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_radix_tree_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;
620 base += 2 * direct_blks;
622 base += direct_index;
625 f2fs_bug_on(F2FS_I_SB(dn->inode), 1);
628 return ((pgofs - base) / skipped_unit + 1) * skipped_unit + base;
632 * The maximum depth is four.
633 * Offset[0] will have raw inode offset.
635 static int get_node_path(struct inode *inode, long block,
636 int offset[4], unsigned int noffset[4])
638 const long direct_index = ADDRS_PER_INODE(inode);
639 const long direct_blks = ADDRS_PER_BLOCK;
640 const long dptrs_per_blk = NIDS_PER_BLOCK;
641 const long indirect_blks = ADDRS_PER_BLOCK * NIDS_PER_BLOCK;
642 const long dindirect_blks = indirect_blks * NIDS_PER_BLOCK;
648 if (block < direct_index) {
652 block -= direct_index;
653 if (block < direct_blks) {
654 offset[n++] = NODE_DIR1_BLOCK;
660 block -= direct_blks;
661 if (block < direct_blks) {
662 offset[n++] = NODE_DIR2_BLOCK;
668 block -= direct_blks;
669 if (block < indirect_blks) {
670 offset[n++] = NODE_IND1_BLOCK;
672 offset[n++] = block / direct_blks;
673 noffset[n] = 4 + offset[n - 1];
674 offset[n] = block % direct_blks;
678 block -= indirect_blks;
679 if (block < indirect_blks) {
680 offset[n++] = NODE_IND2_BLOCK;
681 noffset[n] = 4 + dptrs_per_blk;
682 offset[n++] = block / direct_blks;
683 noffset[n] = 5 + dptrs_per_blk + offset[n - 1];
684 offset[n] = block % direct_blks;
688 block -= indirect_blks;
689 if (block < dindirect_blks) {
690 offset[n++] = NODE_DIND_BLOCK;
691 noffset[n] = 5 + (dptrs_per_blk * 2);
692 offset[n++] = block / indirect_blks;
693 noffset[n] = 6 + (dptrs_per_blk * 2) +
694 offset[n - 1] * (dptrs_per_blk + 1);
695 offset[n++] = (block / direct_blks) % dptrs_per_blk;
696 noffset[n] = 7 + (dptrs_per_blk * 2) +
697 offset[n - 2] * (dptrs_per_blk + 1) +
699 offset[n] = block % direct_blks;
710 * Caller should call f2fs_put_dnode(dn).
711 * Also, it should grab and release a rwsem by calling f2fs_lock_op() and
712 * f2fs_unlock_op() only if ro is not set RDONLY_NODE.
713 * In the case of RDONLY_NODE, we don't need to care about mutex.
715 int f2fs_get_dnode_of_data(struct dnode_of_data *dn, pgoff_t index, int mode)
717 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
718 struct page *npage[4];
719 struct page *parent = NULL;
721 unsigned int noffset[4];
726 level = get_node_path(dn->inode, index, offset, noffset);
730 nids[0] = dn->inode->i_ino;
731 npage[0] = dn->inode_page;
734 npage[0] = f2fs_get_node_page(sbi, nids[0]);
735 if (IS_ERR(npage[0]))
736 return PTR_ERR(npage[0]);
739 /* if inline_data is set, should not report any block indices */
740 if (f2fs_has_inline_data(dn->inode) && index) {
742 f2fs_put_page(npage[0], 1);
748 nids[1] = get_nid(parent, offset[0], true);
749 dn->inode_page = npage[0];
750 dn->inode_page_locked = true;
752 /* get indirect or direct nodes */
753 for (i = 1; i <= level; i++) {
756 if (!nids[i] && mode == ALLOC_NODE) {
758 if (!f2fs_alloc_nid(sbi, &(nids[i]))) {
764 npage[i] = f2fs_new_node_page(dn, noffset[i]);
765 if (IS_ERR(npage[i])) {
766 f2fs_alloc_nid_failed(sbi, nids[i]);
767 err = PTR_ERR(npage[i]);
771 set_nid(parent, offset[i - 1], nids[i], i == 1);
772 f2fs_alloc_nid_done(sbi, nids[i]);
774 } else if (mode == LOOKUP_NODE_RA && i == level && level > 1) {
775 npage[i] = f2fs_get_node_page_ra(parent, offset[i - 1]);
776 if (IS_ERR(npage[i])) {
777 err = PTR_ERR(npage[i]);
783 dn->inode_page_locked = false;
786 f2fs_put_page(parent, 1);
790 npage[i] = f2fs_get_node_page(sbi, nids[i]);
791 if (IS_ERR(npage[i])) {
792 err = PTR_ERR(npage[i]);
793 f2fs_put_page(npage[0], 0);
799 nids[i + 1] = get_nid(parent, offset[i], false);
802 dn->nid = nids[level];
803 dn->ofs_in_node = offset[level];
804 dn->node_page = npage[level];
805 dn->data_blkaddr = datablock_addr(dn->inode,
806 dn->node_page, dn->ofs_in_node);
810 f2fs_put_page(parent, 1);
812 f2fs_put_page(npage[0], 0);
814 dn->inode_page = NULL;
815 dn->node_page = NULL;
816 if (err == -ENOENT) {
818 dn->max_level = level;
819 dn->ofs_in_node = offset[level];
824 static int truncate_node(struct dnode_of_data *dn)
826 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
830 err = f2fs_get_node_info(sbi, dn->nid, &ni);
834 /* Deallocate node address */
835 f2fs_invalidate_blocks(sbi, ni.blk_addr);
836 dec_valid_node_count(sbi, dn->inode, dn->nid == dn->inode->i_ino);
837 set_node_addr(sbi, &ni, NULL_ADDR, false);
839 if (dn->nid == dn->inode->i_ino) {
840 f2fs_remove_orphan_inode(sbi, dn->nid);
841 dec_valid_inode_count(sbi);
842 f2fs_inode_synced(dn->inode);
845 clear_node_page_dirty(dn->node_page);
846 set_sbi_flag(sbi, SBI_IS_DIRTY);
848 f2fs_put_page(dn->node_page, 1);
850 invalidate_mapping_pages(NODE_MAPPING(sbi),
851 dn->node_page->index, dn->node_page->index);
853 dn->node_page = NULL;
854 trace_f2fs_truncate_node(dn->inode, dn->nid, ni.blk_addr);
859 static int truncate_dnode(struct dnode_of_data *dn)
867 /* get direct node */
868 page = f2fs_get_node_page(F2FS_I_SB(dn->inode), dn->nid);
869 if (IS_ERR(page) && PTR_ERR(page) == -ENOENT)
871 else if (IS_ERR(page))
872 return PTR_ERR(page);
874 /* Make dnode_of_data for parameter */
875 dn->node_page = page;
877 f2fs_truncate_data_blocks(dn);
878 err = truncate_node(dn);
885 static int truncate_nodes(struct dnode_of_data *dn, unsigned int nofs,
888 struct dnode_of_data rdn = *dn;
890 struct f2fs_node *rn;
892 unsigned int child_nofs;
897 return NIDS_PER_BLOCK + 1;
899 trace_f2fs_truncate_nodes_enter(dn->inode, dn->nid, dn->data_blkaddr);
901 page = f2fs_get_node_page(F2FS_I_SB(dn->inode), dn->nid);
903 trace_f2fs_truncate_nodes_exit(dn->inode, PTR_ERR(page));
904 return PTR_ERR(page);
907 f2fs_ra_node_pages(page, ofs, NIDS_PER_BLOCK);
909 rn = F2FS_NODE(page);
911 for (i = ofs; i < NIDS_PER_BLOCK; i++, freed++) {
912 child_nid = le32_to_cpu(rn->in.nid[i]);
916 ret = truncate_dnode(&rdn);
919 if (set_nid(page, i, 0, false))
920 dn->node_changed = true;
923 child_nofs = nofs + ofs * (NIDS_PER_BLOCK + 1) + 1;
924 for (i = ofs; i < NIDS_PER_BLOCK; i++) {
925 child_nid = le32_to_cpu(rn->in.nid[i]);
926 if (child_nid == 0) {
927 child_nofs += NIDS_PER_BLOCK + 1;
931 ret = truncate_nodes(&rdn, child_nofs, 0, depth - 1);
