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_LIST] *
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 <= UPDATE_INO; 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);
78 if (!sbi->sb->s_bdi->wb.dirty_exceeded)
84 static void clear_node_page_dirty(struct page *page)
86 struct address_space *mapping = page->mapping;
87 unsigned int long flags;
89 if (PageDirty(page)) {
90 spin_lock_irqsave(&mapping->tree_lock, flags);
91 radix_tree_tag_clear(&mapping->page_tree,
94 spin_unlock_irqrestore(&mapping->tree_lock, flags);
96 clear_page_dirty_for_io(page);
97 dec_page_count(F2FS_M_SB(mapping), F2FS_DIRTY_NODES);
99 ClearPageUptodate(page);
102 static struct page *get_current_nat_page(struct f2fs_sb_info *sbi, nid_t nid)
104 pgoff_t index = current_nat_addr(sbi, nid);
105 return get_meta_page(sbi, index);
108 static struct page *get_next_nat_page(struct f2fs_sb_info *sbi, nid_t nid)
110 struct page *src_page;
111 struct page *dst_page;
116 struct f2fs_nm_info *nm_i = NM_I(sbi);
118 src_off = current_nat_addr(sbi, nid);
119 dst_off = next_nat_addr(sbi, src_off);
121 /* get current nat block page with lock */
122 src_page = get_meta_page(sbi, src_off);
123 dst_page = grab_meta_page(sbi, dst_off);
124 f2fs_bug_on(sbi, PageDirty(src_page));
126 src_addr = page_address(src_page);
127 dst_addr = page_address(dst_page);
128 memcpy(dst_addr, src_addr, PAGE_SIZE);
129 set_page_dirty(dst_page);
130 f2fs_put_page(src_page, 1);
132 set_to_next_nat(nm_i, nid);
137 static struct nat_entry *__lookup_nat_cache(struct f2fs_nm_info *nm_i, nid_t n)
139 return radix_tree_lookup(&nm_i->nat_root, n);
142 static unsigned int __gang_lookup_nat_cache(struct f2fs_nm_info *nm_i,
143 nid_t start, unsigned int nr, struct nat_entry **ep)
145 return radix_tree_gang_lookup(&nm_i->nat_root, (void **)ep, start, nr);
148 static void __del_from_nat_cache(struct f2fs_nm_info *nm_i, struct nat_entry *e)
151 radix_tree_delete(&nm_i->nat_root, nat_get_nid(e));
153 kmem_cache_free(nat_entry_slab, e);
156 static void __set_nat_cache_dirty(struct f2fs_nm_info *nm_i,
157 struct nat_entry *ne)
159 nid_t set = NAT_BLOCK_OFFSET(ne->ni.nid);
160 struct nat_entry_set *head;
162 head = radix_tree_lookup(&nm_i->nat_set_root, set);
164 head = f2fs_kmem_cache_alloc(nat_entry_set_slab, GFP_NOFS);
166 INIT_LIST_HEAD(&head->entry_list);
167 INIT_LIST_HEAD(&head->set_list);
170 f2fs_radix_tree_insert(&nm_i->nat_set_root, set, head);
173 if (get_nat_flag(ne, IS_DIRTY))
176 nm_i->dirty_nat_cnt++;
178 set_nat_flag(ne, IS_DIRTY, true);
180 if (nat_get_blkaddr(ne) == NEW_ADDR)
181 list_del_init(&ne->list);
183 list_move_tail(&ne->list, &head->entry_list);
186 static void __clear_nat_cache_dirty(struct f2fs_nm_info *nm_i,
187 struct nat_entry_set *set, struct nat_entry *ne)
189 list_move_tail(&ne->list, &nm_i->nat_entries);
190 set_nat_flag(ne, IS_DIRTY, false);
192 nm_i->dirty_nat_cnt--;
195 static unsigned int __gang_lookup_nat_set(struct f2fs_nm_info *nm_i,
196 nid_t start, unsigned int nr, struct nat_entry_set **ep)
198 return radix_tree_gang_lookup(&nm_i->nat_set_root, (void **)ep,
202 int need_dentry_mark(struct f2fs_sb_info *sbi, nid_t nid)
204 struct f2fs_nm_info *nm_i = NM_I(sbi);
208 down_read(&nm_i->nat_tree_lock);
209 e = __lookup_nat_cache(nm_i, nid);
211 if (!get_nat_flag(e, IS_CHECKPOINTED) &&
212 !get_nat_flag(e, HAS_FSYNCED_INODE))
215 up_read(&nm_i->nat_tree_lock);
219 bool is_checkpointed_node(struct f2fs_sb_info *sbi, nid_t nid)
221 struct f2fs_nm_info *nm_i = NM_I(sbi);
225 down_read(&nm_i->nat_tree_lock);
226 e = __lookup_nat_cache(nm_i, nid);
227 if (e && !get_nat_flag(e, IS_CHECKPOINTED))
229 up_read(&nm_i->nat_tree_lock);
233 bool need_inode_block_update(struct f2fs_sb_info *sbi, nid_t ino)
235 struct f2fs_nm_info *nm_i = NM_I(sbi);
237 bool need_update = true;
239 down_read(&nm_i->nat_tree_lock);
240 e = __lookup_nat_cache(nm_i, ino);
241 if (e && get_nat_flag(e, HAS_LAST_FSYNC) &&
242 (get_nat_flag(e, IS_CHECKPOINTED) ||
243 get_nat_flag(e, HAS_FSYNCED_INODE)))
245 up_read(&nm_i->nat_tree_lock);
249 static struct nat_entry *grab_nat_entry(struct f2fs_nm_info *nm_i, nid_t nid,
252 struct nat_entry *new;
255 new = f2fs_kmem_cache_alloc(nat_entry_slab, GFP_NOFS);
256 f2fs_radix_tree_insert(&nm_i->nat_root, nid, new);
258 new = kmem_cache_alloc(nat_entry_slab, GFP_NOFS);
261 if (radix_tree_insert(&nm_i->nat_root, nid, new)) {
262 kmem_cache_free(nat_entry_slab, new);
267 memset(new, 0, sizeof(struct nat_entry));
268 nat_set_nid(new, nid);
270 list_add_tail(&new->list, &nm_i->nat_entries);
275 static void cache_nat_entry(struct f2fs_sb_info *sbi, nid_t nid,
276 struct f2fs_nat_entry *ne)
278 struct f2fs_nm_info *nm_i = NM_I(sbi);
281 e = __lookup_nat_cache(nm_i, nid);
283 e = grab_nat_entry(nm_i, nid, false);
285 node_info_from_raw_nat(&e->ni, ne);
287 f2fs_bug_on(sbi, nat_get_ino(e) != le32_to_cpu(ne->ino) ||
288 nat_get_blkaddr(e) !=
289 le32_to_cpu(ne->block_addr) ||
290 nat_get_version(e) != ne->version);
294 static void set_node_addr(struct f2fs_sb_info *sbi, struct node_info *ni,
295 block_t new_blkaddr, bool fsync_done)
297 struct f2fs_nm_info *nm_i = NM_I(sbi);
300 down_write(&nm_i->nat_tree_lock);
301 e = __lookup_nat_cache(nm_i, ni->nid);
303 e = grab_nat_entry(nm_i, ni->nid, true);
304 copy_node_info(&e->ni, ni);
305 f2fs_bug_on(sbi, ni->blk_addr == NEW_ADDR);
306 } else if (new_blkaddr == NEW_ADDR) {
308 * when nid is reallocated,
309 * previous nat entry can be remained in nat cache.
310 * So, reinitialize it with new information.
312 copy_node_info(&e->ni, ni);
313 f2fs_bug_on(sbi, ni->blk_addr != NULL_ADDR);
317 f2fs_bug_on(sbi, nat_get_blkaddr(e) != ni->blk_addr);
318 f2fs_bug_on(sbi, nat_get_blkaddr(e) == NULL_ADDR &&
319 new_blkaddr == NULL_ADDR);
320 f2fs_bug_on(sbi, nat_get_blkaddr(e) == NEW_ADDR &&
321 new_blkaddr == NEW_ADDR);
322 f2fs_bug_on(sbi, nat_get_blkaddr(e) != NEW_ADDR &&
323 nat_get_blkaddr(e) != NULL_ADDR &&
324 new_blkaddr == NEW_ADDR);
326 /* increment version no as node is removed */
327 if (nat_get_blkaddr(e) != NEW_ADDR && new_blkaddr == NULL_ADDR) {
328 unsigned char version = nat_get_version(e);
329 nat_set_version(e, inc_node_version(version));
331 /* in order to reuse the nid */
332 if (nm_i->next_scan_nid > ni->nid)
333 nm_i->next_scan_nid = ni->nid;
337 nat_set_blkaddr(e, new_blkaddr);
338 if (new_blkaddr == NEW_ADDR || new_blkaddr == NULL_ADDR)
339 set_nat_flag(e, IS_CHECKPOINTED, false);
340 __set_nat_cache_dirty(nm_i, e);
342 /* update fsync_mark if its inode nat entry is still alive */
343 if (ni->nid != ni->ino)
344 e = __lookup_nat_cache(nm_i, ni->ino);
346 if (fsync_done && ni->nid == ni->ino)
347 set_nat_flag(e, HAS_FSYNCED_INODE, true);
348 set_nat_flag(e, HAS_LAST_FSYNC, fsync_done);
350 up_write(&nm_i->nat_tree_lock);
353 int try_to_free_nats(struct f2fs_sb_info *sbi, int nr_shrink)
355 struct f2fs_nm_info *nm_i = NM_I(sbi);
358 if (!down_write_trylock(&nm_i->nat_tree_lock))
361 while (nr_shrink && !list_empty(&nm_i->nat_entries)) {
362 struct nat_entry *ne;
363 ne = list_first_entry(&nm_i->nat_entries,
364 struct nat_entry, list);
365 __del_from_nat_cache(nm_i, ne);
368 up_write(&nm_i->nat_tree_lock);
369 return nr - nr_shrink;
373 * This function always returns success
375 void get_node_info(struct f2fs_sb_info *sbi, nid_t nid, struct node_info *ni)
377 struct f2fs_nm_info *nm_i = NM_I(sbi);
378 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
379 struct f2fs_journal *journal = curseg->journal;
380 nid_t start_nid = START_NID(nid);
381 struct f2fs_nat_block *nat_blk;
382 struct page *page = NULL;
383 struct f2fs_nat_entry ne;
390 /* Check nat cache */
391 down_read(&nm_i->nat_tree_lock);
392 e = __lookup_nat_cache(nm_i, nid);
394 ni->ino = nat_get_ino(e);
395 ni->blk_addr = nat_get_blkaddr(e);
396 ni->version = nat_get_version(e);
397 up_read(&nm_i->nat_tree_lock);
401 memset(&ne, 0, sizeof(struct f2fs_nat_entry));
403 /* Check current segment summary */
404 down_read(&curseg->journal_rwsem);
405 i = lookup_journal_in_cursum(journal, NAT_JOURNAL, nid, 0);
407 ne = nat_in_journal(journal, i);
408 node_info_from_raw_nat(ni, &ne);
410 up_read(&curseg->journal_rwsem);
412 up_read(&nm_i->nat_tree_lock);
416 /* Fill node_info from nat page */
417 index = current_nat_addr(sbi, nid);
418 up_read(&nm_i->nat_tree_lock);
420 page = get_meta_page(sbi, index);
421 nat_blk = (struct f2fs_nat_block *)page_address(page);
422 ne = nat_blk->entries[nid - start_nid];
423 node_info_from_raw_nat(ni, &ne);
424 f2fs_put_page(page, 1);
426 /* cache nat entry */
427 down_write(&nm_i->nat_tree_lock);
428 cache_nat_entry(sbi, nid, &ne);
429 up_write(&nm_i->nat_tree_lock);
433 * readahead MAX_RA_NODE number of node pages.
