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 void f2fs_move_node_page(struct page *node_page, int gc_type)
1592 if (gc_type == FG_GC) {
1593 struct writeback_control wbc = {
1594 .sync_mode = WB_SYNC_ALL,
1599 set_page_dirty(node_page);
1600 f2fs_wait_on_page_writeback(node_page, NODE, true);
1602 f2fs_bug_on(F2FS_P_SB(node_page), PageWriteback(node_page));
1603 if (!clear_page_dirty_for_io(node_page))
1606 if (__write_node_page(node_page, false, NULL,
1607 &wbc, false, FS_GC_NODE_IO, NULL))
1608 unlock_page(node_page);
1611 /* set page dirty and write it */
1612 if (!PageWriteback(node_page))
1613 set_page_dirty(node_page);
1616 unlock_page(node_page);
1618 f2fs_put_page(node_page, 0);
1621 static int f2fs_write_node_page(struct page *page,
1622 struct writeback_control *wbc)
1624 return __write_node_page(page, false, NULL, wbc, false,
1628 int f2fs_fsync_node_pages(struct f2fs_sb_info *sbi, struct inode *inode,
1629 struct writeback_control *wbc, bool atomic,
1630 unsigned int *seq_id)
1633 struct pagevec pvec;
1635 struct page *last_page = NULL;
1636 bool marked = false;
1637 nid_t ino = inode->i_ino;
1642 last_page = last_fsync_dnode(sbi, ino);
1643 if (IS_ERR_OR_NULL(last_page))
1644 return PTR_ERR_OR_ZERO(last_page);
1647 pagevec_init(&pvec);
1650 while ((nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1651 PAGECACHE_TAG_DIRTY))) {
1654 for (i = 0; i < nr_pages; i++) {
1655 struct page *page = pvec.pages[i];
1656 bool submitted = false;
1658 if (unlikely(f2fs_cp_error(sbi))) {
1659 f2fs_put_page(last_page, 0);
1660 pagevec_release(&pvec);
1665 if (!IS_DNODE(page) || !is_cold_node(page))
1667 if (ino_of_node(page) != ino)
1672 if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1677 if (ino_of_node(page) != ino)
1678 goto continue_unlock;
1680 if (!PageDirty(page) && page != last_page) {
1681 /* someone wrote it for us */
1682 goto continue_unlock;
1685 f2fs_wait_on_page_writeback(page, NODE, true);
1686 BUG_ON(PageWriteback(page));
1688 set_fsync_mark(page, 0);
1689 set_dentry_mark(page, 0);
1691 if (!atomic || page == last_page) {
1692 set_fsync_mark(page, 1);
1693 if (IS_INODE(page)) {
1694 if (is_inode_flag_set(inode,
1696 f2fs_update_inode(inode, page);
1697 set_dentry_mark(page,
1698 f2fs_need_dentry_mark(sbi, ino));
1700 /* may be written by other thread */
1701 if (!PageDirty(page))
1702 set_page_dirty(page);
1705 if (!clear_page_dirty_for_io(page))
1706 goto continue_unlock;
1708 ret = __write_node_page(page, atomic &&
1710 &submitted, wbc, true,
1711 FS_NODE_IO, seq_id);
1714 f2fs_put_page(last_page, 0);
1716 } else if (submitted) {
1720 if (page == last_page) {
1721 f2fs_put_page(page, 0);
1726 pagevec_release(&pvec);
1732 if (!ret && atomic && !marked) {
1733 f2fs_msg(sbi->sb, KERN_DEBUG,
1734 "Retry to write fsync mark: ino=%u, idx=%lx",
1735 ino, last_page->index);
1736 lock_page(last_page);
1737 f2fs_wait_on_page_writeback(last_page, NODE, true);
1738 set_page_dirty(last_page);
1739 unlock_page(last_page);
1744 f2fs_submit_merged_write_cond(sbi, NULL, NULL, ino, NODE);
1745 return ret ? -EIO: 0;
1748 int f2fs_sync_node_pages(struct f2fs_sb_info *sbi,
1749 struct writeback_control *wbc,
1750 bool do_balance, enum iostat_type io_type)
1753 struct pagevec pvec;
1757 int nr_pages, done = 0;
1759 pagevec_init(&pvec);
1764 while (!done && (nr_pages = pagevec_lookup_tag(&pvec,
1765 NODE_MAPPING(sbi), &index, PAGECACHE_TAG_DIRTY))) {
1768 for (i = 0; i < nr_pages; i++) {
1769 struct page *page = pvec.pages[i];
1770 bool submitted = false;
1772 /* give a priority to WB_SYNC threads */
1773 if (atomic_read(&sbi->wb_sync_req[NODE]) &&
1774 wbc->sync_mode == WB_SYNC_NONE) {
1780 * flushing sequence with step:
1785 if (step == 0 && IS_DNODE(page))
1787 if (step == 1 && (!IS_DNODE(page) ||
1788 is_cold_node(page)))
1790 if (step == 2 && (!IS_DNODE(page) ||
1791 !is_cold_node(page)))
1794 if (wbc->sync_mode == WB_SYNC_ALL)
1796 else if (!trylock_page(page))
1799 if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1805 if (!PageDirty(page)) {
1806 /* someone wrote it for us */
1807 goto continue_unlock;
1810 /* flush inline_data */
1811 if (is_inline_node(page)) {
1812 clear_inline_node(page);
1814 flush_inline_data(sbi, ino_of_node(page));
1818 f2fs_wait_on_page_writeback(page, NODE, true);
1820 BUG_ON(PageWriteback(page));
1821 if (!clear_page_dirty_for_io(page))
1822 goto continue_unlock;
1824 set_fsync_mark(page, 0);
1825 set_dentry_mark(page, 0);
1827 ret = __write_node_page(page, false, &submitted,
1828 wbc, do_balance, io_type, NULL);
1834 if (--wbc->nr_to_write == 0)
1837 pagevec_release(&pvec);
1840 if (wbc->nr_to_write == 0) {
1847 if (wbc->sync_mode == WB_SYNC_NONE && step == 1)
1854 f2fs_submit_merged_write(sbi, NODE);
1856 if (unlikely(f2fs_cp_error(sbi)))
1861 int f2fs_wait_on_node_pages_writeback(struct f2fs_sb_info *sbi,
1862 unsigned int seq_id)
1864 struct fsync_node_entry *fn;
1866 struct list_head *head = &sbi->fsync_node_list;
