1 // SPDX-License-Identifier: GPL-2.0
5 * Copyright (c) 2012 Samsung Electronics Co., Ltd.
6 * http://www.samsung.com/
9 #include <linux/f2fs_fs.h>
10 #include <linux/mpage.h>
11 #include <linux/backing-dev.h>
12 #include <linux/blkdev.h>
13 #include <linux/pagevec.h>
14 #include <linux/swap.h>
21 #include <trace/events/f2fs.h>
23 #define on_f2fs_build_free_nids(nmi) mutex_is_locked(&(nm_i)->build_lock)
25 static struct kmem_cache *nat_entry_slab;
26 static struct kmem_cache *free_nid_slab;
27 static struct kmem_cache *nat_entry_set_slab;
28 static struct kmem_cache *fsync_node_entry_slab;
31 * Check whether the given nid is within node id range.
33 int f2fs_check_nid_range(struct f2fs_sb_info *sbi, nid_t nid)
35 if (unlikely(nid < F2FS_ROOT_INO(sbi) || nid >= NM_I(sbi)->max_nid)) {
36 set_sbi_flag(sbi, SBI_NEED_FSCK);
37 f2fs_msg(sbi->sb, KERN_WARNING,
38 "%s: out-of-range nid=%x, run fsck to fix.",
45 bool f2fs_available_free_memory(struct f2fs_sb_info *sbi, int type)
47 struct f2fs_nm_info *nm_i = NM_I(sbi);
49 unsigned long avail_ram;
50 unsigned long mem_size = 0;
55 /* only uses low memory */
56 avail_ram = val.totalram - val.totalhigh;
59 * give 25%, 25%, 50%, 50%, 50% memory for each components respectively
61 if (type == FREE_NIDS) {
62 mem_size = (nm_i->nid_cnt[FREE_NID] *
63 sizeof(struct free_nid)) >> PAGE_SHIFT;
64 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 2);
65 } else if (type == NAT_ENTRIES) {
66 mem_size = (nm_i->nat_cnt * sizeof(struct nat_entry)) >>
68 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 2);
69 if (excess_cached_nats(sbi))
71 } else if (type == DIRTY_DENTS) {
72 if (sbi->sb->s_bdi->wb.dirty_exceeded)
74 mem_size = get_pages(sbi, F2FS_DIRTY_DENTS);
75 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 1);
76 } else if (type == INO_ENTRIES) {
79 for (i = 0; i < MAX_INO_ENTRY; i++)
80 mem_size += sbi->im[i].ino_num *
81 sizeof(struct ino_entry);
82 mem_size >>= PAGE_SHIFT;
83 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 1);
84 } else if (type == EXTENT_CACHE) {
85 mem_size = (atomic_read(&sbi->total_ext_tree) *
86 sizeof(struct extent_tree) +
87 atomic_read(&sbi->total_ext_node) *
88 sizeof(struct extent_node)) >> PAGE_SHIFT;
89 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 1);
90 } else if (type == INMEM_PAGES) {
91 /* it allows 20% / total_ram for inmemory pages */
92 mem_size = get_pages(sbi, F2FS_INMEM_PAGES);
93 res = mem_size < (val.totalram / 5);
95 if (!sbi->sb->s_bdi->wb.dirty_exceeded)
101 static void clear_node_page_dirty(struct page *page)
103 if (PageDirty(page)) {
104 f2fs_clear_page_cache_dirty_tag(page);
105 clear_page_dirty_for_io(page);
106 dec_page_count(F2FS_P_SB(page), F2FS_DIRTY_NODES);
108 ClearPageUptodate(page);
111 static struct page *get_current_nat_page(struct f2fs_sb_info *sbi, nid_t nid)
113 return f2fs_get_meta_page_nofail(sbi, current_nat_addr(sbi, nid));
116 static struct page *get_next_nat_page(struct f2fs_sb_info *sbi, nid_t nid)
118 struct page *src_page;
119 struct page *dst_page;
123 struct f2fs_nm_info *nm_i = NM_I(sbi);
125 dst_off = next_nat_addr(sbi, current_nat_addr(sbi, nid));
127 /* get current nat block page with lock */
128 src_page = get_current_nat_page(sbi, nid);
129 if (IS_ERR(src_page))
131 dst_page = f2fs_grab_meta_page(sbi, dst_off);
132 f2fs_bug_on(sbi, PageDirty(src_page));
134 src_addr = page_address(src_page);
135 dst_addr = page_address(dst_page);
136 memcpy(dst_addr, src_addr, PAGE_SIZE);
137 set_page_dirty(dst_page);
138 f2fs_put_page(src_page, 1);
140 set_to_next_nat(nm_i, nid);
145 static struct nat_entry *__alloc_nat_entry(nid_t nid, bool no_fail)
147 struct nat_entry *new;
150 new = f2fs_kmem_cache_alloc(nat_entry_slab, GFP_F2FS_ZERO);
152 new = kmem_cache_alloc(nat_entry_slab, GFP_F2FS_ZERO);
154 nat_set_nid(new, nid);
160 static void __free_nat_entry(struct nat_entry *e)
162 kmem_cache_free(nat_entry_slab, e);
165 /* must be locked by nat_tree_lock */
166 static struct nat_entry *__init_nat_entry(struct f2fs_nm_info *nm_i,
167 struct nat_entry *ne, struct f2fs_nat_entry *raw_ne, bool no_fail)
170 f2fs_radix_tree_insert(&nm_i->nat_root, nat_get_nid(ne), ne);
171 else if (radix_tree_insert(&nm_i->nat_root, nat_get_nid(ne), ne))
175 node_info_from_raw_nat(&ne->ni, raw_ne);
177 spin_lock(&nm_i->nat_list_lock);
178 list_add_tail(&ne->list, &nm_i->nat_entries);
179 spin_unlock(&nm_i->nat_list_lock);
185 static struct nat_entry *__lookup_nat_cache(struct f2fs_nm_info *nm_i, nid_t n)
187 struct nat_entry *ne;
189 ne = radix_tree_lookup(&nm_i->nat_root, n);
191 /* for recent accessed nat entry, move it to tail of lru list */
192 if (ne && !get_nat_flag(ne, IS_DIRTY)) {
193 spin_lock(&nm_i->nat_list_lock);
194 if (!list_empty(&ne->list))
195 list_move_tail(&ne->list, &nm_i->nat_entries);
196 spin_unlock(&nm_i->nat_list_lock);
202 static unsigned int __gang_lookup_nat_cache(struct f2fs_nm_info *nm_i,
203 nid_t start, unsigned int nr, struct nat_entry **ep)
205 return radix_tree_gang_lookup(&nm_i->nat_root, (void **)ep, start, nr);
208 static void __del_from_nat_cache(struct f2fs_nm_info *nm_i, struct nat_entry *e)
210 radix_tree_delete(&nm_i->nat_root, nat_get_nid(e));
215 static struct nat_entry_set *__grab_nat_entry_set(struct f2fs_nm_info *nm_i,
216 struct nat_entry *ne)
218 nid_t set = NAT_BLOCK_OFFSET(ne->ni.nid);
219 struct nat_entry_set *head;
221 head = radix_tree_lookup(&nm_i->nat_set_root, set);
223 head = f2fs_kmem_cache_alloc(nat_entry_set_slab, GFP_NOFS);
225 INIT_LIST_HEAD(&head->entry_list);
226 INIT_LIST_HEAD(&head->set_list);
229 f2fs_radix_tree_insert(&nm_i->nat_set_root, set, head);
234 static void __set_nat_cache_dirty(struct f2fs_nm_info *nm_i,
235 struct nat_entry *ne)
237 struct nat_entry_set *head;
238 bool new_ne = nat_get_blkaddr(ne) == NEW_ADDR;
241 head = __grab_nat_entry_set(nm_i, ne);
244 * update entry_cnt in below condition:
245 * 1. update NEW_ADDR to valid block address;
246 * 2. update old block address to new one;
248 if (!new_ne && (get_nat_flag(ne, IS_PREALLOC) ||
249 !get_nat_flag(ne, IS_DIRTY)))
252 set_nat_flag(ne, IS_PREALLOC, new_ne);
254 if (get_nat_flag(ne, IS_DIRTY))
257 nm_i->dirty_nat_cnt++;
258 set_nat_flag(ne, IS_DIRTY, true);
260 spin_lock(&nm_i->nat_list_lock);
262 list_del_init(&ne->list);
264 list_move_tail(&ne->list, &head->entry_list);
265 spin_unlock(&nm_i->nat_list_lock);
268 static void __clear_nat_cache_dirty(struct f2fs_nm_info *nm_i,
269 struct nat_entry_set *set, struct nat_entry *ne)
271 spin_lock(&nm_i->nat_list_lock);
272 list_move_tail(&ne->list, &nm_i->nat_entries);
273 spin_unlock(&nm_i->nat_list_lock);
275 set_nat_flag(ne, IS_DIRTY, false);
277 nm_i->dirty_nat_cnt--;
280 static unsigned int __gang_lookup_nat_set(struct f2fs_nm_info *nm_i,
281 nid_t start, unsigned int nr, struct nat_entry_set **ep)
283 return radix_tree_gang_lookup(&nm_i->nat_set_root, (void **)ep,
287 bool f2fs_in_warm_node_list(struct f2fs_sb_info *sbi, struct page *page)
289 return NODE_MAPPING(sbi) == page->mapping &&
290 IS_DNODE(page) && is_cold_node(page);
293 void f2fs_init_fsync_node_info(struct f2fs_sb_info *sbi)
295 spin_lock_init(&sbi->fsync_node_lock);
296 INIT_LIST_HEAD(&sbi->fsync_node_list);
297 sbi->fsync_seg_id = 0;
298 sbi->fsync_node_num = 0;
301 static unsigned int f2fs_add_fsync_node_entry(struct f2fs_sb_info *sbi,
304 struct fsync_node_entry *fn;
308 fn = f2fs_kmem_cache_alloc(fsync_node_entry_slab, GFP_NOFS);
312 INIT_LIST_HEAD(&fn->list);
314 spin_lock_irqsave(&sbi->fsync_node_lock, flags);
315 list_add_tail(&fn->list, &sbi->fsync_node_list);
316 fn->seq_id = sbi->fsync_seg_id++;
318 sbi->fsync_node_num++;
319 spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
324 void f2fs_del_fsync_node_entry(struct f2fs_sb_info *sbi, struct page *page)
326 struct fsync_node_entry *fn;
329 spin_lock_irqsave(&sbi->fsync_node_lock, flags);
330 list_for_each_entry(fn, &sbi->fsync_node_list, list) {
331 if (fn->page == page) {
333 sbi->fsync_node_num--;
334 spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
335 kmem_cache_free(fsync_node_entry_slab, fn);
340 spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
344 void f2fs_reset_fsync_node_info(struct f2fs_sb_info *sbi)
348 spin_lock_irqsave(&sbi->fsync_node_lock, flags);
349 sbi->fsync_seg_id = 0;
350 spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
353 int f2fs_need_dentry_mark(struct f2fs_sb_info *sbi, nid_t nid)
355 struct f2fs_nm_info *nm_i = NM_I(sbi);
359 down_read(&nm_i->nat_tree_lock);
360 e = __lookup_nat_cache(nm_i, nid);
362 if (!get_nat_flag(e, IS_CHECKPOINTED) &&
363 !get_nat_flag(e, HAS_FSYNCED_INODE))
366 up_read(&nm_i->nat_tree_lock);
370 bool f2fs_is_checkpointed_node(struct f2fs_sb_info *sbi, nid_t nid)
372 struct f2fs_nm_info *nm_i = NM_I(sbi);
376 down_read(&nm_i->nat_tree_lock);
377 e = __lookup_nat_cache(nm_i, nid);
378 if (e && !get_nat_flag(e, IS_CHECKPOINTED))
380 up_read(&nm_i->nat_tree_lock);
384 bool f2fs_need_inode_block_update(struct f2fs_sb_info *sbi, nid_t ino)
386 struct f2fs_nm_info *nm_i = NM_I(sbi);
388 bool need_update = true;
390 down_read(&nm_i->nat_tree_lock);
391 e = __lookup_nat_cache(nm_i, ino);
392 if (e && get_nat_flag(e, HAS_LAST_FSYNC) &&
393 (get_nat_flag(e, IS_CHECKPOINTED) ||
394 get_nat_flag(e, HAS_FSYNCED_INODE)))
396 up_read(&nm_i->nat_tree_lock);
400 /* must be locked by nat_tree_lock */
401 static void cache_nat_entry(struct f2fs_sb_info *sbi, nid_t nid,
402 struct f2fs_nat_entry *ne)
404 struct f2fs_nm_info *nm_i = NM_I(sbi);
405 struct nat_entry *new, *e;
407 new = __alloc_nat_entry(nid, false);
411 down_write(&nm_i->nat_tree_lock);
412 e = __lookup_nat_cache(nm_i, nid);
414 e = __init_nat_entry(nm_i, new, ne, false);
416 f2fs_bug_on(sbi, nat_get_ino(e) != le32_to_cpu(ne->ino) ||
417 nat_get_blkaddr(e) !=
418 le32_to_cpu(ne->block_addr) ||
419 nat_get_version(e) != ne->version);
420 up_write(&nm_i->nat_tree_lock);
422 __free_nat_entry(new);
425 static void set_node_addr(struct f2fs_sb_info *sbi, struct node_info *ni,
426 block_t new_blkaddr, bool fsync_done)
428 struct f2fs_nm_info *nm_i = NM_I(sbi);
430 struct nat_entry *new = __alloc_nat_entry(ni->nid, true);
432 down_write(&nm_i->nat_tree_lock);
433 e = __lookup_nat_cache(nm_i, ni->nid);
435 e = __init_nat_entry(nm_i, new, NULL, true);
436 copy_node_info(&e->ni, ni);
437 f2fs_bug_on(sbi, ni->blk_addr == NEW_ADDR);
438 } else if (new_blkaddr == NEW_ADDR) {
440 * when nid is reallocated,
441 * previous nat entry can be remained in nat cache.
