1 // SPDX-License-Identifier: GPL-2.0
5 * Copyright (c) 2012 Samsung Electronics Co., Ltd.
6 * http://www.samsung.com/
9 #include <linux/f2fs_fs.h>
10 #include <linux/mpage.h>
11 #include <linux/backing-dev.h>
12 #include <linux/blkdev.h>
13 #include <linux/pagevec.h>
14 #include <linux/swap.h>
21 #include <trace/events/f2fs.h>
23 #define on_f2fs_build_free_nids(nmi) mutex_is_locked(&(nm_i)->build_lock)
25 static struct kmem_cache *nat_entry_slab;
26 static struct kmem_cache *free_nid_slab;
27 static struct kmem_cache *nat_entry_set_slab;
28 static struct kmem_cache *fsync_node_entry_slab;
31 * Check whether the given nid is within node id range.
33 int f2fs_check_nid_range(struct f2fs_sb_info *sbi, nid_t nid)
35 if (unlikely(nid < F2FS_ROOT_INO(sbi) || nid >= NM_I(sbi)->max_nid)) {
36 set_sbi_flag(sbi, SBI_NEED_FSCK);
37 f2fs_msg(sbi->sb, KERN_WARNING,
38 "%s: out-of-range nid=%x, run fsck to fix.",
45 bool f2fs_available_free_memory(struct f2fs_sb_info *sbi, int type)
47 struct f2fs_nm_info *nm_i = NM_I(sbi);
49 unsigned long avail_ram;
50 unsigned long mem_size = 0;
55 /* only uses low memory */
56 avail_ram = val.totalram - val.totalhigh;
59 * give 25%, 25%, 50%, 50%, 50% memory for each components respectively
61 if (type == FREE_NIDS) {
62 mem_size = (nm_i->nid_cnt[FREE_NID] *
63 sizeof(struct free_nid)) >> PAGE_SHIFT;
64 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 2);
65 } else if (type == NAT_ENTRIES) {
66 mem_size = (nm_i->nat_cnt * sizeof(struct nat_entry)) >>
68 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 2);
69 if (excess_cached_nats(sbi))
71 } else if (type == DIRTY_DENTS) {
72 if (sbi->sb->s_bdi->wb.dirty_exceeded)
74 mem_size = get_pages(sbi, F2FS_DIRTY_DENTS);
75 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 1);
76 } else if (type == INO_ENTRIES) {
79 for (i = 0; i < MAX_INO_ENTRY; i++)
80 mem_size += sbi->im[i].ino_num *
81 sizeof(struct ino_entry);
82 mem_size >>= PAGE_SHIFT;
83 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 1);
84 } else if (type == EXTENT_CACHE) {
85 mem_size = (atomic_read(&sbi->total_ext_tree) *
86 sizeof(struct extent_tree) +
87 atomic_read(&sbi->total_ext_node) *
88 sizeof(struct extent_node)) >> PAGE_SHIFT;
89 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 1);
90 } else if (type == INMEM_PAGES) {
91 /* it allows 20% / total_ram for inmemory pages */
92 mem_size = get_pages(sbi, F2FS_INMEM_PAGES);
93 res = mem_size < (val.totalram / 5);
95 if (!sbi->sb->s_bdi->wb.dirty_exceeded)
101 static void clear_node_page_dirty(struct page *page)
103 if (PageDirty(page)) {
104 f2fs_clear_page_cache_dirty_tag(page);
105 clear_page_dirty_for_io(page);
106 dec_page_count(F2FS_P_SB(page), F2FS_DIRTY_NODES);
108 ClearPageUptodate(page);
111 static struct page *get_current_nat_page(struct f2fs_sb_info *sbi, nid_t nid)
113 return f2fs_get_meta_page_nofail(sbi, current_nat_addr(sbi, nid));
116 static struct page *get_next_nat_page(struct f2fs_sb_info *sbi, nid_t nid)
118 struct page *src_page;
119 struct page *dst_page;
123 struct f2fs_nm_info *nm_i = NM_I(sbi);
125 dst_off = next_nat_addr(sbi, current_nat_addr(sbi, nid));
127 /* get current nat block page with lock */
128 src_page = get_current_nat_page(sbi, nid);
129 if (IS_ERR(src_page))
131 dst_page = f2fs_grab_meta_page(sbi, dst_off);
132 f2fs_bug_on(sbi, PageDirty(src_page));
134 src_addr = page_address(src_page);
135 dst_addr = page_address(dst_page);
136 memcpy(dst_addr, src_addr, PAGE_SIZE);
137 set_page_dirty(dst_page);
138 f2fs_put_page(src_page, 1);
140 set_to_next_nat(nm_i, nid);
145 static struct nat_entry *__alloc_nat_entry(nid_t nid, bool no_fail)
147 struct nat_entry *new;
150 new = f2fs_kmem_cache_alloc(nat_entry_slab, GFP_F2FS_ZERO);
152 new = kmem_cache_alloc(nat_entry_slab, GFP_F2FS_ZERO);
154 nat_set_nid(new, nid);
160 static void __free_nat_entry(struct nat_entry *e)
162 kmem_cache_free(nat_entry_slab, e);
165 /* must be locked by nat_tree_lock */
166 static struct nat_entry *__init_nat_entry(struct f2fs_nm_info *nm_i,
167 struct nat_entry *ne, struct f2fs_nat_entry *raw_ne, bool no_fail)
170 f2fs_radix_tree_insert(&nm_i->nat_root, nat_get_nid(ne), ne);
171 else if (radix_tree_insert(&nm_i->nat_root, nat_get_nid(ne), ne))
175 node_info_from_raw_nat(&ne->ni, raw_ne);
177 spin_lock(&nm_i->nat_list_lock);
178 list_add_tail(&ne->list, &nm_i->nat_entries);
179 spin_unlock(&nm_i->nat_list_lock);
185 static struct nat_entry *__lookup_nat_cache(struct f2fs_nm_info *nm_i, nid_t n)
187 struct nat_entry *ne;
189 ne = radix_tree_lookup(&nm_i->nat_root, n);
191 /* for recent accessed nat entry, move it to tail of lru list */
192 if (ne && !get_nat_flag(ne, IS_DIRTY)) {
193 spin_lock(&nm_i->nat_list_lock);
194 if (!list_empty(&ne->list))
195 list_move_tail(&ne->list, &nm_i->nat_entries);
196 spin_unlock(&nm_i->nat_list_lock);
202 static unsigned int __gang_lookup_nat_cache(struct f2fs_nm_info *nm_i,
203 nid_t start, unsigned int nr, struct nat_entry **ep)
205 return radix_tree_gang_lookup(&nm_i->nat_root, (void **)ep, start, nr);
208 static void __del_from_nat_cache(struct f2fs_nm_info *nm_i, struct nat_entry *e)
210 radix_tree_delete(&nm_i->nat_root, nat_get_nid(e));
215 static struct nat_entry_set *__grab_nat_entry_set(struct f2fs_nm_info *nm_i,
216 struct nat_entry *ne)
218 nid_t set = NAT_BLOCK_OFFSET(ne->ni.nid);
219 struct nat_entry_set *head;
221 head = radix_tree_lookup(&nm_i->nat_set_root, set);
223 head = f2fs_kmem_cache_alloc(nat_entry_set_slab, GFP_NOFS);
225 INIT_LIST_HEAD(&head->entry_list);
226 INIT_LIST_HEAD(&head->set_list);
229 f2fs_radix_tree_insert(&nm_i->nat_set_root, set, head);
234 static void __set_nat_cache_dirty(struct f2fs_nm_info *nm_i,
235 struct nat_entry *ne)
237 struct nat_entry_set *head;
238 bool new_ne = nat_get_blkaddr(ne) == NEW_ADDR;
241 head = __grab_nat_entry_set(nm_i, ne);
244 * update entry_cnt in below condition:
245 * 1. update NEW_ADDR to valid block address;
246 * 2. update old block address to new one;
248 if (!new_ne && (get_nat_flag(ne, IS_PREALLOC) ||
249 !get_nat_flag(ne, IS_DIRTY)))
252 set_nat_flag(ne, IS_PREALLOC, new_ne);
254 if (get_nat_flag(ne, IS_DIRTY))
