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 struct f2fs_sb_info *sbi = F2FS_SB(mapping->host->i_sb);
58 unsigned int long flags;
60 if (PageDirty(page)) {
61 spin_lock_irqsave(&mapping->tree_lock, flags);
62 radix_tree_tag_clear(&mapping->page_tree,
65 spin_unlock_irqrestore(&mapping->tree_lock, flags);
67 clear_page_dirty_for_io(page);
68 dec_page_count(sbi, F2FS_DIRTY_NODES);
70 ClearPageUptodate(page);
73 static struct page *get_current_nat_page(struct f2fs_sb_info *sbi, nid_t nid)
75 pgoff_t index = current_nat_addr(sbi, nid);
76 return get_meta_page(sbi, index);
79 static struct page *get_next_nat_page(struct f2fs_sb_info *sbi, nid_t nid)
81 struct page *src_page;
82 struct page *dst_page;
87 struct f2fs_nm_info *nm_i = NM_I(sbi);
89 src_off = current_nat_addr(sbi, nid);
90 dst_off = next_nat_addr(sbi, src_off);
92 /* get current nat block page with lock */
93 src_page = get_meta_page(sbi, src_off);
94 dst_page = grab_meta_page(sbi, dst_off);
95 f2fs_bug_on(PageDirty(src_page));
97 src_addr = page_address(src_page);
98 dst_addr = page_address(dst_page);
99 memcpy(dst_addr, src_addr, PAGE_CACHE_SIZE);
100 set_page_dirty(dst_page);
101 f2fs_put_page(src_page, 1);
103 set_to_next_nat(nm_i, nid);
108 static struct nat_entry *__lookup_nat_cache(struct f2fs_nm_info *nm_i, nid_t n)
110 return radix_tree_lookup(&nm_i->nat_root, n);
113 static unsigned int __gang_lookup_nat_cache(struct f2fs_nm_info *nm_i,
114 nid_t start, unsigned int nr, struct nat_entry **ep)
116 return radix_tree_gang_lookup(&nm_i->nat_root, (void **)ep, start, nr);
119 static void __del_from_nat_cache(struct f2fs_nm_info *nm_i, struct nat_entry *e)
122 radix_tree_delete(&nm_i->nat_root, nat_get_nid(e));
124 kmem_cache_free(nat_entry_slab, e);
127 int is_checkpointed_node(struct f2fs_sb_info *sbi, nid_t nid)
129 struct f2fs_nm_info *nm_i = NM_I(sbi);
133 read_lock(&nm_i->nat_tree_lock);
134 e = __lookup_nat_cache(nm_i, nid);
135 if (e && !e->checkpointed)
137 read_unlock(&nm_i->nat_tree_lock);
141 bool fsync_mark_done(struct f2fs_sb_info *sbi, nid_t nid)
143 struct f2fs_nm_info *nm_i = NM_I(sbi);
145 bool fsync_done = false;
147 read_lock(&nm_i->nat_tree_lock);
148 e = __lookup_nat_cache(nm_i, nid);
150 fsync_done = e->fsync_done;
151 read_unlock(&nm_i->nat_tree_lock);
155 void fsync_mark_clear(struct f2fs_sb_info *sbi, nid_t nid)
157 struct f2fs_nm_info *nm_i = NM_I(sbi);
160 write_lock(&nm_i->nat_tree_lock);
161 e = __lookup_nat_cache(nm_i, nid);
163 e->fsync_done = false;
164 write_unlock(&nm_i->nat_tree_lock);
167 static struct nat_entry *grab_nat_entry(struct f2fs_nm_info *nm_i, nid_t nid)
169 struct nat_entry *new;
171 new = kmem_cache_alloc(nat_entry_slab, GFP_ATOMIC);
174 if (radix_tree_insert(&nm_i->nat_root, nid, new)) {
175 kmem_cache_free(nat_entry_slab, new);
178 memset(new, 0, sizeof(struct nat_entry));
179 nat_set_nid(new, nid);
180 new->checkpointed = true;
181 list_add_tail(&new->list, &nm_i->nat_entries);
186 static void cache_nat_entry(struct f2fs_nm_info *nm_i, nid_t nid,
187 struct f2fs_nat_entry *ne)
191 write_lock(&nm_i->nat_tree_lock);
192 e = __lookup_nat_cache(nm_i, nid);
194 e = grab_nat_entry(nm_i, nid);
196 write_unlock(&nm_i->nat_tree_lock);
199 node_info_from_raw_nat(&e->ni, ne);
201 write_unlock(&nm_i->nat_tree_lock);
204 static void set_node_addr(struct f2fs_sb_info *sbi, struct node_info *ni,
205 block_t new_blkaddr, bool fsync_done)
207 struct f2fs_nm_info *nm_i = NM_I(sbi);
210 write_lock(&nm_i->nat_tree_lock);
211 e = __lookup_nat_cache(nm_i, ni->nid);
213 e = grab_nat_entry(nm_i, ni->nid);
215 write_unlock(&nm_i->nat_tree_lock);
219 f2fs_bug_on(ni->blk_addr == NEW_ADDR);
220 } else if (new_blkaddr == NEW_ADDR) {
222 * when nid is reallocated,
223 * previous nat entry can be remained in nat cache.
224 * So, reinitialize it with new information.
227 f2fs_bug_on(ni->blk_addr != NULL_ADDR);
231 f2fs_bug_on(nat_get_blkaddr(e) != ni->blk_addr);
232 f2fs_bug_on(nat_get_blkaddr(e) == NULL_ADDR &&
233 new_blkaddr == NULL_ADDR);
234 f2fs_bug_on(nat_get_blkaddr(e) == NEW_ADDR &&
235 new_blkaddr == NEW_ADDR);
236 f2fs_bug_on(nat_get_blkaddr(e) != NEW_ADDR &&
237 nat_get_blkaddr(e) != NULL_ADDR &&
238 new_blkaddr == NEW_ADDR);
240 /* increment version no as node is removed */
241 if (nat_get_blkaddr(e) != NEW_ADDR && new_blkaddr == NULL_ADDR) {
242 unsigned char version = nat_get_version(e);
243 nat_set_version(e, inc_node_version(version));
247 nat_set_blkaddr(e, new_blkaddr);
248 __set_nat_cache_dirty(nm_i, e);
250 /* update fsync_mark if its inode nat entry is still alive */
251 e = __lookup_nat_cache(nm_i, ni->ino);
253 e->fsync_done = fsync_done;
254 write_unlock(&nm_i->nat_tree_lock);
257 int try_to_free_nats(struct f2fs_sb_info *sbi, int nr_shrink)
259 struct f2fs_nm_info *nm_i = NM_I(sbi);
261 if (available_free_memory(sbi, NAT_ENTRIES))
264 write_lock(&nm_i->nat_tree_lock);
265 while (nr_shrink && !list_empty(&nm_i->nat_entries)) {
266 struct nat_entry *ne;
267 ne = list_first_entry(&nm_i->nat_entries,
268 struct nat_entry, list);
269 __del_from_nat_cache(nm_i, ne);
272 write_unlock(&nm_i->nat_tree_lock);
277 * This function always returns success
279 void get_node_info(struct f2fs_sb_info *sbi, nid_t nid, struct node_info *ni)
281 struct f2fs_nm_info *nm_i = NM_I(sbi);
282 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
283 struct f2fs_summary_block *sum = curseg->sum_blk;
284 nid_t start_nid = START_NID(nid);
285 struct f2fs_nat_block *nat_blk;
286 struct page *page = NULL;
287 struct f2fs_nat_entry ne;
291 memset(&ne, 0, sizeof(struct f2fs_nat_entry));
294 /* Check nat cache */
295 read_lock(&nm_i->nat_tree_lock);
296 e = __lookup_nat_cache(nm_i, nid);
298 ni->ino = nat_get_ino(e);
299 ni->blk_addr = nat_get_blkaddr(e);
300 ni->version = nat_get_version(e);
302 read_unlock(&nm_i->nat_tree_lock);
306 /* Check current segment summary */
307 mutex_lock(&curseg->curseg_mutex);
308 i = lookup_journal_in_cursum(sum, NAT_JOURNAL, nid, 0);
310 ne = nat_in_journal(sum, i);
311 node_info_from_raw_nat(ni, &ne);
313 mutex_unlock(&curseg->curseg_mutex);
317 /* Fill node_info from nat page */
318 page = get_current_nat_page(sbi, start_nid);
319 nat_blk = (struct f2fs_nat_block *)page_address(page);
320 ne = nat_blk->entries[nid - start_nid];
321 node_info_from_raw_nat(ni, &ne);
322 f2fs_put_page(page, 1);
324 /* cache nat entry */
325 cache_nat_entry(NM_I(sbi), nid, &ne);
329 * The maximum depth is four.