442 * So, reinitialize it with new information.
444 copy_node_info(&e->ni, ni);
445 f2fs_bug_on(sbi, ni->blk_addr != NULL_ADDR);
447 /* let's free early to reduce memory consumption */
449 __free_nat_entry(new);
452 f2fs_bug_on(sbi, nat_get_blkaddr(e) != ni->blk_addr);
453 f2fs_bug_on(sbi, nat_get_blkaddr(e) == NULL_ADDR &&
454 new_blkaddr == NULL_ADDR);
455 f2fs_bug_on(sbi, nat_get_blkaddr(e) == NEW_ADDR &&
456 new_blkaddr == NEW_ADDR);
457 f2fs_bug_on(sbi, is_valid_data_blkaddr(sbi, nat_get_blkaddr(e)) &&
458 new_blkaddr == NEW_ADDR);
460 /* increment version no as node is removed */
461 if (nat_get_blkaddr(e) != NEW_ADDR && new_blkaddr == NULL_ADDR) {
462 unsigned char version = nat_get_version(e);
463 nat_set_version(e, inc_node_version(version));
467 nat_set_blkaddr(e, new_blkaddr);
468 if (!is_valid_data_blkaddr(sbi, new_blkaddr))
469 set_nat_flag(e, IS_CHECKPOINTED, false);
470 __set_nat_cache_dirty(nm_i, e);
472 /* update fsync_mark if its inode nat entry is still alive */
473 if (ni->nid != ni->ino)
474 e = __lookup_nat_cache(nm_i, ni->ino);
476 if (fsync_done && ni->nid == ni->ino)
477 set_nat_flag(e, HAS_FSYNCED_INODE, true);
478 set_nat_flag(e, HAS_LAST_FSYNC, fsync_done);
480 up_write(&nm_i->nat_tree_lock);
483 int f2fs_try_to_free_nats(struct f2fs_sb_info *sbi, int nr_shrink)
485 struct f2fs_nm_info *nm_i = NM_I(sbi);
488 if (!down_write_trylock(&nm_i->nat_tree_lock))
491 spin_lock(&nm_i->nat_list_lock);
493 struct nat_entry *ne;
495 if (list_empty(&nm_i->nat_entries))
498 ne = list_first_entry(&nm_i->nat_entries,
499 struct nat_entry, list);
501 spin_unlock(&nm_i->nat_list_lock);
503 __del_from_nat_cache(nm_i, ne);
506 spin_lock(&nm_i->nat_list_lock);
508 spin_unlock(&nm_i->nat_list_lock);
510 up_write(&nm_i->nat_tree_lock);
511 return nr - nr_shrink;
515 * This function always returns success
517 int f2fs_get_node_info(struct f2fs_sb_info *sbi, nid_t nid,
518 struct node_info *ni)
520 struct f2fs_nm_info *nm_i = NM_I(sbi);
521 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
522 struct f2fs_journal *journal = curseg->journal;
523 nid_t start_nid = START_NID(nid);
524 struct f2fs_nat_block *nat_blk;
525 struct page *page = NULL;
526 struct f2fs_nat_entry ne;
533 /* Check nat cache */
534 down_read(&nm_i->nat_tree_lock);
535 e = __lookup_nat_cache(nm_i, nid);
537 ni->ino = nat_get_ino(e);
538 ni->blk_addr = nat_get_blkaddr(e);
539 ni->version = nat_get_version(e);
540 up_read(&nm_i->nat_tree_lock);
544 memset(&ne, 0, sizeof(struct f2fs_nat_entry));
546 /* Check current segment summary */
547 down_read(&curseg->journal_rwsem);
548 i = f2fs_lookup_journal_in_cursum(journal, NAT_JOURNAL, nid, 0);
550 ne = nat_in_journal(journal, i);
551 node_info_from_raw_nat(ni, &ne);
553 up_read(&curseg->journal_rwsem);
555 up_read(&nm_i->nat_tree_lock);
559 /* Fill node_info from nat page */
560 index = current_nat_addr(sbi, nid);
561 up_read(&nm_i->nat_tree_lock);
563 page = f2fs_get_meta_page(sbi, index);
565 return PTR_ERR(page);
567 nat_blk = (struct f2fs_nat_block *)page_address(page);
568 ne = nat_blk->entries[nid - start_nid];
569 node_info_from_raw_nat(ni, &ne);
570 f2fs_put_page(page, 1);
572 /* cache nat entry */
573 cache_nat_entry(sbi, nid, &ne);
578 * readahead MAX_RA_NODE number of node pages.
580 static void f2fs_ra_node_pages(struct page *parent, int start, int n)
582 struct f2fs_sb_info *sbi = F2FS_P_SB(parent);
583 struct blk_plug plug;
587 blk_start_plug(&plug);
589 /* Then, try readahead for siblings of the desired node */
591 end = min(end, NIDS_PER_BLOCK);
592 for (i = start; i < end; i++) {
593 nid = get_nid(parent, i, false);
594 f2fs_ra_node_page(sbi, nid);
597 blk_finish_plug(&plug);
600 pgoff_t f2fs_get_next_page_offset(struct dnode_of_data *dn, pgoff_t pgofs)
602 const long direct_index = ADDRS_PER_INODE(dn->inode);
603 const long direct_blks = ADDRS_PER_BLOCK;
604 const long indirect_blks = ADDRS_PER_BLOCK * NIDS_PER_BLOCK;
605 unsigned int skipped_unit = ADDRS_PER_BLOCK;
606 int cur_level = dn->cur_level;
607 int max_level = dn->max_level;
613 while (max_level-- > cur_level)
614 skipped_unit *= NIDS_PER_BLOCK;
616 switch (dn->max_level) {
618 base += 2 * indirect_blks;
621 base += 2 * direct_blks;
624 base += direct_index;
627 f2fs_bug_on(F2FS_I_SB(dn->inode), 1);
630 return ((pgofs - base) / skipped_unit + 1) * skipped_unit + base;
634 * The maximum depth is four.
635 * Offset[0] will have raw inode offset.
637 static int get_node_path(struct inode *inode, long block,
638 int offset[4], unsigned int noffset[4])
640 const long direct_index = ADDRS_PER_INODE(inode);
641 const long direct_blks = ADDRS_PER_BLOCK;
642 const long dptrs_per_blk = NIDS_PER_BLOCK;
643 const long indirect_blks = ADDRS_PER_BLOCK * NIDS_PER_BLOCK;
644 const long dindirect_blks = indirect_blks * NIDS_PER_BLOCK;
650 if (block < direct_index) {
654 block -= direct_index;
655 if (block < direct_blks) {
656 offset[n++] = NODE_DIR1_BLOCK;
662 block -= direct_blks;
663 if (block < direct_blks) {
664 offset[n++] = NODE_DIR2_BLOCK;
670 block -= direct_blks;
671 if (block < indirect_blks) {
672 offset[n++] = NODE_IND1_BLOCK;
674 offset[n++] = block / direct_blks;
675 noffset[n] = 4 + offset[n - 1];
676 offset[n] = block % direct_blks;
680 block -= indirect_blks;
681 if (block < indirect_blks) {
682 offset[n++] = NODE_IND2_BLOCK;
683 noffset[n] = 4 + dptrs_per_blk;
684 offset[n++] = block / direct_blks;
685 noffset[n] = 5 + dptrs_per_blk + offset[n - 1];
686 offset[n] = block % direct_blks;
690 block -= indirect_blks;
691 if (block < dindirect_blks) {
692 offset[n++] = NODE_DIND_BLOCK;
693 noffset[n] = 5 + (dptrs_per_blk * 2);
694 offset[n++] = block / indirect_blks;
695 noffset[n] = 6 + (dptrs_per_blk * 2) +
696 offset[n - 1] * (dptrs_per_blk + 1);
697 offset[n++] = (block / direct_blks) % dptrs_per_blk;
698 noffset[n] = 7 + (dptrs_per_blk * 2) +
699 offset[n - 2] * (dptrs_per_blk + 1) +
701 offset[n] = block % direct_blks;
712 * Caller should call f2fs_put_dnode(dn).
713 * Also, it should grab and release a rwsem by calling f2fs_lock_op() and
714 * f2fs_unlock_op() only if ro is not set RDONLY_NODE.
715 * In the case of RDONLY_NODE, we don't need to care about mutex.
