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;
28 static struct kmem_cache *nat_entry_set_slab;
30 bool available_free_memory(struct f2fs_sb_info *sbi, int type)
32 struct f2fs_nm_info *nm_i = NM_I(sbi);
34 unsigned long mem_size = 0;
38 /* give 25%, 25%, 50% memory for each components respectively */
39 if (type == FREE_NIDS) {
40 mem_size = (nm_i->fcnt * sizeof(struct free_nid)) >> 12;
41 res = mem_size < ((val.totalram * nm_i->ram_thresh / 100) >> 2);
42 } else if (type == NAT_ENTRIES) {
43 mem_size = (nm_i->nat_cnt * sizeof(struct nat_entry)) >> 12;
44 res = mem_size < ((val.totalram * nm_i->ram_thresh / 100) >> 2);
45 } else if (type == DIRTY_DENTS) {
46 if (sbi->sb->s_bdi->dirty_exceeded)
48 mem_size = get_pages(sbi, F2FS_DIRTY_DENTS);
49 res = mem_size < ((val.totalram * nm_i->ram_thresh / 100) >> 1);
54 static void clear_node_page_dirty(struct page *page)
56 struct address_space *mapping = page->mapping;
57 unsigned int long flags;
59 if (PageDirty(page)) {
60 spin_lock_irqsave(&mapping->tree_lock, flags);
61 radix_tree_tag_clear(&mapping->page_tree,
64 spin_unlock_irqrestore(&mapping->tree_lock, flags);
66 clear_page_dirty_for_io(page);
67 dec_page_count(F2FS_M_SB(mapping), F2FS_DIRTY_NODES);
69 ClearPageUptodate(page);
72 static struct page *get_current_nat_page(struct f2fs_sb_info *sbi, nid_t nid)
74 pgoff_t index = current_nat_addr(sbi, nid);
75 return get_meta_page(sbi, index);
78 static struct page *get_next_nat_page(struct f2fs_sb_info *sbi, nid_t nid)
80 struct page *src_page;
81 struct page *dst_page;
86 struct f2fs_nm_info *nm_i = NM_I(sbi);
88 src_off = current_nat_addr(sbi, nid);
89 dst_off = next_nat_addr(sbi, src_off);
91 /* get current nat block page with lock */
92 src_page = get_meta_page(sbi, src_off);
93 dst_page = grab_meta_page(sbi, dst_off);
94 f2fs_bug_on(sbi, PageDirty(src_page));
96 src_addr = page_address(src_page);
97 dst_addr = page_address(dst_page);
98 memcpy(dst_addr, src_addr, PAGE_CACHE_SIZE);
99 set_page_dirty(dst_page);
100 f2fs_put_page(src_page, 1);
102 set_to_next_nat(nm_i, nid);
107 static struct nat_entry *__lookup_nat_cache(struct f2fs_nm_info *nm_i, nid_t n)
109 return radix_tree_lookup(&nm_i->nat_root, n);
112 static unsigned int __gang_lookup_nat_cache(struct f2fs_nm_info *nm_i,
113 nid_t start, unsigned int nr, struct nat_entry **ep)
115 return radix_tree_gang_lookup(&nm_i->nat_root, (void **)ep, start, nr);
118 static void __del_from_nat_cache(struct f2fs_nm_info *nm_i, struct nat_entry *e)
121 radix_tree_delete(&nm_i->nat_root, nat_get_nid(e));
123 kmem_cache_free(nat_entry_slab, e);
126 static void __set_nat_cache_dirty(struct f2fs_nm_info *nm_i,
127 struct nat_entry *ne)
129 nid_t set = NAT_BLOCK_OFFSET(ne->ni.nid);
130 struct nat_entry_set *head;
132 if (get_nat_flag(ne, IS_DIRTY))
135 head = radix_tree_lookup(&nm_i->nat_set_root, set);
137 head = f2fs_kmem_cache_alloc(nat_entry_set_slab, GFP_ATOMIC);
139 INIT_LIST_HEAD(&head->entry_list);
140 INIT_LIST_HEAD(&head->set_list);
144 if (radix_tree_insert(&nm_i->nat_set_root, set, head)) {
149 list_move_tail(&ne->list, &head->entry_list);
150 nm_i->dirty_nat_cnt++;
152 set_nat_flag(ne, IS_DIRTY, true);
155 static void __clear_nat_cache_dirty(struct f2fs_nm_info *nm_i,
156 struct nat_entry *ne)
158 nid_t set = ne->ni.nid / NAT_ENTRY_PER_BLOCK;
159 struct nat_entry_set *head;
161 head = radix_tree_lookup(&nm_i->nat_set_root, set);
163 list_move_tail(&ne->list, &nm_i->nat_entries);
164 set_nat_flag(ne, IS_DIRTY, false);
166 nm_i->dirty_nat_cnt--;
170 static unsigned int __gang_lookup_nat_set(struct f2fs_nm_info *nm_i,
171 nid_t start, unsigned int nr, struct nat_entry_set **ep)
173 return radix_tree_gang_lookup(&nm_i->nat_set_root, (void **)ep,
177 bool is_checkpointed_node(struct f2fs_sb_info *sbi, nid_t nid)
179 struct f2fs_nm_info *nm_i = NM_I(sbi);
183 read_lock(&nm_i->nat_tree_lock);
184 e = __lookup_nat_cache(nm_i, nid);
185 if (e && !get_nat_flag(e, IS_CHECKPOINTED))
187 read_unlock(&nm_i->nat_tree_lock);
191 bool has_fsynced_inode(struct f2fs_sb_info *sbi, nid_t ino)
193 struct f2fs_nm_info *nm_i = NM_I(sbi);
195 bool fsynced = false;
197 read_lock(&nm_i->nat_tree_lock);
198 e = __lookup_nat_cache(nm_i, ino);
199 if (e && get_nat_flag(e, HAS_FSYNCED_INODE))
201 read_unlock(&nm_i->nat_tree_lock);
205 bool need_inode_block_update(struct f2fs_sb_info *sbi, nid_t ino)
207 struct f2fs_nm_info *nm_i = NM_I(sbi);
209 bool need_update = true;
211 read_lock(&nm_i->nat_tree_lock);
212 e = __lookup_nat_cache(nm_i, ino);
213 if (e && get_nat_flag(e, HAS_LAST_FSYNC) &&
214 (get_nat_flag(e, IS_CHECKPOINTED) ||
215 get_nat_flag(e, HAS_FSYNCED_INODE)))
217 read_unlock(&nm_i->nat_tree_lock);
221 static struct nat_entry *grab_nat_entry(struct f2fs_nm_info *nm_i, nid_t nid)
223 struct nat_entry *new;
225 new = kmem_cache_alloc(nat_entry_slab, GFP_ATOMIC);
228 if (radix_tree_insert(&nm_i->nat_root, nid, new)) {
229 kmem_cache_free(nat_entry_slab, new);
232 memset(new, 0, sizeof(struct nat_entry));
233 nat_set_nid(new, nid);
235 list_add_tail(&new->list, &nm_i->nat_entries);
240 static void cache_nat_entry(struct f2fs_nm_info *nm_i, nid_t nid,
241 struct f2fs_nat_entry *ne)
245 write_lock(&nm_i->nat_tree_lock);
246 e = __lookup_nat_cache(nm_i, nid);
248 e = grab_nat_entry(nm_i, nid);
250 write_unlock(&nm_i->nat_tree_lock);
253 node_info_from_raw_nat(&e->ni, ne);
255 write_unlock(&nm_i->nat_tree_lock);
258 static void set_node_addr(struct f2fs_sb_info *sbi, struct node_info *ni,
259 block_t new_blkaddr, bool fsync_done)
261 struct f2fs_nm_info *nm_i = NM_I(sbi);
264 write_lock(&nm_i->nat_tree_lock);
265 e = __lookup_nat_cache(nm_i, ni->nid);
267 e = grab_nat_entry(nm_i, ni->nid);
269 write_unlock(&nm_i->nat_tree_lock);
273 f2fs_bug_on(sbi, ni->blk_addr == NEW_ADDR);
274 } else if (new_blkaddr == NEW_ADDR) {
276 * when nid is reallocated,
277 * previous nat entry can be remained in nat cache.
