_blk;
551 noffset[n] = 7 + (dptrs_per_blk * 2) +
552 offset[n - 2] * (dptrs_per_blk + 1) +
554 offset[n] = block % direct_blks;
565 * Caller should call f2fs_put_dnode(dn).
566 * Also, it should grab and release a rwsem by calling f2fs_lock_op() and
567 * f2fs_unlock_op() only if ro is not set RDONLY_NODE.
568 * In the case of RDONLY_NODE, we don't need to care about mutex.
570 int get_dnode_of_data(struct dnode_of_data *dn, pgoff_t index, int mode)
572 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
573 struct page *npage[4];
574 struct page *parent = NULL;
576 unsigned int noffset[4];
581 level = get_node_path(dn->inode, index, offset, noffset);
585 nids[0] = dn->inode->i_ino;
586 npage[0] = dn->inode_page;
589 npage[0] = get_node_page(sbi, nids[0]);
590 if (IS_ERR(npage[0]))
591 return PTR_ERR(npage[0]);
594 /* if inline_data is set, should not report any block indices */
595 if (f2fs_has_inline_data(dn->inode) && index) {
597 f2fs_put_page(npage[0], 1);
603 nids[1] = get_nid(parent, offset[0], true);
604 dn->inode_page = npage[0];
605 dn->inode_page_locked = true;
607 /* get indirect or direct nodes */
608 for (i = 1; i <= level; i++) {
611 if (!nids[i] && mode == ALLOC_NODE) {
613 if (!alloc_nid(sbi, &(nids[i]))) {
619 npage[i] = new_node_page(dn, noffset[i]);
620 if (IS_ERR(npage[i])) {
621 alloc_nid_failed(sbi, nids[i]);
622 err = PTR_ERR(npage[i]);
626 set_nid(parent, offset[i - 1], nids[i], i == 1);
627 alloc_nid_done(sbi, nids[i]);
629 } else if (mode == LOOKUP_NODE_RA && i == level && level > 1) {
630 npage[i] = get_node_page_ra(parent, offset[i - 1]);
631 if (IS_ERR(npage[i])) {
632 err = PTR_ERR(npage[i]);
638 dn->inode_page_locked = false;
641 f2fs_put_page(parent, 1);
645 npage[i] = get_node_page(sbi, nids[i]);
646 if (IS_ERR(npage[i])) {
647 err = PTR_ERR(npage[i]);
648 f2fs_put_page(npage[0], 0);
654 nids[i + 1] = get_nid(parent, offset[i], false);
657 dn->nid = nids[level];
658 dn->ofs_in_node = offset[level];
659 dn->node_page = npage[level];
660 dn->data_blkaddr = datablock_addr(dn->inode,
661 dn->node_page, dn->ofs_in_node);
665 f2fs_put_page(parent, 1);
667 f2fs_put_page(npage[0], 0);
669 dn->inode_page = NULL;
670 dn->node_page = NULL;
671 if (err == -ENOENT) {
673 dn->max_level = level;
674 dn->ofs_in_node = offset[level];
679 static void truncate_node(struct dnode_of_data *dn)
681 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
684 get_node_info(sbi, dn->nid, &ni);
685 f2fs_bug_on(sbi, ni.blk_addr == NULL_ADDR);
687 /* Deallocate node address */
688 invalidate_blocks(sbi, ni.blk_addr);
689 dec_valid_node_count(sbi, dn->inode, dn->nid == dn->inode->i_ino);
690 set_node_addr(sbi, &ni, NULL_ADDR, false);
692 if (dn->nid == dn->inode->i_ino) {
693 remove_orphan_inode(sbi, dn->nid);
694 dec_valid_inode_count(sbi);
695 f2fs_inode_synced(dn->inode);
698 clear_node_page_dirty(dn->node_page);
699 set_sbi_flag(sbi, SBI_IS_DIRTY);
701 f2fs_put_page(dn->node_page, 1);
703 invalidate_mapping_pages(NODE_MAPPING(sbi),
704 dn->node_page->index, dn->node_page->index);
706 dn->node_page = NULL;
707 trace_f2fs_truncate_node(dn->inode, dn->nid, ni.blk_addr);
710 static int truncate_dnode(struct dnode_of_data *dn)
717 /* get direct node */
718 page = get_node_page(F2FS_I_SB(dn->inode), dn->nid);
719 if (IS_ERR(page) && PTR_ERR(page) == -ENOENT)
721 else if (IS_ERR(page))
722 return PTR_ERR(page);
724 /* Make dnode_of_data for parameter */
725 dn->node_page = page;
727 truncate_data_blocks(dn);
732 static int truncate_nodes(struct dnode_of_data *dn, unsigned int nofs,
735 struct dnode_of_data rdn = *dn;
737 struct f2fs_node *rn;
739 unsigned int child_nofs;
744 return NIDS_PER_BLOCK + 1;
746 trace_f2fs_truncate_nodes_enter(dn->inode, dn->nid, dn->data_blkaddr);
748 page = get_node_page(F2FS_I_SB(dn->inode), dn->nid);
750 trace_f2fs_truncate_nodes_exit(dn->inode, PTR_ERR(page));
751 return PTR_ERR(page);
754 ra_node_pages(page, ofs, NIDS_PER_BLOCK);
756 rn = F2FS_NODE(page);
758 for (i = ofs; i < NIDS_PER_BLOCK; i++, freed++) {
759 child_nid = le32_to_cpu(rn->in.nid[i]);
763 ret = truncate_dnode(&rdn);
766 if (set_nid(page, i, 0, false))
767 dn->node_changed = true;
770 child_nofs = nofs + ofs * (NIDS_PER_BLOCK + 1) + 1;
771 for (i = ofs; i < NIDS_PER_BLOCK; i++) {
772 child_nid = le32_to_cpu(rn->in.nid[i]);
773 if (child_nid == 0) {
774 child_nofs += NIDS_PER_BLOCK + 1;
778 ret = truncate_nodes(&rdn, child_nofs, 0, depth - 1);
779 if (ret == (NIDS_PER_BLOCK + 1)) {
780 if (set_nid(page, i, 0, false))
781 dn->node_changed = true;
783 } else if (ret < 0 && ret != -ENOENT) {
791 /* remove current indirect node */
792 dn->node_page = page;
796 f2fs_put_page(page, 1);
798 trace_f2fs_truncate_nodes_exit(dn->inode, freed);
802 f2fs_put_page(page, 1);
803 trace_f2fs_truncate_nodes_exit(dn->inode, ret);
807 static int truncate_partial_nodes(struct dnode_of_data *dn,
808 struct f2fs_inode *ri, int *offset, int depth)
810 struct page *pages[2];
817 nid[0] = le32_to_cpu(ri->i_nid[offset[0] - NODE_DIR1_BLOCK]);
821 /* get indirect nodes in the path */
822 for (i = 0; i < idx + 1; i++) {
823 /* reference count'll be increased */
824 pages[i] = get_node_page(F2FS_I_SB(dn->inode), nid[i]);
825 if (IS_ERR(pages[i])) {
826 err = PTR_ERR(pages[i]);
830 nid[i + 1] = get_nid(pages[i], offset[i + 1], false);
833 ra_node_pages(pages[idx], offset[idx + 1], NIDS_PER_BLOCK);
835 /* free direct nodes linked to a partial indirect node */
836 for (i = offset[idx + 1]; i < NIDS_PER_BLOCK; i++) {
837 child_nid = get_nid(pages[idx], i, false);
841 err = truncate_dnode(dn);
844 if (set_nid(pages[idx], i, 0, false))
845 dn->node_changed = true;
848 if (offset[idx + 1] == 0) {
849 dn->node_page = pages[idx];
853 f2fs_put_page(pages[idx], 1);
859 for (i = idx; i >= 0; i--)
860 f2fs_put_page(pages[i], 1);
862 trace_f2fs_truncate_partial_nodes(dn->inode, nid, depth, err);
868 * All the block addresses of data and nodes should be nullified.
