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>
23 #include <trace/events/f2fs.h>
25 #define on_build_free_nids(nmi) mutex_is_locked(&nm_i->build_lock)
27 static struct kmem_cache *nat_entry_slab;
28 static struct kmem_cache *free_nid_slab;
29 static struct kmem_cache *nat_entry_set_slab;
31 bool available_free_memory(struct f2fs_sb_info *sbi, int type)
33 struct f2fs_nm_info *nm_i = NM_I(sbi);
35 unsigned long avail_ram;
36 unsigned long mem_size = 0;
41 /* only uses low memory */
42 avail_ram = val.totalram - val.totalhigh;
45 * give 25%, 25%, 50%, 50%, 50% memory for each components respectively
47 if (type == FREE_NIDS) {
48 mem_size = (nm_i->nid_cnt[FREE_NID_LIST] *
49 sizeof(struct free_nid)) >> PAGE_SHIFT;
50 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 2);
51 } else if (type == NAT_ENTRIES) {
52 mem_size = (nm_i->nat_cnt * sizeof(struct nat_entry)) >>
54 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 2);
55 if (excess_cached_nats(sbi))
57 } else if (type == DIRTY_DENTS) {
58 if (sbi->sb->s_bdi->wb.dirty_exceeded)
60 mem_size = get_pages(sbi, F2FS_DIRTY_DENTS);
61 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 1);
62 } else if (type == INO_ENTRIES) {
65 for (i = 0; i <= UPDATE_INO; i++)
66 mem_size += (sbi->im[i].ino_num *
67 sizeof(struct ino_entry)) >> PAGE_SHIFT;
68 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 1);
69 } else if (type == EXTENT_CACHE) {
70 mem_size = (atomic_read(&sbi->total_ext_tree) *
71 sizeof(struct extent_tree) +
72 atomic_read(&sbi->total_ext_node) *
73 sizeof(struct extent_node)) >> PAGE_SHIFT;
74 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 1);
76 if (!sbi->sb->s_bdi->wb.dirty_exceeded)
82 static void clear_node_page_dirty(struct page *page)
84 struct address_space *mapping = page->mapping;
85 unsigned int long flags;
87 if (PageDirty(page)) {
88 spin_lock_irqsave(&mapping->tree_lock, flags);
89 radix_tree_tag_clear(&mapping->page_tree,
92 spin_unlock_irqrestore(&mapping->tree_lock, flags);
94 clear_page_dirty_for_io(page);
95 dec_page_count(F2FS_M_SB(mapping), F2FS_DIRTY_NODES);
97 ClearPageUptodate(page);
100 static struct page *get_current_nat_page(struct f2fs_sb_info *sbi, nid_t nid)
102 pgoff_t index = current_nat_addr(sbi, nid);
103 return get_meta_page(sbi, index);
106 static struct page *get_next_nat_page(struct f2fs_sb_info *sbi, nid_t nid)
108 struct page *src_page;
109 struct page *dst_page;
114 struct f2fs_nm_info *nm_i = NM_I(sbi);
116 src_off = current_nat_addr(sbi, nid);
117 dst_off = next_nat_addr(sbi, src_off);
119 /* get current nat block page with lock */
120 src_page = get_meta_page(sbi, src_off);
121 dst_page = grab_meta_page(sbi, dst_off);
122 f2fs_bug_on(sbi, PageDirty(src_page));
124 src_addr = page_address(src_page);
125 dst_addr = page_address(dst_page);
126 memcpy(dst_addr, src_addr, PAGE_SIZE);
127 set_page_dirty(dst_page);
128 f2fs_put_page(src_page, 1);
130 set_to_next_nat(nm_i, nid);
135 static struct nat_entry *__lookup_nat_cache(struct f2fs_nm_info *nm_i, nid_t n)
137 return radix_tree_lookup(&nm_i->nat_root, n);
140 static unsigned int __gang_lookup_nat_cache(struct f2fs_nm_info *nm_i,
141 nid_t start, unsigned int nr, struct nat_entry **ep)
143 return radix_tree_gang_lookup(&nm_i->nat_root, (void **)ep, start, nr);
146 static void __del_from_nat_cache(struct f2fs_nm_info *nm_i, struct nat_entry *e)
149 radix_tree_delete(&nm_i->nat_root, nat_get_nid(e));
151 kmem_cache_free(nat_entry_slab, e);
154 static void __set_nat_cache_dirty(struct f2fs_nm_info *nm_i,
155 struct nat_entry *ne)
157 nid_t set = NAT_BLOCK_OFFSET(ne->ni.nid);
158 struct nat_entry_set *head;
160 if (get_nat_flag(ne, IS_DIRTY))
163 head = radix_tree_lookup(&nm_i->nat_set_root, set);
165 head = f2fs_kmem_cache_alloc(nat_entry_set_slab, GFP_NOFS);
167 INIT_LIST_HEAD(&head->entry_list);
168 INIT_LIST_HEAD(&head->set_list);
171 f2fs_radix_tree_insert(&nm_i->nat_set_root, set, head);
173 list_move_tail(&ne->list, &head->entry_list);
174 nm_i->dirty_nat_cnt++;
176 set_nat_flag(ne, IS_DIRTY, true);
179 static void __clear_nat_cache_dirty(struct f2fs_nm_info *nm_i,
180 struct nat_entry *ne)
182 nid_t set = NAT_BLOCK_OFFSET(ne->ni.nid);
183 struct nat_entry_set *head;
185 head = radix_tree_lookup(&nm_i->nat_set_root, set);
187 list_move_tail(&ne->list, &nm_i->nat_entries);
188 set_nat_flag(ne, IS_DIRTY, false);
190 nm_i->dirty_nat_cnt--;
194 static unsigned int __gang_lookup_nat_set(struct f2fs_nm_info *nm_i,
195 nid_t start, unsigned int nr, struct nat_entry_set **ep)
197 return radix_tree_gang_lookup(&nm_i->nat_set_root, (void **)ep,
201 int need_dentry_mark(struct f2fs_sb_info *sbi, nid_t nid)
203 struct f2fs_nm_info *nm_i = NM_I(sbi);
207 down_read(&nm_i->nat_tree_lock);
208 e = __lookup_nat_cache(nm_i, nid);
210 if (!get_nat_flag(e, IS_CHECKPOINTED) &&
211 !get_nat_flag(e, HAS_FSYNCED_INODE))
214 up_read(&nm_i->nat_tree_lock);
218 bool is_checkpointed_node(struct f2fs_sb_info *sbi, nid_t nid)
220 struct f2fs_nm_info *nm_i = NM_I(sbi);
224 down_read(&nm_i->nat_tree_lock);
225 e = __lookup_nat_cache(nm_i, nid);
226 if (e && !get_nat_flag(e, IS_CHECKPOINTED))
228 up_read(&nm_i->nat_tree_lock);
232 bool need_inode_block_update(struct f2fs_sb_info *sbi, nid_t ino)
234 struct f2fs_nm_info *nm_i = NM_I(sbi);
236 bool need_update = true;
238 down_read(&nm_i->nat_tree_lock);
239 e = __lookup_nat_cache(nm_i, ino);
240 if (e && get_nat_flag(e, HAS_LAST_FSYNC) &&
241 (get_nat_flag(e, IS_CHECKPOINTED) ||
242 get_nat_flag(e, HAS_FSYNCED_INODE)))
244 up_read(&nm_i->nat_tree_lock);
248 static struct nat_entry *grab_nat_entry(struct f2fs_nm_info *nm_i, nid_t nid,
251 struct nat_entry *new;
254 new = f2fs_kmem_cache_alloc(nat_entry_slab, GFP_NOFS);
255 f2fs_radix_tree_insert(&nm_i->nat_root, nid, new);
257 new = kmem_cache_alloc(nat_entry_slab, GFP_NOFS);
260 if (radix_tree_insert(&nm_i->nat_root, nid, new)) {
261 kmem_cache_free(nat_entry_slab, new);
266 memset(new, 0, sizeof(struct nat_entry));
267 nat_set_nid(new, nid);
269 list_add_tail(&new->list, &nm_i->nat_entries);
274 static void cache_nat_entry(struct f2fs_sb_info *sbi, nid_t nid,
275 struct f2fs_nat_entry *ne)
277 struct f2fs_nm_info *nm_i = NM_I(sbi);
280 e = __lookup_nat_cache(nm_i, nid);
282 e = grab_nat_entry(nm_i, nid, false);
284 node_info_from_raw_nat(&e->ni, ne);
286 f2fs_bug_on(sbi, nat_get_ino(e) != le32_to_cpu(ne->ino) ||
287 nat_get_blkaddr(e) !=
288 le32_to_cpu(ne->block_addr) ||
289 nat_get_version(e) != ne->version);
293 static void set_node_addr(struct f2fs_sb_info *sbi, struct node_info *ni,
294 block_t new_blkaddr, bool fsync_done)
296 struct f2fs_nm_info *nm_i = NM_I(sbi);
299 down_write(&nm_i->nat_tree_lock);
300 e = __lookup_nat_cache(nm_i, ni->nid);
302 e = grab_nat_entry(nm_i, ni->nid, true);
303 copy_node_info(&e->ni, ni);
304 f2fs_bug_on(sbi, ni->blk_addr == NEW_ADDR);
305 } else if (new_blkaddr == NEW_ADDR) {
307 * when nid is reallocated,
308 * previous nat entry can be remained in nat cache.