278 * So, reinitialize it with new information.
281 f2fs_bug_on(sbi, ni->blk_addr != NULL_ADDR);
285 f2fs_bug_on(sbi, nat_get_blkaddr(e) != ni->blk_addr);
286 f2fs_bug_on(sbi, nat_get_blkaddr(e) == NULL_ADDR &&
287 new_blkaddr == NULL_ADDR);
288 f2fs_bug_on(sbi, nat_get_blkaddr(e) == NEW_ADDR &&
289 new_blkaddr == NEW_ADDR);
290 f2fs_bug_on(sbi, nat_get_blkaddr(e) != NEW_ADDR &&
291 nat_get_blkaddr(e) != NULL_ADDR &&
292 new_blkaddr == NEW_ADDR);
294 /* increment version no as node is removed */
295 if (nat_get_blkaddr(e) != NEW_ADDR && new_blkaddr == NULL_ADDR) {
296 unsigned char version = nat_get_version(e);
297 nat_set_version(e, inc_node_version(version));
301 nat_set_blkaddr(e, new_blkaddr);
302 if (new_blkaddr == NEW_ADDR || new_blkaddr == NULL_ADDR)
303 set_nat_flag(e, IS_CHECKPOINTED, false);
304 __set_nat_cache_dirty(nm_i, e);
306 /* update fsync_mark if its inode nat entry is still alive */
307 e = __lookup_nat_cache(nm_i, ni->ino);
309 if (fsync_done && ni->nid == ni->ino)
310 set_nat_flag(e, HAS_FSYNCED_INODE, true);
311 set_nat_flag(e, HAS_LAST_FSYNC, fsync_done);
313 write_unlock(&nm_i->nat_tree_lock);
316 int try_to_free_nats(struct f2fs_sb_info *sbi, int nr_shrink)
318 struct f2fs_nm_info *nm_i = NM_I(sbi);
320 if (available_free_memory(sbi, NAT_ENTRIES))
323 write_lock(&nm_i->nat_tree_lock);
324 while (nr_shrink && !list_empty(&nm_i->nat_entries)) {
325 struct nat_entry *ne;
326 ne = list_first_entry(&nm_i->nat_entries,
327 struct nat_entry, list);
328 __del_from_nat_cache(nm_i, ne);
331 write_unlock(&nm_i->nat_tree_lock);
336 * This function always returns success
338 void get_node_info(struct f2fs_sb_info *sbi, nid_t nid, struct node_info *ni)
340 struct f2fs_nm_info *nm_i = NM_I(sbi);
341 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
342 struct f2fs_summary_block *sum = curseg->sum_blk;
343 nid_t start_nid = START_NID(nid);
344 struct f2fs_nat_block *nat_blk;
345 struct page *page = NULL;
346 struct f2fs_nat_entry ne;
350 memset(&ne, 0, sizeof(struct f2fs_nat_entry));
353 /* Check nat cache */
354 read_lock(&nm_i->nat_tree_lock);
355 e = __lookup_nat_cache(nm_i, nid);
357 ni->ino = nat_get_ino(e);
358 ni->blk_addr = nat_get_blkaddr(e);
359 ni->version = nat_get_version(e);
361 read_unlock(&nm_i->nat_tree_lock);
365 /* Check current segment summary */
366 mutex_lock(&curseg->curseg_mutex);
367 i = lookup_journal_in_cursum(sum, NAT_JOURNAL, nid, 0);
369 ne = nat_in_journal(sum, i);
370 node_info_from_raw_nat(ni, &ne);
372 mutex_unlock(&curseg->curseg_mutex);
376 /* Fill node_info from nat page */
377 page = get_current_nat_page(sbi, start_nid);
378 nat_blk = (struct f2fs_nat_block *)page_address(page);
379 ne = nat_blk->entries[nid - start_nid];
380 node_info_from_raw_nat(ni, &ne);
381 f2fs_put_page(page, 1);
383 /* cache nat entry */
384 cache_nat_entry(NM_I(sbi), nid, &ne);
388 * The maximum depth is four.
389 * Offset[0] will have raw inode offset.
391 static int get_node_path(struct f2fs_inode_info *fi, long block,
392 int offset[4], unsigned int noffset[4])
394 const long direct_index = ADDRS_PER_INODE(fi);
395 const long direct_blks = ADDRS_PER_BLOCK;
396 const long dptrs_per_blk = NIDS_PER_BLOCK;
397 const long indirect_blks = ADDRS_PER_BLOCK * NIDS_PER_BLOCK;
398 const long dindirect_blks = indirect_blks * NIDS_PER_BLOCK;
404 if (block < direct_index) {
408 block -= direct_index;
409 if (block < direct_blks) {
410 offset[n++] = NODE_DIR1_BLOCK;
416 block -= direct_blks;
417 if (block < direct_blks) {
418 offset[n++] = NODE_DIR2_BLOCK;
424 block -= direct_blks;
425 if (block < indirect_blks) {
426 offset[n++] = NODE_IND1_BLOCK;
428 offset[n++] = block / direct_blks;
429 noffset[n] = 4 + offset[n - 1];
430 offset[n] = block % direct_blks;
434 block -= indirect_blks;
435 if (block < indirect_blks) {
436 offset[n++] = NODE_IND2_BLOCK;
437 noffset[n] = 4 + dptrs_per_blk;
438 offset[n++] = block / direct_blks;
439 noffset[n] = 5 + dptrs_per_blk + offset[n - 1];
440 offset[n] = block % direct_blks;
444 block -= indirect_blks;
445 if (block < dindirect_blks) {
446 offset[n++] = NODE_DIND_BLOCK;
447 noffset[n] = 5 + (dptrs_per_blk * 2);
448 offset[n++] = block / indirect_blks;
449 noffset[n] = 6 + (dptrs_per_blk * 2) +
450 offset[n - 1] * (dptrs_per_blk + 1);
451 offset[n++] = (block / direct_blks) % dptrs_per_blk;
452 noffset[n] = 7 + (dptrs_per_blk * 2) +
453 offset[n - 2] * (dptrs_per_blk + 1) +
455 offset[n] = block % direct_blks;
466 * Caller should call f2fs_put_dnode(dn).
467 * Also, it should grab and release a rwsem by calling f2fs_lock_op() and
468 * f2fs_unlock_op() only if ro is not set RDONLY_NODE.
469 * In the case of RDONLY_NODE, we don't need to care about mutex.