330 * Offset[0] will have raw inode offset.
332 static int get_node_path(struct f2fs_inode_info *fi, long block,
333 int offset[4], unsigned int noffset[4])
335 const long direct_index = ADDRS_PER_INODE(fi);
336 const long direct_blks = ADDRS_PER_BLOCK;
337 const long dptrs_per_blk = NIDS_PER_BLOCK;
338 const long indirect_blks = ADDRS_PER_BLOCK * NIDS_PER_BLOCK;
339 const long dindirect_blks = indirect_blks * NIDS_PER_BLOCK;
345 if (block < direct_index) {
349 block -= direct_index;
350 if (block < direct_blks) {
351 offset[n++] = NODE_DIR1_BLOCK;
357 block -= direct_blks;
358 if (block < direct_blks) {
359 offset[n++] = NODE_DIR2_BLOCK;
365 block -= direct_blks;
366 if (block < indirect_blks) {
367 offset[n++] = NODE_IND1_BLOCK;
369 offset[n++] = block / direct_blks;
370 noffset[n] = 4 + offset[n - 1];
371 offset[n] = block % direct_blks;
375 block -= indirect_blks;
376 if (block < indirect_blks) {
377 offset[n++] = NODE_IND2_BLOCK;
378 noffset[n] = 4 + dptrs_per_blk;
379 offset[n++] = block / direct_blks;
380 noffset[n] = 5 + dptrs_per_blk + offset[n - 1];
381 offset[n] = block % direct_blks;
385 block -= indirect_blks;
386 if (block < dindirect_blks) {
387 offset[n++] = NODE_DIND_BLOCK;
388 noffset[n] = 5 + (dptrs_per_blk * 2);
389 offset[n++] = block / indirect_blks;
390 noffset[n] = 6 + (dptrs_per_blk * 2) +
391 offset[n - 1] * (dptrs_per_blk + 1);
392 offset[n++] = (block / direct_blks) % dptrs_per_blk;
393 noffset[n] = 7 + (dptrs_per_blk * 2) +
394 offset[n - 2] * (dptrs_per_blk + 1) +
396 offset[n] = block % direct_blks;
407 * Caller should call f2fs_put_dnode(dn).
408 * Also, it should grab and release a rwsem by calling f2fs_lock_op() and
409 * f2fs_unlock_op() only if ro is not set RDONLY_NODE.
410 * In the case of RDONLY_NODE, we don't need to care about mutex.
412 int get_dnode_of_data(struct dnode_of_data *dn, pgoff_t index, int mode)
414 struct f2fs_sb_info *sbi = F2FS_SB(dn->inode->i_sb);
415 struct page *npage[4];
418 unsigned int noffset[4];
423 level = get_node_path(F2FS_I(dn->inode), index, offset, noffset);
425 nids[0] = dn->inode->i_ino;
426 npage[0] = dn->inode_page;
429 npage[0] = get_node_page(sbi, nids[0]);
430 if (IS_ERR(npage[0]))
431 return PTR_ERR(npage[0]);
435 nids[1] = get_nid(parent, offset[0], true);
436 dn->inode_page = npage[0];
437 dn->inode_page_locked = true;
439 /* get indirect or direct nodes */
440 for (i = 1; i <= level; i++) {
443 if (!nids[i] && mode == ALLOC_NODE) {
445 if (!alloc_nid(sbi, &(nids[i]))) {
451 npage[i] = new_node_page(dn, noffset[i], NULL);
452 if (IS_ERR(npage[i])) {
453 alloc_nid_failed(sbi, nids[i]);
454 err = PTR_ERR(npage[i]);
458 set_nid(parent, offset[i - 1], nids[i], i == 1);
459 alloc_nid_done(sbi, nids[i]);
461 } else if (mode == LOOKUP_NODE_RA && i == level && level > 1) {
462 npage[i] = get_node_page_ra(parent, offset[i - 1]);
463 if (IS_ERR(npage[i])) {
464 err = PTR_ERR(npage[i]);
470 dn->inode_page_locked = false;
473 f2fs_put_page(parent, 1);
477 npage[i] = get_node_page(sbi, nids[i]);
478 if (IS_ERR(npage[i])) {
479 err = PTR_ERR(npage[i]);
480 f2fs_put_page(npage[0], 0);
486 nids[i + 1] = get_nid(parent, offset[i], false);
489 dn->nid = nids[level];
490 dn->ofs_in_node = offset[level];
491 dn->node_page = npage[level];
492 dn->data_blkaddr = datablock_addr(dn->node_page, dn->ofs_in_node);
496 f2fs_put_page(parent, 1);
498 f2fs_put_page(npage[0], 0);
500 dn->inode_page = NULL;
501 dn->node_page = NULL;
505 static void truncate_node(struct dnode_of_data *dn)
507 struct f2fs_sb_info *sbi = F2FS_SB(dn->inode->i_sb);
510 get_node_info(sbi, dn->nid, &ni);
511 if (dn->inode->i_blocks == 0) {
512 f2fs_bug_on(ni.blk_addr != NULL_ADDR);
515 f2fs_bug_on(ni.blk_addr == NULL_ADDR);
517 /* Deallocate node address */
518 invalidate_blocks(sbi, ni.blk_addr);
519 dec_valid_node_count(sbi, dn->inode);
520 set_node_addr(sbi, &ni, NULL_ADDR, false);
522 if (dn->nid == dn->inode->i_ino) {
523 remove_orphan_inode(sbi, dn->nid);
524 dec_valid_inode_count(sbi);
529 clear_node_page_dirty(dn->node_page);
530 F2FS_SET_SB_DIRT(sbi);
532 f2fs_put_page(dn->node_page, 1);
534 invalidate_mapping_pages(NODE_MAPPING(sbi),
535 dn->node_page->index, dn->node_page->index);
537 dn->node_page = NULL;
538 trace_f2fs_truncate_node(dn->inode, dn->nid, ni.blk_addr);
541 static int truncate_dnode(struct dnode_of_data *dn)
543 struct f2fs_sb_info *sbi = F2FS_SB(dn->inode->i_sb);
549 /* get direct node */
550 page = get_node_page(sbi, dn->nid);
551 if (IS_ERR(page) && PTR_ERR(page) == -ENOENT)
553 else if (IS_ERR(page))
554 return PTR_ERR(page);
556 /* Make dnode_of_data for parameter */
557 dn->node_page = page;
559 truncate_data_blocks(dn);
564 static int truncate_nodes(struct dnode_of_data *dn, unsigned int nofs,
567 struct f2fs_sb_info *sbi = F2FS_SB(dn->inode->i_sb);
568 struct dnode_of_data rdn = *dn;
570 struct f2fs_node *rn;
572 unsigned int child_nofs;
577 return NIDS_PER_BLOCK + 1;
579 trace_f2fs_truncate_nodes_enter(dn->inode, dn->nid, dn->data_blkaddr);
581 page = get_node_page(sbi, dn->nid);
583 trace_f2fs_truncate_nodes_exit(dn->inode, PTR_ERR(page));
584 return PTR_ERR(page);
587 rn = F2FS_NODE(page);
589 for (i = ofs; i < NIDS_PER_BLOCK; i++, freed++) {
590 child_nid = le32_to_cpu(rn->in.nid[i]);
594 ret = truncate_dnode(&rdn);
597 set_nid(page, i, 0, false);
600 child_nofs = nofs + ofs * (NIDS_PER_BLOCK + 1) + 1;
601 for (i = ofs; i < NIDS_PER_BLOCK; i++) {
602 child_nid = le32_to_cpu(rn->in.nid[i]);
603 if (child_nid == 0) {
604 child_nofs += NIDS_PER_BLOCK + 1;
608 ret = truncate_nodes(&rdn, child_nofs, 0, depth - 1);
609 if (ret == (NIDS_PER_BLOCK + 1)) {