309 * So, reinitialize it with new information.
311 copy_node_info(&e->ni, ni);
312 f2fs_bug_on(sbi, ni->blk_addr != NULL_ADDR);
316 f2fs_bug_on(sbi, nat_get_blkaddr(e) != ni->blk_addr);
317 f2fs_bug_on(sbi, nat_get_blkaddr(e) == NULL_ADDR &&
318 new_blkaddr == NULL_ADDR);
319 f2fs_bug_on(sbi, nat_get_blkaddr(e) == NEW_ADDR &&
320 new_blkaddr == NEW_ADDR);
321 f2fs_bug_on(sbi, nat_get_blkaddr(e) != NEW_ADDR &&
322 nat_get_blkaddr(e) != NULL_ADDR &&
323 new_blkaddr == NEW_ADDR);
325 /* increment version no as node is removed */
326 if (nat_get_blkaddr(e) != NEW_ADDR && new_blkaddr == NULL_ADDR) {
327 unsigned char version = nat_get_version(e);
328 nat_set_version(e, inc_node_version(version));
330 /* in order to reuse the nid */
331 if (nm_i->next_scan_nid > ni->nid)
332 nm_i->next_scan_nid = ni->nid;
336 nat_set_blkaddr(e, new_blkaddr);
337 if (new_blkaddr == NEW_ADDR || new_blkaddr == NULL_ADDR)
338 set_nat_flag(e, IS_CHECKPOINTED, false);
339 __set_nat_cache_dirty(nm_i, e);
341 /* update fsync_mark if its inode nat entry is still alive */
342 if (ni->nid != ni->ino)
343 e = __lookup_nat_cache(nm_i, ni->ino);
345 if (fsync_done && ni->nid == ni->ino)
346 set_nat_flag(e, HAS_FSYNCED_INODE, true);
347 set_nat_flag(e, HAS_LAST_FSYNC, fsync_done);
349 up_write(&nm_i->nat_tree_lock);
352 int try_to_free_nats(struct f2fs_sb_info *sbi, int nr_shrink)
354 struct f2fs_nm_info *nm_i = NM_I(sbi);
357 if (!down_write_trylock(&nm_i->nat_tree_lock))
360 while (nr_shrink && !list_empty(&nm_i->nat_entries)) {
361 struct nat_entry *ne;
362 ne = list_first_entry(&nm_i->nat_entries,
363 struct nat_entry, list);
364 __del_from_nat_cache(nm_i, ne);
367 up_write(&nm_i->nat_tree_lock);
368 return nr - nr_shrink;
372 * This function always returns success
374 void get_node_info(struct f2fs_sb_info *sbi, nid_t nid, struct node_info *ni)
376 struct f2fs_nm_info *nm_i = NM_I(sbi);
377 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
378 struct f2fs_journal *journal = curseg->journal;
379 nid_t start_nid = START_NID(nid);
380 struct f2fs_nat_block *nat_blk;
381 struct page *page = NULL;
382 struct f2fs_nat_entry ne;
388 /* Check nat cache */
389 down_read(&nm_i->nat_tree_lock);
390 e = __lookup_nat_cache(nm_i, nid);
392 ni->ino = nat_get_ino(e);
393 ni->blk_addr = nat_get_blkaddr(e);
394 ni->version = nat_get_version(e);
395 up_read(&nm_i->nat_tree_lock);
399 memset(&ne, 0, sizeof(struct f2fs_nat_entry));
401 /* Check current segment summary */
402 down_read(&curseg->journal_rwsem);
403 i = lookup_journal_in_cursum(journal, NAT_JOURNAL, nid, 0);
405 ne = nat_in_journal(journal, i);
406 node_info_from_raw_nat(ni, &ne);
408 up_read(&curseg->journal_rwsem);
412 /* Fill node_info from nat page */
413 page = get_current_nat_page(sbi, start_nid);
414 nat_blk = (struct f2fs_nat_block *)page_address(page);
415 ne = nat_blk->entries[nid - start_nid];
416 node_info_from_raw_nat(ni, &ne);
417 f2fs_put_page(page, 1);
419 up_read(&nm_i->nat_tree_lock);
420 /* cache nat entry */
421 down_write(&nm_i->nat_tree_lock);
422 cache_nat_entry(sbi, nid, &ne);
423 up_write(&nm_i->nat_tree_lock);
427 * readahead MAX_RA_NODE number of node pages.
429 static void ra_node_pages(struct page *parent, int start, int n)
431 struct f2fs_sb_info *sbi = F2FS_P_SB(parent);
432 struct blk_plug plug;
436 blk_start_plug(&plug);
438 /* Then, try readahead for siblings of the desired node */
440 end = min(end, NIDS_PER_BLOCK);
441 for (i = start; i < end; i++) {
442 nid = get_nid(parent, i, false);
443 ra_node_page(sbi, nid);
446 blk_finish_plug(&plug);
449 pgoff_t get_next_page_offset(struct dnode_of_data *dn, pgoff_t pgofs)
451 const long direct_index = ADDRS_PER_INODE(dn->inode);
452 const long direct_blks = ADDRS_PER_BLOCK;
453 const long indirect_blks = ADDRS_PER_BLOCK * NIDS_PER_BLOCK;
454 unsigned int skipped_unit = ADDRS_PER_BLOCK;
455 int cur_level = dn->cur_level;
456 int max_level = dn->max_level;
462 while (max_level-- > cur_level)
463 skipped_unit *= NIDS_PER_BLOCK;
465 switch (dn->max_level) {
467 base += 2 * indirect_blks;
469 base += 2 * direct_blks;
471 base += direct_index;
474 f2fs_bug_on(F2FS_I_SB(dn->inode), 1);
477 return ((pgofs - base) / skipped_unit + 1) * skipped_unit + base;
481 * The maximum depth is four.
482 * Offset[0] will have raw inode offset.
484 static int get_node_path(struct inode *inode, long block,
485 int offset[4], unsigned int noffset[4])
487 const long direct_index = ADDRS_PER_INODE(inode);
488 const long direct_blks = ADDRS_PER_BLOCK;
489 const long dptrs_per_blk = NIDS_PER_BLOCK;
490 const long indirect_blks = ADDRS_PER_BLOCK * NIDS_PER_BLOCK;
491 const long dindirect_blks = indirect_blks * NIDS_PER_BLOCK;
497 if (block < direct_index) {
501 block -= direct_index;
502 if (block < direct_blks) {
503 offset[n++] = NODE_DIR1_BLOCK;
509 block -= direct_blks;
510 if (block < direct_blks) {
511 offset[n++] = NODE_DIR2_BLOCK;
517 block -= direct_blks;
518 if (block < indirect_blks) {
519 offset[n++] = NODE_IND1_BLOCK;
521 offset[n++] = block / direct_blks;
522 noffset[n] = 4 + offset[n - 1];
523 offset[n] = block % direct_blks;
527 block -= indirect_blks;
528 if (block < indirect_blks) {
529 offset[n++] = NODE_IND2_BLOCK;
530 noffset[n] = 4 + dptrs_per_blk;
531 offset[n++] = block / direct_blks;
532 noffset[n] = 5 + dptrs_per_blk + offset[n - 1];
533 offset[n] = block % direct_blks;
537 block -= indirect_blks;
538 if (block < dindirect_blks) {
539 offset[n++] = NODE_DIND_BLOCK;
540 noffset[n] = 5 + (dptrs_per_blk * 2);
541 offset[n++] = block / indirect_blks;
542 noffset[n] = 6 + (dptrs_per_blk * 2) +
543 offset[n - 1] * (dptrs_per_blk + 1);
544 offset[n++] = (block / direct_blks) % dptrs_per_blk;
545 noffset[n] = 7 + (dptrs_per_blk * 2) +
546 offset[n - 2] * (dptrs_per_blk + 1) +
548 offset[n] = block % direct_blks;
559 * Caller should call f2fs_put_dnode(dn).
560 * Also, it should grab and release a rwsem by calling f2fs_lock_op() and
561 * f2fs_unlock_op() only if ro is not set RDONLY_NODE.
562 * In the case of RDONLY_NODE, we don't need to care about mutex.