471 int get_dnode_of_data(struct dnode_of_data *dn, pgoff_t index, int mode)
473 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
474 struct page *npage[4];
477 unsigned int noffset[4];
482 level = get_node_path(F2FS_I(dn->inode), index, offset, noffset);
484 nids[0] = dn->inode->i_ino;
485 npage[0] = dn->inode_page;
488 npage[0] = get_node_page(sbi, nids[0]);
489 if (IS_ERR(npage[0]))
490 return PTR_ERR(npage[0]);
494 nids[1] = get_nid(parent, offset[0], true);
495 dn->inode_page = npage[0];
496 dn->inode_page_locked = true;
498 /* get indirect or direct nodes */
499 for (i = 1; i <= level; i++) {
502 if (!nids[i] && mode == ALLOC_NODE) {
504 if (!alloc_nid(sbi, &(nids[i]))) {
510 npage[i] = new_node_page(dn, noffset[i], NULL);
511 if (IS_ERR(npage[i])) {
512 alloc_nid_failed(sbi, nids[i]);
513 err = PTR_ERR(npage[i]);
517 set_nid(parent, offset[i - 1], nids[i], i == 1);
518 alloc_nid_done(sbi, nids[i]);
520 } else if (mode == LOOKUP_NODE_RA && i == level && level > 1) {
521 npage[i] = get_node_page_ra(parent, offset[i - 1]);
522 if (IS_ERR(npage[i])) {
523 err = PTR_ERR(npage[i]);
529 dn->inode_page_locked = false;
532 f2fs_put_page(parent, 1);
536 npage[i] = get_node_page(sbi, nids[i]);
537 if (IS_ERR(npage[i])) {
538 err = PTR_ERR(npage[i]);
539 f2fs_put_page(npage[0], 0);
545 nids[i + 1] = get_nid(parent, offset[i], false);
548 dn->nid = nids[level];
549 dn->ofs_in_node = offset[level];
550 dn->node_page = npage[level];
551 dn->data_blkaddr = datablock_addr(dn->node_page, dn->ofs_in_node);
555 f2fs_put_page(parent, 1);
557 f2fs_put_page(npage[0], 0);
559 dn->inode_page = NULL;
560 dn->node_page = NULL;
564 static void truncate_node(struct dnode_of_data *dn)
566 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
569 get_node_info(sbi, dn->nid, &ni);
570 if (dn->inode->i_blocks == 0) {
571 f2fs_bug_on(sbi, ni.blk_addr != NULL_ADDR);
574 f2fs_bug_on(sbi, ni.blk_addr == NULL_ADDR);
576 /* Deallocate node address */
577 invalidate_blocks(sbi, ni.blk_addr);
578 dec_valid_node_count(sbi, dn->inode);
579 set_node_addr(sbi, &ni, NULL_ADDR, false);
581 if (dn->nid == dn->inode->i_ino) {
582 remove_orphan_inode(sbi, dn->nid);
583 dec_valid_inode_count(sbi);
588 clear_node_page_dirty(dn->node_page);
589 F2FS_SET_SB_DIRT(sbi);
591 f2fs_put_page(dn->node_page, 1);
593 invalidate_mapping_pages(NODE_MAPPING(sbi),
594 dn->node_page->index, dn->node_page->index);
596 dn->node_page = NULL;
597 trace_f2fs_truncate_node(dn->inode, dn->nid, ni.blk_addr);
600 static int truncate_dnode(struct dnode_of_data *dn)
607 /* get direct node */
608 page = get_node_page(F2FS_I_SB(dn->inode), dn->nid);
609 if (IS_ERR(page) && PTR_ERR(page) == -ENOENT)
611 else if (IS_ERR(page))
612 return PTR_ERR(page);
614 /* Make dnode_of_data for parameter */
615 dn->node_page = page;
617 truncate_data_blocks(dn);
622 static int truncate_nodes(struct dnode_of_data *dn, unsigned int nofs,
625 struct dnode_of_data rdn = *dn;
627 struct f2fs_node *rn;
629 unsigned int child_nofs;
634 return NIDS_PER_BLOCK + 1;
636 trace_f2fs_truncate_nodes_enter(dn->inode, dn->nid, dn->data_blkaddr);
638 page = get_node_page(F2FS_I_SB(dn->inode), dn->nid);
640 trace_f2fs_truncate_nodes_exit(dn->inode, PTR_ERR(page));
641 return PTR_ERR(page);
644 rn = F2FS_NODE(page);
646 for (i = ofs; i < NIDS_PER_BLOCK; i++, freed++) {
647 child_nid = le32_to_cpu(rn->in.nid[i]);
651 ret = truncate_dnode(&rdn);
654 set_nid(page, i, 0, false);
657 child_nofs = nofs + ofs * (NIDS_PER_BLOCK + 1) + 1;
658 for (i = ofs; i < NIDS_PER_BLOCK; i++) {
659 child_nid = le32_to_cpu(rn->in.nid[i]);
660 if (child_nid == 0) {
661 child_nofs += NIDS_PER_BLOCK + 1;
665 ret = truncate_nodes(&rdn, child_nofs, 0, depth - 1);
666 if (ret == (NIDS_PER_BLOCK + 1)) {
667 set_nid(page, i, 0, false);
669 } else if (ret < 0 && ret != -ENOENT) {
677 /* remove current indirect node */
678 dn->node_page = page;
682 f2fs_put_page(page, 1);
684 trace_f2fs_truncate_nodes_exit(dn->inode, freed);
688 f2fs_put_page(page, 1);
689 trace_f2fs_truncate_nodes_exit(dn->inode, ret);
693 static int truncate_partial_nodes(struct dnode_of_data *dn,
694 struct f2fs_inode *ri, int *offset, int depth)
696 struct page *pages[2];
703 nid[0] = le32_to_cpu(ri->i_nid[offset[0] - NODE_DIR1_BLOCK]);
707 /* get indirect nodes in the path */
708 for (i = 0; i < idx + 1; i++) {
709 /* reference count'll be increased */
710 pages[i] = get_node_page(F2FS_I_SB(dn->inode), nid[i]);
711 if (IS_ERR(pages[i])) {
712 err = PTR_ERR(pages[i]);
716 nid[i + 1] = get_nid(pages[i], offset[i + 1], false);
719 /* free direct nodes linked to a partial indirect node */
720 for (i = offset[idx + 1]; i < NIDS_PER_BLOCK; i++) {
721 child_nid = get_nid(pages[idx], i, false);
725 err = truncate_dnode(dn);
728 set_nid(pages[idx], i, 0, false);
731 if (offset[idx + 1] == 0) {
732 dn->node_page = pages[idx];
736 f2fs_put_page(pages[idx], 1);
742 for (i = idx; i >= 0; i--)
743 f2fs_put_page(pages[i], 1);
745 trace_f2fs_truncate_partial_nodes(dn->inode, nid, depth, err);
751 * All the block addresses of data and nodes should be nullified.