435 static void ra_node_pages(struct page *parent, int start, int n)
437 struct f2fs_sb_info *sbi = F2FS_P_SB(parent);
438 struct blk_plug plug;
442 blk_start_plug(&plug);
444 /* Then, try readahead for siblings of the desired node */
446 end = min(end, NIDS_PER_BLOCK);
447 for (i = start; i < end; i++) {
448 nid = get_nid(parent, i, false);
449 ra_node_page(sbi, nid);
452 blk_finish_plug(&plug);
455 pgoff_t get_next_page_offset(struct dnode_of_data *dn, pgoff_t pgofs)
457 const long direct_index = ADDRS_PER_INODE(dn->inode);
458 const long direct_blks = ADDRS_PER_BLOCK;
459 const long indirect_blks = ADDRS_PER_BLOCK * NIDS_PER_BLOCK;
460 unsigned int skipped_unit = ADDRS_PER_BLOCK;
461 int cur_level = dn->cur_level;
462 int max_level = dn->max_level;
468 while (max_level-- > cur_level)
469 skipped_unit *= NIDS_PER_BLOCK;
471 switch (dn->max_level) {
473 base += 2 * indirect_blks;
475 base += 2 * direct_blks;
477 base += direct_index;
480 f2fs_bug_on(F2FS_I_SB(dn->inode), 1);
483 return ((pgofs - base) / skipped_unit + 1) * skipped_unit + base;
487 * The maximum depth is four.
488 * Offset[0] will have raw inode offset.
490 static int get_node_path(struct inode *inode, long block,
491 int offset[4], unsigned int noffset[4])
493 const long direct_index = ADDRS_PER_INODE(inode);
494 const long direct_blks = ADDRS_PER_BLOCK;
495 const long dptrs_per_blk = NIDS_PER_BLOCK;
496 const long indirect_blks = ADDRS_PER_BLOCK * NIDS_PER_BLOCK;
497 const long dindirect_blks = indirect_blks * NIDS_PER_BLOCK;
503 if (block < direct_index) {
507 block -= direct_index;
508 if (block < direct_blks) {
509 offset[n++] = NODE_DIR1_BLOCK;
515 block -= direct_blks;
516 if (block < direct_blks) {
517 offset[n++] = NODE_DIR2_BLOCK;
523 block -= direct_blks;
524 if (block < indirect_blks) {
525 offset[n++] = NODE_IND1_BLOCK;
527 offset[n++] = block / direct_blks;
528 noffset[n] = 4 + offset[n - 1];
529 offset[n] = block % direct_blks;
533 block -= indirect_blks;
534 if (block < indirect_blks) {
535 offset[n++] = NODE_IND2_BLOCK;
536 noffset[n] = 4 + dptrs_per_blk;
537 offset[n++] = block / direct_blks;
538 noffset[n] = 5 + dptrs_per_blk + offset[n - 1];
539 offset[n] = block % direct_blks;
543 block -= indirect_blks;
544 if (block < dindirect_blks) {
545 offset[n++] = NODE_DIND_BLOCK;
546 noffset[n] = 5 + (dptrs_per_blk * 2);
547 offset[n++] = block / indirect_blks;
548 noffset[n] = 6 + (dptrs_per_blk * 2) +
549 offset[n - 1] * (dptrs_per_blk + 1);
550 offset[n++] = (block / direct_blks) % dptrs_per_blk;
551 noffset[n] = 7 + (dptrs_per_blk * 2) +
552 offset[n - 2] * (dptrs_per_blk + 1) +
554 offset[n] = block % direct_blks;
565 * Caller should call f2fs_put_dnode(dn).
566 * Also, it should grab and release a rwsem by calling f2fs_lock_op() and
567 * f2fs_unlock_op() only if ro is not set RDONLY_NODE.
568 * In the case of RDONLY_NODE, we don't need to care about mutex.
570 int get_dnode_of_data(struct dnode_of_data *dn, pgoff_t index, int mode)
572 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
573 struct page *npage[4];
574 struct page *parent = NULL;
576 unsigned int noffset[4];
581 level = get_node_path(dn->inode, index, offset, noffset);
585 nids[0] = dn->inode->i_ino;
586 npage[0] = dn->inode_page;
589 npage[0] = get_node_page(sbi, nids[0]);
590 if (IS_ERR(npage[0]))
591 return PTR_ERR(npage[0]);
594 /* if inline_data is set, should not report any block indices */
595 if (f2fs_has_inline_data(dn->inode) && index) {
597 f2fs_put_page(npage[0], 1);
603 nids[1] = get_nid(parent, offset[0], true);
604 dn->inode_page = npage[0];
605 dn->inode_page_locked = true;
607 /* get indirect or direct nodes */
608 for (i = 1; i <= level; i++) {
611 if (!nids[i] && mode == ALLOC_NODE) {
613 if (!alloc_nid(sbi, &(nids[i]))) {
619 npage[i] = new_node_page(dn, noffset[i]);
620 if (IS_ERR(npage[i])) {
621 alloc_nid_failed(sbi, nids[i]);
622 err = PTR_ERR(npage[i]);
626 set_nid(parent, offset[i - 1], nids[i], i == 1);
627 alloc_nid_done(sbi, nids[i]);
629 } else if (mode == LOOKUP_NODE_RA && i == level && level > 1) {
630 npage[i] = get_node_page_ra(parent, offset[i - 1]);
631 if (IS_ERR(npage[i])) {
632 err = PTR_ERR(npage[i]);
638 dn->inode_page_locked = false;
641 f2fs_put_page(parent, 1);
645 npage[i] = get_node_page(sbi, nids[i]);
646 if (IS_ERR(npage[i])) {
647 err = PTR_ERR(npage[i]);
648 f2fs_put_page(npage[0], 0);
654 nids[i + 1] = get_nid(parent, offset[i], false);
657 dn->nid = nids[level];
658 dn->ofs_in_node = offset[level];
659 dn->node_page = npage[level];
660 dn->data_blkaddr = datablock_addr(dn->inode,
661 dn->node_page, dn->ofs_in_node);
665 f2fs_put_page(parent, 1);
667 f2fs_put_page(npage[0], 0);
669 dn->inode_page = NULL;
670 dn->node_page = NULL;
671 if (err == -ENOENT) {
673 dn->max_level = level;
674 dn->ofs_in_node = offset[level];
679 static void truncate_node(struct dnode_of_data *dn)
681 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
684 get_node_info(sbi, dn->nid, &ni);
685 f2fs_bug_on(sbi, ni.blk_addr == NULL_ADDR);
687 /* Deallocate node address */
688 invalidate_blocks(sbi, ni.blk_addr);
689 dec_valid_node_count(sbi, dn->inode, dn->nid == dn->inode->i_ino);
690 set_node_addr(sbi, &ni, NULL_ADDR, false);
692 if (dn->nid == dn->inode->i_ino) {
693 remove_orphan_inode(sbi, dn->nid);
694 dec_valid_inode_count(sbi);
695 f2fs_inode_synced(dn->inode);
698 clear_node_page_dirty(dn->node_page);
699 set_sbi_flag(sbi, SBI_IS_DIRTY);
701 f2fs_put_page(dn->node_page, 1);
703 invalidate_mapping_pages(NODE_MAPPING(sbi),
704 dn->node_page->index, dn->node_page->index);
706 dn->node_page = NULL;
707 trace_f2fs_truncate_node(dn->inode, dn->nid, ni.blk_addr);
710 static int truncate_dnode(struct dnode_of_data *dn)
717 /* get direct node */
718 page = get_node_page(F2FS_I_SB(dn->inode), dn->nid);
719 if (IS_ERR(page) && PTR_ERR(page) == -ENOENT)
721 else if (IS_ERR(page))
722 return PTR_ERR(page);
724 /* Make dnode_of_data for parameter */
725 dn->node_page = page;
727 truncate_data_blocks(dn);
732 static int truncate_nodes(struct dnode_of_data *dn, unsigned int nofs,
735 struct dnode_of_data rdn = *dn;
737 struct f2fs_node *rn;
739 unsigned int child_nofs;
744 return NIDS_PER_BLOCK + 1;
746 trace_f2fs_truncate_nodes_enter(dn->inode, dn->nid, dn->data_blkaddr);
748 page = get_node_page(F2FS_I_SB(dn->inode), dn->nid);
750 trace_f2fs_truncate_nodes_exit(dn->inode, PTR_ERR(page));
751 return PTR_ERR(page);
754 ra_node_pages(page, ofs, NIDS_PER_BLOCK);
756 rn = F2FS_NODE(page);
758 for (i = ofs; i < NIDS_PER_BLOCK; i++, freed++) {
759 child_nid = le32_to_cpu(rn->in.nid[i]);
763 ret = truncate_dnode(&rdn);
766 if (set_nid(page, i, 0, false))
767 dn->node_changed = true;
770 child_nofs = nofs + ofs * (NIDS_PER_BLOCK + 1) + 1;
771 for (i = ofs; i < NIDS_PER_BLOCK; i++) {
772 child_nid = le32_to_cpu(rn->in.nid[i]);
773 if (child_nid == 0) {
774 child_nofs += NIDS_PER_BLOCK + 1;
778 ret = truncate_nodes(&rdn, child_nofs, 0, depth - 1);
779 if (ret == (NIDS_PER_BLOCK + 1)) {
780 if (set_nid(page, i, 0, false))
781 dn->node_changed = true;
783 } else if (ret < 0 && ret != -ENOENT) {
791 /* remove current indirect node */
792 dn->node_page = page;
796 f2fs_put_page(page, 1);
798 trace_f2fs_truncate_nodes_exit(dn->inode, freed);
802 f2fs_put_page(page, 1);
803 trace_f2fs_truncate_nodes_exit(dn->inode, ret);
807 static int truncate_partial_nodes(struct dnode_of_data *dn,
808 struct f2fs_inode *ri, int *offset, int depth)
810 struct page *pages[2];
817 nid[0] = le32_to_cpu(ri->i_nid[offset[0] - NODE_DIR1_BLOCK]);
821 /* get indirect nodes in the path */
822 for (i = 0; i < idx + 1; i++) {
823 /* reference count'll be increased */
824 pages[i] = get_node_page(F2FS_I_SB(dn->inode), nid[i]);
825 if (IS_ERR(pages[i])) {
826 err = PTR_ERR(pages[i]);
830 nid[i + 1] = get_nid(pages[i], offset[i + 1], false);
833 ra_node_pages(pages[idx], offset[idx + 1], NIDS_PER_BLOCK);
835 /* free direct nodes linked to a partial indirect node */
836 for (i = offset[idx + 1]; i < NIDS_PER_BLOCK; i++) {
837 child_nid = get_nid(pages[idx], i, false);
841 err = truncate_dnode(dn);
844 if (set_nid(pages[idx], i, 0, false))
845 dn->node_changed = true;
848 if (offset[idx + 1] == 0) {
849 dn->node_page = pages[idx];
853 f2fs_put_page(pages[idx], 1);
859 for (i = idx; i >= 0; i--)
860 f2fs_put_page(pages[i], 1);
862 trace_f2fs_truncate_partial_nodes(dn->inode, nid, depth, err);
868 * All the block addresses of data and nodes should be nullified.