1867 unsigned long flags;
1868 unsigned int cur_seq_id = 0;
1871 while (seq_id && cur_seq_id < seq_id) {
1872 spin_lock_irqsave(&sbi->fsync_node_lock, flags);
1873 if (list_empty(head)) {
1874 spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
1877 fn = list_first_entry(head, struct fsync_node_entry, list);
1878 if (fn->seq_id > seq_id) {
1879 spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
1882 cur_seq_id = fn->seq_id;
1885 spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
1887 f2fs_wait_on_page_writeback(page, NODE, true);
1888 if (TestClearPageError(page))
1897 ret2 = filemap_check_errors(NODE_MAPPING(sbi));
1904 static int f2fs_write_node_pages(struct address_space *mapping,
1905 struct writeback_control *wbc)
1907 struct f2fs_sb_info *sbi = F2FS_M_SB(mapping);
1908 struct blk_plug plug;
1911 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
1914 /* balancing f2fs's metadata in background */
1915 f2fs_balance_fs_bg(sbi);
1917 /* collect a number of dirty node pages and write together */
1918 if (get_pages(sbi, F2FS_DIRTY_NODES) < nr_pages_to_skip(sbi, NODE))
1921 if (wbc->sync_mode == WB_SYNC_ALL)
1922 atomic_inc(&sbi->wb_sync_req[NODE]);
1923 else if (atomic_read(&sbi->wb_sync_req[NODE]))
1926 trace_f2fs_writepages(mapping->host, wbc, NODE);
1928 diff = nr_pages_to_write(sbi, NODE, wbc);
1929 blk_start_plug(&plug);
1930 f2fs_sync_node_pages(sbi, wbc, true, FS_NODE_IO);
1931 blk_finish_plug(&plug);
1932 wbc->nr_to_write = max((long)0, wbc->nr_to_write - diff);
1934 if (wbc->sync_mode == WB_SYNC_ALL)
1935 atomic_dec(&sbi->wb_sync_req[NODE]);
1939 wbc->pages_skipped += get_pages(sbi, F2FS_DIRTY_NODES);
1940 trace_f2fs_writepages(mapping->host, wbc, NODE);
1944 static int f2fs_set_node_page_dirty(struct page *page)
1946 trace_f2fs_set_page_dirty(page, NODE);
1948 if (!PageUptodate(page))
1949 SetPageUptodate(page);
1950 #ifdef CONFIG_F2FS_CHECK_FS
1952 f2fs_inode_chksum_set(F2FS_P_SB(page), page);
1954 if (!PageDirty(page)) {
1955 __set_page_dirty_nobuffers(page);
1956 inc_page_count(F2FS_P_SB(page), F2FS_DIRTY_NODES);
1957 SetPagePrivate(page);
1958 f2fs_trace_pid(page);
1965 * Structure of the f2fs node operations
1967 const struct address_space_operations f2fs_node_aops = {
1968 .writepage = f2fs_write_node_page,
1969 .writepages = f2fs_write_node_pages,
1970 .set_page_dirty = f2fs_set_node_page_dirty,
1971 .invalidatepage = f2fs_invalidate_page,
1972 .releasepage = f2fs_release_page,
1973 #ifdef CONFIG_MIGRATION
1974 .migratepage = f2fs_migrate_page,
1978 static struct free_nid *__lookup_free_nid_list(struct f2fs_nm_info *nm_i,
1981 return radix_tree_lookup(&nm_i->free_nid_root, n);
1984 static int __insert_free_nid(struct f2fs_sb_info *sbi,
1985 struct free_nid *i, enum nid_state state)
1987 struct f2fs_nm_info *nm_i = NM_I(sbi);
1989 int err = radix_tree_insert(&nm_i->free_nid_root, i->nid, i);
1993 f2fs_bug_on(sbi, state != i->state);
1994 nm_i->nid_cnt[state]++;
1995 if (state == FREE_NID)
1996 list_add_tail(&i->list, &nm_i->free_nid_list);
2000 static void __remove_free_nid(struct f2fs_sb_info *sbi,
2001 struct free_nid *i, enum nid_state state)
2003 struct f2fs_nm_info *nm_i = NM_I(sbi);
2005 f2fs_bug_on(sbi, state != i->state);
2006 nm_i->nid_cnt[state]--;
2007 if (state == FREE_NID)
2009 radix_tree_delete(&nm_i->free_nid_root, i->nid);
2012 static void __move_free_nid(struct f2fs_sb_info *sbi, struct free_nid *i,
2013 enum nid_state org_state, enum nid_state dst_state)
2015 struct f2fs_nm_info *nm_i = NM_I(sbi);
2017 f2fs_bug_on(sbi, org_state != i->state);
2018 i->state = dst_state;
2019 nm_i->nid_cnt[org_state]--;
2020 nm_i->nid_cnt[dst_state]++;
2022 switch (dst_state) {
2027 list_add_tail(&i->list, &nm_i->free_nid_list);
2034 static void update_free_nid_bitmap(struct f2fs_sb_info *sbi, nid_t nid,
2035 bool set, bool build)
2037 struct f2fs_nm_info *nm_i = NM_I(sbi);
2038 unsigned int nat_ofs = NAT_BLOCK_OFFSET(nid);
2039 unsigned int nid_ofs = nid - START_NID(nid);
2041 if (!test_bit_le(nat_ofs, nm_i->nat_block_bitmap))
2045 if (test_bit_le(nid_ofs, nm_i->free_nid_bitmap[nat_ofs]))
2047 __set_bit_le(nid_ofs, nm_i->free_nid_bitmap[nat_ofs]);
2048 nm_i->free_nid_count[nat_ofs]++;
2050 if (!test_bit_le(nid_ofs, nm_i->free_nid_bitmap[nat_ofs]))
2052 __clear_bit_le(nid_ofs, nm_i->free_nid_bitmap[nat_ofs]);
2054 nm_i->free_nid_count[nat_ofs]--;
2058 /* return if the nid is recognized as free */
2059 static bool add_free_nid(struct f2fs_sb_info *sbi,
2060 nid_t nid, bool build, bool update)
2062 struct f2fs_nm_info *nm_i = NM_I(sbi);
2063 struct free_nid *i, *e;
2064 struct nat_entry *ne;
2068 /* 0 nid should not be used */
2069 if (unlikely(nid == 0))
2072 i = f2fs_kmem_cache_alloc(free_nid_slab, GFP_NOFS);
2074 i->state = FREE_NID;
2076 radix_tree_preload(GFP_NOFS | __GFP_NOFAIL);
2078 spin_lock(&nm_i->nid_list_lock);
2086 * - __insert_nid_to_list(PREALLOC_NID)
2087 * - f2fs_balance_fs_bg
2088 * - f2fs_build_free_nids