717 int f2fs_get_dnode_of_data(struct dnode_of_data *dn, pgoff_t index, int mode)
719 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
720 struct page *npage[4];
721 struct page *parent = NULL;
723 unsigned int noffset[4];
728 level = get_node_path(dn->inode, index, offset, noffset);
732 nids[0] = dn->inode->i_ino;
733 npage[0] = dn->inode_page;
736 npage[0] = f2fs_get_node_page(sbi, nids[0]);
737 if (IS_ERR(npage[0]))
738 return PTR_ERR(npage[0]);
741 /* if inline_data is set, should not report any block indices */
742 if (f2fs_has_inline_data(dn->inode) && index) {
744 f2fs_put_page(npage[0], 1);
750 nids[1] = get_nid(parent, offset[0], true);
751 dn->inode_page = npage[0];
752 dn->inode_page_locked = true;
754 /* get indirect or direct nodes */
755 for (i = 1; i <= level; i++) {
758 if (!nids[i] && mode == ALLOC_NODE) {
760 if (!f2fs_alloc_nid(sbi, &(nids[i]))) {
766 npage[i] = f2fs_new_node_page(dn, noffset[i]);
767 if (IS_ERR(npage[i])) {
768 f2fs_alloc_nid_failed(sbi, nids[i]);
769 err = PTR_ERR(npage[i]);
773 set_nid(parent, offset[i - 1], nids[i], i == 1);
774 f2fs_alloc_nid_done(sbi, nids[i]);
776 } else if (mode == LOOKUP_NODE_RA && i == level && level > 1) {
777 npage[i] = f2fs_get_node_page_ra(parent, offset[i - 1]);
778 if (IS_ERR(npage[i])) {
779 err = PTR_ERR(npage[i]);
785 dn->inode_page_locked = false;
788 f2fs_put_page(parent, 1);
792 npage[i] = f2fs_get_node_page(sbi, nids[i]);
793 if (IS_ERR(npage[i])) {
794 err = PTR_ERR(npage[i]);
795 f2fs_put_page(npage[0], 0);
801 nids[i + 1] = get_nid(parent, offset[i], false);
804 dn->nid = nids[level];
805 dn->ofs_in_node = offset[level];
806 dn->node_page = npage[level];
807 dn->data_blkaddr = datablock_addr(dn->inode,
808 dn->node_page, dn->ofs_in_node);
812 f2fs_put_page(parent, 1);
814 f2fs_put_page(npage[0], 0);
816 dn->inode_page = NULL;
817 dn->node_page = NULL;
818 if (err == -ENOENT) {
820 dn->max_level = level;
821 dn->ofs_in_node = offset[level];
826 static int truncate_node(struct dnode_of_data *dn)
828 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
833 err = f2fs_get_node_info(sbi, dn->nid, &ni);
837 /* Deallocate node address */
838 f2fs_invalidate_blocks(sbi, ni.blk_addr);
839 dec_valid_node_count(sbi, dn->inode, dn->nid == dn->inode->i_ino);
840 set_node_addr(sbi, &ni, NULL_ADDR, false);
842 if (dn->nid == dn->inode->i_ino) {
843 f2fs_remove_orphan_inode(sbi, dn->nid);
844 dec_valid_inode_count(sbi);
845 f2fs_inode_synced(dn->inode);
848 clear_node_page_dirty(dn->node_page);
849 set_sbi_flag(sbi, SBI_IS_DIRTY);
851 index = dn->node_page->index;
852 f2fs_put_page(dn->node_page, 1);
854 invalidate_mapping_pages(NODE_MAPPING(sbi),
857 dn->node_page = NULL;
858 trace_f2fs_truncate_node(dn->inode, dn->nid, ni.blk_addr);
863 static int truncate_dnode(struct dnode_of_data *dn)
871 /* get direct node */
872 page = f2fs_get_node_page(F2FS_I_SB(dn->inode), dn->nid);
873 if (IS_ERR(page) && PTR_ERR(page) == -ENOENT)
875 else if (IS_ERR(page))
876 return PTR_ERR(page);
878 /* Make dnode_of_data for parameter */
879 dn->node_page = page;
881 f2fs_truncate_data_blocks(dn);
882 err = truncate_node(dn);
889 static int truncate_nodes(struct dnode_of_data *dn, unsigned int nofs,
892 struct dnode_of_data rdn = *dn;
894 struct f2fs_node *rn;
896 unsigned int child_nofs;
901 return NIDS_PER_BLOCK + 1;
903 trace_f2fs_truncate_nodes_enter(dn->inode, dn->nid, dn->data_blkaddr);
905 page = f2fs_get_node_page(F2FS_I_SB(dn->inode), dn->nid);
907 trace_f2fs_truncate_nodes_exit(dn->inode, PTR_ERR(page));
908 return PTR_ERR(page);
911 f2fs_ra_node_pages(page, ofs, NIDS_PER_BLOCK);
913 rn = F2FS_NODE(page);
915 for (i = ofs; i < NIDS_PER_BLOCK; i++, freed++) {
916 child_nid = le32_to_cpu(rn->in.nid[i]);
920 ret = truncate_dnode(&rdn);
923 if (set_nid(page, i, 0, false))
924 dn->node_changed = true;
927 child_nofs = nofs + ofs * (NIDS_PER_BLOCK + 1) + 1;
928 for (i = ofs; i < NIDS_PER_BLOCK; i++) {
929 child_nid = le32_to_cpu(rn->in.nid[i]);
930 if (child_nid == 0) {
931 child_nofs += NIDS_PER_BLOCK + 1;
935 ret = truncate_nodes(&rdn, child_nofs, 0, depth - 1);
936 if (ret == (NIDS_PER_BLOCK + 1)) {
937 if (set_nid(page, i, 0, false))
938 dn->node_changed = true;
940 } else if (ret < 0 && ret != -ENOENT) {
948 /* remove current indirect node */
949 dn->node_page = page;
950 ret = truncate_node(dn);
955 f2fs_put_page(page, 1);
957 trace_f2fs_truncate_nodes_exit(dn->inode, freed);
961 f2fs_put_page(page, 1);
962 trace_f2fs_truncate_nodes_exit(dn->inode, ret);
966 static int truncate_partial_nodes(struct dnode_of_data *dn,
967 struct f2fs_inode *ri, int *offset, int depth)
969 struct page *pages[2];
976 nid[0] = le32_to_cpu(ri->i_nid[offset[0] - NODE_DIR1_BLOCK]);
980 /* get indirect nodes in the path */
981 for (i = 0; i < idx + 1; i++) {
982 /* reference count'll be increased */
983 pages[i] = f2fs_get_node_page(F2FS_I_SB(dn->inode), nid[i]);
984 if (IS_ERR(pages[i])) {
985 err = PTR_ERR(pages[i]);
989 nid[i + 1] = get_nid(pages[i], offset[i + 1], false);
992 f2fs_ra_node_pages(pages[idx], offset[idx + 1], NIDS_PER_BLOCK);
994 /* free direct nodes linked to a partial indirect node */
995 for (i = offset[idx + 1]; i < NIDS_PER_BLOCK; i++) {
996 child_nid = get_nid(pages[idx], i, false);
1000 err = truncate_dnode(dn);
1003 if (set_nid(pages[idx], i, 0, false))
1004 dn->node_changed = true;
1007 if (offset[idx + 1] == 0) {
1008 dn->node_page = pages[idx];
1010 err = truncate_node(dn);
1014 f2fs_put_page(pages[idx], 1);
1017 offset[idx + 1] = 0;
1020 for (i = idx; i >= 0; i--)
1021 f2fs_put_page(pages[i], 1);
1023 trace_f2fs_truncate_partial_nodes(dn->inode, nid, depth, err);
1029 * All the block addresses of data and nodes should be nullified.
1031 int f2fs_truncate_inode_blocks(struct inode *inode, pgoff_t from)
1033 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1034 int err = 0, cont = 1;
1035 int level, offset[4], noffset[4];
1036 unsigned int nofs = 0;
1037 struct f2fs_inode *ri;
1038 struct dnode_of_data dn;
1041 trace_f2fs_truncate_inode_blocks_enter(inode, from);
1043 level = get_node_path(inode, from, offset, noffset);
1047 page = f2fs_get_node_page(sbi, inode->i_ino);
1049 trace_f2fs_truncate_inode_blocks_exit(inode, PTR_ERR(page));
1050 return PTR_ERR(page);
1053 set_new_dnode(&dn, inode, page, NULL, 0);
1056 ri = F2FS_INODE(page);
1064 if (!offset[level - 1])
1066 err = truncate_partial_nodes(&dn, ri, offset, level);
1067 if (err < 0 && err != -ENOENT)
1069 nofs += 1 + NIDS_PER_BLOCK;
1072 nofs = 5 + 2 * NIDS_PER_BLOCK;
1073 if (!offset[level - 1])
1075 err = truncate_partial_nodes(&dn, ri, offset, level);
1076 if (err < 0 && err != -ENOENT)
1085 dn.nid = le32_to_cpu(ri->i_nid[offset[0] - NODE_DIR1_BLOCK]);
1086 switch (offset[0]) {
1087 case NODE_DIR1_BLOCK:
1088 case NODE_DIR2_BLOCK:
1089 err = truncate_dnode(&dn);
1092 case NODE_IND1_BLOCK:
1093 case NODE_IND2_BLOCK:
1094 err = truncate_nodes(&dn, nofs, offset[1], 2);
1097 case NODE_DIND_BLOCK:
1098 err = truncate_nodes(&dn, nofs, offset[1], 3);
1105 if (err < 0 && err != -ENOENT)
1107 if (offset[1] == 0 &&
1108 ri->i_nid[offset[0] - NODE_DIR1_BLOCK]) {
1110 BUG_ON(page->mapping != NODE_MAPPING(sbi));
1111 f2fs_wait_on_page_writeback(page, NODE, true, true);
1112 ri->i_nid[offset[0] - NODE_DIR1_BLOCK] = 0;
1113 set_page_dirty(page);
1121 f2fs_put_page(page, 0);
1122 trace_f2fs_truncate_inode_blocks_exit(inode, err);
1123 return err > 0 ? 0 : err;
1126 /* caller must lock inode page */
1127 int f2fs_truncate_xattr_node(struct inode *inode)
1129 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1130 nid_t nid = F2FS_I(inode)->i_xattr_nid;
1131 struct dnode_of_data dn;
1138 npage = f2fs_get_node_page(sbi, nid);
1140 return PTR_ERR(npage);
1142 set_new_dnode(&dn, inode, NULL, npage, nid);
1143 err = truncate_node(&dn);
1145 f2fs_put_page(npage, 1);
1149 f2fs_i_xnid_write(inode, 0);
1155 * Caller should grab and release a rwsem by calling f2fs_lock_op() and
1158 int f2fs_remove_inode_page(struct inode *inode)
1160 struct dnode_of_data dn;
1163 set_new_dnode(&dn, inode, NULL, NULL, inode->i_ino);
1164 err = f2fs_get_dnode_of_data(&dn, 0, LOOKUP_NODE);
1168 err = f2fs_truncate_xattr_node(inode);
1170 f2fs_put_dnode(&dn);
1174 /* remove potential inline_data blocks */
1175 if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
1176 S_ISLNK(inode->i_mode))
1177 f2fs_truncate_data_blocks_range(&dn, 1);
1179 /* 0 is possible, after f2fs_new_inode() has failed */
1180 if (unlikely(f2fs_cp_error(F2FS_I_SB(inode)))) {
1181 f2fs_put_dnode(&dn);
1184 f2fs_bug_on(F2FS_I_SB(inode),
1185 inode->i_blocks != 0 && inode->i_blocks != 8);
1187 /* will put inode & node pages */
1188 err = truncate_node(&dn);
1190 f2fs_put_dnode(&dn);
1196 struct page *f2fs_new_inode_page(struct inode *inode)
1198 struct dnode_of_data dn;
1200 /* allocate inode page for new inode */
1201 set_new_dnode(&dn, inode, NULL, NULL, inode->i_ino);
1203 /* caller should f2fs_put_page(page, 1); */
1204 return f2fs_new_node_page(&dn, 0);
1207 struct page *f2fs_new_node_page(struct dnode_of_data *dn, unsigned int ofs)
1209 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
1210 struct node_info new_ni;
1214 if (unlikely(is_inode_flag_set(dn->inode, FI_NO_ALLOC)))
1215 return ERR_PTR(-EPERM);
1217 page = f2fs_grab_cache_page(NODE_MAPPING(sbi), dn->nid, false);
1219 return ERR_PTR(-ENOMEM);
1221 if (unlikely((err = inc_valid_node_count(sbi, dn->inode, !ofs))))
1224 #ifdef CONFIG_F2FS_CHECK_FS
1225 err = f2fs_get_node_info(sbi, dn->nid, &new_ni);
1227 dec_valid_node_count(sbi, dn->inode, !ofs);
1230 f2fs_bug_on(sbi, new_ni.blk_addr != NULL_ADDR);
1232 new_ni.nid = dn->nid;
1233 new_ni.ino = dn->inode->i_ino;
1234 new_ni.blk_addr = NULL_ADDR;
1237 set_node_addr(sbi, &new_ni, NEW_ADDR, false);
1239 f2fs_wait_on_page_writeback(page, NODE, true, true);
1240 fill_node_footer(page, dn->nid, dn->inode->i_ino, ofs, true);
1241 set_cold_node(page, S_ISDIR(dn->inode->i_mode));
1242 if (!PageUptodate(page))
1243 SetPageUptodate(page);
1244 if (set_page_dirty(page))
1245 dn->node_changed = true;
1247 if (f2fs_has_xattr_block(ofs))
1248 f2fs_i_xnid_write(dn->inode, dn->nid);
1251 inc_valid_inode_count(sbi);
1255 clear_node_page_dirty(page);
1256 f2fs_put_page(page, 1);
1257 return ERR_PTR(err);
1261 * Caller should do after getting the following values.