610 set_nid(page, i, 0, false);
612 } else if (ret < 0 && ret != -ENOENT) {
620 /* remove current indirect node */
621 dn->node_page = page;
625 f2fs_put_page(page, 1);
627 trace_f2fs_truncate_nodes_exit(dn->inode, freed);
631 f2fs_put_page(page, 1);
632 trace_f2fs_truncate_nodes_exit(dn->inode, ret);
636 static int truncate_partial_nodes(struct dnode_of_data *dn,
637 struct f2fs_inode *ri, int *offset, int depth)
639 struct f2fs_sb_info *sbi = F2FS_SB(dn->inode->i_sb);
640 struct page *pages[2];
647 nid[0] = le32_to_cpu(ri->i_nid[offset[0] - NODE_DIR1_BLOCK]);
651 /* get indirect nodes in the path */
652 for (i = 0; i < idx + 1; i++) {
653 /* reference count'll be increased */
654 pages[i] = get_node_page(sbi, nid[i]);
655 if (IS_ERR(pages[i])) {
656 err = PTR_ERR(pages[i]);
660 nid[i + 1] = get_nid(pages[i], offset[i + 1], false);
663 /* free direct nodes linked to a partial indirect node */
664 for (i = offset[idx + 1]; i < NIDS_PER_BLOCK; i++) {
665 child_nid = get_nid(pages[idx], i, false);
669 err = truncate_dnode(dn);
672 set_nid(pages[idx], i, 0, false);
675 if (offset[idx + 1] == 0) {
676 dn->node_page = pages[idx];
680 f2fs_put_page(pages[idx], 1);
686 for (i = idx; i >= 0; i--)
687 f2fs_put_page(pages[i], 1);
689 trace_f2fs_truncate_partial_nodes(dn->inode, nid, depth, err);
695 * All the block addresses of data and nodes should be nullified.
697 int truncate_inode_blocks(struct inode *inode, pgoff_t from)
699 struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
700 int err = 0, cont = 1;
701 int level, offset[4], noffset[4];
702 unsigned int nofs = 0;
703 struct f2fs_inode *ri;
704 struct dnode_of_data dn;
707 trace_f2fs_truncate_inode_blocks_enter(inode, from);
709 level = get_node_path(F2FS_I(inode), from, offset, noffset);
711 page = get_node_page(sbi, inode->i_ino);
713 trace_f2fs_truncate_inode_blocks_exit(inode, PTR_ERR(page));
714 return PTR_ERR(page);
717 set_new_dnode(&dn, inode, page, NULL, 0);
720 ri = F2FS_INODE(page);
728 if (!offset[level - 1])
730 err = truncate_partial_nodes(&dn, ri, offset, level);
731 if (err < 0 && err != -ENOENT)
733 nofs += 1 + NIDS_PER_BLOCK;
736 nofs = 5 + 2 * NIDS_PER_BLOCK;
737 if (!offset[level - 1])
739 err = truncate_partial_nodes(&dn, ri, offset, level);
740 if (err < 0 && err != -ENOENT)
749 dn.nid = le32_to_cpu(ri->i_nid[offset[0] - NODE_DIR1_BLOCK]);
751 case NODE_DIR1_BLOCK:
752 case NODE_DIR2_BLOCK:
753 err = truncate_dnode(&dn);
756 case NODE_IND1_BLOCK:
757 case NODE_IND2_BLOCK:
758 err = truncate_nodes(&dn, nofs, offset[1], 2);
761 case NODE_DIND_BLOCK:
762 err = truncate_nodes(&dn, nofs, offset[1], 3);
769 if (err < 0 && err != -ENOENT)
771 if (offset[1] == 0 &&
772 ri->i_nid[offset[0] - NODE_DIR1_BLOCK]) {
774 if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
775 f2fs_put_page(page, 1);
778 f2fs_wait_on_page_writeback(page, NODE);
779 ri->i_nid[offset[0] - NODE_DIR1_BLOCK] = 0;
780 set_page_dirty(page);
788 f2fs_put_page(page, 0);
789 trace_f2fs_truncate_inode_blocks_exit(inode, err);
790 return err > 0 ? 0 : err;
793 int truncate_xattr_node(struct inode *inode, struct page *page)
795 struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
796 nid_t nid = F2FS_I(inode)->i_xattr_nid;
797 struct dnode_of_data dn;
803 npage = get_node_page(sbi, nid);
805 return PTR_ERR(npage);
807 F2FS_I(inode)->i_xattr_nid = 0;
809 /* need to do checkpoint during fsync */
810 F2FS_I(inode)->xattr_ver = cur_cp_version(F2FS_CKPT(sbi));
812 set_new_dnode(&dn, inode, page, npage, nid);
815 dn.inode_page_locked = true;
821 * Caller should grab and release a rwsem by calling f2fs_lock_op() and
824 void remove_inode_page(struct inode *inode)
826 struct dnode_of_data dn;
828 set_new_dnode(&dn, inode, NULL, NULL, inode->i_ino);
829 if (get_dnode_of_data(&dn, 0, LOOKUP_NODE))
832 if (truncate_xattr_node(inode, dn.inode_page)) {
837 /* remove potential inline_data blocks */
838 if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
839 S_ISLNK(inode->i_mode))
840 truncate_data_blocks_range(&dn, 1);
842 /* 0 is possible, after f2fs_new_inode() has failed */
843 f2fs_bug_on(inode->i_blocks != 0 && inode->i_blocks != 1);
845 /* will put inode & node pages */
849 struct page *new_inode_page(struct inode *inode)
851 struct dnode_of_data dn;
853 /* allocate inode page for new inode */
854 set_new_dnode(&dn, inode, NULL, NULL, inode->i_ino);
856 /* caller should f2fs_put_page(page, 1); */
857 return new_node_page(&dn, 0, NULL);
860 struct page *new_node_page(struct dnode_of_data *dn,
861 unsigned int ofs, struct page *ipage)
863 struct f2fs_sb_info *sbi = F2FS_SB(dn->inode->i_sb);
864 struct node_info old_ni, new_ni;
868 if (unlikely(is_inode_flag_set(F2FS_I(dn->inode), FI_NO_ALLOC)))
869 return ERR_PTR(-EPERM);
871 page = grab_cache_page(NODE_MAPPING(sbi), dn->nid);
873 return ERR_PTR(-ENOMEM);
875 if (unlikely(!inc_valid_node_count(sbi, dn->inode))) {
880 get_node_info(sbi, dn->nid, &old_ni);
882 /* Reinitialize old_ni with new node page */
883 f2fs_bug_on(old_ni.blk_addr != NULL_ADDR);
885 new_ni.ino = dn->inode->i_ino;
886 set_node_addr(sbi, &new_ni, NEW_ADDR, false);
888 f2fs_wait_on_page_writeback(page, NODE);
889 fill_node_footer(page, dn->nid, dn->inode->i_ino, ofs, true);
890 set_cold_node(dn->inode, page);
891 SetPageUptodate(page);
892 set_page_dirty(page);
894 if (f2fs_has_xattr_block(ofs))
895 F2FS_I(dn->inode)->i_xattr_nid = dn->nid;
897 dn->node_page = page;
899 update_inode(dn->inode, ipage);
903 inc_valid_inode_count(sbi);
908 clear_node_page_dirty(page);
909 f2fs_put_page(page, 1);
914 * Caller should do after getting the following values.