870 int truncate_inode_blocks(struct inode *inode, pgoff_t from)
872 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
873 int err = 0, cont = 1;
874 int level, offset[4], noffset[4];
875 unsigned int nofs = 0;
876 struct f2fs_inode *ri;
877 struct dnode_of_data dn;
880 trace_f2fs_truncate_inode_blocks_enter(inode, from);
882 level = get_node_path(inode, from, offset, noffset);
886 page = get_node_page(sbi, inode->i_ino);
888 trace_f2fs_truncate_inode_blocks_exit(inode, PTR_ERR(page));
889 return PTR_ERR(page);
892 set_new_dnode(&dn, inode, page, NULL, 0);
895 ri = F2FS_INODE(page);
903 if (!offset[level - 1])
905 err = truncate_partial_nodes(&dn, ri, offset, level);
906 if (err < 0 && err != -ENOENT)
908 nofs += 1 + NIDS_PER_BLOCK;
911 nofs = 5 + 2 * NIDS_PER_BLOCK;
912 if (!offset[level - 1])
914 err = truncate_partial_nodes(&dn, ri, offset, level);
915 if (err < 0 && err != -ENOENT)
924 dn.nid = le32_to_cpu(ri->i_nid[offset[0] - NODE_DIR1_BLOCK]);
926 case NODE_DIR1_BLOCK:
927 case NODE_DIR2_BLOCK:
928 err = truncate_dnode(&dn);
931 case NODE_IND1_BLOCK:
932 case NODE_IND2_BLOCK:
933 err = truncate_nodes(&dn, nofs, offset[1], 2);
936 case NODE_DIND_BLOCK:
937 err = truncate_nodes(&dn, nofs, offset[1], 3);
944 if (err < 0 && err != -ENOENT)
946 if (offset[1] == 0 &&
947 ri->i_nid[offset[0] - NODE_DIR1_BLOCK]) {
949 BUG_ON(page->mapping != NODE_MAPPING(sbi));
950 f2fs_wait_on_page_writeback(page, NODE, true);
951 ri->i_nid[offset[0] - NODE_DIR1_BLOCK] = 0;
952 set_page_dirty(page);
960 f2fs_put_page(page, 0);
961 trace_f2fs_truncate_inode_blocks_exit(inode, err);
962 return err > 0 ? 0 : err;
965 int truncate_xattr_node(struct inode *inode, struct page *page)
967 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
968 nid_t nid = F2FS_I(inode)->i_xattr_nid;
969 struct dnode_of_data dn;
975 npage = get_node_page(sbi, nid);
977 return PTR_ERR(npage);
979 f2fs_i_xnid_write(inode, 0);
981 set_new_dnode(&dn, inode, page, npage, nid);
984 dn.inode_page_locked = true;
990 * Caller should grab and release a rwsem by calling f2fs_lock_op() and
993 int remove_inode_page(struct inode *inode)
995 struct dnode_of_data dn;
998 set_new_dnode(&dn, inode, NULL, NULL, inode->i_ino);
999 err = get_dnode_of_data(&dn, 0, LOOKUP_NODE);
1003 err = truncate_xattr_node(inode, dn.inode_page);
1005 f2fs_put_dnode(&dn);
1009 /* remove potential inline_data blocks */
1010 if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
1011 S_ISLNK(inode->i_mode))
1012 truncate_data_blocks_range(&dn, 1);
1014 /* 0 is possible, after f2fs_new_inode() has failed */
1015 f2fs_bug_on(F2FS_I_SB(inode),
1016 inode->i_blocks != 0 && inode->i_blocks != 8);
1018 /* will put inode & node pages */
1023 struct page *new_inode_page(struct inode *inode)
1025 struct dnode_of_data dn;
1027 /* allocate inode page for new inode */
1028 set_new_dnode(&dn, inode, NULL, NULL, inode->i_ino);
1030 /* caller should f2fs_put_page(page, 1); */
1031 return new_node_page(&dn, 0);
1034 struct page *new_node_page(struct dnode_of_data *dn, unsigned int ofs)
1036 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
1037 struct node_info new_ni;
1041 if (unlikely(is_inode_flag_set(dn->inode, FI_NO_ALLOC)))
1042 return ERR_PTR(-EPERM);
1044 page = f2fs_grab_cache_page(NODE_MAPPING(sbi), dn->nid, false);
1046 return ERR_PTR(-ENOMEM);
1048 if (unlikely((err = inc_valid_node_count(sbi, dn->inode, !ofs))))
1051 #ifdef CONFIG_F2FS_CHECK_FS
1052 get_node_info(sbi, dn->nid, &new_ni);
1053 f2fs_bug_on(sbi, new_ni.blk_addr != NULL_ADDR);
1055 new_ni.nid = dn->nid;
1056 new_ni.ino = dn->inode->i_ino;
1057 new_ni.blk_addr = NULL_ADDR;
1060 set_node_addr(sbi, &new_ni, NEW_ADDR, false);
1062 f2fs_wait_on_page_writeback(page, NODE, true);
1063 fill_node_footer(page, dn->nid, dn->inode->i_ino, ofs, true);
1064 set_cold_node(dn->inode, page);
1065 if (!PageUptodate(page))
1066 SetPageUptodate(page);
1067 if (set_page_dirty(page))
1068 dn->node_changed = true;
1070 if (f2fs_has_xattr_block(ofs))
1071 f2fs_i_xnid_write(dn->inode, dn->nid);
1074 inc_valid_inode_count(sbi);
1078 clear_node_page_dirty(page);
1079 f2fs_put_page(page, 1);
1080 return ERR_PTR(err);
1084 * Caller should do after getting the following values.
