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 #define on_build_free_nids(nmi) mutex_is_locked(&nm_i->build_lock)
26 static struct kmem_cache *nat_entry_slab;
27 static struct kmem_cache *free_nid_slab;
29 static void clear_node_page_dirty(struct page *page)
31 struct address_space *mapping = page->mapping;
32 struct f2fs_sb_info *sbi = F2FS_SB(mapping->host->i_sb);
33 unsigned int long flags;
35 if (PageDirty(page)) {
36 spin_lock_irqsave(&mapping->tree_lock, flags);
37 radix_tree_tag_clear(&mapping->page_tree,
40 spin_unlock_irqrestore(&mapping->tree_lock, flags);
42 clear_page_dirty_for_io(page);
43 dec_page_count(sbi, F2FS_DIRTY_NODES);
45 ClearPageUptodate(page);
48 static struct page *get_current_nat_page(struct f2fs_sb_info *sbi, nid_t nid)
50 pgoff_t index = current_nat_addr(sbi, nid);
51 return get_meta_page(sbi, index);
54 static struct page *get_next_nat_page(struct f2fs_sb_info *sbi, nid_t nid)
56 struct page *src_page;
57 struct page *dst_page;
62 struct f2fs_nm_info *nm_i = NM_I(sbi);
64 src_off = current_nat_addr(sbi, nid);
65 dst_off = next_nat_addr(sbi, src_off);
67 /* get current nat block page with lock */
68 src_page = get_meta_page(sbi, src_off);
70 /* Dirty src_page means that it is already the new target NAT page. */
71 if (PageDirty(src_page))
74 dst_page = grab_meta_page(sbi, dst_off);
76 src_addr = page_address(src_page);
77 dst_addr = page_address(dst_page);
78 memcpy(dst_addr, src_addr, PAGE_CACHE_SIZE);
79 set_page_dirty(dst_page);
80 f2fs_put_page(src_page, 1);
82 set_to_next_nat(nm_i, nid);
87 static struct nat_entry *__lookup_nat_cache(struct f2fs_nm_info *nm_i, nid_t n)
89 return radix_tree_lookup(&nm_i->nat_root, n);
92 static unsigned int __gang_lookup_nat_cache(struct f2fs_nm_info *nm_i,
93 nid_t start, unsigned int nr, struct nat_entry **ep)
95 return radix_tree_gang_lookup(&nm_i->nat_root, (void **)ep, start, nr);
98 static void __del_from_nat_cache(struct f2fs_nm_info *nm_i, struct nat_entry *e)
101 radix_tree_delete(&nm_i->nat_root, nat_get_nid(e));
103 kmem_cache_free(nat_entry_slab, e);
106 int is_checkpointed_node(struct f2fs_sb_info *sbi, nid_t nid)
108 struct f2fs_nm_info *nm_i = NM_I(sbi);
112 read_lock(&nm_i->nat_tree_lock);
113 e = __lookup_nat_cache(nm_i, nid);
114 if (e && !e->checkpointed)
116 read_unlock(&nm_i->nat_tree_lock);
120 static struct nat_entry *grab_nat_entry(struct f2fs_nm_info *nm_i, nid_t nid)
122 struct nat_entry *new;
124 new = kmem_cache_alloc(nat_entry_slab, GFP_ATOMIC);
127 if (radix_tree_insert(&nm_i->nat_root, nid, new)) {
128 kmem_cache_free(nat_entry_slab, new);
131 memset(new, 0, sizeof(struct nat_entry));
132 nat_set_nid(new, nid);
133 new->checkpointed = true;
134 list_add_tail(&new->list, &nm_i->nat_entries);
139 static void cache_nat_entry(struct f2fs_nm_info *nm_i, nid_t nid,
140 struct f2fs_nat_entry *ne)
144 write_lock(&nm_i->nat_tree_lock);
145 e = __lookup_nat_cache(nm_i, nid);
147 e = grab_nat_entry(nm_i, nid);
149 write_unlock(&nm_i->nat_tree_lock);
152 nat_set_blkaddr(e, le32_to_cpu(ne->block_addr));
153 nat_set_ino(e, le32_to_cpu(ne->ino));
154 nat_set_version(e, ne->version);
156 write_unlock(&nm_i->nat_tree_lock);
159 static void set_node_addr(struct f2fs_sb_info *sbi, struct node_info *ni,
162 struct f2fs_nm_info *nm_i = NM_I(sbi);
165 write_lock(&nm_i->nat_tree_lock);
166 e = __lookup_nat_cache(nm_i, ni->nid);
168 e = grab_nat_entry(nm_i, ni->nid);
170 write_unlock(&nm_i->nat_tree_lock);
174 f2fs_bug_on(ni->blk_addr == NEW_ADDR);
175 } else if (new_blkaddr == NEW_ADDR) {
177 * when nid is reallocated,
178 * previous nat entry can be remained in nat cache.
179 * So, reinitialize it with new information.
182 f2fs_bug_on(ni->blk_addr != NULL_ADDR);
186 f2fs_bug_on(nat_get_blkaddr(e) != ni->blk_addr);
187 f2fs_bug_on(nat_get_blkaddr(e) == NULL_ADDR &&
188 new_blkaddr == NULL_ADDR);
189 f2fs_bug_on(nat_get_blkaddr(e) == NEW_ADDR &&
190 new_blkaddr == NEW_ADDR);
191 f2fs_bug_on(nat_get_blkaddr(e) != NEW_ADDR &&
192 nat_get_blkaddr(e) != NULL_ADDR &&
193 new_blkaddr == NEW_ADDR);
195 /* increament version no as node is removed */
196 if (nat_get_blkaddr(e) != NEW_ADDR && new_blkaddr == NULL_ADDR) {
197 unsigned char version = nat_get_version(e);
198 nat_set_version(e, inc_node_version(version));
202 nat_set_blkaddr(e, new_blkaddr);
203 __set_nat_cache_dirty(nm_i, e);
204 write_unlock(&nm_i->nat_tree_lock);
207 int try_to_free_nats(struct f2fs_sb_info *sbi, int nr_shrink)
209 struct f2fs_nm_info *nm_i = NM_I(sbi);
211 if (nm_i->nat_cnt <= NM_WOUT_THRESHOLD)
214 write_lock(&nm_i->nat_tree_lock);
215 while (nr_shrink && !list_empty(&nm_i->nat_entries)) {
216 struct nat_entry *ne;
217 ne = list_first_entry(&nm_i->nat_entries,
218 struct nat_entry, list);
219 __del_from_nat_cache(nm_i, ne);
222 write_unlock(&nm_i->nat_tree_lock);
227 * This function returns always success
229 void get_node_info(struct f2fs_sb_info *sbi, nid_t nid, struct node_info *ni)
231 struct f2fs_nm_info *nm_i = NM_I(sbi);
232 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
233 struct f2fs_summary_block *sum = curseg->sum_blk;
234 nid_t start_nid = START_NID(nid);
235 struct f2fs_nat_block *nat_blk;
236 struct page *page = NULL;
237 struct f2fs_nat_entry ne;
241 memset(&ne, 0, sizeof(struct f2fs_nat_entry));
244 /* Check nat cache */
245 read_lock(&nm_i->nat_tree_lock);
246 e = __lookup_nat_cache(nm_i, nid);
248 ni->ino = nat_get_ino(e);
249 ni->blk_addr = nat_get_blkaddr(e);
250 ni->version = nat_get_version(e);
252 read_unlock(&nm_i->nat_tree_lock);
256 /* Check current segment summary */
257 mutex_lock(&curseg->curseg_mutex);
258 i = lookup_journal_in_cursum(sum, NAT_JOURNAL, nid, 0);
260 ne = nat_in_journal(sum, i);
261 node_info_from_raw_nat(ni, &ne);
263 mutex_unlock(&curseg->curseg_mutex);
267 /* Fill node_info from nat page */
268 page = get_current_nat_page(sbi, start_nid);
269 nat_blk = (struct f2fs_nat_block *)page_address(page);
270 ne = nat_blk->entries[nid - start_nid];
271 node_info_from_raw_nat(ni, &ne);
272 f2fs_put_page(page, 1);
274 /* cache nat entry */
275 cache_nat_entry(NM_I(sbi), nid, &ne);
279 * The maximum depth is four.