870 int truncate_inode_blocks(struct inode *inode, pgoff_t from)
872 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
873 int err = 0, cont = 1;
874 int level, offset[4], noffset[4];
875 unsigned int nofs = 0;
876 struct f2fs_inode *ri;
877 struct dnode_of_data dn;
880 trace_f2fs_truncate_inode_blocks_enter(inode, from);
882 level = get_node_path(inode, from, offset, noffset);
886 page = get_node_page(sbi, inode->i_ino);
888 trace_f2fs_truncate_inode_blocks_exit(inode, PTR_ERR(page));
889 return PTR_ERR(page);
892 set_new_dnode(&dn, inode, page, NULL, 0);
895 ri = F2FS_INODE(page);
903 if (!offset[level - 1])
905 err = truncate_partial_nodes(&dn, ri, offset, level);
906 if (err < 0 && err != -ENOENT)
908 nofs += 1 + NIDS_PER_BLOCK;
911 nofs = 5 + 2 * NIDS_PER_BLOCK;
912 if (!offset[level - 1])
914 err = truncate_partial_nodes(&dn, ri, offset, level);
915 if (err < 0 && err != -ENOENT)
924 dn.nid = le32_to_cpu(ri->i_nid[offset[0] - NODE_DIR1_BLOCK]);
926 case NODE_DIR1_BLOCK:
927 case NODE_DIR2_BLOCK:
928 err = truncate_dnode(&dn);
931 case NODE_IND1_BLOCK:
932 case NODE_IND2_BLOCK:
933 err = truncate_nodes(&dn, nofs, offset[1], 2);
936 case NODE_DIND_BLOCK:
937 err = truncate_nodes(&dn, nofs, offset[1], 3);
944 if (err < 0 && err != -ENOENT)
946 if (offset[1] == 0 &&
947 ri->i_nid[offset[0] - NODE_DIR1_BLOCK]) {
949 BUG_ON(page->mapping != NODE_MAPPING(sbi));
950 f2fs_wait_on_page_writeback(page, NODE, true);
951 ri->i_nid[offset[0] - NODE_DIR1_BLOCK] = 0;
952 set_page_dirty(page);
960 f2fs_put_page(page, 0);
961 trace_f2fs_truncate_inode_blocks_exit(inode, err);
962 return err > 0 ? 0 : err;
965 /* caller must lock inode page */
966 int truncate_xattr_node(struct inode *inode)
968 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
969 nid_t nid = F2FS_I(inode)->i_xattr_nid;
970 struct dnode_of_data dn;
976 npage = get_node_page(sbi, nid);
978 return PTR_ERR(npage);
980 f2fs_i_xnid_write(inode, 0);
982 set_new_dnode(&dn, inode, NULL, npage, nid);
988 * Caller should grab and release a rwsem by calling f2fs_lock_op() and
991 int remove_inode_page(struct inode *inode)
993 struct dnode_of_data dn;
996 set_new_dnode(&dn, inode, NULL, NULL, inode->i_ino);
997 err = get_dnode_of_data(&dn, 0, LOOKUP_NODE);
1001 err = truncate_xattr_node(inode);
1003 f2fs_put_dnode(&dn);
1007 /* remove potential inline_data blocks */
1008 if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
1009 S_ISLNK(inode->i_mode))
1010 truncate_data_blocks_range(&dn, 1);
1012 /* 0 is possible, after f2fs_new_inode() has failed */
1013 f2fs_bug_on(F2FS_I_SB(inode),
1014 inode->i_blocks != 0 && inode->i_blocks != 8);
1016 /* will put inode & node pages */
1021 struct page *new_inode_page(struct inode *inode)
1023 struct dnode_of_data dn;
1025 /* allocate inode page for new inode */
1026 set_new_dnode(&dn, inode, NULL, NULL, inode->i_ino);
1028 /* caller should f2fs_put_page(page, 1); */
1029 return new_node_page(&dn, 0);
1032 struct page *new_node_page(struct dnode_of_data *dn, unsigned int ofs)
1034 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
1035 struct node_info new_ni;
1039 if (unlikely(is_inode_flag_set(dn->inode, FI_NO_ALLOC)))
1040 return ERR_PTR(-EPERM);
1042 page = f2fs_grab_cache_page(NODE_MAPPING(sbi), dn->nid, false);
1044 return ERR_PTR(-ENOMEM);
1046 if (unlikely((err = inc_valid_node_count(sbi, dn->inode, !ofs))))
1049 #ifdef CONFIG_F2FS_CHECK_FS
1050 get_node_info(sbi, dn->nid, &new_ni);
1051 f2fs_bug_on(sbi, new_ni.blk_addr != NULL_ADDR);
1053 new_ni.nid = dn->nid;
1054 new_ni.ino = dn->inode->i_ino;
1055 new_ni.blk_addr = NULL_ADDR;
1058 set_node_addr(sbi, &new_ni, NEW_ADDR, false);
1060 f2fs_wait_on_page_writeback(page, NODE, true);
1061 fill_node_footer(page, dn->nid, dn->inode->i_ino, ofs, true);
1062 set_cold_node(dn->inode, page);
1063 if (!PageUptodate(page))
1064 SetPageUptodate(page);
1065 if (set_page_dirty(page))
1066 dn->node_changed = true;
1068 if (f2fs_has_xattr_block(ofs))
1069 f2fs_i_xnid_write(dn->inode, dn->nid);
1072 inc_valid_inode_count(sbi);
1076 clear_node_page_dirty(page);
1077 f2fs_put_page(page, 1);
1078 return ERR_PTR(err);
1082 * Caller should do after getting the following values.
1083 * 0: f2fs_put_page(page, 0)
1084 * LOCKED_PAGE or error: f2fs_put_page(page, 1)
1086 static int read_node_page(struct page *page, int op_flags)
1088 struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1089 struct node_info ni;
1090 struct f2fs_io_info fio = {
1094 .op_flags = op_flags,
1096 .encrypted_page = NULL,
1099 if (PageUptodate(page))
1102 get_node_info(sbi, page->index, &ni);
1104 if (unlikely(ni.blk_addr == NULL_ADDR)) {
1105 ClearPageUptodate(page);
1109 fio.new_blkaddr = fio.old_blkaddr = ni.blk_addr;
1110 return f2fs_submit_page_bio(&fio);
1114 * Readahead a node page
1116 void ra_node_page(struct f2fs_sb_info *sbi, nid_t nid)
1123 f2fs_bug_on(sbi, check_nid_range(sbi, nid));
1126 apage = radix_tree_lookup(&NODE_MAPPING(sbi)->page_tree, nid);
1131 apage = f2fs_grab_cache_page(NODE_MAPPING(sbi), nid, false);
1135 err = read_node_page(apage, REQ_RAHEAD);
1136 f2fs_put_page(apage, err ? 1 : 0);
1139 static struct page *__get_node_page(struct f2fs_sb_info *sbi, pgoff_t nid,
1140 struct page *parent, int start)
1146 return ERR_PTR(-ENOENT);
1147 f2fs_bug_on(sbi, check_nid_range(sbi, nid));
1149 page = f2fs_grab_cache_page(NODE_MAPPING(sbi), nid, false);
1151 return ERR_PTR(-ENOMEM);
1153 err = read_node_page(page, 0);
1155 f2fs_put_page(page, 1);
1156 return ERR_PTR(err);
1157 } else if (err == LOCKED_PAGE) {
1163 ra_node_pages(parent, start + 1, MAX_RA_NODE);
1167 if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1168 f2fs_put_page(page, 1);
1172 if (unlikely(!PageUptodate(page))) {
1177 if (!f2fs_inode_chksum_verify(sbi, page)) {
1182 if(unlikely(nid != nid_of_node(page))) {
1183 f2fs_msg(sbi->sb, KERN_WARNING, "inconsistent node block, "
1184 "nid:%lu, node_footer[nid:%u,ino:%u,ofs:%u,cpver:%llu,blkaddr:%u]",
1185 nid, nid_of_node(page), ino_of_node(page),
1186 ofs_of_node(page), cpver_of_node(page),
1187 next_blkaddr_of_node(page));
1190 ClearPageUptodate(page);
1191 f2fs_put_page(page, 1);
1192 return ERR_PTR(err);
1197 struct page *get_node_page(struct f2fs_sb_info *sbi, pgoff_t nid)