915 * 0: f2fs_put_page(page, 0)
916 * LOCKED_PAGE: f2fs_put_page(page, 1)
919 static int read_node_page(struct page *page, int rw)
921 struct f2fs_sb_info *sbi = F2FS_SB(page->mapping->host->i_sb);
924 get_node_info(sbi, page->index, &ni);
926 if (unlikely(ni.blk_addr == NULL_ADDR)) {
927 f2fs_put_page(page, 1);
931 if (PageUptodate(page))
934 return f2fs_submit_page_bio(sbi, page, ni.blk_addr, rw);
938 * Readahead a node page
940 void ra_node_page(struct f2fs_sb_info *sbi, nid_t nid)
945 apage = find_get_page(NODE_MAPPING(sbi), nid);
946 if (apage && PageUptodate(apage)) {
947 f2fs_put_page(apage, 0);
950 f2fs_put_page(apage, 0);
952 apage = grab_cache_page(NODE_MAPPING(sbi), nid);
956 err = read_node_page(apage, READA);
958 f2fs_put_page(apage, 0);
959 else if (err == LOCKED_PAGE)
960 f2fs_put_page(apage, 1);
963 struct page *get_node_page(struct f2fs_sb_info *sbi, pgoff_t nid)
968 page = grab_cache_page(NODE_MAPPING(sbi), nid);
970 return ERR_PTR(-ENOMEM);
972 err = read_node_page(page, READ_SYNC);
975 else if (err == LOCKED_PAGE)
979 if (unlikely(!PageUptodate(page) || nid != nid_of_node(page))) {
980 f2fs_put_page(page, 1);
981 return ERR_PTR(-EIO);
983 if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
984 f2fs_put_page(page, 1);
992 * Return a locked page for the desired node page.
993 * And, readahead MAX_RA_NODE number of node pages.
995 struct page *get_node_page_ra(struct page *parent, int start)
997 struct f2fs_sb_info *sbi = F2FS_SB(parent->mapping->host->i_sb);
998 struct blk_plug plug;
1003 /* First, try getting the desired direct node. */
1004 nid = get_nid(parent, start, false);
1006 return ERR_PTR(-ENOENT);
1008 page = grab_cache_page(NODE_MAPPING(sbi), nid);
1010 return ERR_PTR(-ENOMEM);
1012 err = read_node_page(page, READ_SYNC);
1014 return ERR_PTR(err);
1015 else if (err == LOCKED_PAGE)
1018 blk_start_plug(&plug);
1020 /* Then, try readahead for siblings of the desired node */
1021 end = start + MAX_RA_NODE;
1022 end = min(end, NIDS_PER_BLOCK);
1023 for (i = start + 1; i < end; i++) {
1024 nid = get_nid(parent, i, false);
1027 ra_node_page(sbi, nid);
1030 blk_finish_plug(&plug);
1033 if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1034 f2fs_put_page(page, 1);
1038 if (unlikely(!PageUptodate(page))) {
1039 f2fs_put_page(page, 1);
1040 return ERR_PTR(-EIO);
1045 void sync_inode_page(struct dnode_of_data *dn)
1047 if (IS_INODE(dn->node_page) || dn->inode_page == dn->node_page) {
1048 update_inode(dn->inode, dn->node_page);
1049 } else if (dn->inode_page) {
1050 if (!dn->inode_page_locked)
1051 lock_page(dn->inode_page);
1052 update_inode(dn->inode, dn->inode_page);
1053 if (!dn->inode_page_locked)
1054 unlock_page(dn->inode_page);
1056 update_inode_page(dn->inode);
1060 int sync_node_pages(struct f2fs_sb_info *sbi, nid_t ino,
1061 struct writeback_control *wbc)
1064 struct pagevec pvec;
1065 int step = ino ? 2 : 0;
1066 int nwritten = 0, wrote = 0;
1068 pagevec_init(&pvec, 0);
1074 while (index <= end) {
1076 nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1077 PAGECACHE_TAG_DIRTY,
1078 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
1082 for (i = 0; i < nr_pages; i++) {
1083 struct page *page = pvec.pages[i];
1086 * flushing sequence with step:
1091 if (step == 0 && IS_DNODE(page))
1093 if (step == 1 && (!IS_DNODE(page) ||
1094 is_cold_node(page)))
1096 if (step == 2 && (!IS_DNODE(page) ||
1097 !is_cold_node(page)))
1102 * we should not skip writing node pages.