257 nm_i->dirty_nat_cnt++;
258 set_nat_flag(ne, IS_DIRTY, true);
260 spin_lock(&nm_i->nat_list_lock);
262 list_del_init(&ne->list);
264 list_move_tail(&ne->list, &head->entry_list);
265 spin_unlock(&nm_i->nat_list_lock);
268 static void __clear_nat_cache_dirty(struct f2fs_nm_info *nm_i,
269 struct nat_entry_set *set, struct nat_entry *ne)
271 spin_lock(&nm_i->nat_list_lock);
272 list_move_tail(&ne->list, &nm_i->nat_entries);
273 spin_unlock(&nm_i->nat_list_lock);
275 set_nat_flag(ne, IS_DIRTY, false);
277 nm_i->dirty_nat_cnt--;
280 static unsigned int __gang_lookup_nat_set(struct f2fs_nm_info *nm_i,
281 nid_t start, unsigned int nr, struct nat_entry_set **ep)
283 return radix_tree_gang_lookup(&nm_i->nat_set_root, (void **)ep,
287 bool f2fs_in_warm_node_list(struct f2fs_sb_info *sbi, struct page *page)
289 return NODE_MAPPING(sbi) == page->mapping &&
290 IS_DNODE(page) && is_cold_node(page);
293 void f2fs_init_fsync_node_info(struct f2fs_sb_info *sbi)
295 spin_lock_init(&sbi->fsync_node_lock);
296 INIT_LIST_HEAD(&sbi->fsync_node_list);
297 sbi->fsync_seg_id = 0;
298 sbi->fsync_node_num = 0;
301 static unsigned int f2fs_add_fsync_node_entry(struct f2fs_sb_info *sbi,
304 struct fsync_node_entry *fn;
308 fn = f2fs_kmem_cache_alloc(fsync_node_entry_slab, GFP_NOFS);
312 INIT_LIST_HEAD(&fn->list);
314 spin_lock_irqsave(&sbi->fsync_node_lock, flags);
315 list_add_tail(&fn->list, &sbi->fsync_node_list);
316 fn->seq_id = sbi->fsync_seg_id++;
318 sbi->fsync_node_num++;
319 spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
324 void f2fs_del_fsync_node_entry(struct f2fs_sb_info *sbi, struct page *page)
326 struct fsync_node_entry *fn;
329 spin_lock_irqsave(&sbi->fsync_node_lock, flags);
330 list_for_each_entry(fn, &sbi->fsync_node_list, list) {
331 if (fn->page == page) {
333 sbi->fsync_node_num--;
334 spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
335 kmem_cache_free(fsync_node_entry_slab, fn);
340 spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
344 void f2fs_reset_fsync_node_info(struct f2fs_sb_info *sbi)
348 spin_lock_irqsave(&sbi->fsync_node_lock, flags);
349 sbi->fsync_seg_id = 0;
350 spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
353 int f2fs_need_dentry_mark(struct f2fs_sb_info *sbi, nid_t nid)
355 struct f2fs_nm_info *nm_i = NM_I(sbi);
359 down_read(&nm_i->nat_tree_lock);
360 e = __lookup_nat_cache(nm_i, nid);
362 if (!get_nat_flag(e, IS_CHECKPOINTED) &&
363 !get_nat_flag(e, HAS_FSYNCED_INODE))
366 up_read(&nm_i->nat_tree_lock);
370 bool f2fs_is_checkpointed_node(struct f2fs_sb_info *sbi, nid_t nid)
372 struct f2fs_nm_info *nm_i = NM_I(sbi);
376 down_read(&nm_i->nat_tree_lock);
377 e = __lookup_nat_cache(nm_i, nid);
378 if (e && !get_nat_flag(e, IS_CHECKPOINTED))
380 up_read(&nm_i->nat_tree_lock);
384 bool f2fs_need_inode_block_update(struct f2fs_sb_info *sbi, nid_t ino)
386 struct f2fs_nm_info *nm_i = NM_I(sbi);
388 bool need_update = true;
390 down_read(&nm_i->nat_tree_lock);
391 e = __lookup_nat_cache(nm_i, ino);
392 if (e && get_nat_flag(e, HAS_LAST_FSYNC) &&
393 (get_nat_flag(e, IS_CHECKPOINTED) ||
394 get_nat_flag(e, HAS_FSYNCED_INODE)))
396 up_read(&nm_i->nat_tree_lock);
400 /* must be locked by nat_tree_lock */
401 static void cache_nat_entry(struct f2fs_sb_info *sbi, nid_t nid,
402 struct f2fs_nat_entry *ne)
404 struct f2fs_nm_info *nm_i = NM_I(sbi);
405 struct nat_entry *new, *e;
407 new = __alloc_nat_entry(nid, false);
411 down_write(&nm_i->nat_tree_lock);
412 e = __lookup_nat_cache(nm_i, nid);
414 e = __init_nat_entry(nm_i, new, ne, false);
416 f2fs_bug_on(sbi, nat_get_ino(e) != le32_to_cpu(ne->ino) ||
417 nat_get_blkaddr(e) !=
418 le32_to_cpu(ne->block_addr) ||
419 nat_get_version(e) != ne->version);
420 up_write(&nm_i->nat_tree_lock);
422 __free_nat_entry(new);
425 static void set_node_addr(struct f2fs_sb_info *sbi, struct node_info *ni,
426 block_t new_blkaddr, bool fsync_done)
428 struct f2fs_nm_info *nm_i = NM_I(sbi);
430 struct nat_entry *new = __alloc_nat_entry(ni->nid, true);
432 down_write(&nm_i->nat_tree_lock);
433 e = __lookup_nat_cache(nm_i, ni->nid);
435 e = __init_nat_entry(nm_i, new, NULL, true);
436 copy_node_info(&e->ni, ni);
437 f2fs_bug_on(sbi, ni->blk_addr == NEW_ADDR);
438 } else if (new_blkaddr == NEW_ADDR) {
440 * when nid is reallocated,
441 * previous nat entry can be remained in nat cache.
442 * So, reinitialize it with new information.
444 copy_node_info(&e->ni, ni);
445 f2fs_bug_on(sbi, ni->blk_addr != NULL_ADDR);
447 /* let's free early to reduce memory consumption */
449 __free_nat_entry(new);
452 f2fs_bug_on(sbi, nat_get_blkaddr(e) != ni->blk_addr);
453 f2fs_bug_on(sbi, nat_get_blkaddr(e) == NULL_ADDR &&
454 new_blkaddr == NULL_ADDR);
455 f2fs_bug_on(sbi, nat_get_blkaddr(e) == NEW_ADDR &&
456 new_blkaddr == NEW_ADDR);
457 f2fs_bug_on(sbi, __is_valid_data_blkaddr(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(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;
534 /* Check nat cache */
535 down_read(&nm_i->nat_tree_lock);
536 e = __lookup_nat_cache(nm_i, nid);
538 ni->ino = nat_get_ino(e);
539 ni->blk_addr = nat_get_blkaddr(e);
540 ni->version = nat_get_version(e);
541 up_read(&nm_i->nat_tree_lock);
545 memset(&ne, 0, sizeof(struct f2fs_nat_entry));
547 /* Check current segment summary */
548 down_read(&curseg->journal_rwsem);
549 i = f2fs_lookup_journal_in_cursum(journal, NAT_JOURNAL, nid, 0);
551 ne = nat_in_journal(journal, i);
552 node_info_from_raw_nat(ni, &ne);
554 up_read(&curseg->journal_rwsem);
556 up_read(&nm_i->nat_tree_lock);
560 /* Fill node_info from nat page */
561 index = current_nat_addr(sbi, nid);
562 up_read(&nm_i->nat_tree_lock);
564 page = f2fs_get_meta_page(sbi, index);
566 return PTR_ERR(page);
568 nat_blk = (struct f2fs_nat_block *)page_address(page);
569 ne = nat_blk->entries[nid - start_nid];
570 node_info_from_raw_nat(ni, &ne);
571 f2fs_put_page(page, 1);
573 blkaddr = le32_to_cpu(ne.block_addr);
574 if (__is_valid_data_blkaddr(blkaddr) &&
575 !f2fs_is_valid_blkaddr(sbi, blkaddr, DATA_GENERIC_ENHANCE))
578 /* cache nat entry */
579 cache_nat_entry(sbi, nid, &ne);
584 * readahead MAX_RA_NODE number of node pages.
