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>
22 #include <trace/events/f2fs.h>
24 static struct kmem_cache *nat_entry_slab;
25 static struct kmem_cache *free_nid_slab;
27 static void clear_node_page_dirty(struct page *page)
29 struct address_space *mapping = page->mapping;
30 struct f2fs_sb_info *sbi = F2FS_SB(mapping->host->i_sb);
31 unsigned int long flags;
33 if (PageDirty(page)) {
34 spin_lock_irqsave(&mapping->tree_lock, flags);
35 radix_tree_tag_clear(&mapping->page_tree,
38 spin_unlock_irqrestore(&mapping->tree_lock, flags);
40 clear_page_dirty_for_io(page);
41 dec_page_count(sbi, F2FS_DIRTY_NODES);
43 ClearPageUptodate(page);
46 static struct page *get_current_nat_page(struct f2fs_sb_info *sbi, nid_t nid)
48 pgoff_t index = current_nat_addr(sbi, nid);
49 return get_meta_page(sbi, index);
52 static struct page *get_next_nat_page(struct f2fs_sb_info *sbi, nid_t nid)
54 struct page *src_page;
55 struct page *dst_page;
60 struct f2fs_nm_info *nm_i = NM_I(sbi);
62 src_off = current_nat_addr(sbi, nid);
63 dst_off = next_nat_addr(sbi, src_off);
65 /* get current nat block page with lock */
66 src_page = get_meta_page(sbi, src_off);
68 /* Dirty src_page means that it is already the new target NAT page. */
69 if (PageDirty(src_page))
72 dst_page = grab_meta_page(sbi, dst_off);
74 src_addr = page_address(src_page);
75 dst_addr = page_address(dst_page);
76 memcpy(dst_addr, src_addr, PAGE_CACHE_SIZE);
77 set_page_dirty(dst_page);
78 f2fs_put_page(src_page, 1);
80 set_to_next_nat(nm_i, nid);
88 static void ra_nat_pages(struct f2fs_sb_info *sbi, int nid)
90 struct address_space *mapping = META_MAPPING(sbi);
91 struct f2fs_nm_info *nm_i = NM_I(sbi);
95 struct f2fs_io_info fio = {
97 .rw = READ_SYNC | REQ_META | REQ_PRIO
101 for (i = 0; i < FREE_NID_PAGES; i++, nid += NAT_ENTRY_PER_BLOCK) {
102 if (unlikely(nid >= nm_i->max_nid))
104 index = current_nat_addr(sbi, nid);
106 page = grab_cache_page(mapping, index);
109 if (PageUptodate(page)) {
110 mark_page_accessed(page);
111 f2fs_put_page(page, 1);
114 f2fs_submit_page_mbio(sbi, page, index, &fio);
115 mark_page_accessed(page);
116 f2fs_put_page(page, 0);
118 f2fs_submit_merged_bio(sbi, META, READ);
121 static struct nat_entry *__lookup_nat_cache(struct f2fs_nm_info *nm_i, nid_t n)
123 return radix_tree_lookup(&nm_i->nat_root, n);
126 static unsigned int __gang_lookup_nat_cache(struct f2fs_nm_info *nm_i,
127 nid_t start, unsigned int nr, struct nat_entry **ep)
129 return radix_tree_gang_lookup(&nm_i->nat_root, (void **)ep, start, nr);
132 static void __del_from_nat_cache(struct f2fs_nm_info *nm_i, struct nat_entry *e)
135 radix_tree_delete(&nm_i->nat_root, nat_get_nid(e));
137 kmem_cache_free(nat_entry_slab, e);
140 int is_checkpointed_node(struct f2fs_sb_info *sbi, nid_t nid)
142 struct f2fs_nm_info *nm_i = NM_I(sbi);
146 read_lock(&nm_i->nat_tree_lock);
147 e = __lookup_nat_cache(nm_i, nid);
148 if (e && !e->checkpointed)
150 read_unlock(&nm_i->nat_tree_lock);
154 static struct nat_entry *grab_nat_entry(struct f2fs_nm_info *nm_i, nid_t nid)
156 struct nat_entry *new;
158 new = kmem_cache_alloc(nat_entry_slab, GFP_ATOMIC);
161 if (radix_tree_insert(&nm_i->nat_root, nid, new)) {
162 kmem_cache_free(nat_entry_slab, new);
165 memset(new, 0, sizeof(struct nat_entry));
166 nat_set_nid(new, nid);
167 list_add_tail(&new->list, &nm_i->nat_entries);
172 static void cache_nat_entry(struct f2fs_nm_info *nm_i, nid_t nid,
173 struct f2fs_nat_entry *ne)
177 write_lock(&nm_i->nat_tree_lock);
178 e = __lookup_nat_cache(nm_i, nid);
180 e = grab_nat_entry(nm_i, nid);
182 write_unlock(&nm_i->nat_tree_lock);
185 nat_set_blkaddr(e, le32_to_cpu(ne->block_addr));
186 nat_set_ino(e, le32_to_cpu(ne->ino));
187 nat_set_version(e, ne->version);
188 e->checkpointed = true;
190 write_unlock(&nm_i->nat_tree_lock);
193 static void set_node_addr(struct f2fs_sb_info *sbi, struct node_info *ni,
196 struct f2fs_nm_info *nm_i = NM_I(sbi);
199 write_lock(&nm_i->nat_tree_lock);
200 e = __lookup_nat_cache(nm_i, ni->nid);
202 e = grab_nat_entry(nm_i, ni->nid);
204 write_unlock(&nm_i->nat_tree_lock);
208 e->checkpointed = true;
209 f2fs_bug_on(ni->blk_addr == NEW_ADDR);
210 } else if (new_blkaddr == NEW_ADDR) {
212 * when nid is reallocated,
213 * previous nat entry can be remained in nat cache.
214 * So, reinitialize it with new information.