932 if (ret == (NIDS_PER_BLOCK + 1)) {
933 if (set_nid(page, i, 0, false))
934 dn->node_changed = true;
936 } else if (ret < 0 && ret != -ENOENT) {
944 /* remove current indirect node */
945 dn->node_page = page;
946 ret = truncate_node(dn);
951 f2fs_put_page(page, 1);
953 trace_f2fs_truncate_nodes_exit(dn->inode, freed);
957 f2fs_put_page(page, 1);
958 trace_f2fs_truncate_nodes_exit(dn->inode, ret);
962 static int truncate_partial_nodes(struct dnode_of_data *dn,
963 struct f2fs_inode *ri, int *offset, int depth)
965 struct page *pages[2];
972 nid[0] = le32_to_cpu(ri->i_nid[offset[0] - NODE_DIR1_BLOCK]);
976 /* get indirect nodes in the path */
977 for (i = 0; i < idx + 1; i++) {
978 /* reference count'll be increased */
979 pages[i] = f2fs_get_node_page(F2FS_I_SB(dn->inode), nid[i]);
980 if (IS_ERR(pages[i])) {
981 err = PTR_ERR(pages[i]);
985 nid[i + 1] = get_nid(pages[i], offset[i + 1], false);
988 f2fs_ra_node_pages(pages[idx], offset[idx + 1], NIDS_PER_BLOCK);
990 /* free direct nodes linked to a partial indirect node */
991 for (i = offset[idx + 1]; i < NIDS_PER_BLOCK; i++) {
992 child_nid = get_nid(pages[idx], i, false);
996 err = truncate_dnode(dn);
999 if (set_nid(pages[idx], i, 0, false))
1000 dn->node_changed = true;
1003 if (offset[idx + 1] == 0) {
1004 dn->node_page = pages[idx];
1006 err = truncate_node(dn);
1010 f2fs_put_page(pages[idx], 1);
1013 offset[idx + 1] = 0;
1016 for (i = idx; i >= 0; i--)
1017 f2fs_put_page(pages[i], 1);
1019 trace_f2fs_truncate_partial_nodes(dn->inode, nid, depth, err);
1025 * All the block addresses of data and nodes should be nullified.
1027 int f2fs_truncate_inode_blocks(struct inode *inode, pgoff_t from)
1029 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1030 int err = 0, cont = 1;
1031 int level, offset[4], noffset[4];
1032 unsigned int nofs = 0;
1033 struct f2fs_inode *ri;
1034 struct dnode_of_data dn;
1037 trace_f2fs_truncate_inode_blocks_enter(inode, from);
1039 level = get_node_path(inode, from, offset, noffset);
1043 page = f2fs_get_node_page(sbi, inode->i_ino);
1045 trace_f2fs_truncate_inode_blocks_exit(inode, PTR_ERR(page));
1046 return PTR_ERR(page);
1049 set_new_dnode(&dn, inode, page, NULL, 0);
1052 ri = F2FS_INODE(page);
1060 if (!offset[level - 1])
1062 err = truncate_partial_nodes(&dn, ri, offset, level);
1063 if (err < 0 && err != -ENOENT)
1065 nofs += 1 + NIDS_PER_BLOCK;
1068 nofs = 5 + 2 * NIDS_PER_BLOCK;
1069 if (!offset[level - 1])
1071 err = truncate_partial_nodes(&dn, ri, offset, level);
1072 if (err < 0 && err != -ENOENT)
1081 dn.nid = le32_to_cpu(ri->i_nid[offset[0] - NODE_DIR1_BLOCK]);
1082 switch (offset[0]) {
1083 case NODE_DIR1_BLOCK:
1084 case NODE_DIR2_BLOCK:
1085 err = truncate_dnode(&dn);
1088 case NODE_IND1_BLOCK:
1089 case NODE_IND2_BLOCK:
1090 err = truncate_nodes(&dn, nofs, offset[1], 2);
1093 case NODE_DIND_BLOCK:
1094 err = truncate_nodes(&dn, nofs, offset[1], 3);
1101 if (err < 0 && err != -ENOENT)
1103 if (offset[1] == 0 &&
1104 ri->i_nid[offset[0] - NODE_DIR1_BLOCK]) {
1106 BUG_ON(page->mapping != NODE_MAPPING(sbi));
1107 f2fs_wait_on_page_writeback(page, NODE, true);
1108 ri->i_nid[offset[0] - NODE_DIR1_BLOCK] = 0;
1109 set_page_dirty(page);
1117 f2fs_put_page(page, 0);
1118 trace_f2fs_truncate_inode_blocks_exit(inode, err);
1119 return err > 0 ? 0 : err;
1122 /* caller must lock inode page */
1123 int f2fs_truncate_xattr_node(struct inode *inode)
1125 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1126 nid_t nid = F2FS_I(inode)->i_xattr_nid;
1127 struct dnode_of_data dn;
1134 npage = f2fs_get_node_page(sbi, nid);
1136 return PTR_ERR(npage);
1138 set_new_dnode(&dn, inode, NULL, npage, nid);
1139 err = truncate_node(&dn);
1141 f2fs_put_page(npage, 1);
1145 f2fs_i_xnid_write(inode, 0);
1151 * Caller should grab and release a rwsem by calling f2fs_lock_op() and
1154 int f2fs_remove_inode_page(struct inode *inode)
1156 struct dnode_of_data dn;
1159 set_new_dnode(&dn, inode, NULL, NULL, inode->i_ino);
1160 err = f2fs_get_dnode_of_data(&dn, 0, LOOKUP_NODE);
1164 err = f2fs_truncate_xattr_node(inode);
1166 f2fs_put_dnode(&dn);
1170 /* remove potential inline_data blocks */
1171 if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
1172 S_ISLNK(inode->i_mode))
1173 f2fs_truncate_data_blocks_range(&dn, 1);
1175 /* 0 is possible, after f2fs_new_inode() has failed */
1176 if (unlikely(f2fs_cp_error(F2FS_I_SB(inode)))) {
1177 f2fs_put_dnode(&dn);
1180 f2fs_bug_on(F2FS_I_SB(inode),
1181 inode->i_blocks != 0 && inode->i_blocks != 8);
1183 /* will put inode & node pages */
1184 err = truncate_node(&dn);
1186 f2fs_put_dnode(&dn);
1192 struct page *f2fs_new_inode_page(struct inode *inode)
1194 struct dnode_of_data dn;
1196 /* allocate inode page for new inode */
1197 set_new_dnode(&dn, inode, NULL, NULL, inode->i_ino);
1199 /* caller should f2fs_put_page(page, 1); */
1200 return f2fs_new_node_page(&dn, 0);
1203 struct page *f2fs_new_node_page(struct dnode_of_data *dn, unsigned int ofs)
1205 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
1206 struct node_info new_ni;
1210 if (unlikely(is_inode_flag_set(dn->inode, FI_NO_ALLOC)))
1211 return ERR_PTR(-EPERM);
1213 page = f2fs_grab_cache_page(NODE_MAPPING(sbi), dn->nid, false);
1215 return ERR_PTR(-ENOMEM);
1217 if (unlikely((err = inc_valid_node_count(sbi, dn->inode, !ofs))))
1220 #ifdef CONFIG_F2FS_CHECK_FS
1221 err = f2fs_get_node_info(sbi, dn->nid, &new_ni);
1223 dec_valid_node_count(sbi, dn->inode, !ofs);
1226 f2fs_bug_on(sbi, new_ni.blk_addr != NULL_ADDR);
1228 new_ni.nid = dn->nid;
1229 new_ni.ino = dn->inode->i_ino;
1230 new_ni.blk_addr = NULL_ADDR;
1233 set_node_addr(sbi, &new_ni, NEW_ADDR, false);
1235 f2fs_wait_on_page_writeback(page, NODE, true);
1236 fill_node_footer(page, dn->nid, dn->inode->i_ino, ofs, true);
1237 set_cold_node(page, S_ISDIR(dn->inode->i_mode));
1238 if (!PageUptodate(page))
1239 SetPageUptodate(page);
1240 if (set_page_dirty(page))
1241 dn->node_changed = true;
1243 if (f2fs_has_xattr_block(ofs))
1244 f2fs_i_xnid_write(dn->inode, dn->nid);
1247 inc_valid_inode_count(sbi);
1251 clear_node_page_dirty(page);
1252 f2fs_put_page(page, 1);
1253 return ERR_PTR(err);
1257 * Caller should do after getting the following values.