586 static void f2fs_ra_node_pages(struct page *parent, int start, int n)
588 struct f2fs_sb_info *sbi = F2FS_P_SB(parent);
589 struct blk_plug plug;
593 blk_start_plug(&plug);
595 /* Then, try readahead for siblings of the desired node */
597 end = min(end, NIDS_PER_BLOCK);
598 for (i = start; i < end; i++) {
599 nid = get_nid(parent, i, false);
600 f2fs_ra_node_page(sbi, nid);
603 blk_finish_plug(&plug);
606 pgoff_t f2fs_get_next_page_offset(struct dnode_of_data *dn, pgoff_t pgofs)
608 const long direct_index = ADDRS_PER_INODE(dn->inode);
609 const long direct_blks = ADDRS_PER_BLOCK(dn->inode);
610 const long indirect_blks = ADDRS_PER_BLOCK(dn->inode) * NIDS_PER_BLOCK;
611 unsigned int skipped_unit = ADDRS_PER_BLOCK(dn->inode);
612 int cur_level = dn->cur_level;
613 int max_level = dn->max_level;
619 while (max_level-- > cur_level)
620 skipped_unit *= NIDS_PER_BLOCK;
622 switch (dn->max_level) {
624 base += 2 * indirect_blks;
627 base += 2 * direct_blks;
630 base += direct_index;
633 f2fs_bug_on(F2FS_I_SB(dn->inode), 1);
636 return ((pgofs - base) / skipped_unit + 1) * skipped_unit + base;
640 * The maximum depth is four.
641 * Offset[0] will have raw inode offset.
643 static int get_node_path(struct inode *inode, long block,
644 int offset[4], unsigned int noffset[4])
646 const long direct_index = ADDRS_PER_INODE(inode);
647 const long direct_blks = ADDRS_PER_BLOCK(inode);
648 const long dptrs_per_blk = NIDS_PER_BLOCK;
649 const long indirect_blks = ADDRS_PER_BLOCK(inode) * NIDS_PER_BLOCK;
650 const long dindirect_blks = indirect_blks * NIDS_PER_BLOCK;
656 if (block < direct_index) {
660 block -= direct_index;
661 if (block < direct_blks) {
662 offset[n++] = NODE_DIR1_BLOCK;
668 block -= direct_blks;
669 if (block < direct_blks) {
670 offset[n++] = NODE_DIR2_BLOCK;
676 block -= direct_blks;
677 if (block < indirect_blks) {
678 offset[n++] = NODE_IND1_BLOCK;
680 offset[n++] = block / direct_blks;
681 noffset[n] = 4 + offset[n - 1];
682 offset[n] = block % direct_blks;
686 block -= indirect_blks;
687 if (block < indirect_blks) {
688 offset[n++] = NODE_IND2_BLOCK;
689 noffset[n] = 4 + dptrs_per_blk;
690 offset[n++] = block / direct_blks;
691 noffset[n] = 5 + dptrs_per_blk + offset[n - 1];
692 offset[n] = block % direct_blks;
696 block -= indirect_blks;
697 if (block < dindirect_blks) {
698 offset[n++] = NODE_DIND_BLOCK;
699 noffset[n] = 5 + (dptrs_per_blk * 2);
700 offset[n++] = block / indirect_blks;
701 noffset[n] = 6 + (dptrs_per_blk * 2) +
702 offset[n - 1] * (dptrs_per_blk + 1);
703 offset[n++] = (block / direct_blks) % dptrs_per_blk;
704 noffset[n] = 7 + (dptrs_per_blk * 2) +
705 offset[n - 2] * (dptrs_per_blk + 1) +
707 offset[n] = block % direct_blks;
718 * Caller should call f2fs_put_dnode(dn).
719 * Also, it should grab and release a rwsem by calling f2fs_lock_op() and
720 * f2fs_unlock_op() only if ro is not set RDONLY_NODE.
721 * In the case of RDONLY_NODE, we don't need to care about mutex.
723 int f2fs_get_dnode_of_data(struct dnode_of_data *dn, pgoff_t index, int mode)
725 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
726 struct page *npage[4];
727 struct page *parent = NULL;
729 unsigned int noffset[4];
734 level = get_node_path(dn->inode, index, offset, noffset);
738 nids[0] = dn->inode->i_ino;
739 npage[0] = dn->inode_page;
742 npage[0] = f2fs_get_node_page(sbi, nids[0]);
743 if (IS_ERR(npage[0]))
744 return PTR_ERR(npage[0]);
747 /* if inline_data is set, should not report any block indices */
748 if (f2fs_has_inline_data(dn->inode) && index) {
750 f2fs_put_page(npage[0], 1);
756 nids[1] = get_nid(parent, offset[0], true);
757 dn->inode_page = npage[0];
758 dn->inode_page_locked = true;
760 /* get indirect or direct nodes */
761 for (i = 1; i <= level; i++) {
764 if (!nids[i] && mode == ALLOC_NODE) {
766 if (!f2fs_alloc_nid(sbi, &(nids[i]))) {
772 npage[i] = f2fs_new_node_page(dn, noffset[i]);
773 if (IS_ERR(npage[i])) {
774 f2fs_alloc_nid_failed(sbi, nids[i]);
775 err = PTR_ERR(npage[i]);
779 set_nid(parent, offset[i - 1], nids[i], i == 1);
780 f2fs_alloc_nid_done(sbi, nids[i]);
782 } else if (mode == LOOKUP_NODE_RA && i == level && level > 1) {
783 npage[i] = f2fs_get_node_page_ra(parent, offset[i - 1]);
784 if (IS_ERR(npage[i])) {
785 err = PTR_ERR(npage[i]);
791 dn->inode_page_locked = false;
794 f2fs_put_page(parent, 1);
798 npage[i] = f2fs_get_node_page(sbi, nids[i]);
799 if (IS_ERR(npage[i])) {
800 err = PTR_ERR(npage[i]);
801 f2fs_put_page(npage[0], 0);
807 nids[i + 1] = get_nid(parent, offset[i], false);
810 dn->nid = nids[level];
811 dn->ofs_in_node = offset[level];
812 dn->node_page = npage[level];
813 dn->data_blkaddr = datablock_addr(dn->inode,
814 dn->node_page, dn->ofs_in_node);
818 f2fs_put_page(parent, 1);
820 f2fs_put_page(npage[0], 0);
822 dn->inode_page = NULL;
823 dn->node_page = NULL;
824 if (err == -ENOENT) {
826 dn->max_level = level;
827 dn->ofs_in_node = offset[level];
832 static int truncate_node(struct dnode_of_data *dn)
834 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
839 err = f2fs_get_node_info(sbi, dn->nid, &ni);
843 /* Deallocate node address */
844 f2fs_invalidate_blocks(sbi, ni.blk_addr);
845 dec_valid_node_count(sbi, dn->inode, dn->nid == dn->inode->i_ino);
846 set_node_addr(sbi, &ni, NULL_ADDR, false);
848 if (dn->nid == dn->inode->i_ino) {
849 f2fs_remove_orphan_inode(sbi, dn->nid);
850 dec_valid_inode_count(sbi);
851 f2fs_inode_synced(dn->inode);
854 clear_node_page_dirty(dn->node_page);
855 set_sbi_flag(sbi, SBI_IS_DIRTY);
857 index = dn->node_page->index;
858 f2fs_put_page(dn->node_page, 1);
860 invalidate_mapping_pages(NODE_MAPPING(sbi),
863 dn->node_page = NULL;
864 trace_f2fs_truncate_node(dn->inode, dn->nid, ni.blk_addr);
869 static int truncate_dnode(struct dnode_of_data *dn)
877 /* get direct node */
878 page = f2fs_get_node_page(F2FS_I_SB(dn->inode), dn->nid);
879 if (IS_ERR(page) && PTR_ERR(page) == -ENOENT)
881 else if (IS_ERR(page))