1085 * 0: f2fs_put_page(page, 0)
1086 * LOCKED_PAGE or error: f2fs_put_page(page, 1)
1088 static int read_node_page(struct page *page, int op_flags)
1090 struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1091 struct node_info ni;
1092 struct f2fs_io_info fio = {
1096 .op_flags = op_flags,
1098 .encrypted_page = NULL,
1101 if (PageUptodate(page))
1104 get_node_info(sbi, page->index, &ni);
1106 if (unlikely(ni.blk_addr == NULL_ADDR)) {
1107 ClearPageUptodate(page);
1111 fio.new_blkaddr = fio.old_blkaddr = ni.blk_addr;
1112 return f2fs_submit_page_bio(&fio);
1116 * Readahead a node page
1118 void ra_node_page(struct f2fs_sb_info *sbi, nid_t nid)
1125 f2fs_bug_on(sbi, check_nid_range(sbi, nid));
1128 apage = radix_tree_lookup(&NODE_MAPPING(sbi)->page_tree, nid);
1133 apage = f2fs_grab_cache_page(NODE_MAPPING(sbi), nid, false);
1137 err = read_node_page(apage, REQ_RAHEAD);
1138 f2fs_put_page(apage, err ? 1 : 0);
1141 static struct page *__get_node_page(struct f2fs_sb_info *sbi, pgoff_t nid,
1142 struct page *parent, int start)
1148 return ERR_PTR(-ENOENT);
1149 f2fs_bug_on(sbi, check_nid_range(sbi, nid));
1151 page = f2fs_grab_cache_page(NODE_MAPPING(sbi), nid, false);
1153 return ERR_PTR(-ENOMEM);
1155 err = read_node_page(page, 0);
1157 f2fs_put_page(page, 1);
1158 return ERR_PTR(err);
1159 } else if (err == LOCKED_PAGE) {
1165 ra_node_pages(parent, start + 1, MAX_RA_NODE);
1169 if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1170 f2fs_put_page(page, 1);
1174 if (unlikely(!PageUptodate(page))) {
1179 if (!f2fs_inode_chksum_verify(sbi, page)) {
1184 if(unlikely(nid != nid_of_node(page))) {
1185 f2fs_msg(sbi->sb, KERN_WARNING, "inconsistent node block, "
1186 "nid:%lu, node_footer[nid:%u,ino:%u,ofs:%u,cpver:%llu,blkaddr:%u]",
1187 nid, nid_of_node(page), ino_of_node(page),
1188 ofs_of_node(page), cpver_of_node(page),
1189 next_blkaddr_of_node(page));
1192 ClearPageUptodate(page);
1193 f2fs_put_page(page, 1);
1194 return ERR_PTR(err);
1199 struct page *get_node_page(struct f2fs_sb_info *sbi, pgoff_t nid)
1201 return __get_node_page(sbi, nid, NULL, 0);
1204 struct page *get_node_page_ra(struct page *parent, int start)
1206 struct f2fs_sb_info *sbi = F2FS_P_SB(parent);
1207 nid_t nid = get_nid(parent, start, false);
1209 return __get_node_page(sbi, nid, parent, start);
1212 static void flush_inline_data(struct f2fs_sb_info *sbi, nid_t ino)
1214 struct inode *inode;
1218 /* should flush inline_data before evict_inode */
1219 inode = ilookup(sbi->sb, ino);
1223 page = pagecache_get_page(inode->i_mapping, 0, FGP_LOCK|FGP_NOWAIT, 0);
1227 if (!PageUptodate(page))
1230 if (!PageDirty(page))
1233 if (!clear_page_dirty_for_io(page))
1236 ret = f2fs_write_inline_data(inode, page);
1237 inode_dec_dirty_pages(inode);
1238 remove_dirty_inode(inode);
1240 set_page_dirty(page);
1242 f2fs_put_page(page, 1);
1247 void move_node_page(struct page *node_page, int gc_type)
1249 if (gc_type == FG_GC) {
1250 struct f2fs_sb_info *sbi = F2FS_P_SB(node_page);
1251 struct writeback_control wbc = {
1252 .sync_mode = WB_SYNC_ALL,
1257 set_page_dirty(node_page);
1258 f2fs_wait_on_page_writeback(node_page, NODE, true);
1260 f2fs_bug_on(sbi, PageWriteback(node_page));
1261 if (!clear_page_dirty_for_io(node_page))
1264 if (NODE_MAPPING(sbi)->a_ops->writepage(node_page, &wbc))
1265 unlock_page(node_page);
1268 /* set page dirty and write it */
1269 if (!PageWriteback(node_page))
1270 set_page_dirty(node_page);
1273 unlock_page(node_page);
1275 f2fs_put_page(node_page, 0);
1278 static struct page *last_fsync_dnode(struct f2fs_sb_info *sbi, nid_t ino)
1281 struct pagevec pvec;
1282 struct page *last_page = NULL;
1285 pagevec_init(&pvec, 0);
1288 while ((nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1289 PAGECACHE_TAG_DIRTY))) {
1292 for (i = 0; i < nr_pages; i++) {
1293 struct page *page = pvec.pages[i];
1295 if (unlikely(f2fs_cp_error(sbi))) {
1296 f2fs_put_page(last_page, 0);
1297 pagevec_release(&pvec);
1298 return ERR_PTR(-EIO);
1301 if (!IS_DNODE(page) || !is_cold_node(page))
1303 if (ino_of_node(page) != ino)
1308 if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1313 if (ino_of_node(page) != ino)
1314 goto continue_unlock;
1316 if (!PageDirty(page)) {
1317 /* someone wrote it for us */
1318 goto continue_unlock;
1322 f2fs_put_page(last_page, 0);
1328 pagevec_release(&pvec);
1334 static int __write_node_page(struct page *page, bool atomic, bool *submitted,
1335 struct writeback_control *wbc, bool do_balance,
1336 enum iostat_type io_type)
1338 struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1340 struct node_info ni;
1341 struct f2fs_io_info fio = {
1345 .op_flags = wbc_to_write_flags(wbc),
1347 .encrypted_page = NULL,
1352 trace_f2fs_writepage(page, NODE);
1354 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
1356 if (unlikely(f2fs_cp_error(sbi)))
1359 /* get old block addr of this node page */
1360 nid = nid_of_node(page);
1361 f2fs_bug_on(sbi, page->index != nid);
1363 if (wbc->for_reclaim) {
1364 if (!down_read_trylock(&sbi->node_write))
1367 down_read(&sbi->node_write);
1370 get_node_info(sbi, nid, &ni);
1372 /* This page is already truncated */
1373 if (unlikely(ni.blk_addr == NULL_ADDR)) {
1374 ClearPageUptodate(page);
1375 dec_page_count(sbi, F2FS_DIRTY_NODES);
1376 up_read(&sbi->node_write);
1381 if (atomic && !test_opt(sbi, NOBARRIER))
1382 fio.op_flags |= REQ_PREFLUSH | REQ_FUA;
1384 set_page_writeback(page);
1385 fio.old_blkaddr = ni.blk_addr;
1386 write_node_page(nid, &fio);
1387 set_node_addr(sbi, &ni, fio.new_blkaddr, is_fsync_dnode(page));
1388 dec_page_count(sbi, F2FS_DIRTY_NODES);
1389 up_read(&sbi->node_write);
1391 if (wbc->for_reclaim) {
1392 f2fs_submit_merged_write_cond(sbi, page->mapping->host, 0,
1399 if (unlikely(f2fs_cp_error(sbi))) {
1400 f2fs_submit_merged_write(sbi, NODE);
1404 *submitted = fio.submitted;
1407 f2fs_balance_fs(sbi, false);
1411 redirty_page_for_writepage(wbc, page);
1412 return AOP_WRITEPAGE_ACTIVATE;
1415 static int f2fs_write_node_page(struct page *page,
1416 struct writeback_control *wbc)
1418 return __write_node_page(page, false, NULL, wbc, false, FS_NODE_IO);
1421 int fsync_node_pages(struct f2fs_sb_info *sbi, struct inode *inode,
1422 struct writeback_control *wbc, bool atomic)
1425 pgoff_t last_idx = ULONG_MAX;
1426 struct pagevec pvec;
1428 struct page *last_page = NULL;
1429 bool marked = false;
1430 nid_t ino = inode->i_ino;
1434 last_page = last_fsync_dnode(sbi, ino);
1435 if (IS_ERR_OR_NULL(last_page))
1436 return PTR_ERR_OR_ZERO(last_page);
1439 pagevec_init(&pvec, 0);
1442 while ((nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1443 PAGECACHE_TAG_DIRTY))) {
1446 for (i = 0; i < nr_pages; i++) {
1447 struct page *page = pvec.pages[i];
1448 bool submitted = false;
1450 if (unlikely(f2fs_cp_error(sbi))) {
1451 f2fs_put_page(last_page, 0);
1452 pagevec_release(&pvec);
1457 if (!IS_DNODE(page) || !is_cold_node(page))
1459 if (ino_of_node(page) != ino)
1464 if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1469 if (ino_of_node(page) != ino)
1470 goto continue_unlock;
1472 if (!PageDirty(page) && page != last_page) {
1473 /* someone wrote it for us */
1474 goto continue_unlock;
1477 f2fs_wait_on_page_writeback(page, NODE, true);
1478 BUG_ON(PageWriteback(page));
1480 set_fsync_mark(page, 0);
1481 set_dentry_mark(page, 0);
1483 if (!atomic || page == last_page) {
1484 set_fsync_mark(page, 1);
1485 if (IS_INODE(page)) {
1486 if (is_inode_flag_set(inode,
1488 update_inode(inode, page);
1489 set_dentry_mark(page,
1490 need_dentry_mark(sbi, ino));
1492 /* may be written by other thread */
1493 if (!PageDirty(page))
1494 set_page_dirty(page);
1497 if (!clear_page_dirty_for_io(page))
1498 goto continue_unlock;
1500 ret = __write_node_page(page, atomic &&
1502 &submitted, wbc, true,
1506 f2fs_put_page(last_page, 0);
1508 } else if (submitted) {
1509 last_idx = page->index;
1512 if (page == last_page) {
1513 f2fs_put_page(page, 0);
1518 pagevec_release(&pvec);
1524 if (!ret && atomic && !marked) {
1525 f2fs_msg(sbi->sb, KERN_DEBUG,
1526 "Retry to write fsync mark: ino=%u, idx=%lx",
1527 ino, last_page->index);
1528 lock_page(last_page);
1529 f2fs_wait_on_page_writeback(last_page, NODE, true);
1530 set_page_dirty(last_page);
1531 unlock_page(last_page);
1535 if (last_idx != ULONG_MAX)
1536 f2fs_submit_merged_write_cond(sbi, NULL, ino, last_idx, NODE);
1537 return ret ? -EIO: 0;
1540 int sync_node_pages(struct f2fs_sb_info *sbi, struct writeback_control *wbc,
1541 bool do_balance, enum iostat_type io_type)
1544 struct pagevec pvec;
1550 pagevec_init(&pvec, 0);
1555 while ((nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1556 PAGECACHE_TAG_DIRTY))) {
1559 for (i = 0; i < nr_pages; i++) {
1560 struct page *page = pvec.pages[i];
1561 bool submitted = false;
1563 if (unlikely(f2fs_cp_error(sbi))) {
1564 pagevec_release(&pvec);
1570 * flushing sequence with step:
1575 if (step == 0 && IS_DNODE(page))
1577 if (step == 1 && (!IS_DNODE(page) ||
1578 is_cold_node(page)))
1580 if (step == 2 && (!IS_DNODE(page) ||
1581 !is_cold_node(page)))
1584 if (!trylock_page(page))
1587 if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1593 if (!PageDirty(page)) {
1594 /* someone wrote it for us */
1595 goto continue_unlock;
1598 /* flush inline_data */
1599 if (is_inline_node(page)) {
1600 clear_inline_node(page);