2089 * - __f2fs_build_free_nids
2092 * - __lookup_nat_cache
2094 * - f2fs_init_inode_metadata
2095 * - f2fs_new_inode_page
2096 * - f2fs_new_node_page
2098 * - f2fs_alloc_nid_done
2099 * - __remove_nid_from_list(PREALLOC_NID)
2100 * - __insert_nid_to_list(FREE_NID)
2102 ne = __lookup_nat_cache(nm_i, nid);
2103 if (ne && (!get_nat_flag(ne, IS_CHECKPOINTED) ||
2104 nat_get_blkaddr(ne) != NULL_ADDR))
2107 e = __lookup_free_nid_list(nm_i, nid);
2109 if (e->state == FREE_NID)
2115 err = __insert_free_nid(sbi, i, FREE_NID);
2118 update_free_nid_bitmap(sbi, nid, ret, build);
2120 nm_i->available_nids++;
2122 spin_unlock(&nm_i->nid_list_lock);
2123 radix_tree_preload_end();
2126 kmem_cache_free(free_nid_slab, i);
2130 static void remove_free_nid(struct f2fs_sb_info *sbi, nid_t nid)
2132 struct f2fs_nm_info *nm_i = NM_I(sbi);
2134 bool need_free = false;
2136 spin_lock(&nm_i->nid_list_lock);
2137 i = __lookup_free_nid_list(nm_i, nid);
2138 if (i && i->state == FREE_NID) {
2139 __remove_free_nid(sbi, i, FREE_NID);
2142 spin_unlock(&nm_i->nid_list_lock);
2145 kmem_cache_free(free_nid_slab, i);
2148 static int scan_nat_page(struct f2fs_sb_info *sbi,
2149 struct page *nat_page, nid_t start_nid)
2151 struct f2fs_nm_info *nm_i = NM_I(sbi);
2152 struct f2fs_nat_block *nat_blk = page_address(nat_page);
2154 unsigned int nat_ofs = NAT_BLOCK_OFFSET(start_nid);
2157 __set_bit_le(nat_ofs, nm_i->nat_block_bitmap);
2159 i = start_nid % NAT_ENTRY_PER_BLOCK;
2161 for (; i < NAT_ENTRY_PER_BLOCK; i++, start_nid++) {
2162 if (unlikely(start_nid >= nm_i->max_nid))
2165 blk_addr = le32_to_cpu(nat_blk->entries[i].block_addr);
2167 if (blk_addr == NEW_ADDR)
2170 if (blk_addr == NULL_ADDR) {
2171 add_free_nid(sbi, start_nid, true, true);
2173 spin_lock(&NM_I(sbi)->nid_list_lock);
2174 update_free_nid_bitmap(sbi, start_nid, false, true);
2175 spin_unlock(&NM_I(sbi)->nid_list_lock);
2182 static void scan_curseg_cache(struct f2fs_sb_info *sbi)
2184 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
2185 struct f2fs_journal *journal = curseg->journal;
2188 down_read(&curseg->journal_rwsem);
2189 for (i = 0; i < nats_in_cursum(journal); i++) {
2193 addr = le32_to_cpu(nat_in_journal(journal, i).block_addr);
2194 nid = le32_to_cpu(nid_in_journal(journal, i));
2195 if (addr == NULL_ADDR)
2196 add_free_nid(sbi, nid, true, false);
2198 remove_free_nid(sbi, nid);
2200 up_read(&curseg->journal_rwsem);
2203 static void scan_free_nid_bits(struct f2fs_sb_info *sbi)
2205 struct f2fs_nm_info *nm_i = NM_I(sbi);
2206 unsigned int i, idx;
2209 down_read(&nm_i->nat_tree_lock);
2211 for (i = 0; i < nm_i->nat_blocks; i++) {
2212 if (!test_bit_le(i, nm_i->nat_block_bitmap))
2214 if (!nm_i->free_nid_count[i])
2216 for (idx = 0; idx < NAT_ENTRY_PER_BLOCK; idx++) {
2217 idx = find_next_bit_le(nm_i->free_nid_bitmap[i],
2218 NAT_ENTRY_PER_BLOCK, idx);
2219 if (idx >= NAT_ENTRY_PER_BLOCK)
2222 nid = i * NAT_ENTRY_PER_BLOCK + idx;
2223 add_free_nid(sbi, nid, true, false);
2225 if (nm_i->nid_cnt[FREE_NID] >= MAX_FREE_NIDS)
2230 scan_curseg_cache(sbi);
2232 up_read(&nm_i->nat_tree_lock);
2235 static int __f2fs_build_free_nids(struct f2fs_sb_info *sbi,
2236 bool sync, bool mount)
2238 struct f2fs_nm_info *nm_i = NM_I(sbi);
2240 nid_t nid = nm_i->next_scan_nid;
2242 if (unlikely(nid >= nm_i->max_nid))
2245 /* Enough entries */
2246 if (nm_i->nid_cnt[FREE_NID] >= NAT_ENTRY_PER_BLOCK)
2249 if (!sync && !f2fs_available_free_memory(sbi, FREE_NIDS))
2253 /* try to find free nids in free_nid_bitmap */
2254 scan_free_nid_bits(sbi);
2256 if (nm_i->nid_cnt[FREE_NID] >= NAT_ENTRY_PER_BLOCK)
2260 /* readahead nat pages to be scanned */
2261 f2fs_ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nid), FREE_NID_PAGES,
2264 down_read(&nm_i->nat_tree_lock);
2267 if (!test_bit_le(NAT_BLOCK_OFFSET(nid),
2268 nm_i->nat_block_bitmap)) {
2269 struct page *page = get_current_nat_page(sbi, nid);
2272 ret = PTR_ERR(page);
2274 ret = scan_nat_page(sbi, page, nid);
2275 f2fs_put_page(page, 1);
2279 up_read(&nm_i->nat_tree_lock);
2280 f2fs_bug_on(sbi, !mount);
2281 f2fs_msg(sbi->sb, KERN_ERR,
2282 "NAT is corrupt, run fsck to fix it");
2287 nid += (NAT_ENTRY_PER_BLOCK - (nid % NAT_ENTRY_PER_BLOCK));
2288 if (unlikely(nid >= nm_i->max_nid))
2291 if (++i >= FREE_NID_PAGES)
2295 /* go to the next free nat pages to find free nids abundantly */
2296 nm_i->next_scan_nid = nid;
2298 /* find free nids from current sum_pages */
2299 scan_curseg_cache(sbi);
2301 up_read(&nm_i->nat_tree_lock);
2303 f2fs_ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nm_i->next_scan_nid),
2304 nm_i->ra_nid_pages, META_NAT, false);
2309 int f2fs_build_free_nids(struct f2fs_sb_info *sbi, bool sync, bool mount)
2313 mutex_lock(&NM_I(sbi)->build_lock);
2314 ret = __f2fs_build_free_nids(sbi, sync, mount);
2315 mutex_unlock(&NM_I(sbi)->build_lock);
2321 * If this function returns success, caller can obtain a new nid
2322 * from second parameter of this function.