280 * Offset[0] will have raw inode offset.
282 static int get_node_path(struct f2fs_inode_info *fi, long block,
283 int offset[4], unsigned int noffset[4])
285 const long direct_index = ADDRS_PER_INODE(fi);
286 const long direct_blks = ADDRS_PER_BLOCK;
287 const long dptrs_per_blk = NIDS_PER_BLOCK;
288 const long indirect_blks = ADDRS_PER_BLOCK * NIDS_PER_BLOCK;
289 const long dindirect_blks = indirect_blks * NIDS_PER_BLOCK;
295 if (block < direct_index) {
299 block -= direct_index;
300 if (block < direct_blks) {
301 offset[n++] = NODE_DIR1_BLOCK;
307 block -= direct_blks;
308 if (block < direct_blks) {
309 offset[n++] = NODE_DIR2_BLOCK;
315 block -= direct_blks;
316 if (block < indirect_blks) {
317 offset[n++] = NODE_IND1_BLOCK;
319 offset[n++] = block / direct_blks;
320 noffset[n] = 4 + offset[n - 1];
321 offset[n] = block % direct_blks;
325 block -= indirect_blks;
326 if (block < indirect_blks) {
327 offset[n++] = NODE_IND2_BLOCK;
328 noffset[n] = 4 + dptrs_per_blk;
329 offset[n++] = block / direct_blks;
330 noffset[n] = 5 + dptrs_per_blk + offset[n - 1];
331 offset[n] = block % direct_blks;
335 block -= indirect_blks;
336 if (block < dindirect_blks) {
337 offset[n++] = NODE_DIND_BLOCK;
338 noffset[n] = 5 + (dptrs_per_blk * 2);
339 offset[n++] = block / indirect_blks;
340 noffset[n] = 6 + (dptrs_per_blk * 2) +
341 offset[n - 1] * (dptrs_per_blk + 1);
342 offset[n++] = (block / direct_blks) % dptrs_per_blk;
343 noffset[n] = 7 + (dptrs_per_blk * 2) +
344 offset[n - 2] * (dptrs_per_blk + 1) +
346 offset[n] = block % direct_blks;
357 * Caller should call f2fs_put_dnode(dn).
358 * Also, it should grab and release a rwsem by calling f2fs_lock_op() and
359 * f2fs_unlock_op() only if ro is not set RDONLY_NODE.
360 * In the case of RDONLY_NODE, we don't need to care about mutex.
362 int get_dnode_of_data(struct dnode_of_data *dn, pgoff_t index, int mode)
364 struct f2fs_sb_info *sbi = F2FS_SB(dn->inode->i_sb);
365 struct page *npage[4];
368 unsigned int noffset[4];
373 level = get_node_path(F2FS_I(dn->inode), index, offset, noffset);
375 nids[0] = dn->inode->i_ino;
376 npage[0] = dn->inode_page;
379 npage[0] = get_node_page(sbi, nids[0]);
380 if (IS_ERR(npage[0]))
381 return PTR_ERR(npage[0]);
385 nids[1] = get_nid(parent, offset[0], true);
386 dn->inode_page = npage[0];
387 dn->inode_page_locked = true;
389 /* get indirect or direct nodes */
390 for (i = 1; i <= level; i++) {
393 if (!nids[i] && mode == ALLOC_NODE) {
395 if (!alloc_nid(sbi, &(nids[i]))) {
401 npage[i] = new_node_page(dn, noffset[i], NULL);
402 if (IS_ERR(npage[i])) {
403 alloc_nid_failed(sbi, nids[i]);
404 err = PTR_ERR(npage[i]);
408 set_nid(parent, offset[i - 1], nids[i], i == 1);
409 alloc_nid_done(sbi, nids[i]);
411 } else if (mode == LOOKUP_NODE_RA && i == level && level > 1) {
412 npage[i] = get_node_page_ra(parent, offset[i - 1]);
413 if (IS_ERR(npage[i])) {
414 err = PTR_ERR(npage[i]);
420 dn->inode_page_locked = false;
423 f2fs_put_page(parent, 1);
427 npage[i] = get_node_page(sbi, nids[i]);
428 if (IS_ERR(npage[i])) {
429 err = PTR_ERR(npage[i]);
430 f2fs_put_page(npage[0], 0);
436 nids[i + 1] = get_nid(parent, offset[i], false);
439 dn->nid = nids[level];
440 dn->ofs_in_node = offset[level];
441 dn->node_page = npage[level];
442 dn->data_blkaddr = datablock_addr(dn->node_page, dn->ofs_in_node);
446 f2fs_put_page(parent, 1);
448 f2fs_put_page(npage[0], 0);
450 dn->inode_page = NULL;
451 dn->node_page = NULL;
455 static void truncate_node(struct dnode_of_data *dn)
457 struct f2fs_sb_info *sbi = F2FS_SB(dn->inode->i_sb);
460 get_node_info(sbi, dn->nid, &ni);
461 if (dn->inode->i_blocks == 0) {
462 f2fs_bug_on(ni.blk_addr != NULL_ADDR);
465 f2fs_bug_on(ni.blk_addr == NULL_ADDR);
467 /* Deallocate node address */
468 invalidate_blocks(sbi, ni.blk_addr);
469 dec_valid_node_count(sbi, dn->inode);
470 set_node_addr(sbi, &ni, NULL_ADDR);
472 if (dn->nid == dn->inode->i_ino) {
473 remove_orphan_inode(sbi, dn->nid);
474 dec_valid_inode_count(sbi);
479 clear_node_page_dirty(dn->node_page);
480 F2FS_SET_SB_DIRT(sbi);
482 f2fs_put_page(dn->node_page, 1);
484 invalidate_mapping_pages(NODE_MAPPING(sbi),
485 dn->node_page->index, dn->node_page->index);
487 dn->node_page = NULL;
488 trace_f2fs_truncate_node(dn->inode, dn->nid, ni.blk_addr);
491 static int truncate_dnode(struct dnode_of_data *dn)
493 struct f2fs_sb_info *sbi = F2FS_SB(dn->inode->i_sb);
499 /* get direct node */
500 page = get_node_page(sbi, dn->nid);
501 if (IS_ERR(page) && PTR_ERR(page) == -ENOENT)
503 else if (IS_ERR(page))
504 return PTR_ERR(page);
506 /* Make dnode_of_data for parameter */
507 dn->node_page = page;
509 truncate_data_blocks(dn);
514 static int truncate_nodes(struct dnode_of_data *dn, unsigned int nofs,
517 struct f2fs_sb_info *sbi = F2FS_SB(dn->inode->i_sb);
518 struct dnode_of_data rdn = *dn;
520 struct f2fs_node *rn;