1602 flush_inline_data(sbi, ino_of_node(page));
1606 f2fs_wait_on_page_writeback(page, NODE, true);
1608 BUG_ON(PageWriteback(page));
1609 if (!clear_page_dirty_for_io(page))
1610 goto continue_unlock;
1612 set_fsync_mark(page, 0);
1613 set_dentry_mark(page, 0);
1615 ret = __write_node_page(page, false, &submitted,
1616 wbc, do_balance, io_type);
1622 if (--wbc->nr_to_write == 0)
1625 pagevec_release(&pvec);
1628 if (wbc->nr_to_write == 0) {
1640 f2fs_submit_merged_write(sbi, NODE);
1644 int wait_on_node_pages_writeback(struct f2fs_sb_info *sbi, nid_t ino)
1647 struct pagevec pvec;
1651 pagevec_init(&pvec, 0);
1653 while ((nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1654 PAGECACHE_TAG_WRITEBACK))) {
1657 for (i = 0; i < nr_pages; i++) {
1658 struct page *page = pvec.pages[i];
1660 if (ino && ino_of_node(page) == ino) {
1661 f2fs_wait_on_page_writeback(page, NODE, true);
1662 if (TestClearPageError(page))
1666 pagevec_release(&pvec);
1670 ret2 = filemap_check_errors(NODE_MAPPING(sbi));
1676 static int f2fs_write_node_pages(struct address_space *mapping,
1677 struct writeback_control *wbc)
1679 struct f2fs_sb_info *sbi = F2FS_M_SB(mapping);
1680 struct blk_plug plug;
1683 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
1686 /* balancing f2fs's metadata in background */
1687 f2fs_balance_fs_bg(sbi);
1689 /* collect a number of dirty node pages and write together */
1690 if (get_pages(sbi, F2FS_DIRTY_NODES) < nr_pages_to_skip(sbi, NODE))
1693 trace_f2fs_writepages(mapping->host, wbc, NODE);
1695 diff = nr_pages_to_write(sbi, NODE, wbc);
1696 wbc->sync_mode = WB_SYNC_NONE;
1697 blk_start_plug(&plug);
1698 sync_node_pages(sbi, wbc, true, FS_NODE_IO);
1699 blk_finish_plug(&plug);
1700 wbc->nr_to_write = max((long)0, wbc->nr_to_write - diff);
1704 wbc->pages_skipped += get_pages(sbi, F2FS_DIRTY_NODES);
1705 trace_f2fs_writepages(mapping->host, wbc, NODE);
1709 static int f2fs_set_node_page_dirty(struct page *page)
1711 trace_f2fs_set_page_dirty(page, NODE);
1713 if (!PageUptodate(page))
1714 SetPageUptodate(page);
1715 if (!PageDirty(page)) {
1716 f2fs_set_page_dirty_nobuffers(page);
1717 inc_page_count(F2FS_P_SB(page), F2FS_DIRTY_NODES);
1718 SetPagePrivate(page);
1719 f2fs_trace_pid(page);
1726 * Structure of the f2fs node operations
1728 const struct address_space_operations f2fs_node_aops = {
1729 .writepage = f2fs_write_node_page,
1730 .writepages = f2fs_write_node_pages,
1731 .set_page_dirty = f2fs_set_node_page_dirty,
1732 .invalidatepage = f2fs_invalidate_page,
1733 .releasepage = f2fs_release_page,
1734 #ifdef CONFIG_MIGRATION
1735 .migratepage = f2fs_migrate_page,
1739 static struct free_nid *__lookup_free_nid_list(struct f2fs_nm_info *nm_i,
1742 return radix_tree_lookup(&nm_i->free_nid_root, n);
1745 static int __insert_nid_to_list(struct f2fs_sb_info *sbi,
1746 struct free_nid *i, enum nid_list list, bool new)
1748 struct f2fs_nm_info *nm_i = NM_I(sbi);
1751 int err = radix_tree_insert(&nm_i->free_nid_root, i->nid, i);
1756 f2fs_bug_on(sbi, list == FREE_NID_LIST ? i->state != NID_NEW :
1757 i->state != NID_ALLOC);
1758 nm_i->nid_cnt[list]++;
1759 list_add_tail(&i->list, &nm_i->nid_list[list]);
1763 static void __remove_nid_from_list(struct f2fs_sb_info *sbi,
1764 struct free_nid *i, enum nid_list list, bool reuse)
1766 struct f2fs_nm_info *nm_i = NM_I(sbi);
1768 f2fs_bug_on(sbi, list == FREE_NID_LIST ? i->state != NID_NEW :
1769 i->state != NID_ALLOC);
1770 nm_i->nid_cnt[list]--;
1773 radix_tree_delete(&nm_i->free_nid_root, i->nid);
1776 /* return if the nid is recognized as free */
1777 static bool add_free_nid(struct f2fs_sb_info *sbi, nid_t nid, bool build)
1779 struct f2fs_nm_info *nm_i = NM_I(sbi);
1780 struct free_nid *i, *e;
1781 struct nat_entry *ne;
1785 /* 0 nid should not be used */
1786 if (unlikely(nid == 0))
1789 i = f2fs_kmem_cache_alloc(free_nid_slab, GFP_NOFS);
1793 if (radix_tree_preload(GFP_NOFS))
1796 spin_lock(&nm_i->nid_list_lock);
1804 * - __insert_nid_to_list(ALLOC_NID_LIST)
1805 * - f2fs_balance_fs_bg
1807 * - __build_free_nids
1810 * - __lookup_nat_cache
1812 * - init_inode_metadata
1817 * - __remove_nid_from_list(ALLOC_NID_LIST)
1818 * - __insert_nid_to_list(FREE_NID_LIST)
1820 ne = __lookup_nat_cache(nm_i, nid);
1821 if (ne && (!get_nat_flag(ne, IS_CHECKPOINTED) ||
1822 nat_get_blkaddr(ne) != NULL_ADDR))
1825 e = __lookup_free_nid_list(nm_i, nid);
1827 if (e->state == NID_NEW)
1833 err = __insert_nid_to_list(sbi, i, FREE_NID_LIST, true);
1835 spin_unlock(&nm_i->nid_list_lock);
1836 radix_tree_preload_end();
1839 kmem_cache_free(free_nid_slab, i);
1843 static void remove_free_nid(struct f2fs_sb_info *sbi, nid_t nid)
1845 struct f2fs_nm_info *nm_i = NM_I(sbi);
1847 bool need_free = false;
1849 spin_lock(&nm_i->nid_list_lock);
1850 i = __lookup_free_nid_list(nm_i, nid);
1851 if (i && i->state == NID_NEW) {
1852 __remove_nid_from_list(sbi, i, FREE_NID_LIST, false);
1855 spin_unlock(&nm_i->nid_list_lock);
1858 kmem_cache_free(free_nid_slab, i);
1861 static void update_free_nid_bitmap(struct f2fs_sb_info *sbi, nid_t nid,
1862 bool set, bool build)
1864 struct f2fs_nm_info *nm_i = NM_I(sbi);
1865 unsigned int nat_ofs = NAT_BLOCK_OFFSET(nid);
1866 unsigned int nid_ofs = nid - START_NID(nid);
1868 if (!test_bit_le(nat_ofs, nm_i->nat_block_bitmap))
1872 __set_bit_le(nid_ofs, nm_i->free_nid_bitmap[nat_ofs]);
1874 __clear_bit_le(nid_ofs, nm_i->free_nid_bitmap[nat_ofs]);
1877 nm_i->free_nid_count[nat_ofs]++;
1879 nm_i->free_nid_count[nat_ofs]--;
1882 static void scan_nat_page(struct f2fs_sb_info *sbi,
1883 struct page *nat_page, nid_t start_nid)
1885 struct f2fs_nm_info *nm_i = NM_I(sbi);
1886 struct f2fs_nat_block *nat_blk = page_address(nat_page);
1888 unsigned int nat_ofs = NAT_BLOCK_OFFSET(start_nid);
1891 if (test_bit_le(nat_ofs, nm_i->nat_block_bitmap))
1894 __set_bit_le(nat_ofs, nm_i->nat_block_bitmap);
1896 i = start_nid % NAT_ENTRY_PER_BLOCK;
1898 for (; i < NAT_ENTRY_PER_BLOCK; i++, start_nid++) {
1901 if (unlikely(start_nid >= nm_i->max_nid))
1904 blk_addr = le32_to_cpu(nat_blk->entries[i].block_addr);
1905 f2fs_bug_on(sbi, blk_addr == NEW_ADDR);
1906 if (blk_addr == NULL_ADDR)
1907 freed = add_free_nid(sbi, start_nid, true);
1908 spin_lock(&NM_I(sbi)->nid_list_lock);
1909 update_free_nid_bitmap(sbi, start_nid, freed, true);
1910 spin_unlock(&NM_I(sbi)->nid_list_lock);
1914 static void scan_free_nid_bits(struct f2fs_sb_info *sbi)
1916 struct f2fs_nm_info *nm_i = NM_I(sbi);
1917 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
1918 struct f2fs_journal *journal = curseg->journal;
1919 unsigned int i, idx;
1921 down_read(&nm_i->nat_tree_lock);
1923 for (i = 0; i < nm_i->nat_blocks; i++) {
1924 if (!test_bit_le(i, nm_i->nat_block_bitmap))
1926 if (!nm_i->free_nid_count[i])
1928 for (idx = 0; idx < NAT_ENTRY_PER_BLOCK; idx++) {
1931 if (!test_bit_le(idx, nm_i->free_nid_bitmap[i]))
1934 nid = i * NAT_ENTRY_PER_BLOCK + idx;
1935 add_free_nid(sbi, nid, true);
1937 if (nm_i->nid_cnt[FREE_NID_LIST] >= MAX_FREE_NIDS)
1942 down_read(&curseg->journal_rwsem);
1943 for (i = 0; i < nats_in_cursum(journal); i++) {
1947 addr = le32_to_cpu(nat_in_journal(journal, i).block_addr);
1948 nid = le32_to_cpu(nid_in_journal(journal, i));
1949 if (addr == NULL_ADDR)
1950 add_free_nid(sbi, nid, true);
1952 remove_free_nid(sbi, nid);
1954 up_read(&curseg->journal_rwsem);
1955 up_read(&nm_i->nat_tree_lock);
1958 static void __build_free_nids(struct f2fs_sb_info *sbi, bool sync, bool mount)
1960 struct f2fs_nm_info *nm_i = NM_I(sbi);
1961 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
1962 struct f2fs_journal *journal = curseg->journal;
1964 nid_t nid = nm_i->next_scan_nid;
1966 if (unlikely(nid >= nm_i->max_nid))
1969 /* Enough entries */
1970 if (nm_i->nid_cnt[FREE_NID_LIST] >= NAT_ENTRY_PER_BLOCK)
1973 if (!sync && !available_free_memory(sbi, FREE_NIDS))
1977 /* try to find free nids in free_nid_bitmap */
1978 scan_free_nid_bits(sbi);
1980 if (nm_i->nid_cnt[FREE_NID_LIST])
1984 /* readahead nat pages to be scanned */
1985 ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nid), FREE_NID_PAGES,
1988 down_read(&nm_i->nat_tree_lock);
1991 struct page *page = get_current_nat_page(sbi, nid);
1993 scan_nat_page(sbi, page, nid);
1994 f2fs_put_page(page, 1);
1996 nid += (NAT_ENTRY_PER_BLOCK - (nid % NAT_ENTRY_PER_BLOCK));
1997 if (unlikely(nid >= nm_i->max_nid))
2000 if (++i >= FREE_NID_PAGES)
2004 /* go to the next free nat pages to find free nids abundantly */
2005 nm_i->next_scan_nid = nid;
2007 /* find free nids from current sum_pages */
2008 down_read(&curseg->journal_rwsem);
2009 for (i = 0; i < nats_in_cursum(journal); i++) {
2012 addr = le32_to_cpu(nat_in_journal(journal, i).block_addr);
2013 nid = le32_to_cpu(nid_in_journal(journal, i));
2014 if (addr == NULL_ADDR)
2015 add_free_nid(sbi, nid, true);
2017 remove_free_nid(sbi, nid);
2019 up_read(&curseg->journal_rwsem);
2020 up_read(&nm_i->nat_tree_lock);
2022 ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nm_i->next_scan_nid),
2023 nm_i->ra_nid_pages, META_NAT, false);
2026 void build_free_nids(struct f2fs_sb_info *sbi, bool sync, bool mount)
2028 mutex_lock(&NM_I(sbi)->build_lock);
2029 __build_free_nids(sbi, sync, mount);
2030 mutex_unlock(&NM_I(sbi)->build_lock);
2034 * If this function returns success, caller can obtain a new nid
2035 * from second parameter of this function.