1262 * 0: f2fs_put_page(page, 0)
1263 * LOCKED_PAGE or error: f2fs_put_page(page, 1)
1265 static int read_node_page(struct page *page, int op_flags)
1267 struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1268 struct node_info ni;
1269 struct f2fs_io_info fio = {
1273 .op_flags = op_flags,
1275 .encrypted_page = NULL,
1279 if (PageUptodate(page)) {
1280 #ifdef CONFIG_F2FS_CHECK_FS
1281 f2fs_bug_on(sbi, !f2fs_inode_chksum_verify(sbi, page));
1286 err = f2fs_get_node_info(sbi, page->index, &ni);
1290 if (unlikely(ni.blk_addr == NULL_ADDR) ||
1291 is_sbi_flag_set(sbi, SBI_IS_SHUTDOWN)) {
1292 ClearPageUptodate(page);
1296 fio.new_blkaddr = fio.old_blkaddr = ni.blk_addr;
1297 return f2fs_submit_page_bio(&fio);
1301 * Readahead a node page
1303 void f2fs_ra_node_page(struct f2fs_sb_info *sbi, nid_t nid)
1310 if (f2fs_check_nid_range(sbi, nid))
1313 apage = xa_load(&NODE_MAPPING(sbi)->i_pages, nid);
1317 apage = f2fs_grab_cache_page(NODE_MAPPING(sbi), nid, false);
1321 err = read_node_page(apage, REQ_RAHEAD);
1322 f2fs_put_page(apage, err ? 1 : 0);
1325 static struct page *__get_node_page(struct f2fs_sb_info *sbi, pgoff_t nid,
1326 struct page *parent, int start)
1332 return ERR_PTR(-ENOENT);
1333 if (f2fs_check_nid_range(sbi, nid))
1334 return ERR_PTR(-EINVAL);
1336 page = f2fs_grab_cache_page(NODE_MAPPING(sbi), nid, false);
1338 return ERR_PTR(-ENOMEM);
1340 err = read_node_page(page, 0);
1342 f2fs_put_page(page, 1);
1343 return ERR_PTR(err);
1344 } else if (err == LOCKED_PAGE) {
1350 f2fs_ra_node_pages(parent, start + 1, MAX_RA_NODE);
1354 if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1355 f2fs_put_page(page, 1);
1359 if (unlikely(!PageUptodate(page))) {
1364 if (!f2fs_inode_chksum_verify(sbi, page)) {
1369 if(unlikely(nid != nid_of_node(page))) {
1370 f2fs_msg(sbi->sb, KERN_WARNING, "inconsistent node block, "
1371 "nid:%lu, node_footer[nid:%u,ino:%u,ofs:%u,cpver:%llu,blkaddr:%u]",
1372 nid, nid_of_node(page), ino_of_node(page),
1373 ofs_of_node(page), cpver_of_node(page),
1374 next_blkaddr_of_node(page));
1377 ClearPageUptodate(page);
1378 f2fs_put_page(page, 1);
1379 return ERR_PTR(err);
1384 struct page *f2fs_get_node_page(struct f2fs_sb_info *sbi, pgoff_t nid)
1386 return __get_node_page(sbi, nid, NULL, 0);
1389 struct page *f2fs_get_node_page_ra(struct page *parent, int start)
1391 struct f2fs_sb_info *sbi = F2FS_P_SB(parent);
1392 nid_t nid = get_nid(parent, start, false);
1394 return __get_node_page(sbi, nid, parent, start);
1397 static void flush_inline_data(struct f2fs_sb_info *sbi, nid_t ino)
1399 struct inode *inode;
1403 /* should flush inline_data before evict_inode */
1404 inode = ilookup(sbi->sb, ino);
1408 page = f2fs_pagecache_get_page(inode->i_mapping, 0,
1409 FGP_LOCK|FGP_NOWAIT, 0);
1413 if (!PageUptodate(page))
1416 if (!PageDirty(page))
1419 if (!clear_page_dirty_for_io(page))
1422 ret = f2fs_write_inline_data(inode, page);
1423 inode_dec_dirty_pages(inode);
1424 f2fs_remove_dirty_inode(inode);
1426 set_page_dirty(page);
1428 f2fs_put_page(page, 1);
1433 static struct page *last_fsync_dnode(struct f2fs_sb_info *sbi, nid_t ino)
1436 struct pagevec pvec;
1437 struct page *last_page = NULL;
1440 pagevec_init(&pvec);
1443 while ((nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1444 PAGECACHE_TAG_DIRTY))) {
1447 for (i = 0; i < nr_pages; i++) {
1448 struct page *page = pvec.pages[i];
1450 if (unlikely(f2fs_cp_error(sbi))) {
1451 f2fs_put_page(last_page, 0);
1452 pagevec_release(&pvec);
1453 return ERR_PTR(-EIO);
1456 if (!IS_DNODE(page) || !is_cold_node(page))
1458 if (ino_of_node(page) != ino)
1463 if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1468 if (ino_of_node(page) != ino)
1469 goto continue_unlock;
1471 if (!PageDirty(page)) {
1472 /* someone wrote it for us */
1473 goto continue_unlock;
1477 f2fs_put_page(last_page, 0);
1483 pagevec_release(&pvec);
1489 static int __write_node_page(struct page *page, bool atomic, bool *submitted,
1490 struct writeback_control *wbc, bool do_balance,
1491 enum iostat_type io_type, unsigned int *seq_id)
1493 struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1495 struct node_info ni;
1496 struct f2fs_io_info fio = {
1498 .ino = ino_of_node(page),
1501 .op_flags = wbc_to_write_flags(wbc),
1503 .encrypted_page = NULL,
1510 trace_f2fs_writepage(page, NODE);
1512 if (unlikely(f2fs_cp_error(sbi)))
1515 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
1518 if (wbc->sync_mode == WB_SYNC_NONE &&
1519 IS_DNODE(page) && is_cold_node(page))
1522 /* get old block addr of this node page */
1523 nid = nid_of_node(page);
1524 f2fs_bug_on(sbi, page->index != nid);
1526 if (f2fs_get_node_info(sbi, nid, &ni))
1529 if (wbc->for_reclaim) {
1530 if (!down_read_trylock(&sbi->node_write))
1533 down_read(&sbi->node_write);
1536 /* This page is already truncated */
1537 if (unlikely(ni.blk_addr == NULL_ADDR)) {
1538 ClearPageUptodate(page);
1539 dec_page_count(sbi, F2FS_DIRTY_NODES);
1540 up_read(&sbi->node_write);
1545 if (__is_valid_data_blkaddr(ni.blk_addr) &&
1546 !f2fs_is_valid_blkaddr(sbi, ni.blk_addr, DATA_GENERIC)) {
1547 up_read(&sbi->node_write);
1551 if (atomic && !test_opt(sbi, NOBARRIER))
1552 fio.op_flags |= REQ_PREFLUSH | REQ_FUA;
1554 set_page_writeback(page);
1555 ClearPageError(page);
1557 if (f2fs_in_warm_node_list(sbi, page)) {
1558 seq = f2fs_add_fsync_node_entry(sbi, page);
1563 fio.old_blkaddr = ni.blk_addr;
1564 f2fs_do_write_node_page(nid, &fio);
1565 set_node_addr(sbi, &ni, fio.new_blkaddr, is_fsync_dnode(page));
1566 dec_page_count(sbi, F2FS_DIRTY_NODES);
1567 up_read(&sbi->node_write);
1569 if (wbc->for_reclaim) {
1570 f2fs_submit_merged_write_cond(sbi, NULL, page, 0, NODE);
1576 if (unlikely(f2fs_cp_error(sbi))) {
1577 f2fs_submit_merged_write(sbi, NODE);
1581 *submitted = fio.submitted;
1584 f2fs_balance_fs(sbi, false);
1588 redirty_page_for_writepage(wbc, page);
1589 return AOP_WRITEPAGE_ACTIVATE;
1592 int f2fs_move_node_page(struct page *node_page, int gc_type)
1596 if (gc_type == FG_GC) {
1597 struct writeback_control wbc = {
1598 .sync_mode = WB_SYNC_ALL,
1603 f2fs_wait_on_page_writeback(node_page, NODE, true, true);
1605 set_page_dirty(node_page);
1607 if (!clear_page_dirty_for_io(node_page)) {
1612 if (__write_node_page(node_page, false, NULL,
1613 &wbc, false, FS_GC_NODE_IO, NULL)) {
1615 unlock_page(node_page);
1619 /* set page dirty and write it */
1620 if (!PageWriteback(node_page))
1621 set_page_dirty(node_page);
1624 unlock_page(node_page);
1626 f2fs_put_page(node_page, 0);
1630 static int f2fs_write_node_page(struct page *page,
1631 struct writeback_control *wbc)
1633 return __write_node_page(page, false, NULL, wbc, false,
1637 int f2fs_fsync_node_pages(struct f2fs_sb_info *sbi, struct inode *inode,
1638 struct writeback_control *wbc, bool atomic,
1639 unsigned int *seq_id)
1642 struct pagevec pvec;
1644 struct page *last_page = NULL;
1645 bool marked = false;
1646 nid_t ino = inode->i_ino;
1651 last_page = last_fsync_dnode(sbi, ino);
1652 if (IS_ERR_OR_NULL(last_page))
1653 return PTR_ERR_OR_ZERO(last_page);
1656 pagevec_init(&pvec);
1659 while ((nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1660 PAGECACHE_TAG_DIRTY))) {
1663 for (i = 0; i < nr_pages; i++) {
1664 struct page *page = pvec.pages[i];
1665 bool submitted = false;
1667 if (unlikely(f2fs_cp_error(sbi))) {
1668 f2fs_put_page(last_page, 0);
1669 pagevec_release(&pvec);
1674 if (!IS_DNODE(page) || !is_cold_node(page))
1676 if (ino_of_node(page) != ino)
1681 if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1686 if (ino_of_node(page) != ino)
1687 goto continue_unlock;
1689 if (!PageDirty(page) && page != last_page) {
1690 /* someone wrote it for us */
1691 goto continue_unlock;
1694 f2fs_wait_on_page_writeback(page, NODE, true, true);
1696 set_fsync_mark(page, 0);
1697 set_dentry_mark(page, 0);
1699 if (!atomic || page == last_page) {
1700 set_fsync_mark(page, 1);
1701 if (IS_INODE(page)) {
1702 if (is_inode_flag_set(inode,
1704 f2fs_update_inode(inode, page);
1705 set_dentry_mark(page,
1706 f2fs_need_dentry_mark(sbi, ino));
1708 /* may be written by other thread */
1709 if (!PageDirty(page))
1710 set_page_dirty(page);
1713 if (!clear_page_dirty_for_io(page))