522 unsigned int child_nofs;
527 return NIDS_PER_BLOCK + 1;
529 trace_f2fs_truncate_nodes_enter(dn->inode, dn->nid, dn->data_blkaddr);
531 page = get_node_page(sbi, dn->nid);
533 trace_f2fs_truncate_nodes_exit(dn->inode, PTR_ERR(page));
534 return PTR_ERR(page);
537 rn = F2FS_NODE(page);
539 for (i = ofs; i < NIDS_PER_BLOCK; i++, freed++) {
540 child_nid = le32_to_cpu(rn->in.nid[i]);
544 ret = truncate_dnode(&rdn);
547 set_nid(page, i, 0, false);
550 child_nofs = nofs + ofs * (NIDS_PER_BLOCK + 1) + 1;
551 for (i = ofs; i < NIDS_PER_BLOCK; i++) {
552 child_nid = le32_to_cpu(rn->in.nid[i]);
553 if (child_nid == 0) {
554 child_nofs += NIDS_PER_BLOCK + 1;
558 ret = truncate_nodes(&rdn, child_nofs, 0, depth - 1);
559 if (ret == (NIDS_PER_BLOCK + 1)) {
560 set_nid(page, i, 0, false);
562 } else if (ret < 0 && ret != -ENOENT) {
570 /* remove current indirect node */
571 dn->node_page = page;
575 f2fs_put_page(page, 1);
577 trace_f2fs_truncate_nodes_exit(dn->inode, freed);
581 f2fs_put_page(page, 1);
582 trace_f2fs_truncate_nodes_exit(dn->inode, ret);
586 static int truncate_partial_nodes(struct dnode_of_data *dn,
587 struct f2fs_inode *ri, int *offset, int depth)
589 struct f2fs_sb_info *sbi = F2FS_SB(dn->inode->i_sb);
590 struct page *pages[2];
597 nid[0] = le32_to_cpu(ri->i_nid[offset[0] - NODE_DIR1_BLOCK]);
601 /* get indirect nodes in the path */
602 for (i = 0; i < idx + 1; i++) {
603 /* refernece count'll be increased */
604 pages[i] = get_node_page(sbi, nid[i]);
605 if (IS_ERR(pages[i])) {
606 err = PTR_ERR(pages[i]);
610 nid[i + 1] = get_nid(pages[i], offset[i + 1], false);
613 /* free direct nodes linked to a partial indirect node */
614 for (i = offset[idx + 1]; i < NIDS_PER_BLOCK; i++) {
615 child_nid = get_nid(pages[idx], i, false);
619 err = truncate_dnode(dn);
622 set_nid(pages[idx], i, 0, false);
625 if (offset[idx + 1] == 0) {
626 dn->node_page = pages[idx];
630 f2fs_put_page(pages[idx], 1);
636 for (i = idx; i >= 0; i--)
637 f2fs_put_page(pages[i], 1);
639 trace_f2fs_truncate_partial_nodes(dn->inode, nid, depth, err);
645 * All the block addresses of data and nodes should be nullified.
647 int truncate_inode_blocks(struct inode *inode, pgoff_t from)
649 struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
650 int err = 0, cont = 1;
651 int level, offset[4], noffset[4];
652 unsigned int nofs = 0;
653 struct f2fs_inode *ri;
654 struct dnode_of_data dn;
657 trace_f2fs_truncate_inode_blocks_enter(inode, from);
659 level = get_node_path(F2FS_I(inode), from, offset, noffset);
661 page = get_node_page(sbi, inode->i_ino);
663 trace_f2fs_truncate_inode_blocks_exit(inode, PTR_ERR(page));
664 return PTR_ERR(page);
667 set_new_dnode(&dn, inode, page, NULL, 0);
670 ri = F2FS_INODE(page);
678 if (!offset[level - 1])
680 err = truncate_partial_nodes(&dn, ri, offset, level);
681 if (err < 0 && err != -ENOENT)
683 nofs += 1 + NIDS_PER_BLOCK;
686 nofs = 5 + 2 * NIDS_PER_BLOCK;
687 if (!offset[level - 1])
689 err = truncate_partial_nodes(&dn, ri, offset, level);
690 if (err < 0 && err != -ENOENT)
699 dn.nid = le32_to_cpu(ri->i_nid[offset[0] - NODE_DIR1_BLOCK]);
701 case NODE_DIR1_BLOCK:
702 case NODE_DIR2_BLOCK:
703 err = truncate_dnode(&dn);
706 case NODE_IND1_BLOCK:
707 case NODE_IND2_BLOCK:
708 err = truncate_nodes(&dn, nofs, offset[1], 2);
711 case NODE_DIND_BLOCK:
712 err = truncate_nodes(&dn, nofs, offset[1], 3);
719 if (err < 0 && err != -ENOENT)
721 if (offset[1] == 0 &&
722 ri->i_nid[offset[0] - NODE_DIR1_BLOCK]) {
724 if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
725 f2fs_put_page(page, 1);
728 wait_on_page_writeback(page);
729 ri->i_nid[offset[0] - NODE_DIR1_BLOCK] = 0;
730 set_page_dirty(page);
738 f2fs_put_page(page, 0);
739 trace_f2fs_truncate_inode_blocks_exit(inode, err);
740 return err > 0 ? 0 : err;
743 int truncate_xattr_node(struct inode *inode, struct page *page)
745 struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
746 nid_t nid = F2FS_I(inode)->i_xattr_nid;
747 struct dnode_of_data dn;
753 npage = get_node_page(sbi, nid);
755 return PTR_ERR(npage);
757 F2FS_I(inode)->i_xattr_nid = 0;
759 /* need to do checkpoint during fsync */
760 F2FS_I(inode)->xattr_ver = cur_cp_version(F2FS_CKPT(sbi));
762 set_new_dnode(&dn, inode, page, npage, nid);
765 dn.inode_page_locked = true;
771 * Caller should grab and release a rwsem by calling f2fs_lock_op() and
774 void remove_inode_page(struct inode *inode)
776 struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
778 nid_t ino = inode->i_ino;
779 struct dnode_of_data dn;
781 page = get_node_page(sbi, ino);
785 if (truncate_xattr_node(inode, page)) {
786 f2fs_put_page(page, 1);
789 /* 0 is possible, after f2fs_new_inode() is failed */
790 f2fs_bug_on(inode->i_blocks != 0 && inode->i_blocks != 1);
791 set_new_dnode(&dn, inode, page, page, ino);