1258 * 0: f2fs_put_page(page, 0)
1259 * LOCKED_PAGE or error: f2fs_put_page(page, 1)
1261 static int read_node_page(struct page *page, int op_flags)
1263 struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1264 struct node_info ni;
1265 struct f2fs_io_info fio = {
1269 .op_flags = op_flags,
1271 .encrypted_page = NULL,
1275 if (PageUptodate(page)) {
1276 #ifdef CONFIG_F2FS_CHECK_FS
1277 f2fs_bug_on(sbi, !f2fs_inode_chksum_verify(sbi, page));
1282 err = f2fs_get_node_info(sbi, page->index, &ni);
1286 if (unlikely(ni.blk_addr == NULL_ADDR) ||
1287 is_sbi_flag_set(sbi, SBI_IS_SHUTDOWN)) {
1288 ClearPageUptodate(page);
1292 fio.new_blkaddr = fio.old_blkaddr = ni.blk_addr;
1293 return f2fs_submit_page_bio(&fio);
1297 * Readahead a node page
1299 void f2fs_ra_node_page(struct f2fs_sb_info *sbi, nid_t nid)
1306 if (f2fs_check_nid_range(sbi, nid))
1310 apage = radix_tree_lookup(&NODE_MAPPING(sbi)->i_pages, nid);
1315 apage = f2fs_grab_cache_page(NODE_MAPPING(sbi), nid, false);
1319 err = read_node_page(apage, REQ_RAHEAD);
1320 f2fs_put_page(apage, err ? 1 : 0);
1323 static struct page *__get_node_page(struct f2fs_sb_info *sbi, pgoff_t nid,
1324 struct page *parent, int start)
1330 return ERR_PTR(-ENOENT);
1331 if (f2fs_check_nid_range(sbi, nid))
1332 return ERR_PTR(-EINVAL);
1334 page = f2fs_grab_cache_page(NODE_MAPPING(sbi), nid, false);
1336 return ERR_PTR(-ENOMEM);
1338 err = read_node_page(page, 0);
1340 f2fs_put_page(page, 1);
1341 return ERR_PTR(err);
1342 } else if (err == LOCKED_PAGE) {
1348 f2fs_ra_node_pages(parent, start + 1, MAX_RA_NODE);
1352 if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1353 f2fs_put_page(page, 1);
1357 if (unlikely(!PageUptodate(page))) {
1362 if (!f2fs_inode_chksum_verify(sbi, page)) {
1367 if(unlikely(nid != nid_of_node(page))) {
1368 f2fs_msg(sbi->sb, KERN_WARNING, "inconsistent node block, "
1369 "nid:%lu, node_footer[nid:%u,ino:%u,ofs:%u,cpver:%llu,blkaddr:%u]",
1370 nid, nid_of_node(page), ino_of_node(page),
1371 ofs_of_node(page), cpver_of_node(page),
1372 next_blkaddr_of_node(page));
1375 ClearPageUptodate(page);
1376 f2fs_put_page(page, 1);
1377 return ERR_PTR(err);
1382 struct page *f2fs_get_node_page(struct f2fs_sb_info *sbi, pgoff_t nid)
1384 return __get_node_page(sbi, nid, NULL, 0);
1387 struct page *f2fs_get_node_page_ra(struct page *parent, int start)
1389 struct f2fs_sb_info *sbi = F2FS_P_SB(parent);
1390 nid_t nid = get_nid(parent, start, false);
1392 return __get_node_page(sbi, nid, parent, start);
1395 static void flush_inline_data(struct f2fs_sb_info *sbi, nid_t ino)
1397 struct inode *inode;
1401 /* should flush inline_data before evict_inode */
1402 inode = ilookup(sbi->sb, ino);
1406 page = f2fs_pagecache_get_page(inode->i_mapping, 0,
1407 FGP_LOCK|FGP_NOWAIT, 0);
1411 if (!PageUptodate(page))
1414 if (!PageDirty(page))
1417 if (!clear_page_dirty_for_io(page))
1420 ret = f2fs_write_inline_data(inode, page);
1421 inode_dec_dirty_pages(inode);
1422 f2fs_remove_dirty_inode(inode);
1424 set_page_dirty(page);
1426 f2fs_put_page(page, 1);
1431 static struct page *last_fsync_dnode(struct f2fs_sb_info *sbi, nid_t ino)
1434 struct pagevec pvec;
1435 struct page *last_page = NULL;
1438 pagevec_init(&pvec);
1441 while ((nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1442 PAGECACHE_TAG_DIRTY))) {
1445 for (i = 0; i < nr_pages; i++) {
1446 struct page *page = pvec.pages[i];
1448 if (unlikely(f2fs_cp_error(sbi))) {
1449 f2fs_put_page(last_page, 0);
1450 pagevec_release(&pvec);
1451 return ERR_PTR(-EIO);
1454 if (!IS_DNODE(page) || !is_cold_node(page))
1456 if (ino_of_node(page) != ino)
1461 if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1466 if (ino_of_node(page) != ino)
1467 goto continue_unlock;
1469 if (!PageDirty(page)) {
1470 /* someone wrote it for us */
1471 goto continue_unlock;
1475 f2fs_put_page(last_page, 0);
1481 pagevec_release(&pvec);
1487 static int __write_node_page(struct page *page, bool atomic, bool *submitted,
1488 struct writeback_control *wbc, bool do_balance,
1489 enum iostat_type io_type, unsigned int *seq_id)
1491 struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1493 struct node_info ni;
1494 struct f2fs_io_info fio = {
1496 .ino = ino_of_node(page),
1499 .op_flags = wbc_to_write_flags(wbc),
1501 .encrypted_page = NULL,
1508 trace_f2fs_writepage(page, NODE);
1510 if (unlikely(f2fs_cp_error(sbi)))
1513 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
1516 if (wbc->sync_mode == WB_SYNC_NONE &&
1517 IS_DNODE(page) && is_cold_node(page))
1520 /* get old block addr of this node page */
1521 nid = nid_of_node(page);
1522 f2fs_bug_on(sbi, page->index != nid);
1524 if (f2fs_get_node_info(sbi, nid, &ni))
1527 if (wbc->for_reclaim) {
1528 if (!down_read_trylock(&sbi->node_write))
1531 down_read(&sbi->node_write);
1534 /* This page is already truncated */
1535 if (unlikely(ni.blk_addr == NULL_ADDR)) {
1536 ClearPageUptodate(page);
1537 dec_page_count(sbi, F2FS_DIRTY_NODES);
1538 up_read(&sbi->node_write);
1543 if (__is_valid_data_blkaddr(ni.blk_addr) &&
1544 !f2fs_is_valid_blkaddr(sbi, ni.blk_addr, DATA_GENERIC)) {
1545 up_read(&sbi->node_write);
1549 if (atomic && !test_opt(sbi, NOBARRIER))