564 int get_dnode_of_data(struct dnode_of_data *dn, pgoff_t index, int mode)
566 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
567 struct page *npage[4];
568 struct page *parent = NULL;
570 unsigned int noffset[4];
575 level = get_node_path(dn->inode, index, offset, noffset);
577 nids[0] = dn->inode->i_ino;
578 npage[0] = dn->inode_page;
581 npage[0] = get_node_page(sbi, nids[0]);
582 if (IS_ERR(npage[0]))
583 return PTR_ERR(npage[0]);
586 /* if inline_data is set, should not report any block indices */
587 if (f2fs_has_inline_data(dn->inode) && index) {
589 f2fs_put_page(npage[0], 1);
595 nids[1] = get_nid(parent, offset[0], true);
596 dn->inode_page = npage[0];
597 dn->inode_page_locked = true;
599 /* get indirect or direct nodes */
600 for (i = 1; i <= level; i++) {
603 if (!nids[i] && mode == ALLOC_NODE) {
605 if (!alloc_nid(sbi, &(nids[i]))) {
611 npage[i] = new_node_page(dn, noffset[i], NULL);
612 if (IS_ERR(npage[i])) {
613 alloc_nid_failed(sbi, nids[i]);
614 err = PTR_ERR(npage[i]);
618 set_nid(parent, offset[i - 1], nids[i], i == 1);
619 alloc_nid_done(sbi, nids[i]);
621 } else if (mode == LOOKUP_NODE_RA && i == level && level > 1) {
622 npage[i] = get_node_page_ra(parent, offset[i - 1]);
623 if (IS_ERR(npage[i])) {
624 err = PTR_ERR(npage[i]);
630 dn->inode_page_locked = false;
633 f2fs_put_page(parent, 1);
637 npage[i] = get_node_page(sbi, nids[i]);
638 if (IS_ERR(npage[i])) {
639 err = PTR_ERR(npage[i]);
640 f2fs_put_page(npage[0], 0);
646 nids[i + 1] = get_nid(parent, offset[i], false);
649 dn->nid = nids[level];
650 dn->ofs_in_node = offset[level];
651 dn->node_page = npage[level];
652 dn->data_blkaddr = datablock_addr(dn->node_page, dn->ofs_in_node);
656 f2fs_put_page(parent, 1);
658 f2fs_put_page(npage[0], 0);
660 dn->inode_page = NULL;
661 dn->node_page = NULL;
662 if (err == -ENOENT) {
664 dn->max_level = level;
665 dn->ofs_in_node = offset[level];
670 static void truncate_node(struct dnode_of_data *dn)
672 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
675 get_node_info(sbi, dn->nid, &ni);
676 if (dn->inode->i_blocks == 0) {
677 f2fs_bug_on(sbi, ni.blk_addr != NULL_ADDR);
680 f2fs_bug_on(sbi, ni.blk_addr == NULL_ADDR);
682 /* Deallocate node address */
683 invalidate_blocks(sbi, ni.blk_addr);
684 dec_valid_node_count(sbi, dn->inode);
685 set_node_addr(sbi, &ni, NULL_ADDR, false);
687 if (dn->nid == dn->inode->i_ino) {
688 remove_orphan_inode(sbi, dn->nid);
689 dec_valid_inode_count(sbi);
690 f2fs_inode_synced(dn->inode);
693 clear_node_page_dirty(dn->node_page);
694 set_sbi_flag(sbi, SBI_IS_DIRTY);
696 f2fs_put_page(dn->node_page, 1);
698 invalidate_mapping_pages(NODE_MAPPING(sbi),
699 dn->node_page->index, dn->node_page->index);
701 dn->node_page = NULL;
702 trace_f2fs_truncate_node(dn->inode, dn->nid, ni.blk_addr);
705 static int truncate_dnode(struct dnode_of_data *dn)
712 /* get direct node */
713 page = get_node_page(F2FS_I_SB(dn->inode), dn->nid);
714 if (IS_ERR(page) && PTR_ERR(page) == -ENOENT)
716 else if (IS_ERR(page))
717 return PTR_ERR(page);
719 /* Make dnode_of_data for parameter */
720 dn->node_page = page;
722 truncate_data_blocks(dn);
727 static int truncate_nodes(struct dnode_of_data *dn, unsigned int nofs,
730 struct dnode_of_data rdn = *dn;
732 struct f2fs_node *rn;
734 unsigned int child_nofs;
739 return NIDS_PER_BLOCK + 1;
741 trace_f2fs_truncate_nodes_enter(dn->inode, dn->nid, dn->data_blkaddr);
743 page = get_node_page(F2FS_I_SB(dn->inode), dn->nid);
745 trace_f2fs_truncate_nodes_exit(dn->inode, PTR_ERR(page));
746 return PTR_ERR(page);
749 ra_node_pages(page, ofs, NIDS_PER_BLOCK);
751 rn = F2FS_NODE(page);
753 for (i = ofs; i < NIDS_PER_BLOCK; i++, freed++) {
754 child_nid = le32_to_cpu(rn->in.nid[i]);
758 ret = truncate_dnode(&rdn);
761 if (set_nid(page, i, 0, false))
762 dn->node_changed = true;
765 child_nofs = nofs + ofs * (NIDS_PER_BLOCK + 1) + 1;
766 for (i = ofs; i < NIDS_PER_BLOCK; i++) {
767 child_nid = le32_to_cpu(rn->in.nid[i]);
768 if (child_nid == 0) {
769 child_nofs += NIDS_PER_BLOCK + 1;
773 ret = truncate_nodes(&rdn, child_nofs, 0, depth - 1);
774 if (ret == (NIDS_PER_BLOCK + 1)) {
775 if (set_nid(page, i, 0, false))
776 dn->node_changed = true;
778 } else if (ret < 0 && ret != -ENOENT) {
786 /* remove current indirect node */
787 dn->node_page = page;
791 f2fs_put_page(page, 1);
793 trace_f2fs_truncate_nodes_exit(dn->inode, freed);
797 f2fs_put_page(page, 1);
798 trace_f2fs_truncate_nodes_exit(dn->inode, ret);
802 static int truncate_partial_nodes(struct dnode_of_data *dn,
803 struct f2fs_inode *ri, int *offset, int depth)
805 struct page *pages[2];
812 nid[0] = le32_to_cpu(ri->i_nid[offset[0] - NODE_DIR1_BLOCK]);
816 /* get indirect nodes in the path */
817 for (i = 0; i < idx + 1; i++) {
818 /* reference count'll be increased */
819 pages[i] = get_node_page(F2FS_I_SB(dn->inode), nid[i]);
820 if (IS_ERR(pages[i])) {
821 err = PTR_ERR(pages[i]);
825 nid[i + 1] = get_nid(pages[i], offset[i + 1], false);
828 ra_node_pages(pages[idx], offset[idx + 1], NIDS_PER_BLOCK);
830 /* free direct nodes linked to a partial indirect node */
831 for (i = offset[idx + 1]; i < NIDS_PER_BLOCK; i++) {
832 child_nid = get_nid(pages[idx], i, false);
836 err = truncate_dnode(dn);
839 if (set_nid(pages[idx], i, 0, false))
840 dn->node_changed = true;
843 if (offset[idx + 1] == 0) {
844 dn->node_page = pages[idx];
848 f2fs_put_page(pages[idx], 1);
854 for (i = idx; i >= 0; i--)
855 f2fs_put_page(pages[i], 1);
857 trace_f2fs_truncate_partial_nodes(dn->inode, nid, depth, err);
863 * All the block addresses of data and nodes should be nullified.