2036 * The returned nid could be used ino as well as nid when inode is created.
2038 bool alloc_nid(struct f2fs_sb_info *sbi, nid_t *nid)
2040 struct f2fs_nm_info *nm_i = NM_I(sbi);
2041 struct free_nid *i = NULL;
2043 #ifdef CONFIG_F2FS_FAULT_INJECTION
2044 if (time_to_inject(sbi, FAULT_ALLOC_NID)) {
2045 f2fs_show_injection_info(FAULT_ALLOC_NID);
2049 spin_lock(&nm_i->nid_list_lock);
2051 if (unlikely(nm_i->available_nids == 0)) {
2052 spin_unlock(&nm_i->nid_list_lock);
2056 /* We should not use stale free nids created by build_free_nids */
2057 if (nm_i->nid_cnt[FREE_NID_LIST] && !on_build_free_nids(nm_i)) {
2058 f2fs_bug_on(sbi, list_empty(&nm_i->nid_list[FREE_NID_LIST]));
2059 i = list_first_entry(&nm_i->nid_list[FREE_NID_LIST],
2060 struct free_nid, list);
2063 __remove_nid_from_list(sbi, i, FREE_NID_LIST, true);
2064 i->state = NID_ALLOC;
2065 __insert_nid_to_list(sbi, i, ALLOC_NID_LIST, false);
2066 nm_i->available_nids--;
2068 update_free_nid_bitmap(sbi, *nid, false, false);
2070 spin_unlock(&nm_i->nid_list_lock);
2073 spin_unlock(&nm_i->nid_list_lock);
2075 /* Let's scan nat pages and its caches to get free nids */
2076 build_free_nids(sbi, true, false);
2081 * alloc_nid() should be called prior to this function.
2083 void alloc_nid_done(struct f2fs_sb_info *sbi, nid_t nid)
2085 struct f2fs_nm_info *nm_i = NM_I(sbi);
2088 spin_lock(&nm_i->nid_list_lock);
2089 i = __lookup_free_nid_list(nm_i, nid);
2090 f2fs_bug_on(sbi, !i);
2091 __remove_nid_from_list(sbi, i, ALLOC_NID_LIST, false);
2092 spin_unlock(&nm_i->nid_list_lock);
2094 kmem_cache_free(free_nid_slab, i);
2098 * alloc_nid() should be called prior to this function.
2100 void alloc_nid_failed(struct f2fs_sb_info *sbi, nid_t nid)
2102 struct f2fs_nm_info *nm_i = NM_I(sbi);
2104 bool need_free = false;
2109 spin_lock(&nm_i->nid_list_lock);
2110 i = __lookup_free_nid_list(nm_i, nid);
2111 f2fs_bug_on(sbi, !i);
2113 if (!available_free_memory(sbi, FREE_NIDS)) {
2114 __remove_nid_from_list(sbi, i, ALLOC_NID_LIST, false);
2117 __remove_nid_from_list(sbi, i, ALLOC_NID_LIST, true);
2119 __insert_nid_to_list(sbi, i, FREE_NID_LIST, false);
2122 nm_i->available_nids++;
2124 update_free_nid_bitmap(sbi, nid, true, false);
2126 spin_unlock(&nm_i->nid_list_lock);
2129 kmem_cache_free(free_nid_slab, i);
2132 int try_to_free_nids(struct f2fs_sb_info *sbi, int nr_shrink)
2134 struct f2fs_nm_info *nm_i = NM_I(sbi);
2135 struct free_nid *i, *next;
2138 if (nm_i->nid_cnt[FREE_NID_LIST] <= MAX_FREE_NIDS)
2141 if (!mutex_trylock(&nm_i->build_lock))
2144 spin_lock(&nm_i->nid_list_lock);
2145 list_for_each_entry_safe(i, next, &nm_i->nid_list[FREE_NID_LIST],
2147 if (nr_shrink <= 0 ||
2148 nm_i->nid_cnt[FREE_NID_LIST] <= MAX_FREE_NIDS)
2151 __remove_nid_from_list(sbi, i, FREE_NID_LIST, false);
2152 kmem_cache_free(free_nid_slab, i);
2155 spin_unlock(&nm_i->nid_list_lock);
2156 mutex_unlock(&nm_i->build_lock);
2158 return nr - nr_shrink;
2161 void recover_inline_xattr(struct inode *inode, struct page *page)
2163 void *src_addr, *dst_addr;
2166 struct f2fs_inode *ri;
2168 ipage = get_node_page(F2FS_I_SB(inode), inode->i_ino);
2169 f2fs_bug_on(F2FS_I_SB(inode), IS_ERR(ipage));
2171 ri = F2FS_INODE(page);
2172 if (!(ri->i_inline & F2FS_INLINE_XATTR)) {
2173 clear_inode_flag(inode, FI_INLINE_XATTR);
2177 dst_addr = inline_xattr_addr(ipage);
2178 src_addr = inline_xattr_addr(page);
2179 inline_size = inline_xattr_size(inode);
2181 f2fs_wait_on_page_writeback(ipage, NODE, true);
2182 memcpy(dst_addr, src_addr, inline_size);
2184 update_inode(inode, ipage);
2185 f2fs_put_page(ipage, 1);
2188 int recover_xattr_data(struct inode *inode, struct page *page, block_t blkaddr)
2190 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
2191 nid_t prev_xnid = F2FS_I(inode)->i_xattr_nid;
2193 struct dnode_of_data dn;
2194 struct node_info ni;
2200 /* 1: invalidate the previous xattr nid */
2201 get_node_info(sbi, prev_xnid, &ni);
2202 f2fs_bug_on(sbi, ni.blk_addr == NULL_ADDR);
2203 invalidate_blocks(sbi, ni.blk_addr);
2204 dec_valid_node_count(sbi, inode, false);
2205 set_node_addr(sbi, &ni, NULL_ADDR, false);
2208 /* 2: update xattr nid in inode */
2209 if (!alloc_nid(sbi, &new_xnid))
2212 set_new_dnode(&dn, inode, NULL, NULL, new_xnid);
2213 xpage = new_node_page(&dn, XATTR_NODE_OFFSET);
2214 if (IS_ERR(xpage)) {
2215 alloc_nid_failed(sbi, new_xnid);
2216 return PTR_ERR(xpage);
2219 alloc_nid_done(sbi, new_xnid);
2220 update_inode_page(inode);
2222 /* 3: update and set xattr node page dirty */
2223 memcpy(F2FS_NODE(xpage), F2FS_NODE(page), VALID_XATTR_BLOCK_SIZE);
2225 set_page_dirty(xpage);
2226 f2fs_put_page(xpage, 1);
2231 int recover_inode_page(struct f2fs_sb_info *sbi, struct page *page)
2233 struct f2fs_inode *src, *dst;
2234 nid_t ino = ino_of_node(page);
2235 struct node_info old_ni, new_ni;
2238 get_node_info(sbi, ino, &old_ni);
2240 if (unlikely(old_ni.blk_addr != NULL_ADDR))
2243 ipage = f2fs_grab_cache_page(NODE_MAPPING(sbi), ino, false);
2245 congestion_wait(BLK_RW_ASYNC, HZ/50);
2249 /* Should not use this inode from free nid list */
2250 remove_free_nid(sbi, ino);
2252 if (!PageUptodate(ipage))
2253 SetPageUptodate(ipage);
2254 fill_node_footer(ipage, ino, ino, 0, true);
2256 src = F2FS_INODE(page);
2257 dst = F2FS_INODE(ipage);
2259 memcpy(dst, src, (unsigned long)&src->i_ext - (unsigned long)src);
2261 dst->i_blocks = cpu_to_le64(1);
2262 dst->i_links = cpu_to_le32(1);
2263 dst->i_xattr_nid = 0;
2264 dst->i_inline = src->i_inline & (F2FS_INLINE_XATTR | F2FS_EXTRA_ATTR);
2265 if (dst->i_inline & F2FS_EXTRA_ATTR) {
2266 dst->i_extra_isize = src->i_extra_isize;
2267 if (f2fs_sb_has_project_quota(sbi->sb) &&
2268 F2FS_FITS_IN_INODE(src, le16_to_cpu(src->i_extra_isize),
2270 dst->i_projid = src->i_projid;
2276 if (unlikely(inc_valid_node_count(sbi, NULL, true)))
2278 set_node_addr(sbi, &new_ni, NEW_ADDR, false);
2279 inc_valid_inode_count(sbi);
2280 set_page_dirty(ipage);
2281 f2fs_put_page(ipage, 1);
2285 int restore_node_summary(struct f2fs_sb_info *sbi,
2286 unsigned int segno, struct f2fs_summary_block *sum)
2288 struct f2fs_node *rn;
2289 struct f2fs_summary *sum_entry;
2291 int i, idx, last_offset, nrpages;
2293 /* scan the node segment */
2294 last_offset = sbi->blocks_per_seg;
2295 addr = START_BLOCK(sbi, segno);
2296 sum_entry = &sum->entries[0];
2298 for (i = 0; i < last_offset; i += nrpages, addr += nrpages) {
2299 nrpages = min(last_offset - i, BIO_MAX_PAGES);
2301 /* readahead node pages */
2302 ra_meta_pages(sbi, addr, nrpages, META_POR, true);
2304 for (idx = addr; idx < addr + nrpages; idx++) {
2305 struct page *page = get_tmp_page(sbi, idx);
2307 rn = F2FS_NODE(page);
2308 sum_entry->nid = rn->footer.nid;
2309 sum_entry->version = 0;
2310 sum_entry->ofs_in_node = 0;
2312 f2fs_put_page(page, 1);
2315 invalidate_mapping_pages(META_MAPPING(sbi), addr,
2321 static void remove_nats_in_journal(struct f2fs_sb_info *sbi)
2323 struct f2fs_nm_info *nm_i = NM_I(sbi);
2324 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
2325 struct f2fs_journal *journal = curseg->journal;
2328 down_write(&curseg->journal_rwsem);
2329 for (i = 0; i < nats_in_cursum(journal); i++) {
2330 struct nat_entry *ne;
2331 struct f2fs_nat_entry raw_ne;
2332 nid_t nid = le32_to_cpu(nid_in_journal(journal, i));
2334 raw_ne = nat_in_journal(journal, i);
2336 ne = __lookup_nat_cache(nm_i, nid);
2338 ne = grab_nat_entry(nm_i, nid, true);
2339 node_info_from_raw_nat(&ne->ni, &raw_ne);
2343 * if a free nat in journal has not been used after last
2344 * checkpoint, we should remove it from available nids,
2345 * since later we will add it again.