2323 * The returned nid could be used ino as well as nid when inode is created.
2325 bool f2fs_alloc_nid(struct f2fs_sb_info *sbi, nid_t *nid)
2327 struct f2fs_nm_info *nm_i = NM_I(sbi);
2328 struct free_nid *i = NULL;
2330 if (time_to_inject(sbi, FAULT_ALLOC_NID)) {
2331 f2fs_show_injection_info(FAULT_ALLOC_NID);
2335 spin_lock(&nm_i->nid_list_lock);
2337 if (unlikely(nm_i->available_nids == 0)) {
2338 spin_unlock(&nm_i->nid_list_lock);
2342 /* We should not use stale free nids created by f2fs_build_free_nids */
2343 if (nm_i->nid_cnt[FREE_NID] && !on_f2fs_build_free_nids(nm_i)) {
2344 f2fs_bug_on(sbi, list_empty(&nm_i->free_nid_list));
2345 i = list_first_entry(&nm_i->free_nid_list,
2346 struct free_nid, list);
2349 __move_free_nid(sbi, i, FREE_NID, PREALLOC_NID);
2350 nm_i->available_nids--;
2352 update_free_nid_bitmap(sbi, *nid, false, false);
2354 spin_unlock(&nm_i->nid_list_lock);
2357 spin_unlock(&nm_i->nid_list_lock);
2359 /* Let's scan nat pages and its caches to get free nids */
2360 if (!f2fs_build_free_nids(sbi, true, false))
2366 * f2fs_alloc_nid() should be called prior to this function.
2368 void f2fs_alloc_nid_done(struct f2fs_sb_info *sbi, nid_t nid)
2370 struct f2fs_nm_info *nm_i = NM_I(sbi);
2373 spin_lock(&nm_i->nid_list_lock);
2374 i = __lookup_free_nid_list(nm_i, nid);
2375 f2fs_bug_on(sbi, !i);
2376 __remove_free_nid(sbi, i, PREALLOC_NID);
2377 spin_unlock(&nm_i->nid_list_lock);
2379 kmem_cache_free(free_nid_slab, i);
2383 * f2fs_alloc_nid() should be called prior to this function.
2385 void f2fs_alloc_nid_failed(struct f2fs_sb_info *sbi, nid_t nid)
2387 struct f2fs_nm_info *nm_i = NM_I(sbi);
2389 bool need_free = false;
2394 spin_lock(&nm_i->nid_list_lock);
2395 i = __lookup_free_nid_list(nm_i, nid);
2396 f2fs_bug_on(sbi, !i);
2398 if (!f2fs_available_free_memory(sbi, FREE_NIDS)) {
2399 __remove_free_nid(sbi, i, PREALLOC_NID);
2402 __move_free_nid(sbi, i, PREALLOC_NID, FREE_NID);
2405 nm_i->available_nids++;
2407 update_free_nid_bitmap(sbi, nid, true, false);
2409 spin_unlock(&nm_i->nid_list_lock);
2412 kmem_cache_free(free_nid_slab, i);
2415 int f2fs_try_to_free_nids(struct f2fs_sb_info *sbi, int nr_shrink)
2417 struct f2fs_nm_info *nm_i = NM_I(sbi);
2418 struct free_nid *i, *next;
2421 if (nm_i->nid_cnt[FREE_NID] <= MAX_FREE_NIDS)
2424 if (!mutex_trylock(&nm_i->build_lock))
2427 spin_lock(&nm_i->nid_list_lock);
2428 list_for_each_entry_safe(i, next, &nm_i->free_nid_list, list) {
2429 if (nr_shrink <= 0 ||
2430 nm_i->nid_cnt[FREE_NID] <= MAX_FREE_NIDS)
2433 __remove_free_nid(sbi, i, FREE_NID);
2434 kmem_cache_free(free_nid_slab, i);
2437 spin_unlock(&nm_i->nid_list_lock);
2438 mutex_unlock(&nm_i->build_lock);
2440 return nr - nr_shrink;
2443 void f2fs_recover_inline_xattr(struct inode *inode, struct page *page)
2445 void *src_addr, *dst_addr;
2448 struct f2fs_inode *ri;
2450 ipage = f2fs_get_node_page(F2FS_I_SB(inode), inode->i_ino);
2451 f2fs_bug_on(F2FS_I_SB(inode), IS_ERR(ipage));
2453 ri = F2FS_INODE(page);
2454 if (ri->i_inline & F2FS_INLINE_XATTR) {
2455 set_inode_flag(inode, FI_INLINE_XATTR);
2457 clear_inode_flag(inode, FI_INLINE_XATTR);
2461 dst_addr = inline_xattr_addr(inode, ipage);
2462 src_addr = inline_xattr_addr(inode, page);
2463 inline_size = inline_xattr_size(inode);
2465 f2fs_wait_on_page_writeback(ipage, NODE, true);
2466 memcpy(dst_addr, src_addr, inline_size);
2468 f2fs_update_inode(inode, ipage);
2469 f2fs_put_page(ipage, 1);
2472 int f2fs_recover_xattr_data(struct inode *inode, struct page *page)
2474 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
2475 nid_t prev_xnid = F2FS_I(inode)->i_xattr_nid;
2477 struct dnode_of_data dn;
2478 struct node_info ni;
2485 /* 1: invalidate the previous xattr nid */
2486 err = f2fs_get_node_info(sbi, prev_xnid, &ni);
2490 f2fs_invalidate_blocks(sbi, ni.blk_addr);
2491 dec_valid_node_count(sbi, inode, false);
2492 set_node_addr(sbi, &ni, NULL_ADDR, false);
2495 /* 2: update xattr nid in inode */