865 int truncate_inode_blocks(struct inode *inode, pgoff_t from)
867 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
868 int err = 0, cont = 1;
869 int level, offset[4], noffset[4];
870 unsigned int nofs = 0;
871 struct f2fs_inode *ri;
872 struct dnode_of_data dn;
875 trace_f2fs_truncate_inode_blocks_enter(inode, from);
877 level = get_node_path(inode, from, offset, noffset);
879 page = get_node_page(sbi, inode->i_ino);
881 trace_f2fs_truncate_inode_blocks_exit(inode, PTR_ERR(page));
882 return PTR_ERR(page);
885 set_new_dnode(&dn, inode, page, NULL, 0);
888 ri = F2FS_INODE(page);
896 if (!offset[level - 1])
898 err = truncate_partial_nodes(&dn, ri, offset, level);
899 if (err < 0 && err != -ENOENT)
901 nofs += 1 + NIDS_PER_BLOCK;
904 nofs = 5 + 2 * NIDS_PER_BLOCK;
905 if (!offset[level - 1])
907 err = truncate_partial_nodes(&dn, ri, offset, level);
908 if (err < 0 && err != -ENOENT)
917 dn.nid = le32_to_cpu(ri->i_nid[offset[0] - NODE_DIR1_BLOCK]);
919 case NODE_DIR1_BLOCK:
920 case NODE_DIR2_BLOCK:
921 err = truncate_dnode(&dn);
924 case NODE_IND1_BLOCK:
925 case NODE_IND2_BLOCK:
926 err = truncate_nodes(&dn, nofs, offset[1], 2);
929 case NODE_DIND_BLOCK:
930 err = truncate_nodes(&dn, nofs, offset[1], 3);
937 if (err < 0 && err != -ENOENT)
939 if (offset[1] == 0 &&
940 ri->i_nid[offset[0] - NODE_DIR1_BLOCK]) {
942 BUG_ON(page->mapping != NODE_MAPPING(sbi));
943 f2fs_wait_on_page_writeback(page, NODE, true);
944 ri->i_nid[offset[0] - NODE_DIR1_BLOCK] = 0;
945 set_page_dirty(page);
953 f2fs_put_page(page, 0);
954 trace_f2fs_truncate_inode_blocks_exit(inode, err);
955 return err > 0 ? 0 : err;
958 int truncate_xattr_node(struct inode *inode, struct page *page)
960 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
961 nid_t nid = F2FS_I(inode)->i_xattr_nid;
962 struct dnode_of_data dn;
968 npage = get_node_page(sbi, nid);
970 return PTR_ERR(npage);
972 f2fs_i_xnid_write(inode, 0);
974 /* need to do checkpoint during fsync */
975 F2FS_I(inode)->xattr_ver = cur_cp_version(F2FS_CKPT(sbi));
977 set_new_dnode(&dn, inode, page, npage, nid);
980 dn.inode_page_locked = true;
986 * Caller should grab and release a rwsem by calling f2fs_lock_op() and
989 int remove_inode_page(struct inode *inode)
991 struct dnode_of_data dn;
994 set_new_dnode(&dn, inode, NULL, NULL, inode->i_ino);
995 err = get_dnode_of_data(&dn, 0, LOOKUP_NODE);
999 err = truncate_xattr_node(inode, dn.inode_page);
1001 f2fs_put_dnode(&dn);
1005 /* remove potential inline_data blocks */
1006 if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
1007 S_ISLNK(inode->i_mode))
1008 truncate_data_blocks_range(&dn, 1);
1010 /* 0 is possible, after f2fs_new_inode() has failed */
1011 f2fs_bug_on(F2FS_I_SB(inode),
1012 inode->i_blocks != 0 && inode->i_blocks != 1);
1014 /* will put inode & node pages */
1019 struct page *new_inode_page(struct inode *inode)
1021 struct dnode_of_data dn;
1023 /* allocate inode page for new inode */
1024 set_new_dnode(&dn, inode, NULL, NULL, inode->i_ino);
1026 /* caller should f2fs_put_page(page, 1); */
1027 return new_node_page(&dn, 0, NULL);
1030 struct page *new_node_page(struct dnode_of_data *dn,
1031 unsigned int ofs, struct page *ipage)
1033 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
1034 struct node_info old_ni, new_ni;
1038 if (unlikely(is_inode_flag_set(dn->inode, FI_NO_ALLOC)))
1039 return ERR_PTR(-EPERM);
1041 page = f2fs_grab_cache_page(NODE_MAPPING(sbi), dn->nid, false);
1043 return ERR_PTR(-ENOMEM);
1045 if (unlikely(!inc_valid_node_count(sbi, dn->inode))) {
1050 get_node_info(sbi, dn->nid, &old_ni);
1052 /* Reinitialize old_ni with new node page */
1053 f2fs_bug_on(sbi, old_ni.blk_addr != NULL_ADDR);
1055 new_ni.ino = dn->inode->i_ino;
1056 set_node_addr(sbi, &new_ni, NEW_ADDR, false);
1058 f2fs_wait_on_page_writeback(page, NODE, true);
1059 fill_node_footer(page, dn->nid, dn->inode->i_ino, ofs, true);
1060 set_cold_node(dn->inode, page);
1061 if (!PageUptodate(page))
1062 SetPageUptodate(page);
1063 if (set_page_dirty(page))
1064 dn->node_changed = true;
1066 if (f2fs_has_xattr_block(ofs))
1067 f2fs_i_xnid_write(dn->inode, dn->nid);
1070 inc_valid_inode_count(sbi);
1074 clear_node_page_dirty(page);
1075 f2fs_put_page(page, 1);
1076 return ERR_PTR(err);
1080 * Caller should do after getting the following values.
1081 * 0: f2fs_put_page(page, 0)
1082 * LOCKED_PAGE or error: f2fs_put_page(page, 1)
1084 static int read_node_page(struct page *page, int op_flags)
1086 struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1087 struct node_info ni;
1088 struct f2fs_io_info fio = {
1092 .op_flags = op_flags,
1094 .encrypted_page = NULL,
1097 if (PageUptodate(page))
1100 get_node_info(sbi, page->index, &ni);
1102 if (unlikely(ni.blk_addr == NULL_ADDR)) {
1103 ClearPageUptodate(page);
1107 fio.new_blkaddr = fio.old_blkaddr = ni.blk_addr;
1108 return f2fs_submit_page_bio(&fio);
1112 * Readahead a node page
1114 void ra_node_page(struct f2fs_sb_info *sbi, nid_t nid)
1121 f2fs_bug_on(sbi, check_nid_range(sbi, nid));
1124 apage = radix_tree_lookup(&NODE_MAPPING(sbi)->page_tree, nid);
1129 apage = f2fs_grab_cache_page(NODE_MAPPING(sbi), nid, false);
1133 err = read_node_page(apage, REQ_RAHEAD);
1134 f2fs_put_page(apage, err ? 1 : 0);
1137 static struct page *__get_node_page(struct f2fs_sb_info *sbi, pgoff_t nid,
1138 struct page *parent, int start)
1144 return ERR_PTR(-ENOENT);
1145 f2fs_bug_on(sbi, check_nid_range(sbi, nid));
1147 page = f2fs_grab_cache_page(NODE_MAPPING(sbi), nid, false);
1149 return ERR_PTR(-ENOMEM);
1151 err = read_node_page(page, 0);
1153 f2fs_put_page(page, 1);
1154 return ERR_PTR(err);
1155 } else if (err == LOCKED_PAGE) {
1160 ra_node_pages(parent, start + 1, MAX_RA_NODE);
1164 if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1165 f2fs_put_page(page, 1);
1169 if (unlikely(!PageUptodate(page)))
1172 if(unlikely(nid != nid_of_node(page))) {
1173 f2fs_bug_on(sbi, 1);
1174 ClearPageUptodate(page);
1176 f2fs_put_page(page, 1);
1177 return ERR_PTR(-EIO);
1182 struct page *get_node_page(struct f2fs_sb_info *sbi, pgoff_t nid)
1184 return __get_node_page(sbi, nid, NULL, 0);
1187 struct page *get_node_page_ra(struct page *parent, int start)
1189 struct f2fs_sb_info *sbi = F2FS_P_SB(parent);
1190 nid_t nid = get_nid(parent, start, false);
1192 return __get_node_page(sbi, nid, parent, start);
1195 static void flush_inline_data(struct f2fs_sb_info *sbi, nid_t ino)
1197 struct inode *inode;
1201 /* should flush inline_data before evict_inode */
1202 inode = ilookup(sbi->sb, ino);
1206 page = pagecache_get_page(inode->i_mapping, 0, FGP_LOCK|FGP_NOWAIT, 0);
1210 if (!PageUptodate(page))
1213 if (!PageDirty(page))
1216 if (!clear_page_dirty_for_io(page))
1219 ret = f2fs_write_inline_data(inode, page);
1220 inode_dec_dirty_pages(inode);
1221 remove_dirty_inode(inode);
1223 set_page_dirty(page);
1225 f2fs_put_page(page, 1);
1230 void move_node_page(struct page *node_page, int gc_type)
1232 if (gc_type == FG_GC) {
1233 struct f2fs_sb_info *sbi = F2FS_P_SB(node_page);
1234 struct writeback_control wbc = {
1235 .sync_mode = WB_SYNC_ALL,
1240 set_page_dirty(node_page);
1241 f2fs_wait_on_page_writeback(node_page, NODE, true);
1243 f2fs_bug_on(sbi, PageWriteback(node_page));
1244 if (!clear_page_dirty_for_io(node_page))