1199 return __get_node_page(sbi, nid, NULL, 0);
1202 struct page *get_node_page_ra(struct page *parent, int start)
1204 struct f2fs_sb_info *sbi = F2FS_P_SB(parent);
1205 nid_t nid = get_nid(parent, start, false);
1207 return __get_node_page(sbi, nid, parent, start);
1210 static void flush_inline_data(struct f2fs_sb_info *sbi, nid_t ino)
1212 struct inode *inode;
1216 /* should flush inline_data before evict_inode */
1217 inode = ilookup(sbi->sb, ino);
1221 page = pagecache_get_page(inode->i_mapping, 0, FGP_LOCK|FGP_NOWAIT, 0);
1225 if (!PageUptodate(page))
1228 if (!PageDirty(page))
1231 if (!clear_page_dirty_for_io(page))
1234 ret = f2fs_write_inline_data(inode, page);
1235 inode_dec_dirty_pages(inode);
1236 remove_dirty_inode(inode);
1238 set_page_dirty(page);
1240 f2fs_put_page(page, 1);
1245 void move_node_page(struct page *node_page, int gc_type)
1247 if (gc_type == FG_GC) {
1248 struct f2fs_sb_info *sbi = F2FS_P_SB(node_page);
1249 struct writeback_control wbc = {
1250 .sync_mode = WB_SYNC_ALL,
1255 set_page_dirty(node_page);
1256 f2fs_wait_on_page_writeback(node_page, NODE, true);
1258 f2fs_bug_on(sbi, PageWriteback(node_page));
1259 if (!clear_page_dirty_for_io(node_page))
1262 if (NODE_MAPPING(sbi)->a_ops->writepage(node_page, &wbc))
1263 unlock_page(node_page);
1266 /* set page dirty and write it */
1267 if (!PageWriteback(node_page))
1268 set_page_dirty(node_page);
1271 unlock_page(node_page);
1273 f2fs_put_page(node_page, 0);
1276 static struct page *last_fsync_dnode(struct f2fs_sb_info *sbi, nid_t ino)
1279 struct pagevec pvec;
1280 struct page *last_page = NULL;
1282 pagevec_init(&pvec, 0);
1286 while (index <= end) {
1288 nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1289 PAGECACHE_TAG_DIRTY,
1290 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
1294 for (i = 0; i < nr_pages; i++) {
1295 struct page *page = pvec.pages[i];
1297 if (unlikely(f2fs_cp_error(sbi))) {
1298 f2fs_put_page(last_page, 0);
1299 pagevec_release(&pvec);
1300 return ERR_PTR(-EIO);
1303 if (!IS_DNODE(page) || !is_cold_node(page))
1305 if (ino_of_node(page) != ino)
1310 if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1315 if (ino_of_node(page) != ino)
1316 goto continue_unlock;
1318 if (!PageDirty(page)) {
1319 /* someone wrote it for us */
1320 goto continue_unlock;
1324 f2fs_put_page(last_page, 0);
1330 pagevec_release(&pvec);
1336 static int __write_node_page(struct page *page, bool atomic, bool *submitted,
1337 struct writeback_control *wbc, bool do_balance,
1338 enum iostat_type io_type)
1340 struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1342 struct node_info ni;
1343 struct f2fs_io_info fio = {
1345 .ino = ino_of_node(page),
1348 .op_flags = wbc_to_write_flags(wbc),
1350 .encrypted_page = NULL,
1355 trace_f2fs_writepage(page, NODE);
1357 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
1359 if (unlikely(f2fs_cp_error(sbi)))
1362 /* get old block addr of this node page */
1363 nid = nid_of_node(page);
1364 f2fs_bug_on(sbi, page->index != nid);
1366 if (wbc->for_reclaim) {
1367 if (!down_read_trylock(&sbi->node_write))
1370 down_read(&sbi->node_write);
1373 get_node_info(sbi, nid, &ni);
1375 /* This page is already truncated */
1376 if (unlikely(ni.blk_addr == NULL_ADDR)) {
1377 ClearPageUptodate(page);
1378 dec_page_count(sbi, F2FS_DIRTY_NODES);
1379 up_read(&sbi->node_write);
1384 if (atomic && !test_opt(sbi, NOBARRIER))
1385 fio.op_flags |= REQ_PREFLUSH | REQ_FUA;
1387 set_page_writeback(page);
1388 fio.old_blkaddr = ni.blk_addr;
1389 write_node_page(nid, &fio);
1390 set_node_addr(sbi, &ni, fio.new_blkaddr, is_fsync_dnode(page));
1391 dec_page_count(sbi, F2FS_DIRTY_NODES);
1392 up_read(&sbi->node_write);
1394 if (wbc->for_reclaim) {
1395 f2fs_submit_merged_write_cond(sbi, page->mapping->host, 0,
1402 if (unlikely(f2fs_cp_error(sbi))) {
1403 f2fs_submit_merged_write(sbi, NODE);
1407 *submitted = fio.submitted;
1410 f2fs_balance_fs(sbi, false);
1414 redirty_page_for_writepage(wbc, page);
1415 return AOP_WRITEPAGE_ACTIVATE;
1418 static int f2fs_write_node_page(struct page *page,
1419 struct writeback_control *wbc)
1421 return __write_node_page(page, false, NULL, wbc, false, FS_NODE_IO);
1424 int fsync_node_pages(struct f2fs_sb_info *sbi, struct inode *inode,
1425 struct writeback_control *wbc, bool atomic)
1428 pgoff_t last_idx = ULONG_MAX;
1429 struct pagevec pvec;
1431 struct page *last_page = NULL;
1432 bool marked = false;
1433 nid_t ino = inode->i_ino;
1436 last_page = last_fsync_dnode(sbi, ino);
1437 if (IS_ERR_OR_NULL(last_page))
1438 return PTR_ERR_OR_ZERO(last_page);
1441 pagevec_init(&pvec, 0);
1445 while (index <= end) {
1447 nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1448 PAGECACHE_TAG_DIRTY,
1449 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
1453 for (i = 0; i < nr_pages; i++) {
1454 struct page *page = pvec.pages[i];
1455 bool submitted = false;
1457 if (unlikely(f2fs_cp_error(sbi))) {
1458 f2fs_put_page(last_page, 0);
1459 pagevec_release(&pvec);
1464 if (!IS_DNODE(page) || !is_cold_node(page))
1466 if (ino_of_node(page) != ino)
1471 if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1476 if (ino_of_node(page) != ino)
1477 goto continue_unlock;
1479 if (!PageDirty(page) && page != last_page) {
1480 /* someone wrote it for us */
1481 goto continue_unlock;
1484 f2fs_wait_on_page_writeback(page, NODE, true);
1485 BUG_ON(PageWriteback(page));
1487 set_fsync_mark(page, 0);
1488 set_dentry_mark(page, 0);
1490 if (!atomic || page == last_page) {
1491 set_fsync_mark(page, 1);
1492 if (IS_INODE(page)) {
1493 if (is_inode_flag_set(inode,
1495 update_inode(inode, page);
1496 set_dentry_mark(page,
1497 need_dentry_mark(sbi, ino));
1499 /* may be written by other thread */
1500 if (!PageDirty(page))
1501 set_page_dirty(page);
1504 if (!clear_page_dirty_for_io(page))
1505 goto continue_unlock;
1507 ret = __write_node_page(page, atomic &&
1509 &submitted, wbc, true,
1513 f2fs_put_page(last_page, 0);
1515 } else if (submitted) {
1516 last_idx = page->index;
1519 if (page == last_page) {
1520 f2fs_put_page(page, 0);
1525 pagevec_release(&pvec);
1531 if (!ret && atomic && !marked) {
1532 f2fs_msg(sbi->sb, KERN_DEBUG,
1533 "Retry to write fsync mark: ino=%u, idx=%lx",