882 return PTR_ERR(page);
884 /* Make dnode_of_data for parameter */
885 dn->node_page = page;
887 f2fs_truncate_data_blocks(dn);
888 err = truncate_node(dn);
895 static int truncate_nodes(struct dnode_of_data *dn, unsigned int nofs,
898 struct dnode_of_data rdn = *dn;
900 struct f2fs_node *rn;
902 unsigned int child_nofs;
907 return NIDS_PER_BLOCK + 1;
909 trace_f2fs_truncate_nodes_enter(dn->inode, dn->nid, dn->data_blkaddr);
911 page = f2fs_get_node_page(F2FS_I_SB(dn->inode), dn->nid);
913 trace_f2fs_truncate_nodes_exit(dn->inode, PTR_ERR(page));
914 return PTR_ERR(page);
917 f2fs_ra_node_pages(page, ofs, NIDS_PER_BLOCK);
919 rn = F2FS_NODE(page);
921 for (i = ofs; i < NIDS_PER_BLOCK; i++, freed++) {
922 child_nid = le32_to_cpu(rn->in.nid[i]);
926 ret = truncate_dnode(&rdn);
929 if (set_nid(page, i, 0, false))
930 dn->node_changed = true;
933 child_nofs = nofs + ofs * (NIDS_PER_BLOCK + 1) + 1;
934 for (i = ofs; i < NIDS_PER_BLOCK; i++) {
935 child_nid = le32_to_cpu(rn->in.nid[i]);
936 if (child_nid == 0) {
937 child_nofs += NIDS_PER_BLOCK + 1;
941 ret = truncate_nodes(&rdn, child_nofs, 0, depth - 1);
942 if (ret == (NIDS_PER_BLOCK + 1)) {
943 if (set_nid(page, i, 0, false))
944 dn->node_changed = true;
946 } else if (ret < 0 && ret != -ENOENT) {
954 /* remove current indirect node */
955 dn->node_page = page;
956 ret = truncate_node(dn);
961 f2fs_put_page(page, 1);
963 trace_f2fs_truncate_nodes_exit(dn->inode, freed);
967 f2fs_put_page(page, 1);
968 trace_f2fs_truncate_nodes_exit(dn->inode, ret);
972 static int truncate_partial_nodes(struct dnode_of_data *dn,
973 struct f2fs_inode *ri, int *offset, int depth)
975 struct page *pages[2];
982 nid[0] = le32_to_cpu(ri->i_nid[offset[0] - NODE_DIR1_BLOCK]);
986 /* get indirect nodes in the path */
987 for (i = 0; i < idx + 1; i++) {
988 /* reference count'll be increased */
989 pages[i] = f2fs_get_node_page(F2FS_I_SB(dn->inode), nid[i]);
990 if (IS_ERR(pages[i])) {
991 err = PTR_ERR(pages[i]);
995 nid[i + 1] = get_nid(pages[i], offset[i + 1], false);
998 f2fs_ra_node_pages(pages[idx], offset[idx + 1], NIDS_PER_BLOCK);
1000 /* free direct nodes linked to a partial indirect node */
1001 for (i = offset[idx + 1]; i < NIDS_PER_BLOCK; i++) {
1002 child_nid = get_nid(pages[idx], i, false);
1005 dn->nid = child_nid;
1006 err = truncate_dnode(dn);
1009 if (set_nid(pages[idx], i, 0, false))
1010 dn->node_changed = true;
1013 if (offset[idx + 1] == 0) {
1014 dn->node_page = pages[idx];
1016 err = truncate_node(dn);
1020 f2fs_put_page(pages[idx], 1);
1023 offset[idx + 1] = 0;
1026 for (i = idx; i >= 0; i--)
1027 f2fs_put_page(pages[i], 1);
1029 trace_f2fs_truncate_partial_nodes(dn->inode, nid, depth, err);
1035 * All the block addresses of data and nodes should be nullified.
1037 int f2fs_truncate_inode_blocks(struct inode *inode, pgoff_t from)
1039 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1040 int err = 0, cont = 1;
1041 int level, offset[4], noffset[4];
1042 unsigned int nofs = 0;
1043 struct f2fs_inode *ri;
1044 struct dnode_of_data dn;
1047 trace_f2fs_truncate_inode_blocks_enter(inode, from);
1049 level = get_node_path(inode, from, offset, noffset);
1053 page = f2fs_get_node_page(sbi, inode->i_ino);
1055 trace_f2fs_truncate_inode_blocks_exit(inode, PTR_ERR(page));
1056 return PTR_ERR(page);
1059 set_new_dnode(&dn, inode, page, NULL, 0);
1062 ri = F2FS_INODE(page);
1070 if (!offset[level - 1])
1072 err = truncate_partial_nodes(&dn, ri, offset, level);
1073 if (err < 0 && err != -ENOENT)
1075 nofs += 1 + NIDS_PER_BLOCK;
1078 nofs = 5 + 2 * NIDS_PER_BLOCK;
1079 if (!offset[level - 1])
1081 err = truncate_partial_nodes(&dn, ri, offset, level);
1082 if (err < 0 && err != -ENOENT)
1091 dn.nid = le32_to_cpu(ri->i_nid[offset[0] - NODE_DIR1_BLOCK]);
1092 switch (offset[0]) {
1093 case NODE_DIR1_BLOCK:
1094 case NODE_DIR2_BLOCK:
1095 err = truncate_dnode(&dn);
1098 case NODE_IND1_BLOCK:
1099 case NODE_IND2_BLOCK:
1100 err = truncate_nodes(&dn, nofs, offset[1], 2);
1103 case NODE_DIND_BLOCK:
1104 err = truncate_nodes(&dn, nofs, offset[1], 3);
1111 if (err < 0 && err != -ENOENT)
1113 if (offset[1] == 0 &&
1114 ri->i_nid[offset[0] - NODE_DIR1_BLOCK]) {
1116 BUG_ON(page->mapping != NODE_MAPPING(sbi));
1117 f2fs_wait_on_page_writeback(page, NODE, true, true);
1118 ri->i_nid[offset[0] - NODE_DIR1_BLOCK] = 0;
1119 set_page_dirty(page);
1127 f2fs_put_page(page, 0);
1128 trace_f2fs_truncate_inode_blocks_exit(inode, err);
1129 return err > 0 ? 0 : err;
1132 /* caller must lock inode page */
1133 int f2fs_truncate_xattr_node(struct inode *inode)
1135 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1136 nid_t nid = F2FS_I(inode)->i_xattr_nid;
1137 struct dnode_of_data dn;
1144 npage = f2fs_get_node_page(sbi, nid);
1146 return PTR_ERR(npage);
1148 set_new_dnode(&dn, inode, NULL, npage, nid);
1149 err = truncate_node(&dn);
1151 f2fs_put_page(npage, 1);
1155 f2fs_i_xnid_write(inode, 0);
1161 * Caller should grab and release a rwsem by calling f2fs_lock_op() and
1164 int f2fs_remove_inode_page(struct inode *inode)
1166 struct dnode_of_data dn;
1169 set_new_dnode(&dn, inode, NULL, NULL, inode->i_ino);
1170 err = f2fs_get_dnode_of_data(&dn, 0, LOOKUP_NODE);
1174 err = f2fs_truncate_xattr_node(inode);
1176 f2fs_put_dnode(&dn);
1180 /* remove potential inline_data blocks */
1181 if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
1182 S_ISLNK(inode->i_mode))
1183 f2fs_truncate_data_blocks_range(&dn, 1);
1185 /* 0 is possible, after f2fs_new_inode() has failed */
1186 if (unlikely(f2fs_cp_error(F2FS_I_SB(inode)))) {
1187 f2fs_put_dnode(&dn);
1191 if (unlikely(inode->i_blocks != 0 && inode->i_blocks != 8)) {
1192 f2fs_msg(F2FS_I_SB(inode)->sb, KERN_WARNING,
1193 "Inconsistent i_blocks, ino:%lu, iblocks:%llu",
1195 (unsigned long long)inode->i_blocks);
1196 set_sbi_flag(F2FS_I_SB(inode), SBI_NEED_FSCK);
1199 /* will put inode & node pages */
1200 err = truncate_node(&dn);