1247 if (NODE_MAPPING(sbi)->a_ops->writepage(node_page, &wbc))
1248 unlock_page(node_page);
1251 /* set page dirty and write it */
1252 if (!PageWriteback(node_page))
1253 set_page_dirty(node_page);
1256 unlock_page(node_page);
1258 f2fs_put_page(node_page, 0);
1261 static struct page *last_fsync_dnode(struct f2fs_sb_info *sbi, nid_t ino)
1264 struct pagevec pvec;
1265 struct page *last_page = NULL;
1267 pagevec_init(&pvec, 0);
1271 while (index <= end) {
1273 nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1274 PAGECACHE_TAG_DIRTY,
1275 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
1279 for (i = 0; i < nr_pages; i++) {
1280 struct page *page = pvec.pages[i];
1282 if (unlikely(f2fs_cp_error(sbi))) {
1283 f2fs_put_page(last_page, 0);
1284 pagevec_release(&pvec);
1285 return ERR_PTR(-EIO);
1288 if (!IS_DNODE(page) || !is_cold_node(page))
1290 if (ino_of_node(page) != ino)
1295 if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1300 if (ino_of_node(page) != ino)
1301 goto continue_unlock;
1303 if (!PageDirty(page)) {
1304 /* someone wrote it for us */
1305 goto continue_unlock;
1309 f2fs_put_page(last_page, 0);
1315 pagevec_release(&pvec);
1321 static int __write_node_page(struct page *page, bool atomic,
1322 struct writeback_control *wbc)
1324 struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1326 struct node_info ni;
1327 struct f2fs_io_info fio = {
1331 .op_flags = wbc_to_write_flags(wbc),
1333 .encrypted_page = NULL,
1336 trace_f2fs_writepage(page, NODE);
1338 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
1340 if (unlikely(f2fs_cp_error(sbi)))
1343 /* get old block addr of this node page */
1344 nid = nid_of_node(page);
1345 f2fs_bug_on(sbi, page->index != nid);
1347 if (wbc->for_reclaim) {
1348 if (!down_read_trylock(&sbi->node_write))
1351 down_read(&sbi->node_write);
1354 get_node_info(sbi, nid, &ni);
1356 /* This page is already truncated */
1357 if (unlikely(ni.blk_addr == NULL_ADDR)) {
1358 ClearPageUptodate(page);
1359 dec_page_count(sbi, F2FS_DIRTY_NODES);
1360 up_read(&sbi->node_write);
1365 if (atomic && !test_opt(sbi, NOBARRIER))
1366 fio.op_flags |= REQ_PREFLUSH | REQ_FUA;
1368 set_page_writeback(page);
1369 fio.old_blkaddr = ni.blk_addr;
1370 write_node_page(nid, &fio);
1371 set_node_addr(sbi, &ni, fio.new_blkaddr, is_fsync_dnode(page));
1372 dec_page_count(sbi, F2FS_DIRTY_NODES);
1373 up_read(&sbi->node_write);
1375 if (wbc->for_reclaim)
1376 f2fs_submit_merged_bio_cond(sbi, NULL, page, 0, NODE, WRITE);
1380 if (unlikely(f2fs_cp_error(sbi)))
1381 f2fs_submit_merged_bio(sbi, NODE, WRITE);
1386 redirty_page_for_writepage(wbc, page);
1387 return AOP_WRITEPAGE_ACTIVATE;
1390 static int f2fs_write_node_page(struct page *page,
1391 struct writeback_control *wbc)
1393 return __write_node_page(page, false, wbc);
1396 int fsync_node_pages(struct f2fs_sb_info *sbi, struct inode *inode,
1397 struct writeback_control *wbc, bool atomic)
1400 struct pagevec pvec;
1402 struct page *last_page = NULL;
1403 bool marked = false;
1404 nid_t ino = inode->i_ino;
1408 last_page = last_fsync_dnode(sbi, ino);
1409 if (IS_ERR_OR_NULL(last_page))
1410 return PTR_ERR_OR_ZERO(last_page);
1413 pagevec_init(&pvec, 0);
1417 while (index <= end) {
1419 nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1420 PAGECACHE_TAG_DIRTY,
1421 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
1425 for (i = 0; i < nr_pages; i++) {
1426 struct page *page = pvec.pages[i];
1428 if (unlikely(f2fs_cp_error(sbi))) {
1429 f2fs_put_page(last_page, 0);
1430 pagevec_release(&pvec);
1435 if (!IS_DNODE(page) || !is_cold_node(page))
1437 if (ino_of_node(page) != ino)
1442 if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1447 if (ino_of_node(page) != ino)
1448 goto continue_unlock;
1450 if (!PageDirty(page) && page != last_page) {
1451 /* someone wrote it for us */
1452 goto continue_unlock;
1455 f2fs_wait_on_page_writeback(page, NODE, true);
1456 BUG_ON(PageWriteback(page));
1458 if (!atomic || page == last_page) {
1459 set_fsync_mark(page, 1);
1460 if (IS_INODE(page)) {
1461 if (is_inode_flag_set(inode,
1463 update_inode(inode, page);
1464 set_dentry_mark(page,
1465 need_dentry_mark(sbi, ino));
1467 /* may be written by other thread */
1468 if (!PageDirty(page))
1469 set_page_dirty(page);
1472 if (!clear_page_dirty_for_io(page))
1473 goto continue_unlock;
1475 ret = __write_node_page(page, atomic &&
1476 page == last_page, wbc);
1479 f2fs_put_page(last_page, 0);
1485 if (page == last_page) {
1486 f2fs_put_page(page, 0);
1491 pagevec_release(&pvec);
1497 if (!ret && atomic && !marked) {
1498 f2fs_msg(sbi->sb, KERN_DEBUG,
1499 "Retry to write fsync mark: ino=%u, idx=%lx",
1500 ino, last_page->index);
1501 lock_page(last_page);
1502 f2fs_wait_on_page_writeback(last_page, NODE, true);
1503 set_page_dirty(last_page);
1504 unlock_page(last_page);
1509 f2fs_submit_merged_bio_cond(sbi, NULL, NULL, ino, NODE, WRITE);
1510 return ret ? -EIO: 0;
1513 int sync_node_pages(struct f2fs_sb_info *sbi, struct writeback_control *wbc)
1516 struct pagevec pvec;
1521 pagevec_init(&pvec, 0);
1527 while (index <= end) {
1529 nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1530 PAGECACHE_TAG_DIRTY,
1531 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
1535 for (i = 0; i < nr_pages; i++) {
1536 struct page *page = pvec.pages[i];
1538 if (unlikely(f2fs_cp_error(sbi))) {
1539 pagevec_release(&pvec);
1545 * flushing sequence with step:
1550 if (step == 0 && IS_DNODE(page))
1552 if (step == 1 && (!IS_DNODE(page) ||
1553 is_cold_node(page)))
1555 if (step == 2 && (!IS_DNODE(page) ||
1556 !is_cold_node(page)))
1559 if (!trylock_page(page))
1562 if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1568 if (!PageDirty(page)) {
1569 /* someone wrote it for us */
1570 goto continue_unlock;
1573 /* flush inline_data */
1574 if (is_inline_node(page)) {
1575 clear_inline_node(page);
1577 flush_inline_data(sbi, ino_of_node(page));
1581 f2fs_wait_on_page_writeback(page, NODE, true);
1583 BUG_ON(PageWriteback(page));
1584 if (!clear_page_dirty_for_io(page))
1585 goto continue_unlock;
1587 set_fsync_mark(page, 0);
1588 set_dentry_mark(page, 0);
1590 if (NODE_MAPPING(sbi)->a_ops->writepage(page, wbc))
1595 if (--wbc->nr_to_write == 0)
1598 pagevec_release(&pvec);
1601 if (wbc->nr_to_write == 0) {
1613 f2fs_submit_merged_bio(sbi, NODE, WRITE);
1617 int wait_on_node_pages_writeback(struct f2fs_sb_info *sbi, nid_t ino)
1619 pgoff_t index = 0, end = ULONG_MAX;
1620 struct pagevec pvec;
1623 pagevec_init(&pvec, 0);
1625 while (index <= end) {
1627 nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1628 PAGECACHE_TAG_WRITEBACK,
1629 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
1633 for (i = 0; i < nr_pages; i++) {
1634 struct page *page = pvec.pages[i];
1636 /* until radix tree lookup accepts end_index */
1637 if (unlikely(page->index > end))
1640 if (ino && ino_of_node(page) == ino) {
1641 f2fs_wait_on_page_writeback(page, NODE, true);
1642 if (TestClearPageError(page))
1646 pagevec_release(&pvec);
1650 ret2 = filemap_check_errors(NODE_MAPPING(sbi));
1656 static int f2fs_write_node_pages(struct address_space *mapping,
1657 struct writeback_control *wbc)
1659 struct f2fs_sb_info *sbi = F2FS_M_SB(mapping);
1660 struct blk_plug plug;