217 f2fs_bug_on(ni->blk_addr != NULL_ADDR);
220 if (new_blkaddr == NEW_ADDR)
221 e->checkpointed = false;
224 f2fs_bug_on(nat_get_blkaddr(e) != ni->blk_addr);
225 f2fs_bug_on(nat_get_blkaddr(e) == NULL_ADDR &&
226 new_blkaddr == NULL_ADDR);
227 f2fs_bug_on(nat_get_blkaddr(e) == NEW_ADDR &&
228 new_blkaddr == NEW_ADDR);
229 f2fs_bug_on(nat_get_blkaddr(e) != NEW_ADDR &&
230 nat_get_blkaddr(e) != NULL_ADDR &&
231 new_blkaddr == NEW_ADDR);
233 /* increament version no as node is removed */
234 if (nat_get_blkaddr(e) != NEW_ADDR && new_blkaddr == NULL_ADDR) {
235 unsigned char version = nat_get_version(e);
236 nat_set_version(e, inc_node_version(version));
240 nat_set_blkaddr(e, new_blkaddr);
241 __set_nat_cache_dirty(nm_i, e);
242 write_unlock(&nm_i->nat_tree_lock);
245 int try_to_free_nats(struct f2fs_sb_info *sbi, int nr_shrink)
247 struct f2fs_nm_info *nm_i = NM_I(sbi);
249 if (nm_i->nat_cnt <= NM_WOUT_THRESHOLD)
252 write_lock(&nm_i->nat_tree_lock);
253 while (nr_shrink && !list_empty(&nm_i->nat_entries)) {
254 struct nat_entry *ne;
255 ne = list_first_entry(&nm_i->nat_entries,
256 struct nat_entry, list);
257 __del_from_nat_cache(nm_i, ne);
260 write_unlock(&nm_i->nat_tree_lock);
265 * This function returns always success
267 void get_node_info(struct f2fs_sb_info *sbi, nid_t nid, struct node_info *ni)
269 struct f2fs_nm_info *nm_i = NM_I(sbi);
270 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
271 struct f2fs_summary_block *sum = curseg->sum_blk;
272 nid_t start_nid = START_NID(nid);
273 struct f2fs_nat_block *nat_blk;
274 struct page *page = NULL;
275 struct f2fs_nat_entry ne;
279 memset(&ne, 0, sizeof(struct f2fs_nat_entry));
282 /* Check nat cache */
283 read_lock(&nm_i->nat_tree_lock);
284 e = __lookup_nat_cache(nm_i, nid);
286 ni->ino = nat_get_ino(e);
287 ni->blk_addr = nat_get_blkaddr(e);
288 ni->version = nat_get_version(e);
290 read_unlock(&nm_i->nat_tree_lock);
294 /* Check current segment summary */
295 mutex_lock(&curseg->curseg_mutex);
296 i = lookup_journal_in_cursum(sum, NAT_JOURNAL, nid, 0);
298 ne = nat_in_journal(sum, i);
299 node_info_from_raw_nat(ni, &ne);
301 mutex_unlock(&curseg->curseg_mutex);
305 /* Fill node_info from nat page */
306 page = get_current_nat_page(sbi, start_nid);
307 nat_blk = (struct f2fs_nat_block *)page_address(page);
308 ne = nat_blk->entries[nid - start_nid];
309 node_info_from_raw_nat(ni, &ne);
310 f2fs_put_page(page, 1);
312 /* cache nat entry */
313 cache_nat_entry(NM_I(sbi), nid, &ne);
317 * The maximum depth is four.
318 * Offset[0] will have raw inode offset.
320 static int get_node_path(struct f2fs_inode_info *fi, long block,
321 int offset[4], unsigned int noffset[4])
323 const long direct_index = ADDRS_PER_INODE(fi);
324 const long direct_blks = ADDRS_PER_BLOCK;
325 const long dptrs_per_blk = NIDS_PER_BLOCK;
326 const long indirect_blks = ADDRS_PER_BLOCK * NIDS_PER_BLOCK;
327 const long dindirect_blks = indirect_blks * NIDS_PER_BLOCK;
333 if (block < direct_index) {
337 block -= direct_index;
338 if (block < direct_blks) {
339 offset[n++] = NODE_DIR1_BLOCK;
345 block -= direct_blks;
346 if (block < direct_blks) {
347 offset[n++] = NODE_DIR2_BLOCK;
353 block -= direct_blks;
354 if (block < indirect_blks) {
355 offset[n++] = NODE_IND1_BLOCK;
357 offset[n++] = block / direct_blks;
358 noffset[n] = 4 + offset[n - 1];
359 offset[n] = block % direct_blks;
363 block -= indirect_blks;
364 if (block < indirect_blks) {
365 offset[n++] = NODE_IND2_BLOCK;
366 noffset[n] = 4 + dptrs_per_blk;
367 offset[n++] = block / direct_blks;
368 noffset[n] = 5 + dptrs_per_blk + offset[n - 1];
369 offset[n] = block % direct_blks;
373 block -= indirect_blks;
374 if (block < dindirect_blks) {
375 offset[n++] = NODE_DIND_BLOCK;
376 noffset[n] = 5 + (dptrs_per_blk * 2);
377 offset[n++] = block / indirect_blks;
378 noffset[n] = 6 + (dptrs_per_blk * 2) +
379 offset[n - 1] * (dptrs_per_blk + 1);
380 offset[n++] = (block / direct_blks) % dptrs_per_blk;
381 noffset[n] = 7 + (dptrs_per_blk * 2) +
382 offset[n - 2] * (dptrs_per_blk + 1) +
384 offset[n] = block % direct_blks;
395 * Caller should call f2fs_put_dnode(dn).
396 * Also, it should grab and release a rwsem by calling f2fs_lock_op() and
397 * f2fs_unlock_op() only if ro is not set RDONLY_NODE.
398 * In the case of RDONLY_NODE, we don't need to care about mutex.
