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 bool available_free_memory(struct f2fs_sb_info *sbi, int type)
31 struct f2fs_nm_info *nm_i = NM_I(sbi);
33 unsigned long mem_size = 0;
37 /* give 25%, 25%, 50% memory for each components respectively */
38 if (type == FREE_NIDS) {
39 mem_size = (nm_i->fcnt * sizeof(struct free_nid)) >> 12;
40 res = mem_size < ((val.totalram * nm_i->ram_thresh / 100) >> 2);
41 } else if (type == NAT_ENTRIES) {
42 mem_size = (nm_i->nat_cnt * sizeof(struct nat_entry)) >> 12;
43 res = mem_size < ((val.totalram * nm_i->ram_thresh / 100) >> 2);
44 } else if (type == DIRTY_DENTS) {
45 mem_size = get_pages(sbi, F2FS_DIRTY_DENTS);
46 res = mem_size < ((val.totalram * nm_i->ram_thresh / 100) >> 1);
51 static void clear_node_page_dirty(struct page *page)
53 struct address_space *mapping = page->mapping;
54 struct f2fs_sb_info *sbi = F2FS_SB(mapping->host->i_sb);
55 unsigned int long flags;
57 if (PageDirty(page)) {
58 spin_lock_irqsave(&mapping->tree_lock, flags);
59 radix_tree_tag_clear(&mapping->page_tree,
62 spin_unlock_irqrestore(&mapping->tree_lock, flags);
64 clear_page_dirty_for_io(page);
65 dec_page_count(sbi, F2FS_DIRTY_NODES);
67 ClearPageUptodate(page);
70 static struct page *get_current_nat_page(struct f2fs_sb_info *sbi, nid_t nid)
72 pgoff_t index = current_nat_addr(sbi, nid);
73 return get_meta_page(sbi, index);
76 static struct page *get_next_nat_page(struct f2fs_sb_info *sbi, nid_t nid)
78 struct page *src_page;
79 struct page *dst_page;
84 struct f2fs_nm_info *nm_i = NM_I(sbi);
86 src_off = current_nat_addr(sbi, nid);
87 dst_off = next_nat_addr(sbi, src_off);
89 /* get current nat block page with lock */
90 src_page = get_meta_page(sbi, src_off);
92 /* Dirty src_page means that it is already the new target NAT page. */
93 if (PageDirty(src_page))
96 dst_page = grab_meta_page(sbi, dst_off);
98 src_addr = page_address(src_page);
99 dst_addr = page_address(dst_page);
100 memcpy(dst_addr, src_addr, PAGE_CACHE_SIZE);
101 set_page_dirty(dst_page);
102 f2fs_put_page(src_page, 1);
104 set_to_next_nat(nm_i, nid);
109 static struct nat_entry *__lookup_nat_cache(struct f2fs_nm_info *nm_i, nid_t n)
111 return radix_tree_lookup(&nm_i->nat_root, n);
114 static unsigned int __gang_lookup_nat_cache(struct f2fs_nm_info *nm_i,
115 nid_t start, unsigned int nr, struct nat_entry **ep)
117 return radix_tree_gang_lookup(&nm_i->nat_root, (void **)ep, start, nr);
120 static void __del_from_nat_cache(struct f2fs_nm_info *nm_i, struct nat_entry *e)
123 radix_tree_delete(&nm_i->nat_root, nat_get_nid(e));
125 kmem_cache_free(nat_entry_slab, e);
128 int is_checkpointed_node(struct f2fs_sb_info *sbi, nid_t nid)
130 struct f2fs_nm_info *nm_i = NM_I(sbi);
134 read_lock(&nm_i->nat_tree_lock);
135 e = __lookup_nat_cache(nm_i, nid);
136 if (e && !e->checkpointed)
138 read_unlock(&nm_i->nat_tree_lock);
142 bool fsync_mark_done(struct f2fs_sb_info *sbi, nid_t nid)
144 struct f2fs_nm_info *nm_i = NM_I(sbi);
146 bool fsync_done = false;
148 read_lock(&nm_i->nat_tree_lock);
149 e = __lookup_nat_cache(nm_i, nid);
151 fsync_done = e->fsync_done;
152 read_unlock(&nm_i->nat_tree_lock);
156 static struct nat_entry *grab_nat_entry(struct f2fs_nm_info *nm_i, nid_t nid)
158 struct nat_entry *new;
160 new = kmem_cache_alloc(nat_entry_slab, GFP_ATOMIC);
163 if (radix_tree_insert(&nm_i->nat_root, nid, new)) {
164 kmem_cache_free(nat_entry_slab, new);
167 memset(new, 0, sizeof(struct nat_entry));
168 nat_set_nid(new, nid);
169 new->checkpointed = true;
170 list_add_tail(&new->list, &nm_i->nat_entries);
175 static void cache_nat_entry(struct f2fs_nm_info *nm_i, nid_t nid,
176 struct f2fs_nat_entry *ne)
180 write_lock(&nm_i->nat_tree_lock);
181 e = __lookup_nat_cache(nm_i, nid);
183 e = grab_nat_entry(nm_i, nid);
185 write_unlock(&nm_i->nat_tree_lock);
188 node_info_from_raw_nat(&e->ni, ne);
190 write_unlock(&nm_i->nat_tree_lock);
193 static void set_node_addr(struct f2fs_sb_info *sbi, struct node_info *ni,
194 block_t new_blkaddr, bool fsync_done)
196 struct f2fs_nm_info *nm_i = NM_I(sbi);
199 write_lock(&nm_i->nat_tree_lock);
200 e = __lookup_nat_cache(nm_i, ni->nid);
202 e = grab_nat_entry(nm_i, ni->nid);
204 write_unlock(&nm_i->nat_tree_lock);
208 f2fs_bug_on(ni->blk_addr == NEW_ADDR);
209 } else if (new_blkaddr == NEW_ADDR) {
211 * when nid is reallocated,
212 * previous nat entry can be remained in nat cache.
213 * So, reinitialize it with new information.