795 struct page *new_inode_page(struct inode *inode, const struct qstr *name)
797 struct dnode_of_data dn;
799 /* allocate inode page for new inode */
800 set_new_dnode(&dn, inode, NULL, NULL, inode->i_ino);
802 /* caller should f2fs_put_page(page, 1); */
803 return new_node_page(&dn, 0, NULL);
806 struct page *new_node_page(struct dnode_of_data *dn,
807 unsigned int ofs, struct page *ipage)
809 struct f2fs_sb_info *sbi = F2FS_SB(dn->inode->i_sb);
810 struct node_info old_ni, new_ni;
814 if (unlikely(is_inode_flag_set(F2FS_I(dn->inode), FI_NO_ALLOC)))
815 return ERR_PTR(-EPERM);
817 page = grab_cache_page(NODE_MAPPING(sbi), dn->nid);
819 return ERR_PTR(-ENOMEM);
821 if (unlikely(!inc_valid_node_count(sbi, dn->inode))) {
826 get_node_info(sbi, dn->nid, &old_ni);
828 /* Reinitialize old_ni with new node page */
829 f2fs_bug_on(old_ni.blk_addr != NULL_ADDR);
831 new_ni.ino = dn->inode->i_ino;
832 set_node_addr(sbi, &new_ni, NEW_ADDR);
834 fill_node_footer(page, dn->nid, dn->inode->i_ino, ofs, true);
835 set_cold_node(dn->inode, page);
836 SetPageUptodate(page);
837 set_page_dirty(page);
839 if (ofs == XATTR_NODE_OFFSET)
840 F2FS_I(dn->inode)->i_xattr_nid = dn->nid;
842 dn->node_page = page;
844 update_inode(dn->inode, ipage);
848 inc_valid_inode_count(sbi);
853 clear_node_page_dirty(page);
854 f2fs_put_page(page, 1);
859 * Caller should do after getting the following values.
860 * 0: f2fs_put_page(page, 0)
861 * LOCKED_PAGE: f2fs_put_page(page, 1)
864 static int read_node_page(struct page *page, int rw)
866 struct f2fs_sb_info *sbi = F2FS_SB(page->mapping->host->i_sb);
869 get_node_info(sbi, page->index, &ni);
871 if (unlikely(ni.blk_addr == NULL_ADDR)) {
872 f2fs_put_page(page, 1);
876 if (PageUptodate(page))
879 return f2fs_submit_page_bio(sbi, page, ni.blk_addr, rw);
883 * Readahead a node page
885 void ra_node_page(struct f2fs_sb_info *sbi, nid_t nid)
890 apage = find_get_page(NODE_MAPPING(sbi), nid);
891 if (apage && PageUptodate(apage)) {
892 f2fs_put_page(apage, 0);
895 f2fs_put_page(apage, 0);
897 apage = grab_cache_page(NODE_MAPPING(sbi), nid);
901 err = read_node_page(apage, READA);
903 f2fs_put_page(apage, 0);
904 else if (err == LOCKED_PAGE)
905 f2fs_put_page(apage, 1);
908 struct page *get_node_page(struct f2fs_sb_info *sbi, pgoff_t nid)
913 page = grab_cache_page(NODE_MAPPING(sbi), nid);
915 return ERR_PTR(-ENOMEM);
917 err = read_node_page(page, READ_SYNC);
920 else if (err == LOCKED_PAGE)
924 if (unlikely(!PageUptodate(page))) {
925 f2fs_put_page(page, 1);
926 return ERR_PTR(-EIO);
928 if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
929 f2fs_put_page(page, 1);
933 f2fs_bug_on(nid != nid_of_node(page));
934 mark_page_accessed(page);
939 * Return a locked page for the desired node page.
940 * And, readahead MAX_RA_NODE number of node pages.
942 struct page *get_node_page_ra(struct page *parent, int start)
944 struct f2fs_sb_info *sbi = F2FS_SB(parent->mapping->host->i_sb);
945 struct blk_plug plug;
950 /* First, try getting the desired direct node. */
951 nid = get_nid(parent, start, false);
953 return ERR_PTR(-ENOENT);
955 page = grab_cache_page(NODE_MAPPING(sbi), nid);
957 return ERR_PTR(-ENOMEM);
959 err = read_node_page(page, READ_SYNC);
962 else if (err == LOCKED_PAGE)
965 blk_start_plug(&plug);
967 /* Then, try readahead for siblings of the desired node */
968 end = start + MAX_RA_NODE;
969 end = min(end, NIDS_PER_BLOCK);
970 for (i = start + 1; i < end; i++) {
971 nid = get_nid(parent, i, false);
974 ra_node_page(sbi, nid);
977 blk_finish_plug(&plug);
980 if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
981 f2fs_put_page(page, 1);
985 if (unlikely(!PageUptodate(page))) {
986 f2fs_put_page(page, 1);
987 return ERR_PTR(-EIO);
989 mark_page_accessed(page);
993 void sync_inode_page(struct dnode_of_data *dn)
995 if (IS_INODE(dn->node_page) || dn->inode_page == dn->node_page) {
996 update_inode(dn->inode, dn->node_page);
997 } else if (dn->inode_page) {
998 if (!dn->inode_page_locked)
999 lock_page(dn->inode_page);
1000 update_inode(dn->inode, dn->inode_page);
1001 if (!dn->inode_page_locked)
1002 unlock_page(dn->inode_page);
1004 update_inode_page(dn->inode);
1008 int sync_node_pages(struct f2fs_sb_info *sbi, nid_t ino,
1009 struct writeback_control *wbc)
1012 struct pagevec pvec;
1013 int step = ino ? 2 : 0;
1014 int nwritten = 0, wrote = 0;
1016 pagevec_init(&pvec, 0);
1022 while (index <= end) {
1024 nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1025 PAGECACHE_TAG_DIRTY,
1026 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
1030 for (i = 0; i < nr_pages; i++) {
1031 struct page *page = pvec.pages[i];
1034 * flushing sequence with step:
1039 if (step == 0 && IS_DNODE(page))
1041 if (step == 1 && (!IS_DNODE(page) ||
1042 is_cold_node(page)))
1044 if (step == 2 && (!IS_DNODE(page) ||
1045 !is_cold_node(page)))
1050 * we should not skip writing node pages.