1534 ino, last_page->index);
1535 lock_page(last_page);
1536 f2fs_wait_on_page_writeback(last_page, NODE, true);
1537 set_page_dirty(last_page);
1538 unlock_page(last_page);
1542 if (last_idx != ULONG_MAX)
1543 f2fs_submit_merged_write_cond(sbi, NULL, ino, last_idx, NODE);
1544 return ret ? -EIO: 0;
1547 int sync_node_pages(struct f2fs_sb_info *sbi, struct writeback_control *wbc,
1548 bool do_balance, enum iostat_type io_type)
1551 struct pagevec pvec;
1556 pagevec_init(&pvec, 0);
1562 while (index <= end) {
1564 nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1565 PAGECACHE_TAG_DIRTY,
1566 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
1570 for (i = 0; i < nr_pages; i++) {
1571 struct page *page = pvec.pages[i];
1572 bool submitted = false;
1574 if (unlikely(f2fs_cp_error(sbi))) {
1575 pagevec_release(&pvec);
1581 * flushing sequence with step:
1586 if (step == 0 && IS_DNODE(page))
1588 if (step == 1 && (!IS_DNODE(page) ||
1589 is_cold_node(page)))
1591 if (step == 2 && (!IS_DNODE(page) ||
1592 !is_cold_node(page)))
1595 if (!trylock_page(page))
1598 if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1604 if (!PageDirty(page)) {
1605 /* someone wrote it for us */
1606 goto continue_unlock;
1609 /* flush inline_data */
1610 if (is_inline_node(page)) {
1611 clear_inline_node(page);
1613 flush_inline_data(sbi, ino_of_node(page));
1617 f2fs_wait_on_page_writeback(page, NODE, true);
1619 BUG_ON(PageWriteback(page));
1620 if (!clear_page_dirty_for_io(page))
1621 goto continue_unlock;
1623 set_fsync_mark(page, 0);
1624 set_dentry_mark(page, 0);
1626 ret = __write_node_page(page, false, &submitted,
1627 wbc, do_balance, io_type);
1633 if (--wbc->nr_to_write == 0)
1636 pagevec_release(&pvec);
1639 if (wbc->nr_to_write == 0) {
1651 f2fs_submit_merged_write(sbi, NODE);
1655 int wait_on_node_pages_writeback(struct f2fs_sb_info *sbi, nid_t ino)
1657 pgoff_t index = 0, end = ULONG_MAX;
1658 struct pagevec pvec;
1661 pagevec_init(&pvec, 0);
1663 while (index <= end) {
1665 nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1666 PAGECACHE_TAG_WRITEBACK,
1667 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
1671 for (i = 0; i < nr_pages; i++) {
1672 struct page *page = pvec.pages[i];
1674 /* until radix tree lookup accepts end_index */
1675 if (unlikely(page->index > end))
1678 if (ino && ino_of_node(page) == ino) {
1679 f2fs_wait_on_page_writeback(page, NODE, true);
1680 if (TestClearPageError(page))
1684 pagevec_release(&pvec);
1688 ret2 = filemap_check_errors(NODE_MAPPING(sbi));
1694 static int f2fs_write_node_pages(struct address_space *mapping,
1695 struct writeback_control *wbc)
1697 struct f2fs_sb_info *sbi = F2FS_M_SB(mapping);
1698 struct blk_plug plug;
1701 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
1704 /* balancing f2fs's metadata in background */
1705 f2fs_balance_fs_bg(sbi);
1707 /* collect a number of dirty node pages and write together */
1708 if (get_pages(sbi, F2FS_DIRTY_NODES) < nr_pages_to_skip(sbi, NODE))
1711 trace_f2fs_writepages(mapping->host, wbc, NODE);
1713 diff = nr_pages_to_write(sbi, NODE, wbc);
1714 wbc->sync_mode = WB_SYNC_NONE;
1715 blk_start_plug(&plug);
1716 sync_node_pages(sbi, wbc, true, FS_NODE_IO);
1717 blk_finish_plug(&plug);
1718 wbc->nr_to_write = max((long)0, wbc->nr_to_write - diff);
1722 wbc->pages_skipped += get_pages(sbi, F2FS_DIRTY_NODES);
1723 trace_f2fs_writepages(mapping->host, wbc, NODE);
1727 static int f2fs_set_node_page_dirty(struct page *page)
1729 trace_f2fs_set_page_dirty(page, NODE);
1731 if (!PageUptodate(page))
1732 SetPageUptodate(page);
1733 if (!PageDirty(page)) {
1734 f2fs_set_page_dirty_nobuffers(page);
1735 inc_page_count(F2FS_P_SB(page), F2FS_DIRTY_NODES);
1736 SetPagePrivate(page);
1737 f2fs_trace_pid(page);
1744 * Structure of the f2fs node operations
1746 const struct address_space_operations f2fs_node_aops = {
1747 .writepage = f2fs_write_node_page,
1748 .writepages = f2fs_write_node_pages,
1749 .set_page_dirty = f2fs_set_node_page_dirty,
1750 .invalidatepage = f2fs_invalidate_page,
1751 .releasepage = f2fs_release_page,
1752 #ifdef CONFIG_MIGRATION
1753 .migratepage = f2fs_migrate_page,
1757 static struct free_nid *__lookup_free_nid_list(struct f2fs_nm_info *nm_i,
1760 return radix_tree_lookup(&nm_i->free_nid_root, n);
1763 static int __insert_free_nid(struct f2fs_sb_info *sbi,
1764 struct free_nid *i, enum nid_state state, bool new)
1766 struct f2fs_nm_info *nm_i = NM_I(sbi);
1769 int err = radix_tree_insert(&nm_i->free_nid_root, i->nid, i);
1774 f2fs_bug_on(sbi, state != i->state);
1775 nm_i->nid_cnt[state]++;
1776 if (state == FREE_NID)
1777 list_add_tail(&i->list, &nm_i->free_nid_list);
1781 static void __remove_free_nid(struct f2fs_sb_info *sbi,
1782 struct free_nid *i, enum nid_state state, bool reuse)
1784 struct f2fs_nm_info *nm_i = NM_I(sbi);
1786 f2fs_bug_on(sbi, state != i->state);
1787 nm_i->nid_cnt[state]--;
1788 if (state == FREE_NID)
1791 radix_tree_delete(&nm_i->free_nid_root, i->nid);
1794 /* return if the nid is recognized as free */
1795 static bool add_free_nid(struct f2fs_sb_info *sbi, nid_t nid, bool build)
1797 struct f2fs_nm_info *nm_i = NM_I(sbi);
1798 struct free_nid *i, *e;
1799 struct nat_entry *ne;
1803 /* 0 nid should not be used */
1804 if (unlikely(nid == 0))
1807 i = f2fs_kmem_cache_alloc(free_nid_slab, GFP_NOFS);
1809 i->state = FREE_NID;
1811 if (radix_tree_preload(GFP_NOFS))
1814 spin_lock(&nm_i->nid_list_lock);
1822 * - __insert_nid_to_list(PREALLOC_NID)
1823 * - f2fs_balance_fs_bg
1825 * - __build_free_nids
1828 * - __lookup_nat_cache
1830 * - init_inode_metadata
1835 * - __remove_nid_from_list(PREALLOC_NID)
1836 * - __insert_nid_to_list(FREE_NID)
1838 ne = __lookup_nat_cache(nm_i, nid);
1839 if (ne && (!get_nat_flag(ne, IS_CHECKPOINTED) ||
1840 nat_get_blkaddr(ne) != NULL_ADDR))
1843 e = __lookup_free_nid_list(nm_i, nid);
1845 if (e->state == FREE_NID)
1851 err = __insert_free_nid(sbi, i, FREE_NID, true);
1853 spin_unlock(&nm_i->nid_list_lock);
1854 radix_tree_preload_end();
1857 kmem_cache_free(free_nid_slab, i);
1861 static void remove_free_nid(struct f2fs_sb_info *sbi, nid_t nid)
1863 struct f2fs_nm_info *nm_i = NM_I(sbi);
1865 bool need_free = false;
