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 inline bool available_free_memory(struct f2fs_nm_info *nm_i, int type)
32 unsigned long mem_size = 0;
35 if (type == FREE_NIDS)
36 mem_size = nm_i->fcnt * sizeof(struct free_nid);
37 else if (type == NAT_ENTRIES)
38 mem_size += nm_i->nat_cnt * sizeof(struct nat_entry);
41 /* give 50:50 memory for free nids and nat caches respectively */
42 return (mem_size < ((val.totalram * nm_i->ram_thresh) >> 11));
45 static void clear_node_page_dirty(struct page *page)
47 struct address_space *mapping = page->mapping;
48 struct f2fs_sb_info *sbi = F2FS_SB(mapping->host->i_sb);
49 unsigned int long flags;
51 if (PageDirty(page)) {
52 spin_lock_irqsave(&mapping->tree_lock, flags);
53 radix_tree_tag_clear(&mapping->page_tree,
56 spin_unlock_irqrestore(&mapping->tree_lock, flags);
58 clear_page_dirty_for_io(page);
59 dec_page_count(sbi, F2FS_DIRTY_NODES);
61 ClearPageUptodate(page);
64 static struct page *get_current_nat_page(struct f2fs_sb_info *sbi, nid_t nid)
66 pgoff_t index = current_nat_addr(sbi, nid);
67 return get_meta_page(sbi, index);
70 static struct page *get_next_nat_page(struct f2fs_sb_info *sbi, nid_t nid)
72 struct page *src_page;
73 struct page *dst_page;
78 struct f2fs_nm_info *nm_i = NM_I(sbi);
80 src_off = current_nat_addr(sbi, nid);
81 dst_off = next_nat_addr(sbi, src_off);
83 /* get current nat block page with lock */
84 src_page = get_meta_page(sbi, src_off);
86 /* Dirty src_page means that it is already the new target NAT page. */
87 if (PageDirty(src_page))
90 dst_page = grab_meta_page(sbi, dst_off);
92 src_addr = page_address(src_page);
93 dst_addr = page_address(dst_page);
94 memcpy(dst_addr, src_addr, PAGE_CACHE_SIZE);
95 set_page_dirty(dst_page);
96 f2fs_put_page(src_page, 1);
98 set_to_next_nat(nm_i, nid);
103 static struct nat_entry *__lookup_nat_cache(struct f2fs_nm_info *nm_i, nid_t n)
105 return radix_tree_lookup(&nm_i->nat_root, n);
108 static unsigned int __gang_lookup_nat_cache(struct f2fs_nm_info *nm_i,
109 nid_t start, unsigned int nr, struct nat_entry **ep)
111 return radix_tree_gang_lookup(&nm_i->nat_root, (void **)ep, start, nr);
114 static void __del_from_nat_cache(struct f2fs_nm_info *nm_i, struct nat_entry *e)
117 radix_tree_delete(&nm_i->nat_root, nat_get_nid(e));
119 kmem_cache_free(nat_entry_slab, e);
122 int is_checkpointed_node(struct f2fs_sb_info *sbi, nid_t nid)
124 struct f2fs_nm_info *nm_i = NM_I(sbi);
128 read_lock(&nm_i->nat_tree_lock);
129 e = __lookup_nat_cache(nm_i, nid);
130 if (e && !e->checkpointed)
132 read_unlock(&nm_i->nat_tree_lock);
136 bool fsync_mark_done(struct f2fs_sb_info *sbi, nid_t nid)
138 struct f2fs_nm_info *nm_i = NM_I(sbi);
140 bool fsync_done = false;
142 read_lock(&nm_i->nat_tree_lock);
143 e = __lookup_nat_cache(nm_i, nid);
145 fsync_done = e->fsync_done;
146 read_unlock(&nm_i->nat_tree_lock);
150 static struct nat_entry *grab_nat_entry(struct f2fs_nm_info *nm_i, nid_t nid)
152 struct nat_entry *new;
154 new = kmem_cache_alloc(nat_entry_slab, GFP_ATOMIC);
157 if (radix_tree_insert(&nm_i->nat_root, nid, new)) {
158 kmem_cache_free(nat_entry_slab, new);
161 memset(new, 0, sizeof(struct nat_entry));
162 nat_set_nid(new, nid);
163 new->checkpointed = true;
164 list_add_tail(&new->list, &nm_i->nat_entries);
169 static void cache_nat_entry(struct f2fs_nm_info *nm_i, nid_t nid,
170 struct f2fs_nat_entry *ne)
174 write_lock(&nm_i->nat_tree_lock);
175 e = __lookup_nat_cache(nm_i, nid);
177 e = grab_nat_entry(nm_i, nid);
179 write_unlock(&nm_i->nat_tree_lock);
182 nat_set_blkaddr(e, le32_to_cpu(ne->block_addr));
183 nat_set_ino(e, le32_to_cpu(ne->ino));
184 nat_set_version(e, ne->version);
186 write_unlock(&nm_i->nat_tree_lock);
189 static void set_node_addr(struct f2fs_sb_info *sbi, struct node_info *ni,
190 block_t new_blkaddr, bool fsync_done)
192 struct f2fs_nm_info *nm_i = NM_I(sbi);
195 write_lock(&nm_i->nat_tree_lock);
196 e = __lookup_nat_cache(nm_i, ni->nid);
198 e = grab_nat_entry(nm_i, ni->nid);
200 write_unlock(&nm_i->nat_tree_lock);
204 f2fs_bug_on(ni->blk_addr == NEW_ADDR);
205 } else if (new_blkaddr == NEW_ADDR) {
207 * when nid is reallocated,
208 * previous nat entry can be remained in nat cache.
209 * So, reinitialize it with new information.