1052 if (ino && ino_of_node(page) == ino)
1054 else if (!trylock_page(page))
1057 if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1062 if (ino && ino_of_node(page) != ino)
1063 goto continue_unlock;
1065 if (!PageDirty(page)) {
1066 /* someone wrote it for us */
1067 goto continue_unlock;
1070 if (!clear_page_dirty_for_io(page))
1071 goto continue_unlock;
1073 /* called by fsync() */
1074 if (ino && IS_DNODE(page)) {
1075 int mark = !is_checkpointed_node(sbi, ino);
1076 set_fsync_mark(page, 1);
1078 set_dentry_mark(page, mark);
1081 set_fsync_mark(page, 0);
1082 set_dentry_mark(page, 0);
1084 NODE_MAPPING(sbi)->a_ops->writepage(page, wbc);
1087 if (--wbc->nr_to_write == 0)
1090 pagevec_release(&pvec);
1093 if (wbc->nr_to_write == 0) {
1105 f2fs_submit_merged_bio(sbi, NODE, WRITE);
1109 int wait_on_node_pages_writeback(struct f2fs_sb_info *sbi, nid_t ino)
1111 pgoff_t index = 0, end = LONG_MAX;
1112 struct pagevec pvec;
1113 int ret2 = 0, ret = 0;
1115 pagevec_init(&pvec, 0);
1117 while (index <= end) {
1119 nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1120 PAGECACHE_TAG_WRITEBACK,
1121 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
1125 for (i = 0; i < nr_pages; i++) {
1126 struct page *page = pvec.pages[i];
1128 /* until radix tree lookup accepts end_index */
1129 if (unlikely(page->index > end))
1132 if (ino && ino_of_node(page) == ino) {
1133 wait_on_page_writeback(page);
1134 if (TestClearPageError(page))
1138 pagevec_release(&pvec);
1142 if (unlikely(test_and_clear_bit(AS_ENOSPC, &NODE_MAPPING(sbi)->flags)))
1144 if (unlikely(test_and_clear_bit(AS_EIO, &NODE_MAPPING(sbi)->flags)))
1151 static int f2fs_write_node_page(struct page *page,
1152 struct writeback_control *wbc)
1154 struct f2fs_sb_info *sbi = F2FS_SB(page->mapping->host->i_sb);
1157 struct node_info ni;
1158 struct f2fs_io_info fio = {
1160 .rw = (wbc->sync_mode == WB_SYNC_ALL) ? WRITE_SYNC : WRITE,
1163 if (unlikely(sbi->por_doing))
1166 wait_on_page_writeback(page);
1168 /* get old block addr of this node page */
1169 nid = nid_of_node(page);
1170 f2fs_bug_on(page->index != nid);
1172 get_node_info(sbi, nid, &ni);
1174 /* This page is already truncated */
1175 if (unlikely(ni.blk_addr == NULL_ADDR)) {
1176 dec_page_count(sbi, F2FS_DIRTY_NODES);
1181 if (wbc->for_reclaim)
1184 mutex_lock(&sbi->node_write);
1185 set_page_writeback(page);
1186 write_node_page(sbi, page, &fio, nid, ni.blk_addr, &new_addr);
1187 set_node_addr(sbi, &ni, new_addr);
1188 dec_page_count(sbi, F2FS_DIRTY_NODES);
1189 mutex_unlock(&sbi->node_write);
1194 dec_page_count(sbi, F2FS_DIRTY_NODES);
1195 wbc->pages_skipped++;
1196 account_page_redirty(page);
1197 set_page_dirty(page);
1198 return AOP_WRITEPAGE_ACTIVATE;
1202 * It is very important to gather dirty pages and write at once, so that we can
1203 * submit a big bio without interfering other data writes.
1204 * Be default, 512 pages (2MB) * 3 node types, is more reasonable.
1206 #define COLLECT_DIRTY_NODES 1536
1207 static int f2fs_write_node_pages(struct address_space *mapping,
1208 struct writeback_control *wbc)
1210 struct f2fs_sb_info *sbi = F2FS_SB(mapping->host->i_sb);
1211 long nr_to_write = wbc->nr_to_write;
1213 /* balancing f2fs's metadata in background */
1214 f2fs_balance_fs_bg(sbi);
1216 /* collect a number of dirty node pages and write together */
1217 if (get_pages(sbi, F2FS_DIRTY_NODES) < COLLECT_DIRTY_NODES)
1220 /* if mounting is failed, skip writing node pages */
1221 wbc->nr_to_write = 3 * max_hw_blocks(sbi);
1222 wbc->sync_mode = WB_SYNC_NONE;
1223 sync_node_pages(sbi, 0, wbc);
1224 wbc->nr_to_write = nr_to_write - (3 * max_hw_blocks(sbi) -
1229 static int f2fs_set_node_page_dirty(struct page *page)
1231 struct address_space *mapping = page->mapping;
1232 struct f2fs_sb_info *sbi = F2FS_SB(mapping->host->i_sb);
1234 trace_f2fs_set_page_dirty(page, NODE);
1236 SetPageUptodate(page);
1237 if (!PageDirty(page)) {
1238 __set_page_dirty_nobuffers(page);
1239 inc_page_count(sbi, F2FS_DIRTY_NODES);
1240 SetPagePrivate(page);
1246 static void f2fs_invalidate_node_page(struct page *page, unsigned int offset,
1247 unsigned int length)
1249 struct inode *inode = page->mapping->host;
1250 struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
1251 if (PageDirty(page))
1252 dec_page_count(sbi, F2FS_DIRTY_NODES);
1253 ClearPagePrivate(page);
1256 static int f2fs_release_node_page(struct page *page, gfp_t wait)
1258 ClearPagePrivate(page);
1263 * Structure of the f2fs node operations
1265 const struct address_space_operations f2fs_node_aops = {
1266 .writepage = f2fs_write_node_page,
1267 .writepages = f2fs_write_node_pages,
1268 .set_page_dirty = f2fs_set_node_page_dirty,