1867 spin_lock(&nm_i->nid_list_lock);
1868 i = __lookup_free_nid_list(nm_i, nid);
1869 if (i && i->state == FREE_NID) {
1870 __remove_free_nid(sbi, i, FREE_NID, false);
1873 spin_unlock(&nm_i->nid_list_lock);
1876 kmem_cache_free(free_nid_slab, i);
1879 static void update_free_nid_bitmap(struct f2fs_sb_info *sbi, nid_t nid,
1880 bool set, bool build)
1882 struct f2fs_nm_info *nm_i = NM_I(sbi);
1883 unsigned int nat_ofs = NAT_BLOCK_OFFSET(nid);
1884 unsigned int nid_ofs = nid - START_NID(nid);
1886 if (!test_bit_le(nat_ofs, nm_i->nat_block_bitmap))
1890 __set_bit_le(nid_ofs, nm_i->free_nid_bitmap[nat_ofs]);
1892 __clear_bit_le(nid_ofs, nm_i->free_nid_bitmap[nat_ofs]);
1895 nm_i->free_nid_count[nat_ofs]++;
1897 nm_i->free_nid_count[nat_ofs]--;
1900 static void scan_nat_page(struct f2fs_sb_info *sbi,
1901 struct page *nat_page, nid_t start_nid)
1903 struct f2fs_nm_info *nm_i = NM_I(sbi);
1904 struct f2fs_nat_block *nat_blk = page_address(nat_page);
1906 unsigned int nat_ofs = NAT_BLOCK_OFFSET(start_nid);
1909 if (test_bit_le(nat_ofs, nm_i->nat_block_bitmap))
1912 __set_bit_le(nat_ofs, nm_i->nat_block_bitmap);
1914 i = start_nid % NAT_ENTRY_PER_BLOCK;
1916 for (; i < NAT_ENTRY_PER_BLOCK; i++, start_nid++) {
1919 if (unlikely(start_nid >= nm_i->max_nid))
1922 blk_addr = le32_to_cpu(nat_blk->entries[i].block_addr);
1923 f2fs_bug_on(sbi, blk_addr == NEW_ADDR);
1924 if (blk_addr == NULL_ADDR)
1925 freed = add_free_nid(sbi, start_nid, true);
1926 spin_lock(&NM_I(sbi)->nid_list_lock);
1927 update_free_nid_bitmap(sbi, start_nid, freed, true);
1928 spin_unlock(&NM_I(sbi)->nid_list_lock);
1932 static void scan_free_nid_bits(struct f2fs_sb_info *sbi)
1934 struct f2fs_nm_info *nm_i = NM_I(sbi);
1935 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
1936 struct f2fs_journal *journal = curseg->journal;
1937 unsigned int i, idx;
1939 down_read(&nm_i->nat_tree_lock);
1941 for (i = 0; i < nm_i->nat_blocks; i++) {
1942 if (!test_bit_le(i, nm_i->nat_block_bitmap))
1944 if (!nm_i->free_nid_count[i])
1946 for (idx = 0; idx < NAT_ENTRY_PER_BLOCK; idx++) {
1949 if (!test_bit_le(idx, nm_i->free_nid_bitmap[i]))
1952 nid = i * NAT_ENTRY_PER_BLOCK + idx;
1953 add_free_nid(sbi, nid, true);
1955 if (nm_i->nid_cnt[FREE_NID] >= MAX_FREE_NIDS)
1960 down_read(&curseg->journal_rwsem);
1961 for (i = 0; i < nats_in_cursum(journal); i++) {
1965 addr = le32_to_cpu(nat_in_journal(journal, i).block_addr);
1966 nid = le32_to_cpu(nid_in_journal(journal, i));
1967 if (addr == NULL_ADDR)
1968 add_free_nid(sbi, nid, true);
1970 remove_free_nid(sbi, nid);
1972 up_read(&curseg->journal_rwsem);
1973 up_read(&nm_i->nat_tree_lock);
1976 static void __build_free_nids(struct f2fs_sb_info *sbi, bool sync, bool mount)
1978 struct f2fs_nm_info *nm_i = NM_I(sbi);
1979 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
1980 struct f2fs_journal *journal = curseg->journal;
1982 nid_t nid = nm_i->next_scan_nid;
1984 if (unlikely(nid >= nm_i->max_nid))
1987 /* Enough entries */
1988 if (nm_i->nid_cnt[FREE_NID] >= NAT_ENTRY_PER_BLOCK)
1991 if (!sync && !available_free_memory(sbi, FREE_NIDS))
1995 /* try to find free nids in free_nid_bitmap */
1996 scan_free_nid_bits(sbi);
1998 if (nm_i->nid_cnt[FREE_NID])
2002 /* readahead nat pages to be scanned */
2003 ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nid), FREE_NID_PAGES,
2006 down_read(&nm_i->nat_tree_lock);
2009 struct page *page = get_current_nat_page(sbi, nid);
2011 scan_nat_page(sbi, page, nid);
2012 f2fs_put_page(page, 1);
2014 nid += (NAT_ENTRY_PER_BLOCK - (nid % NAT_ENTRY_PER_BLOCK));
2015 if (unlikely(nid >= nm_i->max_nid))
2018 if (++i >= FREE_NID_PAGES)
2022 /* go to the next free nat pages to find free nids abundantly */
2023 nm_i->next_scan_nid = nid;
2025 /* find free nids from current sum_pages */
2026 down_read(&curseg->journal_rwsem);
2027 for (i = 0; i < nats_in_cursum(journal); i++) {
2030 addr = le32_to_cpu(nat_in_journal(journal, i).block_addr);
2031 nid = le32_to_cpu(nid_in_journal(journal, i));
2032 if (addr == NULL_ADDR)
2033 add_free_nid(sbi, nid, true);
2035 remove_free_nid(sbi, nid);
2037 up_read(&curseg->journal_rwsem);
2038 up_read(&nm_i->nat_tree_lock);
2040 ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nm_i->next_scan_nid),
2041 nm_i->ra_nid_pages, META_NAT, false);
2044 void build_free_nids(struct f2fs_sb_info *sbi, bool sync, bool mount)
2046 mutex_lock(&NM_I(sbi)->build_lock);
2047 __build_free_nids(sbi, sync, mount);
2048 mutex_unlock(&NM_I(sbi)->build_lock);
2052 * If this function returns success, caller can obtain a new nid
2053 * from second parameter of this function.
2054 * The returned nid could be used ino as well as nid when inode is created.
2056 bool alloc_nid(struct f2fs_sb_info *sbi, nid_t *nid)
2058 struct f2fs_nm_info *nm_i = NM_I(sbi);
2059 struct free_nid *i = NULL;
2061 #ifdef CONFIG_F2FS_FAULT_INJECTION
2062 if (time_to_inject(sbi, FAULT_ALLOC_NID)) {
2063 f2fs_show_injection_info(FAULT_ALLOC_NID);
2067 spin_lock(&nm_i->nid_list_lock);
2069 if (unlikely(nm_i->available_nids == 0)) {
2070 spin_unlock(&nm_i->nid_list_lock);
2074 /* We should not use stale free nids created by build_free_nids */
2075 if (nm_i->nid_cnt[FREE_NID] && !on_build_free_nids(nm_i)) {
2076 f2fs_bug_on(sbi, list_empty(&nm_i->free_nid_list));
2077 i = list_first_entry(&nm_i->free_nid_list,
2078 struct free_nid, list);
2081 __remove_free_nid(sbi, i, FREE_NID, true);
2082 i->state = PREALLOC_NID;
2083 __insert_free_nid(sbi, i, PREALLOC_NID, false);
2084 nm_i->available_nids--;
2086 update_free_nid_bitmap(sbi, *nid, false, false);
2088 spin_unlock(&nm_i->nid_list_lock);
2091 spin_unlock(&nm_i->nid_list_lock);
2093 /* Let's scan nat pages and its caches to get free nids */
2094 build_free_nids(sbi, true, false);
2099 * alloc_nid() should be called prior to this function.
2101 void alloc_nid_done(struct f2fs_sb_info *sbi, nid_t nid)
2103 struct f2fs_nm_info *nm_i = NM_I(sbi);
2106 spin_lock(&nm_i->nid_list_lock);
2107 i = __lookup_free_nid_list(nm_i, nid);
2108 f2fs_bug_on(sbi, !i);
2109 __remove_free_nid(sbi, i, PREALLOC_NID, false);
2110 spin_unlock(&nm_i->nid_list_lock);
2112 kmem_cache_free(free_nid_slab, i);
2116 * alloc_nid() should be called prior to this function.