216 f2fs_bug_on(ni->blk_addr != NULL_ADDR);
220 f2fs_bug_on(nat_get_blkaddr(e) != ni->blk_addr);
221 f2fs_bug_on(nat_get_blkaddr(e) == NULL_ADDR &&
222 new_blkaddr == NULL_ADDR);
223 f2fs_bug_on(nat_get_blkaddr(e) == NEW_ADDR &&
224 new_blkaddr == NEW_ADDR);
225 f2fs_bug_on(nat_get_blkaddr(e) != NEW_ADDR &&
226 nat_get_blkaddr(e) != NULL_ADDR &&
227 new_blkaddr == NEW_ADDR);
229 /* increament version no as node is removed */
230 if (nat_get_blkaddr(e) != NEW_ADDR && new_blkaddr == NULL_ADDR) {
231 unsigned char version = nat_get_version(e);
232 nat_set_version(e, inc_node_version(version));
236 nat_set_blkaddr(e, new_blkaddr);
237 __set_nat_cache_dirty(nm_i, e);
239 /* update fsync_mark if its inode nat entry is still alive */
240 e = __lookup_nat_cache(nm_i, ni->ino);
242 e->fsync_done = fsync_done;
243 write_unlock(&nm_i->nat_tree_lock);
246 int try_to_free_nats(struct f2fs_sb_info *sbi, int nr_shrink)
248 struct f2fs_nm_info *nm_i = NM_I(sbi);
250 if (available_free_memory(sbi, NAT_ENTRIES))
253 write_lock(&nm_i->nat_tree_lock);
254 while (nr_shrink && !list_empty(&nm_i->nat_entries)) {
255 struct nat_entry *ne;
256 ne = list_first_entry(&nm_i->nat_entries,
257 struct nat_entry, list);
258 __del_from_nat_cache(nm_i, ne);
261 write_unlock(&nm_i->nat_tree_lock);
266 * This function returns always success
268 void get_node_info(struct f2fs_sb_info *sbi, nid_t nid, struct node_info *ni)
270 struct f2fs_nm_info *nm_i = NM_I(sbi);
271 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
272 struct f2fs_summary_block *sum = curseg->sum_blk;
273 nid_t start_nid = START_NID(nid);
274 struct f2fs_nat_block *nat_blk;
275 struct page *page = NULL;
276 struct f2fs_nat_entry ne;
280 memset(&ne, 0, sizeof(struct f2fs_nat_entry));
283 /* Check nat cache */
284 read_lock(&nm_i->nat_tree_lock);
285 e = __lookup_nat_cache(nm_i, nid);
287 ni->ino = nat_get_ino(e);
288 ni->blk_addr = nat_get_blkaddr(e);
289 ni->version = nat_get_version(e);
291 read_unlock(&nm_i->nat_tree_lock);
295 /* Check current segment summary */
296 mutex_lock(&curseg->curseg_mutex);
297 i = lookup_journal_in_cursum(sum, NAT_JOURNAL, nid, 0);
299 ne = nat_in_journal(sum, i);
300 node_info_from_raw_nat(ni, &ne);
302 mutex_unlock(&curseg->curseg_mutex);
306 /* Fill node_info from nat page */
307 page = get_current_nat_page(sbi, start_nid);
308 nat_blk = (struct f2fs_nat_block *)page_address(page);
309 ne = nat_blk->entries[nid - start_nid];
310 node_info_from_raw_nat(ni, &ne);
311 f2fs_put_page(page, 1);
313 /* cache nat entry */
314 cache_nat_entry(NM_I(sbi), nid, &ne);
318 * The maximum depth is four.
319 * Offset[0] will have raw inode offset.
321 static int get_node_path(struct f2fs_inode_info *fi, long block,
322 int offset[4], unsigned int noffset[4])
324 const long direct_index = ADDRS_PER_INODE(fi);
325 const long direct_blks = ADDRS_PER_BLOCK;
326 const long dptrs_per_blk = NIDS_PER_BLOCK;
327 const long indirect_blks = ADDRS_PER_BLOCK * NIDS_PER_BLOCK;
328 const long dindirect_blks = indirect_blks * NIDS_PER_BLOCK;
334 if (block < direct_index) {
338 block -= direct_index;
339 if (block < direct_blks) {
340 offset[n++] = NODE_DIR1_BLOCK;
346 block -= direct_blks;
347 if (block < direct_blks) {
348 offset[n++] = NODE_DIR2_BLOCK;
354 block -= direct_blks;
355 if (block < indirect_blks) {
356 offset[n++] = NODE_IND1_BLOCK;
358 offset[n++] = block / direct_blks;
359 noffset[n] = 4 + offset[n - 1];
360 offset[n] = block % direct_blks;
364 block -= indirect_blks;
365 if (block < indirect_blks) {
366 offset[n++] = NODE_IND2_BLOCK;
367 noffset[n] = 4 + dptrs_per_blk;
368 offset[n++] = block / direct_blks;
369 noffset[n] = 5 + dptrs_per_blk + offset[n - 1];
370 offset[n] = block % direct_blks;
374 block -= indirect_blks;
375 if (block < dindirect_blks) {
376 offset[n++] = NODE_DIND_BLOCK;
377 noffset[n] = 5 + (dptrs_per_blk * 2);
378 offset[n++] = block / indirect_blks;
379 noffset[n] = 6 + (dptrs_per_blk * 2) +
380 offset[n - 1] * (dptrs_per_blk + 1);
381 offset[n++] = (block / direct_blks) % dptrs_per_blk;
382 noffset[n] = 7 + (dptrs_per_blk * 2) +
383 offset[n - 2] * (dptrs_per_blk + 1) +
385 offset[n] = block % direct_blks;
396 * Caller should call f2fs_put_dnode(dn).
397 * Also, it should grab and release a rwsem by calling f2fs_lock_op() and
398 * f2fs_unlock_op() only if ro is not set RDONLY_NODE.
399 * In the case of RDONLY_NODE, we don't need to care about mutex.