2496 if (!f2fs_alloc_nid(sbi, &new_xnid))
2499 set_new_dnode(&dn, inode, NULL, NULL, new_xnid);
2500 xpage = f2fs_new_node_page(&dn, XATTR_NODE_OFFSET);
2501 if (IS_ERR(xpage)) {
2502 f2fs_alloc_nid_failed(sbi, new_xnid);
2503 return PTR_ERR(xpage);
2506 f2fs_alloc_nid_done(sbi, new_xnid);
2507 f2fs_update_inode_page(inode);
2509 /* 3: update and set xattr node page dirty */
2510 memcpy(F2FS_NODE(xpage), F2FS_NODE(page), VALID_XATTR_BLOCK_SIZE);
2512 set_page_dirty(xpage);
2513 f2fs_put_page(xpage, 1);
2518 int f2fs_recover_inode_page(struct f2fs_sb_info *sbi, struct page *page)
2520 struct f2fs_inode *src, *dst;
2521 nid_t ino = ino_of_node(page);
2522 struct node_info old_ni, new_ni;
2526 err = f2fs_get_node_info(sbi, ino, &old_ni);
2530 if (unlikely(old_ni.blk_addr != NULL_ADDR))
2533 ipage = f2fs_grab_cache_page(NODE_MAPPING(sbi), ino, false);
2535 congestion_wait(BLK_RW_ASYNC, HZ/50);
2539 /* Should not use this inode from free nid list */
2540 remove_free_nid(sbi, ino);
2542 if (!PageUptodate(ipage))
2543 SetPageUptodate(ipage);
2544 fill_node_footer(ipage, ino, ino, 0, true);
2545 set_cold_node(page, false);
2547 src = F2FS_INODE(page);
2548 dst = F2FS_INODE(ipage);
2550 memcpy(dst, src, (unsigned long)&src->i_ext - (unsigned long)src);
2552 dst->i_blocks = cpu_to_le64(1);
2553 dst->i_links = cpu_to_le32(1);
2554 dst->i_xattr_nid = 0;
2555 dst->i_inline = src->i_inline & (F2FS_INLINE_XATTR | F2FS_EXTRA_ATTR);
2556 if (dst->i_inline & F2FS_EXTRA_ATTR) {
2557 dst->i_extra_isize = src->i_extra_isize;
2559 if (f2fs_sb_has_flexible_inline_xattr(sbi->sb) &&
2560 F2FS_FITS_IN_INODE(src, le16_to_cpu(src->i_extra_isize),
2561 i_inline_xattr_size))
2562 dst->i_inline_xattr_size = src->i_inline_xattr_size;
2564 if (f2fs_sb_has_project_quota(sbi->sb) &&
2565 F2FS_FITS_IN_INODE(src, le16_to_cpu(src->i_extra_isize),
2567 dst->i_projid = src->i_projid;
2569 if (f2fs_sb_has_inode_crtime(sbi->sb) &&
2570 F2FS_FITS_IN_INODE(src, le16_to_cpu(src->i_extra_isize),
2572 dst->i_crtime = src->i_crtime;
2573 dst->i_crtime_nsec = src->i_crtime_nsec;
2580 if (unlikely(inc_valid_node_count(sbi, NULL, true)))
2582 set_node_addr(sbi, &new_ni, NEW_ADDR, false);
2583 inc_valid_inode_count(sbi);
2584 set_page_dirty(ipage);
2585 f2fs_put_page(ipage, 1);
2589 int f2fs_restore_node_summary(struct f2fs_sb_info *sbi,
2590 unsigned int segno, struct f2fs_summary_block *sum)
2592 struct f2fs_node *rn;
2593 struct f2fs_summary *sum_entry;
2595 int i, idx, last_offset, nrpages;
2597 /* scan the node segment */
2598 last_offset = sbi->blocks_per_seg;
2599 addr = START_BLOCK(sbi, segno);
2600 sum_entry = &sum->entries[0];
2602 for (i = 0; i < last_offset; i += nrpages, addr += nrpages) {
2603 nrpages = min(last_offset - i, BIO_MAX_PAGES);
2605 /* readahead node pages */
2606 f2fs_ra_meta_pages(sbi, addr, nrpages, META_POR, true);
2608 for (idx = addr; idx < addr + nrpages; idx++) {
2609 struct page *page = f2fs_get_tmp_page(sbi, idx);
2612 return PTR_ERR(page);
2614 rn = F2FS_NODE(page);
2615 sum_entry->nid = rn->footer.nid;
2616 sum_entry->version = 0;
2617 sum_entry->ofs_in_node = 0;
2619 f2fs_put_page(page, 1);
2622 invalidate_mapping_pages(META_MAPPING(sbi), addr,
2628 static void remove_nats_in_journal(struct f2fs_sb_info *sbi)
2630 struct f2fs_nm_info *nm_i = NM_I(sbi);
2631 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
2632 struct f2fs_journal *journal = curseg->journal;
2635 down_write(&curseg->journal_rwsem);
2636 for (i = 0; i < nats_in_cursum(journal); i++) {
2637 struct nat_entry *ne;
2638 struct f2fs_nat_entry raw_ne;
2639 nid_t nid = le32_to_cpu(nid_in_journal(journal, i));
2641 raw_ne = nat_in_journal(journal, i);
2643 ne = __lookup_nat_cache(nm_i, nid);
2645 ne = __alloc_nat_entry(nid, true);
2646 __init_nat_entry(nm_i, ne, &raw_ne, true);
2650 * if a free nat in journal has not been used after last
2651 * checkpoint, we should remove it from available nids,
2652 * since later we will add it again.