1663 /* balancing f2fs's metadata in background */
1664 f2fs_balance_fs_bg(sbi);
1666 /* collect a number of dirty node pages and write together */
1667 if (get_pages(sbi, F2FS_DIRTY_NODES) < nr_pages_to_skip(sbi, NODE))
1670 trace_f2fs_writepages(mapping->host, wbc, NODE);
1672 diff = nr_pages_to_write(sbi, NODE, wbc);
1673 wbc->sync_mode = WB_SYNC_NONE;
1674 blk_start_plug(&plug);
1675 sync_node_pages(sbi, wbc);
1676 blk_finish_plug(&plug);
1677 wbc->nr_to_write = max((long)0, wbc->nr_to_write - diff);
1681 wbc->pages_skipped += get_pages(sbi, F2FS_DIRTY_NODES);
1682 trace_f2fs_writepages(mapping->host, wbc, NODE);
1686 static int f2fs_set_node_page_dirty(struct page *page)
1688 trace_f2fs_set_page_dirty(page, NODE);
1690 if (!PageUptodate(page))
1691 SetPageUptodate(page);
1692 if (!PageDirty(page)) {
1693 f2fs_set_page_dirty_nobuffers(page);
1694 inc_page_count(F2FS_P_SB(page), F2FS_DIRTY_NODES);
1695 SetPagePrivate(page);
1696 f2fs_trace_pid(page);
1703 * Structure of the f2fs node operations
1705 const struct address_space_operations f2fs_node_aops = {
1706 .writepage = f2fs_write_node_page,
1707 .writepages = f2fs_write_node_pages,
1708 .set_page_dirty = f2fs_set_node_page_dirty,
1709 .invalidatepage = f2fs_invalidate_page,
1710 .releasepage = f2fs_release_page,
1711 #ifdef CONFIG_MIGRATION
1712 .migratepage = f2fs_migrate_page,
1716 static struct free_nid *__lookup_free_nid_list(struct f2fs_nm_info *nm_i,
1719 return radix_tree_lookup(&nm_i->free_nid_root, n);
1722 static int __insert_nid_to_list(struct f2fs_sb_info *sbi,
1723 struct free_nid *i, enum nid_list list, bool new)
1725 struct f2fs_nm_info *nm_i = NM_I(sbi);
1728 int err = radix_tree_insert(&nm_i->free_nid_root, i->nid, i);
1733 f2fs_bug_on(sbi, list == FREE_NID_LIST ? i->state != NID_NEW :
1734 i->state != NID_ALLOC);
1735 nm_i->nid_cnt[list]++;
1736 list_add_tail(&i->list, &nm_i->nid_list[list]);
1740 static void __remove_nid_from_list(struct f2fs_sb_info *sbi,
1741 struct free_nid *i, enum nid_list list, bool reuse)
1743 struct f2fs_nm_info *nm_i = NM_I(sbi);
1745 f2fs_bug_on(sbi, list == FREE_NID_LIST ? i->state != NID_NEW :
1746 i->state != NID_ALLOC);
1747 nm_i->nid_cnt[list]--;
1750 radix_tree_delete(&nm_i->free_nid_root, i->nid);
1753 static int add_free_nid(struct f2fs_sb_info *sbi, nid_t nid, bool build)
1755 struct f2fs_nm_info *nm_i = NM_I(sbi);
1757 struct nat_entry *ne;
1760 /* 0 nid should not be used */
1761 if (unlikely(nid == 0))
1765 /* do not add allocated nids */
1766 ne = __lookup_nat_cache(nm_i, nid);
1767 if (ne && (!get_nat_flag(ne, IS_CHECKPOINTED) ||
1768 nat_get_blkaddr(ne) != NULL_ADDR))
1772 i = f2fs_kmem_cache_alloc(free_nid_slab, GFP_NOFS);
1776 if (radix_tree_preload(GFP_NOFS)) {
1777 kmem_cache_free(free_nid_slab, i);
1781 spin_lock(&nm_i->nid_list_lock);
1782 err = __insert_nid_to_list(sbi, i, FREE_NID_LIST, true);
1783 spin_unlock(&nm_i->nid_list_lock);
1784 radix_tree_preload_end();
1786 kmem_cache_free(free_nid_slab, i);
1792 static void remove_free_nid(struct f2fs_sb_info *sbi, nid_t nid)
1794 struct f2fs_nm_info *nm_i = NM_I(sbi);
1796 bool need_free = false;
1798 spin_lock(&nm_i->nid_list_lock);
1799 i = __lookup_free_nid_list(nm_i, nid);
1800 if (i && i->state == NID_NEW) {
1801 __remove_nid_from_list(sbi, i, FREE_NID_LIST, false);
1804 spin_unlock(&nm_i->nid_list_lock);
1807 kmem_cache_free(free_nid_slab, i);
1810 static void scan_nat_page(struct f2fs_sb_info *sbi,
1811 struct page *nat_page, nid_t start_nid)
1813 struct f2fs_nm_info *nm_i = NM_I(sbi);
1814 struct f2fs_nat_block *nat_blk = page_address(nat_page);
1818 i = start_nid % NAT_ENTRY_PER_BLOCK;
1820 for (; i < NAT_ENTRY_PER_BLOCK; i++, start_nid++) {
1822 if (unlikely(start_nid >= nm_i->max_nid))
1825 blk_addr = le32_to_cpu(nat_blk->entries[i].block_addr);
1826 f2fs_bug_on(sbi, blk_addr == NEW_ADDR);
1827 if (blk_addr == NULL_ADDR)
1828 add_free_nid(sbi, start_nid, true);
1832 static void __build_free_nids(struct f2fs_sb_info *sbi, bool sync)
1834 struct f2fs_nm_info *nm_i = NM_I(sbi);
1835 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
1836 struct f2fs_journal *journal = curseg->journal;
1838 nid_t nid = nm_i->next_scan_nid;
1840 /* Enough entries */
1841 if (nm_i->nid_cnt[FREE_NID_LIST] >= NAT_ENTRY_PER_BLOCK)
1844 if (!sync && !available_free_memory(sbi, FREE_NIDS))
1847 /* readahead nat pages to be scanned */
1848 ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nid), FREE_NID_PAGES,
1851 down_read(&nm_i->nat_tree_lock);
1854 struct page *page = get_current_nat_page(sbi, nid);
1856 scan_nat_page(sbi, page, nid);
1857 f2fs_put_page(page, 1);
1859 nid += (NAT_ENTRY_PER_BLOCK - (nid % NAT_ENTRY_PER_BLOCK));
1860 if (unlikely(nid >= nm_i->max_nid))
1863 if (++i >= FREE_NID_PAGES)
1867 /* go to the next free nat pages to find free nids abundantly */
1868 nm_i->next_scan_nid = nid;
1870 /* find free nids from current sum_pages */
1871 down_read(&curseg->journal_rwsem);
1872 for (i = 0; i < nats_in_cursum(journal); i++) {
1875 addr = le32_to_cpu(nat_in_journal(journal, i).block_addr);
1876 nid = le32_to_cpu(nid_in_journal(journal, i));
1877 if (addr == NULL_ADDR)
1878 add_free_nid(sbi, nid, true);
1880 remove_free_nid(sbi, nid);
1882 up_read(&curseg->journal_rwsem);
1883 up_read(&nm_i->nat_tree_lock);
1885 ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nm_i->next_scan_nid),
1886 nm_i->ra_nid_pages, META_NAT, false);
1889 void build_free_nids(struct f2fs_sb_info *sbi, bool sync)
1891 mutex_lock(&NM_I(sbi)->build_lock);
1892 __build_free_nids(sbi, sync);
1893 mutex_unlock(&NM_I(sbi)->build_lock);
1897 * If this function returns success, caller can obtain a new nid
1898 * from second parameter of this function.
1899 * The returned nid could be used ino as well as nid when inode is created.
1901 bool alloc_nid(struct f2fs_sb_info *sbi, nid_t *nid)
1903 struct f2fs_nm_info *nm_i = NM_I(sbi);
1904 struct free_nid *i = NULL;
1906 #ifdef CONFIG_F2FS_FAULT_INJECTION
1907 if (time_to_inject(sbi, FAULT_ALLOC_NID))
1910 spin_lock(&nm_i->nid_list_lock);
1912 if (unlikely(nm_i->available_nids == 0)) {
1913 spin_unlock(&nm_i->nid_list_lock);
1917 /* We should not use stale free nids created by build_free_nids */
1918 if (nm_i->nid_cnt[FREE_NID_LIST] && !on_build_free_nids(nm_i)) {
1919 f2fs_bug_on(sbi, list_empty(&nm_i->nid_list[FREE_NID_LIST]));
1920 i = list_first_entry(&nm_i->nid_list[FREE_NID_LIST],
1921 struct free_nid, list);
1924 __remove_nid_from_list(sbi, i, FREE_NID_LIST, true);
1925 i->state = NID_ALLOC;
1926 __insert_nid_to_list(sbi, i, ALLOC_NID_LIST, false);
1927 nm_i->available_nids--;
1928 spin_unlock(&nm_i->nid_list_lock);
1931 spin_unlock(&nm_i->nid_list_lock);
1933 /* Let's scan nat pages and its caches to get free nids */
1934 build_free_nids(sbi, true);
1939 * alloc_nid() should be called prior to this function.
1941 void alloc_nid_done(struct f2fs_sb_info *sbi, nid_t nid)
1943 struct f2fs_nm_info *nm_i = NM_I(sbi);
1946 spin_lock(&nm_i->nid_list_lock);
1947 i = __lookup_free_nid_list(nm_i, nid);
1948 f2fs_bug_on(sbi, !i);
1949 __remove_nid_from_list(sbi, i, ALLOC_NID_LIST, false);
1950 spin_unlock(&nm_i->nid_list_lock);
1952 kmem_cache_free(free_nid_slab, i);
1956 * alloc_nid() should be called prior to this function.