1550 fio.op_flags |= REQ_PREFLUSH | REQ_FUA;
1552 set_page_writeback(page);
1553 ClearPageError(page);
1555 if (f2fs_in_warm_node_list(sbi, page)) {
1556 seq = f2fs_add_fsync_node_entry(sbi, page);
1561 fio.old_blkaddr = ni.blk_addr;
1562 f2fs_do_write_node_page(nid, &fio);
1563 set_node_addr(sbi, &ni, fio.new_blkaddr, is_fsync_dnode(page));
1564 dec_page_count(sbi, F2FS_DIRTY_NODES);
1565 up_read(&sbi->node_write);
1567 if (wbc->for_reclaim) {
1568 f2fs_submit_merged_write_cond(sbi, NULL, page, 0, NODE);
1574 if (unlikely(f2fs_cp_error(sbi))) {
1575 f2fs_submit_merged_write(sbi, NODE);
1579 *submitted = fio.submitted;
1582 f2fs_balance_fs(sbi, false);
1586 redirty_page_for_writepage(wbc, page);
1587 return AOP_WRITEPAGE_ACTIVATE;
1590 int f2fs_move_node_page(struct page *node_page, int gc_type)
1594 if (gc_type == FG_GC) {
1595 struct writeback_control wbc = {
1596 .sync_mode = WB_SYNC_ALL,
1601 set_page_dirty(node_page);
1602 f2fs_wait_on_page_writeback(node_page, NODE, true);
1604 f2fs_bug_on(F2FS_P_SB(node_page), PageWriteback(node_page));
1605 if (!clear_page_dirty_for_io(node_page)) {
1610 if (__write_node_page(node_page, false, NULL,
1611 &wbc, false, FS_GC_NODE_IO, NULL)) {
1613 unlock_page(node_page);
1617 /* set page dirty and write it */
1618 if (!PageWriteback(node_page))
1619 set_page_dirty(node_page);
1622 unlock_page(node_page);
1624 f2fs_put_page(node_page, 0);
1628 static int f2fs_write_node_page(struct page *page,
1629 struct writeback_control *wbc)
1631 return __write_node_page(page, false, NULL, wbc, false,
1635 int f2fs_fsync_node_pages(struct f2fs_sb_info *sbi, struct inode *inode,
1636 struct writeback_control *wbc, bool atomic,
1637 unsigned int *seq_id)
1640 struct pagevec pvec;
1642 struct page *last_page = NULL;
1643 bool marked = false;
1644 nid_t ino = inode->i_ino;
1649 last_page = last_fsync_dnode(sbi, ino);
1650 if (IS_ERR_OR_NULL(last_page))
1651 return PTR_ERR_OR_ZERO(last_page);
1654 pagevec_init(&pvec);
1657 while ((nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1658 PAGECACHE_TAG_DIRTY))) {
1661 for (i = 0; i < nr_pages; i++) {
1662 struct page *page = pvec.pages[i];
1663 bool submitted = false;
1665 if (unlikely(f2fs_cp_error(sbi))) {
1666 f2fs_put_page(last_page, 0);
1667 pagevec_release(&pvec);
1672 if (!IS_DNODE(page) || !is_cold_node(page))
1674 if (ino_of_node(page) != ino)
1679 if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1684 if (ino_of_node(page) != ino)
1685 goto continue_unlock;
1687 if (!PageDirty(page) && page != last_page) {
1688 /* someone wrote it for us */
1689 goto continue_unlock;
1692 f2fs_wait_on_page_writeback(page, NODE, true);
1693 BUG_ON(PageWriteback(page));
1695 set_fsync_mark(page, 0);
1696 set_dentry_mark(page, 0);
1698 if (!atomic || page == last_page) {
1699 set_fsync_mark(page, 1);
1700 if (IS_INODE(page)) {
1701 if (is_inode_flag_set(inode,
1703 f2fs_update_inode(inode, page);
1704 set_dentry_mark(page,
1705 f2fs_need_dentry_mark(sbi, ino));
1707 /* may be written by other thread */
1708 if (!PageDirty(page))
1709 set_page_dirty(page);
1712 if (!clear_page_dirty_for_io(page))
1713 goto continue_unlock;
1715 ret = __write_node_page(page, atomic &&
1717 &submitted, wbc, true,
1718 FS_NODE_IO, seq_id);
1721 f2fs_put_page(last_page, 0);
1723 } else if (submitted) {
1727 if (page == last_page) {
1728 f2fs_put_page(page, 0);
1733 pagevec_release(&pvec);
1739 if (!ret && atomic && !marked) {
1740 f2fs_msg(sbi->sb, KERN_DEBUG,
1741 "Retry to write fsync mark: ino=%u, idx=%lx",
1742 ino, last_page->index);
1743 lock_page(last_page);
1744 f2fs_wait_on_page_writeback(last_page, NODE, true);
1745 set_page_dirty(last_page);
1746 unlock_page(last_page);
1751 f2fs_submit_merged_write_cond(sbi, NULL, NULL, ino, NODE);
1752 return ret ? -EIO: 0;
1755 int f2fs_sync_node_pages(struct f2fs_sb_info *sbi,
1756 struct writeback_control *wbc,
1757 bool do_balance, enum iostat_type io_type)
1760 struct pagevec pvec;
1764 int nr_pages, done = 0;
1766 pagevec_init(&pvec);
1771 while (!done && (nr_pages = pagevec_lookup_tag(&pvec,
1772 NODE_MAPPING(sbi), &index, PAGECACHE_TAG_DIRTY))) {
1775 for (i = 0; i < nr_pages; i++) {
1776 struct page *page = pvec.pages[i];
1777 bool submitted = false;
1779 /* give a priority to WB_SYNC threads */
1780 if (atomic_read(&sbi->wb_sync_req[NODE]) &&
1781 wbc->sync_mode == WB_SYNC_NONE) {
1787 * flushing sequence with step:
1792 if (step == 0 && IS_DNODE(page))
1794 if (step == 1 && (!IS_DNODE(page) ||
1795 is_cold_node(page)))
1797 if (step == 2 && (!IS_DNODE(page) ||
1798 !is_cold_node(page)))
1801 if (wbc->sync_mode == WB_SYNC_ALL)
1803 else if (!trylock_page(page))
1806 if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1812 if (!PageDirty(page)) {
1813 /* someone wrote it for us */
1814 goto continue_unlock;
1817 /* flush inline_data */
1818 if (is_inline_node(page)) {
1819 clear_inline_node(page);
1821 flush_inline_data(sbi, ino_of_node(page));
1825 f2fs_wait_on_page_writeback(page, NODE, true);
1827 BUG_ON(PageWriteback(page));
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);
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 (get_pages(sbi, F2FS_DIRTY_NODES) < nr_pages_to_skip(sbi, NODE))