1104 if (ino && ino_of_node(page) == ino)
1106 else if (!trylock_page(page))
1109 if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1114 if (ino && ino_of_node(page) != ino)
1115 goto continue_unlock;
1117 if (!PageDirty(page)) {
1118 /* someone wrote it for us */
1119 goto continue_unlock;
1122 if (!clear_page_dirty_for_io(page))
1123 goto continue_unlock;
1125 /* called by fsync() */
1126 if (ino && IS_DNODE(page)) {
1127 int mark = !is_checkpointed_node(sbi, ino);
1128 set_fsync_mark(page, 1);
1130 set_dentry_mark(page, mark);
1133 set_fsync_mark(page, 0);
1134 set_dentry_mark(page, 0);
1137 if (NODE_MAPPING(sbi)->a_ops->writepage(page, wbc))
1142 if (--wbc->nr_to_write == 0)
1145 pagevec_release(&pvec);
1148 if (wbc->nr_to_write == 0) {
1160 f2fs_submit_merged_bio(sbi, NODE, WRITE);
1164 int wait_on_node_pages_writeback(struct f2fs_sb_info *sbi, nid_t ino)
1166 pgoff_t index = 0, end = LONG_MAX;
1167 struct pagevec pvec;
1168 int ret2 = 0, ret = 0;
1170 pagevec_init(&pvec, 0);
1172 while (index <= end) {
1174 nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1175 PAGECACHE_TAG_WRITEBACK,
1176 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
1180 for (i = 0; i < nr_pages; i++) {
1181 struct page *page = pvec.pages[i];
1183 /* until radix tree lookup accepts end_index */
1184 if (unlikely(page->index > end))
1187 if (ino && ino_of_node(page) == ino) {
1188 f2fs_wait_on_page_writeback(page, NODE);
1189 if (TestClearPageError(page))
1193 pagevec_release(&pvec);
1197 if (unlikely(test_and_clear_bit(AS_ENOSPC, &NODE_MAPPING(sbi)->flags)))
1199 if (unlikely(test_and_clear_bit(AS_EIO, &NODE_MAPPING(sbi)->flags)))
1206 static int f2fs_write_node_page(struct page *page,
1207 struct writeback_control *wbc)
1209 struct f2fs_sb_info *sbi = F2FS_SB(page->mapping->host->i_sb);
1212 struct node_info ni;
1213 struct f2fs_io_info fio = {
1215 .rw = (wbc->sync_mode == WB_SYNC_ALL) ? WRITE_SYNC : WRITE,
1218 trace_f2fs_writepage(page, NODE);
1220 if (unlikely(sbi->por_doing))
1222 if (unlikely(f2fs_cp_error(sbi)))
1225 f2fs_wait_on_page_writeback(page, NODE);
1227 /* get old block addr of this node page */
1228 nid = nid_of_node(page);
1229 f2fs_bug_on(page->index != nid);
1231 get_node_info(sbi, nid, &ni);
1233 /* This page is already truncated */
1234 if (unlikely(ni.blk_addr == NULL_ADDR)) {
1235 dec_page_count(sbi, F2FS_DIRTY_NODES);
1240 if (wbc->for_reclaim)
1243 down_read(&sbi->node_write);
1244 set_page_writeback(page);
1245 write_node_page(sbi, page, &fio, nid, ni.blk_addr, &new_addr);
1246 set_node_addr(sbi, &ni, new_addr, is_fsync_dnode(page));
1247 dec_page_count(sbi, F2FS_DIRTY_NODES);
1248 up_read(&sbi->node_write);
1253 redirty_page_for_writepage(wbc, page);
1254 return AOP_WRITEPAGE_ACTIVATE;
1257 static int f2fs_write_node_pages(struct address_space *mapping,
1258 struct writeback_control *wbc)
1260 struct f2fs_sb_info *sbi = F2FS_SB(mapping->host->i_sb);
1263 trace_f2fs_writepages(mapping->host, wbc, NODE);
1265 /* balancing f2fs's metadata in background */
1266 f2fs_balance_fs_bg(sbi);
1268 /* collect a number of dirty node pages and write together */
1269 if (get_pages(sbi, F2FS_DIRTY_NODES) < nr_pages_to_skip(sbi, NODE))
1272 diff = nr_pages_to_write(sbi, NODE, wbc);
1273 wbc->sync_mode = WB_SYNC_NONE;
1274 sync_node_pages(sbi, 0, wbc);
1275 wbc->nr_to_write = max((long)0, wbc->nr_to_write - diff);
1279 wbc->pages_skipped += get_pages(sbi, F2FS_DIRTY_NODES);
1283 static int f2fs_set_node_page_dirty(struct page *page)
1285 struct address_space *mapping = page->mapping;
1286 struct f2fs_sb_info *sbi = F2FS_SB(mapping->host->i_sb);
1288 trace_f2fs_set_page_dirty(page, NODE);
1290 SetPageUptodate(page);
1291 if (!PageDirty(page)) {
1292 __set_page_dirty_nobuffers(page);
1293 inc_page_count(sbi, F2FS_DIRTY_NODES);
1294 SetPagePrivate(page);
1300 static void f2fs_invalidate_node_page(struct page *page, unsigned int offset,
1301 unsigned int length)
1303 struct inode *inode = page->mapping->host;
1304 struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
1305 if (PageDirty(page))
1306 dec_page_count(sbi, F2FS_DIRTY_NODES);
1307 ClearPagePrivate(page);
1310 static int f2fs_release_node_page(struct page *page, gfp_t wait)
1312 ClearPagePrivate(page);
1317 * Structure of the f2fs node operations
1319 const struct address_space_operations f2fs_node_aops = {
1320 .writepage = f2fs_write_node_page,
1321 .writepages = f2fs_write_node_pages,
1322 .set_page_dirty = f2fs_set_node_page_dirty,
1323 .invalidatepage = f2fs_invalidate_node_page,
1324 .releasepage = f2fs_release_node_page,
1327 static struct free_nid *__lookup_free_nid_list(struct f2fs_nm_info *nm_i,
1330 return radix_tree_lookup(&nm_i->free_nid_root, n);
1333 static void __del_from_free_nid_list(struct f2fs_nm_info *nm_i,
1337 radix_tree_delete(&nm_i->free_nid_root, i->nid);
1340 static int add_free_nid(struct f2fs_sb_info *sbi, nid_t nid, bool build)
1342 struct f2fs_nm_info *nm_i = NM_I(sbi);
1344 struct nat_entry *ne;
1345 bool allocated = false;
1347 if (!available_free_memory(sbi, FREE_NIDS))
1350 /* 0 nid should not be used */
1351 if (unlikely(nid == 0))
1355 /* do not add allocated nids */
1356 read_lock(&nm_i->nat_tree_lock);
1357 ne = __lookup_nat_cache(nm_i, nid);
1359 (!ne->checkpointed || nat_get_blkaddr(ne) != NULL_ADDR))
1361 read_unlock(&nm_i->nat_tree_lock);
1366 i = f2fs_kmem_cache_alloc(free_nid_slab, GFP_NOFS);
1370 spin_lock(&nm_i->free_nid_list_lock);
1371 if (radix_tree_insert(&nm_i->free_nid_root, i->nid, i)) {
1372 spin_unlock(&nm_i->free_nid_list_lock);
1373 kmem_cache_free(free_nid_slab, i);
1376 list_add_tail(&i->list, &nm_i->free_nid_list);
1378 spin_unlock(&nm_i->free_nid_list_lock);
1382 static void remove_free_nid(struct f2fs_nm_info *nm_i, nid_t nid)
1385 bool need_free = false;
1387 spin_lock(&nm_i->free_nid_list_lock);
1388 i = __lookup_free_nid_list(nm_i, nid);
1389 if (i && i->state == NID_NEW) {
1390 __del_from_free_nid_list(nm_i, i);
1394 spin_unlock(&nm_i->free_nid_list_lock);
1397 kmem_cache_free(free_nid_slab, i);
1400 static void scan_nat_page(struct f2fs_sb_info *sbi,
1401 struct page *nat_page, nid_t start_nid)
1403 struct f2fs_nm_info *nm_i = NM_I(sbi);
1404 struct f2fs_nat_block *nat_blk = page_address(nat_page);
1408 i = start_nid % NAT_ENTRY_PER_BLOCK;
1410 for (; i < NAT_ENTRY_PER_BLOCK; i++, start_nid++) {
1412 if (unlikely(start_nid >= nm_i->max_nid))
1415 blk_addr = le32_to_cpu(nat_blk->entries[i].block_addr);
1416 f2fs_bug_on(blk_addr == NEW_ADDR);
1417 if (blk_addr == NULL_ADDR) {
1418 if (add_free_nid(sbi, start_nid, true) < 0)
1424 static void build_free_nids(struct f2fs_sb_info *sbi)
1426 struct f2fs_nm_info *nm_i = NM_I(sbi);
1427 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
1428 struct f2fs_summary_block *sum = curseg->sum_blk;
1430 nid_t nid = nm_i->next_scan_nid;
1432 /* Enough entries */
1433 if (nm_i->fcnt > NAT_ENTRY_PER_BLOCK)
1436 /* readahead nat pages to be scanned */
1437 ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nid), FREE_NID_PAGES, META_NAT);
1440 struct page *page = get_current_nat_page(sbi, nid);
1442 scan_nat_page(sbi, page, nid);
1443 f2fs_put_page(page, 1);
1445 nid += (NAT_ENTRY_PER_BLOCK - (nid % NAT_ENTRY_PER_BLOCK));
1446 if (unlikely(nid >= nm_i->max_nid))
1449 if (i++ == FREE_NID_PAGES)
1453 /* go to the next free nat pages to find free nids abundantly */
1454 nm_i->next_scan_nid = nid;
1456 /* find free nids from current sum_pages */
1457 mutex_lock(&curseg->curseg_mutex);
1458 for (i = 0; i < nats_in_cursum(sum); i++) {
1459 block_t addr = le32_to_cpu(nat_in_journal(sum, i).block_addr);
1460 nid = le32_to_cpu(nid_in_journal(sum, i));
1461 if (addr == NULL_ADDR)
1462 add_free_nid(sbi, nid, true);
1464 remove_free_nid(nm_i, nid);
1466 mutex_unlock(&curseg->curseg_mutex);
1470 * If this function returns success, caller can obtain a new nid
1471 * from second parameter of this function.