212 f2fs_bug_on(ni->blk_addr != NULL_ADDR);
216 f2fs_bug_on(nat_get_blkaddr(e) != ni->blk_addr);
217 f2fs_bug_on(nat_get_blkaddr(e) == NULL_ADDR &&
218 new_blkaddr == NULL_ADDR);
219 f2fs_bug_on(nat_get_blkaddr(e) == NEW_ADDR &&
220 new_blkaddr == NEW_ADDR);
221 f2fs_bug_on(nat_get_blkaddr(e) != NEW_ADDR &&
222 nat_get_blkaddr(e) != NULL_ADDR &&
223 new_blkaddr == NEW_ADDR);
225 /* increament version no as node is removed */
226 if (nat_get_blkaddr(e) != NEW_ADDR && new_blkaddr == NULL_ADDR) {
227 unsigned char version = nat_get_version(e);
228 nat_set_version(e, inc_node_version(version));
232 nat_set_blkaddr(e, new_blkaddr);
233 __set_nat_cache_dirty(nm_i, e);
235 /* update fsync_mark if its inode nat entry is still alive */
236 e = __lookup_nat_cache(nm_i, ni->ino);
238 e->fsync_done = fsync_done;
239 write_unlock(&nm_i->nat_tree_lock);
242 int try_to_free_nats(struct f2fs_sb_info *sbi, int nr_shrink)
244 struct f2fs_nm_info *nm_i = NM_I(sbi);
246 if (available_free_memory(nm_i, NAT_ENTRIES))
249 write_lock(&nm_i->nat_tree_lock);
250 while (nr_shrink && !list_empty(&nm_i->nat_entries)) {
251 struct nat_entry *ne;
252 ne = list_first_entry(&nm_i->nat_entries,
253 struct nat_entry, list);
254 __del_from_nat_cache(nm_i, ne);
257 write_unlock(&nm_i->nat_tree_lock);
262 * This function returns always success
264 void get_node_info(struct f2fs_sb_info *sbi, nid_t nid, struct node_info *ni)
266 struct f2fs_nm_info *nm_i = NM_I(sbi);
267 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
268 struct f2fs_summary_block *sum = curseg->sum_blk;
269 nid_t start_nid = START_NID(nid);
270 struct f2fs_nat_block *nat_blk;
271 struct page *page = NULL;
272 struct f2fs_nat_entry ne;
276 memset(&ne, 0, sizeof(struct f2fs_nat_entry));
279 /* Check nat cache */
280 read_lock(&nm_i->nat_tree_lock);
281 e = __lookup_nat_cache(nm_i, nid);
283 ni->ino = nat_get_ino(e);
284 ni->blk_addr = nat_get_blkaddr(e);
285 ni->version = nat_get_version(e);
287 read_unlock(&nm_i->nat_tree_lock);
291 /* Check current segment summary */
292 mutex_lock(&curseg->curseg_mutex);
293 i = lookup_journal_in_cursum(sum, NAT_JOURNAL, nid, 0);
295 ne = nat_in_journal(sum, i);
296 node_info_from_raw_nat(ni, &ne);
298 mutex_unlock(&curseg->curseg_mutex);
302 /* Fill node_info from nat page */
303 page = get_current_nat_page(sbi, start_nid);
304 nat_blk = (struct f2fs_nat_block *)page_address(page);
305 ne = nat_blk->entries[nid - start_nid];
306 node_info_from_raw_nat(ni, &ne);
307 f2fs_put_page(page, 1);
309 /* cache nat entry */
310 cache_nat_entry(NM_I(sbi), nid, &ne);
314 * The maximum depth is four.
315 * Offset[0] will have raw inode offset.
317 static int get_node_path(struct f2fs_inode_info *fi, long block,
318 int offset[4], unsigned int noffset[4])
320 const long direct_index = ADDRS_PER_INODE(fi);
321 const long direct_blks = ADDRS_PER_BLOCK;
322 const long dptrs_per_blk = NIDS_PER_BLOCK;
323 const long indirect_blks = ADDRS_PER_BLOCK * NIDS_PER_BLOCK;
324 const long dindirect_blks = indirect_blks * NIDS_PER_BLOCK;
330 if (block < direct_index) {
334 block -= direct_index;
335 if (block < direct_blks) {
336 offset[n++] = NODE_DIR1_BLOCK;
342 block -= direct_blks;
343 if (block < direct_blks) {
344 offset[n++] = NODE_DIR2_BLOCK;
350 block -= direct_blks;
351 if (block < indirect_blks) {
352 offset[n++] = NODE_IND1_BLOCK;
354 offset[n++] = block / direct_blks;
355 noffset[n] = 4 + offset[n - 1];
356 offset[n] = block % direct_blks;
360 block -= indirect_blks;
361 if (block < indirect_blks) {
362 offset[n++] = NODE_IND2_BLOCK;
363 noffset[n] = 4 + dptrs_per_blk;
364 offset[n++] = block / direct_blks;
365 noffset[n] = 5 + dptrs_per_blk + offset[n - 1];
366 offset[n] = block % direct_blks;
370 block -= indirect_blks;
371 if (block < dindirect_blks) {
372 offset[n++] = NODE_DIND_BLOCK;
373 noffset[n] = 5 + (dptrs_per_blk * 2);
374 offset[n++] = block / indirect_blks;
375 noffset[n] = 6 + (dptrs_per_blk * 2) +
376 offset[n - 1] * (dptrs_per_blk + 1);
377 offset[n++] = (block / direct_blks) % dptrs_per_blk;
378 noffset[n] = 7 + (dptrs_per_blk * 2) +
379 offset[n - 2] * (dptrs_per_blk + 1) +
381 offset[n] = block % direct_blks;
392 * Caller should call f2fs_put_dnode(dn).
393 * Also, it should grab and release a rwsem by calling f2fs_lock_op() and
394 * f2fs_unlock_op() only if ro is not set RDONLY_NODE.
395 * In the case of RDONLY_NODE, we don't need to care about mutex.
397 int get_dnode_of_data(struct dnode_of_data *dn, pgoff_t index, int mode)
399 struct f2fs_sb_info *sbi = F2FS_SB(dn->inode->i_sb);
400 struct page *npage[4];
403 unsigned int noffset[4];
408 level = get_node_path(F2FS_I(dn->inode), index, offset, noffset);
410 nids[0] = dn->inode->i_ino;
411 npage[0] = dn->inode_page;
414 npage[0] = get_node_page(sbi, nids[0]);
415 if (IS_ERR(npage[0]))
416 return PTR_ERR(npage[0]);
420 nids[1] = get_nid(parent, offset[0], true);
421 dn->inode_page = npage[0];
422 dn->inode_page_locked = true;
424 /* get indirect or direct nodes */
425 for (i = 1; i <= level; i++) {
428 if (!nids[i] && mode == ALLOC_NODE) {
430 if (!alloc_nid(sbi, &(nids[i]))) {
436 npage[i] = new_node_page(dn, noffset[i], NULL);
437 if (IS_ERR(npage[i])) {
438 alloc_nid_failed(sbi, nids[i]);
439 err = PTR_ERR(npage[i]);
443 set_nid(parent, offset[i - 1], nids[i], i == 1);
444 alloc_nid_done(sbi, nids[i]);
446 } else if (mode == LOOKUP_NODE_RA && i == level && level > 1) {
447 npage[i] = get_node_page_ra(parent, offset[i - 1]);
448 if (IS_ERR(npage[i])) {
449 err = PTR_ERR(npage[i]);
455 dn->inode_page_locked = false;
458 f2fs_put_page(parent, 1);
462 npage[i] = get_node_page(sbi, nids[i]);
463 if (IS_ERR(npage[i])) {
464 err = PTR_ERR(npage[i]);
465 f2fs_put_page(npage[0], 0);
471 nids[i + 1] = get_nid(parent, offset[i], false);