1928 if (wbc->sync_mode == WB_SYNC_ALL)
1929 atomic_inc(&sbi->wb_sync_req[NODE]);
1930 else if (atomic_read(&sbi->wb_sync_req[NODE]))
1933 trace_f2fs_writepages(mapping->host, wbc, NODE);
1935 diff = nr_pages_to_write(sbi, NODE, wbc);
1936 blk_start_plug(&plug);
1937 f2fs_sync_node_pages(sbi, wbc, true, FS_NODE_IO);
1938 blk_finish_plug(&plug);
1939 wbc->nr_to_write = max((long)0, wbc->nr_to_write - diff);
1941 if (wbc->sync_mode == WB_SYNC_ALL)
1942 atomic_dec(&sbi->wb_sync_req[NODE]);
1946 wbc->pages_skipped += get_pages(sbi, F2FS_DIRTY_NODES);
1947 trace_f2fs_writepages(mapping->host, wbc, NODE);
1951 static int f2fs_set_node_page_dirty(struct page *page)
1953 trace_f2fs_set_page_dirty(page, NODE);
1955 if (!PageUptodate(page))
1956 SetPageUptodate(page);
1957 #ifdef CONFIG_F2FS_CHECK_FS
1959 f2fs_inode_chksum_set(F2FS_P_SB(page), page);
1961 if (!PageDirty(page)) {
1962 __set_page_dirty_nobuffers(page);
1963 inc_page_count(F2FS_P_SB(page), F2FS_DIRTY_NODES);
1964 SetPagePrivate(page);
1965 f2fs_trace_pid(page);
1972 * Structure of the f2fs node operations
1974 const struct address_space_operations f2fs_node_aops = {
1975 .writepage = f2fs_write_node_page,
1976 .writepages = f2fs_write_node_pages,
1977 .set_page_dirty = f2fs_set_node_page_dirty,
1978 .invalidatepage = f2fs_invalidate_page,
1979 .releasepage = f2fs_release_page,
1980 #ifdef CONFIG_MIGRATION
1981 .migratepage = f2fs_migrate_page,
1985 static struct free_nid *__lookup_free_nid_list(struct f2fs_nm_info *nm_i,
1988 return radix_tree_lookup(&nm_i->free_nid_root, n);
1991 static int __insert_free_nid(struct f2fs_sb_info *sbi,
1992 struct free_nid *i, enum nid_state state)
1994 struct f2fs_nm_info *nm_i = NM_I(sbi);
1996 int err = radix_tree_insert(&nm_i->free_nid_root, i->nid, i);
2000 f2fs_bug_on(sbi, state != i->state);
2001 nm_i->nid_cnt[state]++;
2002 if (state == FREE_NID)
2003 list_add_tail(&i->list, &nm_i->free_nid_list);
2007 static void __remove_free_nid(struct f2fs_sb_info *sbi,
2008 struct free_nid *i, enum nid_state state)
2010 struct f2fs_nm_info *nm_i = NM_I(sbi);
2012 f2fs_bug_on(sbi, state != i->state);
2013 nm_i->nid_cnt[state]--;
2014 if (state == FREE_NID)
2016 radix_tree_delete(&nm_i->free_nid_root, i->nid);
2019 static void __move_free_nid(struct f2fs_sb_info *sbi, struct free_nid *i,
2020 enum nid_state org_state, enum nid_state dst_state)
2022 struct f2fs_nm_info *nm_i = NM_I(sbi);
2024 f2fs_bug_on(sbi, org_state != i->state);
2025 i->state = dst_state;
2026 nm_i->nid_cnt[org_state]--;
2027 nm_i->nid_cnt[dst_state]++;
2029 switch (dst_state) {
2034 list_add_tail(&i->list, &nm_i->free_nid_list);
2041 static void update_free_nid_bitmap(struct f2fs_sb_info *sbi, nid_t nid,
2042 bool set, bool build)
2044 struct f2fs_nm_info *nm_i = NM_I(sbi);
2045 unsigned int nat_ofs = NAT_BLOCK_OFFSET(nid);
2046 unsigned int nid_ofs = nid - START_NID(nid);
2048 if (!test_bit_le(nat_ofs, nm_i->nat_block_bitmap))
2052 if (test_bit_le(nid_ofs, nm_i->free_nid_bitmap[nat_ofs]))
2054 __set_bit_le(nid_ofs, nm_i->free_nid_bitmap[nat_ofs]);
2055 nm_i->free_nid_count[nat_ofs]++;
2057 if (!test_bit_le(nid_ofs, nm_i->free_nid_bitmap[nat_ofs]))
2059 __clear_bit_le(nid_ofs, nm_i->free_nid_bitmap[nat_ofs]);
2061 nm_i->free_nid_count[nat_ofs]--;
2065 /* return if the nid is recognized as free */
2066 static bool add_free_nid(struct f2fs_sb_info *sbi,
2067 nid_t nid, bool build, bool update)
2069 struct f2fs_nm_info *nm_i = NM_I(sbi);
2070 struct free_nid *i, *e;
2071 struct nat_entry *ne;
2075 /* 0 nid should not be used */
2076 if (unlikely(nid == 0))
2079 i = f2fs_kmem_cache_alloc(free_nid_slab, GFP_NOFS);
2081 i->state = FREE_NID;
2083 radix_tree_preload(GFP_NOFS | __GFP_NOFAIL);
2085 spin_lock(&nm_i->nid_list_lock);
2093 * - __insert_nid_to_list(PREALLOC_NID)
2094 * - f2fs_balance_fs_bg
2095 * - f2fs_build_free_nids
2096 * - __f2fs_build_free_nids
2099 * - __lookup_nat_cache
2101 * - f2fs_init_inode_metadata
2102 * - f2fs_new_inode_page
2103 * - f2fs_new_node_page
2105 * - f2fs_alloc_nid_done
2106 * - __remove_nid_from_list(PREALLOC_NID)
2107 * - __insert_nid_to_list(FREE_NID)
2109 ne = __lookup_nat_cache(nm_i, nid);
2110 if (ne && (!get_nat_flag(ne, IS_CHECKPOINTED) ||
2111 nat_get_blkaddr(ne) != NULL_ADDR))
2114 e = __lookup_free_nid_list(nm_i, nid);
2116 if (e->state == FREE_NID)
2122 err = __insert_free_nid(sbi, i, FREE_NID);
2125 update_free_nid_bitmap(sbi, nid, ret, build);
2127 nm_i->available_nids++;
2129 spin_unlock(&nm_i->nid_list_lock);
2130 radix_tree_preload_end();
2133 kmem_cache_free(free_nid_slab, i);
2137 static void remove_free_nid(struct f2fs_sb_info *sbi, nid_t nid)
2139 struct f2fs_nm_info *nm_i = NM_I(sbi);
2141 bool need_free = false;
2143 spin_lock(&nm_i->nid_list_lock);
2144 i = __lookup_free_nid_list(nm_i, nid);
2145 if (i && i->state == FREE_NID) {
2146 __remove_free_nid(sbi, i, FREE_NID);
2149 spin_unlock(&nm_i->nid_list_lock);
2152 kmem_cache_free(free_nid_slab, i);
2155 static int scan_nat_page(struct f2fs_sb_info *sbi,
2156 struct page *nat_page, nid_t start_nid)
2158 struct f2fs_nm_info *nm_i = NM_I(sbi);
2159 struct f2fs_nat_block *nat_blk = page_address(nat_page);