1202 f2fs_put_dnode(&dn);
1208 struct page *f2fs_new_inode_page(struct inode *inode)
1210 struct dnode_of_data dn;
1212 /* allocate inode page for new inode */
1213 set_new_dnode(&dn, inode, NULL, NULL, inode->i_ino);
1215 /* caller should f2fs_put_page(page, 1); */
1216 return f2fs_new_node_page(&dn, 0);
1219 struct page *f2fs_new_node_page(struct dnode_of_data *dn, unsigned int ofs)
1221 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
1222 struct node_info new_ni;
1226 if (unlikely(is_inode_flag_set(dn->inode, FI_NO_ALLOC)))
1227 return ERR_PTR(-EPERM);
1229 page = f2fs_grab_cache_page(NODE_MAPPING(sbi), dn->nid, false);
1231 return ERR_PTR(-ENOMEM);
1233 if (unlikely((err = inc_valid_node_count(sbi, dn->inode, !ofs))))
1236 #ifdef CONFIG_F2FS_CHECK_FS
1237 err = f2fs_get_node_info(sbi, dn->nid, &new_ni);
1239 dec_valid_node_count(sbi, dn->inode, !ofs);
1242 f2fs_bug_on(sbi, new_ni.blk_addr != NULL_ADDR);
1244 new_ni.nid = dn->nid;
1245 new_ni.ino = dn->inode->i_ino;
1246 new_ni.blk_addr = NULL_ADDR;
1249 set_node_addr(sbi, &new_ni, NEW_ADDR, false);
1251 f2fs_wait_on_page_writeback(page, NODE, true, true);
1252 fill_node_footer(page, dn->nid, dn->inode->i_ino, ofs, true);
1253 set_cold_node(page, S_ISDIR(dn->inode->i_mode));
1254 if (!PageUptodate(page))
1255 SetPageUptodate(page);
1256 if (set_page_dirty(page))
1257 dn->node_changed = true;
1259 if (f2fs_has_xattr_block(ofs))
1260 f2fs_i_xnid_write(dn->inode, dn->nid);
1263 inc_valid_inode_count(sbi);
1267 clear_node_page_dirty(page);
1268 f2fs_put_page(page, 1);
1269 return ERR_PTR(err);
1273 * Caller should do after getting the following values.
1274 * 0: f2fs_put_page(page, 0)
1275 * LOCKED_PAGE or error: f2fs_put_page(page, 1)
1277 static int read_node_page(struct page *page, int op_flags)
1279 struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1280 struct node_info ni;
1281 struct f2fs_io_info fio = {
1285 .op_flags = op_flags,
1287 .encrypted_page = NULL,
1291 if (PageUptodate(page)) {
1292 if (!f2fs_inode_chksum_verify(sbi, page)) {
1293 ClearPageUptodate(page);
1299 err = f2fs_get_node_info(sbi, page->index, &ni);
1303 if (unlikely(ni.blk_addr == NULL_ADDR) ||
1304 is_sbi_flag_set(sbi, SBI_IS_SHUTDOWN)) {
1305 ClearPageUptodate(page);
1309 fio.new_blkaddr = fio.old_blkaddr = ni.blk_addr;
1310 return f2fs_submit_page_bio(&fio);
1314 * Readahead a node page
1316 void f2fs_ra_node_page(struct f2fs_sb_info *sbi, nid_t nid)
1323 if (f2fs_check_nid_range(sbi, nid))
1326 apage = xa_load(&NODE_MAPPING(sbi)->i_pages, nid);
1330 apage = f2fs_grab_cache_page(NODE_MAPPING(sbi), nid, false);
1334 err = read_node_page(apage, REQ_RAHEAD);
1335 f2fs_put_page(apage, err ? 1 : 0);
1338 static struct page *__get_node_page(struct f2fs_sb_info *sbi, pgoff_t nid,
1339 struct page *parent, int start)
1345 return ERR_PTR(-ENOENT);
1346 if (f2fs_check_nid_range(sbi, nid))
1347 return ERR_PTR(-EINVAL);
1349 page = f2fs_grab_cache_page(NODE_MAPPING(sbi), nid, false);
1351 return ERR_PTR(-ENOMEM);
1353 err = read_node_page(page, 0);
1355 f2fs_put_page(page, 1);
1356 return ERR_PTR(err);
1357 } else if (err == LOCKED_PAGE) {
1363 f2fs_ra_node_pages(parent, start + 1, MAX_RA_NODE);
1367 if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1368 f2fs_put_page(page, 1);
1372 if (unlikely(!PageUptodate(page))) {
1377 if (!f2fs_inode_chksum_verify(sbi, page)) {
1382 if(unlikely(nid != nid_of_node(page))) {
1383 f2fs_msg(sbi->sb, KERN_WARNING, "inconsistent node block, "
1384 "nid:%lu, node_footer[nid:%u,ino:%u,ofs:%u,cpver:%llu,blkaddr:%u]",
1385 nid, nid_of_node(page), ino_of_node(page),
1386 ofs_of_node(page), cpver_of_node(page),
1387 next_blkaddr_of_node(page));
1390 ClearPageUptodate(page);
1391 f2fs_put_page(page, 1);
1392 return ERR_PTR(err);
1397 struct page *f2fs_get_node_page(struct f2fs_sb_info *sbi, pgoff_t nid)
1399 return __get_node_page(sbi, nid, NULL, 0);
1402 struct page *f2fs_get_node_page_ra(struct page *parent, int start)
1404 struct f2fs_sb_info *sbi = F2FS_P_SB(parent);
1405 nid_t nid = get_nid(parent, start, false);
1407 return __get_node_page(sbi, nid, parent, start);
1410 static void flush_inline_data(struct f2fs_sb_info *sbi, nid_t ino)
1412 struct inode *inode;
1416 /* should flush inline_data before evict_inode */
1417 inode = ilookup(sbi->sb, ino);
1421 page = f2fs_pagecache_get_page(inode->i_mapping, 0,
1422 FGP_LOCK|FGP_NOWAIT, 0);
1426 if (!PageUptodate(page))
1429 if (!PageDirty(page))
1432 if (!clear_page_dirty_for_io(page))
1435 ret = f2fs_write_inline_data(inode, page);
1436 inode_dec_dirty_pages(inode);
1437 f2fs_remove_dirty_inode(inode);
1439 set_page_dirty(page);
1441 f2fs_put_page(page, 1);
1446 static struct page *last_fsync_dnode(struct f2fs_sb_info *sbi, nid_t ino)
1449 struct pagevec pvec;
1450 struct page *last_page = NULL;
1453 pagevec_init(&pvec);
1456 while ((nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1457 PAGECACHE_TAG_DIRTY))) {
1460 for (i = 0; i < nr_pages; i++) {
1461 struct page *page = pvec.pages[i];
1463 if (unlikely(f2fs_cp_error(sbi))) {
1464 f2fs_put_page(last_page, 0);
1465 pagevec_release(&pvec);
1466 return ERR_PTR(-EIO);
1469 if (!IS_DNODE(page) || !is_cold_node(page))
1471 if (ino_of_node(page) != ino)
1476 if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1481 if (ino_of_node(page) != ino)
1482 goto continue_unlock;
1484 if (!PageDirty(page)) {
1485 /* someone wrote it for us */
1486 goto continue_unlock;
1490 f2fs_put_page(last_page, 0);
1496 pagevec_release(&pvec);
1502 static int __write_node_page(struct page *page, bool atomic, bool *submitted,
1503 struct writeback_control *wbc, bool do_balance,
1504 enum iostat_type io_type, unsigned int *seq_id)
1506 struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1508 struct node_info ni;
1509 struct f2fs_io_info fio = {
1511 .ino = ino_of_node(page),
1514 .op_flags = wbc_to_write_flags(wbc),
1516 .encrypted_page = NULL,
1523 trace_f2fs_writepage(page, NODE);