474 dn->nid = nids[level];
475 dn->ofs_in_node = offset[level];
476 dn->node_page = npage[level];
477 dn->data_blkaddr = datablock_addr(dn->node_page, dn->ofs_in_node);
481 f2fs_put_page(parent, 1);
483 f2fs_put_page(npage[0], 0);
485 dn->inode_page = NULL;
486 dn->node_page = NULL;
490 static void truncate_node(struct dnode_of_data *dn)
492 struct f2fs_sb_info *sbi = F2FS_SB(dn->inode->i_sb);
495 get_node_info(sbi, dn->nid, &ni);
496 if (dn->inode->i_blocks == 0) {
497 f2fs_bug_on(ni.blk_addr != NULL_ADDR);
500 f2fs_bug_on(ni.blk_addr == NULL_ADDR);
502 /* Deallocate node address */
503 invalidate_blocks(sbi, ni.blk_addr);
504 dec_valid_node_count(sbi, dn->inode);
505 set_node_addr(sbi, &ni, NULL_ADDR, false);
507 if (dn->nid == dn->inode->i_ino) {
508 remove_orphan_inode(sbi, dn->nid);
509 dec_valid_inode_count(sbi);
514 clear_node_page_dirty(dn->node_page);
515 F2FS_SET_SB_DIRT(sbi);
517 f2fs_put_page(dn->node_page, 1);
519 invalidate_mapping_pages(NODE_MAPPING(sbi),
520 dn->node_page->index, dn->node_page->index);
522 dn->node_page = NULL;
523 trace_f2fs_truncate_node(dn->inode, dn->nid, ni.blk_addr);
526 static int truncate_dnode(struct dnode_of_data *dn)
528 struct f2fs_sb_info *sbi = F2FS_SB(dn->inode->i_sb);
534 /* get direct node */
535 page = get_node_page(sbi, dn->nid);
536 if (IS_ERR(page) && PTR_ERR(page) == -ENOENT)
538 else if (IS_ERR(page))
539 return PTR_ERR(page);
541 /* Make dnode_of_data for parameter */
542 dn->node_page = page;
544 truncate_data_blocks(dn);
549 static int truncate_nodes(struct dnode_of_data *dn, unsigned int nofs,
552 struct f2fs_sb_info *sbi = F2FS_SB(dn->inode->i_sb);
553 struct dnode_of_data rdn = *dn;
555 struct f2fs_node *rn;
557 unsigned int child_nofs;
562 return NIDS_PER_BLOCK + 1;
564 trace_f2fs_truncate_nodes_enter(dn->inode, dn->nid, dn->data_blkaddr);
566 page = get_node_page(sbi, dn->nid);
568 trace_f2fs_truncate_nodes_exit(dn->inode, PTR_ERR(page));
569 return PTR_ERR(page);
572 rn = F2FS_NODE(page);
574 for (i = ofs; i < NIDS_PER_BLOCK; i++, freed++) {
575 child_nid = le32_to_cpu(rn->in.nid[i]);
579 ret = truncate_dnode(&rdn);
582 set_nid(page, i, 0, false);
585 child_nofs = nofs + ofs * (NIDS_PER_BLOCK + 1) + 1;
586 for (i = ofs; i < NIDS_PER_BLOCK; i++) {
587 child_nid = le32_to_cpu(rn->in.nid[i]);
588 if (child_nid == 0) {
589 child_nofs += NIDS_PER_BLOCK + 1;
593 ret = truncate_nodes(&rdn, child_nofs, 0, depth - 1);
594 if (ret == (NIDS_PER_BLOCK + 1)) {
595 set_nid(page, i, 0, false);
597 } else if (ret < 0 && ret != -ENOENT) {
605 /* remove current indirect node */
606 dn->node_page = page;
610 f2fs_put_page(page, 1);
612 trace_f2fs_truncate_nodes_exit(dn->inode, freed);
616 f2fs_put_page(page, 1);
617 trace_f2fs_truncate_nodes_exit(dn->inode, ret);
621 static int truncate_partial_nodes(struct dnode_of_data *dn,
622 struct f2fs_inode *ri, int *offset, int depth)
624 struct f2fs_sb_info *sbi = F2FS_SB(dn->inode->i_sb);
625 struct page *pages[2];
632 nid[0] = le32_to_cpu(ri->i_nid[offset[0] - NODE_DIR1_BLOCK]);
636 /* get indirect nodes in the path */
637 for (i = 0; i < idx + 1; i++) {
638 /* refernece count'll be increased */
639 pages[i] = get_node_page(sbi, nid[i]);
640 if (IS_ERR(pages[i])) {
641 err = PTR_ERR(pages[i]);
645 nid[i + 1] = get_nid(pages[i], offset[i + 1], false);
648 /* free direct nodes linked to a partial indirect node */
649 for (i = offset[idx + 1]; i < NIDS_PER_BLOCK; i++) {
650 child_nid = get_nid(pages[idx], i, false);
654 err = truncate_dnode(dn);
657 set_nid(pages[idx], i, 0, false);
660 if (offset[idx + 1] == 0) {
661 dn->node_page = pages[idx];
665 f2fs_put_page(pages[idx], 1);
671 for (i = idx; i >= 0; i--)
672 f2fs_put_page(pages[i], 1);
674 trace_f2fs_truncate_partial_nodes(dn->inode, nid, depth, err);
680 * All the block addresses of data and nodes should be nullified.
682 int truncate_inode_blocks(struct inode *inode, pgoff_t from)
684 struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
685 int err = 0, cont = 1;
686 int level, offset[4], noffset[4];
687 unsigned int nofs = 0;
688 struct f2fs_inode *ri;
689 struct dnode_of_data dn;
692 trace_f2fs_truncate_inode_blocks_enter(inode, from);
694 level = get_node_path(F2FS_I(inode), from, offset, noffset);
696 page = get_node_page(sbi, inode->i_ino);
698 trace_f2fs_truncate_inode_blocks_exit(inode, PTR_ERR(page));
699 return PTR_ERR(page);
702 set_new_dnode(&dn, inode, page, NULL, 0);
705 ri = F2FS_INODE(page);
713 if (!offset[level - 1])
715 err = truncate_partial_nodes(&dn, ri, offset, level);
716 if (err < 0 && err != -ENOENT)
718 nofs += 1 + NIDS_PER_BLOCK;
721 nofs = 5 + 2 * NIDS_PER_BLOCK;
722 if (!offset[level - 1])
724 err = truncate_partial_nodes(&dn, ri, offset, level);
725 if (err < 0 && err != -ENOENT)
734 dn.nid = le32_to_cpu(ri->i_nid[offset[0] - NODE_DIR1_BLOCK]);
736 case NODE_DIR1_BLOCK:
737 case NODE_DIR2_BLOCK:
738 err = truncate_dnode(&dn);
741 case NODE_IND1_BLOCK:
742 case NODE_IND2_BLOCK:
743 err = truncate_nodes(&dn, nofs, offset[1], 2);
746 case NODE_DIND_BLOCK:
747 err = truncate_nodes(&dn, nofs, offset[1], 3);
754 if (err < 0 && err != -ENOENT)
756 if (offset[1] == 0 &&
757 ri->i_nid[offset[0] - NODE_DIR1_BLOCK]) {
759 if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
760 f2fs_put_page(page, 1);
763 f2fs_wait_on_page_writeback(page, NODE);
764 ri->i_nid[offset[0] - NODE_DIR1_BLOCK] = 0;
765 set_page_dirty(page);
773 f2fs_put_page(page, 0);
774 trace_f2fs_truncate_inode_blocks_exit(inode, err);
775 return err > 0 ? 0 : err;
778 int truncate_xattr_node(struct inode *inode, struct page *page)
780 struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
781 nid_t nid = F2FS_I(inode)->i_xattr_nid;
782 struct dnode_of_data dn;