1269 .invalidatepage = f2fs_invalidate_node_page,
1270 .releasepage = f2fs_release_node_page,
1273 static struct free_nid *__lookup_free_nid_list(struct f2fs_nm_info *nm_i,
1276 return radix_tree_lookup(&nm_i->free_nid_root, n);
1279 static void __del_from_free_nid_list(struct f2fs_nm_info *nm_i,
1283 radix_tree_delete(&nm_i->free_nid_root, i->nid);
1284 kmem_cache_free(free_nid_slab, i);
1287 static int add_free_nid(struct f2fs_nm_info *nm_i, nid_t nid, bool build)
1290 struct nat_entry *ne;
1291 bool allocated = false;
1293 if (nm_i->fcnt > 2 * MAX_FREE_NIDS)
1296 /* 0 nid should not be used */
1297 if (unlikely(nid == 0))
1301 /* do not add allocated nids */
1302 read_lock(&nm_i->nat_tree_lock);
1303 ne = __lookup_nat_cache(nm_i, nid);
1305 (!ne->checkpointed || nat_get_blkaddr(ne) != NULL_ADDR))
1307 read_unlock(&nm_i->nat_tree_lock);
1312 i = f2fs_kmem_cache_alloc(free_nid_slab, GFP_NOFS);
1316 spin_lock(&nm_i->free_nid_list_lock);
1317 if (radix_tree_insert(&nm_i->free_nid_root, i->nid, i)) {
1318 spin_unlock(&nm_i->free_nid_list_lock);
1319 kmem_cache_free(free_nid_slab, i);
1322 list_add_tail(&i->list, &nm_i->free_nid_list);
1324 spin_unlock(&nm_i->free_nid_list_lock);
1328 static void remove_free_nid(struct f2fs_nm_info *nm_i, nid_t nid)
1331 spin_lock(&nm_i->free_nid_list_lock);
1332 i = __lookup_free_nid_list(nm_i, nid);
1333 if (i && i->state == NID_NEW) {
1334 __del_from_free_nid_list(nm_i, i);
1337 spin_unlock(&nm_i->free_nid_list_lock);
1340 static void scan_nat_page(struct f2fs_nm_info *nm_i,
1341 struct page *nat_page, nid_t start_nid)
1343 struct f2fs_nat_block *nat_blk = page_address(nat_page);
1347 i = start_nid % NAT_ENTRY_PER_BLOCK;
1349 for (; i < NAT_ENTRY_PER_BLOCK; i++, start_nid++) {
1351 if (unlikely(start_nid >= nm_i->max_nid))
1354 blk_addr = le32_to_cpu(nat_blk->entries[i].block_addr);
1355 f2fs_bug_on(blk_addr == NEW_ADDR);
1356 if (blk_addr == NULL_ADDR) {
1357 if (add_free_nid(nm_i, start_nid, true) < 0)
1363 static void build_free_nids(struct f2fs_sb_info *sbi)
1365 struct f2fs_nm_info *nm_i = NM_I(sbi);
1366 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
1367 struct f2fs_summary_block *sum = curseg->sum_blk;
1369 nid_t nid = nm_i->next_scan_nid;
1371 /* Enough entries */
1372 if (nm_i->fcnt > NAT_ENTRY_PER_BLOCK)
1375 /* readahead nat pages to be scanned */
1376 ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nid), FREE_NID_PAGES, META_NAT);
1379 struct page *page = get_current_nat_page(sbi, nid);
1381 scan_nat_page(nm_i, page, nid);
1382 f2fs_put_page(page, 1);
1384 nid += (NAT_ENTRY_PER_BLOCK - (nid % NAT_ENTRY_PER_BLOCK));
1385 if (unlikely(nid >= nm_i->max_nid))
1388 if (i++ == FREE_NID_PAGES)
1392 /* go to the next free nat pages to find free nids abundantly */
1393 nm_i->next_scan_nid = nid;
1395 /* find free nids from current sum_pages */
1396 mutex_lock(&curseg->curseg_mutex);
1397 for (i = 0; i < nats_in_cursum(sum); i++) {
1398 block_t addr = le32_to_cpu(nat_in_journal(sum, i).block_addr);
1399 nid = le32_to_cpu(nid_in_journal(sum, i));
1400 if (addr == NULL_ADDR)
1401 add_free_nid(nm_i, nid, true);
1403 remove_free_nid(nm_i, nid);
1405 mutex_unlock(&curseg->curseg_mutex);
1409 * If this function returns success, caller can obtain a new nid
1410 * from second parameter of this function.
1411 * The returned nid could be used ino as well as nid when inode is created.
1413 bool alloc_nid(struct f2fs_sb_info *sbi, nid_t *nid)
1415 struct f2fs_nm_info *nm_i = NM_I(sbi);
1416 struct free_nid *i = NULL;
1417 struct list_head *this;
1419 if (unlikely(sbi->total_valid_node_count + 1 >= nm_i->max_nid))
1422 spin_lock(&nm_i->free_nid_list_lock);
1424 /* We should not use stale free nids created by build_free_nids */
1425 if (nm_i->fcnt && !on_build_free_nids(nm_i)) {
1426 f2fs_bug_on(list_empty(&nm_i->free_nid_list));
1427 list_for_each(this, &nm_i->free_nid_list) {
1428 i = list_entry(this, struct free_nid, list);
1429 if (i->state == NID_NEW)
1433 f2fs_bug_on(i->state != NID_NEW);
1435 i->state = NID_ALLOC;
1437 spin_unlock(&nm_i->free_nid_list_lock);
1440 spin_unlock(&nm_i->free_nid_list_lock);
1442 /* Let's scan nat pages and its caches to get free nids */
1443 mutex_lock(&nm_i->build_lock);
1444 build_free_nids(sbi);
1445 mutex_unlock(&nm_i->build_lock);
1450 * alloc_nid() should be called prior to this function.
1452 void alloc_nid_done(struct f2fs_sb_info *sbi, nid_t nid)
1454 struct f2fs_nm_info *nm_i = NM_I(sbi);
1457 spin_lock(&nm_i->free_nid_list_lock);
1458 i = __lookup_free_nid_list(nm_i, nid);
1459 f2fs_bug_on(!i || i->state != NID_ALLOC);
1460 __del_from_free_nid_list(nm_i, i);
1461 spin_unlock(&nm_i->free_nid_list_lock);
1465 * alloc_nid() should be called prior to this function.