2654 if (!get_nat_flag(ne, IS_DIRTY) &&
2655 le32_to_cpu(raw_ne.block_addr) == NULL_ADDR) {
2656 spin_lock(&nm_i->nid_list_lock);
2657 nm_i->available_nids--;
2658 spin_unlock(&nm_i->nid_list_lock);
2661 __set_nat_cache_dirty(nm_i, ne);
2663 update_nats_in_cursum(journal, -i);
2664 up_write(&curseg->journal_rwsem);
2667 static void __adjust_nat_entry_set(struct nat_entry_set *nes,
2668 struct list_head *head, int max)
2670 struct nat_entry_set *cur;
2672 if (nes->entry_cnt >= max)
2675 list_for_each_entry(cur, head, set_list) {
2676 if (cur->entry_cnt >= nes->entry_cnt) {
2677 list_add(&nes->set_list, cur->set_list.prev);
2682 list_add_tail(&nes->set_list, head);
2685 static void __update_nat_bits(struct f2fs_sb_info *sbi, nid_t start_nid,
2688 struct f2fs_nm_info *nm_i = NM_I(sbi);
2689 unsigned int nat_index = start_nid / NAT_ENTRY_PER_BLOCK;
2690 struct f2fs_nat_block *nat_blk = page_address(page);
2694 if (!enabled_nat_bits(sbi, NULL))
2697 if (nat_index == 0) {
2701 for (; i < NAT_ENTRY_PER_BLOCK; i++) {
2702 if (nat_blk->entries[i].block_addr != NULL_ADDR)
2706 __set_bit_le(nat_index, nm_i->empty_nat_bits);
2707 __clear_bit_le(nat_index, nm_i->full_nat_bits);
2711 __clear_bit_le(nat_index, nm_i->empty_nat_bits);
2712 if (valid == NAT_ENTRY_PER_BLOCK)
2713 __set_bit_le(nat_index, nm_i->full_nat_bits);
2715 __clear_bit_le(nat_index, nm_i->full_nat_bits);
2718 static int __flush_nat_entry_set(struct f2fs_sb_info *sbi,
2719 struct nat_entry_set *set, struct cp_control *cpc)
2721 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
2722 struct f2fs_journal *journal = curseg->journal;
2723 nid_t start_nid = set->set * NAT_ENTRY_PER_BLOCK;
2724 bool to_journal = true;
2725 struct f2fs_nat_block *nat_blk;
2726 struct nat_entry *ne, *cur;
2727 struct page *page = NULL;
2730 * there are two steps to flush nat entries:
2731 * #1, flush nat entries to journal in current hot data summary block.
2732 * #2, flush nat entries to nat page.
2734 if (enabled_nat_bits(sbi, cpc) ||
2735 !__has_cursum_space(journal, set->entry_cnt, NAT_JOURNAL))
2739 down_write(&curseg->journal_rwsem);
2741 page = get_next_nat_page(sbi, start_nid);
2743 return PTR_ERR(page);
2745 nat_blk = page_address(page);
2746 f2fs_bug_on(sbi, !nat_blk);
2749 /* flush dirty nats in nat entry set */
2750 list_for_each_entry_safe(ne, cur, &set->entry_list, list) {
2751 struct f2fs_nat_entry *raw_ne;
2752 nid_t nid = nat_get_nid(ne);
2755 f2fs_bug_on(sbi, nat_get_blkaddr(ne) == NEW_ADDR);
2758 offset = f2fs_lookup_journal_in_cursum(journal,
2759 NAT_JOURNAL, nid, 1);
2760 f2fs_bug_on(sbi, offset < 0);
2761 raw_ne = &nat_in_journal(journal, offset);
2762 nid_in_journal(journal, offset) = cpu_to_le32(nid);
2764 raw_ne = &nat_blk->entries[nid - start_nid];
2766 raw_nat_from_node_info(raw_ne, &ne->ni);
2768 __clear_nat_cache_dirty(NM_I(sbi), set, ne);
2769 if (nat_get_blkaddr(ne) == NULL_ADDR) {
2770 add_free_nid(sbi, nid, false, true);
2772 spin_lock(&NM_I(sbi)->nid_list_lock);
2773 update_free_nid_bitmap(sbi, nid, false, false);
2774 spin_unlock(&NM_I(sbi)->nid_list_lock);
2779 up_write(&curseg->journal_rwsem);
2781 __update_nat_bits(sbi, start_nid, page);
2782 f2fs_put_page(page, 1);
2785 /* Allow dirty nats by node block allocation in write_begin */
2786 if (!set->entry_cnt) {
2787 radix_tree_delete(&NM_I(sbi)->nat_set_root, set->set);
2788 kmem_cache_free(nat_entry_set_slab, set);
2794 * This function is called during the checkpointing process.
2796 int f2fs_flush_nat_entries(struct f2fs_sb_info *sbi, struct cp_control *cpc)
2798 struct f2fs_nm_info *nm_i = NM_I(sbi);
2799 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
2800 struct f2fs_journal *journal = curseg->journal;
2801 struct nat_entry_set *setvec[SETVEC_SIZE];
2802 struct nat_entry_set *set, *tmp;
2808 /* during unmount, let's flush nat_bits before checking dirty_nat_cnt */
2809 if (enabled_nat_bits(sbi, cpc)) {
2810 down_write(&nm_i->nat_tree_lock);
2811 remove_nats_in_journal(sbi);
2812 up_write(&nm_i->nat_tree_lock);
2815 if (!nm_i->dirty_nat_cnt)
2818 down_write(&nm_i->nat_tree_lock);
2821 * if there are no enough space in journal to store dirty nat
2822 * entries, remove all entries from journal and merge them
2823 * into nat entry set.