788 npage = get_node_page(sbi, nid);
790 return PTR_ERR(npage);
792 F2FS_I(inode)->i_xattr_nid = 0;
794 /* need to do checkpoint during fsync */
795 F2FS_I(inode)->xattr_ver = cur_cp_version(F2FS_CKPT(sbi));
797 set_new_dnode(&dn, inode, page, npage, nid);
800 dn.inode_page_locked = true;
806 * Caller should grab and release a rwsem by calling f2fs_lock_op() and
809 void remove_inode_page(struct inode *inode)
811 struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
813 nid_t ino = inode->i_ino;
814 struct dnode_of_data dn;
816 page = get_node_page(sbi, ino);
820 if (truncate_xattr_node(inode, page)) {
821 f2fs_put_page(page, 1);
824 /* 0 is possible, after f2fs_new_inode() is failed */
825 f2fs_bug_on(inode->i_blocks != 0 && inode->i_blocks != 1);
826 set_new_dnode(&dn, inode, page, page, ino);
830 struct page *new_inode_page(struct inode *inode, const struct qstr *name)
832 struct dnode_of_data dn;
834 /* allocate inode page for new inode */
835 set_new_dnode(&dn, inode, NULL, NULL, inode->i_ino);
837 /* caller should f2fs_put_page(page, 1); */
838 return new_node_page(&dn, 0, NULL);
841 struct page *new_node_page(struct dnode_of_data *dn,
842 unsigned int ofs, struct page *ipage)
844 struct f2fs_sb_info *sbi = F2FS_SB(dn->inode->i_sb);
845 struct node_info old_ni, new_ni;
849 if (unlikely(is_inode_flag_set(F2FS_I(dn->inode), FI_NO_ALLOC)))
850 return ERR_PTR(-EPERM);
852 page = grab_cache_page_write_begin(NODE_MAPPING(sbi),
853 dn->nid, AOP_FLAG_NOFS);
855 return ERR_PTR(-ENOMEM);
857 if (unlikely(!inc_valid_node_count(sbi, dn->inode))) {
862 get_node_info(sbi, dn->nid, &old_ni);
864 /* Reinitialize old_ni with new node page */
865 f2fs_bug_on(old_ni.blk_addr != NULL_ADDR);
867 new_ni.ino = dn->inode->i_ino;
868 set_node_addr(sbi, &new_ni, NEW_ADDR, false);
870 fill_node_footer(page, dn->nid, dn->inode->i_ino, ofs, true);
871 set_cold_node(dn->inode, page);
872 SetPageUptodate(page);
873 set_page_dirty(page);
875 if (f2fs_has_xattr_block(ofs))
876 F2FS_I(dn->inode)->i_xattr_nid = dn->nid;
878 dn->node_page = page;
880 update_inode(dn->inode, ipage);
884 inc_valid_inode_count(sbi);
889 clear_node_page_dirty(page);
890 f2fs_put_page(page, 1);
895 * Caller should do after getting the following values.
896 * 0: f2fs_put_page(page, 0)
897 * LOCKED_PAGE: f2fs_put_page(page, 1)
900 static int read_node_page(struct page *page, int rw)
902 struct f2fs_sb_info *sbi = F2FS_SB(page->mapping->host->i_sb);
905 get_node_info(sbi, page->index, &ni);
907 if (unlikely(ni.blk_addr == NULL_ADDR)) {
908 f2fs_put_page(page, 1);
912 if (PageUptodate(page))
915 return f2fs_submit_page_bio(sbi, page, ni.blk_addr, rw);
919 * Readahead a node page
921 void ra_node_page(struct f2fs_sb_info *sbi, nid_t nid)
926 apage = find_get_page(NODE_MAPPING(sbi), nid);
927 if (apage && PageUptodate(apage)) {
928 f2fs_put_page(apage, 0);
931 f2fs_put_page(apage, 0);
933 apage = grab_cache_page(NODE_MAPPING(sbi), nid);
937 err = read_node_page(apage, READA);
939 f2fs_put_page(apage, 0);
940 else if (err == LOCKED_PAGE)
941 f2fs_put_page(apage, 1);
944 struct page *get_node_page(struct f2fs_sb_info *sbi, pgoff_t nid)
949 page = grab_cache_page_write_begin(NODE_MAPPING(sbi),
952 return ERR_PTR(-ENOMEM);
954 err = read_node_page(page, READ_SYNC);
957 else if (err == LOCKED_PAGE)
961 if (unlikely(!PageUptodate(page) || nid != nid_of_node(page))) {
962 f2fs_put_page(page, 1);
963 return ERR_PTR(-EIO);
965 if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
966 f2fs_put_page(page, 1);
970 mark_page_accessed(page);
975 * Return a locked page for the desired node page.
976 * And, readahead MAX_RA_NODE number of node pages.
978 struct page *get_node_page_ra(struct page *parent, int start)
980 struct f2fs_sb_info *sbi = F2FS_SB(parent->mapping->host->i_sb);
981 struct blk_plug plug;
986 /* First, try getting the desired direct node. */
987 nid = get_nid(parent, start, false);
989 return ERR_PTR(-ENOENT);
991 page = grab_cache_page(NODE_MAPPING(sbi), nid);
993 return ERR_PTR(-ENOMEM);
995 err = read_node_page(page, READ_SYNC);
998 else if (err == LOCKED_PAGE)
1001 blk_start_plug(&plug);
1003 /* Then, try readahead for siblings of the desired node */
1004 end = start + MAX_RA_NODE;
1005 end = min(end, NIDS_PER_BLOCK);
1006 for (i = start + 1; i < end; i++) {
1007 nid = get_nid(parent, i, false);
1010 ra_node_page(sbi, nid);
1013 blk_finish_plug(&plug);
1016 if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1017 f2fs_put_page(page, 1);
1021 if (unlikely(!PageUptodate(page))) {
1022 f2fs_put_page(page, 1);
1023 return ERR_PTR(-EIO);
1025 mark_page_accessed(page);
1029 void sync_inode_page(struct dnode_of_data *dn)
1031 if (IS_INODE(dn->node_page) || dn->inode_page == dn->node_page) {
1032 update_inode(dn->inode, dn->node_page);
1033 } else if (dn->inode_page) {
1034 if (!dn->inode_page_locked)
1035 lock_page(dn->inode_page);
1036 update_inode(dn->inode, dn->inode_page);
1037 if (!dn->inode_page_locked)
1038 unlock_page(dn->inode_page);
1040 update_inode_page(dn->inode);
1044 int sync_node_pages(struct f2fs_sb_info *sbi, nid_t ino,
1045 struct writeback_control *wbc)
1048 struct pagevec pvec;
1049 int step = ino ? 2 : 0;
1050 int nwritten = 0, wrote = 0;
1052 pagevec_init(&pvec, 0);
1058 while (index <= end) {
1060 nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1061 PAGECACHE_TAG_DIRTY,
1062 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
1066 for (i = 0; i < nr_pages; i++) {
1067 struct page *page = pvec.pages[i];
1070 * flushing sequence with step:
1075 if (step == 0 && IS_DNODE(page))
1077 if (step == 1 && (!IS_DNODE(page) ||
1078 is_cold_node(page)))
1080 if (step == 2 && (!IS_DNODE(page) ||
1081 !is_cold_node(page)))
1086 * we should not skip writing node pages.