2161 unsigned int nat_ofs = NAT_BLOCK_OFFSET(start_nid);
2164 __set_bit_le(nat_ofs, nm_i->nat_block_bitmap);
2166 i = start_nid % NAT_ENTRY_PER_BLOCK;
2168 for (; i < NAT_ENTRY_PER_BLOCK; i++, start_nid++) {
2169 if (unlikely(start_nid >= nm_i->max_nid))
2172 blk_addr = le32_to_cpu(nat_blk->entries[i].block_addr);
2174 if (blk_addr == NEW_ADDR)
2177 if (blk_addr == NULL_ADDR) {
2178 add_free_nid(sbi, start_nid, true, true);
2180 spin_lock(&NM_I(sbi)->nid_list_lock);
2181 update_free_nid_bitmap(sbi, start_nid, false, true);
2182 spin_unlock(&NM_I(sbi)->nid_list_lock);
2189 static void scan_curseg_cache(struct f2fs_sb_info *sbi)
2191 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
2192 struct f2fs_journal *journal = curseg->journal;
2195 down_read(&curseg->journal_rwsem);
2196 for (i = 0; i < nats_in_cursum(journal); i++) {
2200 addr = le32_to_cpu(nat_in_journal(journal, i).block_addr);
2201 nid = le32_to_cpu(nid_in_journal(journal, i));
2202 if (addr == NULL_ADDR)
2203 add_free_nid(sbi, nid, true, false);
2205 remove_free_nid(sbi, nid);
2207 up_read(&curseg->journal_rwsem);
2210 static void scan_free_nid_bits(struct f2fs_sb_info *sbi)
2212 struct f2fs_nm_info *nm_i = NM_I(sbi);
2213 unsigned int i, idx;
2216 down_read(&nm_i->nat_tree_lock);
2218 for (i = 0; i < nm_i->nat_blocks; i++) {
2219 if (!test_bit_le(i, nm_i->nat_block_bitmap))
2221 if (!nm_i->free_nid_count[i])
2223 for (idx = 0; idx < NAT_ENTRY_PER_BLOCK; idx++) {
2224 idx = find_next_bit_le(nm_i->free_nid_bitmap[i],
2225 NAT_ENTRY_PER_BLOCK, idx);
2226 if (idx >= NAT_ENTRY_PER_BLOCK)
2229 nid = i * NAT_ENTRY_PER_BLOCK + idx;
2230 add_free_nid(sbi, nid, true, false);
2232 if (nm_i->nid_cnt[FREE_NID] >= MAX_FREE_NIDS)
2237 scan_curseg_cache(sbi);
2239 up_read(&nm_i->nat_tree_lock);
2242 static int __f2fs_build_free_nids(struct f2fs_sb_info *sbi,
2243 bool sync, bool mount)
2245 struct f2fs_nm_info *nm_i = NM_I(sbi);
2247 nid_t nid = nm_i->next_scan_nid;
2249 if (unlikely(nid >= nm_i->max_nid))
2252 /* Enough entries */
2253 if (nm_i->nid_cnt[FREE_NID] >= NAT_ENTRY_PER_BLOCK)
2256 if (!sync && !f2fs_available_free_memory(sbi, FREE_NIDS))
2260 /* try to find free nids in free_nid_bitmap */
2261 scan_free_nid_bits(sbi);
2263 if (nm_i->nid_cnt[FREE_NID] >= NAT_ENTRY_PER_BLOCK)
2267 /* readahead nat pages to be scanned */
2268 f2fs_ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nid), FREE_NID_PAGES,
2271 down_read(&nm_i->nat_tree_lock);
2274 if (!test_bit_le(NAT_BLOCK_OFFSET(nid),
2275 nm_i->nat_block_bitmap)) {
2276 struct page *page = get_current_nat_page(sbi, nid);
2279 ret = PTR_ERR(page);
2281 ret = scan_nat_page(sbi, page, nid);
2282 f2fs_put_page(page, 1);
2286 up_read(&nm_i->nat_tree_lock);
2287 f2fs_bug_on(sbi, !mount);
2288 f2fs_msg(sbi->sb, KERN_ERR,
2289 "NAT is corrupt, run fsck to fix it");
2294 nid += (NAT_ENTRY_PER_BLOCK - (nid % NAT_ENTRY_PER_BLOCK));
2295 if (unlikely(nid >= nm_i->max_nid))
2298 if (++i >= FREE_NID_PAGES)
2302 /* go to the next free nat pages to find free nids abundantly */
2303 nm_i->next_scan_nid = nid;
2305 /* find free nids from current sum_pages */
2306 scan_curseg_cache(sbi);
2308 up_read(&nm_i->nat_tree_lock);
2310 f2fs_ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nm_i->next_scan_nid),
2311 nm_i->ra_nid_pages, META_NAT, false);
2316 int f2fs_build_free_nids(struct f2fs_sb_info *sbi, bool sync, bool mount)
2320 mutex_lock(&NM_I(sbi)->build_lock);
2321 ret = __f2fs_build_free_nids(sbi, sync, mount);
2322 mutex_unlock(&NM_I(sbi)->build_lock);
2328 * If this function returns success, caller can obtain a new nid
2329 * from second parameter of this function.
2330 * The returned nid could be used ino as well as nid when inode is created.
2332 bool f2fs_alloc_nid(struct f2fs_sb_info *sbi, nid_t *nid)
2334 struct f2fs_nm_info *nm_i = NM_I(sbi);
2335 struct free_nid *i = NULL;
2337 if (time_to_inject(sbi, FAULT_ALLOC_NID)) {
2338 f2fs_show_injection_info(FAULT_ALLOC_NID);
2342 spin_lock(&nm_i->nid_list_lock);
2344 if (unlikely(nm_i->available_nids == 0)) {
2345 spin_unlock(&nm_i->nid_list_lock);
2349 /* We should not use stale free nids created by f2fs_build_free_nids */
2350 if (nm_i->nid_cnt[FREE_NID] && !on_f2fs_build_free_nids(nm_i)) {
2351 f2fs_bug_on(sbi, list_empty(&nm_i->free_nid_list));
2352 i = list_first_entry(&nm_i->free_nid_list,
2353 struct free_nid, list);
2356 __move_free_nid(sbi, i, FREE_NID, PREALLOC_NID);
2357 nm_i->available_nids--;
2359 update_free_nid_bitmap(sbi, *nid, false, false);
2361 spin_unlock(&nm_i->nid_list_lock);
2364 spin_unlock(&nm_i->nid_list_lock);
2366 /* Let's scan nat pages and its caches to get free nids */
2367 if (!f2fs_build_free_nids(sbi, true, false))
2373 * f2fs_alloc_nid() should be called prior to this function.
2375 void f2fs_alloc_nid_done(struct f2fs_sb_info *sbi, nid_t nid)
2377 struct f2fs_nm_info *nm_i = NM_I(sbi);
2380 spin_lock(&nm_i->nid_list_lock);
2381 i = __lookup_free_nid_list(nm_i, nid);
2382 f2fs_bug_on(sbi, !i);
2383 __remove_free_nid(sbi, i, PREALLOC_NID);
2384 spin_unlock(&nm_i->nid_list_lock);
2386 kmem_cache_free(free_nid_slab, i);
2390 * f2fs_alloc_nid() should be called prior to this function.