1472 * The returned nid could be used ino as well as nid when inode is created.
1474 bool alloc_nid(struct f2fs_sb_info *sbi, nid_t *nid)
1476 struct f2fs_nm_info *nm_i = NM_I(sbi);
1477 struct free_nid *i = NULL;
1479 if (unlikely(sbi->total_valid_node_count + 1 > nm_i->available_nids))
1482 spin_lock(&nm_i->free_nid_list_lock);
1484 /* We should not use stale free nids created by build_free_nids */
1485 if (nm_i->fcnt && !on_build_free_nids(nm_i)) {
1486 f2fs_bug_on(list_empty(&nm_i->free_nid_list));
1487 list_for_each_entry(i, &nm_i->free_nid_list, list)
1488 if (i->state == NID_NEW)
1491 f2fs_bug_on(i->state != NID_NEW);
1493 i->state = NID_ALLOC;
1495 spin_unlock(&nm_i->free_nid_list_lock);
1498 spin_unlock(&nm_i->free_nid_list_lock);
1500 /* Let's scan nat pages and its caches to get free nids */
1501 mutex_lock(&nm_i->build_lock);
1502 build_free_nids(sbi);
1503 mutex_unlock(&nm_i->build_lock);
1508 * alloc_nid() should be called prior to this function.
1510 void alloc_nid_done(struct f2fs_sb_info *sbi, nid_t nid)
1512 struct f2fs_nm_info *nm_i = NM_I(sbi);
1515 spin_lock(&nm_i->free_nid_list_lock);
1516 i = __lookup_free_nid_list(nm_i, nid);
1517 f2fs_bug_on(!i || i->state != NID_ALLOC);
1518 __del_from_free_nid_list(nm_i, i);
1519 spin_unlock(&nm_i->free_nid_list_lock);
1521 kmem_cache_free(free_nid_slab, i);
1525 * alloc_nid() should be called prior to this function.
1527 void alloc_nid_failed(struct f2fs_sb_info *sbi, nid_t nid)
1529 struct f2fs_nm_info *nm_i = NM_I(sbi);
1531 bool need_free = false;
1536 spin_lock(&nm_i->free_nid_list_lock);
1537 i = __lookup_free_nid_list(nm_i, nid);
1538 f2fs_bug_on(!i || i->state != NID_ALLOC);
1539 if (!available_free_memory(sbi, FREE_NIDS)) {
1540 __del_from_free_nid_list(nm_i, i);
1546 spin_unlock(&nm_i->free_nid_list_lock);
1549 kmem_cache_free(free_nid_slab, i);
1552 void recover_inline_xattr(struct inode *inode, struct page *page)
1554 struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
1555 void *src_addr, *dst_addr;
1558 struct f2fs_inode *ri;
1560 ipage = get_node_page(sbi, inode->i_ino);
1561 f2fs_bug_on(IS_ERR(ipage));
1563 ri = F2FS_INODE(page);
1564 if (!(ri->i_inline & F2FS_INLINE_XATTR)) {
1565 clear_inode_flag(F2FS_I(inode), FI_INLINE_XATTR);
1569 dst_addr = inline_xattr_addr(ipage);
1570 src_addr = inline_xattr_addr(page);
1571 inline_size = inline_xattr_size(inode);
1573 f2fs_wait_on_page_writeback(ipage, NODE);
1574 memcpy(dst_addr, src_addr, inline_size);
1576 update_inode(inode, ipage);
1577 f2fs_put_page(ipage, 1);
1580 void recover_xattr_data(struct inode *inode, struct page *page, block_t blkaddr)
1582 struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
1583 nid_t prev_xnid = F2FS_I(inode)->i_xattr_nid;
1584 nid_t new_xnid = nid_of_node(page);
1585 struct node_info ni;
1587 /* 1: invalidate the previous xattr nid */
1591 /* Deallocate node address */
1592 get_node_info(sbi, prev_xnid, &ni);
1593 f2fs_bug_on(ni.blk_addr == NULL_ADDR);
1594 invalidate_blocks(sbi, ni.blk_addr);
1595 dec_valid_node_count(sbi, inode);
1596 set_node_addr(sbi, &ni, NULL_ADDR, false);
1599 /* 2: allocate new xattr nid */
1600 if (unlikely(!inc_valid_node_count(sbi, inode)))
1603 remove_free_nid(NM_I(sbi), new_xnid);
1604 get_node_info(sbi, new_xnid, &ni);
1605 ni.ino = inode->i_ino;
1606 set_node_addr(sbi, &ni, NEW_ADDR, false);
1607 F2FS_I(inode)->i_xattr_nid = new_xnid;
1609 /* 3: update xattr blkaddr */
1610 refresh_sit_entry(sbi, NEW_ADDR, blkaddr);
1611 set_node_addr(sbi, &ni, blkaddr, false);
1613 update_inode_page(inode);
1616 int recover_inode_page(struct f2fs_sb_info *sbi, struct page *page)
1618 struct f2fs_inode *src, *dst;
1619 nid_t ino = ino_of_node(page);
1620 struct node_info old_ni, new_ni;
1623 get_node_info(sbi, ino, &old_ni);
1625 if (unlikely(old_ni.blk_addr != NULL_ADDR))
1628 ipage = grab_cache_page(NODE_MAPPING(sbi), ino);
1632 /* Should not use this inode from free nid list */
1633 remove_free_nid(NM_I(sbi), ino);
1635 SetPageUptodate(ipage);
1636 fill_node_footer(ipage, ino, ino, 0, true);
1638 src = F2FS_INODE(page);
1639 dst = F2FS_INODE(ipage);
1641 memcpy(dst, src, (unsigned long)&src->i_ext - (unsigned long)src);
1643 dst->i_blocks = cpu_to_le64(1);
1644 dst->i_links = cpu_to_le32(1);
1645 dst->i_xattr_nid = 0;
1646 dst->i_inline = src->i_inline & F2FS_INLINE_XATTR;
1651 if (unlikely(!inc_valid_node_count(sbi, NULL)))
1653 set_node_addr(sbi, &new_ni, NEW_ADDR, false);
1654 inc_valid_inode_count(sbi);
1655 set_page_dirty(ipage);
1656 f2fs_put_page(ipage, 1);
1661 * ra_sum_pages() merge contiguous pages into one bio and submit.