1525 if (unlikely(f2fs_cp_error(sbi)))
1528 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
1531 if (wbc->sync_mode == WB_SYNC_NONE &&
1532 IS_DNODE(page) && is_cold_node(page))
1535 /* get old block addr of this node page */
1536 nid = nid_of_node(page);
1537 f2fs_bug_on(sbi, page->index != nid);
1539 if (f2fs_get_node_info(sbi, nid, &ni))
1542 if (wbc->for_reclaim) {
1543 if (!down_read_trylock(&sbi->node_write))
1546 down_read(&sbi->node_write);
1549 /* This page is already truncated */
1550 if (unlikely(ni.blk_addr == NULL_ADDR)) {
1551 ClearPageUptodate(page);
1552 dec_page_count(sbi, F2FS_DIRTY_NODES);
1553 up_read(&sbi->node_write);
1558 if (__is_valid_data_blkaddr(ni.blk_addr) &&
1559 !f2fs_is_valid_blkaddr(sbi, ni.blk_addr,
1560 DATA_GENERIC_ENHANCE)) {
1561 up_read(&sbi->node_write);
1565 if (atomic && !test_opt(sbi, NOBARRIER))
1566 fio.op_flags |= REQ_PREFLUSH | REQ_FUA;
1568 set_page_writeback(page);
1569 ClearPageError(page);
1571 if (f2fs_in_warm_node_list(sbi, page)) {
1572 seq = f2fs_add_fsync_node_entry(sbi, page);
1577 fio.old_blkaddr = ni.blk_addr;
1578 f2fs_do_write_node_page(nid, &fio);
1579 set_node_addr(sbi, &ni, fio.new_blkaddr, is_fsync_dnode(page));
1580 dec_page_count(sbi, F2FS_DIRTY_NODES);
1581 up_read(&sbi->node_write);
1583 if (wbc->for_reclaim) {
1584 f2fs_submit_merged_write_cond(sbi, NULL, page, 0, NODE);
1590 if (unlikely(f2fs_cp_error(sbi))) {
1591 f2fs_submit_merged_write(sbi, NODE);
1595 *submitted = fio.submitted;
1598 f2fs_balance_fs(sbi, false);
1602 redirty_page_for_writepage(wbc, page);
1603 return AOP_WRITEPAGE_ACTIVATE;
1606 int f2fs_move_node_page(struct page *node_page, int gc_type)
1610 if (gc_type == FG_GC) {
1611 struct writeback_control wbc = {
1612 .sync_mode = WB_SYNC_ALL,
1617 f2fs_wait_on_page_writeback(node_page, NODE, true, true);
1619 set_page_dirty(node_page);
1621 if (!clear_page_dirty_for_io(node_page)) {
1626 if (__write_node_page(node_page, false, NULL,
1627 &wbc, false, FS_GC_NODE_IO, NULL)) {
1629 unlock_page(node_page);
1633 /* set page dirty and write it */
1634 if (!PageWriteback(node_page))
1635 set_page_dirty(node_page);
1638 unlock_page(node_page);
1640 f2fs_put_page(node_page, 0);
1644 static int f2fs_write_node_page(struct page *page,
1645 struct writeback_control *wbc)
1647 return __write_node_page(page, false, NULL, wbc, false,
1651 int f2fs_fsync_node_pages(struct f2fs_sb_info *sbi, struct inode *inode,
1652 struct writeback_control *wbc, bool atomic,
1653 unsigned int *seq_id)
1656 struct pagevec pvec;
1658 struct page *last_page = NULL;
1659 bool marked = false;
1660 nid_t ino = inode->i_ino;
1665 last_page = last_fsync_dnode(sbi, ino);
1666 if (IS_ERR_OR_NULL(last_page))
1667 return PTR_ERR_OR_ZERO(last_page);
1670 pagevec_init(&pvec);
1673 while ((nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1674 PAGECACHE_TAG_DIRTY))) {
1677 for (i = 0; i < nr_pages; i++) {
1678 struct page *page = pvec.pages[i];
1679 bool submitted = false;
1681 if (unlikely(f2fs_cp_error(sbi))) {
1682 f2fs_put_page(last_page, 0);
1683 pagevec_release(&pvec);
1688 if (!IS_DNODE(page) || !is_cold_node(page))
1690 if (ino_of_node(page) != ino)
1695 if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1700 if (ino_of_node(page) != ino)
1701 goto continue_unlock;
1703 if (!PageDirty(page) && page != last_page) {
1704 /* someone wrote it for us */
1705 goto continue_unlock;
1708 f2fs_wait_on_page_writeback(page, NODE, true, true);
1710 set_fsync_mark(page, 0);
1711 set_dentry_mark(page, 0);
1713 if (!atomic || page == last_page) {
1714 set_fsync_mark(page, 1);
1715 if (IS_INODE(page)) {
1716 if (is_inode_flag_set(inode,
1718 f2fs_update_inode(inode, page);
1719 set_dentry_mark(page,
1720 f2fs_need_dentry_mark(sbi, ino));
1722 /* may be written by other thread */
1723 if (!PageDirty(page))
1724 set_page_dirty(page);
1727 if (!clear_page_dirty_for_io(page))
1728 goto continue_unlock;
1730 ret = __write_node_page(page, atomic &&
1732 &submitted, wbc, true,
1733 FS_NODE_IO, seq_id);
1736 f2fs_put_page(last_page, 0);
1738 } else if (submitted) {
1742 if (page == last_page) {
1743 f2fs_put_page(page, 0);
1748 pagevec_release(&pvec);
1754 if (!ret && atomic && !marked) {
1755 f2fs_msg(sbi->sb, KERN_DEBUG,
1756 "Retry to write fsync mark: ino=%u, idx=%lx",
1757 ino, last_page->index);
1758 lock_page(last_page);
1759 f2fs_wait_on_page_writeback(last_page, NODE, true, true);
1760 set_page_dirty(last_page);
1761 unlock_page(last_page);
1766 f2fs_submit_merged_write_cond(sbi, NULL, NULL, ino, NODE);
1767 return ret ? -EIO: 0;
1770 int f2fs_sync_node_pages(struct f2fs_sb_info *sbi,
1771 struct writeback_control *wbc,
1772 bool do_balance, enum iostat_type io_type)
1775 struct pagevec pvec;
1779 int nr_pages, done = 0;
1781 pagevec_init(&pvec);
1786 while (!done && (nr_pages = pagevec_lookup_tag(&pvec,
1787 NODE_MAPPING(sbi), &index, PAGECACHE_TAG_DIRTY))) {
1790 for (i = 0; i < nr_pages; i++) {
1791 struct page *page = pvec.pages[i];
1792 bool submitted = false;
1794 /* give a priority to WB_SYNC threads */
1795 if (atomic_read(&sbi->wb_sync_req[NODE]) &&
1796 wbc->sync_mode == WB_SYNC_NONE) {
1802 * flushing sequence with step:
1807 if (step == 0 && IS_DNODE(page))
1809 if (step == 1 && (!IS_DNODE(page) ||
1810 is_cold_node(page)))
1812 if (step == 2 && (!IS_DNODE(page) ||
1813 !is_cold_node(page)))
1816 if (wbc->sync_mode == WB_SYNC_ALL)
1818 else if (!trylock_page(page))
1821 if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1827 if (!PageDirty(page)) {
1828 /* someone wrote it for us */
1829 goto continue_unlock;
1832 /* flush inline_data */
1833 if (is_inline_node(page)) {
1834 clear_inline_node(page);
1836 flush_inline_data(sbi, ino_of_node(page));
1840 f2fs_wait_on_page_writeback(page, NODE, true, true);
1842 if (!clear_page_dirty_for_io(page))
1843 goto continue_unlock;
1845 set_fsync_mark(page, 0);
1846 set_dentry_mark(page, 0);