1958 void alloc_nid_failed(struct f2fs_sb_info *sbi, nid_t nid)
1960 struct f2fs_nm_info *nm_i = NM_I(sbi);
1962 bool need_free = false;
1967 spin_lock(&nm_i->nid_list_lock);
1968 i = __lookup_free_nid_list(nm_i, nid);
1969 f2fs_bug_on(sbi, !i);
1971 if (!available_free_memory(sbi, FREE_NIDS)) {
1972 __remove_nid_from_list(sbi, i, ALLOC_NID_LIST, false);
1975 __remove_nid_from_list(sbi, i, ALLOC_NID_LIST, true);
1977 __insert_nid_to_list(sbi, i, FREE_NID_LIST, false);
1980 nm_i->available_nids++;
1982 spin_unlock(&nm_i->nid_list_lock);
1985 kmem_cache_free(free_nid_slab, i);
1988 int try_to_free_nids(struct f2fs_sb_info *sbi, int nr_shrink)
1990 struct f2fs_nm_info *nm_i = NM_I(sbi);
1991 struct free_nid *i, *next;
1994 if (nm_i->nid_cnt[FREE_NID_LIST] <= MAX_FREE_NIDS)
1997 if (!mutex_trylock(&nm_i->build_lock))
2000 spin_lock(&nm_i->nid_list_lock);
2001 list_for_each_entry_safe(i, next, &nm_i->nid_list[FREE_NID_LIST],
2003 if (nr_shrink <= 0 ||
2004 nm_i->nid_cnt[FREE_NID_LIST] <= MAX_FREE_NIDS)
2007 __remove_nid_from_list(sbi, i, FREE_NID_LIST, false);
2008 kmem_cache_free(free_nid_slab, i);
2011 spin_unlock(&nm_i->nid_list_lock);
2012 mutex_unlock(&nm_i->build_lock);
2014 return nr - nr_shrink;
2017 void recover_inline_xattr(struct inode *inode, struct page *page)
2019 void *src_addr, *dst_addr;
2022 struct f2fs_inode *ri;
2024 ipage = get_node_page(F2FS_I_SB(inode), inode->i_ino);
2025 f2fs_bug_on(F2FS_I_SB(inode), IS_ERR(ipage));
2027 ri = F2FS_INODE(page);
2028 if (!(ri->i_inline & F2FS_INLINE_XATTR)) {
2029 clear_inode_flag(inode, FI_INLINE_XATTR);
2033 dst_addr = inline_xattr_addr(ipage);
2034 src_addr = inline_xattr_addr(page);
2035 inline_size = inline_xattr_size(inode);
2037 f2fs_wait_on_page_writeback(ipage, NODE, true);
2038 memcpy(dst_addr, src_addr, inline_size);
2040 update_inode(inode, ipage);
2041 f2fs_put_page(ipage, 1);
2044 void recover_xattr_data(struct inode *inode, struct page *page, block_t blkaddr)
2046 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
2047 nid_t prev_xnid = F2FS_I(inode)->i_xattr_nid;
2048 nid_t new_xnid = nid_of_node(page);
2049 struct node_info ni;
2051 /* 1: invalidate the previous xattr nid */
2055 /* Deallocate node address */
2056 get_node_info(sbi, prev_xnid, &ni);
2057 f2fs_bug_on(sbi, ni.blk_addr == NULL_ADDR);
2058 invalidate_blocks(sbi, ni.blk_addr);
2059 dec_valid_node_count(sbi, inode);
2060 set_node_addr(sbi, &ni, NULL_ADDR, false);
2063 /* 2: allocate new xattr nid */
2064 if (unlikely(!inc_valid_node_count(sbi, inode)))
2065 f2fs_bug_on(sbi, 1);
2067 remove_free_nid(sbi, new_xnid);
2068 get_node_info(sbi, new_xnid, &ni);
2069 ni.ino = inode->i_ino;
2070 set_node_addr(sbi, &ni, NEW_ADDR, false);
2071 f2fs_i_xnid_write(inode, new_xnid);
2073 /* 3: update xattr blkaddr */
2074 refresh_sit_entry(sbi, NEW_ADDR, blkaddr);
2075 set_node_addr(sbi, &ni, blkaddr, false);
2078 int recover_inode_page(struct f2fs_sb_info *sbi, struct page *page)
2080 struct f2fs_inode *src, *dst;
2081 nid_t ino = ino_of_node(page);
2082 struct node_info old_ni, new_ni;
2085 get_node_info(sbi, ino, &old_ni);
2087 if (unlikely(old_ni.blk_addr != NULL_ADDR))
2090 ipage = f2fs_grab_cache_page(NODE_MAPPING(sbi), ino, false);
2092 congestion_wait(BLK_RW_ASYNC, HZ/50);
2096 /* Should not use this inode from free nid list */
2097 remove_free_nid(sbi, ino);
2099 if (!PageUptodate(ipage))
2100 SetPageUptodate(ipage);
2101 fill_node_footer(ipage, ino, ino, 0, true);
2103 src = F2FS_INODE(page);
2104 dst = F2FS_INODE(ipage);
2106 memcpy(dst, src, (unsigned long)&src->i_ext - (unsigned long)src);
2108 dst->i_blocks = cpu_to_le64(1);
2109 dst->i_links = cpu_to_le32(1);
2110 dst->i_xattr_nid = 0;
2111 dst->i_inline = src->i_inline & F2FS_INLINE_XATTR;
2116 if (unlikely(!inc_valid_node_count(sbi, NULL)))
2118 set_node_addr(sbi, &new_ni, NEW_ADDR, false);
2119 inc_valid_inode_count(sbi);
2120 set_page_dirty(ipage);
2121 f2fs_put_page(ipage, 1);
2125 int restore_node_summary(struct f2fs_sb_info *sbi,
2126 unsigned int segno, struct f2fs_summary_block *sum)
2128 struct f2fs_node *rn;
2129 struct f2fs_summary *sum_entry;
2131 int i, idx, last_offset, nrpages;
2133 /* scan the node segment */
2134 last_offset = sbi->blocks_per_seg;
2135 addr = START_BLOCK(sbi, segno);
2136 sum_entry = &sum->entries[0];
2138 for (i = 0; i < last_offset; i += nrpages, addr += nrpages) {
2139 nrpages = min(last_offset - i, BIO_MAX_PAGES);
2141 /* readahead node pages */
2142 ra_meta_pages(sbi, addr, nrpages, META_POR, true);
2144 for (idx = addr; idx < addr + nrpages; idx++) {
2145 struct page *page = get_tmp_page(sbi, idx);
2147 rn = F2FS_NODE(page);
2148 sum_entry->nid = rn->footer.nid;
2149 sum_entry->version = 0;
2150 sum_entry->ofs_in_node = 0;
2152 f2fs_put_page(page, 1);
2155 invalidate_mapping_pages(META_MAPPING(sbi), addr,
2161 static void remove_nats_in_journal(struct f2fs_sb_info *sbi)
2163 struct f2fs_nm_info *nm_i = NM_I(sbi);
2164 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
2165 struct f2fs_journal *journal = curseg->journal;
2168 down_write(&curseg->journal_rwsem);
2169 for (i = 0; i < nats_in_cursum(journal); i++) {
2170 struct nat_entry *ne;
2171 struct f2fs_nat_entry raw_ne;
2172 nid_t nid = le32_to_cpu(nid_in_journal(journal, i));
2174 raw_ne = nat_in_journal(journal, i);
2176 ne = __lookup_nat_cache(nm_i, nid);
2178 ne = grab_nat_entry(nm_i, nid, true);
2179 node_info_from_raw_nat(&ne->ni, &raw_ne);
2183 * if a free nat in journal has not been used after last
2184 * checkpoint, we should remove it from available nids,
2185 * since later we will add it again.
2187 if (!get_nat_flag(ne, IS_DIRTY) &&
2188 le32_to_cpu(raw_ne.block_addr) == NULL_ADDR) {
2189 spin_lock(&nm_i->nid_list_lock);
2190 nm_i->available_nids--;
2191 spin_unlock(&nm_i->nid_list_lock);
2194 __set_nat_cache_dirty(nm_i, ne);
2196 update_nats_in_cursum(journal, -i);
2197 up_write(&curseg->journal_rwsem);
2200 static void __adjust_nat_entry_set(struct nat_entry_set *nes,
2201 struct list_head *head, int max)
2203 struct nat_entry_set *cur;
2205 if (nes->entry_cnt >= max)
2208 list_for_each_entry(cur, head, set_list) {
2209 if (cur->entry_cnt >= nes->entry_cnt) {
2210 list_add(&nes->set_list, cur->set_list.prev);
2215 list_add_tail(&nes->set_list, head);
2218 static void __flush_nat_entry_set(struct f2fs_sb_info *sbi,
2219 struct nat_entry_set *set)
2221 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
2222 struct f2fs_journal *journal = curseg->journal;
2223 nid_t start_nid = set->set * NAT_ENTRY_PER_BLOCK;
2224 bool to_journal = true;
2225 struct f2fs_nat_block *nat_blk;
2226 struct nat_entry *ne, *cur;
2227 struct page *page = NULL;
2230 * there are two steps to flush nat entries:
2231 * #1, flush nat entries to journal in current hot data summary block.
2232 * #2, flush nat entries to nat page.