401 int get_dnode_of_data(struct dnode_of_data *dn, pgoff_t index, int mode)
403 struct f2fs_sb_info *sbi = F2FS_SB(dn->inode->i_sb);
404 struct page *npage[4];
407 unsigned int noffset[4];
412 level = get_node_path(F2FS_I(dn->inode), index, offset, noffset);
414 nids[0] = dn->inode->i_ino;
415 npage[0] = dn->inode_page;
418 npage[0] = get_node_page(sbi, nids[0]);
419 if (IS_ERR(npage[0]))
420 return PTR_ERR(npage[0]);
424 nids[1] = get_nid(parent, offset[0], true);
425 dn->inode_page = npage[0];
426 dn->inode_page_locked = true;
428 /* get indirect or direct nodes */
429 for (i = 1; i <= level; i++) {
432 if (!nids[i] && mode == ALLOC_NODE) {
434 if (!alloc_nid(sbi, &(nids[i]))) {
440 npage[i] = new_node_page(dn, noffset[i], NULL);
441 if (IS_ERR(npage[i])) {
442 alloc_nid_failed(sbi, nids[i]);
443 err = PTR_ERR(npage[i]);
447 set_nid(parent, offset[i - 1], nids[i], i == 1);
448 alloc_nid_done(sbi, nids[i]);
450 } else if (mode == LOOKUP_NODE_RA && i == level && level > 1) {
451 npage[i] = get_node_page_ra(parent, offset[i - 1]);
452 if (IS_ERR(npage[i])) {
453 err = PTR_ERR(npage[i]);
459 dn->inode_page_locked = false;
462 f2fs_put_page(parent, 1);
466 npage[i] = get_node_page(sbi, nids[i]);
467 if (IS_ERR(npage[i])) {
468 err = PTR_ERR(npage[i]);
469 f2fs_put_page(npage[0], 0);
475 nids[i + 1] = get_nid(parent, offset[i], false);
478 dn->nid = nids[level];
479 dn->ofs_in_node = offset[level];
480 dn->node_page = npage[level];
481 dn->data_blkaddr = datablock_addr(dn->node_page, dn->ofs_in_node);
485 f2fs_put_page(parent, 1);
487 f2fs_put_page(npage[0], 0);
489 dn->inode_page = NULL;
490 dn->node_page = NULL;
494 static void truncate_node(struct dnode_of_data *dn)
496 struct f2fs_sb_info *sbi = F2FS_SB(dn->inode->i_sb);
499 get_node_info(sbi, dn->nid, &ni);
500 if (dn->inode->i_blocks == 0) {
501 f2fs_bug_on(ni.blk_addr != NULL_ADDR);
504 f2fs_bug_on(ni.blk_addr == NULL_ADDR);
506 /* Deallocate node address */
507 invalidate_blocks(sbi, ni.blk_addr);
508 dec_valid_node_count(sbi, dn->inode);
509 set_node_addr(sbi, &ni, NULL_ADDR, false);
511 if (dn->nid == dn->inode->i_ino) {
512 remove_orphan_inode(sbi, dn->nid);
513 dec_valid_inode_count(sbi);
518 clear_node_page_dirty(dn->node_page);
519 F2FS_SET_SB_DIRT(sbi);
521 f2fs_put_page(dn->node_page, 1);
523 invalidate_mapping_pages(NODE_MAPPING(sbi),
524 dn->node_page->index, dn->node_page->index);
526 dn->node_page = NULL;
527 trace_f2fs_truncate_node(dn->inode, dn->nid, ni.blk_addr);
530 static int truncate_dnode(struct dnode_of_data *dn)
532 struct f2fs_sb_info *sbi = F2FS_SB(dn->inode->i_sb);
538 /* get direct node */
539 page = get_node_page(sbi, dn->nid);
540 if (IS_ERR(page) && PTR_ERR(page) == -ENOENT)
542 else if (IS_ERR(page))
543 return PTR_ERR(page);
545 /* Make dnode_of_data for parameter */
546 dn->node_page = page;
548 truncate_data_blocks(dn);
553 static int truncate_nodes(struct dnode_of_data *dn, unsigned int nofs,
556 struct f2fs_sb_info *sbi = F2FS_SB(dn->inode->i_sb);
557 struct dnode_of_data rdn = *dn;
559 struct f2fs_node *rn;
561 unsigned int child_nofs;
566 return NIDS_PER_BLOCK + 1;
568 trace_f2fs_truncate_nodes_enter(dn->inode, dn->nid, dn->data_blkaddr);
570 page = get_node_page(sbi, dn->nid);
572 trace_f2fs_truncate_nodes_exit(dn->inode, PTR_ERR(page));
573 return PTR_ERR(page);
576 rn = F2FS_NODE(page);
578 for (i = ofs; i < NIDS_PER_BLOCK; i++, freed++) {
579 child_nid = le32_to_cpu(rn->in.nid[i]);
583 ret = truncate_dnode(&rdn);
586 set_nid(page, i, 0, false);
589 child_nofs = nofs + ofs * (NIDS_PER_BLOCK + 1) + 1;
590 for (i = ofs; i < NIDS_PER_BLOCK; i++) {
591 child_nid = le32_to_cpu(rn->in.nid[i]);
592 if (child_nid == 0) {
593 child_nofs += NIDS_PER_BLOCK + 1;
597 ret = truncate_nodes(&rdn, child_nofs, 0, depth - 1);
598 if (ret == (NIDS_PER_BLOCK + 1)) {
599 set_nid(page, i, 0, false);
601 } else if (ret < 0 && ret != -ENOENT) {
609 /* remove current indirect node */
610 dn->node_page = page;
614 f2fs_put_page(page, 1);
616 trace_f2fs_truncate_nodes_exit(dn->inode, freed);
620 f2fs_put_page(page, 1);
621 trace_f2fs_truncate_nodes_exit(dn->inode, ret);
625 static int truncate_partial_nodes(struct dnode_of_data *dn,
626 struct f2fs_inode *ri, int *offset, int depth)
628 struct f2fs_sb_info *sbi = F2FS_SB(dn->inode->i_sb);
629 struct page *pages[2];
636 nid[0] = le32_to_cpu(ri->i_nid[offset[0] - NODE_DIR1_BLOCK]);
640 /* get indirect nodes in the path */
641 for (i = 0; i < idx + 1; i++) {
642 /* refernece count'll be increased */
643 pages[i] = get_node_page(sbi, nid[i]);
644 if (IS_ERR(pages[i])) {
645 err = PTR_ERR(pages[i]);
649 nid[i + 1] = get_nid(pages[i], offset[i + 1], false);
652 /* free direct nodes linked to a partial indirect node */
653 for (i = offset[idx + 1]; i < NIDS_PER_BLOCK; i++) {
654 child_nid = get_nid(pages[idx], i, false);
658 err = truncate_dnode(dn);
661 set_nid(pages[idx], i, 0, false);
664 if (offset[idx + 1] == 0) {
665 dn->node_page = pages[idx];
669 f2fs_put_page(pages[idx], 1);
675 for (i = idx; i >= 0; i--)
676 f2fs_put_page(pages[i], 1);
678 trace_f2fs_truncate_partial_nodes(dn->inode, nid, depth, err);
684 * All the block addresses of data and nodes should be nullified.