1088 if (ino && ino_of_node(page) == ino)
1090 else if (!trylock_page(page))
1093 if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1098 if (ino && ino_of_node(page) != ino)
1099 goto continue_unlock;
1101 if (!PageDirty(page)) {
1102 /* someone wrote it for us */
1103 goto continue_unlock;
1106 if (!clear_page_dirty_for_io(page))
1107 goto continue_unlock;
1109 /* called by fsync() */
1110 if (ino && IS_DNODE(page)) {
1111 int mark = !is_checkpointed_node(sbi, ino);
1112 set_fsync_mark(page, 1);
1114 set_dentry_mark(page, mark);
1117 set_fsync_mark(page, 0);
1118 set_dentry_mark(page, 0);
1120 NODE_MAPPING(sbi)->a_ops->writepage(page, wbc);
1123 if (--wbc->nr_to_write == 0)
1126 pagevec_release(&pvec);
1129 if (wbc->nr_to_write == 0) {
1141 f2fs_submit_merged_bio(sbi, NODE, WRITE);
1145 int wait_on_node_pages_writeback(struct f2fs_sb_info *sbi, nid_t ino)
1147 pgoff_t index = 0, end = LONG_MAX;
1148 struct pagevec pvec;
1149 int ret2 = 0, ret = 0;
1151 pagevec_init(&pvec, 0);
1153 while (index <= end) {
1155 nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1156 PAGECACHE_TAG_WRITEBACK,
1157 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
1161 for (i = 0; i < nr_pages; i++) {
1162 struct page *page = pvec.pages[i];
1164 /* until radix tree lookup accepts end_index */
1165 if (unlikely(page->index > end))
1168 if (ino && ino_of_node(page) == ino) {
1169 f2fs_wait_on_page_writeback(page, NODE);
1170 if (TestClearPageError(page))
1174 pagevec_release(&pvec);
1178 if (unlikely(test_and_clear_bit(AS_ENOSPC, &NODE_MAPPING(sbi)->flags)))
1180 if (unlikely(test_and_clear_bit(AS_EIO, &NODE_MAPPING(sbi)->flags)))
1187 static int f2fs_write_node_page(struct page *page,
1188 struct writeback_control *wbc)
1190 struct f2fs_sb_info *sbi = F2FS_SB(page->mapping->host->i_sb);
1193 struct node_info ni;
1194 struct f2fs_io_info fio = {
1196 .rw = (wbc->sync_mode == WB_SYNC_ALL) ? WRITE_SYNC : WRITE,
1199 if (unlikely(sbi->por_doing))
1202 f2fs_wait_on_page_writeback(page, NODE);
1204 /* get old block addr of this node page */
1205 nid = nid_of_node(page);
1206 f2fs_bug_on(page->index != nid);
1208 get_node_info(sbi, nid, &ni);
1210 /* This page is already truncated */
1211 if (unlikely(ni.blk_addr == NULL_ADDR)) {
1212 dec_page_count(sbi, F2FS_DIRTY_NODES);
1217 if (wbc->for_reclaim)
1220 mutex_lock(&sbi->node_write);
1221 set_page_writeback(page);
1222 write_node_page(sbi, page, &fio, nid, ni.blk_addr, &new_addr);
1223 set_node_addr(sbi, &ni, new_addr, is_fsync_dnode(page));
1224 dec_page_count(sbi, F2FS_DIRTY_NODES);
1225 mutex_unlock(&sbi->node_write);
1230 dec_page_count(sbi, F2FS_DIRTY_NODES);
1231 wbc->pages_skipped++;
1232 account_page_redirty(page);
1233 set_page_dirty(page);
1234 return AOP_WRITEPAGE_ACTIVATE;
1237 static int f2fs_write_node_pages(struct address_space *mapping,
1238 struct writeback_control *wbc)
1240 struct f2fs_sb_info *sbi = F2FS_SB(mapping->host->i_sb);
1243 /* balancing f2fs's metadata in background */
1244 f2fs_balance_fs_bg(sbi);
1246 /* collect a number of dirty node pages and write together */
1247 if (get_pages(sbi, F2FS_DIRTY_NODES) < nr_pages_to_skip(sbi, NODE))
1250 diff = nr_pages_to_write(sbi, NODE, wbc);
1251 wbc->sync_mode = WB_SYNC_NONE;
1252 sync_node_pages(sbi, 0, wbc);
1253 wbc->nr_to_write = max((long)0, wbc->nr_to_write - diff);
1257 wbc->pages_skipped += get_pages(sbi, F2FS_DIRTY_NODES);
1261 static int f2fs_set_node_page_dirty(struct page *page)
1263 struct address_space *mapping = page->mapping;
1264 struct f2fs_sb_info *sbi = F2FS_SB(mapping->host->i_sb);
1266 trace_f2fs_set_page_dirty(page, NODE);
1268 SetPageUptodate(page);
1269 if (!PageDirty(page)) {
1270 __set_page_dirty_nobuffers(page);
1271 inc_page_count(sbi, F2FS_DIRTY_NODES);
1272 SetPagePrivate(page);
1278 static void f2fs_invalidate_node_page(struct page *page, unsigned int offset,
1279 unsigned int length)
1281 struct inode *inode = page->mapping->host;
1282 struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
1283 if (PageDirty(page))
1284 dec_page_count(sbi, F2FS_DIRTY_NODES);
1285 ClearPagePrivate(page);
1288 static int f2fs_release_node_page(struct page *page, gfp_t wait)
1290 ClearPagePrivate(page);
1295 * Structure of the f2fs node operations
1297 const struct address_space_operations f2fs_node_aops = {
1298 .writepage = f2fs_write_node_page,
1299 .writepages = f2fs_write_node_pages,
1300 .set_page_dirty = f2fs_set_node_page_dirty,
1301 .invalidatepage = f2fs_invalidate_node_page,
1302 .releasepage = f2fs_release_node_page,
1305 static struct free_nid *__lookup_free_nid_list(struct f2fs_nm_info *nm_i,