400 int get_dnode_of_data(struct dnode_of_data *dn, pgoff_t index, int mode)
402 struct f2fs_sb_info *sbi = F2FS_SB(dn->inode->i_sb);
403 struct page *npage[4];
406 unsigned int noffset[4];
411 level = get_node_path(F2FS_I(dn->inode), index, offset, noffset);
413 nids[0] = dn->inode->i_ino;
414 npage[0] = dn->inode_page;
417 npage[0] = get_node_page(sbi, nids[0]);
418 if (IS_ERR(npage[0]))
419 return PTR_ERR(npage[0]);
423 nids[1] = get_nid(parent, offset[0], true);
424 dn->inode_page = npage[0];
425 dn->inode_page_locked = true;
427 /* get indirect or direct nodes */
428 for (i = 1; i <= level; i++) {
431 if (!nids[i] && mode == ALLOC_NODE) {
433 if (!alloc_nid(sbi, &(nids[i]))) {
439 npage[i] = new_node_page(dn, noffset[i], NULL);
440 if (IS_ERR(npage[i])) {
441 alloc_nid_failed(sbi, nids[i]);
442 err = PTR_ERR(npage[i]);
446 set_nid(parent, offset[i - 1], nids[i], i == 1);
447 alloc_nid_done(sbi, nids[i]);
449 } else if (mode == LOOKUP_NODE_RA && i == level && level > 1) {
450 npage[i] = get_node_page_ra(parent, offset[i - 1]);
451 if (IS_ERR(npage[i])) {
452 err = PTR_ERR(npage[i]);
458 dn->inode_page_locked = false;
461 f2fs_put_page(parent, 1);
465 npage[i] = get_node_page(sbi, nids[i]);
466 if (IS_ERR(npage[i])) {
467 err = PTR_ERR(npage[i]);
468 f2fs_put_page(npage[0], 0);
474 nids[i + 1] = get_nid(parent, offset[i], false);
477 dn->nid = nids[level];
478 dn->ofs_in_node = offset[level];
479 dn->node_page = npage[level];
480 dn->data_blkaddr = datablock_addr(dn->node_page, dn->ofs_in_node);
484 f2fs_put_page(parent, 1);
486 f2fs_put_page(npage[0], 0);
488 dn->inode_page = NULL;
489 dn->node_page = NULL;
493 static void truncate_node(struct dnode_of_data *dn)
495 struct f2fs_sb_info *sbi = F2FS_SB(dn->inode->i_sb);
498 get_node_info(sbi, dn->nid, &ni);
499 if (dn->inode->i_blocks == 0) {
500 f2fs_bug_on(ni.blk_addr != NULL_ADDR);
503 f2fs_bug_on(ni.blk_addr == NULL_ADDR);
505 /* Deallocate node address */
506 invalidate_blocks(sbi, ni.blk_addr);
507 dec_valid_node_count(sbi, dn->inode);
508 set_node_addr(sbi, &ni, NULL_ADDR);
510 if (dn->nid == dn->inode->i_ino) {
511 remove_orphan_inode(sbi, dn->nid);
512 dec_valid_inode_count(sbi);
517 clear_node_page_dirty(dn->node_page);
518 F2FS_SET_SB_DIRT(sbi);
520 f2fs_put_page(dn->node_page, 1);
522 invalidate_mapping_pages(NODE_MAPPING(sbi),
523 dn->node_page->index, dn->node_page->index);
525 dn->node_page = NULL;
526 trace_f2fs_truncate_node(dn->inode, dn->nid, ni.blk_addr);
529 static int truncate_dnode(struct dnode_of_data *dn)
531 struct f2fs_sb_info *sbi = F2FS_SB(dn->inode->i_sb);
537 /* get direct node */
538 page = get_node_page(sbi, dn->nid);
539 if (IS_ERR(page) && PTR_ERR(page) == -ENOENT)
541 else if (IS_ERR(page))
542 return PTR_ERR(page);
544 /* Make dnode_of_data for parameter */
545 dn->node_page = page;
547 truncate_data_blocks(dn);
552 static int truncate_nodes(struct dnode_of_data *dn, unsigned int nofs,
555 struct f2fs_sb_info *sbi = F2FS_SB(dn->inode->i_sb);
556 struct dnode_of_data rdn = *dn;
558 struct f2fs_node *rn;
560 unsigned int child_nofs;
565 return NIDS_PER_BLOCK + 1;
567 trace_f2fs_truncate_nodes_enter(dn->inode, dn->nid, dn->data_blkaddr);
569 page = get_node_page(sbi, dn->nid);
571 trace_f2fs_truncate_nodes_exit(dn->inode, PTR_ERR(page));
572 return PTR_ERR(page);
575 rn = F2FS_NODE(page);
577 for (i = ofs; i < NIDS_PER_BLOCK; i++, freed++) {
578 child_nid = le32_to_cpu(rn->in.nid[i]);
582 ret = truncate_dnode(&rdn);
585 set_nid(page, i, 0, false);
588 child_nofs = nofs + ofs * (NIDS_PER_BLOCK + 1) + 1;
589 for (i = ofs; i < NIDS_PER_BLOCK; i++) {
590 child_nid = le32_to_cpu(rn->in.nid[i]);
591 if (child_nid == 0) {
592 child_nofs += NIDS_PER_BLOCK + 1;
596 ret = truncate_nodes(&rdn, child_nofs, 0, depth - 1);
597 if (ret == (NIDS_PER_BLOCK + 1)) {
598 set_nid(page, i, 0, false);
600 } else if (ret < 0 && ret != -ENOENT) {
608 /* remove current indirect node */
609 dn->node_page = page;
613 f2fs_put_page(page, 1);
615 trace_f2fs_truncate_nodes_exit(dn->inode, freed);
619 f2fs_put_page(page, 1);
620 trace_f2fs_truncate_nodes_exit(dn->inode, ret);
624 static int truncate_partial_nodes(struct dnode_of_data *dn,
625 struct f2fs_inode *ri, int *offset, int depth)
627 struct f2fs_sb_info *sbi = F2FS_SB(dn->inode->i_sb);
628 struct page *pages[2];
635 nid[0] = le32_to_cpu(ri->i_nid[offset[0] - NODE_DIR1_BLOCK]);
639 /* get indirect nodes in the path */
640 for (i = 0; i < idx + 1; i++) {
641 /* refernece count'll be increased */
642 pages[i] = get_node_page(sbi, nid[i]);
643 if (IS_ERR(pages[i])) {
644 err = PTR_ERR(pages[i]);
648 nid[i + 1] = get_nid(pages[i], offset[i + 1], false);
651 /* free direct nodes linked to a partial indirect node */
652 for (i = offset[idx + 1]; i < NIDS_PER_BLOCK; i++) {
653 child_nid = get_nid(pages[idx], i, false);
657 err = truncate_dnode(dn);
660 set_nid(pages[idx], i, 0, false);
663 if (offset[idx + 1] == 0) {
664 dn->node_page = pages[idx];
668 f2fs_put_page(pages[idx], 1);
674 for (i = idx; i >= 0; i--)
675 f2fs_put_page(pages[i], 1);
677 trace_f2fs_truncate_partial_nodes(dn->inode, nid, depth, err);
683 * All the block addresses of data and nodes should be nullified.