753 int truncate_inode_blocks(struct inode *inode, pgoff_t from)
755 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
756 int err = 0, cont = 1;
757 int level, offset[4], noffset[4];
758 unsigned int nofs = 0;
759 struct f2fs_inode *ri;
760 struct dnode_of_data dn;
763 trace_f2fs_truncate_inode_blocks_enter(inode, from);
765 level = get_node_path(F2FS_I(inode), from, offset, noffset);
767 page = get_node_page(sbi, inode->i_ino);
769 trace_f2fs_truncate_inode_blocks_exit(inode, PTR_ERR(page));
770 return PTR_ERR(page);
773 set_new_dnode(&dn, inode, page, NULL, 0);
776 ri = F2FS_INODE(page);
784 if (!offset[level - 1])
786 err = truncate_partial_nodes(&dn, ri, offset, level);
787 if (err < 0 && err != -ENOENT)
789 nofs += 1 + NIDS_PER_BLOCK;
792 nofs = 5 + 2 * NIDS_PER_BLOCK;
793 if (!offset[level - 1])
795 err = truncate_partial_nodes(&dn, ri, offset, level);
796 if (err < 0 && err != -ENOENT)
805 dn.nid = le32_to_cpu(ri->i_nid[offset[0] - NODE_DIR1_BLOCK]);
807 case NODE_DIR1_BLOCK:
808 case NODE_DIR2_BLOCK:
809 err = truncate_dnode(&dn);
812 case NODE_IND1_BLOCK:
813 case NODE_IND2_BLOCK:
814 err = truncate_nodes(&dn, nofs, offset[1], 2);
817 case NODE_DIND_BLOCK:
818 err = truncate_nodes(&dn, nofs, offset[1], 3);
825 if (err < 0 && err != -ENOENT)
827 if (offset[1] == 0 &&
828 ri->i_nid[offset[0] - NODE_DIR1_BLOCK]) {
830 if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
831 f2fs_put_page(page, 1);
834 f2fs_wait_on_page_writeback(page, NODE);
835 ri->i_nid[offset[0] - NODE_DIR1_BLOCK] = 0;
836 set_page_dirty(page);
844 f2fs_put_page(page, 0);
845 trace_f2fs_truncate_inode_blocks_exit(inode, err);
846 return err > 0 ? 0 : err;
849 int truncate_xattr_node(struct inode *inode, struct page *page)
851 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
852 nid_t nid = F2FS_I(inode)->i_xattr_nid;
853 struct dnode_of_data dn;
859 npage = get_node_page(sbi, nid);
861 return PTR_ERR(npage);
863 F2FS_I(inode)->i_xattr_nid = 0;
865 /* need to do checkpoint during fsync */
866 F2FS_I(inode)->xattr_ver = cur_cp_version(F2FS_CKPT(sbi));
868 set_new_dnode(&dn, inode, page, npage, nid);
871 dn.inode_page_locked = true;
877 * Caller should grab and release a rwsem by calling f2fs_lock_op() and
880 void remove_inode_page(struct inode *inode)
882 struct dnode_of_data dn;
884 set_new_dnode(&dn, inode, NULL, NULL, inode->i_ino);
885 if (get_dnode_of_data(&dn, 0, LOOKUP_NODE))
888 if (truncate_xattr_node(inode, dn.inode_page)) {
893 /* remove potential inline_data blocks */
894 if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
895 S_ISLNK(inode->i_mode))
896 truncate_data_blocks_range(&dn, 1);
898 /* 0 is possible, after f2fs_new_inode() has failed */
899 f2fs_bug_on(F2FS_I_SB(inode),
900 inode->i_blocks != 0 && inode->i_blocks != 1);
902 /* will put inode & node pages */
906 struct page *new_inode_page(struct inode *inode)
908 struct dnode_of_data dn;
910 /* allocate inode page for new inode */
911 set_new_dnode(&dn, inode, NULL, NULL, inode->i_ino);
913 /* caller should f2fs_put_page(page, 1); */
914 return new_node_page(&dn, 0, NULL);
917 struct page *new_node_page(struct dnode_of_data *dn,
918 unsigned int ofs, struct page *ipage)
920 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
921 struct node_info old_ni, new_ni;
925 if (unlikely(is_inode_flag_set(F2FS_I(dn->inode), FI_NO_ALLOC)))
926 return ERR_PTR(-EPERM);
928 page = grab_cache_page(NODE_MAPPING(sbi), dn->nid);
930 return ERR_PTR(-ENOMEM);
932 if (unlikely(!inc_valid_node_count(sbi, dn->inode))) {
937 get_node_info(sbi, dn->nid, &old_ni);
939 /* Reinitialize old_ni with new node page */
940 f2fs_bug_on(sbi, old_ni.blk_addr != NULL_ADDR);
942 new_ni.ino = dn->inode->i_ino;
943 set_node_addr(sbi, &new_ni, NEW_ADDR, false);
945 f2fs_wait_on_page_writeback(page, NODE);
946 fill_node_footer(page, dn->nid, dn->inode->i_ino, ofs, true);
947 set_cold_node(dn->inode, page);
948 SetPageUptodate(page);
949 set_page_dirty(page);
951 if (f2fs_has_xattr_block(ofs))
952 F2FS_I(dn->inode)->i_xattr_nid = dn->nid;
954 dn->node_page = page;
956 update_inode(dn->inode, ipage);
960 inc_valid_inode_count(sbi);
965 clear_node_page_dirty(page);
966 f2fs_put_page(page, 1);
971 * Caller should do after getting the following values.
972 * 0: f2fs_put_page(page, 0)
973 * LOCKED_PAGE: f2fs_put_page(page, 1)
976 static int read_node_page(struct page *page, int rw)
978 struct f2fs_sb_info *sbi = F2FS_P_SB(page);
981 get_node_info(sbi, page->index, &ni);
983 if (unlikely(ni.blk_addr == NULL_ADDR)) {
984 f2fs_put_page(page, 1);
988 if (PageUptodate(page))
991 return f2fs_submit_page_bio(sbi, page, ni.blk_addr, rw);
995 * Readahead a node page
997 void ra_node_page(struct f2fs_sb_info *sbi, nid_t nid)
1002 apage = find_get_page(NODE_MAPPING(sbi), nid);
1003 if (apage && PageUptodate(apage)) {
1004 f2fs_put_page(apage, 0);
1007 f2fs_put_page(apage, 0);
1009 apage = grab_cache_page(NODE_MAPPING(sbi), nid);
1013 err = read_node_page(apage, READA);
1015 f2fs_put_page(apage, 0);
1016 else if (err == LOCKED_PAGE)
1017 f2fs_put_page(apage, 1);
1020 struct page *get_node_page(struct f2fs_sb_info *sbi, pgoff_t nid)
1025 page = grab_cache_page(NODE_MAPPING(sbi), nid);
1027 return ERR_PTR(-ENOMEM);
1029 err = read_node_page(page, READ_SYNC);
1031 return ERR_PTR(err);
1032 else if (err == LOCKED_PAGE)
1036 if (unlikely(!PageUptodate(page) || nid != nid_of_node(page))) {
1037 f2fs_put_page(page, 1);
1038 return ERR_PTR(-EIO);
1040 if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1041 f2fs_put_page(page, 1);
1049 * Return a locked page for the desired node page.
1050 * And, readahead MAX_RA_NODE number of node pages.
1052 struct page *get_node_page_ra(struct page *parent, int start)
1054 struct f2fs_sb_info *sbi = F2FS_P_SB(parent);
1055 struct blk_plug plug;
1060 /* First, try getting the desired direct node. */
1061 nid = get_nid(parent, start, false);
1063 return ERR_PTR(-ENOENT);
1065 page = grab_cache_page(NODE_MAPPING(sbi), nid);
1067 return ERR_PTR(-ENOMEM);
1069 err = read_node_page(page, READ_SYNC);
1071 return ERR_PTR(err);
1072 else if (err == LOCKED_PAGE)
1075 blk_start_plug(&plug);
1077 /* Then, try readahead for siblings of the desired node */
1078 end = start + MAX_RA_NODE;
1079 end = min(end, NIDS_PER_BLOCK);
1080 for (i = start + 1; i < end; i++) {
1081 nid = get_nid(parent, i, false);
1084 ra_node_page(sbi, nid);
1087 blk_finish_plug(&plug);
1090 if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1091 f2fs_put_page(page, 1);
1095 if (unlikely(!PageUptodate(page))) {
1096 f2fs_put_page(page, 1);
1097 return ERR_PTR(-EIO);
1102 void sync_inode_page(struct dnode_of_data *dn)
1104 if (IS_INODE(dn->node_page) || dn->inode_page == dn->node_page) {
1105 update_inode(dn->inode, dn->node_page);
1106 } else if (dn->inode_page) {
1107 if (!dn->inode_page_locked)
1108 lock_page(dn->inode_page);
1109 update_inode(dn->inode, dn->inode_page);
1110 if (!dn->inode_page_locked)
1111 unlock_page(dn->inode_page);
1113 update_inode_page(dn->inode);
1117 int sync_node_pages(struct f2fs_sb_info *sbi, nid_t ino,
1118 struct writeback_control *wbc)
1121 struct pagevec pvec;
1122 int step = ino ? 2 : 0;
1123 int nwritten = 0, wrote = 0;
1125 pagevec_init(&pvec, 0);
1131 while (index <= end) {
1133 nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1134 PAGECACHE_TAG_DIRTY,
1135 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
1139 for (i = 0; i < nr_pages; i++) {
1140 struct page *page = pvec.pages[i];
1143 * flushing sequence with step:
1148 if (step == 0 && IS_DNODE(page))
1150 if (step == 1 && (!IS_DNODE(page) ||
1151 is_cold_node(page)))
1153 if (step == 2 && (!IS_DNODE(page) ||
1154 !is_cold_node(page)))
1159 * we should not skip writing node pages.