2825 if (enabled_nat_bits(sbi, cpc) ||
2826 !__has_cursum_space(journal, nm_i->dirty_nat_cnt, NAT_JOURNAL))
2827 remove_nats_in_journal(sbi);
2829 while ((found = __gang_lookup_nat_set(nm_i,
2830 set_idx, SETVEC_SIZE, setvec))) {
2832 set_idx = setvec[found - 1]->set + 1;
2833 for (idx = 0; idx < found; idx++)
2834 __adjust_nat_entry_set(setvec[idx], &sets,
2835 MAX_NAT_JENTRIES(journal));
2838 /* flush dirty nats in nat entry set */
2839 list_for_each_entry_safe(set, tmp, &sets, set_list) {
2840 err = __flush_nat_entry_set(sbi, set, cpc);
2845 up_write(&nm_i->nat_tree_lock);
2846 /* Allow dirty nats by node block allocation in write_begin */
2851 static int __get_nat_bitmaps(struct f2fs_sb_info *sbi)
2853 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
2854 struct f2fs_nm_info *nm_i = NM_I(sbi);
2855 unsigned int nat_bits_bytes = nm_i->nat_blocks / BITS_PER_BYTE;
2857 __u64 cp_ver = cur_cp_version(ckpt);
2858 block_t nat_bits_addr;
2860 if (!enabled_nat_bits(sbi, NULL))
2863 nm_i->nat_bits_blocks = F2FS_BLK_ALIGN((nat_bits_bytes << 1) + 8);
2864 nm_i->nat_bits = f2fs_kzalloc(sbi,
2865 nm_i->nat_bits_blocks << F2FS_BLKSIZE_BITS, GFP_KERNEL);
2866 if (!nm_i->nat_bits)
2869 nat_bits_addr = __start_cp_addr(sbi) + sbi->blocks_per_seg -
2870 nm_i->nat_bits_blocks;
2871 for (i = 0; i < nm_i->nat_bits_blocks; i++) {
2874 page = f2fs_get_meta_page(sbi, nat_bits_addr++);
2876 disable_nat_bits(sbi, true);
2877 return PTR_ERR(page);
2880 memcpy(nm_i->nat_bits + (i << F2FS_BLKSIZE_BITS),
2881 page_address(page), F2FS_BLKSIZE);
2882 f2fs_put_page(page, 1);
2885 cp_ver |= (cur_cp_crc(ckpt) << 32);
2886 if (cpu_to_le64(cp_ver) != *(__le64 *)nm_i->nat_bits) {
2887 disable_nat_bits(sbi, true);
2891 nm_i->full_nat_bits = nm_i->nat_bits + 8;
2892 nm_i->empty_nat_bits = nm_i->full_nat_bits + nat_bits_bytes;
2894 f2fs_msg(sbi->sb, KERN_NOTICE, "Found nat_bits in checkpoint");
2898 static inline void load_free_nid_bitmap(struct f2fs_sb_info *sbi)
2900 struct f2fs_nm_info *nm_i = NM_I(sbi);
2902 nid_t nid, last_nid;
2904 if (!enabled_nat_bits(sbi, NULL))
2907 for (i = 0; i < nm_i->nat_blocks; i++) {
2908 i = find_next_bit_le(nm_i->empty_nat_bits, nm_i->nat_blocks, i);
2909 if (i >= nm_i->nat_blocks)
2912 __set_bit_le(i, nm_i->nat_block_bitmap);
2914 nid = i * NAT_ENTRY_PER_BLOCK;
2915 last_nid = nid + NAT_ENTRY_PER_BLOCK;
2917 spin_lock(&NM_I(sbi)->nid_list_lock);
2918 for (; nid < last_nid; nid++)
2919 update_free_nid_bitmap(sbi, nid, true, true);
2920 spin_unlock(&NM_I(sbi)->nid_list_lock);
2923 for (i = 0; i < nm_i->nat_blocks; i++) {
2924 i = find_next_bit_le(nm_i->full_nat_bits, nm_i->nat_blocks, i);
2925 if (i >= nm_i->nat_blocks)
2928 __set_bit_le(i, nm_i->nat_block_bitmap);
2932 static int init_node_manager(struct f2fs_sb_info *sbi)
2934 struct f2fs_super_block *sb_raw = F2FS_RAW_SUPER(sbi);
2935 struct f2fs_nm_info *nm_i = NM_I(sbi);
2936 unsigned char *version_bitmap;
2937 unsigned int nat_segs;
2940 nm_i->nat_blkaddr = le32_to_cpu(sb_raw->nat_blkaddr);
2942 /* segment_count_nat includes pair segment so divide to 2. */
2943 nat_segs = le32_to_cpu(sb_raw->segment_count_nat) >> 1;
2944 nm_i->nat_blocks = nat_segs << le32_to_cpu(sb_raw->log_blocks_per_seg);
2945 nm_i->max_nid = NAT_ENTRY_PER_BLOCK * nm_i->nat_blocks;
2947 /* not used nids: 0, node, meta, (and root counted as valid node) */
2948 nm_i->available_nids = nm_i->max_nid - sbi->total_valid_node_count -
2949 sbi->nquota_files - F2FS_RESERVED_NODE_NUM;
2950 nm_i->nid_cnt[FREE_NID] = 0;
2951 nm_i->nid_cnt[PREALLOC_NID] = 0;
2953 nm_i->ram_thresh = DEF_RAM_THRESHOLD;
2954 nm_i->ra_nid_pages = DEF_RA_NID_PAGES;
2955 nm_i->dirty_nats_ratio = DEF_DIRTY_NAT_RATIO_THRESHOLD;
2957 INIT_RADIX_TREE(&nm_i->free_nid_root, GFP_ATOMIC);
2958 INIT_LIST_HEAD(&nm_i->free_nid_list);
2959 INIT_RADIX_TREE(&nm_i->nat_root, GFP_NOIO);
2960 INIT_RADIX_TREE(&nm_i->nat_set_root, GFP_NOIO);
2961 INIT_LIST_HEAD(&nm_i->nat_entries);
2962 spin_lock_init(&nm_i->nat_list_lock);
2964 mutex_init(&nm_i->build_lock);
2965 spin_lock_init(&nm_i->nid_list_lock);
2966 init_rwsem(&nm_i->nat_tree_lock);
2968 nm_i->next_scan_nid = le32_to_cpu(sbi->ckpt->next_free_nid);
2969 nm_i->bitmap_size = __bitmap_size(sbi, NAT_BITMAP);
2970 version_bitmap = __bitmap_ptr(sbi, NAT_BITMAP);