1714 goto continue_unlock;
1716 ret = __write_node_page(page, atomic &&
1718 &submitted, wbc, true,
1719 FS_NODE_IO, seq_id);
1722 f2fs_put_page(last_page, 0);
1724 } else if (submitted) {
1728 if (page == last_page) {
1729 f2fs_put_page(page, 0);
1734 pagevec_release(&pvec);
1740 if (!ret && atomic && !marked) {
1741 f2fs_msg(sbi->sb, KERN_DEBUG,
1742 "Retry to write fsync mark: ino=%u, idx=%lx",
1743 ino, last_page->index);
1744 lock_page(last_page);
1745 f2fs_wait_on_page_writeback(last_page, NODE, true, true);
1746 set_page_dirty(last_page);
1747 unlock_page(last_page);
1752 f2fs_submit_merged_write_cond(sbi, NULL, NULL, ino, NODE);
1753 return ret ? -EIO: 0;
1756 int f2fs_sync_node_pages(struct f2fs_sb_info *sbi,
1757 struct writeback_control *wbc,
1758 bool do_balance, enum iostat_type io_type)
1761 struct pagevec pvec;
1765 int nr_pages, done = 0;
1767 pagevec_init(&pvec);
1772 while (!done && (nr_pages = pagevec_lookup_tag(&pvec,
1773 NODE_MAPPING(sbi), &index, PAGECACHE_TAG_DIRTY))) {
1776 for (i = 0; i < nr_pages; i++) {
1777 struct page *page = pvec.pages[i];
1778 bool submitted = false;
1780 /* give a priority to WB_SYNC threads */
1781 if (atomic_read(&sbi->wb_sync_req[NODE]) &&
1782 wbc->sync_mode == WB_SYNC_NONE) {
1788 * flushing sequence with step:
1793 if (step == 0 && IS_DNODE(page))
1795 if (step == 1 && (!IS_DNODE(page) ||
1796 is_cold_node(page)))
1798 if (step == 2 && (!IS_DNODE(page) ||
1799 !is_cold_node(page)))
1802 if (wbc->sync_mode == WB_SYNC_ALL)
1804 else if (!trylock_page(page))
1807 if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1813 if (!PageDirty(page)) {
1814 /* someone wrote it for us */
1815 goto continue_unlock;
1818 /* flush inline_data */
1819 if (is_inline_node(page)) {
1820 clear_inline_node(page);
1822 flush_inline_data(sbi, ino_of_node(page));
1826 f2fs_wait_on_page_writeback(page, NODE, true, true);
1828 if (!clear_page_dirty_for_io(page))
1829 goto continue_unlock;
1831 set_fsync_mark(page, 0);
1832 set_dentry_mark(page, 0);
1834 ret = __write_node_page(page, false, &submitted,
1835 wbc, do_balance, io_type, NULL);
1841 if (--wbc->nr_to_write == 0)
1844 pagevec_release(&pvec);
1847 if (wbc->nr_to_write == 0) {
1854 if (wbc->sync_mode == WB_SYNC_NONE && step == 1)
1861 f2fs_submit_merged_write(sbi, NODE);
1863 if (unlikely(f2fs_cp_error(sbi)))
1868 int f2fs_wait_on_node_pages_writeback(struct f2fs_sb_info *sbi,
1869 unsigned int seq_id)
1871 struct fsync_node_entry *fn;
1873 struct list_head *head = &sbi->fsync_node_list;
1874 unsigned long flags;
1875 unsigned int cur_seq_id = 0;
1878 while (seq_id && cur_seq_id < seq_id) {
1879 spin_lock_irqsave(&sbi->fsync_node_lock, flags);
1880 if (list_empty(head)) {
1881 spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
1884 fn = list_first_entry(head, struct fsync_node_entry, list);
1885 if (fn->seq_id > seq_id) {
1886 spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
1889 cur_seq_id = fn->seq_id;
1892 spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
1894 f2fs_wait_on_page_writeback(page, NODE, true, false);
1895 if (TestClearPageError(page))
1904 ret2 = filemap_check_errors(NODE_MAPPING(sbi));
1911 static int f2fs_write_node_pages(struct address_space *mapping,
1912 struct writeback_control *wbc)
1914 struct f2fs_sb_info *sbi = F2FS_M_SB(mapping);
1915 struct blk_plug plug;
1918 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
1921 /* balancing f2fs's metadata in background */
1922 f2fs_balance_fs_bg(sbi);
1924 /* collect a number of dirty node pages and write together */
1925 if (wbc->sync_mode != WB_SYNC_ALL &&
1926 get_pages(sbi, F2FS_DIRTY_NODES) <
1927 nr_pages_to_skip(sbi, NODE))
1930 if (wbc->sync_mode == WB_SYNC_ALL)
1931 atomic_inc(&sbi->wb_sync_req[NODE]);
1932 else if (atomic_read(&sbi->wb_sync_req[NODE]))
1935 trace_f2fs_writepages(mapping->host, wbc, NODE);
1937 diff = nr_pages_to_write(sbi, NODE, wbc);
1938 blk_start_plug(&plug);
1939 f2fs_sync_node_pages(sbi, wbc, true, FS_NODE_IO);
1940 blk_finish_plug(&plug);
1941 wbc->nr_to_write = max((long)0, wbc->nr_to_write - diff);
1943 if (wbc->sync_mode == WB_SYNC_ALL)
1944 atomic_dec(&sbi->wb_sync_req[NODE]);
1948 wbc->pages_skipped += get_pages(sbi, F2FS_DIRTY_NODES);
1949 trace_f2fs_writepages(mapping->host, wbc, NODE);
1953 static int f2fs_set_node_page_dirty(struct page *page)
1955 trace_f2fs_set_page_dirty(page, NODE);
1957 if (!PageUptodate(page))
1958 SetPageUptodate(page);
1959 #ifdef CONFIG_F2FS_CHECK_FS
1961 f2fs_inode_chksum_set(F2FS_P_SB(page), page);
1963 if (!PageDirty(page)) {
1964 __set_page_dirty_nobuffers(page);
1965 inc_page_count(F2FS_P_SB(page), F2FS_DIRTY_NODES);
1966 f2fs_set_page_private(page, 0);
1967 f2fs_trace_pid(page);
1974 * Structure of the f2fs node operations
1976 const struct address_space_operations f2fs_node_aops = {
1977 .writepage = f2fs_write_node_page,
1978 .writepages = f2fs_write_node_pages,
1979 .set_page_dirty = f2fs_set_node_page_dirty,
1980 .invalidatepage = f2fs_invalidate_page,
1981 .releasepage = f2fs_release_page,
1982 #ifdef CONFIG_MIGRATION
1983 .migratepage = f2fs_migrate_page,
1987 static struct free_nid *__lookup_free_nid_list(struct f2fs_nm_info *nm_i,
1990 return radix_tree_lookup(&nm_i->free_nid_root, n);
1993 static int __insert_free_nid(struct f2fs_sb_info *sbi,
1994 struct free_nid *i, enum nid_state state)
1996 struct f2fs_nm_info *nm_i = NM_I(sbi);
1998 int err = radix_tree_insert(&nm_i->free_nid_root, i->nid, i);
2002 f2fs_bug_on(sbi, state != i->state);
2003 nm_i->nid_cnt[state]++;
2004 if (state == FREE_NID)
2005 list_add_tail(&i->list, &nm_i->free_nid_list);
2009 static void __remove_free_nid(struct f2fs_sb_info *sbi,
2010 struct free_nid *i, enum nid_state state)
2012 struct f2fs_nm_info *nm_i = NM_I(sbi);
2014 f2fs_bug_on(sbi, state != i->state);
2015 nm_i->nid_cnt[state]--;
2016 if (state == FREE_NID)
2018 radix_tree_delete(&nm_i->free_nid_root, i->nid);
2021 static void __move_free_nid(struct f2fs_sb_info *sbi, struct free_nid *i,
2022 enum nid_state org_state, enum nid_state dst_state)
2024 struct f2fs_nm_info *nm_i = NM_I(sbi);
2026 f2fs_bug_on(sbi, org_state != i->state);
2027 i->state = dst_state;
2028 nm_i->nid_cnt[org_state]--;
2029 nm_i->nid_cnt[dst_state]++;
2031 switch (dst_state) {
2036 list_add_tail(&i->list, &nm_i->free_nid_list);
2043 static void update_free_nid_bitmap(struct f2fs_sb_info *sbi, nid_t nid,
2044 bool set, bool build)
2046 struct f2fs_nm_info *nm_i = NM_I(sbi);
2047 unsigned int nat_ofs = NAT_BLOCK_OFFSET(nid);
2048 unsigned int nid_ofs = nid - START_NID(nid);
2050 if (!test_bit_le(nat_ofs, nm_i->nat_block_bitmap))
2054 if (test_bit_le(nid_ofs, nm_i->free_nid_bitmap[nat_ofs]))
2056 __set_bit_le(nid_ofs, nm_i->free_nid_bitmap[nat_ofs]);
2057 nm_i->free_nid_count[nat_ofs]++;
2059 if (!test_bit_le(nid_ofs, nm_i->free_nid_bitmap[nat_ofs]))
2061 __clear_bit_le(nid_ofs, nm_i->free_nid_bitmap[nat_ofs]);
2063 nm_i->free_nid_count[nat_ofs]--;
2067 /* return if the nid is recognized as free */
2068 static bool add_free_nid(struct f2fs_sb_info *sbi,
2069 nid_t nid, bool build, bool update)
2071 struct f2fs_nm_info *nm_i = NM_I(sbi);
2072 struct free_nid *i, *e;
2073 struct nat_entry *ne;
2077 /* 0 nid should not be used */
2078 if (unlikely(nid == 0))
2081 i = f2fs_kmem_cache_alloc(free_nid_slab, GFP_NOFS);
2083 i->state = FREE_NID;
2085 radix_tree_preload(GFP_NOFS | __GFP_NOFAIL);
2087 spin_lock(&nm_i->nid_list_lock);
2095 * - __insert_nid_to_list(PREALLOC_NID)
2096 * - f2fs_balance_fs_bg
2097 * - f2fs_build_free_nids
2098 * - __f2fs_build_free_nids
2101 * - __lookup_nat_cache
2103 * - f2fs_init_inode_metadata
2104 * - f2fs_new_inode_page
2105 * - f2fs_new_node_page
2107 * - f2fs_alloc_nid_done
2108 * - __remove_nid_from_list(PREALLOC_NID)
2109 * - __insert_nid_to_list(FREE_NID)
2111 ne = __lookup_nat_cache(nm_i, nid);
2112 if (ne && (!get_nat_flag(ne, IS_CHECKPOINTED) ||
2113 nat_get_blkaddr(ne) != NULL_ADDR))
2116 e = __lookup_free_nid_list(nm_i, nid);
2118 if (e->state == FREE_NID)
2124 err = __insert_free_nid(sbi, i, FREE_NID);
2127 update_free_nid_bitmap(sbi, nid, ret, build);
2129 nm_i->available_nids++;
2131 spin_unlock(&nm_i->nid_list_lock);
2132 radix_tree_preload_end();