1161 if (ino && ino_of_node(page) == ino)
1163 else if (!trylock_page(page))
1166 if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1171 if (ino && ino_of_node(page) != ino)
1172 goto continue_unlock;
1174 if (!PageDirty(page)) {
1175 /* someone wrote it for us */
1176 goto continue_unlock;
1179 if (!clear_page_dirty_for_io(page))
1180 goto continue_unlock;
1182 /* called by fsync() */
1183 if (ino && IS_DNODE(page)) {
1184 set_fsync_mark(page, 1);
1185 if (IS_INODE(page)) {
1186 if (!is_checkpointed_node(sbi, ino) &&
1187 !has_fsynced_inode(sbi, ino))
1188 set_dentry_mark(page, 1);
1190 set_dentry_mark(page, 0);
1194 set_fsync_mark(page, 0);
1195 set_dentry_mark(page, 0);
1198 if (NODE_MAPPING(sbi)->a_ops->writepage(page, wbc))
1203 if (--wbc->nr_to_write == 0)
1206 pagevec_release(&pvec);
1209 if (wbc->nr_to_write == 0) {
1221 f2fs_submit_merged_bio(sbi, NODE, WRITE);
1225 int wait_on_node_pages_writeback(struct f2fs_sb_info *sbi, nid_t ino)
1227 pgoff_t index = 0, end = LONG_MAX;
1228 struct pagevec pvec;
1229 int ret2 = 0, ret = 0;
1231 pagevec_init(&pvec, 0);
1233 while (index <= end) {
1235 nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1236 PAGECACHE_TAG_WRITEBACK,
1237 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
1241 for (i = 0; i < nr_pages; i++) {
1242 struct page *page = pvec.pages[i];
1244 /* until radix tree lookup accepts end_index */
1245 if (unlikely(page->index > end))
1248 if (ino && ino_of_node(page) == ino) {
1249 f2fs_wait_on_page_writeback(page, NODE);
1250 if (TestClearPageError(page))
1254 pagevec_release(&pvec);
1258 if (unlikely(test_and_clear_bit(AS_ENOSPC, &NODE_MAPPING(sbi)->flags)))
1260 if (unlikely(test_and_clear_bit(AS_EIO, &NODE_MAPPING(sbi)->flags)))
1267 static int f2fs_write_node_page(struct page *page,
1268 struct writeback_control *wbc)
1270 struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1273 struct node_info ni;
1274 struct f2fs_io_info fio = {
1276 .rw = (wbc->sync_mode == WB_SYNC_ALL) ? WRITE_SYNC : WRITE,
1279 trace_f2fs_writepage(page, NODE);
1281 if (unlikely(sbi->por_doing))
1283 if (unlikely(f2fs_cp_error(sbi)))
1286 f2fs_wait_on_page_writeback(page, NODE);
1288 /* get old block addr of this node page */
1289 nid = nid_of_node(page);
1290 f2fs_bug_on(sbi, page->index != nid);
1292 get_node_info(sbi, nid, &ni);
1294 /* This page is already truncated */
1295 if (unlikely(ni.blk_addr == NULL_ADDR)) {
1296 dec_page_count(sbi, F2FS_DIRTY_NODES);
1301 if (wbc->for_reclaim)
1304 down_read(&sbi->node_write);
1305 set_page_writeback(page);
1306 write_node_page(sbi, page, &fio, nid, ni.blk_addr, &new_addr);
1307 set_node_addr(sbi, &ni, new_addr, is_fsync_dnode(page));
1308 dec_page_count(sbi, F2FS_DIRTY_NODES);
1309 up_read(&sbi->node_write);
1314 redirty_page_for_writepage(wbc, page);
1315 return AOP_WRITEPAGE_ACTIVATE;
1318 static int f2fs_write_node_pages(struct address_space *mapping,
1319 struct writeback_control *wbc)
1321 struct f2fs_sb_info *sbi = F2FS_M_SB(mapping);
1324 trace_f2fs_writepages(mapping->host, wbc, NODE);
1326 /* balancing f2fs's metadata in background */
1327 f2fs_balance_fs_bg(sbi);
1329 /* collect a number of dirty node pages and write together */
1330 if (get_pages(sbi, F2FS_DIRTY_NODES) < nr_pages_to_skip(sbi, NODE))
1333 diff = nr_pages_to_write(sbi, NODE, wbc);
1334 wbc->sync_mode = WB_SYNC_NONE;
1335 sync_node_pages(sbi, 0, wbc);
1336 wbc->nr_to_write = max((long)0, wbc->nr_to_write - diff);
1340 wbc->pages_skipped += get_pages(sbi, F2FS_DIRTY_NODES);
1344 static int f2fs_set_node_page_dirty(struct page *page)
1346 trace_f2fs_set_page_dirty(page, NODE);
1348 SetPageUptodate(page);
1349 if (!PageDirty(page)) {
1350 __set_page_dirty_nobuffers(page);
1351 inc_page_count(F2FS_P_SB(page), F2FS_DIRTY_NODES);
1352 SetPagePrivate(page);
1358 static void f2fs_invalidate_node_page(struct page *page, unsigned int offset,
1359 unsigned int length)
1361 struct inode *inode = page->mapping->host;
1362 if (PageDirty(page))
1363 dec_page_count(F2FS_I_SB(inode), F2FS_DIRTY_NODES);
1364 ClearPagePrivate(page);
1367 static int f2fs_release_node_page(struct page *page, gfp_t wait)
1369 ClearPagePrivate(page);
1374 * Structure of the f2fs node operations
1376 const struct address_space_operations f2fs_node_aops = {
1377 .writepage = f2fs_write_node_page,
1378 .writepages = f2fs_write_node_pages,
1379 .set_page_dirty = f2fs_set_node_page_dirty,
1380 .invalidatepage = f2fs_invalidate_node_page,
1381 .releasepage = f2fs_release_node_page,
1384 static struct free_nid *__lookup_free_nid_list(struct f2fs_nm_info *nm_i,
1387 return radix_tree_lookup(&nm_i->free_nid_root, n);