2347 if (!get_nat_flag(ne, IS_DIRTY) &&
2348 le32_to_cpu(raw_ne.block_addr) == NULL_ADDR) {
2349 spin_lock(&nm_i->nid_list_lock);
2350 nm_i->available_nids--;
2351 spin_unlock(&nm_i->nid_list_lock);
2354 __set_nat_cache_dirty(nm_i, ne);
2356 update_nats_in_cursum(journal, -i);
2357 up_write(&curseg->journal_rwsem);
2360 static void __adjust_nat_entry_set(struct nat_entry_set *nes,
2361 struct list_head *head, int max)
2363 struct nat_entry_set *cur;
2365 if (nes->entry_cnt >= max)
2368 list_for_each_entry(cur, head, set_list) {
2369 if (cur->entry_cnt >= nes->entry_cnt) {
2370 list_add(&nes->set_list, cur->set_list.prev);
2375 list_add_tail(&nes->set_list, head);
2378 static void __update_nat_bits(struct f2fs_sb_info *sbi, nid_t start_nid,
2381 struct f2fs_nm_info *nm_i = NM_I(sbi);
2382 unsigned int nat_index = start_nid / NAT_ENTRY_PER_BLOCK;
2383 struct f2fs_nat_block *nat_blk = page_address(page);
2387 if (!enabled_nat_bits(sbi, NULL))
2390 for (i = 0; i < NAT_ENTRY_PER_BLOCK; i++) {
2391 if (start_nid == 0 && i == 0)
2393 if (nat_blk->entries[i].block_addr)
2397 __set_bit_le(nat_index, nm_i->empty_nat_bits);
2398 __clear_bit_le(nat_index, nm_i->full_nat_bits);
2402 __clear_bit_le(nat_index, nm_i->empty_nat_bits);
2403 if (valid == NAT_ENTRY_PER_BLOCK)
2404 __set_bit_le(nat_index, nm_i->full_nat_bits);
2406 __clear_bit_le(nat_index, nm_i->full_nat_bits);
2409 static void __flush_nat_entry_set(struct f2fs_sb_info *sbi,
2410 struct nat_entry_set *set, struct cp_control *cpc)
2412 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
2413 struct f2fs_journal *journal = curseg->journal;
2414 nid_t start_nid = set->set * NAT_ENTRY_PER_BLOCK;
2415 bool to_journal = true;
2416 struct f2fs_nat_block *nat_blk;
2417 struct nat_entry *ne, *cur;
2418 struct page *page = NULL;
2421 * there are two steps to flush nat entries:
2422 * #1, flush nat entries to journal in current hot data summary block.
2423 * #2, flush nat entries to nat page.
2425 if (enabled_nat_bits(sbi, cpc) ||
2426 !__has_cursum_space(journal, set->entry_cnt, NAT_JOURNAL))
2430 down_write(&curseg->journal_rwsem);
2432 page = get_next_nat_page(sbi, start_nid);
2433 nat_blk = page_address(page);
2434 f2fs_bug_on(sbi, !nat_blk);
2437 /* flush dirty nats in nat entry set */
2438 list_for_each_entry_safe(ne, cur, &set->entry_list, list) {
2439 struct f2fs_nat_entry *raw_ne;
2440 nid_t nid = nat_get_nid(ne);
2443 f2fs_bug_on(sbi, nat_get_blkaddr(ne) == NEW_ADDR);
2446 offset = lookup_journal_in_cursum(journal,
2447 NAT_JOURNAL, nid, 1);
2448 f2fs_bug_on(sbi, offset < 0);
2449 raw_ne = &nat_in_journal(journal, offset);
2450 nid_in_journal(journal, offset) = cpu_to_le32(nid);
2452 raw_ne = &nat_blk->entries[nid - start_nid];
2454 raw_nat_from_node_info(raw_ne, &ne->ni);
2456 __clear_nat_cache_dirty(NM_I(sbi), set, ne);
2457 if (nat_get_blkaddr(ne) == NULL_ADDR) {
2458 add_free_nid(sbi, nid, false);
2459 spin_lock(&NM_I(sbi)->nid_list_lock);
2460 NM_I(sbi)->available_nids++;
2461 update_free_nid_bitmap(sbi, nid, true, false);
2462 spin_unlock(&NM_I(sbi)->nid_list_lock);
2464 spin_lock(&NM_I(sbi)->nid_list_lock);
2465 update_free_nid_bitmap(sbi, nid, false, false);
2466 spin_unlock(&NM_I(sbi)->nid_list_lock);
2471 up_write(&curseg->journal_rwsem);
2473 __update_nat_bits(sbi, start_nid, page);
2474 f2fs_put_page(page, 1);
2477 /* Allow dirty nats by node block allocation in write_begin */
2478 if (!set->entry_cnt) {
2479 radix_tree_delete(&NM_I(sbi)->nat_set_root, set->set);
2480 kmem_cache_free(nat_entry_set_slab, set);
2485 * This function is called during the checkpointing process.
2487 void flush_nat_entries(struct f2fs_sb_info *sbi, struct cp_control *cpc)
2489 struct f2fs_nm_info *nm_i = NM_I(sbi);
2490 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
2491 struct f2fs_journal *journal = curseg->journal;
2492 struct nat_entry_set *setvec[SETVEC_SIZE];
2493 struct nat_entry_set *set, *tmp;
2498 if (!nm_i->dirty_nat_cnt)
2501 down_write(&nm_i->nat_tree_lock);
2504 * if there are no enough space in journal to store dirty nat
2505 * entries, remove all entries from journal and merge them
2506 * into nat entry set.