2118 void alloc_nid_failed(struct f2fs_sb_info *sbi, nid_t nid)
2120 struct f2fs_nm_info *nm_i = NM_I(sbi);
2122 bool need_free = false;
2127 spin_lock(&nm_i->nid_list_lock);
2128 i = __lookup_free_nid_list(nm_i, nid);
2129 f2fs_bug_on(sbi, !i);
2131 if (!available_free_memory(sbi, FREE_NIDS)) {
2132 __remove_free_nid(sbi, i, PREALLOC_NID, false);
2135 __remove_free_nid(sbi, i, PREALLOC_NID, true);
2136 i->state = FREE_NID;
2137 __insert_free_nid(sbi, i, FREE_NID, false);
2140 nm_i->available_nids++;
2142 update_free_nid_bitmap(sbi, nid, true, false);
2144 spin_unlock(&nm_i->nid_list_lock);
2147 kmem_cache_free(free_nid_slab, i);
2150 int try_to_free_nids(struct f2fs_sb_info *sbi, int nr_shrink)
2152 struct f2fs_nm_info *nm_i = NM_I(sbi);
2153 struct free_nid *i, *next;
2156 if (nm_i->nid_cnt[FREE_NID] <= MAX_FREE_NIDS)
2159 if (!mutex_trylock(&nm_i->build_lock))
2162 spin_lock(&nm_i->nid_list_lock);
2163 list_for_each_entry_safe(i, next, &nm_i->free_nid_list, list) {
2164 if (nr_shrink <= 0 ||
2165 nm_i->nid_cnt[FREE_NID] <= MAX_FREE_NIDS)
2168 __remove_free_nid(sbi, i, FREE_NID, false);
2169 kmem_cache_free(free_nid_slab, i);
2172 spin_unlock(&nm_i->nid_list_lock);
2173 mutex_unlock(&nm_i->build_lock);
2175 return nr - nr_shrink;
2178 void recover_inline_xattr(struct inode *inode, struct page *page)
2180 void *src_addr, *dst_addr;
2183 struct f2fs_inode *ri;
2185 ipage = get_node_page(F2FS_I_SB(inode), inode->i_ino);
2186 f2fs_bug_on(F2FS_I_SB(inode), IS_ERR(ipage));
2188 ri = F2FS_INODE(page);
2189 if (!(ri->i_inline & F2FS_INLINE_XATTR)) {
2190 clear_inode_flag(inode, FI_INLINE_XATTR);
2194 dst_addr = inline_xattr_addr(inode, ipage);
2195 src_addr = inline_xattr_addr(inode, page);
2196 inline_size = inline_xattr_size(inode);
2198 f2fs_wait_on_page_writeback(ipage, NODE, true);
2199 memcpy(dst_addr, src_addr, inline_size);
2201 update_inode(inode, ipage);
2202 f2fs_put_page(ipage, 1);
2205 int recover_xattr_data(struct inode *inode, struct page *page, block_t blkaddr)
2207 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
2208 nid_t prev_xnid = F2FS_I(inode)->i_xattr_nid;
2210 struct dnode_of_data dn;
2211 struct node_info ni;
2217 /* 1: invalidate the previous xattr nid */
2218 get_node_info(sbi, prev_xnid, &ni);
2219 f2fs_bug_on(sbi, ni.blk_addr == NULL_ADDR);
2220 invalidate_blocks(sbi, ni.blk_addr);
2221 dec_valid_node_count(sbi, inode, false);
2222 set_node_addr(sbi, &ni, NULL_ADDR, false);
2225 /* 2: update xattr nid in inode */
2226 if (!alloc_nid(sbi, &new_xnid))
2229 set_new_dnode(&dn, inode, NULL, NULL, new_xnid);
2230 xpage = new_node_page(&dn, XATTR_NODE_OFFSET);
2231 if (IS_ERR(xpage)) {
2232 alloc_nid_failed(sbi, new_xnid);
2233 return PTR_ERR(xpage);
2236 alloc_nid_done(sbi, new_xnid);
2237 update_inode_page(inode);
2239 /* 3: update and set xattr node page dirty */
2240 memcpy(F2FS_NODE(xpage), F2FS_NODE(page), VALID_XATTR_BLOCK_SIZE);
2242 set_page_dirty(xpage);
2243 f2fs_put_page(xpage, 1);
2248 int recover_inode_page(struct f2fs_sb_info *sbi, struct page *page)
2250 struct f2fs_inode *src, *dst;
2251 nid_t ino = ino_of_node(page);
2252 struct node_info old_ni, new_ni;
2255 get_node_info(sbi, ino, &old_ni);
2257 if (unlikely(old_ni.blk_addr != NULL_ADDR))
2260 ipage = f2fs_grab_cache_page(NODE_MAPPING(sbi), ino, false);
2262 congestion_wait(BLK_RW_ASYNC, HZ/50);
2266 /* Should not use this inode from free nid list */
2267 remove_free_nid(sbi, ino);
2269 if (!PageUptodate(ipage))
2270 SetPageUptodate(ipage);
2271 fill_node_footer(ipage, ino, ino, 0, true);
2273 src = F2FS_INODE(page);
2274 dst = F2FS_INODE(ipage);
2276 memcpy(dst, src, (unsigned long)&src->i_ext - (unsigned long)src);
2278 dst->i_blocks = cpu_to_le64(1);
2279 dst->i_links = cpu_to_le32(1);
2280 dst->i_xattr_nid = 0;
2281 dst->i_inline = src->i_inline & (F2FS_INLINE_XATTR | F2FS_EXTRA_ATTR);
2282 if (dst->i_inline & F2FS_EXTRA_ATTR) {
2283 dst->i_extra_isize = src->i_extra_isize;
2285 if (f2fs_sb_has_flexible_inline_xattr(sbi->sb) &&
2286 F2FS_FITS_IN_INODE(src, le16_to_cpu(src->i_extra_isize),
2287 i_inline_xattr_size))
2288 dst->i_inline_xattr_size = src->i_inline_xattr_size;
2290 if (f2fs_sb_has_project_quota(sbi->sb) &&
2291 F2FS_FITS_IN_INODE(src, le16_to_cpu(src->i_extra_isize),
2293 dst->i_projid = src->i_projid;
2299 if (unlikely(inc_valid_node_count(sbi, NULL, true)))
2301 set_node_addr(sbi, &new_ni, NEW_ADDR, false);
2302 inc_valid_inode_count(sbi);
2303 set_page_dirty(ipage);
2304 f2fs_put_page(ipage, 1);
2308 int restore_node_summary(struct f2fs_sb_info *sbi,
2309 unsigned int segno, struct f2fs_summary_block *sum)
2311 struct f2fs_node *rn;
2312 struct f2fs_summary *sum_entry;
2314 int i, idx, last_offset, nrpages;
2316 /* scan the node segment */
2317 last_offset = sbi->blocks_per_seg;
2318 addr = START_BLOCK(sbi, segno);
2319 sum_entry = &sum->entries[0];
2321 for (i = 0; i < last_offset; i += nrpages, addr += nrpages) {
2322 nrpages = min(last_offset - i, BIO_MAX_PAGES);
2324 /* readahead node pages */
2325 ra_meta_pages(sbi, addr, nrpages, META_POR, true);
2327 for (idx = addr; idx < addr + nrpages; idx++) {
2328 struct page *page = get_tmp_page(sbi, idx);
2330 rn = F2FS_NODE(page);
2331 sum_entry->nid = rn->footer.nid;
2332 sum_entry->version = 0;
2333 sum_entry->ofs_in_node = 0;
2335 f2fs_put_page(page, 1);
2338 invalidate_mapping_pages(META_MAPPING(sbi), addr,
2344 static void remove_nats_in_journal(struct f2fs_sb_info *sbi)
2346 struct f2fs_nm_info *nm_i = NM_I(sbi);
2347 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
2348 struct f2fs_journal *journal = curseg->journal;
2351 down_write(&curseg->journal_rwsem);
2352 for (i = 0; i < nats_in_cursum(journal); i++) {
2353 struct nat_entry *ne;
2354 struct f2fs_nat_entry raw_ne;
2355 nid_t nid = le32_to_cpu(nid_in_journal(journal, i));
2357 raw_ne = nat_in_journal(journal, i);
2359 ne = __lookup_nat_cache(nm_i, nid);
2361 ne = grab_nat_entry(nm_i, nid, true);
2362 node_info_from_raw_nat(&ne->ni, &raw_ne);
2366 * if a free nat in journal has not been used after last
2367 * checkpoint, we should remove it from available nids,
2368 * since later we will add it again.