1308 return radix_tree_lookup(&nm_i->free_nid_root, n);
1311 static void __del_from_free_nid_list(struct f2fs_nm_info *nm_i,
1315 radix_tree_delete(&nm_i->free_nid_root, i->nid);
1316 kmem_cache_free(free_nid_slab, i);
1319 static int add_free_nid(struct f2fs_nm_info *nm_i, nid_t nid, bool build)
1322 struct nat_entry *ne;
1323 bool allocated = false;
1325 if (!available_free_memory(nm_i, FREE_NIDS))
1328 /* 0 nid should not be used */
1329 if (unlikely(nid == 0))
1333 /* do not add allocated nids */
1334 read_lock(&nm_i->nat_tree_lock);
1335 ne = __lookup_nat_cache(nm_i, nid);
1337 (!ne->checkpointed || nat_get_blkaddr(ne) != NULL_ADDR))
1339 read_unlock(&nm_i->nat_tree_lock);
1344 i = f2fs_kmem_cache_alloc(free_nid_slab, GFP_NOFS);
1348 spin_lock(&nm_i->free_nid_list_lock);
1349 if (radix_tree_insert(&nm_i->free_nid_root, i->nid, i)) {
1350 spin_unlock(&nm_i->free_nid_list_lock);
1351 kmem_cache_free(free_nid_slab, i);
1354 list_add_tail(&i->list, &nm_i->free_nid_list);
1356 spin_unlock(&nm_i->free_nid_list_lock);
1360 static void remove_free_nid(struct f2fs_nm_info *nm_i, nid_t nid)
1363 spin_lock(&nm_i->free_nid_list_lock);
1364 i = __lookup_free_nid_list(nm_i, nid);
1365 if (i && i->state == NID_NEW) {
1366 __del_from_free_nid_list(nm_i, i);
1369 spin_unlock(&nm_i->free_nid_list_lock);
1372 static void scan_nat_page(struct f2fs_nm_info *nm_i,
1373 struct page *nat_page, nid_t start_nid)
1375 struct f2fs_nat_block *nat_blk = page_address(nat_page);
1379 i = start_nid % NAT_ENTRY_PER_BLOCK;
1381 for (; i < NAT_ENTRY_PER_BLOCK; i++, start_nid++) {
1383 if (unlikely(start_nid >= nm_i->max_nid))
1386 blk_addr = le32_to_cpu(nat_blk->entries[i].block_addr);
1387 f2fs_bug_on(blk_addr == NEW_ADDR);
1388 if (blk_addr == NULL_ADDR) {
1389 if (add_free_nid(nm_i, start_nid, true) < 0)
1395 static void build_free_nids(struct f2fs_sb_info *sbi)
1397 struct f2fs_nm_info *nm_i = NM_I(sbi);
1398 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
1399 struct f2fs_summary_block *sum = curseg->sum_blk;
1401 nid_t nid = nm_i->next_scan_nid;
1403 /* Enough entries */
1404 if (nm_i->fcnt > NAT_ENTRY_PER_BLOCK)
1407 /* readahead nat pages to be scanned */
1408 ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nid), FREE_NID_PAGES, META_NAT);
1411 struct page *page = get_current_nat_page(sbi, nid);
1413 scan_nat_page(nm_i, page, nid);
1414 f2fs_put_page(page, 1);
1416 nid += (NAT_ENTRY_PER_BLOCK - (nid % NAT_ENTRY_PER_BLOCK));
1417 if (unlikely(nid >= nm_i->max_nid))
1420 if (i++ == FREE_NID_PAGES)
1424 /* go to the next free nat pages to find free nids abundantly */
1425 nm_i->next_scan_nid = nid;
1427 /* find free nids from current sum_pages */
1428 mutex_lock(&curseg->curseg_mutex);
1429 for (i = 0; i < nats_in_cursum(sum); i++) {
1430 block_t addr = le32_to_cpu(nat_in_journal(sum, i).block_addr);
1431 nid = le32_to_cpu(nid_in_journal(sum, i));
1432 if (addr == NULL_ADDR)
1433 add_free_nid(nm_i, nid, true);
1435 remove_free_nid(nm_i, nid);
1437 mutex_unlock(&curseg->curseg_mutex);
1441 * If this function returns success, caller can obtain a new nid
1442 * from second parameter of this function.
1443 * The returned nid could be used ino as well as nid when inode is created.
1445 bool alloc_nid(struct f2fs_sb_info *sbi, nid_t *nid)
1447 struct f2fs_nm_info *nm_i = NM_I(sbi);
1448 struct free_nid *i = NULL;
1449 struct list_head *this;
1451 if (unlikely(sbi->total_valid_node_count + 1 >= nm_i->max_nid))
1454 spin_lock(&nm_i->free_nid_list_lock);
1456 /* We should not use stale free nids created by build_free_nids */
1457 if (nm_i->fcnt && !on_build_free_nids(nm_i)) {
1458 f2fs_bug_on(list_empty(&nm_i->free_nid_list));
1459 list_for_each(this, &nm_i->free_nid_list) {
1460 i = list_entry(this, struct free_nid, list);
1461 if (i->state == NID_NEW)
1465 f2fs_bug_on(i->state != NID_NEW);
1467 i->state = NID_ALLOC;
1469 spin_unlock(&nm_i->free_nid_list_lock);
1472 spin_unlock(&nm_i->free_nid_list_lock);
1474 /* Let's scan nat pages and its caches to get free nids */
1475 mutex_lock(&nm_i->build_lock);
1476 build_free_nids(sbi);
1477 mutex_unlock(&nm_i->build_lock);
1482 * alloc_nid() should be called prior to this function.
1484 void alloc_nid_done(struct f2fs_sb_info *sbi, nid_t nid)
1486 struct f2fs_nm_info *nm_i = NM_I(sbi);
1489 spin_lock(&nm_i->free_nid_list_lock);
1490 i = __lookup_free_nid_list(nm_i, nid);
1491 f2fs_bug_on(!i || i->state != NID_ALLOC);
1492 __del_from_free_nid_list(nm_i, i);
1493 spin_unlock(&nm_i->free_nid_list_lock);
1497 * alloc_nid() should be called prior to this function.