686 int truncate_inode_blocks(struct inode *inode, pgoff_t from)
688 struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
689 int err = 0, cont = 1;
690 int level, offset[4], noffset[4];
691 unsigned int nofs = 0;
692 struct f2fs_inode *ri;
693 struct dnode_of_data dn;
696 trace_f2fs_truncate_inode_blocks_enter(inode, from);
698 level = get_node_path(F2FS_I(inode), from, offset, noffset);
700 page = get_node_page(sbi, inode->i_ino);
702 trace_f2fs_truncate_inode_blocks_exit(inode, PTR_ERR(page));
703 return PTR_ERR(page);
706 set_new_dnode(&dn, inode, page, NULL, 0);
709 ri = F2FS_INODE(page);
717 if (!offset[level - 1])
719 err = truncate_partial_nodes(&dn, ri, offset, level);
720 if (err < 0 && err != -ENOENT)
722 nofs += 1 + NIDS_PER_BLOCK;
725 nofs = 5 + 2 * NIDS_PER_BLOCK;
726 if (!offset[level - 1])
728 err = truncate_partial_nodes(&dn, ri, offset, level);
729 if (err < 0 && err != -ENOENT)
738 dn.nid = le32_to_cpu(ri->i_nid[offset[0] - NODE_DIR1_BLOCK]);
740 case NODE_DIR1_BLOCK:
741 case NODE_DIR2_BLOCK:
742 err = truncate_dnode(&dn);
745 case NODE_IND1_BLOCK:
746 case NODE_IND2_BLOCK:
747 err = truncate_nodes(&dn, nofs, offset[1], 2);
750 case NODE_DIND_BLOCK:
751 err = truncate_nodes(&dn, nofs, offset[1], 3);
758 if (err < 0 && err != -ENOENT)
760 if (offset[1] == 0 &&
761 ri->i_nid[offset[0] - NODE_DIR1_BLOCK]) {
763 if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
764 f2fs_put_page(page, 1);
767 f2fs_wait_on_page_writeback(page, NODE);
768 ri->i_nid[offset[0] - NODE_DIR1_BLOCK] = 0;
769 set_page_dirty(page);
777 f2fs_put_page(page, 0);
778 trace_f2fs_truncate_inode_blocks_exit(inode, err);
779 return err > 0 ? 0 : err;
782 int truncate_xattr_node(struct inode *inode, struct page *page)
784 struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
785 nid_t nid = F2FS_I(inode)->i_xattr_nid;
786 struct dnode_of_data dn;
792 npage = get_node_page(sbi, nid);
794 return PTR_ERR(npage);
796 F2FS_I(inode)->i_xattr_nid = 0;
798 /* need to do checkpoint during fsync */
799 F2FS_I(inode)->xattr_ver = cur_cp_version(F2FS_CKPT(sbi));
801 set_new_dnode(&dn, inode, page, npage, nid);
804 dn.inode_page_locked = true;
810 * Caller should grab and release a rwsem by calling f2fs_lock_op() and
813 void remove_inode_page(struct inode *inode)
815 struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
817 nid_t ino = inode->i_ino;
818 struct dnode_of_data dn;
820 page = get_node_page(sbi, ino);
824 if (truncate_xattr_node(inode, page)) {
825 f2fs_put_page(page, 1);
828 /* 0 is possible, after f2fs_new_inode() is failed */
829 f2fs_bug_on(inode->i_blocks != 0 && inode->i_blocks != 1);
830 set_new_dnode(&dn, inode, page, page, ino);
834 struct page *new_inode_page(struct inode *inode, const struct qstr *name)
836 struct dnode_of_data dn;
838 /* allocate inode page for new inode */
839 set_new_dnode(&dn, inode, NULL, NULL, inode->i_ino);
841 /* caller should f2fs_put_page(page, 1); */
842 return new_node_page(&dn, 0, NULL);
845 struct page *new_node_page(struct dnode_of_data *dn,
846 unsigned int ofs, struct page *ipage)
848 struct f2fs_sb_info *sbi = F2FS_SB(dn->inode->i_sb);
849 struct node_info old_ni, new_ni;
853 if (unlikely(is_inode_flag_set(F2FS_I(dn->inode), FI_NO_ALLOC)))
854 return ERR_PTR(-EPERM);
856 page = grab_cache_page(NODE_MAPPING(sbi), dn->nid);
858 return ERR_PTR(-ENOMEM);
860 if (unlikely(!inc_valid_node_count(sbi, dn->inode))) {
865 get_node_info(sbi, dn->nid, &old_ni);
867 /* Reinitialize old_ni with new node page */
868 f2fs_bug_on(old_ni.blk_addr != NULL_ADDR);
870 new_ni.ino = dn->inode->i_ino;
871 set_node_addr(sbi, &new_ni, NEW_ADDR, false);
873 f2fs_wait_on_page_writeback(page, NODE);
874 fill_node_footer(page, dn->nid, dn->inode->i_ino, ofs, true);
875 set_cold_node(dn->inode, page);
876 SetPageUptodate(page);
877 set_page_dirty(page);
879 if (f2fs_has_xattr_block(ofs))
880 F2FS_I(dn->inode)->i_xattr_nid = dn->nid;
882 dn->node_page = page;
884 update_inode(dn->inode, ipage);
888 inc_valid_inode_count(sbi);
893 clear_node_page_dirty(page);
894 f2fs_put_page(page, 1);
899 * Caller should do after getting the following values.
900 * 0: f2fs_put_page(page, 0)
901 * LOCKED_PAGE: f2fs_put_page(page, 1)
904 static int read_node_page(struct page *page, int rw)
906 struct f2fs_sb_info *sbi = F2FS_SB(page->mapping->host->i_sb);
909 get_node_info(sbi, page->index, &ni);
911 if (unlikely(ni.blk_addr == NULL_ADDR)) {
912 f2fs_put_page(page, 1);
916 if (PageUptodate(page))
919 return f2fs_submit_page_bio(sbi, page, ni.blk_addr, rw);
923 * Readahead a node page
925 void ra_node_page(struct f2fs_sb_info *sbi, nid_t nid)
930 apage = find_get_page(NODE_MAPPING(sbi), nid);
931 if (apage && PageUptodate(apage)) {
932 f2fs_put_page(apage, 0);
935 f2fs_put_page(apage, 0);
937 apage = grab_cache_page(NODE_MAPPING(sbi), nid);
941 err = read_node_page(apage, READA);
943 f2fs_put_page(apage, 0);
944 else if (err == LOCKED_PAGE)
945 f2fs_put_page(apage, 1);
948 struct page *get_node_page(struct f2fs_sb_info *sbi, pgoff_t nid)
953 page = grab_cache_page(NODE_MAPPING(sbi), nid);
955 return ERR_PTR(-ENOMEM);
957 err = read_node_page(page, READ_SYNC);
960 else if (err == LOCKED_PAGE)
964 if (unlikely(!PageUptodate(page) || nid != nid_of_node(page))) {
965 f2fs_put_page(page, 1);
966 return ERR_PTR(-EIO);
968 if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
969 f2fs_put_page(page, 1);
973 mark_page_accessed(page);
978 * Return a locked page for the desired node page.
979 * And, readahead MAX_RA_NODE number of node pages.