2370 if (!get_nat_flag(ne, IS_DIRTY) &&
2371 le32_to_cpu(raw_ne.block_addr) == NULL_ADDR) {
2372 spin_lock(&nm_i->nid_list_lock);
2373 nm_i->available_nids--;
2374 spin_unlock(&nm_i->nid_list_lock);
2377 __set_nat_cache_dirty(nm_i, ne);
2379 update_nats_in_cursum(journal, -i);
2380 up_write(&curseg->journal_rwsem);
2383 static void __adjust_nat_entry_set(struct nat_entry_set *nes,
2384 struct list_head *head, int max)
2386 struct nat_entry_set *cur;
2388 if (nes->entry_cnt >= max)
2391 list_for_each_entry(cur, head, set_list) {
2392 if (cur->entry_cnt >= nes->entry_cnt) {
2393 list_add(&nes->set_list, cur->set_list.prev);
2398 list_add_tail(&nes->set_list, head);
2401 static void __update_nat_bits(struct f2fs_sb_info *sbi, nid_t start_nid,
2404 struct f2fs_nm_info *nm_i = NM_I(sbi);
2405 unsigned int nat_index = start_nid / NAT_ENTRY_PER_BLOCK;
2406 struct f2fs_nat_block *nat_blk = page_address(page);
2410 if (!enabled_nat_bits(sbi, NULL))
2413 for (i = 0; i < NAT_ENTRY_PER_BLOCK; i++) {
2414 if (start_nid == 0 && i == 0)
2416 if (nat_blk->entries[i].block_addr)
2420 __set_bit_le(nat_index, nm_i->empty_nat_bits);
2421 __clear_bit_le(nat_index, nm_i->full_nat_bits);
2425 __clear_bit_le(nat_index, nm_i->empty_nat_bits);
2426 if (valid == NAT_ENTRY_PER_BLOCK)
2427 __set_bit_le(nat_index, nm_i->full_nat_bits);
2429 __clear_bit_le(nat_index, nm_i->full_nat_bits);
2432 static void __flush_nat_entry_set(struct f2fs_sb_info *sbi,
2433 struct nat_entry_set *set, struct cp_control *cpc)
2435 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
2436 struct f2fs_journal *journal = curseg->journal;
2437 nid_t start_nid = set->set * NAT_ENTRY_PER_BLOCK;
2438 bool to_journal = true;
2439 struct f2fs_nat_block *nat_blk;
2440 struct nat_entry *ne, *cur;
2441 struct page *page = NULL;
2444 * there are two steps to flush nat entries:
2445 * #1, flush nat entries to journal in current hot data summary block.
2446 * #2, flush nat entries to nat page.
2448 if (enabled_nat_bits(sbi, cpc) ||
2449 !__has_cursum_space(journal, set->entry_cnt, NAT_JOURNAL))
2453 down_write(&curseg->journal_rwsem);
2455 page = get_next_nat_page(sbi, start_nid);
2456 nat_blk = page_address(page);
2457 f2fs_bug_on(sbi, !nat_blk);
2460 /* flush dirty nats in nat entry set */
2461 list_for_each_entry_safe(ne, cur, &set->entry_list, list) {
2462 struct f2fs_nat_entry *raw_ne;
2463 nid_t nid = nat_get_nid(ne);
2466 f2fs_bug_on(sbi, nat_get_blkaddr(ne) == NEW_ADDR);
2469 offset = lookup_journal_in_cursum(journal,
2470 NAT_JOURNAL, nid, 1);
2471 f2fs_bug_on(sbi, offset < 0);
2472 raw_ne = &nat_in_journal(journal, offset);
2473 nid_in_journal(journal, offset) = cpu_to_le32(nid);
2475 raw_ne = &nat_blk->entries[nid - start_nid];
2477 raw_nat_from_node_info(raw_ne, &ne->ni);
2479 __clear_nat_cache_dirty(NM_I(sbi), set, ne);
2480 if (nat_get_blkaddr(ne) == NULL_ADDR) {
2481 add_free_nid(sbi, nid, false);
2482 spin_lock(&NM_I(sbi)->nid_list_lock);
2483 NM_I(sbi)->available_nids++;
2484 update_free_nid_bitmap(sbi, nid, true, false);
2485 spin_unlock(&NM_I(sbi)->nid_list_lock);
2487 spin_lock(&NM_I(sbi)->nid_list_lock);
2488 update_free_nid_bitmap(sbi, nid, false, false);
2489 spin_unlock(&NM_I(sbi)->nid_list_lock);
2494 up_write(&curseg->journal_rwsem);
2496 __update_nat_bits(sbi, start_nid, page);
2497 f2fs_put_page(page, 1);
2500 /* Allow dirty nats by node block allocation in write_begin */
2501 if (!set->entry_cnt) {
2502 radix_tree_delete(&NM_I(sbi)->nat_set_root, set->set);
2503 kmem_cache_free(nat_entry_set_slab, set);
2508 * This function is called during the checkpointing process.
2510 void flush_nat_entries(struct f2fs_sb_info *sbi, struct cp_control *cpc)
2512 struct f2fs_nm_info *nm_i = NM_I(sbi);
2513 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
2514 struct f2fs_journal *journal = curseg->journal;
2515 struct nat_entry_set *setvec[SETVEC_SIZE];
2516 struct nat_entry_set *set, *tmp;
2521 if (!nm_i->dirty_nat_cnt)
2524 down_write(&nm_i->nat_tree_lock);
2527 * if there are no enough space in journal to store dirty nat
2528 * entries, remove all entries from journal and merge them
2529 * into nat entry set.
2531 if (enabled_nat_bits(sbi, cpc) ||
2532 !__has_cursum_space(journal, nm_i->dirty_nat_cnt, NAT_JOURNAL))
2533 remove_nats_in_journal(sbi);
2535 while ((found = __gang_lookup_nat_set(nm_i,
2536 set_idx, SETVEC_SIZE, setvec))) {
2538 set_idx = setvec[found - 1]->set + 1;
2539 for (idx = 0; idx < found; idx++)
2540 __adjust_nat_entry_set(setvec[idx], &sets,
2541 MAX_NAT_JENTRIES(journal));
2544 /* flush dirty nats in nat entry set */
2545 list_for_each_entry_safe(set, tmp, &sets, set_list)
2546 __flush_nat_entry_set(sbi, set, cpc);
2548 up_write(&nm_i->nat_tree_lock);
2549 /* Allow dirty nats by node block allocation in write_begin */
2552 static int __get_nat_bitmaps(struct f2fs_sb_info *sbi)
2554 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
2555 struct f2fs_nm_info *nm_i = NM_I(sbi);
2556 unsigned int nat_bits_bytes = nm_i->nat_blocks / BITS_PER_BYTE;
2558 __u64 cp_ver = cur_cp_version(ckpt);
2559 block_t nat_bits_addr;
2561 if (!enabled_nat_bits(sbi, NULL))
2564 nm_i->nat_bits_blocks = F2FS_BYTES_TO_BLK((nat_bits_bytes << 1) + 8 +
2566 nm_i->nat_bits = kzalloc(nm_i->nat_bits_blocks << F2FS_BLKSIZE_BITS,
2568 if (!nm_i->nat_bits)
2571 nat_bits_addr = __start_cp_addr(sbi) + sbi->blocks_per_seg -
2572 nm_i->nat_bits_blocks;
2573 for (i = 0; i < nm_i->nat_bits_blocks; i++) {
2574 struct page *page = get_meta_page(sbi, nat_bits_addr++);
2576 memcpy(nm_i->nat_bits + (i << F2FS_BLKSIZE_BITS),
2577 page_address(page), F2FS_BLKSIZE);
2578 f2fs_put_page(page, 1);
2581 cp_ver |= (cur_cp_crc(ckpt) << 32);
2582 if (cpu_to_le64(cp_ver) != *(__le64 *)nm_i->nat_bits) {
2583 disable_nat_bits(sbi, true);
2587 nm_i->full_nat_bits = nm_i->nat_bits + 8;
2588 nm_i->empty_nat_bits = nm_i->full_nat_bits + nat_bits_bytes;