1390 static void __del_from_free_nid_list(struct f2fs_nm_info *nm_i,
1394 radix_tree_delete(&nm_i->free_nid_root, i->nid);
1397 static int add_free_nid(struct f2fs_sb_info *sbi, nid_t nid, bool build)
1399 struct f2fs_nm_info *nm_i = NM_I(sbi);
1401 struct nat_entry *ne;
1402 bool allocated = false;
1404 if (!available_free_memory(sbi, FREE_NIDS))
1407 /* 0 nid should not be used */
1408 if (unlikely(nid == 0))
1412 /* do not add allocated nids */
1413 read_lock(&nm_i->nat_tree_lock);
1414 ne = __lookup_nat_cache(nm_i, nid);
1416 (!get_nat_flag(ne, IS_CHECKPOINTED) ||
1417 nat_get_blkaddr(ne) != NULL_ADDR))
1419 read_unlock(&nm_i->nat_tree_lock);
1424 i = f2fs_kmem_cache_alloc(free_nid_slab, GFP_NOFS);
1428 spin_lock(&nm_i->free_nid_list_lock);
1429 if (radix_tree_insert(&nm_i->free_nid_root, i->nid, i)) {
1430 spin_unlock(&nm_i->free_nid_list_lock);
1431 kmem_cache_free(free_nid_slab, i);
1434 list_add_tail(&i->list, &nm_i->free_nid_list);
1436 spin_unlock(&nm_i->free_nid_list_lock);
1440 static void remove_free_nid(struct f2fs_nm_info *nm_i, nid_t nid)
1443 bool need_free = false;
1445 spin_lock(&nm_i->free_nid_list_lock);
1446 i = __lookup_free_nid_list(nm_i, nid);
1447 if (i && i->state == NID_NEW) {
1448 __del_from_free_nid_list(nm_i, i);
1452 spin_unlock(&nm_i->free_nid_list_lock);
1455 kmem_cache_free(free_nid_slab, i);
1458 static void scan_nat_page(struct f2fs_sb_info *sbi,
1459 struct page *nat_page, nid_t start_nid)
1461 struct f2fs_nm_info *nm_i = NM_I(sbi);
1462 struct f2fs_nat_block *nat_blk = page_address(nat_page);
1466 i = start_nid % NAT_ENTRY_PER_BLOCK;
1468 for (; i < NAT_ENTRY_PER_BLOCK; i++, start_nid++) {
1470 if (unlikely(start_nid >= nm_i->max_nid))
1473 blk_addr = le32_to_cpu(nat_blk->entries[i].block_addr);
1474 f2fs_bug_on(sbi, blk_addr == NEW_ADDR);
1475 if (blk_addr == NULL_ADDR) {
1476 if (add_free_nid(sbi, start_nid, true) < 0)
1482 static void build_free_nids(struct f2fs_sb_info *sbi)
1484 struct f2fs_nm_info *nm_i = NM_I(sbi);
1485 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
1486 struct f2fs_summary_block *sum = curseg->sum_blk;
1488 nid_t nid = nm_i->next_scan_nid;
1490 /* Enough entries */
1491 if (nm_i->fcnt > NAT_ENTRY_PER_BLOCK)
1494 /* readahead nat pages to be scanned */
1495 ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nid), FREE_NID_PAGES, META_NAT);
1498 struct page *page = get_current_nat_page(sbi, nid);
1500 scan_nat_page(sbi, page, nid);
1501 f2fs_put_page(page, 1);
1503 nid += (NAT_ENTRY_PER_BLOCK - (nid % NAT_ENTRY_PER_BLOCK));
1504 if (unlikely(nid >= nm_i->max_nid))
1507 if (i++ == FREE_NID_PAGES)
1511 /* go to the next free nat pages to find free nids abundantly */
1512 nm_i->next_scan_nid = nid;
1514 /* find free nids from current sum_pages */
1515 mutex_lock(&curseg->curseg_mutex);
1516 for (i = 0; i < nats_in_cursum(sum); i++) {
1517 block_t addr = le32_to_cpu(nat_in_journal(sum, i).block_addr);
1518 nid = le32_to_cpu(nid_in_journal(sum, i));
1519 if (addr == NULL_ADDR)
1520 add_free_nid(sbi, nid, true);
1522 remove_free_nid(nm_i, nid);
1524 mutex_unlock(&curseg->curseg_mutex);
1528 * If this function returns success, caller can obtain a new nid
1529 * from second parameter of this function.
1530 * The returned nid could be used ino as well as nid when inode is created.
1532 bool alloc_nid(struct f2fs_sb_info *sbi, nid_t *nid)
1534 struct f2fs_nm_info *nm_i = NM_I(sbi);
1535 struct free_nid *i = NULL;
1537 if (unlikely(sbi->total_valid_node_count + 1 > nm_i->available_nids))
1540 spin_lock(&nm_i->free_nid_list_lock);
1542 /* We should not use stale free nids created by build_free_nids */
1543 if (nm_i->fcnt && !on_build_free_nids(nm_i)) {
1544 f2fs_bug_on(sbi, list_empty(&nm_i->free_nid_list));
1545 list_for_each_entry(i, &nm_i->free_nid_list, list)
1546 if (i->state == NID_NEW)
1549 f2fs_bug_on(sbi, i->state != NID_NEW);
1551 i->state = NID_ALLOC;
1553 spin_unlock(&nm_i->free_nid_list_lock);
1556 spin_unlock(&nm_i->free_nid_list_lock);
1558 /* Let's scan nat pages and its caches to get free nids */
1559 mutex_lock(&nm_i->build_lock);
1560 build_free_nids(sbi);
1561 mutex_unlock(&nm_i->build_lock);
1566 * alloc_nid() should be called prior to this function.
1568 void alloc_nid_done(struct f2fs_sb_info *sbi, nid_t nid)
1570 struct f2fs_nm_info *nm_i = NM_I(sbi);
1573 spin_lock(&nm_i->free_nid_list_lock);
1574 i = __lookup_free_nid_list(nm_i, nid);
1575 f2fs_bug_on(sbi, !i || i->state != NID_ALLOC);
1576 __del_from_free_nid_list(nm_i, i);
1577 spin_unlock(&nm_i->free_nid_list_lock);
1579 kmem_cache_free(free_nid_slab, i);
1583 * alloc_nid() should be called prior to this function.