2508 if (enabled_nat_bits(sbi, cpc) ||
2509 !__has_cursum_space(journal, nm_i->dirty_nat_cnt, NAT_JOURNAL))
2510 remove_nats_in_journal(sbi);
2512 while ((found = __gang_lookup_nat_set(nm_i,
2513 set_idx, SETVEC_SIZE, setvec))) {
2515 set_idx = setvec[found - 1]->set + 1;
2516 for (idx = 0; idx < found; idx++)
2517 __adjust_nat_entry_set(setvec[idx], &sets,
2518 MAX_NAT_JENTRIES(journal));
2521 /* flush dirty nats in nat entry set */
2522 list_for_each_entry_safe(set, tmp, &sets, set_list)
2523 __flush_nat_entry_set(sbi, set, cpc);
2525 up_write(&nm_i->nat_tree_lock);
2526 /* Allow dirty nats by node block allocation in write_begin */
2529 static int __get_nat_bitmaps(struct f2fs_sb_info *sbi)
2531 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
2532 struct f2fs_nm_info *nm_i = NM_I(sbi);
2533 unsigned int nat_bits_bytes = nm_i->nat_blocks / BITS_PER_BYTE;
2535 __u64 cp_ver = cur_cp_version(ckpt);
2536 block_t nat_bits_addr;
2538 if (!enabled_nat_bits(sbi, NULL))
2541 nm_i->nat_bits_blocks = F2FS_BYTES_TO_BLK((nat_bits_bytes << 1) + 8 +
2543 nm_i->nat_bits = kzalloc(nm_i->nat_bits_blocks << F2FS_BLKSIZE_BITS,
2545 if (!nm_i->nat_bits)
2548 nat_bits_addr = __start_cp_addr(sbi) + sbi->blocks_per_seg -
2549 nm_i->nat_bits_blocks;
2550 for (i = 0; i < nm_i->nat_bits_blocks; i++) {
2551 struct page *page = get_meta_page(sbi, nat_bits_addr++);
2553 memcpy(nm_i->nat_bits + (i << F2FS_BLKSIZE_BITS),
2554 page_address(page), F2FS_BLKSIZE);
2555 f2fs_put_page(page, 1);
2558 cp_ver |= (cur_cp_crc(ckpt) << 32);
2559 if (cpu_to_le64(cp_ver) != *(__le64 *)nm_i->nat_bits) {
2560 disable_nat_bits(sbi, true);
2564 nm_i->full_nat_bits = nm_i->nat_bits + 8;
2565 nm_i->empty_nat_bits = nm_i->full_nat_bits + nat_bits_bytes;
2567 f2fs_msg(sbi->sb, KERN_NOTICE, "Found nat_bits in checkpoint");
2571 static inline void load_free_nid_bitmap(struct f2fs_sb_info *sbi)
2573 struct f2fs_nm_info *nm_i = NM_I(sbi);
2575 nid_t nid, last_nid;
2577 if (!enabled_nat_bits(sbi, NULL))
2580 for (i = 0; i < nm_i->nat_blocks; i++) {
2581 i = find_next_bit_le(nm_i->empty_nat_bits, nm_i->nat_blocks, i);
2582 if (i >= nm_i->nat_blocks)
2585 __set_bit_le(i, nm_i->nat_block_bitmap);
2587 nid = i * NAT_ENTRY_PER_BLOCK;
2588 last_nid = (i + 1) * NAT_ENTRY_PER_BLOCK;
2590 spin_lock(&NM_I(sbi)->nid_list_lock);
2591 for (; nid < last_nid; nid++)
2592 update_free_nid_bitmap(sbi, nid, true, true);
2593 spin_unlock(&NM_I(sbi)->nid_list_lock);
2596 for (i = 0; i < nm_i->nat_blocks; i++) {
2597 i = find_next_bit_le(nm_i->full_nat_bits, nm_i->nat_blocks, i);
2598 if (i >= nm_i->nat_blocks)
2601 __set_bit_le(i, nm_i->nat_block_bitmap);
2605 static int init_node_manager(struct f2fs_sb_info *sbi)
2607 struct f2fs_super_block *sb_raw = F2FS_RAW_SUPER(sbi);
2608 struct f2fs_nm_info *nm_i = NM_I(sbi);
2609 unsigned char *version_bitmap;
2610 unsigned int nat_segs;
2613 nm_i->nat_blkaddr = le32_to_cpu(sb_raw->nat_blkaddr);
2615 /* segment_count_nat includes pair segment so divide to 2. */
2616 nat_segs = le32_to_cpu(sb_raw->segment_count_nat) >> 1;
2617 nm_i->nat_blocks = nat_segs << le32_to_cpu(sb_raw->log_blocks_per_seg);
2618 nm_i->max_nid = NAT_ENTRY_PER_BLOCK * nm_i->nat_blocks;
2620 /* not used nids: 0, node, meta, (and root counted as valid node) */
2621 nm_i->available_nids = nm_i->max_nid - sbi->total_valid_node_count -
2622 F2FS_RESERVED_NODE_NUM;
2623 nm_i->nid_cnt[FREE_NID_LIST] = 0;
2624 nm_i->nid_cnt[ALLOC_NID_LIST] = 0;
2626 nm_i->ram_thresh = DEF_RAM_THRESHOLD;
2627 nm_i->ra_nid_pages = DEF_RA_NID_PAGES;
2628 nm_i->dirty_nats_ratio = DEF_DIRTY_NAT_RATIO_THRESHOLD;
2630 INIT_RADIX_TREE(&nm_i->free_nid_root, GFP_ATOMIC);
2631 INIT_LIST_HEAD(&nm_i->nid_list[FREE_NID_LIST]);
2632 INIT_LIST_HEAD(&nm_i->nid_list[ALLOC_NID_LIST]);
2633 INIT_RADIX_TREE(&nm_i->nat_root, GFP_NOIO);
2634 INIT_RADIX_TREE(&nm_i->nat_set_root, GFP_NOIO);
2635 INIT_LIST_HEAD(&nm_i->nat_entries);
2637 mutex_init(&nm_i->build_lock);
2638 spin_lock_init(&nm_i->nid_list_lock);
2639 init_rwsem(&nm_i->nat_tree_lock);
2641 nm_i->next_scan_nid = le32_to_cpu(sbi->ckpt->next_free_nid);
2642 nm_i->bitmap_size = __bitmap_size(sbi, NAT_BITMAP);
2643 version_bitmap = __bitmap_ptr(sbi, NAT_BITMAP);
2644 if (!version_bitmap)
2647 nm_i->nat_bitmap = kmemdup(version_bitmap, nm_i->bitmap_size,
2649 if (!nm_i->nat_bitmap)
2652 err = __get_nat_bitmaps(sbi);
2656 #ifdef CONFIG_F2FS_CHECK_FS
2657 nm_i->nat_bitmap_mir = kmemdup(version_bitmap, nm_i->bitmap_size,
2659 if (!nm_i->nat_bitmap_mir)
2666 static int init_free_nid_cache(struct f2fs_sb_info *sbi)
2668 struct f2fs_nm_info *nm_i = NM_I(sbi);
2670 nm_i->free_nid_bitmap = kvzalloc(nm_i->nat_blocks *
2671 NAT_ENTRY_BITMAP_SIZE, GFP_KERNEL);
2672 if (!nm_i->free_nid_bitmap)
2675 nm_i->nat_block_bitmap = kvzalloc(nm_i->nat_blocks / 8,
2677 if (!nm_i->nat_block_bitmap)
2680 nm_i->free_nid_count = kvzalloc(nm_i->nat_blocks *
2681 sizeof(unsigned short), GFP_KERNEL);
2682 if (!nm_i->free_nid_count)
2687 int build_node_manager(struct f2fs_sb_info *sbi)
2691 sbi->nm_info = kzalloc(sizeof(struct f2fs_nm_info), GFP_KERNEL);
2695 err = init_node_manager(sbi);
2699 err = init_free_nid_cache(sbi);
2703 /* load free nid status from nat_bits table */
2704 load_free_nid_bitmap(sbi);
2706 build_free_nids(sbi, true, true);
2710 void destroy_node_manager(struct f2fs_sb_info *sbi)
2712 struct f2fs_nm_info *nm_i = NM_I(sbi);
2713 struct free_nid *i, *next_i;
2714 struct nat_entry *natvec[NATVEC_SIZE];
2715 struct nat_entry_set *setvec[SETVEC_SIZE];
2722 /* destroy free nid list */
2723 spin_lock(&nm_i->nid_list_lock);
2724 list_for_each_entry_safe(i, next_i, &nm_i->nid_list[FREE_NID_LIST],
2726 __remove_nid_from_list(sbi, i, FREE_NID_LIST, false);
2727 spin_unlock(&nm_i->nid_list_lock);
2728 kmem_cache_free(free_nid_slab, i);
2729 spin_lock(&nm_i->nid_list_lock);
2731 f2fs_bug_on(sbi, nm_i->nid_cnt[FREE_NID_LIST]);
2732 f2fs_bug_on(sbi, nm_i->nid_cnt[ALLOC_NID_LIST]);
2733 f2fs_bug_on(sbi, !list_empty(&nm_i->nid_list[ALLOC_NID_LIST]));
2734 spin_unlock(&nm_i->nid_list_lock);
2736 /* destroy nat cache */
2737 down_write(&nm_i->nat_tree_lock);
2738 while ((found = __gang_lookup_nat_cache(nm_i,
2739 nid, NATVEC_SIZE, natvec))) {
2742 nid = nat_get_nid(natvec[found - 1]) + 1;
2743 for (idx = 0; idx < found; idx++)
2744 __del_from_nat_cache(nm_i, natvec[idx]);
2746 f2fs_bug_on(sbi, nm_i->nat_cnt);
2748 /* destroy nat set cache */
2750 while ((found = __gang_lookup_nat_set(nm_i,
2751 nid, SETVEC_SIZE, setvec))) {
2754 nid = setvec[found - 1]->set + 1;
2755 for (idx = 0; idx < found; idx++) {
2756 /* entry_cnt is not zero, when cp_error was occurred */
2757 f2fs_bug_on(sbi, !list_empty(&setvec[idx]->entry_list));
2758 radix_tree_delete(&nm_i->nat_set_root, setvec[idx]->set);
2759 kmem_cache_free(nat_entry_set_slab, setvec[idx]);
2762 up_write(&nm_i->nat_tree_lock);
2764 kvfree(nm_i->nat_block_bitmap);
2765 kvfree(nm_i->free_nid_bitmap);
2766 kvfree(nm_i->free_nid_count);
2768 kfree(nm_i->nat_bitmap);
2769 kfree(nm_i->nat_bits);
2770 #ifdef CONFIG_F2FS_CHECK_FS
2771 kfree(nm_i->nat_bitmap_mir);
2773 sbi->nm_info = NULL;
2777 int __init create_node_manager_caches(void)
2779 nat_entry_slab = f2fs_kmem_cache_create("nat_entry",
2780 sizeof(struct nat_entry));
2781 if (!nat_entry_slab)
2784 free_nid_slab = f2fs_kmem_cache_create("free_nid",
2785 sizeof(struct free_nid));
2787 goto destroy_nat_entry;
2789 nat_entry_set_slab = f2fs_kmem_cache_create("nat_entry_set",
2790 sizeof(struct nat_entry_set));
2791 if (!nat_entry_set_slab)
2792 goto destroy_free_nid;
2796 kmem_cache_destroy(free_nid_slab);
2798 kmem_cache_destroy(nat_entry_slab);
2803 void destroy_node_manager_caches(void)
2805 kmem_cache_destroy(nat_entry_set_slab);
2806 kmem_cache_destroy(free_nid_slab);
2807 kmem_cache_destroy(nat_entry_slab);