1848 ret = __write_node_page(page, false, &submitted,
1849 wbc, do_balance, io_type, NULL);
1855 if (--wbc->nr_to_write == 0)
1858 pagevec_release(&pvec);
1861 if (wbc->nr_to_write == 0) {
1868 if (wbc->sync_mode == WB_SYNC_NONE && step == 1)
1875 f2fs_submit_merged_write(sbi, NODE);
1877 if (unlikely(f2fs_cp_error(sbi)))
1882 int f2fs_wait_on_node_pages_writeback(struct f2fs_sb_info *sbi,
1883 unsigned int seq_id)
1885 struct fsync_node_entry *fn;
1887 struct list_head *head = &sbi->fsync_node_list;
1888 unsigned long flags;
1889 unsigned int cur_seq_id = 0;
1892 while (seq_id && cur_seq_id < seq_id) {
1893 spin_lock_irqsave(&sbi->fsync_node_lock, flags);
1894 if (list_empty(head)) {
1895 spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
1898 fn = list_first_entry(head, struct fsync_node_entry, list);
1899 if (fn->seq_id > seq_id) {
1900 spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
1903 cur_seq_id = fn->seq_id;
1906 spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
1908 f2fs_wait_on_page_writeback(page, NODE, true, false);
1909 if (TestClearPageError(page))
1918 ret2 = filemap_check_errors(NODE_MAPPING(sbi));
1925 static int f2fs_write_node_pages(struct address_space *mapping,
1926 struct writeback_control *wbc)
1928 struct f2fs_sb_info *sbi = F2FS_M_SB(mapping);
1929 struct blk_plug plug;
1932 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
1935 /* balancing f2fs's metadata in background */
1936 f2fs_balance_fs_bg(sbi);
1938 /* collect a number of dirty node pages and write together */
1939 if (wbc->sync_mode != WB_SYNC_ALL &&
1940 get_pages(sbi, F2FS_DIRTY_NODES) <
1941 nr_pages_to_skip(sbi, NODE))
1944 if (wbc->sync_mode == WB_SYNC_ALL)
1945 atomic_inc(&sbi->wb_sync_req[NODE]);
1946 else if (atomic_read(&sbi->wb_sync_req[NODE]))
1949 trace_f2fs_writepages(mapping->host, wbc, NODE);
1951 diff = nr_pages_to_write(sbi, NODE, wbc);
1952 blk_start_plug(&plug);
1953 f2fs_sync_node_pages(sbi, wbc, true, FS_NODE_IO);
1954 blk_finish_plug(&plug);
1955 wbc->nr_to_write = max((long)0, wbc->nr_to_write - diff);
1957 if (wbc->sync_mode == WB_SYNC_ALL)
1958 atomic_dec(&sbi->wb_sync_req[NODE]);
1962 wbc->pages_skipped += get_pages(sbi, F2FS_DIRTY_NODES);
1963 trace_f2fs_writepages(mapping->host, wbc, NODE);
1967 static int f2fs_set_node_page_dirty(struct page *page)
1969 trace_f2fs_set_page_dirty(page, NODE);
1971 if (!PageUptodate(page))
1972 SetPageUptodate(page);
1973 #ifdef CONFIG_F2FS_CHECK_FS
1975 f2fs_inode_chksum_set(F2FS_P_SB(page), page);
1977 if (!PageDirty(page)) {
1978 __set_page_dirty_nobuffers(page);
1979 inc_page_count(F2FS_P_SB(page), F2FS_DIRTY_NODES);
1980 f2fs_set_page_private(page, 0);
1981 f2fs_trace_pid(page);
1988 * Structure of the f2fs node operations
1990 const struct address_space_operations f2fs_node_aops = {
1991 .writepage = f2fs_write_node_page,
1992 .writepages = f2fs_write_node_pages,
1993 .set_page_dirty = f2fs_set_node_page_dirty,
1994 .invalidatepage = f2fs_invalidate_page,
1995 .releasepage = f2fs_release_page,
1996 #ifdef CONFIG_MIGRATION
1997 .migratepage = f2fs_migrate_page,
2001 static struct free_nid *__lookup_free_nid_list(struct f2fs_nm_info *nm_i,
2004 return radix_tree_lookup(&nm_i->free_nid_root, n);
2007 static int __insert_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 int err = radix_tree_insert(&nm_i->free_nid_root, i->nid, i);
2016 f2fs_bug_on(sbi, state != i->state);
2017 nm_i->nid_cnt[state]++;
2018 if (state == FREE_NID)
2019 list_add_tail(&i->list, &nm_i->free_nid_list);
2023 static void __remove_free_nid(struct f2fs_sb_info *sbi,
2024 struct free_nid *i, enum nid_state state)
2026 struct f2fs_nm_info *nm_i = NM_I(sbi);
2028 f2fs_bug_on(sbi, state != i->state);
2029 nm_i->nid_cnt[state]--;
2030 if (state == FREE_NID)
2032 radix_tree_delete(&nm_i->free_nid_root, i->nid);
2035 static void __move_free_nid(struct f2fs_sb_info *sbi, struct free_nid *i,
2036 enum nid_state org_state, enum nid_state dst_state)
2038 struct f2fs_nm_info *nm_i = NM_I(sbi);
2040 f2fs_bug_on(sbi, org_state != i->state);
2041 i->state = dst_state;
2042 nm_i->nid_cnt[org_state]--;
2043 nm_i->nid_cnt[dst_state]++;
2045 switch (dst_state) {
2050 list_add_tail(&i->list, &nm_i->free_nid_list);
2057 static void update_free_nid_bitmap(struct f2fs_sb_info *sbi, nid_t nid,
2058 bool set, bool build)
2060 struct f2fs_nm_info *nm_i = NM_I(sbi);
2061 unsigned int nat_ofs = NAT_BLOCK_OFFSET(nid);
2062 unsigned int nid_ofs = nid - START_NID(nid);
2064 if (!test_bit_le(nat_ofs, nm_i->nat_block_bitmap))
2068 if (test_bit_le(nid_ofs, nm_i->free_nid_bitmap[nat_ofs]))
2070 __set_bit_le(nid_ofs, nm_i->free_nid_bitmap[nat_ofs]);
2071 nm_i->free_nid_count[nat_ofs]++;
2073 if (!test_bit_le(nid_ofs, nm_i->free_nid_bitmap[nat_ofs]))
2075 __clear_bit_le(nid_ofs, nm_i->free_nid_bitmap[nat_ofs]);
2077 nm_i->free_nid_count[nat_ofs]--;
2081 /* return if the nid is recognized as free */
2082 static bool add_free_nid(struct f2fs_sb_info *sbi,
2083 nid_t nid, bool build, bool update)
2085 struct f2fs_nm_info *nm_i = NM_I(sbi);
2086 struct free_nid *i, *e;
2087 struct nat_entry *ne;
2091 /* 0 nid should not be used */
2092 if (unlikely(nid == 0))
2095 if (unlikely(f2fs_check_nid_range(sbi, nid)))
2098 i = f2fs_kmem_cache_alloc(free_nid_slab, GFP_NOFS);
2100 i->state = FREE_NID;
2102 radix_tree_preload(GFP_NOFS | __GFP_NOFAIL);
2104 spin_lock(&nm_i->nid_list_lock);
2112 * - __insert_nid_to_list(PREALLOC_NID)
2113 * - f2fs_balance_fs_bg
2114 * - f2fs_build_free_nids
2115 * - __f2fs_build_free_nids
2118 * - __lookup_nat_cache
2120 * - f2fs_init_inode_metadata
2121 * - f2fs_new_inode_page
2122 * - f2fs_new_node_page
2124 * - f2fs_alloc_nid_done
2125 * - __remove_nid_from_list(PREALLOC_NID)
2126 * - __insert_nid_to_list(FREE_NID)
2128 ne = __lookup_nat_cache(nm_i, nid);
2129 if (ne && (!get_nat_flag(ne, IS_CHECKPOINTED) ||
2130 nat_get_blkaddr(ne) != NULL_ADDR))
2133 e = __lookup_free_nid_list(nm_i, nid);