981 struct page *get_node_page_ra(struct page *parent, int start)
983 struct f2fs_sb_info *sbi = F2FS_SB(parent->mapping->host->i_sb);
984 struct blk_plug plug;
989 /* First, try getting the desired direct node. */
990 nid = get_nid(parent, start, false);
992 return ERR_PTR(-ENOENT);
994 page = grab_cache_page(NODE_MAPPING(sbi), nid);
996 return ERR_PTR(-ENOMEM);
998 err = read_node_page(page, READ_SYNC);
1000 return ERR_PTR(err);
1001 else if (err == LOCKED_PAGE)
1004 blk_start_plug(&plug);
1006 /* Then, try readahead for siblings of the desired node */
1007 end = start + MAX_RA_NODE;
1008 end = min(end, NIDS_PER_BLOCK);
1009 for (i = start + 1; i < end; i++) {
1010 nid = get_nid(parent, i, false);
1013 ra_node_page(sbi, nid);
1016 blk_finish_plug(&plug);
1019 if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1020 f2fs_put_page(page, 1);
1024 if (unlikely(!PageUptodate(page))) {
1025 f2fs_put_page(page, 1);
1026 return ERR_PTR(-EIO);
1028 mark_page_accessed(page);
1032 void sync_inode_page(struct dnode_of_data *dn)
1034 if (IS_INODE(dn->node_page) || dn->inode_page == dn->node_page) {
1035 update_inode(dn->inode, dn->node_page);
1036 } else if (dn->inode_page) {
1037 if (!dn->inode_page_locked)
1038 lock_page(dn->inode_page);
1039 update_inode(dn->inode, dn->inode_page);
1040 if (!dn->inode_page_locked)
1041 unlock_page(dn->inode_page);
1043 update_inode_page(dn->inode);
1047 int sync_node_pages(struct f2fs_sb_info *sbi, nid_t ino,
1048 struct writeback_control *wbc)
1051 struct pagevec pvec;
1052 int step = ino ? 2 : 0;
1053 int nwritten = 0, wrote = 0;
1055 pagevec_init(&pvec, 0);
1061 while (index <= end) {
1063 nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1064 PAGECACHE_TAG_DIRTY,
1065 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
1069 for (i = 0; i < nr_pages; i++) {
1070 struct page *page = pvec.pages[i];
1073 * flushing sequence with step:
1078 if (step == 0 && IS_DNODE(page))
1080 if (step == 1 && (!IS_DNODE(page) ||
1081 is_cold_node(page)))
1083 if (step == 2 && (!IS_DNODE(page) ||
1084 !is_cold_node(page)))
1089 * we should not skip writing node pages.
1091 if (ino && ino_of_node(page) == ino)
1093 else if (!trylock_page(page))
1096 if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1101 if (ino && ino_of_node(page) != ino)
1102 goto continue_unlock;
1104 if (!PageDirty(page)) {
1105 /* someone wrote it for us */
1106 goto continue_unlock;
1109 if (!clear_page_dirty_for_io(page))
1110 goto continue_unlock;
1112 /* called by fsync() */
1113 if (ino && IS_DNODE(page)) {
1114 int mark = !is_checkpointed_node(sbi, ino);
1115 set_fsync_mark(page, 1);
1117 set_dentry_mark(page, mark);
1120 set_fsync_mark(page, 0);
1121 set_dentry_mark(page, 0);
1123 NODE_MAPPING(sbi)->a_ops->writepage(page, wbc);
1126 if (--wbc->nr_to_write == 0)
1129 pagevec_release(&pvec);
1132 if (wbc->nr_to_write == 0) {
1144 f2fs_submit_merged_bio(sbi, NODE, WRITE);
1148 int wait_on_node_pages_writeback(struct f2fs_sb_info *sbi, nid_t ino)
1150 pgoff_t index = 0, end = LONG_MAX;
1151 struct pagevec pvec;
1152 int ret2 = 0, ret = 0;
1154 pagevec_init(&pvec, 0);
1156 while (index <= end) {
1158 nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1159 PAGECACHE_TAG_WRITEBACK,
1160 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
1164 for (i = 0; i < nr_pages; i++) {
1165 struct page *page = pvec.pages[i];
1167 /* until radix tree lookup accepts end_index */
1168 if (unlikely(page->index > end))
1171 if (ino && ino_of_node(page) == ino) {
1172 f2fs_wait_on_page_writeback(page, NODE);
1173 if (TestClearPageError(page))
1177 pagevec_release(&pvec);
1181 if (unlikely(test_and_clear_bit(AS_ENOSPC, &NODE_MAPPING(sbi)->flags)))
1183 if (unlikely(test_and_clear_bit(AS_EIO, &NODE_MAPPING(sbi)->flags)))
1190 static int f2fs_write_node_page(struct page *page,
1191 struct writeback_control *wbc)
1193 struct f2fs_sb_info *sbi = F2FS_SB(page->mapping->host->i_sb);
1196 struct node_info ni;
1197 struct f2fs_io_info fio = {
1199 .rw = (wbc->sync_mode == WB_SYNC_ALL) ? WRITE_SYNC : WRITE,
1202 if (unlikely(sbi->por_doing))
1205 f2fs_wait_on_page_writeback(page, NODE);
1207 /* get old block addr of this node page */
1208 nid = nid_of_node(page);
1209 f2fs_bug_on(page->index != nid);
1211 get_node_info(sbi, nid, &ni);
1213 /* This page is already truncated */
1214 if (unlikely(ni.blk_addr == NULL_ADDR)) {
1215 dec_page_count(sbi, F2FS_DIRTY_NODES);
1220 if (wbc->for_reclaim)
1223 mutex_lock(&sbi->node_write);
1224 set_page_writeback(page);
1225 write_node_page(sbi, page, &fio, nid, ni.blk_addr, &new_addr);
1226 set_node_addr(sbi, &ni, new_addr, is_fsync_dnode(page));
1227 dec_page_count(sbi, F2FS_DIRTY_NODES);
1228 mutex_unlock(&sbi->node_write);
1233 redirty_page_for_writepage(wbc, 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);
1318 static int add_free_nid(struct f2fs_sb_info *sbi, nid_t nid, bool build)
1320 struct f2fs_nm_info *nm_i = NM_I(sbi);
1322 struct nat_entry *ne;