685 int truncate_inode_blocks(struct inode *inode, pgoff_t from)
687 struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
688 int err = 0, cont = 1;
689 int level, offset[4], noffset[4];
690 unsigned int nofs = 0;
691 struct f2fs_inode *ri;
692 struct dnode_of_data dn;
695 trace_f2fs_truncate_inode_blocks_enter(inode, from);
697 level = get_node_path(F2FS_I(inode), from, offset, noffset);
699 page = get_node_page(sbi, inode->i_ino);
701 trace_f2fs_truncate_inode_blocks_exit(inode, PTR_ERR(page));
702 return PTR_ERR(page);
705 set_new_dnode(&dn, inode, page, NULL, 0);
708 ri = F2FS_INODE(page);
716 if (!offset[level - 1])
718 err = truncate_partial_nodes(&dn, ri, offset, level);
719 if (err < 0 && err != -ENOENT)
721 nofs += 1 + NIDS_PER_BLOCK;
724 nofs = 5 + 2 * NIDS_PER_BLOCK;
725 if (!offset[level - 1])
727 err = truncate_partial_nodes(&dn, ri, offset, level);
728 if (err < 0 && err != -ENOENT)
737 dn.nid = le32_to_cpu(ri->i_nid[offset[0] - NODE_DIR1_BLOCK]);
739 case NODE_DIR1_BLOCK:
740 case NODE_DIR2_BLOCK:
741 err = truncate_dnode(&dn);
744 case NODE_IND1_BLOCK:
745 case NODE_IND2_BLOCK:
746 err = truncate_nodes(&dn, nofs, offset[1], 2);
749 case NODE_DIND_BLOCK:
750 err = truncate_nodes(&dn, nofs, offset[1], 3);
757 if (err < 0 && err != -ENOENT)
759 if (offset[1] == 0 &&
760 ri->i_nid[offset[0] - NODE_DIR1_BLOCK]) {
762 if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
763 f2fs_put_page(page, 1);
766 wait_on_page_writeback(page);
767 ri->i_nid[offset[0] - NODE_DIR1_BLOCK] = 0;
768 set_page_dirty(page);
776 f2fs_put_page(page, 0);
777 trace_f2fs_truncate_inode_blocks_exit(inode, err);
778 return err > 0 ? 0 : err;
781 int truncate_xattr_node(struct inode *inode, struct page *page)
783 struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
784 nid_t nid = F2FS_I(inode)->i_xattr_nid;
785 struct dnode_of_data dn;
791 npage = get_node_page(sbi, nid);
793 return PTR_ERR(npage);
795 F2FS_I(inode)->i_xattr_nid = 0;
797 /* need to do checkpoint during fsync */
798 F2FS_I(inode)->xattr_ver = cur_cp_version(F2FS_CKPT(sbi));
800 set_new_dnode(&dn, inode, page, npage, nid);
803 dn.inode_page_locked = true;
809 * Caller should grab and release a rwsem by calling f2fs_lock_op() and
812 void remove_inode_page(struct inode *inode)
814 struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
816 nid_t ino = inode->i_ino;
817 struct dnode_of_data dn;
819 page = get_node_page(sbi, ino);
823 if (truncate_xattr_node(inode, page)) {
824 f2fs_put_page(page, 1);
827 /* 0 is possible, after f2fs_new_inode() is failed */
828 f2fs_bug_on(inode->i_blocks != 0 && inode->i_blocks != 1);
829 set_new_dnode(&dn, inode, page, page, ino);
833 struct page *new_inode_page(struct inode *inode, const struct qstr *name)
835 struct dnode_of_data dn;
837 /* allocate inode page for new inode */
838 set_new_dnode(&dn, inode, NULL, NULL, inode->i_ino);
840 /* caller should f2fs_put_page(page, 1); */
841 return new_node_page(&dn, 0, NULL);
844 struct page *new_node_page(struct dnode_of_data *dn,
845 unsigned int ofs, struct page *ipage)
847 struct f2fs_sb_info *sbi = F2FS_SB(dn->inode->i_sb);
848 struct node_info old_ni, new_ni;
852 if (unlikely(is_inode_flag_set(F2FS_I(dn->inode), FI_NO_ALLOC)))
853 return ERR_PTR(-EPERM);
855 page = grab_cache_page(NODE_MAPPING(sbi), dn->nid);
857 return ERR_PTR(-ENOMEM);
859 if (unlikely(!inc_valid_node_count(sbi, dn->inode))) {
864 get_node_info(sbi, dn->nid, &old_ni);
866 /* Reinitialize old_ni with new node page */
867 f2fs_bug_on(old_ni.blk_addr != NULL_ADDR);
869 new_ni.ino = dn->inode->i_ino;
870 set_node_addr(sbi, &new_ni, NEW_ADDR);
872 fill_node_footer(page, dn->nid, dn->inode->i_ino, ofs, true);
873 set_cold_node(dn->inode, page);
874 SetPageUptodate(page);
875 set_page_dirty(page);
877 if (ofs == XATTR_NODE_OFFSET)
878 F2FS_I(dn->inode)->i_xattr_nid = dn->nid;
880 dn->node_page = page;
882 update_inode(dn->inode, ipage);
886 inc_valid_inode_count(sbi);
891 clear_node_page_dirty(page);
892 f2fs_put_page(page, 1);
897 * Caller should do after getting the following values.
898 * 0: f2fs_put_page(page, 0)
899 * LOCKED_PAGE: f2fs_put_page(page, 1)
902 static int read_node_page(struct page *page, int rw)
904 struct f2fs_sb_info *sbi = F2FS_SB(page->mapping->host->i_sb);
907 get_node_info(sbi, page->index, &ni);
909 if (unlikely(ni.blk_addr == NULL_ADDR)) {
910 f2fs_put_page(page, 1);
914 if (PageUptodate(page))
917 return f2fs_submit_page_bio(sbi, page, ni.blk_addr, rw);
921 * Readahead a node page
923 void ra_node_page(struct f2fs_sb_info *sbi, nid_t nid)
928 apage = find_get_page(NODE_MAPPING(sbi), nid);
929 if (apage && PageUptodate(apage)) {
930 f2fs_put_page(apage, 0);
933 f2fs_put_page(apage, 0);
935 apage = grab_cache_page(NODE_MAPPING(sbi), nid);
939 err = read_node_page(apage, READA);
941 f2fs_put_page(apage, 0);
942 else if (err == LOCKED_PAGE)
943 f2fs_put_page(apage, 1);
946 struct page *get_node_page(struct f2fs_sb_info *sbi, pgoff_t nid)
951 page = grab_cache_page(NODE_MAPPING(sbi), nid);
953 return ERR_PTR(-ENOMEM);
955 err = read_node_page(page, READ_SYNC);
958 else if (err == LOCKED_PAGE)
962 if (unlikely(!PageUptodate(page))) {
963 f2fs_put_page(page, 1);
964 return ERR_PTR(-EIO);
966 if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
967 f2fs_put_page(page, 1);
971 f2fs_bug_on(nid != nid_of_node(page));
972 mark_page_accessed(page);
977 * Return a locked page for the desired node page.
978 * And, readahead MAX_RA_NODE number of node pages.