2971 if (!version_bitmap)
2974 nm_i->nat_bitmap = kmemdup(version_bitmap, nm_i->bitmap_size,
2976 if (!nm_i->nat_bitmap)
2979 err = __get_nat_bitmaps(sbi);
2983 #ifdef CONFIG_F2FS_CHECK_FS
2984 nm_i->nat_bitmap_mir = kmemdup(version_bitmap, nm_i->bitmap_size,
2986 if (!nm_i->nat_bitmap_mir)
2993 static int init_free_nid_cache(struct f2fs_sb_info *sbi)
2995 struct f2fs_nm_info *nm_i = NM_I(sbi);
2998 nm_i->free_nid_bitmap =
2999 f2fs_kzalloc(sbi, array_size(sizeof(unsigned char *),
3002 if (!nm_i->free_nid_bitmap)
3005 for (i = 0; i < nm_i->nat_blocks; i++) {
3006 nm_i->free_nid_bitmap[i] = f2fs_kvzalloc(sbi,
3007 f2fs_bitmap_size(NAT_ENTRY_PER_BLOCK), GFP_KERNEL);
3008 if (!nm_i->free_nid_bitmap[i])
3012 nm_i->nat_block_bitmap = f2fs_kvzalloc(sbi, nm_i->nat_blocks / 8,
3014 if (!nm_i->nat_block_bitmap)
3017 nm_i->free_nid_count =
3018 f2fs_kvzalloc(sbi, array_size(sizeof(unsigned short),
3021 if (!nm_i->free_nid_count)
3026 int f2fs_build_node_manager(struct f2fs_sb_info *sbi)
3030 sbi->nm_info = f2fs_kzalloc(sbi, sizeof(struct f2fs_nm_info),
3035 err = init_node_manager(sbi);
3039 err = init_free_nid_cache(sbi);
3043 /* load free nid status from nat_bits table */
3044 load_free_nid_bitmap(sbi);
3046 return f2fs_build_free_nids(sbi, true, true);
3049 void f2fs_destroy_node_manager(struct f2fs_sb_info *sbi)
3051 struct f2fs_nm_info *nm_i = NM_I(sbi);
3052 struct free_nid *i, *next_i;
3053 struct nat_entry *natvec[NATVEC_SIZE];
3054 struct nat_entry_set *setvec[SETVEC_SIZE];
3061 /* destroy free nid list */
3062 spin_lock(&nm_i->nid_list_lock);
3063 list_for_each_entry_safe(i, next_i, &nm_i->free_nid_list, list) {
3064 __remove_free_nid(sbi, i, FREE_NID);
3065 spin_unlock(&nm_i->nid_list_lock);
3066 kmem_cache_free(free_nid_slab, i);
3067 spin_lock(&nm_i->nid_list_lock);
3069 f2fs_bug_on(sbi, nm_i->nid_cnt[FREE_NID]);
3070 f2fs_bug_on(sbi, nm_i->nid_cnt[PREALLOC_NID]);
3071 f2fs_bug_on(sbi, !list_empty(&nm_i->free_nid_list));
3072 spin_unlock(&nm_i->nid_list_lock);
3074 /* destroy nat cache */
3075 down_write(&nm_i->nat_tree_lock);
3076 while ((found = __gang_lookup_nat_cache(nm_i,
3077 nid, NATVEC_SIZE, natvec))) {
3080 nid = nat_get_nid(natvec[found - 1]) + 1;
3081 for (idx = 0; idx < found; idx++) {
3082 spin_lock(&nm_i->nat_list_lock);
3083 list_del(&natvec[idx]->list);
3084 spin_unlock(&nm_i->nat_list_lock);
3086 __del_from_nat_cache(nm_i, natvec[idx]);
3089 f2fs_bug_on(sbi, nm_i->nat_cnt);
3091 /* destroy nat set cache */
3093 while ((found = __gang_lookup_nat_set(nm_i,
3094 nid, SETVEC_SIZE, setvec))) {
3097 nid = setvec[found - 1]->set + 1;
3098 for (idx = 0; idx < found; idx++) {
3099 /* entry_cnt is not zero, when cp_error was occurred */
3100 f2fs_bug_on(sbi, !list_empty(&setvec[idx]->entry_list));
3101 radix_tree_delete(&nm_i->nat_set_root, setvec[idx]->set);
3102 kmem_cache_free(nat_entry_set_slab, setvec[idx]);
3105 up_write(&nm_i->nat_tree_lock);
3107 kvfree(nm_i->nat_block_bitmap);
3108 if (nm_i->free_nid_bitmap) {
3111 for (i = 0; i < nm_i->nat_blocks; i++)
3112 kvfree(nm_i->free_nid_bitmap[i]);
3113 kfree(nm_i->free_nid_bitmap);
3115 kvfree(nm_i->free_nid_count);
3117 kfree(nm_i->nat_bitmap);
3118 kfree(nm_i->nat_bits);
3119 #ifdef CONFIG_F2FS_CHECK_FS
3120 kfree(nm_i->nat_bitmap_mir);
3122 sbi->nm_info = NULL;
3126 int __init f2fs_create_node_manager_caches(void)
3128 nat_entry_slab = f2fs_kmem_cache_create("nat_entry",
3129 sizeof(struct nat_entry));
3130 if (!nat_entry_slab)
3133 free_nid_slab = f2fs_kmem_cache_create("free_nid",
3134 sizeof(struct free_nid));
3136 goto destroy_nat_entry;
3138 nat_entry_set_slab = f2fs_kmem_cache_create("nat_entry_set",
3139 sizeof(struct nat_entry_set));
3140 if (!nat_entry_set_slab)
3141 goto destroy_free_nid;
3143 fsync_node_entry_slab = f2fs_kmem_cache_create("fsync_node_entry",
3144 sizeof(struct fsync_node_entry));
3145 if (!fsync_node_entry_slab)
3146 goto destroy_nat_entry_set;
3149 destroy_nat_entry_set:
3150 kmem_cache_destroy(nat_entry_set_slab);
3152 kmem_cache_destroy(free_nid_slab);
3154 kmem_cache_destroy(nat_entry_slab);
3159 void f2fs_destroy_node_manager_caches(void)
3161 kmem_cache_destroy(fsync_node_entry_slab);
3162 kmem_cache_destroy(nat_entry_set_slab);
3163 kmem_cache_destroy(free_nid_slab);
3164 kmem_cache_destroy(nat_entry_slab);