2135 kmem_cache_free(free_nid_slab, i);
2139 static void remove_free_nid(struct f2fs_sb_info *sbi, nid_t nid)
2141 struct f2fs_nm_info *nm_i = NM_I(sbi);
2143 bool need_free = false;
2145 spin_lock(&nm_i->nid_list_lock);
2146 i = __lookup_free_nid_list(nm_i, nid);
2147 if (i && i->state == FREE_NID) {
2148 __remove_free_nid(sbi, i, FREE_NID);
2151 spin_unlock(&nm_i->nid_list_lock);
2154 kmem_cache_free(free_nid_slab, i);
2157 static int scan_nat_page(struct f2fs_sb_info *sbi,
2158 struct page *nat_page, nid_t start_nid)
2160 struct f2fs_nm_info *nm_i = NM_I(sbi);
2161 struct f2fs_nat_block *nat_blk = page_address(nat_page);
2163 unsigned int nat_ofs = NAT_BLOCK_OFFSET(start_nid);
2166 __set_bit_le(nat_ofs, nm_i->nat_block_bitmap);
2168 i = start_nid % NAT_ENTRY_PER_BLOCK;
2170 for (; i < NAT_ENTRY_PER_BLOCK; i++, start_nid++) {
2171 if (unlikely(start_nid >= nm_i->max_nid))
2174 blk_addr = le32_to_cpu(nat_blk->entries[i].block_addr);
2176 if (blk_addr == NEW_ADDR)
2179 if (blk_addr == NULL_ADDR) {
2180 add_free_nid(sbi, start_nid, true, true);
2182 spin_lock(&NM_I(sbi)->nid_list_lock);
2183 update_free_nid_bitmap(sbi, start_nid, false, true);
2184 spin_unlock(&NM_I(sbi)->nid_list_lock);
2191 static void scan_curseg_cache(struct f2fs_sb_info *sbi)
2193 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
2194 struct f2fs_journal *journal = curseg->journal;
2197 down_read(&curseg->journal_rwsem);
2198 for (i = 0; i < nats_in_cursum(journal); i++) {
2202 addr = le32_to_cpu(nat_in_journal(journal, i).block_addr);
2203 nid = le32_to_cpu(nid_in_journal(journal, i));
2204 if (addr == NULL_ADDR)
2205 add_free_nid(sbi, nid, true, false);
2207 remove_free_nid(sbi, nid);
2209 up_read(&curseg->journal_rwsem);
2212 static void scan_free_nid_bits(struct f2fs_sb_info *sbi)
2214 struct f2fs_nm_info *nm_i = NM_I(sbi);
2215 unsigned int i, idx;
2218 down_read(&nm_i->nat_tree_lock);
2220 for (i = 0; i < nm_i->nat_blocks; i++) {
2221 if (!test_bit_le(i, nm_i->nat_block_bitmap))
2223 if (!nm_i->free_nid_count[i])
2225 for (idx = 0; idx < NAT_ENTRY_PER_BLOCK; idx++) {
2226 idx = find_next_bit_le(nm_i->free_nid_bitmap[i],
2227 NAT_ENTRY_PER_BLOCK, idx);
2228 if (idx >= NAT_ENTRY_PER_BLOCK)
2231 nid = i * NAT_ENTRY_PER_BLOCK + idx;
2232 add_free_nid(sbi, nid, true, false);
2234 if (nm_i->nid_cnt[FREE_NID] >= MAX_FREE_NIDS)
2239 scan_curseg_cache(sbi);
2241 up_read(&nm_i->nat_tree_lock);
2244 static int __f2fs_build_free_nids(struct f2fs_sb_info *sbi,
2245 bool sync, bool mount)
2247 struct f2fs_nm_info *nm_i = NM_I(sbi);
2249 nid_t nid = nm_i->next_scan_nid;
2251 if (unlikely(nid >= nm_i->max_nid))
2254 /* Enough entries */
2255 if (nm_i->nid_cnt[FREE_NID] >= NAT_ENTRY_PER_BLOCK)
2258 if (!sync && !f2fs_available_free_memory(sbi, FREE_NIDS))
2262 /* try to find free nids in free_nid_bitmap */
2263 scan_free_nid_bits(sbi);
2265 if (nm_i->nid_cnt[FREE_NID] >= NAT_ENTRY_PER_BLOCK)
2269 /* readahead nat pages to be scanned */
2270 f2fs_ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nid), FREE_NID_PAGES,
2273 down_read(&nm_i->nat_tree_lock);
2276 if (!test_bit_le(NAT_BLOCK_OFFSET(nid),
2277 nm_i->nat_block_bitmap)) {
2278 struct page *page = get_current_nat_page(sbi, nid);
2281 ret = PTR_ERR(page);
2283 ret = scan_nat_page(sbi, page, nid);
2284 f2fs_put_page(page, 1);
2288 up_read(&nm_i->nat_tree_lock);
2289 f2fs_bug_on(sbi, !mount);
2290 f2fs_msg(sbi->sb, KERN_ERR,
2291 "NAT is corrupt, run fsck to fix it");
2296 nid += (NAT_ENTRY_PER_BLOCK - (nid % NAT_ENTRY_PER_BLOCK));
2297 if (unlikely(nid >= nm_i->max_nid))
2300 if (++i >= FREE_NID_PAGES)
2304 /* go to the next free nat pages to find free nids abundantly */
2305 nm_i->next_scan_nid = nid;
2307 /* find free nids from current sum_pages */
2308 scan_curseg_cache(sbi);
2310 up_read(&nm_i->nat_tree_lock);
2312 f2fs_ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nm_i->next_scan_nid),
2313 nm_i->ra_nid_pages, META_NAT, false);
2318 int f2fs_build_free_nids(struct f2fs_sb_info *sbi, bool sync, bool mount)
2322 mutex_lock(&NM_I(sbi)->build_lock);
2323 ret = __f2fs_build_free_nids(sbi, sync, mount);
2324 mutex_unlock(&NM_I(sbi)->build_lock);
2330 * If this function returns success, caller can obtain a new nid
2331 * from second parameter of this function.
2332 * The returned nid could be used ino as well as nid when inode is created.
2334 bool f2fs_alloc_nid(struct f2fs_sb_info *sbi, nid_t *nid)
2336 struct f2fs_nm_info *nm_i = NM_I(sbi);
2337 struct free_nid *i = NULL;
2339 if (time_to_inject(sbi, FAULT_ALLOC_NID)) {
2340 f2fs_show_injection_info(FAULT_ALLOC_NID);
2344 spin_lock(&nm_i->nid_list_lock);
2346 if (unlikely(nm_i->available_nids == 0)) {
2347 spin_unlock(&nm_i->nid_list_lock);
2351 /* We should not use stale free nids created by f2fs_build_free_nids */
2352 if (nm_i->nid_cnt[FREE_NID] && !on_f2fs_build_free_nids(nm_i)) {
2353 f2fs_bug_on(sbi, list_empty(&nm_i->free_nid_list));
2354 i = list_first_entry(&nm_i->free_nid_list,
2355 struct free_nid, list);
2358 __move_free_nid(sbi, i, FREE_NID, PREALLOC_NID);
2359 nm_i->available_nids--;
2361 update_free_nid_bitmap(sbi, *nid, false, false);
2363 spin_unlock(&nm_i->nid_list_lock);
2366 spin_unlock(&nm_i->nid_list_lock);
2368 /* Let's scan nat pages and its caches to get free nids */
2369 if (!f2fs_build_free_nids(sbi, true, false))
2375 * f2fs_alloc_nid() should be called prior to this function.
2377 void f2fs_alloc_nid_done(struct f2fs_sb_info *sbi, nid_t nid)
2379 struct f2fs_nm_info *nm_i = NM_I(sbi);
2382 spin_lock(&nm_i->nid_list_lock);
2383 i = __lookup_free_nid_list(nm_i, nid);
2384 f2fs_bug_on(sbi, !i);
2385 __remove_free_nid(sbi, i, PREALLOC_NID);
2386 spin_unlock(&nm_i->nid_list_lock);
2388 kmem_cache_free(free_nid_slab, i);
2392 * f2fs_alloc_nid() should be called prior to this function.
2394 void f2fs_alloc_nid_failed(struct f2fs_sb_info *sbi, nid_t nid)
2396 struct f2fs_nm_info *nm_i = NM_I(sbi);
2398 bool need_free = false;
2403 spin_lock(&nm_i->nid_list_lock);
2404 i = __lookup_free_nid_list(nm_i, nid);
2405 f2fs_bug_on(sbi, !i);
2407 if (!f2fs_available_free_memory(sbi, FREE_NIDS)) {
2408 __remove_free_nid(sbi, i, PREALLOC_NID);
2411 __move_free_nid(sbi, i, PREALLOC_NID, FREE_NID);
2414 nm_i->available_nids++;
2416 update_free_nid_bitmap(sbi, nid, true, false);
2418 spin_unlock(&nm_i->nid_list_lock);
2421 kmem_cache_free(free_nid_slab, i);
2424 int f2fs_try_to_free_nids(struct f2fs_sb_info *sbi, int nr_shrink)
2426 struct f2fs_nm_info *nm_i = NM_I(sbi);
2427 struct free_nid *i, *next;
2430 if (nm_i->nid_cnt[FREE_NID] <= MAX_FREE_NIDS)
2433 if (!mutex_trylock(&nm_i->build_lock))
2436 spin_lock(&nm_i->nid_list_lock);
2437 list_for_each_entry_safe(i, next, &nm_i->free_nid_list, list) {
2438 if (nr_shrink <= 0 ||
2439 nm_i->nid_cnt[FREE_NID] <= MAX_FREE_NIDS)
2442 __remove_free_nid(sbi, i, FREE_NID);
2443 kmem_cache_free(free_nid_slab, i);
2446 spin_unlock(&nm_i->nid_list_lock);
2447 mutex_unlock(&nm_i->build_lock);
2449 return nr - nr_shrink;
2452 void f2fs_recover_inline_xattr(struct inode *inode, struct page *page)
2454 void *src_addr, *dst_addr;
2457 struct f2fs_inode *ri;
2459 ipage = f2fs_get_node_page(F2FS_I_SB(inode), inode->i_ino);
2460 f2fs_bug_on(F2FS_I_SB(inode), IS_ERR(ipage));
2462 ri = F2FS_INODE(page);
2463 if (ri->i_inline & F2FS_INLINE_XATTR) {
2464 set_inode_flag(inode, FI_INLINE_XATTR);
2466 clear_inode_flag(inode, FI_INLINE_XATTR);
2470 dst_addr = inline_xattr_addr(inode, ipage);
2471 src_addr = inline_xattr_addr(inode, page);
2472 inline_size = inline_xattr_size(inode);
2474 f2fs_wait_on_page_writeback(ipage, NODE, true, true);
2475 memcpy(dst_addr, src_addr, inline_size);
2477 f2fs_update_inode(inode, ipage);
2478 f2fs_put_page(ipage, 1);
2481 int f2fs_recover_xattr_data(struct inode *inode, struct page *page)
2483 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
2484 nid_t prev_xnid = F2FS_I(inode)->i_xattr_nid;
2486 struct dnode_of_data dn;
2487 struct node_info ni;
2494 /* 1: invalidate the previous xattr nid */
2495 err = f2fs_get_node_info(sbi, prev_xnid, &ni);
2499 f2fs_invalidate_blocks(sbi, ni.blk_addr);
2500 dec_valid_node_count(sbi, inode, false);
2501 set_node_addr(sbi, &ni, NULL_ADDR, false);
2504 /* 2: update xattr nid in inode */