1467 void alloc_nid_failed(struct f2fs_sb_info *sbi, nid_t nid)
1469 struct f2fs_nm_info *nm_i = NM_I(sbi);
1475 spin_lock(&nm_i->free_nid_list_lock);
1476 i = __lookup_free_nid_list(nm_i, nid);
1477 f2fs_bug_on(!i || i->state != NID_ALLOC);
1478 if (nm_i->fcnt > 2 * MAX_FREE_NIDS) {
1479 __del_from_free_nid_list(nm_i, i);
1484 spin_unlock(&nm_i->free_nid_list_lock);
1487 void recover_node_page(struct f2fs_sb_info *sbi, struct page *page,
1488 struct f2fs_summary *sum, struct node_info *ni,
1489 block_t new_blkaddr)
1491 rewrite_node_page(sbi, page, sum, ni->blk_addr, new_blkaddr);
1492 set_node_addr(sbi, ni, new_blkaddr);
1493 clear_node_page_dirty(page);
1496 bool recover_xattr_data(struct inode *inode, struct page *page, block_t blkaddr)
1498 struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
1499 nid_t prev_xnid = F2FS_I(inode)->i_xattr_nid;
1500 nid_t new_xnid = nid_of_node(page);
1501 struct node_info ni;
1503 if (ofs_of_node(page) != XATTR_NODE_OFFSET)
1506 /* 1: invalidate the previous xattr nid */
1510 /* Deallocate node address */
1511 get_node_info(sbi, prev_xnid, &ni);
1512 f2fs_bug_on(ni.blk_addr == NULL_ADDR);
1513 invalidate_blocks(sbi, ni.blk_addr);
1514 dec_valid_node_count(sbi, inode);
1515 set_node_addr(sbi, &ni, NULL_ADDR);
1518 /* 2: allocate new xattr nid */
1519 if (unlikely(!inc_valid_node_count(sbi, inode)))
1522 remove_free_nid(NM_I(sbi), new_xnid);
1523 get_node_info(sbi, new_xnid, &ni);
1524 ni.ino = inode->i_ino;
1525 set_node_addr(sbi, &ni, NEW_ADDR);
1526 F2FS_I(inode)->i_xattr_nid = new_xnid;
1528 /* 3: update xattr blkaddr */
1529 refresh_sit_entry(sbi, NEW_ADDR, blkaddr);
1530 set_node_addr(sbi, &ni, blkaddr);
1532 update_inode_page(inode);
1536 int recover_inode_page(struct f2fs_sb_info *sbi, struct page *page)
1538 struct f2fs_inode *src, *dst;
1539 nid_t ino = ino_of_node(page);
1540 struct node_info old_ni, new_ni;
1543 ipage = grab_cache_page(NODE_MAPPING(sbi), ino);
1547 /* Should not use this inode from free nid list */
1548 remove_free_nid(NM_I(sbi), ino);
1550 get_node_info(sbi, ino, &old_ni);
1551 SetPageUptodate(ipage);
1552 fill_node_footer(ipage, ino, ino, 0, true);
1554 src = F2FS_INODE(page);
1555 dst = F2FS_INODE(ipage);
1557 memcpy(dst, src, (unsigned long)&src->i_ext - (unsigned long)src);
1559 dst->i_blocks = cpu_to_le64(1);
1560 dst->i_links = cpu_to_le32(1);
1561 dst->i_xattr_nid = 0;
1566 if (unlikely(!inc_valid_node_count(sbi, NULL)))
1568 set_node_addr(sbi, &new_ni, NEW_ADDR);
1569 inc_valid_inode_count(sbi);
1570 f2fs_put_page(ipage, 1);
1575 * ra_sum_pages() merge contiguous pages into one bio and submit.
1576 * these pre-readed pages are linked in pages list.
1578 static int ra_sum_pages(struct f2fs_sb_info *sbi, struct list_head *pages,
1579 int start, int nrpages)
1582 int page_idx = start;
1583 struct f2fs_io_info fio = {
1585 .rw = READ_SYNC | REQ_META | REQ_PRIO
1588 for (; page_idx < start + nrpages; page_idx++) {
1589 /* alloc temporal page for read node summary info*/
1590 page = alloc_page(GFP_F2FS_ZERO);
1595 page->index = page_idx;
1596 list_add_tail(&page->lru, pages);
1599 list_for_each_entry(page, pages, lru)
1600 f2fs_submit_page_mbio(sbi, page, page->index, &fio);
1602 f2fs_submit_merged_bio(sbi, META, READ);
1604 return page_idx - start;
1607 int restore_node_summary(struct f2fs_sb_info *sbi,
1608 unsigned int segno, struct f2fs_summary_block *sum)
1610 struct f2fs_node *rn;
1611 struct f2fs_summary *sum_entry;
1612 struct page *page, *tmp;
1614 int bio_blocks = MAX_BIO_BLOCKS(max_hw_blocks(sbi));
1615 int i, last_offset, nrpages, err = 0;
1616 LIST_HEAD(page_list);
1618 /* scan the node segment */
1619 last_offset = sbi->blocks_per_seg;
1620 addr = START_BLOCK(sbi, segno);
1621 sum_entry = &sum->entries[0];
1623 for (i = 0; !err && i < last_offset; i += nrpages, addr += nrpages) {
1624 nrpages = min(last_offset - i, bio_blocks);
1626 /* read ahead node pages */
1627 nrpages = ra_sum_pages(sbi, &page_list, addr, nrpages);
1631 list_for_each_entry_safe(page, tmp, &page_list, lru) {
1636 if (unlikely(!PageUptodate(page))) {
1639 rn = F2FS_NODE(page);
1640 sum_entry->nid = rn->footer.nid;
1641 sum_entry->version = 0;
1642 sum_entry->ofs_in_node = 0;
1647 list_del(&page->lru);
1648 __free_pages(page, 0);
1654 static bool flush_nats_in_journal(struct f2fs_sb_info *sbi)
1656 struct f2fs_nm_info *nm_i = NM_I(sbi);
1657 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
1658 struct f2fs_summary_block *sum = curseg->sum_blk;
1661 mutex_lock(&curseg->curseg_mutex);
1663 if (nats_in_cursum(sum) < NAT_JOURNAL_ENTRIES) {
1664 mutex_unlock(&curseg->curseg_mutex);
1668 for (i = 0; i < nats_in_cursum(sum); i++) {
1669 struct nat_entry *ne;
1670 struct f2fs_nat_entry raw_ne;
1671 nid_t nid = le32_to_cpu(nid_in_journal(sum, i));
1673 raw_ne = nat_in_journal(sum, i);
1675 write_lock(&nm_i->nat_tree_lock);
1676 ne = __lookup_nat_cache(nm_i, nid);
1678 __set_nat_cache_dirty(nm_i, ne);
1679 write_unlock(&nm_i->nat_tree_lock);
1682 ne = grab_nat_entry(nm_i, nid);
1684 write_unlock(&nm_i->nat_tree_lock);
1687 nat_set_blkaddr(ne, le32_to_cpu(raw_ne.block_addr));
1688 nat_set_ino(ne, le32_to_cpu(raw_ne.ino));
1689 nat_set_version(ne, raw_ne.version);
1690 __set_nat_cache_dirty(nm_i, ne);
1691 write_unlock(&nm_i->nat_tree_lock);
1693 update_nats_in_cursum(sum, -i);
1694 mutex_unlock(&curseg->curseg_mutex);
1699 * This function is called during the checkpointing process.