1499 void alloc_nid_failed(struct f2fs_sb_info *sbi, nid_t nid)
1501 struct f2fs_nm_info *nm_i = NM_I(sbi);
1507 spin_lock(&nm_i->free_nid_list_lock);
1508 i = __lookup_free_nid_list(nm_i, nid);
1509 f2fs_bug_on(!i || i->state != NID_ALLOC);
1510 if (!available_free_memory(nm_i, FREE_NIDS)) {
1511 __del_from_free_nid_list(nm_i, i);
1516 spin_unlock(&nm_i->free_nid_list_lock);
1519 void recover_node_page(struct f2fs_sb_info *sbi, struct page *page,
1520 struct f2fs_summary *sum, struct node_info *ni,
1521 block_t new_blkaddr)
1523 rewrite_node_page(sbi, page, sum, ni->blk_addr, new_blkaddr);
1524 set_node_addr(sbi, ni, new_blkaddr, false);
1525 clear_node_page_dirty(page);
1528 void recover_inline_xattr(struct inode *inode, struct page *page)
1530 struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
1531 void *src_addr, *dst_addr;
1534 struct f2fs_inode *ri;
1536 if (!f2fs_has_inline_xattr(inode))
1539 if (!IS_INODE(page))
1542 ri = F2FS_INODE(page);
1543 if (!(ri->i_inline & F2FS_INLINE_XATTR))
1546 ipage = get_node_page(sbi, inode->i_ino);
1547 f2fs_bug_on(IS_ERR(ipage));
1549 dst_addr = inline_xattr_addr(ipage);
1550 src_addr = inline_xattr_addr(page);
1551 inline_size = inline_xattr_size(inode);
1553 memcpy(dst_addr, src_addr, inline_size);
1555 update_inode(inode, ipage);
1556 f2fs_put_page(ipage, 1);
1559 bool recover_xattr_data(struct inode *inode, struct page *page, block_t blkaddr)
1561 struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
1562 nid_t prev_xnid = F2FS_I(inode)->i_xattr_nid;
1563 nid_t new_xnid = nid_of_node(page);
1564 struct node_info ni;
1566 recover_inline_xattr(inode, page);
1568 if (!f2fs_has_xattr_block(ofs_of_node(page)))
1571 /* 1: invalidate the previous xattr nid */
1575 /* Deallocate node address */
1576 get_node_info(sbi, prev_xnid, &ni);
1577 f2fs_bug_on(ni.blk_addr == NULL_ADDR);
1578 invalidate_blocks(sbi, ni.blk_addr);
1579 dec_valid_node_count(sbi, inode);
1580 set_node_addr(sbi, &ni, NULL_ADDR, false);
1583 /* 2: allocate new xattr nid */
1584 if (unlikely(!inc_valid_node_count(sbi, inode)))
1587 remove_free_nid(NM_I(sbi), new_xnid);
1588 get_node_info(sbi, new_xnid, &ni);
1589 ni.ino = inode->i_ino;
1590 set_node_addr(sbi, &ni, NEW_ADDR, false);
1591 F2FS_I(inode)->i_xattr_nid = new_xnid;
1593 /* 3: update xattr blkaddr */
1594 refresh_sit_entry(sbi, NEW_ADDR, blkaddr);
1595 set_node_addr(sbi, &ni, blkaddr, false);
1597 update_inode_page(inode);
1601 int recover_inode_page(struct f2fs_sb_info *sbi, struct page *page)
1603 struct f2fs_inode *src, *dst;
1604 nid_t ino = ino_of_node(page);
1605 struct node_info old_ni, new_ni;
1608 ipage = grab_cache_page(NODE_MAPPING(sbi), ino);
1612 /* Should not use this inode from free nid list */
1613 remove_free_nid(NM_I(sbi), ino);
1615 get_node_info(sbi, ino, &old_ni);
1616 SetPageUptodate(ipage);
1617 fill_node_footer(ipage, ino, ino, 0, true);
1619 src = F2FS_INODE(page);
1620 dst = F2FS_INODE(ipage);
1622 memcpy(dst, src, (unsigned long)&src->i_ext - (unsigned long)src);
1624 dst->i_blocks = cpu_to_le64(1);
1625 dst->i_links = cpu_to_le32(1);
1626 dst->i_xattr_nid = 0;
1631 if (unlikely(!inc_valid_node_count(sbi, NULL)))
1633 set_node_addr(sbi, &new_ni, NEW_ADDR, false);
1634 inc_valid_inode_count(sbi);
1635 f2fs_put_page(ipage, 1);
1640 * ra_sum_pages() merge contiguous pages into one bio and submit.
1641 * these pre-readed pages are linked in pages list.
1643 static int ra_sum_pages(struct f2fs_sb_info *sbi, struct list_head *pages,
1644 int start, int nrpages)
1647 int page_idx = start;
1648 struct f2fs_io_info fio = {
1650 .rw = READ_SYNC | REQ_META | REQ_PRIO
1653 for (; page_idx < start + nrpages; page_idx++) {
1654 /* alloc temporal page for read node summary info*/
1655 page = alloc_page(GFP_F2FS_ZERO);
1660 page->index = page_idx;
1661 list_add_tail(&page->lru, pages);
1664 list_for_each_entry(page, pages, lru)
1665 f2fs_submit_page_mbio(sbi, page, page->index, &fio);
1667 f2fs_submit_merged_bio(sbi, META, READ);
1669 return page_idx - start;
1672 int restore_node_summary(struct f2fs_sb_info *sbi,
1673 unsigned int segno, struct f2fs_summary_block *sum)
1675 struct f2fs_node *rn;
1676 struct f2fs_summary *sum_entry;
1677 struct page *page, *tmp;
1679 int bio_blocks = MAX_BIO_BLOCKS(max_hw_blocks(sbi));
1680 int i, last_offset, nrpages, err = 0;
1681 LIST_HEAD(page_list);
1683 /* scan the node segment */
1684 last_offset = sbi->blocks_per_seg;
1685 addr = START_BLOCK(sbi, segno);
1686 sum_entry = &sum->entries[0];
1688 for (i = 0; !err && i < last_offset; i += nrpages, addr += nrpages) {
1689 nrpages = min(last_offset - i, bio_blocks);
1691 /* read ahead node pages */
1692 nrpages = ra_sum_pages(sbi, &page_list, addr, nrpages);
1696 list_for_each_entry_safe(page, tmp, &page_list, lru) {
1701 if (unlikely(!PageUptodate(page))) {
1704 rn = F2FS_NODE(page);
1705 sum_entry->nid = rn->footer.nid;
1706 sum_entry->version = 0;
1707 sum_entry->ofs_in_node = 0;
1712 list_del(&page->lru);
1713 __free_pages(page, 0);
1719 static bool flush_nats_in_journal(struct f2fs_sb_info *sbi)
1721 struct f2fs_nm_info *nm_i = NM_I(sbi);
1722 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
1723 struct f2fs_summary_block *sum = curseg->sum_blk;
1726 mutex_lock(&curseg->curseg_mutex);
1728 if (nats_in_cursum(sum) < NAT_JOURNAL_ENTRIES) {
1729 mutex_unlock(&curseg->curseg_mutex);
1733 for (i = 0; i < nats_in_cursum(sum); i++) {
1734 struct nat_entry *ne;
1735 struct f2fs_nat_entry raw_ne;
1736 nid_t nid = le32_to_cpu(nid_in_journal(sum, i));
1738 raw_ne = nat_in_journal(sum, i);
1740 write_lock(&nm_i->nat_tree_lock);
1741 ne = __lookup_nat_cache(nm_i, nid);
1743 __set_nat_cache_dirty(nm_i, ne);
1744 write_unlock(&nm_i->nat_tree_lock);
1747 ne = grab_nat_entry(nm_i, nid);
1749 write_unlock(&nm_i->nat_tree_lock);
1752 nat_set_blkaddr(ne, le32_to_cpu(raw_ne.block_addr));
1753 nat_set_ino(ne, le32_to_cpu(raw_ne.ino));
1754 nat_set_version(ne, raw_ne.version);
1755 __set_nat_cache_dirty(nm_i, ne);
1756 write_unlock(&nm_i->nat_tree_lock);
1758 update_nats_in_cursum(sum, -i);
1759 mutex_unlock(&curseg->curseg_mutex);
1764 * This function is called during the checkpointing process.