1585 void alloc_nid_failed(struct f2fs_sb_info *sbi, nid_t nid)
1587 struct f2fs_nm_info *nm_i = NM_I(sbi);
1589 bool need_free = false;
1594 spin_lock(&nm_i->free_nid_list_lock);
1595 i = __lookup_free_nid_list(nm_i, nid);
1596 f2fs_bug_on(sbi, !i || i->state != NID_ALLOC);
1597 if (!available_free_memory(sbi, FREE_NIDS)) {
1598 __del_from_free_nid_list(nm_i, i);
1604 spin_unlock(&nm_i->free_nid_list_lock);
1607 kmem_cache_free(free_nid_slab, i);
1610 void recover_inline_xattr(struct inode *inode, struct page *page)
1612 void *src_addr, *dst_addr;
1615 struct f2fs_inode *ri;
1617 ipage = get_node_page(F2FS_I_SB(inode), inode->i_ino);
1618 f2fs_bug_on(F2FS_I_SB(inode), IS_ERR(ipage));
1620 ri = F2FS_INODE(page);
1621 if (!(ri->i_inline & F2FS_INLINE_XATTR)) {
1622 clear_inode_flag(F2FS_I(inode), FI_INLINE_XATTR);
1626 dst_addr = inline_xattr_addr(ipage);
1627 src_addr = inline_xattr_addr(page);
1628 inline_size = inline_xattr_size(inode);
1630 f2fs_wait_on_page_writeback(ipage, NODE);
1631 memcpy(dst_addr, src_addr, inline_size);
1633 update_inode(inode, ipage);
1634 f2fs_put_page(ipage, 1);
1637 void recover_xattr_data(struct inode *inode, struct page *page, block_t blkaddr)
1639 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1640 nid_t prev_xnid = F2FS_I(inode)->i_xattr_nid;
1641 nid_t new_xnid = nid_of_node(page);
1642 struct node_info ni;
1644 /* 1: invalidate the previous xattr nid */
1648 /* Deallocate node address */
1649 get_node_info(sbi, prev_xnid, &ni);
1650 f2fs_bug_on(sbi, ni.blk_addr == NULL_ADDR);
1651 invalidate_blocks(sbi, ni.blk_addr);
1652 dec_valid_node_count(sbi, inode);
1653 set_node_addr(sbi, &ni, NULL_ADDR, false);
1656 /* 2: allocate new xattr nid */
1657 if (unlikely(!inc_valid_node_count(sbi, inode)))
1658 f2fs_bug_on(sbi, 1);
1660 remove_free_nid(NM_I(sbi), new_xnid);
1661 get_node_info(sbi, new_xnid, &ni);
1662 ni.ino = inode->i_ino;
1663 set_node_addr(sbi, &ni, NEW_ADDR, false);
1664 F2FS_I(inode)->i_xattr_nid = new_xnid;
1666 /* 3: update xattr blkaddr */
1667 refresh_sit_entry(sbi, NEW_ADDR, blkaddr);
1668 set_node_addr(sbi, &ni, blkaddr, false);
1670 update_inode_page(inode);
1673 int recover_inode_page(struct f2fs_sb_info *sbi, struct page *page)
1675 struct f2fs_inode *src, *dst;
1676 nid_t ino = ino_of_node(page);
1677 struct node_info old_ni, new_ni;
1680 get_node_info(sbi, ino, &old_ni);
1682 if (unlikely(old_ni.blk_addr != NULL_ADDR))
1685 ipage = grab_cache_page(NODE_MAPPING(sbi), ino);
1689 /* Should not use this inode from free nid list */
1690 remove_free_nid(NM_I(sbi), ino);
1692 SetPageUptodate(ipage);
1693 fill_node_footer(ipage, ino, ino, 0, true);
1695 src = F2FS_INODE(page);
1696 dst = F2FS_INODE(ipage);
1698 memcpy(dst, src, (unsigned long)&src->i_ext - (unsigned long)src);
1700 dst->i_blocks = cpu_to_le64(1);
1701 dst->i_links = cpu_to_le32(1);
1702 dst->i_xattr_nid = 0;
1703 dst->i_inline = src->i_inline & F2FS_INLINE_XATTR;
1708 if (unlikely(!inc_valid_node_count(sbi, NULL)))
1710 set_node_addr(sbi, &new_ni, NEW_ADDR, false);
1711 inc_valid_inode_count(sbi);
1712 set_page_dirty(ipage);
1713 f2fs_put_page(ipage, 1);
1718 * ra_sum_pages() merge contiguous pages into one bio and submit.
1719 * these pre-read pages are allocated in bd_inode's mapping tree.
1721 static int ra_sum_pages(struct f2fs_sb_info *sbi, struct page **pages,
1722 int start, int nrpages)
1724 struct inode *inode = sbi->sb->s_bdev->bd_inode;
1725 struct address_space *mapping = inode->i_mapping;
1726 int i, page_idx = start;
1727 struct f2fs_io_info fio = {
1729 .rw = READ_SYNC | REQ_META | REQ_PRIO
1732 for (i = 0; page_idx < start + nrpages; page_idx++, i++) {
1733 /* alloc page in bd_inode for reading node summary info */
1734 pages[i] = grab_cache_page(mapping, page_idx);
1737 f2fs_submit_page_mbio(sbi, pages[i], page_idx, &fio);
1740 f2fs_submit_merged_bio(sbi, META, READ);
1744 int restore_node_summary(struct f2fs_sb_info *sbi,
1745 unsigned int segno, struct f2fs_summary_block *sum)
1747 struct f2fs_node *rn;
1748 struct f2fs_summary *sum_entry;
1749 struct inode *inode = sbi->sb->s_bdev->bd_inode;
1751 int bio_blocks = MAX_BIO_BLOCKS(sbi);
1752 struct page *pages[bio_blocks];
1753 int i, idx, last_offset, nrpages, err = 0;
1755 /* scan the node segment */
1756 last_offset = sbi->blocks_per_seg;
1757 addr = START_BLOCK(sbi, segno);
1758 sum_entry = &sum->entries[0];
1760 for (i = 0; !err && i < last_offset; i += nrpages, addr += nrpages) {
1761 nrpages = min(last_offset - i, bio_blocks);
1763 /* readahead node pages */
1764 nrpages = ra_sum_pages(sbi, pages, addr, nrpages);
1768 for (idx = 0; idx < nrpages; idx++) {
1772 lock_page(pages[idx]);
1773 if (unlikely(!PageUptodate(pages[idx]))) {
1776 rn = F2FS_NODE(pages[idx]);
1777 sum_entry->nid = rn->footer.nid;
1778 sum_entry->version = 0;
1779 sum_entry->ofs_in_node = 0;
1782 unlock_page(pages[idx]);
1784 page_cache_release(pages[idx]);
1787 invalidate_mapping_pages(inode->i_mapping, addr,
1793 static void remove_nats_in_journal(struct f2fs_sb_info *sbi)
1795 struct f2fs_nm_info *nm_i = NM_I(sbi);
1796 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
1797 struct f2fs_summary_block *sum = curseg->sum_blk;
1800 mutex_lock(&curseg->curseg_mutex);
1801 for (i = 0; i < nats_in_cursum(sum); i++) {
1802 struct nat_entry *ne;
1803 struct f2fs_nat_entry raw_ne;
1804 nid_t nid = le32_to_cpu(nid_in_journal(sum, i));
1806 raw_ne = nat_in_journal(sum, i);
1808 write_lock(&nm_i->nat_tree_lock);
1809 ne = __lookup_nat_cache(nm_i, nid);
1813 ne = grab_nat_entry(nm_i, nid);
1815 write_unlock(&nm_i->nat_tree_lock);
1818 node_info_from_raw_nat(&ne->ni, &raw_ne);
1820 __set_nat_cache_dirty(nm_i, ne);
1821 write_unlock(&nm_i->nat_tree_lock);
1823 update_nats_in_cursum(sum, -i);
1824 mutex_unlock(&curseg->curseg_mutex);
1827 static void __adjust_nat_entry_set(struct nat_entry_set *nes,
1828 struct list_head *head, int max)
1830 struct nat_entry_set *cur;
1832 if (nes->entry_cnt >= max)
1835 list_for_each_entry(cur, head, set_list) {
1836 if (cur->entry_cnt >= nes->entry_cnt) {
1837 list_add(&nes->set_list, cur->set_list.prev);
1842 list_add_tail(&nes->set_list, head);
1845 static void __flush_nat_entry_set(struct f2fs_sb_info *sbi,
1846 struct nat_entry_set *set)
1848 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
1849 struct f2fs_summary_block *sum = curseg->sum_blk;
1850 nid_t start_nid = set->set * NAT_ENTRY_PER_BLOCK;
1851 bool to_journal = true;
1852 struct f2fs_nat_block *nat_blk;
1853 struct nat_entry *ne, *cur;
1854 struct page *page = NULL;
1857 * there are two steps to flush nat entries:
1858 * #1, flush nat entries to journal in current hot data summary block.
1859 * #2, flush nat entries to nat page.