1662 * these pre-read pages are allocated in bd_inode's mapping tree.
1664 static int ra_sum_pages(struct f2fs_sb_info *sbi, struct page **pages,
1665 int start, int nrpages)
1667 struct inode *inode = sbi->sb->s_bdev->bd_inode;
1668 struct address_space *mapping = inode->i_mapping;
1669 int i, page_idx = start;
1670 struct f2fs_io_info fio = {
1672 .rw = READ_SYNC | REQ_META | REQ_PRIO
1675 for (i = 0; page_idx < start + nrpages; page_idx++, i++) {
1676 /* alloc page in bd_inode for reading node summary info */
1677 pages[i] = grab_cache_page(mapping, page_idx);
1680 f2fs_submit_page_mbio(sbi, pages[i], page_idx, &fio);
1683 f2fs_submit_merged_bio(sbi, META, READ);
1687 int restore_node_summary(struct f2fs_sb_info *sbi,
1688 unsigned int segno, struct f2fs_summary_block *sum)
1690 struct f2fs_node *rn;
1691 struct f2fs_summary *sum_entry;
1692 struct inode *inode = sbi->sb->s_bdev->bd_inode;
1694 int bio_blocks = MAX_BIO_BLOCKS(max_hw_blocks(sbi));
1695 struct page *pages[bio_blocks];
1696 int i, idx, last_offset, nrpages, err = 0;
1698 /* scan the node segment */
1699 last_offset = sbi->blocks_per_seg;
1700 addr = START_BLOCK(sbi, segno);
1701 sum_entry = &sum->entries[0];
1703 for (i = 0; !err && i < last_offset; i += nrpages, addr += nrpages) {
1704 nrpages = min(last_offset - i, bio_blocks);
1706 /* readahead node pages */
1707 nrpages = ra_sum_pages(sbi, pages, addr, nrpages);
1711 for (idx = 0; idx < nrpages; idx++) {
1715 lock_page(pages[idx]);
1716 if (unlikely(!PageUptodate(pages[idx]))) {
1719 rn = F2FS_NODE(pages[idx]);
1720 sum_entry->nid = rn->footer.nid;
1721 sum_entry->version = 0;
1722 sum_entry->ofs_in_node = 0;
1725 unlock_page(pages[idx]);
1727 page_cache_release(pages[idx]);
1730 invalidate_mapping_pages(inode->i_mapping, addr,
1736 static struct nat_entry_set *grab_nat_entry_set(void)
1738 struct nat_entry_set *nes =
1739 f2fs_kmem_cache_alloc(nat_entry_set_slab, GFP_ATOMIC);
1742 INIT_LIST_HEAD(&nes->set_list);
1743 INIT_LIST_HEAD(&nes->entry_list);
1747 static void release_nat_entry_set(struct nat_entry_set *nes,
1748 struct f2fs_nm_info *nm_i)
1750 f2fs_bug_on(!list_empty(&nes->entry_list));
1752 nm_i->dirty_nat_cnt -= nes->entry_cnt;
1753 list_del(&nes->set_list);
1754 kmem_cache_free(nat_entry_set_slab, nes);
1757 static void adjust_nat_entry_set(struct nat_entry_set *nes,
1758 struct list_head *head)
1760 struct nat_entry_set *next = nes;
1762 if (list_is_last(&nes->set_list, head))
1765 list_for_each_entry_continue(next, head, set_list)
1766 if (nes->entry_cnt <= next->entry_cnt)
1769 list_move_tail(&nes->set_list, &next->set_list);
1772 static void add_nat_entry(struct nat_entry *ne, struct list_head *head)
1774 struct nat_entry_set *nes;
1775 nid_t start_nid = START_NID(ne->ni.nid);
1777 list_for_each_entry(nes, head, set_list) {
1778 if (nes->start_nid == start_nid) {
1779 list_move_tail(&ne->list, &nes->entry_list);
1781 adjust_nat_entry_set(nes, head);
1786 nes = grab_nat_entry_set();
1788 nes->start_nid = start_nid;
1789 list_move_tail(&ne->list, &nes->entry_list);
1791 list_add(&nes->set_list, head);
1794 static void merge_nats_in_set(struct f2fs_sb_info *sbi)
1796 struct f2fs_nm_info *nm_i = NM_I(sbi);
1797 struct list_head *dirty_list = &nm_i->dirty_nat_entries;
1798 struct list_head *set_list = &nm_i->nat_entry_set;
1799 struct nat_entry *ne, *tmp;
1801 write_lock(&nm_i->nat_tree_lock);
1802 list_for_each_entry_safe(ne, tmp, dirty_list, list) {
1803 if (nat_get_blkaddr(ne) == NEW_ADDR)
1805 add_nat_entry(ne, set_list);
1806 nm_i->dirty_nat_cnt++;
1808 write_unlock(&nm_i->nat_tree_lock);
1811 static bool __has_cursum_space(struct f2fs_summary_block *sum, int size)
1813 if (nats_in_cursum(sum) + size <= NAT_JOURNAL_ENTRIES)
1819 static void remove_nats_in_journal(struct f2fs_sb_info *sbi)
1821 struct f2fs_nm_info *nm_i = NM_I(sbi);
1822 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
1823 struct f2fs_summary_block *sum = curseg->sum_blk;
1826 mutex_lock(&curseg->curseg_mutex);
1827 for (i = 0; i < nats_in_cursum(sum); i++) {
1828 struct nat_entry *ne;
1829 struct f2fs_nat_entry raw_ne;
1830 nid_t nid = le32_to_cpu(nid_in_journal(sum, i));
1832 raw_ne = nat_in_journal(sum, i);
1834 write_lock(&nm_i->nat_tree_lock);
1835 ne = __lookup_nat_cache(nm_i, nid);
1839 ne = grab_nat_entry(nm_i, nid);
1841 write_unlock(&nm_i->nat_tree_lock);
1844 node_info_from_raw_nat(&ne->ni, &raw_ne);
1846 __set_nat_cache_dirty(nm_i, ne);
1847 write_unlock(&nm_i->nat_tree_lock);
1849 update_nats_in_cursum(sum, -i);
1850 mutex_unlock(&curseg->curseg_mutex);
1854 * This function is called during the checkpointing process.
1856 void flush_nat_entries(struct f2fs_sb_info *sbi)
1858 struct f2fs_nm_info *nm_i = NM_I(sbi);
1859 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
1860 struct f2fs_summary_block *sum = curseg->sum_blk;
1861 struct nat_entry_set *nes, *tmp;
1862 struct list_head *head = &nm_i->nat_entry_set;
1863 bool to_journal = true;
1865 /* merge nat entries of dirty list to nat entry set temporarily */
1866 merge_nats_in_set(sbi);
1869 * if there are no enough space in journal to store dirty nat
1870 * entries, remove all entries from journal and merge them
1871 * into nat entry set.