2392 void f2fs_alloc_nid_failed(struct f2fs_sb_info *sbi, nid_t nid)
2394 struct f2fs_nm_info *nm_i = NM_I(sbi);
2396 bool need_free = false;
2401 spin_lock(&nm_i->nid_list_lock);
2402 i = __lookup_free_nid_list(nm_i, nid);
2403 f2fs_bug_on(sbi, !i);
2405 if (!f2fs_available_free_memory(sbi, FREE_NIDS)) {
2406 __remove_free_nid(sbi, i, PREALLOC_NID);
2409 __move_free_nid(sbi, i, PREALLOC_NID, FREE_NID);
2412 nm_i->available_nids++;
2414 update_free_nid_bitmap(sbi, nid, true, false);
2416 spin_unlock(&nm_i->nid_list_lock);
2419 kmem_cache_free(free_nid_slab, i);
2422 int f2fs_try_to_free_nids(struct f2fs_sb_info *sbi, int nr_shrink)
2424 struct f2fs_nm_info *nm_i = NM_I(sbi);
2425 struct free_nid *i, *next;
2428 if (nm_i->nid_cnt[FREE_NID] <= MAX_FREE_NIDS)
2431 if (!mutex_trylock(&nm_i->build_lock))
2434 spin_lock(&nm_i->nid_list_lock);
2435 list_for_each_entry_safe(i, next, &nm_i->free_nid_list, list) {
2436 if (nr_shrink <= 0 ||
2437 nm_i->nid_cnt[FREE_NID] <= MAX_FREE_NIDS)
2440 __remove_free_nid(sbi, i, FREE_NID);
2441 kmem_cache_free(free_nid_slab, i);
2444 spin_unlock(&nm_i->nid_list_lock);
2445 mutex_unlock(&nm_i->build_lock);
2447 return nr - nr_shrink;
2450 void f2fs_recover_inline_xattr(struct inode *inode, struct page *page)
2452 void *src_addr, *dst_addr;
2455 struct f2fs_inode *ri;
2457 ipage = f2fs_get_node_page(F2FS_I_SB(inode), inode->i_ino);
2458 f2fs_bug_on(F2FS_I_SB(inode), IS_ERR(ipage));
2460 ri = F2FS_INODE(page);
2461 if (ri->i_inline & F2FS_INLINE_XATTR) {
2462 set_inode_flag(inode, FI_INLINE_XATTR);
2464 clear_inode_flag(inode, FI_INLINE_XATTR);
2468 dst_addr = inline_xattr_addr(inode, ipage);
2469 src_addr = inline_xattr_addr(inode, page);
2470 inline_size = inline_xattr_size(inode);
2472 f2fs_wait_on_page_writeback(ipage, NODE, true);
2473 memcpy(dst_addr, src_addr, inline_size);
2475 f2fs_update_inode(inode, ipage);
2476 f2fs_put_page(ipage, 1);
2479 int f2fs_recover_xattr_data(struct inode *inode, struct page *page)
2481 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
2482 nid_t prev_xnid = F2FS_I(inode)->i_xattr_nid;
2484 struct dnode_of_data dn;
2485 struct node_info ni;
2492 /* 1: invalidate the previous xattr nid */
2493 err = f2fs_get_node_info(sbi, prev_xnid, &ni);
2497 f2fs_invalidate_blocks(sbi, ni.blk_addr);
2498 dec_valid_node_count(sbi, inode, false);
2499 set_node_addr(sbi, &ni, NULL_ADDR, false);
2502 /* 2: update xattr nid in inode */
2503 if (!f2fs_alloc_nid(sbi, &new_xnid))
2506 set_new_dnode(&dn, inode, NULL, NULL, new_xnid);
2507 xpage = f2fs_new_node_page(&dn, XATTR_NODE_OFFSET);
2508 if (IS_ERR(xpage)) {
2509 f2fs_alloc_nid_failed(sbi, new_xnid);
2510 return PTR_ERR(xpage);
2513 f2fs_alloc_nid_done(sbi, new_xnid);
2514 f2fs_update_inode_page(inode);
2516 /* 3: update and set xattr node page dirty */
2517 memcpy(F2FS_NODE(xpage), F2FS_NODE(page), VALID_XATTR_BLOCK_SIZE);
2519 set_page_dirty(xpage);
2520 f2fs_put_page(xpage, 1);
2525 int f2fs_recover_inode_page(struct f2fs_sb_info *sbi, struct page *page)
2527 struct f2fs_inode *src, *dst;
2528 nid_t ino = ino_of_node(page);
2529 struct node_info old_ni, new_ni;
2533 err = f2fs_get_node_info(sbi, ino, &old_ni);
2537 if (unlikely(old_ni.blk_addr != NULL_ADDR))
2540 ipage = f2fs_grab_cache_page(NODE_MAPPING(sbi), ino, false);
2542 congestion_wait(BLK_RW_ASYNC, HZ/50);
2546 /* Should not use this inode from free nid list */
2547 remove_free_nid(sbi, ino);
2549 if (!PageUptodate(ipage))
2550 SetPageUptodate(ipage);
2551 fill_node_footer(ipage, ino, ino, 0, true);
2552 set_cold_node(page, false);
2554 src = F2FS_INODE(page);
2555 dst = F2FS_INODE(ipage);
2557 memcpy(dst, src, (unsigned long)&src->i_ext - (unsigned long)src);
2559 dst->i_blocks = cpu_to_le64(1);
2560 dst->i_links = cpu_to_le32(1);
2561 dst->i_xattr_nid = 0;
2562 dst->i_inline = src->i_inline & (F2FS_INLINE_XATTR | F2FS_EXTRA_ATTR);
2563 if (dst->i_inline & F2FS_EXTRA_ATTR) {
2564 dst->i_extra_isize = src->i_extra_isize;
2566 if (f2fs_sb_has_flexible_inline_xattr(sbi->sb) &&
2567 F2FS_FITS_IN_INODE(src, le16_to_cpu(src->i_extra_isize),
2568 i_inline_xattr_size))
2569 dst->i_inline_xattr_size = src->i_inline_xattr_size;
2571 if (f2fs_sb_has_project_quota(sbi->sb) &&
2572 F2FS_FITS_IN_INODE(src, le16_to_cpu(src->i_extra_isize),
2574 dst->i_projid = src->i_projid;
2576 if (f2fs_sb_has_inode_crtime(sbi->sb) &&
2577 F2FS_FITS_IN_INODE(src, le16_to_cpu(src->i_extra_isize),
2579 dst->i_crtime = src->i_crtime;
2580 dst->i_crtime_nsec = src->i_crtime_nsec;
2587 if (unlikely(inc_valid_node_count(sbi, NULL, true)))
2589 set_node_addr(sbi, &new_ni, NEW_ADDR, false);
2590 inc_valid_inode_count(sbi);
2591 set_page_dirty(ipage);
2592 f2fs_put_page(ipage, 1);
2596 int f2fs_restore_node_summary(struct f2fs_sb_info *sbi,
2597 unsigned int segno, struct f2fs_summary_block *sum)
2599 struct f2fs_node *rn;
2600 struct f2fs_summary *sum_entry;
2602 int i, idx, last_offset, nrpages;
2604 /* scan the node segment */
2605 last_offset = sbi->blocks_per_seg;
2606 addr = START_BLOCK(sbi, segno);
2607 sum_entry = &sum->entries[0];
2609 for (i = 0; i < last_offset; i += nrpages, addr += nrpages) {
2610 nrpages = min(last_offset - i, BIO_MAX_PAGES);
2612 /* readahead node pages */
2613 f2fs_ra_meta_pages(sbi, addr, nrpages, META_POR, true);
2615 for (idx = addr; idx < addr + nrpages; idx++) {
2616 struct page *page = f2fs_get_tmp_page(sbi, idx);
2619 return PTR_ERR(page);
2621 rn = F2FS_NODE(page);
2622 sum_entry->nid = rn->footer.nid;
2623 sum_entry->version = 0;
2624 sum_entry->ofs_in_node = 0;
2626 f2fs_put_page(page, 1);
2629 invalidate_mapping_pages(META_MAPPING(sbi), addr,
2635 static void remove_nats_in_journal(struct f2fs_sb_info *sbi)
2637 struct f2fs_nm_info *nm_i = NM_I(sbi);
2638 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
2639 struct f2fs_journal *journal = curseg->journal;
2642 down_write(&curseg->journal_rwsem);
2643 for (i = 0; i < nats_in_cursum(journal); i++) {
2644 struct nat_entry *ne;
2645 struct f2fs_nat_entry raw_ne;
2646 nid_t nid = le32_to_cpu(nid_in_journal(journal, i));
2648 raw_ne = nat_in_journal(journal, i);
2650 ne = __lookup_nat_cache(nm_i, nid);
2652 ne = __alloc_nat_entry(nid, true);
2653 __init_nat_entry(nm_i, ne, &raw_ne, true);
2657 * if a free nat in journal has not been used after last
2658 * checkpoint, we should remove it from available nids,
2659 * since later we will add it again.