980 struct page *get_node_page_ra(struct page *parent, int start)
982 struct f2fs_sb_info *sbi = F2FS_SB(parent->mapping->host->i_sb);
983 struct blk_plug plug;
988 /* First, try getting the desired direct node. */
989 nid = get_nid(parent, start, false);
991 return ERR_PTR(-ENOENT);
993 page = grab_cache_page(NODE_MAPPING(sbi), nid);
995 return ERR_PTR(-ENOMEM);
997 err = read_node_page(page, READ_SYNC);
1000 else if (err == LOCKED_PAGE)
1003 blk_start_plug(&plug);
1005 /* Then, try readahead for siblings of the desired node */
1006 end = start + MAX_RA_NODE;
1007 end = min(end, NIDS_PER_BLOCK);
1008 for (i = start + 1; i < end; i++) {
1009 nid = get_nid(parent, i, false);
1012 ra_node_page(sbi, nid);
1015 blk_finish_plug(&plug);
1018 if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1019 f2fs_put_page(page, 1);
1023 if (unlikely(!PageUptodate(page))) {
1024 f2fs_put_page(page, 1);
1025 return ERR_PTR(-EIO);
1027 mark_page_accessed(page);
1031 void sync_inode_page(struct dnode_of_data *dn)
1033 if (IS_INODE(dn->node_page) || dn->inode_page == dn->node_page) {
1034 update_inode(dn->inode, dn->node_page);
1035 } else if (dn->inode_page) {
1036 if (!dn->inode_page_locked)
1037 lock_page(dn->inode_page);
1038 update_inode(dn->inode, dn->inode_page);
1039 if (!dn->inode_page_locked)
1040 unlock_page(dn->inode_page);
1042 update_inode_page(dn->inode);
1046 int sync_node_pages(struct f2fs_sb_info *sbi, nid_t ino,
1047 struct writeback_control *wbc)
1050 struct pagevec pvec;
1051 int step = ino ? 2 : 0;
1052 int nwritten = 0, wrote = 0;
1054 pagevec_init(&pvec, 0);
1060 while (index <= end) {
1062 nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1063 PAGECACHE_TAG_DIRTY,
1064 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
1068 for (i = 0; i < nr_pages; i++) {
1069 struct page *page = pvec.pages[i];
1072 * flushing sequence with step:
1077 if (step == 0 && IS_DNODE(page))
1079 if (step == 1 && (!IS_DNODE(page) ||
1080 is_cold_node(page)))
1082 if (step == 2 && (!IS_DNODE(page) ||
1083 !is_cold_node(page)))
1088 * we should not skip writing node pages.
1090 if (ino && ino_of_node(page) == ino)
1092 else if (!trylock_page(page))
1095 if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1100 if (ino && ino_of_node(page) != ino)
1101 goto continue_unlock;
1103 if (!PageDirty(page)) {
1104 /* someone wrote it for us */
1105 goto continue_unlock;
1108 if (!clear_page_dirty_for_io(page))
1109 goto continue_unlock;
1111 /* called by fsync() */
1112 if (ino && IS_DNODE(page)) {
1113 int mark = !is_checkpointed_node(sbi, ino);
1114 set_fsync_mark(page, 1);
1116 set_dentry_mark(page, mark);
1119 set_fsync_mark(page, 0);
1120 set_dentry_mark(page, 0);
1122 NODE_MAPPING(sbi)->a_ops->writepage(page, wbc);
1125 if (--wbc->nr_to_write == 0)
1128 pagevec_release(&pvec);
1131 if (wbc->nr_to_write == 0) {
1143 f2fs_submit_merged_bio(sbi, NODE, WRITE);
1147 int wait_on_node_pages_writeback(struct f2fs_sb_info *sbi, nid_t ino)
1149 pgoff_t index = 0, end = LONG_MAX;
1150 struct pagevec pvec;
1151 int ret2 = 0, ret = 0;
1153 pagevec_init(&pvec, 0);
1155 while (index <= end) {
1157 nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1158 PAGECACHE_TAG_WRITEBACK,
1159 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
1163 for (i = 0; i < nr_pages; i++) {
1164 struct page *page = pvec.pages[i];
1166 /* until radix tree lookup accepts end_index */
1167 if (unlikely(page->index > end))
1170 if (ino && ino_of_node(page) == ino) {
1171 wait_on_page_writeback(page);
1172 if (TestClearPageError(page))
1176 pagevec_release(&pvec);
1180 if (unlikely(test_and_clear_bit(AS_ENOSPC, &NODE_MAPPING(sbi)->flags)))
1182 if (unlikely(test_and_clear_bit(AS_EIO, &NODE_MAPPING(sbi)->flags)))
1189 static int f2fs_write_node_page(struct page *page,
1190 struct writeback_control *wbc)
1192 struct f2fs_sb_info *sbi = F2FS_SB(page->mapping->host->i_sb);
1195 struct node_info ni;
1196 struct f2fs_io_info fio = {
1198 .rw = (wbc->sync_mode == WB_SYNC_ALL) ? WRITE_SYNC : WRITE,
1201 if (unlikely(sbi->por_doing))
1204 wait_on_page_writeback(page);
1206 /* get old block addr of this node page */
1207 nid = nid_of_node(page);
1208 f2fs_bug_on(page->index != nid);
1210 get_node_info(sbi, nid, &ni);
1212 /* This page is already truncated */
1213 if (unlikely(ni.blk_addr == NULL_ADDR)) {
1214 dec_page_count(sbi, F2FS_DIRTY_NODES);
1219 if (wbc->for_reclaim)
1222 mutex_lock(&sbi->node_write);
1223 set_page_writeback(page);
1224 write_node_page(sbi, page, &fio, nid, ni.blk_addr, &new_addr);
1225 set_node_addr(sbi, &ni, new_addr);
1226 dec_page_count(sbi, F2FS_DIRTY_NODES);
1227 mutex_unlock(&sbi->node_write);
1232 dec_page_count(sbi, F2FS_DIRTY_NODES);
1233 wbc->pages_skipped++;
1234 set_page_dirty(page);
1235 return AOP_WRITEPAGE_ACTIVATE;
1239 * It is very important to gather dirty pages and write at once, so that we can
1240 * submit a big bio without interfering other data writes.
1241 * Be default, 512 pages (2MB) * 3 node types, is more reasonable.