1701 void flush_nat_entries(struct f2fs_sb_info *sbi)
1703 struct f2fs_nm_info *nm_i = NM_I(sbi);
1704 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
1705 struct f2fs_summary_block *sum = curseg->sum_blk;
1706 struct list_head *cur, *n;
1707 struct page *page = NULL;
1708 struct f2fs_nat_block *nat_blk = NULL;
1709 nid_t start_nid = 0, end_nid = 0;
1712 flushed = flush_nats_in_journal(sbi);
1715 mutex_lock(&curseg->curseg_mutex);
1717 /* 1) flush dirty nat caches */
1718 list_for_each_safe(cur, n, &nm_i->dirty_nat_entries) {
1719 struct nat_entry *ne;
1721 struct f2fs_nat_entry raw_ne;
1723 block_t new_blkaddr;
1725 ne = list_entry(cur, struct nat_entry, list);
1726 nid = nat_get_nid(ne);
1728 if (nat_get_blkaddr(ne) == NEW_ADDR)
1733 /* if there is room for nat enries in curseg->sumpage */
1734 offset = lookup_journal_in_cursum(sum, NAT_JOURNAL, nid, 1);
1736 raw_ne = nat_in_journal(sum, offset);
1740 if (!page || (start_nid > nid || nid > end_nid)) {
1742 f2fs_put_page(page, 1);
1745 start_nid = START_NID(nid);
1746 end_nid = start_nid + NAT_ENTRY_PER_BLOCK - 1;
1749 * get nat block with dirty flag, increased reference
1750 * count, mapped and lock
1752 page = get_next_nat_page(sbi, start_nid);
1753 nat_blk = page_address(page);
1756 f2fs_bug_on(!nat_blk);
1757 raw_ne = nat_blk->entries[nid - start_nid];
1759 new_blkaddr = nat_get_blkaddr(ne);
1761 raw_ne.ino = cpu_to_le32(nat_get_ino(ne));
1762 raw_ne.block_addr = cpu_to_le32(new_blkaddr);
1763 raw_ne.version = nat_get_version(ne);
1766 nat_blk->entries[nid - start_nid] = raw_ne;
1768 nat_in_journal(sum, offset) = raw_ne;
1769 nid_in_journal(sum, offset) = cpu_to_le32(nid);
1772 if (nat_get_blkaddr(ne) == NULL_ADDR &&
1773 add_free_nid(NM_I(sbi), nid, false) <= 0) {
1774 write_lock(&nm_i->nat_tree_lock);
1775 __del_from_nat_cache(nm_i, ne);
1776 write_unlock(&nm_i->nat_tree_lock);
1778 write_lock(&nm_i->nat_tree_lock);
1779 __clear_nat_cache_dirty(nm_i, ne);
1780 write_unlock(&nm_i->nat_tree_lock);
1784 mutex_unlock(&curseg->curseg_mutex);
1785 f2fs_put_page(page, 1);
1787 /* 2) shrink nat caches if necessary */
1788 try_to_free_nats(sbi, nm_i->nat_cnt - NM_WOUT_THRESHOLD);
1791 static int init_node_manager(struct f2fs_sb_info *sbi)
1793 struct f2fs_super_block *sb_raw = F2FS_RAW_SUPER(sbi);
1794 struct f2fs_nm_info *nm_i = NM_I(sbi);
1795 unsigned char *version_bitmap;
1796 unsigned int nat_segs, nat_blocks;
1798 nm_i->nat_blkaddr = le32_to_cpu(sb_raw->nat_blkaddr);
1800 /* segment_count_nat includes pair segment so divide to 2. */
1801 nat_segs = le32_to_cpu(sb_raw->segment_count_nat) >> 1;
1802 nat_blocks = nat_segs << le32_to_cpu(sb_raw->log_blocks_per_seg);
1804 /* not used nids: 0, node, meta, (and root counted as valid node) */
1805 nm_i->max_nid = NAT_ENTRY_PER_BLOCK * nat_blocks - 3;
1809 INIT_RADIX_TREE(&nm_i->free_nid_root, GFP_ATOMIC);
1810 INIT_LIST_HEAD(&nm_i->free_nid_list);
1811 INIT_RADIX_TREE(&nm_i->nat_root, GFP_ATOMIC);
1812 INIT_LIST_HEAD(&nm_i->nat_entries);
1813 INIT_LIST_HEAD(&nm_i->dirty_nat_entries);
1815 mutex_init(&nm_i->build_lock);
1816 spin_lock_init(&nm_i->free_nid_list_lock);
1817 rwlock_init(&nm_i->nat_tree_lock);
1819 nm_i->next_scan_nid = le32_to_cpu(sbi->ckpt->next_free_nid);
1820 nm_i->bitmap_size = __bitmap_size(sbi, NAT_BITMAP);
1821 version_bitmap = __bitmap_ptr(sbi, NAT_BITMAP);
1822 if (!version_bitmap)
1825 nm_i->nat_bitmap = kmemdup(version_bitmap, nm_i->bitmap_size,
1827 if (!nm_i->nat_bitmap)
1832 int build_node_manager(struct f2fs_sb_info *sbi)
1836 sbi->nm_info = kzalloc(sizeof(struct f2fs_nm_info), GFP_KERNEL);
1840 err = init_node_manager(sbi);
1844 build_free_nids(sbi);
1848 void destroy_node_manager(struct f2fs_sb_info *sbi)
1850 struct f2fs_nm_info *nm_i = NM_I(sbi);
1851 struct free_nid *i, *next_i;
1852 struct nat_entry *natvec[NATVEC_SIZE];
1859 /* destroy free nid list */
1860 spin_lock(&nm_i->free_nid_list_lock);
1861 list_for_each_entry_safe(i, next_i, &nm_i->free_nid_list, list) {
1862 f2fs_bug_on(i->state == NID_ALLOC);
1863 __del_from_free_nid_list(nm_i, i);
1866 f2fs_bug_on(nm_i->fcnt);
1867 spin_unlock(&nm_i->free_nid_list_lock);
1869 /* destroy nat cache */
1870 write_lock(&nm_i->nat_tree_lock);
1871 while ((found = __gang_lookup_nat_cache(nm_i,
1872 nid, NATVEC_SIZE, natvec))) {
1874 nid = nat_get_nid(natvec[found - 1]) + 1;
1875 for (idx = 0; idx < found; idx++)
1876 __del_from_nat_cache(nm_i, natvec[idx]);
1878 f2fs_bug_on(nm_i->nat_cnt);
1879 write_unlock(&nm_i->nat_tree_lock);
1881 kfree(nm_i->nat_bitmap);
1882 sbi->nm_info = NULL;
1886 int __init create_node_manager_caches(void)
1888 nat_entry_slab = f2fs_kmem_cache_create("nat_entry",
1889 sizeof(struct nat_entry));
1890 if (!nat_entry_slab)
1893 free_nid_slab = f2fs_kmem_cache_create("free_nid",
1894 sizeof(struct free_nid));
1895 if (!free_nid_slab) {
1896 kmem_cache_destroy(nat_entry_slab);
1902 void destroy_node_manager_caches(void)
1904 kmem_cache_destroy(free_nid_slab);
1905 kmem_cache_destroy(nat_entry_slab);