2661 if (!get_nat_flag(ne, IS_DIRTY) &&
2662 le32_to_cpu(raw_ne.block_addr) == NULL_ADDR) {
2663 spin_lock(&nm_i->nid_list_lock);
2664 nm_i->available_nids--;
2665 spin_unlock(&nm_i->nid_list_lock);
2668 __set_nat_cache_dirty(nm_i, ne);
2670 update_nats_in_cursum(journal, -i);
2671 up_write(&curseg->journal_rwsem);
2674 static void __adjust_nat_entry_set(struct nat_entry_set *nes,
2675 struct list_head *head, int max)
2677 struct nat_entry_set *cur;
2679 if (nes->entry_cnt >= max)
2682 list_for_each_entry(cur, head, set_list) {
2683 if (cur->entry_cnt >= nes->entry_cnt) {
2684 list_add(&nes->set_list, cur->set_list.prev);
2689 list_add_tail(&nes->set_list, head);
2692 static void __update_nat_bits(struct f2fs_sb_info *sbi, nid_t start_nid,
2695 struct f2fs_nm_info *nm_i = NM_I(sbi);
2696 unsigned int nat_index = start_nid / NAT_ENTRY_PER_BLOCK;
2697 struct f2fs_nat_block *nat_blk = page_address(page);
2701 if (!enabled_nat_bits(sbi, NULL))
2704 if (nat_index == 0) {
2708 for (; i < NAT_ENTRY_PER_BLOCK; i++) {
2709 if (nat_blk->entries[i].block_addr != NULL_ADDR)
2713 __set_bit_le(nat_index, nm_i->empty_nat_bits);
2714 __clear_bit_le(nat_index, nm_i->full_nat_bits);
2718 __clear_bit_le(nat_index, nm_i->empty_nat_bits);
2719 if (valid == NAT_ENTRY_PER_BLOCK)
2720 __set_bit_le(nat_index, nm_i->full_nat_bits);
2722 __clear_bit_le(nat_index, nm_i->full_nat_bits);
2725 static int __flush_nat_entry_set(struct f2fs_sb_info *sbi,
2726 struct nat_entry_set *set, struct cp_control *cpc)
2728 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
2729 struct f2fs_journal *journal = curseg->journal;
2730 nid_t start_nid = set->set * NAT_ENTRY_PER_BLOCK;
2731 bool to_journal = true;
2732 struct f2fs_nat_block *nat_blk;
2733 struct nat_entry *ne, *cur;
2734 struct page *page = NULL;
2737 * there are two steps to flush nat entries:
2738 * #1, flush nat entries to journal in current hot data summary block.
2739 * #2, flush nat entries to nat page.
2741 if (enabled_nat_bits(sbi, cpc) ||
2742 !__has_cursum_space(journal, set->entry_cnt, NAT_JOURNAL))
2746 down_write(&curseg->journal_rwsem);
2748 page = get_next_nat_page(sbi, start_nid);
2750 return PTR_ERR(page);
2752 nat_blk = page_address(page);
2753 f2fs_bug_on(sbi, !nat_blk);
2756 /* flush dirty nats in nat entry set */
2757 list_for_each_entry_safe(ne, cur, &set->entry_list, list) {
2758 struct f2fs_nat_entry *raw_ne;
2759 nid_t nid = nat_get_nid(ne);
2762 f2fs_bug_on(sbi, nat_get_blkaddr(ne) == NEW_ADDR);
2765 offset = f2fs_lookup_journal_in_cursum(journal,
2766 NAT_JOURNAL, nid, 1);
2767 f2fs_bug_on(sbi, offset < 0);
2768 raw_ne = &nat_in_journal(journal, offset);
2769 nid_in_journal(journal, offset) = cpu_to_le32(nid);
2771 raw_ne = &nat_blk->entries[nid - start_nid];
2773 raw_nat_from_node_info(raw_ne, &ne->ni);
2775 __clear_nat_cache_dirty(NM_I(sbi), set, ne);
2776 if (nat_get_blkaddr(ne) == NULL_ADDR) {
2777 add_free_nid(sbi, nid, false, true);
2779 spin_lock(&NM_I(sbi)->nid_list_lock);
2780 update_free_nid_bitmap(sbi, nid, false, false);
2781 spin_unlock(&NM_I(sbi)->nid_list_lock);
2786 up_write(&curseg->journal_rwsem);
2788 __update_nat_bits(sbi, start_nid, page);
2789 f2fs_put_page(page, 1);
2792 /* Allow dirty nats by node block allocation in write_begin */
2793 if (!set->entry_cnt) {
2794 radix_tree_delete(&NM_I(sbi)->nat_set_root, set->set);
2795 kmem_cache_free(nat_entry_set_slab, set);
2801 * This function is called during the checkpointing process.
2803 int f2fs_flush_nat_entries(struct f2fs_sb_info *sbi, struct cp_control *cpc)
2805 struct f2fs_nm_info *nm_i = NM_I(sbi);
2806 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
2807 struct f2fs_journal *journal = curseg->journal;
2808 struct nat_entry_set *setvec[SETVEC_SIZE];
2809 struct nat_entry_set *set, *tmp;
2815 /* during unmount, let's flush nat_bits before checking dirty_nat_cnt */
2816 if (enabled_nat_bits(sbi, cpc)) {
2817 down_write(&nm_i->nat_tree_lock);
2818 remove_nats_in_journal(sbi);
2819 up_write(&nm_i->nat_tree_lock);
2822 if (!nm_i->dirty_nat_cnt)
2825 down_write(&nm_i->nat_tree_lock);
2828 * if there are no enough space in journal to store dirty nat
2829 * entries, remove all entries from journal and merge them
2830 * into nat entry set.
2832 if (enabled_nat_bits(sbi, cpc) ||
2833 !__has_cursum_space(journal, nm_i->dirty_nat_cnt, NAT_JOURNAL))
2834 remove_nats_in_journal(sbi);
2836 while ((found = __gang_lookup_nat_set(nm_i,
2837 set_idx, SETVEC_SIZE, setvec))) {
2839 set_idx = setvec[found - 1]->set + 1;
2840 for (idx = 0; idx < found; idx++)
2841 __adjust_nat_entry_set(setvec[idx], &sets,
2842 MAX_NAT_JENTRIES(journal));
2845 /* flush dirty nats in nat entry set */
2846 list_for_each_entry_safe(set, tmp, &sets, set_list) {
2847 err = __flush_nat_entry_set(sbi, set, cpc);
2852 up_write(&nm_i->nat_tree_lock);
2853 /* Allow dirty nats by node block allocation in write_begin */
2858 static int __get_nat_bitmaps(struct f2fs_sb_info *sbi)
2860 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
2861 struct f2fs_nm_info *nm_i = NM_I(sbi);
2862 unsigned int nat_bits_bytes = nm_i->nat_blocks / BITS_PER_BYTE;
2864 __u64 cp_ver = cur_cp_version(ckpt);
2865 block_t nat_bits_addr;
2867 if (!enabled_nat_bits(sbi, NULL))
2870 nm_i->nat_bits_blocks = F2FS_BLK_ALIGN((nat_bits_bytes << 1) + 8);
2871 nm_i->nat_bits = f2fs_kzalloc(sbi,
2872 nm_i->nat_bits_blocks << F2FS_BLKSIZE_BITS, GFP_KERNEL);
2873 if (!nm_i->nat_bits)
2876 nat_bits_addr = __start_cp_addr(sbi) + sbi->blocks_per_seg -
2877 nm_i->nat_bits_blocks;
2878 for (i = 0; i < nm_i->nat_bits_blocks; i++) {
2881 page = f2fs_get_meta_page(sbi, nat_bits_addr++);
2883 disable_nat_bits(sbi, true);
2884 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 kfree(nm_i->free_nid_bitmap);
3122 kvfree(nm_i->free_nid_count);
3124 kfree(nm_i->nat_bitmap);
3125 kfree(nm_i->nat_bits);
3126 #ifdef CONFIG_F2FS_CHECK_FS
3127 kfree(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);