1323 bool allocated = false;
1325 if (!available_free_memory(sbi, 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 bool need_free = false;
1365 spin_lock(&nm_i->free_nid_list_lock);
1366 i = __lookup_free_nid_list(nm_i, nid);
1367 if (i && i->state == NID_NEW) {
1368 __del_from_free_nid_list(nm_i, i);
1372 spin_unlock(&nm_i->free_nid_list_lock);
1375 kmem_cache_free(free_nid_slab, i);
1378 static void scan_nat_page(struct f2fs_sb_info *sbi,
1379 struct page *nat_page, nid_t start_nid)
1381 struct f2fs_nm_info *nm_i = NM_I(sbi);
1382 struct f2fs_nat_block *nat_blk = page_address(nat_page);
1386 i = start_nid % NAT_ENTRY_PER_BLOCK;
1388 for (; i < NAT_ENTRY_PER_BLOCK; i++, start_nid++) {
1390 if (unlikely(start_nid >= nm_i->max_nid))
1393 blk_addr = le32_to_cpu(nat_blk->entries[i].block_addr);
1394 f2fs_bug_on(blk_addr == NEW_ADDR);
1395 if (blk_addr == NULL_ADDR) {
1396 if (add_free_nid(sbi, start_nid, true) < 0)
1402 static void build_free_nids(struct f2fs_sb_info *sbi)
1404 struct f2fs_nm_info *nm_i = NM_I(sbi);
1405 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
1406 struct f2fs_summary_block *sum = curseg->sum_blk;
1408 nid_t nid = nm_i->next_scan_nid;
1410 /* Enough entries */
1411 if (nm_i->fcnt > NAT_ENTRY_PER_BLOCK)
1414 /* readahead nat pages to be scanned */
1415 ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nid), FREE_NID_PAGES, META_NAT);
1418 struct page *page = get_current_nat_page(sbi, nid);
1420 scan_nat_page(sbi, page, nid);
1421 f2fs_put_page(page, 1);
1423 nid += (NAT_ENTRY_PER_BLOCK - (nid % NAT_ENTRY_PER_BLOCK));
1424 if (unlikely(nid >= nm_i->max_nid))
1427 if (i++ == FREE_NID_PAGES)
1431 /* go to the next free nat pages to find free nids abundantly */
1432 nm_i->next_scan_nid = nid;
1434 /* find free nids from current sum_pages */
1435 mutex_lock(&curseg->curseg_mutex);
1436 for (i = 0; i < nats_in_cursum(sum); i++) {
1437 block_t addr = le32_to_cpu(nat_in_journal(sum, i).block_addr);
1438 nid = le32_to_cpu(nid_in_journal(sum, i));
1439 if (addr == NULL_ADDR)
1440 add_free_nid(sbi, nid, true);
1442 remove_free_nid(nm_i, nid);
1444 mutex_unlock(&curseg->curseg_mutex);
1448 * If this function returns success, caller can obtain a new nid
1449 * from second parameter of this function.
1450 * The returned nid could be used ino as well as nid when inode is created.
1452 bool alloc_nid(struct f2fs_sb_info *sbi, nid_t *nid)
1454 struct f2fs_nm_info *nm_i = NM_I(sbi);
1455 struct free_nid *i = NULL;
1457 if (unlikely(sbi->total_valid_node_count + 1 > nm_i->available_nids))
1460 spin_lock(&nm_i->free_nid_list_lock);
1462 /* We should not use stale free nids created by build_free_nids */
1463 if (nm_i->fcnt && !on_build_free_nids(nm_i)) {
1464 f2fs_bug_on(list_empty(&nm_i->free_nid_list));
1465 list_for_each_entry(i, &nm_i->free_nid_list, list)
1466 if (i->state == NID_NEW)
1469 f2fs_bug_on(i->state != NID_NEW);
1471 i->state = NID_ALLOC;
1473 spin_unlock(&nm_i->free_nid_list_lock);
1476 spin_unlock(&nm_i->free_nid_list_lock);
1478 /* Let's scan nat pages and its caches to get free nids */
1479 mutex_lock(&nm_i->build_lock);
1480 build_free_nids(sbi);
1481 mutex_unlock(&nm_i->build_lock);
1486 * alloc_nid() should be called prior to this function.
1488 void alloc_nid_done(struct f2fs_sb_info *sbi, nid_t nid)
1490 struct f2fs_nm_info *nm_i = NM_I(sbi);
1493 spin_lock(&nm_i->free_nid_list_lock);
1494 i = __lookup_free_nid_list(nm_i, nid);
1495 f2fs_bug_on(!i || i->state != NID_ALLOC);
1496 __del_from_free_nid_list(nm_i, i);
1497 spin_unlock(&nm_i->free_nid_list_lock);
1499 kmem_cache_free(free_nid_slab, i);
1503 * alloc_nid() should be called prior to this function.
1505 void alloc_nid_failed(struct f2fs_sb_info *sbi, nid_t nid)
1507 struct f2fs_nm_info *nm_i = NM_I(sbi);
1509 bool need_free = false;
1514 spin_lock(&nm_i->free_nid_list_lock);
1515 i = __lookup_free_nid_list(nm_i, nid);
1516 f2fs_bug_on(!i || i->state != NID_ALLOC);
1517 if (!available_free_memory(sbi, FREE_NIDS)) {
1518 __del_from_free_nid_list(nm_i, i);
1524 spin_unlock(&nm_i->free_nid_list_lock);
1527 kmem_cache_free(free_nid_slab, i);
1530 void recover_node_page(struct f2fs_sb_info *sbi, struct page *page,
1531 struct f2fs_summary *sum, struct node_info *ni,
1532 block_t new_blkaddr)
1534 rewrite_node_page(sbi, page, sum, ni->blk_addr, new_blkaddr);
1535 set_node_addr(sbi, ni, new_blkaddr, false);
1536 clear_node_page_dirty(page);
1539 static void recover_inline_xattr(struct inode *inode, struct page *page)
1541 struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
1542 void *src_addr, *dst_addr;
1545 struct f2fs_inode *ri;
1547 if (!f2fs_has_inline_xattr(inode))
1550 if (!IS_INODE(page))
1553 ri = F2FS_INODE(page);
1554 if (!(ri->i_inline & F2FS_INLINE_XATTR))
1557 ipage = get_node_page(sbi, inode->i_ino);
1558 f2fs_bug_on(IS_ERR(ipage));
1560 dst_addr = inline_xattr_addr(ipage);
1561 src_addr = inline_xattr_addr(page);
1562 inline_size = inline_xattr_size(inode);