1243 #define COLLECT_DIRTY_NODES 1536
1244 static int f2fs_write_node_pages(struct address_space *mapping,
1245 struct writeback_control *wbc)
1247 struct f2fs_sb_info *sbi = F2FS_SB(mapping->host->i_sb);
1248 long nr_to_write = wbc->nr_to_write;
1250 /* balancing f2fs's metadata in background */
1251 f2fs_balance_fs_bg(sbi);
1253 /* collect a number of dirty node pages and write together */
1254 if (get_pages(sbi, F2FS_DIRTY_NODES) < COLLECT_DIRTY_NODES)
1257 /* if mounting is failed, skip writing node pages */
1258 wbc->nr_to_write = 3 * max_hw_blocks(sbi);
1259 wbc->sync_mode = WB_SYNC_NONE;
1260 sync_node_pages(sbi, 0, wbc);
1261 wbc->nr_to_write = nr_to_write - (3 * max_hw_blocks(sbi) -
1266 static int f2fs_set_node_page_dirty(struct page *page)
1268 struct address_space *mapping = page->mapping;
1269 struct f2fs_sb_info *sbi = F2FS_SB(mapping->host->i_sb);
1271 trace_f2fs_set_page_dirty(page, NODE);
1273 SetPageUptodate(page);
1274 if (!PageDirty(page)) {
1275 __set_page_dirty_nobuffers(page);
1276 inc_page_count(sbi, F2FS_DIRTY_NODES);
1277 SetPagePrivate(page);
1283 static void f2fs_invalidate_node_page(struct page *page, unsigned int offset,
1284 unsigned int length)
1286 struct inode *inode = page->mapping->host;
1287 struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
1288 if (PageDirty(page))
1289 dec_page_count(sbi, F2FS_DIRTY_NODES);
1290 ClearPagePrivate(page);
1293 static int f2fs_release_node_page(struct page *page, gfp_t wait)
1295 ClearPagePrivate(page);
1300 * Structure of the f2fs node operations
1302 const struct address_space_operations f2fs_node_aops = {
1303 .writepage = f2fs_write_node_page,
1304 .writepages = f2fs_write_node_pages,
1305 .set_page_dirty = f2fs_set_node_page_dirty,
1306 .invalidatepage = f2fs_invalidate_node_page,
1307 .releasepage = f2fs_release_node_page,
1310 static struct free_nid *__lookup_free_nid_list(nid_t n, struct list_head *head)
1312 struct list_head *this;
1314 list_for_each(this, head) {
1315 i = list_entry(this, struct free_nid, list);
1322 static void __del_from_free_nid_list(struct free_nid *i)
1325 kmem_cache_free(free_nid_slab, i);
1328 static int add_free_nid(struct f2fs_nm_info *nm_i, nid_t nid, bool build)
1331 struct nat_entry *ne;
1332 bool allocated = false;
1334 if (nm_i->fcnt > 2 * MAX_FREE_NIDS)
1337 /* 0 nid should not be used */
1338 if (unlikely(nid == 0))
1342 /* do not add allocated nids */
1343 read_lock(&nm_i->nat_tree_lock);
1344 ne = __lookup_nat_cache(nm_i, nid);
1345 if (ne && nat_get_blkaddr(ne) != NULL_ADDR)
1347 read_unlock(&nm_i->nat_tree_lock);
1352 i = f2fs_kmem_cache_alloc(free_nid_slab, GFP_NOFS);
1356 spin_lock(&nm_i->free_nid_list_lock);
1357 if (__lookup_free_nid_list(nid, &nm_i->free_nid_list)) {
1358 spin_unlock(&nm_i->free_nid_list_lock);
1359 kmem_cache_free(free_nid_slab, i);
1362 list_add_tail(&i->list, &nm_i->free_nid_list);
1364 spin_unlock(&nm_i->free_nid_list_lock);
1368 static void remove_free_nid(struct f2fs_nm_info *nm_i, nid_t nid)
1371 spin_lock(&nm_i->free_nid_list_lock);
1372 i = __lookup_free_nid_list(nid, &nm_i->free_nid_list);
1373 if (i && i->state == NID_NEW) {
1374 __del_from_free_nid_list(i);
1377 spin_unlock(&nm_i->free_nid_list_lock);
1380 static void scan_nat_page(struct f2fs_nm_info *nm_i,
1381 struct page *nat_page, nid_t start_nid)
1383 struct f2fs_nat_block *nat_blk = page_address(nat_page);
1387 i = start_nid % NAT_ENTRY_PER_BLOCK;
1389 for (; i < NAT_ENTRY_PER_BLOCK; i++, start_nid++) {
1391 if (unlikely(start_nid >= nm_i->max_nid))
1394 blk_addr = le32_to_cpu(nat_blk->entries[i].block_addr);
1395 f2fs_bug_on(blk_addr == NEW_ADDR);
1396 if (blk_addr == NULL_ADDR) {
1397 if (add_free_nid(nm_i, start_nid, true) < 0)
1403 static void build_free_nids(struct f2fs_sb_info *sbi)
1405 struct f2fs_nm_info *nm_i = NM_I(sbi);
1406 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
1407 struct f2fs_summary_block *sum = curseg->sum_blk;
1409 nid_t nid = nm_i->next_scan_nid;
1411 /* Enough entries */
1412 if (nm_i->fcnt > NAT_ENTRY_PER_BLOCK)
1415 /* readahead nat pages to be scanned */
1416 ra_nat_pages(sbi, nid);
1419 struct page *page = get_current_nat_page(sbi, nid);
1421 scan_nat_page(nm_i, page, nid);
1422 f2fs_put_page(page, 1);
1424 nid += (NAT_ENTRY_PER_BLOCK - (nid % NAT_ENTRY_PER_BLOCK));
1425 if (unlikely(nid >= nm_i->max_nid))
1428 if (i++ == FREE_NID_PAGES)
1432 /* go to the next free nat pages to find free nids abundantly */
1433 nm_i->next_scan_nid = nid;
1435 /* find free nids from current sum_pages */
1436 mutex_lock(&curseg->curseg_mutex);
1437 for (i = 0; i < nats_in_cursum(sum); i++) {
1438 block_t addr = le32_to_cpu(nat_in_journal(sum, i).block_addr);
1439 nid = le32_to_cpu(nid_in_journal(sum, i));
1440 if (addr == NULL_ADDR)
1441 add_free_nid(nm_i, nid, true);
1443 remove_free_nid(nm_i, nid);
1445 mutex_unlock(&curseg->curseg_mutex);
1449 * If this function returns success, caller can obtain a new nid
1450 * from second parameter of this function.
1451 * The returned nid could be used ino as well as nid when inode is created.
1453 bool alloc_nid(struct f2fs_sb_info *sbi, nid_t *nid)
1455 struct f2fs_nm_info *nm_i = NM_I(sbi);
1456 struct free_nid *i = NULL;
1457 struct list_head *this;
1459 if (unlikely(sbi->total_valid_node_count + 1 >= nm_i->max_nid))
1462 spin_lock(&nm_i->free_nid_list_lock);
1464 /* We should not use stale free nids created by build_free_nids */
1465 if (nm_i->fcnt && !sbi->on_build_free_nids) {
1466 f2fs_bug_on(list_empty(&nm_i->free_nid_list));
1467 list_for_each(this, &nm_i->free_nid_list) {
1468 i = list_entry(this, struct free_nid, list);
1469 if (i->state == NID_NEW)
1473 f2fs_bug_on(i->state != NID_NEW);
1475 i->state = NID_ALLOC;
1477 spin_unlock(&nm_i->free_nid_list_lock);
1480 spin_unlock(&nm_i->free_nid_list_lock);
1482 /* Let's scan nat pages and its caches to get free nids */
1483 mutex_lock(&nm_i->build_lock);
1484 sbi->on_build_free_nids = true;
1485 build_free_nids(sbi);
1486 sbi->on_build_free_nids = false;
1487 mutex_unlock(&nm_i->build_lock);
1492 * alloc_nid() should be called prior to this function.