1873 if (!__has_cursum_space(sum, nm_i->dirty_nat_cnt)) {
1874 remove_nats_in_journal(sbi);
1877 * merge nat entries of dirty list to nat entry set temporarily
1879 merge_nats_in_set(sbi);
1882 if (!nm_i->dirty_nat_cnt)
1886 * there are two steps to flush nat entries:
1887 * #1, flush nat entries to journal in current hot data summary block.
1888 * #2, flush nat entries to nat page.
1890 list_for_each_entry_safe(nes, tmp, head, set_list) {
1891 struct f2fs_nat_block *nat_blk;
1892 struct nat_entry *ne, *cur;
1894 nid_t start_nid = nes->start_nid;
1896 if (to_journal && !__has_cursum_space(sum, nes->entry_cnt))
1900 mutex_lock(&curseg->curseg_mutex);
1902 page = get_next_nat_page(sbi, start_nid);
1903 nat_blk = page_address(page);
1904 f2fs_bug_on(!nat_blk);
1907 /* flush dirty nats in nat entry set */
1908 list_for_each_entry_safe(ne, cur, &nes->entry_list, list) {
1909 struct f2fs_nat_entry *raw_ne;
1910 nid_t nid = nat_get_nid(ne);
1914 offset = lookup_journal_in_cursum(sum,
1915 NAT_JOURNAL, nid, 1);
1916 f2fs_bug_on(offset < 0);
1917 raw_ne = &nat_in_journal(sum, offset);
1918 nid_in_journal(sum, offset) = cpu_to_le32(nid);
1920 raw_ne = &nat_blk->entries[nid - start_nid];
1922 raw_nat_from_node_info(raw_ne, &ne->ni);
1924 if (nat_get_blkaddr(ne) == NULL_ADDR &&
1925 add_free_nid(sbi, nid, false) <= 0) {
1926 write_lock(&nm_i->nat_tree_lock);
1927 __del_from_nat_cache(nm_i, ne);
1928 write_unlock(&nm_i->nat_tree_lock);
1930 write_lock(&nm_i->nat_tree_lock);
1931 __clear_nat_cache_dirty(nm_i, ne);
1932 write_unlock(&nm_i->nat_tree_lock);
1937 mutex_unlock(&curseg->curseg_mutex);
1939 f2fs_put_page(page, 1);
1941 release_nat_entry_set(nes, nm_i);
1944 f2fs_bug_on(!list_empty(head));
1945 f2fs_bug_on(nm_i->dirty_nat_cnt);
1948 static int init_node_manager(struct f2fs_sb_info *sbi)
1950 struct f2fs_super_block *sb_raw = F2FS_RAW_SUPER(sbi);
1951 struct f2fs_nm_info *nm_i = NM_I(sbi);
1952 unsigned char *version_bitmap;
1953 unsigned int nat_segs, nat_blocks;
1955 nm_i->nat_blkaddr = le32_to_cpu(sb_raw->nat_blkaddr);
1957 /* segment_count_nat includes pair segment so divide to 2. */
1958 nat_segs = le32_to_cpu(sb_raw->segment_count_nat) >> 1;
1959 nat_blocks = nat_segs << le32_to_cpu(sb_raw->log_blocks_per_seg);
1961 nm_i->max_nid = NAT_ENTRY_PER_BLOCK * nat_blocks;
1963 /* not used nids: 0, node, meta, (and root counted as valid node) */
1964 nm_i->available_nids = nm_i->max_nid - F2FS_RESERVED_NODE_NUM;
1967 nm_i->ram_thresh = DEF_RAM_THRESHOLD;
1969 INIT_RADIX_TREE(&nm_i->free_nid_root, GFP_ATOMIC);
1970 INIT_LIST_HEAD(&nm_i->free_nid_list);
1971 INIT_RADIX_TREE(&nm_i->nat_root, GFP_ATOMIC);
1972 INIT_LIST_HEAD(&nm_i->nat_entries);
1973 INIT_LIST_HEAD(&nm_i->dirty_nat_entries);
1974 INIT_LIST_HEAD(&nm_i->nat_entry_set);
1976 mutex_init(&nm_i->build_lock);
1977 spin_lock_init(&nm_i->free_nid_list_lock);
1978 rwlock_init(&nm_i->nat_tree_lock);
1980 nm_i->next_scan_nid = le32_to_cpu(sbi->ckpt->next_free_nid);
1981 nm_i->bitmap_size = __bitmap_size(sbi, NAT_BITMAP);
1982 version_bitmap = __bitmap_ptr(sbi, NAT_BITMAP);
1983 if (!version_bitmap)
1986 nm_i->nat_bitmap = kmemdup(version_bitmap, nm_i->bitmap_size,
1988 if (!nm_i->nat_bitmap)
1993 int build_node_manager(struct f2fs_sb_info *sbi)
1997 sbi->nm_info = kzalloc(sizeof(struct f2fs_nm_info), GFP_KERNEL);
2001 err = init_node_manager(sbi);
2005 build_free_nids(sbi);
2009 void destroy_node_manager(struct f2fs_sb_info *sbi)
2011 struct f2fs_nm_info *nm_i = NM_I(sbi);
2012 struct free_nid *i, *next_i;
2013 struct nat_entry *natvec[NATVEC_SIZE];
2020 /* destroy free nid list */
2021 spin_lock(&nm_i->free_nid_list_lock);
2022 list_for_each_entry_safe(i, next_i, &nm_i->free_nid_list, list) {
2023 f2fs_bug_on(i->state == NID_ALLOC);
2024 __del_from_free_nid_list(nm_i, i);
2026 spin_unlock(&nm_i->free_nid_list_lock);
2027 kmem_cache_free(free_nid_slab, i);
2028 spin_lock(&nm_i->free_nid_list_lock);
2030 f2fs_bug_on(nm_i->fcnt);
2031 spin_unlock(&nm_i->free_nid_list_lock);
2033 /* destroy nat cache */
2034 write_lock(&nm_i->nat_tree_lock);
2035 while ((found = __gang_lookup_nat_cache(nm_i,
2036 nid, NATVEC_SIZE, natvec))) {
2038 nid = nat_get_nid(natvec[found - 1]) + 1;
2039 for (idx = 0; idx < found; idx++)
2040 __del_from_nat_cache(nm_i, natvec[idx]);
2042 f2fs_bug_on(nm_i->nat_cnt);
2043 write_unlock(&nm_i->nat_tree_lock);
2045 kfree(nm_i->nat_bitmap);
2046 sbi->nm_info = NULL;
2050 int __init create_node_manager_caches(void)
2052 nat_entry_slab = f2fs_kmem_cache_create("nat_entry",
2053 sizeof(struct nat_entry));
2054 if (!nat_entry_slab)
2057 free_nid_slab = f2fs_kmem_cache_create("free_nid",
2058 sizeof(struct free_nid));
2060 goto destory_nat_entry;
2062 nat_entry_set_slab = f2fs_kmem_cache_create("nat_entry_set",
2063 sizeof(struct nat_entry_set));
2064 if (!nat_entry_set_slab)
2065 goto destory_free_nid;
2069 kmem_cache_destroy(free_nid_slab);
2071 kmem_cache_destroy(nat_entry_slab);
2076 void destroy_node_manager_caches(void)
2078 kmem_cache_destroy(nat_entry_set_slab);
2079 kmem_cache_destroy(free_nid_slab);
2080 kmem_cache_destroy(nat_entry_slab);