1564 f2fs_wait_on_page_writeback(ipage, NODE);
1565 memcpy(dst_addr, src_addr, inline_size);
1567 update_inode(inode, ipage);
1568 f2fs_put_page(ipage, 1);
1571 bool recover_xattr_data(struct inode *inode, struct page *page, block_t blkaddr)
1573 struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
1574 nid_t prev_xnid = F2FS_I(inode)->i_xattr_nid;
1575 nid_t new_xnid = nid_of_node(page);
1576 struct node_info ni;
1578 recover_inline_xattr(inode, page);
1580 if (!f2fs_has_xattr_block(ofs_of_node(page)))
1583 /* 1: invalidate the previous xattr nid */
1587 /* Deallocate node address */
1588 get_node_info(sbi, prev_xnid, &ni);
1589 f2fs_bug_on(ni.blk_addr == NULL_ADDR);
1590 invalidate_blocks(sbi, ni.blk_addr);
1591 dec_valid_node_count(sbi, inode);
1592 set_node_addr(sbi, &ni, NULL_ADDR, false);
1595 /* 2: allocate new xattr nid */
1596 if (unlikely(!inc_valid_node_count(sbi, inode)))
1599 remove_free_nid(NM_I(sbi), new_xnid);
1600 get_node_info(sbi, new_xnid, &ni);
1601 ni.ino = inode->i_ino;
1602 set_node_addr(sbi, &ni, NEW_ADDR, false);
1603 F2FS_I(inode)->i_xattr_nid = new_xnid;
1605 /* 3: update xattr blkaddr */
1606 refresh_sit_entry(sbi, NEW_ADDR, blkaddr);
1607 set_node_addr(sbi, &ni, blkaddr, false);
1609 update_inode_page(inode);
1613 int recover_inode_page(struct f2fs_sb_info *sbi, struct page *page)
1615 struct f2fs_inode *src, *dst;
1616 nid_t ino = ino_of_node(page);
1617 struct node_info old_ni, new_ni;
1620 get_node_info(sbi, ino, &old_ni);
1622 if (unlikely(old_ni.blk_addr != NULL_ADDR))
1625 ipage = grab_cache_page(NODE_MAPPING(sbi), ino);
1629 /* Should not use this inode from free nid list */
1630 remove_free_nid(NM_I(sbi), ino);
1632 SetPageUptodate(ipage);
1633 fill_node_footer(ipage, ino, ino, 0, true);
1635 src = F2FS_INODE(page);
1636 dst = F2FS_INODE(ipage);
1638 memcpy(dst, src, (unsigned long)&src->i_ext - (unsigned long)src);
1640 dst->i_blocks = cpu_to_le64(1);
1641 dst->i_links = cpu_to_le32(1);
1642 dst->i_xattr_nid = 0;
1647 if (unlikely(!inc_valid_node_count(sbi, NULL)))
1649 set_node_addr(sbi, &new_ni, NEW_ADDR, false);
1650 inc_valid_inode_count(sbi);
1651 f2fs_put_page(ipage, 1);
1656 * ra_sum_pages() merge contiguous pages into one bio and submit.
1657 * these pre-readed pages are linked in pages list.
1659 static int ra_sum_pages(struct f2fs_sb_info *sbi, struct list_head *pages,
1660 int start, int nrpages)
1663 int page_idx = start;
1664 struct f2fs_io_info fio = {
1666 .rw = READ_SYNC | REQ_META | REQ_PRIO
1669 for (; page_idx < start + nrpages; page_idx++) {
1670 /* alloc temporal page for read node summary info*/
1671 page = alloc_page(GFP_F2FS_ZERO);
1676 page->index = page_idx;
1677 list_add_tail(&page->lru, pages);
1680 list_for_each_entry(page, pages, lru)
1681 f2fs_submit_page_mbio(sbi, page, page->index, &fio);
1683 f2fs_submit_merged_bio(sbi, META, READ);
1685 return page_idx - start;
1688 int restore_node_summary(struct f2fs_sb_info *sbi,
1689 unsigned int segno, struct f2fs_summary_block *sum)
1691 struct f2fs_node *rn;
1692 struct f2fs_summary *sum_entry;
1693 struct page *page, *tmp;
1695 int bio_blocks = MAX_BIO_BLOCKS(max_hw_blocks(sbi));
1696 int i, last_offset, nrpages, err = 0;
1697 LIST_HEAD(page_list);
1699 /* scan the node segment */
1700 last_offset = sbi->blocks_per_seg;
1701 addr = START_BLOCK(sbi, segno);
1702 sum_entry = &sum->entries[0];
1704 for (i = 0; !err && i < last_offset; i += nrpages, addr += nrpages) {
1705 nrpages = min(last_offset - i, bio_blocks);
1707 /* read ahead node pages */
1708 nrpages = ra_sum_pages(sbi, &page_list, addr, nrpages);
1712 list_for_each_entry_safe(page, tmp, &page_list, lru) {
1717 if (unlikely(!PageUptodate(page))) {
1720 rn = F2FS_NODE(page);
1721 sum_entry->nid = rn->footer.nid;
1722 sum_entry->version = 0;
1723 sum_entry->ofs_in_node = 0;
1728 list_del(&page->lru);
1729 __free_pages(page, 0);
1735 static bool flush_nats_in_journal(struct f2fs_sb_info *sbi)
1737 struct f2fs_nm_info *nm_i = NM_I(sbi);
1738 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
1739 struct f2fs_summary_block *sum = curseg->sum_blk;
1742 mutex_lock(&curseg->curseg_mutex);
1744 if (nats_in_cursum(sum) < NAT_JOURNAL_ENTRIES) {
1745 mutex_unlock(&curseg->curseg_mutex);
1749 for (i = 0; i < nats_in_cursum(sum); i++) {
1750 struct nat_entry *ne;
1751 struct f2fs_nat_entry raw_ne;
1752 nid_t nid = le32_to_cpu(nid_in_journal(sum, i));
1754 raw_ne = nat_in_journal(sum, i);
1756 write_lock(&nm_i->nat_tree_lock);
1757 ne = __lookup_nat_cache(nm_i, nid);
1759 __set_nat_cache_dirty(nm_i, ne);
1760 write_unlock(&nm_i->nat_tree_lock);
1763 ne = grab_nat_entry(nm_i, nid);
1765 write_unlock(&nm_i->nat_tree_lock);
1768 node_info_from_raw_nat(&ne->ni, &raw_ne);
1769 __set_nat_cache_dirty(nm_i, ne);
1770 write_unlock(&nm_i->nat_tree_lock);
1772 update_nats_in_cursum(sum, -i);
1773 mutex_unlock(&curseg->curseg_mutex);
1778 * This function is called during the checkpointing process.