1494 void alloc_nid_done(struct f2fs_sb_info *sbi, nid_t nid)
1496 struct f2fs_nm_info *nm_i = NM_I(sbi);
1499 spin_lock(&nm_i->free_nid_list_lock);
1500 i = __lookup_free_nid_list(nid, &nm_i->free_nid_list);
1501 f2fs_bug_on(!i || i->state != NID_ALLOC);
1502 __del_from_free_nid_list(i);
1503 spin_unlock(&nm_i->free_nid_list_lock);
1507 * alloc_nid() should be called prior to this function.
1509 void alloc_nid_failed(struct f2fs_sb_info *sbi, nid_t nid)
1511 struct f2fs_nm_info *nm_i = NM_I(sbi);
1517 spin_lock(&nm_i->free_nid_list_lock);
1518 i = __lookup_free_nid_list(nid, &nm_i->free_nid_list);
1519 f2fs_bug_on(!i || i->state != NID_ALLOC);
1520 if (nm_i->fcnt > 2 * MAX_FREE_NIDS) {
1521 __del_from_free_nid_list(i);
1526 spin_unlock(&nm_i->free_nid_list_lock);
1529 void recover_node_page(struct f2fs_sb_info *sbi, struct page *page,
1530 struct f2fs_summary *sum, struct node_info *ni,
1531 block_t new_blkaddr)
1533 rewrite_node_page(sbi, page, sum, ni->blk_addr, new_blkaddr);
1534 set_node_addr(sbi, ni, new_blkaddr);
1535 clear_node_page_dirty(page);
1538 int recover_inode_page(struct f2fs_sb_info *sbi, struct page *page)
1540 struct f2fs_inode *src, *dst;
1541 nid_t ino = ino_of_node(page);
1542 struct node_info old_ni, new_ni;
1545 ipage = grab_cache_page(NODE_MAPPING(sbi), ino);
1549 /* Should not use this inode from free nid list */
1550 remove_free_nid(NM_I(sbi), ino);
1552 get_node_info(sbi, ino, &old_ni);
1553 SetPageUptodate(ipage);
1554 fill_node_footer(ipage, ino, ino, 0, true);
1556 src = F2FS_INODE(page);
1557 dst = F2FS_INODE(ipage);
1559 memcpy(dst, src, (unsigned long)&src->i_ext - (unsigned long)src);
1561 dst->i_blocks = cpu_to_le64(1);
1562 dst->i_links = cpu_to_le32(1);
1563 dst->i_xattr_nid = 0;
1568 if (unlikely(!inc_valid_node_count(sbi, NULL)))
1570 set_node_addr(sbi, &new_ni, NEW_ADDR);
1571 inc_valid_inode_count(sbi);
1572 f2fs_put_page(ipage, 1);
1577 * ra_sum_pages() merge contiguous pages into one bio and submit.
1578 * these pre-readed pages are linked in pages list.
1580 static int ra_sum_pages(struct f2fs_sb_info *sbi, struct list_head *pages,
1581 int start, int nrpages)
1584 int page_idx = start;
1585 struct f2fs_io_info fio = {
1587 .rw = READ_SYNC | REQ_META | REQ_PRIO
1590 for (; page_idx < start + nrpages; page_idx++) {
1591 /* alloc temporal page for read node summary info*/
1592 page = alloc_page(GFP_F2FS_ZERO);
1595 list_for_each_entry_safe(page, tmp, pages, lru) {
1596 list_del(&page->lru);
1598 __free_pages(page, 0);
1604 page->index = page_idx;
1605 list_add_tail(&page->lru, pages);
1608 list_for_each_entry(page, pages, lru)
1609 f2fs_submit_page_mbio(sbi, page, page->index, &fio);
1611 f2fs_submit_merged_bio(sbi, META, READ);
1615 int restore_node_summary(struct f2fs_sb_info *sbi,
1616 unsigned int segno, struct f2fs_summary_block *sum)
1618 struct f2fs_node *rn;
1619 struct f2fs_summary *sum_entry;
1620 struct page *page, *tmp;
1622 int bio_blocks = MAX_BIO_BLOCKS(max_hw_blocks(sbi));
1623 int i, last_offset, nrpages, err = 0;
1624 LIST_HEAD(page_list);
1626 /* scan the node segment */
1627 last_offset = sbi->blocks_per_seg;
1628 addr = START_BLOCK(sbi, segno);
1629 sum_entry = &sum->entries[0];
1631 for (i = 0; i < last_offset; i += nrpages, addr += nrpages) {
1632 nrpages = min(last_offset - i, bio_blocks);
1634 /* read ahead node pages */
1635 err = ra_sum_pages(sbi, &page_list, addr, nrpages);
1639 list_for_each_entry_safe(page, tmp, &page_list, lru) {
1642 if (unlikely(!PageUptodate(page))) {
1645 rn = F2FS_NODE(page);
1646 sum_entry->nid = rn->footer.nid;
1647 sum_entry->version = 0;
1648 sum_entry->ofs_in_node = 0;
1652 list_del(&page->lru);
1654 __free_pages(page, 0);
1660 static bool flush_nats_in_journal(struct f2fs_sb_info *sbi)
1662 struct f2fs_nm_info *nm_i = NM_I(sbi);
1663 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
1664 struct f2fs_summary_block *sum = curseg->sum_blk;
1667 mutex_lock(&curseg->curseg_mutex);
1669 if (nats_in_cursum(sum) < NAT_JOURNAL_ENTRIES) {
1670 mutex_unlock(&curseg->curseg_mutex);
1674 for (i = 0; i < nats_in_cursum(sum); i++) {
1675 struct nat_entry *ne;
1676 struct f2fs_nat_entry raw_ne;
1677 nid_t nid = le32_to_cpu(nid_in_journal(sum, i));
1679 raw_ne = nat_in_journal(sum, i);
1681 write_lock(&nm_i->nat_tree_lock);
1682 ne = __lookup_nat_cache(nm_i, nid);
1684 __set_nat_cache_dirty(nm_i, ne);
1685 write_unlock(&nm_i->nat_tree_lock);
1688 ne = grab_nat_entry(nm_i, nid);
1690 write_unlock(&nm_i->nat_tree_lock);
1693 nat_set_blkaddr(ne, le32_to_cpu(raw_ne.block_addr));
1694 nat_set_ino(ne, le32_to_cpu(raw_ne.ino));
1695 nat_set_version(ne, raw_ne.version);
1696 __set_nat_cache_dirty(nm_i, ne);
1697 write_unlock(&nm_i->nat_tree_lock);
1699 update_nats_in_cursum(sum, -i);
1700 mutex_unlock(&curseg->curseg_mutex);
1705 * This function is called during the checkpointing process.