2234 if (!__has_cursum_space(journal, set->entry_cnt, NAT_JOURNAL))
2238 down_write(&curseg->journal_rwsem);
2240 page = get_next_nat_page(sbi, start_nid);
2241 nat_blk = page_address(page);
2242 f2fs_bug_on(sbi, !nat_blk);
2245 /* flush dirty nats in nat entry set */
2246 list_for_each_entry_safe(ne, cur, &set->entry_list, list) {
2247 struct f2fs_nat_entry *raw_ne;
2248 nid_t nid = nat_get_nid(ne);
2251 if (nat_get_blkaddr(ne) == NEW_ADDR)
2255 offset = lookup_journal_in_cursum(journal,
2256 NAT_JOURNAL, nid, 1);
2257 f2fs_bug_on(sbi, offset < 0);
2258 raw_ne = &nat_in_journal(journal, offset);
2259 nid_in_journal(journal, offset) = cpu_to_le32(nid);
2261 raw_ne = &nat_blk->entries[nid - start_nid];
2263 raw_nat_from_node_info(raw_ne, &ne->ni);
2265 __clear_nat_cache_dirty(NM_I(sbi), ne);
2266 if (nat_get_blkaddr(ne) == NULL_ADDR) {
2267 add_free_nid(sbi, nid, false);
2268 spin_lock(&NM_I(sbi)->nid_list_lock);
2269 NM_I(sbi)->available_nids++;
2270 spin_unlock(&NM_I(sbi)->nid_list_lock);
2275 up_write(&curseg->journal_rwsem);
2277 f2fs_put_page(page, 1);
2279 f2fs_bug_on(sbi, set->entry_cnt);
2281 radix_tree_delete(&NM_I(sbi)->nat_set_root, set->set);
2282 kmem_cache_free(nat_entry_set_slab, set);
2286 * This function is called during the checkpointing process.
2288 void flush_nat_entries(struct f2fs_sb_info *sbi)
2290 struct f2fs_nm_info *nm_i = NM_I(sbi);
2291 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
2292 struct f2fs_journal *journal = curseg->journal;
2293 struct nat_entry_set *setvec[SETVEC_SIZE];
2294 struct nat_entry_set *set, *tmp;
2299 if (!nm_i->dirty_nat_cnt)
2302 down_write(&nm_i->nat_tree_lock);
2305 * if there are no enough space in journal to store dirty nat
2306 * entries, remove all entries from journal and merge them
2307 * into nat entry set.
2309 if (!__has_cursum_space(journal, nm_i->dirty_nat_cnt, NAT_JOURNAL))
2310 remove_nats_in_journal(sbi);
2312 while ((found = __gang_lookup_nat_set(nm_i,
2313 set_idx, SETVEC_SIZE, setvec))) {
2315 set_idx = setvec[found - 1]->set + 1;
2316 for (idx = 0; idx < found; idx++)
2317 __adjust_nat_entry_set(setvec[idx], &sets,
2318 MAX_NAT_JENTRIES(journal));
2321 /* flush dirty nats in nat entry set */
2322 list_for_each_entry_safe(set, tmp, &sets, set_list)
2323 __flush_nat_entry_set(sbi, set);
2325 up_write(&nm_i->nat_tree_lock);
2327 f2fs_bug_on(sbi, nm_i->dirty_nat_cnt);
2330 static int init_node_manager(struct f2fs_sb_info *sbi)
2332 struct f2fs_super_block *sb_raw = F2FS_RAW_SUPER(sbi);
2333 struct f2fs_nm_info *nm_i = NM_I(sbi);
2334 unsigned char *version_bitmap;
2335 unsigned int nat_segs, nat_blocks;
2337 nm_i->nat_blkaddr = le32_to_cpu(sb_raw->nat_blkaddr);
2339 /* segment_count_nat includes pair segment so divide to 2. */
2340 nat_segs = le32_to_cpu(sb_raw->segment_count_nat) >> 1;
2341 nat_blocks = nat_segs << le32_to_cpu(sb_raw->log_blocks_per_seg);
2343 nm_i->max_nid = NAT_ENTRY_PER_BLOCK * nat_blocks;
2345 /* not used nids: 0, node, meta, (and root counted as valid node) */
2346 nm_i->available_nids = nm_i->max_nid - sbi->total_valid_node_count -
2347 F2FS_RESERVED_NODE_NUM;
2348 nm_i->nid_cnt[FREE_NID_LIST] = 0;
2349 nm_i->nid_cnt[ALLOC_NID_LIST] = 0;
2351 nm_i->ram_thresh = DEF_RAM_THRESHOLD;
2352 nm_i->ra_nid_pages = DEF_RA_NID_PAGES;
2353 nm_i->dirty_nats_ratio = DEF_DIRTY_NAT_RATIO_THRESHOLD;
2355 INIT_RADIX_TREE(&nm_i->free_nid_root, GFP_ATOMIC);
2356 INIT_LIST_HEAD(&nm_i->nid_list[FREE_NID_LIST]);
2357 INIT_LIST_HEAD(&nm_i->nid_list[ALLOC_NID_LIST]);
2358 INIT_RADIX_TREE(&nm_i->nat_root, GFP_NOIO);
2359 INIT_RADIX_TREE(&nm_i->nat_set_root, GFP_NOIO);
2360 INIT_LIST_HEAD(&nm_i->nat_entries);
2362 mutex_init(&nm_i->build_lock);
2363 spin_lock_init(&nm_i->nid_list_lock);
2364 init_rwsem(&nm_i->nat_tree_lock);
2366 nm_i->next_scan_nid = le32_to_cpu(sbi->ckpt->next_free_nid);
2367 nm_i->bitmap_size = __bitmap_size(sbi, NAT_BITMAP);
2368 version_bitmap = __bitmap_ptr(sbi, NAT_BITMAP);
2369 if (!version_bitmap)
2372 nm_i->nat_bitmap = kmemdup(version_bitmap, nm_i->bitmap_size,
2374 if (!nm_i->nat_bitmap)
2377 #ifdef CONFIG_F2FS_CHECK_FS
2378 nm_i->nat_bitmap_mir = kmemdup(version_bitmap, nm_i->bitmap_size,
2380 if (!nm_i->nat_bitmap_mir)
2387 int build_node_manager(struct f2fs_sb_info *sbi)
2391 sbi->nm_info = kzalloc(sizeof(struct f2fs_nm_info), GFP_KERNEL);
2395 err = init_node_manager(sbi);
2399 build_free_nids(sbi, true);
2403 void destroy_node_manager(struct f2fs_sb_info *sbi)
2405 struct f2fs_nm_info *nm_i = NM_I(sbi);
2406 struct free_nid *i, *next_i;
2407 struct nat_entry *natvec[NATVEC_SIZE];
2408 struct nat_entry_set *setvec[SETVEC_SIZE];
2415 /* destroy free nid list */
2416 spin_lock(&nm_i->nid_list_lock);
2417 list_for_each_entry_safe(i, next_i, &nm_i->nid_list[FREE_NID_LIST],
2419 __remove_nid_from_list(sbi, i, FREE_NID_LIST, false);
2420 spin_unlock(&nm_i->nid_list_lock);
2421 kmem_cache_free(free_nid_slab, i);
2422 spin_lock(&nm_i->nid_list_lock);
2424 f2fs_bug_on(sbi, nm_i->nid_cnt[FREE_NID_LIST]);
2425 f2fs_bug_on(sbi, nm_i->nid_cnt[ALLOC_NID_LIST]);
2426 f2fs_bug_on(sbi, !list_empty(&nm_i->nid_list[ALLOC_NID_LIST]));
2427 spin_unlock(&nm_i->nid_list_lock);
2429 /* destroy nat cache */
2430 down_write(&nm_i->nat_tree_lock);
2431 while ((found = __gang_lookup_nat_cache(nm_i,
2432 nid, NATVEC_SIZE, natvec))) {
2435 nid = nat_get_nid(natvec[found - 1]) + 1;
2436 for (idx = 0; idx < found; idx++)
2437 __del_from_nat_cache(nm_i, natvec[idx]);
2439 f2fs_bug_on(sbi, nm_i->nat_cnt);
2441 /* destroy nat set cache */
2443 while ((found = __gang_lookup_nat_set(nm_i,
2444 nid, SETVEC_SIZE, setvec))) {
2447 nid = setvec[found - 1]->set + 1;
2448 for (idx = 0; idx < found; idx++) {
2449 /* entry_cnt is not zero, when cp_error was occurred */
2450 f2fs_bug_on(sbi, !list_empty(&setvec[idx]->entry_list));
2451 radix_tree_delete(&nm_i->nat_set_root, setvec[idx]->set);
2452 kmem_cache_free(nat_entry_set_slab, setvec[idx]);
2455 up_write(&nm_i->nat_tree_lock);
2457 kfree(nm_i->nat_bitmap);
2458 #ifdef CONFIG_F2FS_CHECK_FS
2459 kfree(nm_i->nat_bitmap_mir);
2461 sbi->nm_info = NULL;
2465 int __init create_node_manager_caches(void)
2467 nat_entry_slab = f2fs_kmem_cache_create("nat_entry",
2468 sizeof(struct nat_entry));
2469 if (!nat_entry_slab)
2472 free_nid_slab = f2fs_kmem_cache_create("free_nid",
2473 sizeof(struct free_nid));
2475 goto destroy_nat_entry;
2477 nat_entry_set_slab = f2fs_kmem_cache_create("nat_entry_set",
2478 sizeof(struct nat_entry_set));
2479 if (!nat_entry_set_slab)
2480 goto destroy_free_nid;
2484 kmem_cache_destroy(free_nid_slab);
2486 kmem_cache_destroy(nat_entry_slab);
2491 void destroy_node_manager_caches(void)
2493 kmem_cache_destroy(nat_entry_set_slab);
2494 kmem_cache_destroy(free_nid_slab);
2495 kmem_cache_destroy(nat_entry_slab);