1780 void flush_nat_entries(struct f2fs_sb_info *sbi)
1782 struct f2fs_nm_info *nm_i = NM_I(sbi);
1783 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
1784 struct f2fs_summary_block *sum = curseg->sum_blk;
1785 struct nat_entry *ne, *cur;
1786 struct page *page = NULL;
1787 struct f2fs_nat_block *nat_blk = NULL;
1788 nid_t start_nid = 0, end_nid = 0;
1791 flushed = flush_nats_in_journal(sbi);
1794 mutex_lock(&curseg->curseg_mutex);
1796 /* 1) flush dirty nat caches */
1797 list_for_each_entry_safe(ne, cur, &nm_i->dirty_nat_entries, list) {
1799 struct f2fs_nat_entry raw_ne;
1802 if (nat_get_blkaddr(ne) == NEW_ADDR)
1805 nid = nat_get_nid(ne);
1810 /* if there is room for nat enries in curseg->sumpage */
1811 offset = lookup_journal_in_cursum(sum, NAT_JOURNAL, nid, 1);
1813 raw_ne = nat_in_journal(sum, offset);
1817 if (!page || (start_nid > nid || nid > end_nid)) {
1819 f2fs_put_page(page, 1);
1822 start_nid = START_NID(nid);
1823 end_nid = start_nid + NAT_ENTRY_PER_BLOCK - 1;
1826 * get nat block with dirty flag, increased reference
1827 * count, mapped and lock
1829 page = get_next_nat_page(sbi, start_nid);
1830 nat_blk = page_address(page);
1833 f2fs_bug_on(!nat_blk);
1834 raw_ne = nat_blk->entries[nid - start_nid];
1836 raw_nat_from_node_info(&raw_ne, &ne->ni);
1839 nat_blk->entries[nid - start_nid] = raw_ne;
1841 nat_in_journal(sum, offset) = raw_ne;
1842 nid_in_journal(sum, offset) = cpu_to_le32(nid);
1845 if (nat_get_blkaddr(ne) == NULL_ADDR &&
1846 add_free_nid(sbi, nid, false) <= 0) {
1847 write_lock(&nm_i->nat_tree_lock);
1848 __del_from_nat_cache(nm_i, ne);
1849 write_unlock(&nm_i->nat_tree_lock);
1851 write_lock(&nm_i->nat_tree_lock);
1852 __clear_nat_cache_dirty(nm_i, ne);
1853 write_unlock(&nm_i->nat_tree_lock);
1857 mutex_unlock(&curseg->curseg_mutex);
1858 f2fs_put_page(page, 1);
1861 static int init_node_manager(struct f2fs_sb_info *sbi)
1863 struct f2fs_super_block *sb_raw = F2FS_RAW_SUPER(sbi);
1864 struct f2fs_nm_info *nm_i = NM_I(sbi);
1865 unsigned char *version_bitmap;
1866 unsigned int nat_segs, nat_blocks;
1868 nm_i->nat_blkaddr = le32_to_cpu(sb_raw->nat_blkaddr);
1870 /* segment_count_nat includes pair segment so divide to 2. */
1871 nat_segs = le32_to_cpu(sb_raw->segment_count_nat) >> 1;
1872 nat_blocks = nat_segs << le32_to_cpu(sb_raw->log_blocks_per_seg);
1874 nm_i->max_nid = NAT_ENTRY_PER_BLOCK * nat_blocks;
1876 /* not used nids: 0, node, meta, (and root counted as valid node) */
1877 nm_i->available_nids = nm_i->max_nid - 3;
1880 nm_i->ram_thresh = DEF_RAM_THRESHOLD;
1882 INIT_RADIX_TREE(&nm_i->free_nid_root, GFP_ATOMIC);
1883 INIT_LIST_HEAD(&nm_i->free_nid_list);
1884 INIT_RADIX_TREE(&nm_i->nat_root, GFP_ATOMIC);
1885 INIT_LIST_HEAD(&nm_i->nat_entries);
1886 INIT_LIST_HEAD(&nm_i->dirty_nat_entries);
1888 mutex_init(&nm_i->build_lock);
1889 spin_lock_init(&nm_i->free_nid_list_lock);
1890 rwlock_init(&nm_i->nat_tree_lock);
1892 nm_i->next_scan_nid = le32_to_cpu(sbi->ckpt->next_free_nid);
1893 nm_i->bitmap_size = __bitmap_size(sbi, NAT_BITMAP);
1894 version_bitmap = __bitmap_ptr(sbi, NAT_BITMAP);
1895 if (!version_bitmap)
1898 nm_i->nat_bitmap = kmemdup(version_bitmap, nm_i->bitmap_size,
1900 if (!nm_i->nat_bitmap)
1905 int build_node_manager(struct f2fs_sb_info *sbi)
1909 sbi->nm_info = kzalloc(sizeof(struct f2fs_nm_info), GFP_KERNEL);
1913 err = init_node_manager(sbi);
1917 build_free_nids(sbi);
1921 void destroy_node_manager(struct f2fs_sb_info *sbi)
1923 struct f2fs_nm_info *nm_i = NM_I(sbi);
1924 struct free_nid *i, *next_i;
1925 struct nat_entry *natvec[NATVEC_SIZE];
1932 /* destroy free nid list */
1933 spin_lock(&nm_i->free_nid_list_lock);
1934 list_for_each_entry_safe(i, next_i, &nm_i->free_nid_list, list) {
1935 f2fs_bug_on(i->state == NID_ALLOC);
1936 __del_from_free_nid_list(nm_i, i);
1938 spin_unlock(&nm_i->free_nid_list_lock);
1939 kmem_cache_free(free_nid_slab, i);
1940 spin_lock(&nm_i->free_nid_list_lock);
1942 f2fs_bug_on(nm_i->fcnt);
1943 spin_unlock(&nm_i->free_nid_list_lock);
1945 /* destroy nat cache */
1946 write_lock(&nm_i->nat_tree_lock);
1947 while ((found = __gang_lookup_nat_cache(nm_i,
1948 nid, NATVEC_SIZE, natvec))) {
1950 nid = nat_get_nid(natvec[found - 1]) + 1;
1951 for (idx = 0; idx < found; idx++)
1952 __del_from_nat_cache(nm_i, natvec[idx]);
1954 f2fs_bug_on(nm_i->nat_cnt);
1955 write_unlock(&nm_i->nat_tree_lock);
1957 kfree(nm_i->nat_bitmap);
1958 sbi->nm_info = NULL;
1962 int __init create_node_manager_caches(void)
1964 nat_entry_slab = f2fs_kmem_cache_create("nat_entry",
1965 sizeof(struct nat_entry));
1966 if (!nat_entry_slab)
1969 free_nid_slab = f2fs_kmem_cache_create("free_nid",
1970 sizeof(struct free_nid));
1971 if (!free_nid_slab) {
1972 kmem_cache_destroy(nat_entry_slab);
1978 void destroy_node_manager_caches(void)
1980 kmem_cache_destroy(free_nid_slab);
1981 kmem_cache_destroy(nat_entry_slab);