2590 f2fs_msg(sbi->sb, KERN_NOTICE, "Found nat_bits in checkpoint");
2594 static inline void load_free_nid_bitmap(struct f2fs_sb_info *sbi)
2596 struct f2fs_nm_info *nm_i = NM_I(sbi);
2598 nid_t nid, last_nid;
2600 if (!enabled_nat_bits(sbi, NULL))
2603 for (i = 0; i < nm_i->nat_blocks; i++) {
2604 i = find_next_bit_le(nm_i->empty_nat_bits, nm_i->nat_blocks, i);
2605 if (i >= nm_i->nat_blocks)
2608 __set_bit_le(i, nm_i->nat_block_bitmap);
2610 nid = i * NAT_ENTRY_PER_BLOCK;
2611 last_nid = (i + 1) * NAT_ENTRY_PER_BLOCK;
2613 spin_lock(&NM_I(sbi)->nid_list_lock);
2614 for (; nid < last_nid; nid++)
2615 update_free_nid_bitmap(sbi, nid, true, true);
2616 spin_unlock(&NM_I(sbi)->nid_list_lock);
2619 for (i = 0; i < nm_i->nat_blocks; i++) {
2620 i = find_next_bit_le(nm_i->full_nat_bits, nm_i->nat_blocks, i);
2621 if (i >= nm_i->nat_blocks)
2624 __set_bit_le(i, nm_i->nat_block_bitmap);
2628 static int init_node_manager(struct f2fs_sb_info *sbi)
2630 struct f2fs_super_block *sb_raw = F2FS_RAW_SUPER(sbi);
2631 struct f2fs_nm_info *nm_i = NM_I(sbi);
2632 unsigned char *version_bitmap;
2633 unsigned int nat_segs;
2636 nm_i->nat_blkaddr = le32_to_cpu(sb_raw->nat_blkaddr);
2638 /* segment_count_nat includes pair segment so divide to 2. */
2639 nat_segs = le32_to_cpu(sb_raw->segment_count_nat) >> 1;
2640 nm_i->nat_blocks = nat_segs << le32_to_cpu(sb_raw->log_blocks_per_seg);
2641 nm_i->max_nid = NAT_ENTRY_PER_BLOCK * nm_i->nat_blocks;
2643 /* not used nids: 0, node, meta, (and root counted as valid node) */
2644 nm_i->available_nids = nm_i->max_nid - sbi->total_valid_node_count -
2645 F2FS_RESERVED_NODE_NUM;
2646 nm_i->nid_cnt[FREE_NID] = 0;
2647 nm_i->nid_cnt[PREALLOC_NID] = 0;
2649 nm_i->ram_thresh = DEF_RAM_THRESHOLD;
2650 nm_i->ra_nid_pages = DEF_RA_NID_PAGES;
2651 nm_i->dirty_nats_ratio = DEF_DIRTY_NAT_RATIO_THRESHOLD;
2653 INIT_RADIX_TREE(&nm_i->free_nid_root, GFP_ATOMIC);
2654 INIT_LIST_HEAD(&nm_i->free_nid_list);
2655 INIT_RADIX_TREE(&nm_i->nat_root, GFP_NOIO);
2656 INIT_RADIX_TREE(&nm_i->nat_set_root, GFP_NOIO);
2657 INIT_LIST_HEAD(&nm_i->nat_entries);
2659 mutex_init(&nm_i->build_lock);
2660 spin_lock_init(&nm_i->nid_list_lock);
2661 init_rwsem(&nm_i->nat_tree_lock);
2663 nm_i->next_scan_nid = le32_to_cpu(sbi->ckpt->next_free_nid);
2664 nm_i->bitmap_size = __bitmap_size(sbi, NAT_BITMAP);
2665 version_bitmap = __bitmap_ptr(sbi, NAT_BITMAP);
2666 if (!version_bitmap)
2669 nm_i->nat_bitmap = kmemdup(version_bitmap, nm_i->bitmap_size,
2671 if (!nm_i->nat_bitmap)
2674 err = __get_nat_bitmaps(sbi);
2678 #ifdef CONFIG_F2FS_CHECK_FS
2679 nm_i->nat_bitmap_mir = kmemdup(version_bitmap, nm_i->bitmap_size,
2681 if (!nm_i->nat_bitmap_mir)
2688 static int init_free_nid_cache(struct f2fs_sb_info *sbi)
2690 struct f2fs_nm_info *nm_i = NM_I(sbi);
2692 nm_i->free_nid_bitmap = kvzalloc(nm_i->nat_blocks *
2693 NAT_ENTRY_BITMAP_SIZE, GFP_KERNEL);
2694 if (!nm_i->free_nid_bitmap)
2697 nm_i->nat_block_bitmap = kvzalloc(nm_i->nat_blocks / 8,
2699 if (!nm_i->nat_block_bitmap)
2702 nm_i->free_nid_count = kvzalloc(nm_i->nat_blocks *
2703 sizeof(unsigned short), GFP_KERNEL);
2704 if (!nm_i->free_nid_count)
2709 int build_node_manager(struct f2fs_sb_info *sbi)
2713 sbi->nm_info = kzalloc(sizeof(struct f2fs_nm_info), GFP_KERNEL);
2717 err = init_node_manager(sbi);
2721 err = init_free_nid_cache(sbi);
2725 /* load free nid status from nat_bits table */
2726 load_free_nid_bitmap(sbi);
2728 build_free_nids(sbi, true, true);
2732 void destroy_node_manager(struct f2fs_sb_info *sbi)
2734 struct f2fs_nm_info *nm_i = NM_I(sbi);
2735 struct free_nid *i, *next_i;
2736 struct nat_entry *natvec[NATVEC_SIZE];
2737 struct nat_entry_set *setvec[SETVEC_SIZE];
2744 /* destroy free nid list */
2745 spin_lock(&nm_i->nid_list_lock);
2746 list_for_each_entry_safe(i, next_i, &nm_i->free_nid_list, list) {
2747 __remove_free_nid(sbi, i, FREE_NID, false);
2748 spin_unlock(&nm_i->nid_list_lock);
2749 kmem_cache_free(free_nid_slab, i);
2750 spin_lock(&nm_i->nid_list_lock);
2752 f2fs_bug_on(sbi, nm_i->nid_cnt[FREE_NID]);
2753 f2fs_bug_on(sbi, nm_i->nid_cnt[PREALLOC_NID]);
2754 f2fs_bug_on(sbi, !list_empty(&nm_i->free_nid_list));
2755 spin_unlock(&nm_i->nid_list_lock);
2757 /* destroy nat cache */
2758 down_write(&nm_i->nat_tree_lock);
2759 while ((found = __gang_lookup_nat_cache(nm_i,
2760 nid, NATVEC_SIZE, natvec))) {
2763 nid = nat_get_nid(natvec[found - 1]) + 1;
2764 for (idx = 0; idx < found; idx++)
2765 __del_from_nat_cache(nm_i, natvec[idx]);
2767 f2fs_bug_on(sbi, nm_i->nat_cnt);
2769 /* destroy nat set cache */
2771 while ((found = __gang_lookup_nat_set(nm_i,
2772 nid, SETVEC_SIZE, setvec))) {
2775 nid = setvec[found - 1]->set + 1;
2776 for (idx = 0; idx < found; idx++) {
2777 /* entry_cnt is not zero, when cp_error was occurred */
2778 f2fs_bug_on(sbi, !list_empty(&setvec[idx]->entry_list));
2779 radix_tree_delete(&nm_i->nat_set_root, setvec[idx]->set);
2780 kmem_cache_free(nat_entry_set_slab, setvec[idx]);
2783 up_write(&nm_i->nat_tree_lock);
2785 kvfree(nm_i->nat_block_bitmap);
2786 kvfree(nm_i->free_nid_bitmap);
2787 kvfree(nm_i->free_nid_count);
2789 kfree(nm_i->nat_bitmap);
2790 kfree(nm_i->nat_bits);
2791 #ifdef CONFIG_F2FS_CHECK_FS
2792 kfree(nm_i->nat_bitmap_mir);
2794 sbi->nm_info = NULL;
2798 int __init create_node_manager_caches(void)
2800 nat_entry_slab = f2fs_kmem_cache_create("nat_entry",
2801 sizeof(struct nat_entry));
2802 if (!nat_entry_slab)
2805 free_nid_slab = f2fs_kmem_cache_create("free_nid",
2806 sizeof(struct free_nid));
2808 goto destroy_nat_entry;
2810 nat_entry_set_slab = f2fs_kmem_cache_create("nat_entry_set",
2811 sizeof(struct nat_entry_set));
2812 if (!nat_entry_set_slab)
2813 goto destroy_free_nid;
2817 kmem_cache_destroy(free_nid_slab);
2819 kmem_cache_destroy(nat_entry_slab);
2824 void destroy_node_manager_caches(void)
2826 kmem_cache_destroy(nat_entry_set_slab);
2827 kmem_cache_destroy(free_nid_slab);
2828 kmem_cache_destroy(nat_entry_slab);