1766 void flush_nat_entries(struct f2fs_sb_info *sbi)
1768 struct f2fs_nm_info *nm_i = NM_I(sbi);
1769 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
1770 struct f2fs_summary_block *sum = curseg->sum_blk;
1771 struct list_head *cur, *n;
1772 struct page *page = NULL;
1773 struct f2fs_nat_block *nat_blk = NULL;
1774 nid_t start_nid = 0, end_nid = 0;
1777 flushed = flush_nats_in_journal(sbi);
1780 mutex_lock(&curseg->curseg_mutex);
1782 /* 1) flush dirty nat caches */
1783 list_for_each_safe(cur, n, &nm_i->dirty_nat_entries) {
1784 struct nat_entry *ne;
1786 struct f2fs_nat_entry raw_ne;
1788 block_t new_blkaddr;
1790 ne = list_entry(cur, struct nat_entry, list);
1791 nid = nat_get_nid(ne);
1793 if (nat_get_blkaddr(ne) == NEW_ADDR)
1798 /* if there is room for nat enries in curseg->sumpage */
1799 offset = lookup_journal_in_cursum(sum, NAT_JOURNAL, nid, 1);
1801 raw_ne = nat_in_journal(sum, offset);
1805 if (!page || (start_nid > nid || nid > end_nid)) {
1807 f2fs_put_page(page, 1);
1810 start_nid = START_NID(nid);
1811 end_nid = start_nid + NAT_ENTRY_PER_BLOCK - 1;
1814 * get nat block with dirty flag, increased reference
1815 * count, mapped and lock
1817 page = get_next_nat_page(sbi, start_nid);
1818 nat_blk = page_address(page);
1821 f2fs_bug_on(!nat_blk);
1822 raw_ne = nat_blk->entries[nid - start_nid];
1824 new_blkaddr = nat_get_blkaddr(ne);
1826 raw_ne.ino = cpu_to_le32(nat_get_ino(ne));
1827 raw_ne.block_addr = cpu_to_le32(new_blkaddr);
1828 raw_ne.version = nat_get_version(ne);
1831 nat_blk->entries[nid - start_nid] = raw_ne;
1833 nat_in_journal(sum, offset) = raw_ne;
1834 nid_in_journal(sum, offset) = cpu_to_le32(nid);
1837 if (nat_get_blkaddr(ne) == NULL_ADDR &&
1838 add_free_nid(NM_I(sbi), nid, false) <= 0) {
1839 write_lock(&nm_i->nat_tree_lock);
1840 __del_from_nat_cache(nm_i, ne);
1841 write_unlock(&nm_i->nat_tree_lock);
1843 write_lock(&nm_i->nat_tree_lock);
1844 __clear_nat_cache_dirty(nm_i, ne);
1845 write_unlock(&nm_i->nat_tree_lock);
1849 mutex_unlock(&curseg->curseg_mutex);
1850 f2fs_put_page(page, 1);
1853 static int init_node_manager(struct f2fs_sb_info *sbi)
1855 struct f2fs_super_block *sb_raw = F2FS_RAW_SUPER(sbi);
1856 struct f2fs_nm_info *nm_i = NM_I(sbi);
1857 unsigned char *version_bitmap;
1858 unsigned int nat_segs, nat_blocks;
1860 nm_i->nat_blkaddr = le32_to_cpu(sb_raw->nat_blkaddr);
1862 /* segment_count_nat includes pair segment so divide to 2. */
1863 nat_segs = le32_to_cpu(sb_raw->segment_count_nat) >> 1;
1864 nat_blocks = nat_segs << le32_to_cpu(sb_raw->log_blocks_per_seg);
1866 /* not used nids: 0, node, meta, (and root counted as valid node) */
1867 nm_i->max_nid = NAT_ENTRY_PER_BLOCK * nat_blocks - 3;
1870 nm_i->ram_thresh = DEF_RAM_THRESHOLD;
1872 INIT_RADIX_TREE(&nm_i->free_nid_root, GFP_ATOMIC);
1873 INIT_LIST_HEAD(&nm_i->free_nid_list);
1874 INIT_RADIX_TREE(&nm_i->nat_root, GFP_ATOMIC);
1875 INIT_LIST_HEAD(&nm_i->nat_entries);
1876 INIT_LIST_HEAD(&nm_i->dirty_nat_entries);
1878 mutex_init(&nm_i->build_lock);
1879 spin_lock_init(&nm_i->free_nid_list_lock);
1880 rwlock_init(&nm_i->nat_tree_lock);
1882 nm_i->next_scan_nid = le32_to_cpu(sbi->ckpt->next_free_nid);
1883 nm_i->bitmap_size = __bitmap_size(sbi, NAT_BITMAP);
1884 version_bitmap = __bitmap_ptr(sbi, NAT_BITMAP);
1885 if (!version_bitmap)
1888 nm_i->nat_bitmap = kmemdup(version_bitmap, nm_i->bitmap_size,
1890 if (!nm_i->nat_bitmap)
1895 int build_node_manager(struct f2fs_sb_info *sbi)
1899 sbi->nm_info = kzalloc(sizeof(struct f2fs_nm_info), GFP_KERNEL);
1903 err = init_node_manager(sbi);
1907 build_free_nids(sbi);
1911 void destroy_node_manager(struct f2fs_sb_info *sbi)
1913 struct f2fs_nm_info *nm_i = NM_I(sbi);
1914 struct free_nid *i, *next_i;
1915 struct nat_entry *natvec[NATVEC_SIZE];
1922 /* destroy free nid list */
1923 spin_lock(&nm_i->free_nid_list_lock);
1924 list_for_each_entry_safe(i, next_i, &nm_i->free_nid_list, list) {
1925 f2fs_bug_on(i->state == NID_ALLOC);
1926 __del_from_free_nid_list(nm_i, i);
1929 f2fs_bug_on(nm_i->fcnt);
1930 spin_unlock(&nm_i->free_nid_list_lock);
1932 /* destroy nat cache */
1933 write_lock(&nm_i->nat_tree_lock);
1934 while ((found = __gang_lookup_nat_cache(nm_i,
1935 nid, NATVEC_SIZE, natvec))) {
1937 nid = nat_get_nid(natvec[found - 1]) + 1;
1938 for (idx = 0; idx < found; idx++)
1939 __del_from_nat_cache(nm_i, natvec[idx]);
1941 f2fs_bug_on(nm_i->nat_cnt);
1942 write_unlock(&nm_i->nat_tree_lock);
1944 kfree(nm_i->nat_bitmap);
1945 sbi->nm_info = NULL;
1949 int __init create_node_manager_caches(void)
1951 nat_entry_slab = f2fs_kmem_cache_create("nat_entry",
1952 sizeof(struct nat_entry));
1953 if (!nat_entry_slab)
1956 free_nid_slab = f2fs_kmem_cache_create("free_nid",
1957 sizeof(struct free_nid));
1958 if (!free_nid_slab) {
1959 kmem_cache_destroy(nat_entry_slab);
1965 void destroy_node_manager_caches(void)
1967 kmem_cache_destroy(free_nid_slab);
1968 kmem_cache_destroy(nat_entry_slab);