1861 if (!__has_cursum_space(sum, set->entry_cnt, NAT_JOURNAL))
1865 mutex_lock(&curseg->curseg_mutex);
1867 page = get_next_nat_page(sbi, start_nid);
1868 nat_blk = page_address(page);
1869 f2fs_bug_on(sbi, !nat_blk);
1872 /* flush dirty nats in nat entry set */
1873 list_for_each_entry_safe(ne, cur, &set->entry_list, list) {
1874 struct f2fs_nat_entry *raw_ne;
1875 nid_t nid = nat_get_nid(ne);
1878 if (nat_get_blkaddr(ne) == NEW_ADDR)
1882 offset = lookup_journal_in_cursum(sum,
1883 NAT_JOURNAL, nid, 1);
1884 f2fs_bug_on(sbi, offset < 0);
1885 raw_ne = &nat_in_journal(sum, offset);
1886 nid_in_journal(sum, offset) = cpu_to_le32(nid);
1888 raw_ne = &nat_blk->entries[nid - start_nid];
1890 raw_nat_from_node_info(raw_ne, &ne->ni);
1892 write_lock(&NM_I(sbi)->nat_tree_lock);
1894 __clear_nat_cache_dirty(NM_I(sbi), ne);
1895 write_unlock(&NM_I(sbi)->nat_tree_lock);
1897 if (nat_get_blkaddr(ne) == NULL_ADDR)
1898 add_free_nid(sbi, nid, false);
1902 mutex_unlock(&curseg->curseg_mutex);
1904 f2fs_put_page(page, 1);
1906 if (!set->entry_cnt) {
1907 radix_tree_delete(&NM_I(sbi)->nat_set_root, set->set);
1908 kmem_cache_free(nat_entry_set_slab, set);
1913 * This function is called during the checkpointing process.
1915 void flush_nat_entries(struct f2fs_sb_info *sbi)
1917 struct f2fs_nm_info *nm_i = NM_I(sbi);
1918 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
1919 struct f2fs_summary_block *sum = curseg->sum_blk;
1920 struct nat_entry_set *setvec[NATVEC_SIZE];
1921 struct nat_entry_set *set, *tmp;
1927 * if there are no enough space in journal to store dirty nat
1928 * entries, remove all entries from journal and merge them
1929 * into nat entry set.
1931 if (!__has_cursum_space(sum, nm_i->dirty_nat_cnt, NAT_JOURNAL))
1932 remove_nats_in_journal(sbi);
1934 if (!nm_i->dirty_nat_cnt)
1937 while ((found = __gang_lookup_nat_set(nm_i,
1938 set_idx, NATVEC_SIZE, setvec))) {
1940 set_idx = setvec[found - 1]->set + 1;
1941 for (idx = 0; idx < found; idx++)
1942 __adjust_nat_entry_set(setvec[idx], &sets,
1943 MAX_NAT_JENTRIES(sum));
1946 /* flush dirty nats in nat entry set */
1947 list_for_each_entry_safe(set, tmp, &sets, set_list)
1948 __flush_nat_entry_set(sbi, set);
1950 f2fs_bug_on(sbi, nm_i->dirty_nat_cnt);
1953 static int init_node_manager(struct f2fs_sb_info *sbi)
1955 struct f2fs_super_block *sb_raw = F2FS_RAW_SUPER(sbi);
1956 struct f2fs_nm_info *nm_i = NM_I(sbi);
1957 unsigned char *version_bitmap;
1958 unsigned int nat_segs, nat_blocks;
1960 nm_i->nat_blkaddr = le32_to_cpu(sb_raw->nat_blkaddr);
1962 /* segment_count_nat includes pair segment so divide to 2. */
1963 nat_segs = le32_to_cpu(sb_raw->segment_count_nat) >> 1;
1964 nat_blocks = nat_segs << le32_to_cpu(sb_raw->log_blocks_per_seg);
1966 nm_i->max_nid = NAT_ENTRY_PER_BLOCK * nat_blocks;
1968 /* not used nids: 0, node, meta, (and root counted as valid node) */
1969 nm_i->available_nids = nm_i->max_nid - F2FS_RESERVED_NODE_NUM;
1972 nm_i->ram_thresh = DEF_RAM_THRESHOLD;
1974 INIT_RADIX_TREE(&nm_i->free_nid_root, GFP_ATOMIC);
1975 INIT_LIST_HEAD(&nm_i->free_nid_list);
1976 INIT_RADIX_TREE(&nm_i->nat_root, GFP_ATOMIC);
1977 INIT_RADIX_TREE(&nm_i->nat_set_root, GFP_ATOMIC);
1978 INIT_LIST_HEAD(&nm_i->nat_entries);
1980 mutex_init(&nm_i->build_lock);
1981 spin_lock_init(&nm_i->free_nid_list_lock);
1982 rwlock_init(&nm_i->nat_tree_lock);
1984 nm_i->next_scan_nid = le32_to_cpu(sbi->ckpt->next_free_nid);
1985 nm_i->bitmap_size = __bitmap_size(sbi, NAT_BITMAP);
1986 version_bitmap = __bitmap_ptr(sbi, NAT_BITMAP);
1987 if (!version_bitmap)
1990 nm_i->nat_bitmap = kmemdup(version_bitmap, nm_i->bitmap_size,
1992 if (!nm_i->nat_bitmap)
1997 int build_node_manager(struct f2fs_sb_info *sbi)
2001 sbi->nm_info = kzalloc(sizeof(struct f2fs_nm_info), GFP_KERNEL);
2005 err = init_node_manager(sbi);
2009 build_free_nids(sbi);
2013 void destroy_node_manager(struct f2fs_sb_info *sbi)
2015 struct f2fs_nm_info *nm_i = NM_I(sbi);
2016 struct free_nid *i, *next_i;
2017 struct nat_entry *natvec[NATVEC_SIZE];
2024 /* destroy free nid list */
2025 spin_lock(&nm_i->free_nid_list_lock);
2026 list_for_each_entry_safe(i, next_i, &nm_i->free_nid_list, list) {
2027 f2fs_bug_on(sbi, i->state == NID_ALLOC);
2028 __del_from_free_nid_list(nm_i, i);
2030 spin_unlock(&nm_i->free_nid_list_lock);
2031 kmem_cache_free(free_nid_slab, i);
2032 spin_lock(&nm_i->free_nid_list_lock);
2034 f2fs_bug_on(sbi, nm_i->fcnt);
2035 spin_unlock(&nm_i->free_nid_list_lock);
2037 /* destroy nat cache */
2038 write_lock(&nm_i->nat_tree_lock);
2039 while ((found = __gang_lookup_nat_cache(nm_i,
2040 nid, NATVEC_SIZE, natvec))) {
2042 nid = nat_get_nid(natvec[found - 1]) + 1;
2043 for (idx = 0; idx < found; idx++)
2044 __del_from_nat_cache(nm_i, natvec[idx]);
2046 f2fs_bug_on(sbi, nm_i->nat_cnt);
2047 write_unlock(&nm_i->nat_tree_lock);
2049 kfree(nm_i->nat_bitmap);
2050 sbi->nm_info = NULL;
2054 int __init create_node_manager_caches(void)
2056 nat_entry_slab = f2fs_kmem_cache_create("nat_entry",
2057 sizeof(struct nat_entry));
2058 if (!nat_entry_slab)
2061 free_nid_slab = f2fs_kmem_cache_create("free_nid",
2062 sizeof(struct free_nid));
2064 goto destory_nat_entry;
2066 nat_entry_set_slab = f2fs_kmem_cache_create("nat_entry_set",
2067 sizeof(struct nat_entry_set));
2068 if (!nat_entry_set_slab)
2069 goto destory_free_nid;
2073 kmem_cache_destroy(free_nid_slab);
2075 kmem_cache_destroy(nat_entry_slab);
2080 void destroy_node_manager_caches(void)
2082 kmem_cache_destroy(nat_entry_set_slab);
2083 kmem_cache_destroy(free_nid_slab);
2084 kmem_cache_destroy(nat_entry_slab);