1707 void flush_nat_entries(struct f2fs_sb_info *sbi)
1709 struct f2fs_nm_info *nm_i = NM_I(sbi);
1710 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
1711 struct f2fs_summary_block *sum = curseg->sum_blk;
1712 struct list_head *cur, *n;
1713 struct page *page = NULL;
1714 struct f2fs_nat_block *nat_blk = NULL;
1715 nid_t start_nid = 0, end_nid = 0;
1718 flushed = flush_nats_in_journal(sbi);
1721 mutex_lock(&curseg->curseg_mutex);
1723 /* 1) flush dirty nat caches */
1724 list_for_each_safe(cur, n, &nm_i->dirty_nat_entries) {
1725 struct nat_entry *ne;
1727 struct f2fs_nat_entry raw_ne;
1729 block_t new_blkaddr;
1731 ne = list_entry(cur, struct nat_entry, list);
1732 nid = nat_get_nid(ne);
1734 if (nat_get_blkaddr(ne) == NEW_ADDR)
1739 /* if there is room for nat enries in curseg->sumpage */
1740 offset = lookup_journal_in_cursum(sum, NAT_JOURNAL, nid, 1);
1742 raw_ne = nat_in_journal(sum, offset);
1746 if (!page || (start_nid > nid || nid > end_nid)) {
1748 f2fs_put_page(page, 1);
1751 start_nid = START_NID(nid);
1752 end_nid = start_nid + NAT_ENTRY_PER_BLOCK - 1;
1755 * get nat block with dirty flag, increased reference
1756 * count, mapped and lock
1758 page = get_next_nat_page(sbi, start_nid);
1759 nat_blk = page_address(page);
1762 f2fs_bug_on(!nat_blk);
1763 raw_ne = nat_blk->entries[nid - start_nid];
1765 new_blkaddr = nat_get_blkaddr(ne);
1767 raw_ne.ino = cpu_to_le32(nat_get_ino(ne));
1768 raw_ne.block_addr = cpu_to_le32(new_blkaddr);
1769 raw_ne.version = nat_get_version(ne);
1772 nat_blk->entries[nid - start_nid] = raw_ne;
1774 nat_in_journal(sum, offset) = raw_ne;
1775 nid_in_journal(sum, offset) = cpu_to_le32(nid);
1778 if (nat_get_blkaddr(ne) == NULL_ADDR &&
1779 add_free_nid(NM_I(sbi), nid, false) <= 0) {
1780 write_lock(&nm_i->nat_tree_lock);
1781 __del_from_nat_cache(nm_i, ne);
1782 write_unlock(&nm_i->nat_tree_lock);
1784 write_lock(&nm_i->nat_tree_lock);
1785 __clear_nat_cache_dirty(nm_i, ne);
1786 ne->checkpointed = true;
1787 write_unlock(&nm_i->nat_tree_lock);
1791 mutex_unlock(&curseg->curseg_mutex);
1792 f2fs_put_page(page, 1);
1794 /* 2) shrink nat caches if necessary */
1795 try_to_free_nats(sbi, nm_i->nat_cnt - NM_WOUT_THRESHOLD);
1798 static int init_node_manager(struct f2fs_sb_info *sbi)
1800 struct f2fs_super_block *sb_raw = F2FS_RAW_SUPER(sbi);
1801 struct f2fs_nm_info *nm_i = NM_I(sbi);
1802 unsigned char *version_bitmap;
1803 unsigned int nat_segs, nat_blocks;
1805 nm_i->nat_blkaddr = le32_to_cpu(sb_raw->nat_blkaddr);
1807 /* segment_count_nat includes pair segment so divide to 2. */
1808 nat_segs = le32_to_cpu(sb_raw->segment_count_nat) >> 1;
1809 nat_blocks = nat_segs << le32_to_cpu(sb_raw->log_blocks_per_seg);
1810 nm_i->max_nid = NAT_ENTRY_PER_BLOCK * nat_blocks;
1814 INIT_LIST_HEAD(&nm_i->free_nid_list);
1815 INIT_RADIX_TREE(&nm_i->nat_root, GFP_ATOMIC);
1816 INIT_LIST_HEAD(&nm_i->nat_entries);
1817 INIT_LIST_HEAD(&nm_i->dirty_nat_entries);
1819 mutex_init(&nm_i->build_lock);
1820 spin_lock_init(&nm_i->free_nid_list_lock);
1821 rwlock_init(&nm_i->nat_tree_lock);
1823 nm_i->next_scan_nid = le32_to_cpu(sbi->ckpt->next_free_nid);
1824 nm_i->bitmap_size = __bitmap_size(sbi, NAT_BITMAP);
1825 version_bitmap = __bitmap_ptr(sbi, NAT_BITMAP);
1826 if (!version_bitmap)
1829 nm_i->nat_bitmap = kmemdup(version_bitmap, nm_i->bitmap_size,
1831 if (!nm_i->nat_bitmap)
1836 int build_node_manager(struct f2fs_sb_info *sbi)
1840 sbi->nm_info = kzalloc(sizeof(struct f2fs_nm_info), GFP_KERNEL);
1844 err = init_node_manager(sbi);
1848 build_free_nids(sbi);
1852 void destroy_node_manager(struct f2fs_sb_info *sbi)
1854 struct f2fs_nm_info *nm_i = NM_I(sbi);
1855 struct free_nid *i, *next_i;
1856 struct nat_entry *natvec[NATVEC_SIZE];
1863 /* destroy free nid list */
1864 spin_lock(&nm_i->free_nid_list_lock);
1865 list_for_each_entry_safe(i, next_i, &nm_i->free_nid_list, list) {
1866 f2fs_bug_on(i->state == NID_ALLOC);
1867 __del_from_free_nid_list(i);
1870 f2fs_bug_on(nm_i->fcnt);
1871 spin_unlock(&nm_i->free_nid_list_lock);
1873 /* destroy nat cache */
1874 write_lock(&nm_i->nat_tree_lock);
1875 while ((found = __gang_lookup_nat_cache(nm_i,
1876 nid, NATVEC_SIZE, natvec))) {
1878 for (idx = 0; idx < found; idx++) {
1879 struct nat_entry *e = natvec[idx];
1880 nid = nat_get_nid(e) + 1;
1881 __del_from_nat_cache(nm_i, e);
1884 f2fs_bug_on(nm_i->nat_cnt);
1885 write_unlock(&nm_i->nat_tree_lock);
1887 kfree(nm_i->nat_bitmap);
1888 sbi->nm_info = NULL;
1892 int __init create_node_manager_caches(void)
1894 nat_entry_slab = f2fs_kmem_cache_create("nat_entry",
1895 sizeof(struct nat_entry), NULL);
1896 if (!nat_entry_slab)
1899 free_nid_slab = f2fs_kmem_cache_create("free_nid",
1900 sizeof(struct free_nid), NULL);
1901 if (!free_nid_slab) {
1902 kmem_cache_destroy(nat_entry_slab);
1908 void destroy_node_manager_caches(void)
1910 kmem_cache_destroy(free_nid_slab);
1911 kmem_cache_destroy(nat_entry_slab);