2505 if (!f2fs_alloc_nid(sbi, &new_xnid))
2508 set_new_dnode(&dn, inode, NULL, NULL, new_xnid);
2509 xpage = f2fs_new_node_page(&dn, XATTR_NODE_OFFSET);
2510 if (IS_ERR(xpage)) {
2511 f2fs_alloc_nid_failed(sbi, new_xnid);
2512 return PTR_ERR(xpage);
2515 f2fs_alloc_nid_done(sbi, new_xnid);
2516 f2fs_update_inode_page(inode);
2518 /* 3: update and set xattr node page dirty */
2519 memcpy(F2FS_NODE(xpage), F2FS_NODE(page), VALID_XATTR_BLOCK_SIZE);
2521 set_page_dirty(xpage);
2522 f2fs_put_page(xpage, 1);
2527 int f2fs_recover_inode_page(struct f2fs_sb_info *sbi, struct page *page)
2529 struct f2fs_inode *src, *dst;
2530 nid_t ino = ino_of_node(page);
2531 struct node_info old_ni, new_ni;
2535 err = f2fs_get_node_info(sbi, ino, &old_ni);
2539 if (unlikely(old_ni.blk_addr != NULL_ADDR))
2542 ipage = f2fs_grab_cache_page(NODE_MAPPING(sbi), ino, false);
2544 congestion_wait(BLK_RW_ASYNC, HZ/50);
2548 /* Should not use this inode from free nid list */
2549 remove_free_nid(sbi, ino);
2551 if (!PageUptodate(ipage))
2552 SetPageUptodate(ipage);
2553 fill_node_footer(ipage, ino, ino, 0, true);
2554 set_cold_node(ipage, false);
2556 src = F2FS_INODE(page);
2557 dst = F2FS_INODE(ipage);
2559 memcpy(dst, src, (unsigned long)&src->i_ext - (unsigned long)src);
2561 dst->i_blocks = cpu_to_le64(1);
2562 dst->i_links = cpu_to_le32(1);
2563 dst->i_xattr_nid = 0;
2564 dst->i_inline = src->i_inline & (F2FS_INLINE_XATTR | F2FS_EXTRA_ATTR);
2565 if (dst->i_inline & F2FS_EXTRA_ATTR) {
2566 dst->i_extra_isize = src->i_extra_isize;
2568 if (f2fs_sb_has_flexible_inline_xattr(sbi) &&
2569 F2FS_FITS_IN_INODE(src, le16_to_cpu(src->i_extra_isize),
2570 i_inline_xattr_size))
2571 dst->i_inline_xattr_size = src->i_inline_xattr_size;
2573 if (f2fs_sb_has_project_quota(sbi) &&
2574 F2FS_FITS_IN_INODE(src, le16_to_cpu(src->i_extra_isize),
2576 dst->i_projid = src->i_projid;
2578 if (f2fs_sb_has_inode_crtime(sbi) &&
2579 F2FS_FITS_IN_INODE(src, le16_to_cpu(src->i_extra_isize),
2581 dst->i_crtime = src->i_crtime;
2582 dst->i_crtime_nsec = src->i_crtime_nsec;
2589 if (unlikely(inc_valid_node_count(sbi, NULL, true)))
2591 set_node_addr(sbi, &new_ni, NEW_ADDR, false);
2592 inc_valid_inode_count(sbi);
2593 set_page_dirty(ipage);
2594 f2fs_put_page(ipage, 1);
2598 int f2fs_restore_node_summary(struct f2fs_sb_info *sbi,
2599 unsigned int segno, struct f2fs_summary_block *sum)
2601 struct f2fs_node *rn;
2602 struct f2fs_summary *sum_entry;
2604 int i, idx, last_offset, nrpages;
2606 /* scan the node segment */
2607 last_offset = sbi->blocks_per_seg;
2608 addr = START_BLOCK(sbi, segno);
2609 sum_entry = &sum->entries[0];
2611 for (i = 0; i < last_offset; i += nrpages, addr += nrpages) {
2612 nrpages = min(last_offset - i, BIO_MAX_PAGES);
2614 /* readahead node pages */
2615 f2fs_ra_meta_pages(sbi, addr, nrpages, META_POR, true);
2617 for (idx = addr; idx < addr + nrpages; idx++) {
2618 struct page *page = f2fs_get_tmp_page(sbi, idx);
2621 return PTR_ERR(page);
2623 rn = F2FS_NODE(page);
2624 sum_entry->nid = rn->footer.nid;
2625 sum_entry->version = 0;
2626 sum_entry->ofs_in_node = 0;
2628 f2fs_put_page(page, 1);
2631 invalidate_mapping_pages(META_MAPPING(sbi), addr,
2637 static void remove_nats_in_journal(struct f2fs_sb_info *sbi)
2639 struct f2fs_nm_info *nm_i = NM_I(sbi);
2640 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
2641 struct f2fs_journal *journal = curseg->journal;
2644 down_write(&curseg->journal_rwsem);
2645 for (i = 0; i < nats_in_cursum(journal); i++) {
2646 struct nat_entry *ne;
2647 struct f2fs_nat_entry raw_ne;
2648 nid_t nid = le32_to_cpu(nid_in_journal(journal, i));
2650 raw_ne = nat_in_journal(journal, i);
2652 ne = __lookup_nat_cache(nm_i, nid);
2654 ne = __alloc_nat_entry(nid, true);
2655 __init_nat_entry(nm_i, ne, &raw_ne, true);
2659 * if a free nat in journal has not been used after last
2660 * checkpoint, we should remove it from available nids,
2661 * since later we will add it again.
2663 if (!get_nat_flag(ne, IS_DIRTY) &&
2664 le32_to_cpu(raw_ne.block_addr) == NULL_ADDR) {
2665 spin_lock(&nm_i->nid_list_lock);
2666 nm_i->available_nids--;
2667 spin_unlock(&nm_i->nid_list_lock);
2670 __set_nat_cache_dirty(nm_i, ne);
2672 update_nats_in_cursum(journal, -i);
2673 up_write(&curseg->journal_rwsem);
2676 static void __adjust_nat_entry_set(struct nat_entry_set *nes,
2677 struct list_head *head, int max)
2679 struct nat_entry_set *cur;
2681 if (nes->entry_cnt >= max)
2684 list_for_each_entry(cur, head, set_list) {
2685 if (cur->entry_cnt >= nes->entry_cnt) {
2686 list_add(&nes->set_list, cur->set_list.prev);
2691 list_add_tail(&nes->set_list, head);
2694 static void __update_nat_bits(struct f2fs_sb_info *sbi, nid_t start_nid,
2697 struct f2fs_nm_info *nm_i = NM_I(sbi);
2698 unsigned int nat_index = start_nid / NAT_ENTRY_PER_BLOCK;
2699 struct f2fs_nat_block *nat_blk = page_address(page);
2703 if (!enabled_nat_bits(sbi, NULL))
2706 if (nat_index == 0) {
2710 for (; i < NAT_ENTRY_PER_BLOCK; i++) {
2711 if (nat_blk->entries[i].block_addr != NULL_ADDR)
2715 __set_bit_le(nat_index, nm_i->empty_nat_bits);
2716 __clear_bit_le(nat_index, nm_i->full_nat_bits);
2720 __clear_bit_le(nat_index, nm_i->empty_nat_bits);
2721 if (valid == NAT_ENTRY_PER_BLOCK)
2722 __set_bit_le(nat_index, nm_i->full_nat_bits);
2724 __clear_bit_le(nat_index, nm_i->full_nat_bits);
2727 static int __flush_nat_entry_set(struct f2fs_sb_info *sbi,
2728 struct nat_entry_set *set, struct cp_control *cpc)
2730 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
2731 struct f2fs_journal *journal = curseg->journal;
2732 nid_t start_nid = set->set * NAT_ENTRY_PER_BLOCK;
2733 bool to_journal = true;
2734 struct f2fs_nat_block *nat_blk;
2735 struct nat_entry *ne, *cur;
2736 struct page *page = NULL;
2739 * there are two steps to flush nat entries:
2740 * #1, flush nat entries to journal in current hot data summary block.
2741 * #2, flush nat entries to nat page.
2743 if (enabled_nat_bits(sbi, cpc) ||
2744 !__has_cursum_space(journal, set->entry_cnt, NAT_JOURNAL))
2748 down_write(&curseg->journal_rwsem);
2750 page = get_next_nat_page(sbi, start_nid);
2752 return PTR_ERR(page);
2754 nat_blk = page_address(page);
2755 f2fs_bug_on(sbi, !nat_blk);
2758 /* flush dirty nats in nat entry set */
2759 list_for_each_entry_safe(ne, cur, &set->entry_list, list) {
2760 struct f2fs_nat_entry *raw_ne;
2761 nid_t nid = nat_get_nid(ne);
2764 f2fs_bug_on(sbi, nat_get_blkaddr(ne) == NEW_ADDR);
2767 offset = f2fs_lookup_journal_in_cursum(journal,
2768 NAT_JOURNAL, nid, 1);
2769 f2fs_bug_on(sbi, offset < 0);
2770 raw_ne = &nat_in_journal(journal, offset);
2771 nid_in_journal(journal, offset) = cpu_to_le32(nid);
2773 raw_ne = &nat_blk->entries[nid - start_nid];
2775 raw_nat_from_node_info(raw_ne, &ne->ni);
2777 __clear_nat_cache_dirty(NM_I(sbi), set, ne);
2778 if (nat_get_blkaddr(ne) == NULL_ADDR) {
2779 add_free_nid(sbi, nid, false, true);
2781 spin_lock(&NM_I(sbi)->nid_list_lock);
2782 update_free_nid_bitmap(sbi, nid, false, false);
2783 spin_unlock(&NM_I(sbi)->nid_list_lock);
2788 up_write(&curseg->journal_rwsem);
2790 __update_nat_bits(sbi, start_nid, page);
2791 f2fs_put_page(page, 1);
2794 /* Allow dirty nats by node block allocation in write_begin */
2795 if (!set->entry_cnt) {
2796 radix_tree_delete(&NM_I(sbi)->nat_set_root, set->set);
2797 kmem_cache_free(nat_entry_set_slab, set);
2803 * This function is called during the checkpointing process.
2805 int f2fs_flush_nat_entries(struct f2fs_sb_info *sbi, struct cp_control *cpc)
2807 struct f2fs_nm_info *nm_i = NM_I(sbi);
2808 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
2809 struct f2fs_journal *journal = curseg->journal;
2810 struct nat_entry_set *setvec[SETVEC_SIZE];
2811 struct nat_entry_set *set, *tmp;
2817 /* during unmount, let's flush nat_bits before checking dirty_nat_cnt */
2818 if (enabled_nat_bits(sbi, cpc)) {
2819 down_write(&nm_i->nat_tree_lock);
2820 remove_nats_in_journal(sbi);
2821 up_write(&nm_i->nat_tree_lock);
2824 if (!nm_i->dirty_nat_cnt)
2827 down_write(&nm_i->nat_tree_lock);
2830 * if there are no enough space in journal to store dirty nat
2831 * entries, remove all entries from journal and merge them
2832 * into nat entry set.
2834 if (enabled_nat_bits(sbi, cpc) ||
2835 !__has_cursum_space(journal, nm_i->dirty_nat_cnt, NAT_JOURNAL))
2836 remove_nats_in_journal(sbi);
2838 while ((found = __gang_lookup_nat_set(nm_i,
2839 set_idx, SETVEC_SIZE, setvec))) {
2841 set_idx = setvec[found - 1]->set + 1;
2842 for (idx = 0; idx < found; idx++)
2843 __adjust_nat_entry_set(setvec[idx], &sets,
2844 MAX_NAT_JENTRIES(journal));
2847 /* flush dirty nats in nat entry set */
2848 list_for_each_entry_safe(set, tmp, &sets, set_list) {
2849 err = __flush_nat_entry_set(sbi, set, cpc);
2854 up_write(&nm_i->nat_tree_lock);
2855 /* Allow dirty nats by node block allocation in write_begin */
2860 static int __get_nat_bitmaps(struct f2fs_sb_info *sbi)
2862 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
2863 struct f2fs_nm_info *nm_i = NM_I(sbi);
2864 unsigned int nat_bits_bytes = nm_i->nat_blocks / BITS_PER_BYTE;
2866 __u64 cp_ver = cur_cp_version(ckpt);
2867 block_t nat_bits_addr;
2869 if (!enabled_nat_bits(sbi, NULL))
2872 nm_i->nat_bits_blocks = F2FS_BLK_ALIGN((nat_bits_bytes << 1) + 8);
2873 nm_i->nat_bits = f2fs_kzalloc(sbi,
2874 nm_i->nat_bits_blocks << F2FS_BLKSIZE_BITS, GFP_KERNEL);
2875 if (!nm_i->nat_bits)
2878 nat_bits_addr = __start_cp_addr(sbi) + sbi->blocks_per_seg -
2879 nm_i->nat_bits_blocks;
2880 for (i = 0; i < nm_i->nat_bits_blocks; i++) {
2883 page = f2fs_get_meta_page(sbi, nat_bits_addr++);
2885 return PTR_ERR(page);
2887 memcpy(nm_i->nat_bits + (i << F2FS_BLKSIZE_BITS),
2888 page_address(page), F2FS_BLKSIZE);
2889 f2fs_put_page(page, 1);
2892 cp_ver |= (cur_cp_crc(ckpt) << 32);
2893 if (cpu_to_le64(cp_ver) != *(__le64 *)nm_i->nat_bits) {
2894 disable_nat_bits(sbi, true);
2898 nm_i->full_nat_bits = nm_i->nat_bits + 8;
2899 nm_i->empty_nat_bits = nm_i->full_nat_bits + nat_bits_bytes;
2901 f2fs_msg(sbi->sb, KERN_NOTICE, "Found nat_bits in checkpoint");
2905 static inline void load_free_nid_bitmap(struct f2fs_sb_info *sbi)
2907 struct f2fs_nm_info *nm_i = NM_I(sbi);
2909 nid_t nid, last_nid;
2911 if (!enabled_nat_bits(sbi, NULL))
2914 for (i = 0; i < nm_i->nat_blocks; i++) {
2915 i = find_next_bit_le(nm_i->empty_nat_bits, nm_i->nat_blocks, i);
2916 if (i >= nm_i->nat_blocks)
2919 __set_bit_le(i, nm_i->nat_block_bitmap);
2921 nid = i * NAT_ENTRY_PER_BLOCK;
2922 last_nid = nid + NAT_ENTRY_PER_BLOCK;
2924 spin_lock(&NM_I(sbi)->nid_list_lock);
2925 for (; nid < last_nid; nid++)
2926 update_free_nid_bitmap(sbi, nid, true, true);
2927 spin_unlock(&NM_I(sbi)->nid_list_lock);
2930 for (i = 0; i < nm_i->nat_blocks; i++) {
2931 i = find_next_bit_le(nm_i->full_nat_bits, nm_i->nat_blocks, i);
2932 if (i >= nm_i->nat_blocks)
2935 __set_bit_le(i, nm_i->nat_block_bitmap);
2939 static int init_node_manager(struct f2fs_sb_info *sbi)
2941 struct f2fs_super_block *sb_raw = F2FS_RAW_SUPER(sbi);
2942 struct f2fs_nm_info *nm_i = NM_I(sbi);
2943 unsigned char *version_bitmap;
2944 unsigned int nat_segs;
2947 nm_i->nat_blkaddr = le32_to_cpu(sb_raw->nat_blkaddr);
2949 /* segment_count_nat includes pair segment so divide to 2. */
2950 nat_segs = le32_to_cpu(sb_raw->segment_count_nat) >> 1;
2951 nm_i->nat_blocks = nat_segs << le32_to_cpu(sb_raw->log_blocks_per_seg);
2952 nm_i->max_nid = NAT_ENTRY_PER_BLOCK * nm_i->nat_blocks;
2954 /* not used nids: 0, node, meta, (and root counted as valid node) */
2955 nm_i->available_nids = nm_i->max_nid - sbi->total_valid_node_count -
2956 sbi->nquota_files - F2FS_RESERVED_NODE_NUM;
2957 nm_i->nid_cnt[FREE_NID] = 0;
2958 nm_i->nid_cnt[PREALLOC_NID] = 0;
2960 nm_i->ram_thresh = DEF_RAM_THRESHOLD;
2961 nm_i->ra_nid_pages = DEF_RA_NID_PAGES;
2962 nm_i->dirty_nats_ratio = DEF_DIRTY_NAT_RATIO_THRESHOLD;
2964 INIT_RADIX_TREE(&nm_i->free_nid_root, GFP_ATOMIC);
2965 INIT_LIST_HEAD(&nm_i->free_nid_list);
2966 INIT_RADIX_TREE(&nm_i->nat_root, GFP_NOIO);
2967 INIT_RADIX_TREE(&nm_i->nat_set_root, GFP_NOIO);
2968 INIT_LIST_HEAD(&nm_i->nat_entries);
2969 spin_lock_init(&nm_i->nat_list_lock);
2971 mutex_init(&nm_i->build_lock);
2972 spin_lock_init(&nm_i->nid_list_lock);
2973 init_rwsem(&nm_i->nat_tree_lock);
2975 nm_i->next_scan_nid = le32_to_cpu(sbi->ckpt->next_free_nid);
2976 nm_i->bitmap_size = __bitmap_size(sbi, NAT_BITMAP);
2977 version_bitmap = __bitmap_ptr(sbi, NAT_BITMAP);
2978 if (!version_bitmap)
2981 nm_i->nat_bitmap = kmemdup(version_bitmap, nm_i->bitmap_size,
2983 if (!nm_i->nat_bitmap)
2986 err = __get_nat_bitmaps(sbi);
2990 #ifdef CONFIG_F2FS_CHECK_FS
2991 nm_i->nat_bitmap_mir = kmemdup(version_bitmap, nm_i->bitmap_size,
2993 if (!nm_i->nat_bitmap_mir)
3000 static int init_free_nid_cache(struct f2fs_sb_info *sbi)
3002 struct f2fs_nm_info *nm_i = NM_I(sbi);
3005 nm_i->free_nid_bitmap =
3006 f2fs_kzalloc(sbi, array_size(sizeof(unsigned char *),
3009 if (!nm_i->free_nid_bitmap)
3012 for (i = 0; i < nm_i->nat_blocks; i++) {
3013 nm_i->free_nid_bitmap[i] = f2fs_kvzalloc(sbi,
3014 f2fs_bitmap_size(NAT_ENTRY_PER_BLOCK), GFP_KERNEL);
3015 if (!nm_i->free_nid_bitmap[i])
3019 nm_i->nat_block_bitmap = f2fs_kvzalloc(sbi, nm_i->nat_blocks / 8,
3021 if (!nm_i->nat_block_bitmap)
3024 nm_i->free_nid_count =
3025 f2fs_kvzalloc(sbi, array_size(sizeof(unsigned short),
3028 if (!nm_i->free_nid_count)
3033 int f2fs_build_node_manager(struct f2fs_sb_info *sbi)
3037 sbi->nm_info = f2fs_kzalloc(sbi, sizeof(struct f2fs_nm_info),
3042 err = init_node_manager(sbi);
3046 err = init_free_nid_cache(sbi);
3050 /* load free nid status from nat_bits table */
3051 load_free_nid_bitmap(sbi);
3053 return f2fs_build_free_nids(sbi, true, true);
3056 void f2fs_destroy_node_manager(struct f2fs_sb_info *sbi)
3058 struct f2fs_nm_info *nm_i = NM_I(sbi);
3059 struct free_nid *i, *next_i;
3060 struct nat_entry *natvec[NATVEC_SIZE];
3061 struct nat_entry_set *setvec[SETVEC_SIZE];
3068 /* destroy free nid list */
3069 spin_lock(&nm_i->nid_list_lock);
3070 list_for_each_entry_safe(i, next_i, &nm_i->free_nid_list, list) {
3071 __remove_free_nid(sbi, i, FREE_NID);
3072 spin_unlock(&nm_i->nid_list_lock);
3073 kmem_cache_free(free_nid_slab, i);
3074 spin_lock(&nm_i->nid_list_lock);
3076 f2fs_bug_on(sbi, nm_i->nid_cnt[FREE_NID]);
3077 f2fs_bug_on(sbi, nm_i->nid_cnt[PREALLOC_NID]);
3078 f2fs_bug_on(sbi, !list_empty(&nm_i->free_nid_list));
3079 spin_unlock(&nm_i->nid_list_lock);
3081 /* destroy nat cache */
3082 down_write(&nm_i->nat_tree_lock);
3083 while ((found = __gang_lookup_nat_cache(nm_i,
3084 nid, NATVEC_SIZE, natvec))) {
3087 nid = nat_get_nid(natvec[found - 1]) + 1;
3088 for (idx = 0; idx < found; idx++) {
3089 spin_lock(&nm_i->nat_list_lock);
3090 list_del(&natvec[idx]->list);
3091 spin_unlock(&nm_i->nat_list_lock);
3093 __del_from_nat_cache(nm_i, natvec[idx]);
3096 f2fs_bug_on(sbi, nm_i->nat_cnt);
3098 /* destroy nat set cache */
3100 while ((found = __gang_lookup_nat_set(nm_i,
3101 nid, SETVEC_SIZE, setvec))) {
3104 nid = setvec[found - 1]->set + 1;
3105 for (idx = 0; idx < found; idx++) {
3106 /* entry_cnt is not zero, when cp_error was occurred */
3107 f2fs_bug_on(sbi, !list_empty(&setvec[idx]->entry_list));
3108 radix_tree_delete(&nm_i->nat_set_root, setvec[idx]->set);
3109 kmem_cache_free(nat_entry_set_slab, setvec[idx]);
3112 up_write(&nm_i->nat_tree_lock);
3114 kvfree(nm_i->nat_block_bitmap);
3115 if (nm_i->free_nid_bitmap) {
3118 for (i = 0; i < nm_i->nat_blocks; i++)
3119 kvfree(nm_i->free_nid_bitmap[i]);
3120 kvfree(nm_i->free_nid_bitmap);
3122 kvfree(nm_i->free_nid_count);
3124 kvfree(nm_i->nat_bitmap);
3125 kvfree(nm_i->nat_bits);
3126 #ifdef CONFIG_F2FS_CHECK_FS
3127 kvfree(nm_i->nat_bitmap_mir);
3129 sbi->nm_info = NULL;
3133 int __init f2fs_create_node_manager_caches(void)
3135 nat_entry_slab = f2fs_kmem_cache_create("nat_entry",
3136 sizeof(struct nat_entry));
3137 if (!nat_entry_slab)
3140 free_nid_slab = f2fs_kmem_cache_create("free_nid",
3141 sizeof(struct free_nid));
3143 goto destroy_nat_entry;
3145 nat_entry_set_slab = f2fs_kmem_cache_create("nat_entry_set",
3146 sizeof(struct nat_entry_set));
3147 if (!nat_entry_set_slab)
3148 goto destroy_free_nid;
3150 fsync_node_entry_slab = f2fs_kmem_cache_create("fsync_node_entry",
3151 sizeof(struct fsync_node_entry));
3152 if (!fsync_node_entry_slab)
3153 goto destroy_nat_entry_set;
3156 destroy_nat_entry_set:
3157 kmem_cache_destroy(nat_entry_set_slab);
3159 kmem_cache_destroy(free_nid_slab);
3161 kmem_cache_destroy(nat_entry_slab);
3166 void f2fs_destroy_node_manager_caches(void)
3168 kmem_cache_destroy(fsync_node_entry_slab);
3169 kmem_cache_destroy(nat_entry_set_slab);
3170 kmem_cache_destroy(free_nid_slab);
3171 kmem_cache_destroy(nat_entry_slab);