2135 if (e->state == FREE_NID)
2141 err = __insert_free_nid(sbi, i, FREE_NID);
2144 update_free_nid_bitmap(sbi, nid, ret, build);
2146 nm_i->available_nids++;
2148 spin_unlock(&nm_i->nid_list_lock);
2149 radix_tree_preload_end();
2152 kmem_cache_free(free_nid_slab, i);
2156 static void remove_free_nid(struct f2fs_sb_info *sbi, nid_t nid)
2158 struct f2fs_nm_info *nm_i = NM_I(sbi);
2160 bool need_free = false;
2162 spin_lock(&nm_i->nid_list_lock);
2163 i = __lookup_free_nid_list(nm_i, nid);
2164 if (i && i->state == FREE_NID) {
2165 __remove_free_nid(sbi, i, FREE_NID);
2168 spin_unlock(&nm_i->nid_list_lock);
2171 kmem_cache_free(free_nid_slab, i);
2174 static int scan_nat_page(struct f2fs_sb_info *sbi,
2175 struct page *nat_page, nid_t start_nid)
2177 struct f2fs_nm_info *nm_i = NM_I(sbi);
2178 struct f2fs_nat_block *nat_blk = page_address(nat_page);
2180 unsigned int nat_ofs = NAT_BLOCK_OFFSET(start_nid);
2183 __set_bit_le(nat_ofs, nm_i->nat_block_bitmap);
2185 i = start_nid % NAT_ENTRY_PER_BLOCK;
2187 for (; i < NAT_ENTRY_PER_BLOCK; i++, start_nid++) {
2188 if (unlikely(start_nid >= nm_i->max_nid))
2191 blk_addr = le32_to_cpu(nat_blk->entries[i].block_addr);
2193 if (blk_addr == NEW_ADDR)
2196 if (blk_addr == NULL_ADDR) {
2197 add_free_nid(sbi, start_nid, true, true);
2199 spin_lock(&NM_I(sbi)->nid_list_lock);
2200 update_free_nid_bitmap(sbi, start_nid, false, true);
2201 spin_unlock(&NM_I(sbi)->nid_list_lock);
2208 static void scan_curseg_cache(struct f2fs_sb_info *sbi)
2210 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
2211 struct f2fs_journal *journal = curseg->journal;
2214 down_read(&curseg->journal_rwsem);
2215 for (i = 0; i < nats_in_cursum(journal); i++) {
2219 addr = le32_to_cpu(nat_in_journal(journal, i).block_addr);
2220 nid = le32_to_cpu(nid_in_journal(journal, i));
2221 if (addr == NULL_ADDR)
2222 add_free_nid(sbi, nid, true, false);
2224 remove_free_nid(sbi, nid);
2226 up_read(&curseg->journal_rwsem);
2229 static void scan_free_nid_bits(struct f2fs_sb_info *sbi)
2231 struct f2fs_nm_info *nm_i = NM_I(sbi);
2232 unsigned int i, idx;
2235 down_read(&nm_i->nat_tree_lock);
2237 for (i = 0; i < nm_i->nat_blocks; i++) {
2238 if (!test_bit_le(i, nm_i->nat_block_bitmap))
2240 if (!nm_i->free_nid_count[i])
2242 for (idx = 0; idx < NAT_ENTRY_PER_BLOCK; idx++) {
2243 idx = find_next_bit_le(nm_i->free_nid_bitmap[i],
2244 NAT_ENTRY_PER_BLOCK, idx);
2245 if (idx >= NAT_ENTRY_PER_BLOCK)
2248 nid = i * NAT_ENTRY_PER_BLOCK + idx;
2249 add_free_nid(sbi, nid, true, false);
2251 if (nm_i->nid_cnt[FREE_NID] >= MAX_FREE_NIDS)
2256 scan_curseg_cache(sbi);
2258 up_read(&nm_i->nat_tree_lock);
2261 static int __f2fs_build_free_nids(struct f2fs_sb_info *sbi,
2262 bool sync, bool mount)
2264 struct f2fs_nm_info *nm_i = NM_I(sbi);
2266 nid_t nid = nm_i->next_scan_nid;
2268 if (unlikely(nid >= nm_i->max_nid))
2271 /* Enough entries */
2272 if (nm_i->nid_cnt[FREE_NID] >= NAT_ENTRY_PER_BLOCK)
2275 if (!sync && !f2fs_available_free_memory(sbi, FREE_NIDS))
2279 /* try to find free nids in free_nid_bitmap */
2280 scan_free_nid_bits(sbi);
2282 if (nm_i->nid_cnt[FREE_NID] >= NAT_ENTRY_PER_BLOCK)
2286 /* readahead nat pages to be scanned */
2287 f2fs_ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nid), FREE_NID_PAGES,
2290 down_read(&nm_i->nat_tree_lock);
2293 if (!test_bit_le(NAT_BLOCK_OFFSET(nid),
2294 nm_i->nat_block_bitmap)) {
2295 struct page *page = get_current_nat_page(sbi, nid);
2298 ret = PTR_ERR(page);
2300 ret = scan_nat_page(sbi, page, nid);
2301 f2fs_put_page(page, 1);
2305 up_read(&nm_i->nat_tree_lock);
2306 f2fs_bug_on(sbi, !mount);
2307 f2fs_msg(sbi->sb, KERN_ERR,
2308 "NAT is corrupt, run fsck to fix it");
2313 nid += (NAT_ENTRY_PER_BLOCK - (nid % NAT_ENTRY_PER_BLOCK));
2314 if (unlikely(nid >= nm_i->max_nid))
2317 if (++i >= FREE_NID_PAGES)
2321 /* go to the next free nat pages to find free nids abundantly */
2322 nm_i->next_scan_nid = nid;
2324 /* find free nids from current sum_pages */
2325 scan_curseg_cache(sbi);
2327 up_read(&nm_i->nat_tree_lock);
2329 f2fs_ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nm_i->next_scan_nid),
2330 nm_i->ra_nid_pages, META_NAT, false);
2335 int f2fs_build_free_nids(struct f2fs_sb_info *sbi, bool sync, bool mount)
2339 mutex_lock(&NM_I(sbi)->build_lock);
2340 ret = __f2fs_build_free_nids(sbi, sync, mount);
2341 mutex_unlock(&NM_I(sbi)->build_lock);
2347 * If this function returns success, caller can obtain a new nid
2348 * from second parameter of this function.
2349 * The returned nid could be used ino as well as nid when inode is created.
2351 bool f2fs_alloc_nid(struct f2fs_sb_info *sbi, nid_t *nid)
2353 struct f2fs_nm_info *nm_i = NM_I(sbi);
2354 struct free_nid *i = NULL;
2356 if (time_to_inject(sbi, FAULT_ALLOC_NID)) {
2357 f2fs_show_injection_info(FAULT_ALLOC_NID);
2361 spin_lock(&nm_i->nid_list_lock);
2363 if (unlikely(nm_i->available_nids == 0)) {
2364 spin_unlock(&nm_i->nid_list_lock);
2368 /* We should not use stale free nids created by f2fs_build_free_nids */
2369 if (nm_i->nid_cnt[FREE_NID] && !on_f2fs_build_free_nids(nm_i)) {
2370 f2fs_bug_on(sbi, list_empty(&nm_i->free_nid_list));
2371 i = list_first_entry(&nm_i->free_nid_list,
2372 struct free_nid, list);
2375 __move_free_nid(sbi, i, FREE_NID, PREALLOC_NID);
2376 nm_i->available_nids--;
2378 update_free_nid_bitmap(sbi, *nid, false, false);
2380 spin_unlock(&nm_i->nid_list_lock);
2383 spin_unlock(&nm_i->nid_list_lock);
2385 /* Let's scan nat pages and its caches to get free nids */
2386 if (!f2fs_build_free_nids(sbi, true, false))
2392 * f2fs_alloc_nid() should be called prior to this function.
2394 void f2fs_alloc_nid_done(struct f2fs_sb_info *sbi, nid_t nid)
2396 struct f2fs_nm_info *nm_i = NM_I(sbi);
2399 spin_lock(&nm_i->nid_list_lock);
2400 i = __lookup_free_nid_list(nm_i, nid);
2401 f2fs_bug_on(sbi, !i);
2402 __remove_free_nid(sbi, i, PREALLOC_NID);
2403 spin_unlock(&nm_i->nid_list_lock);
2405 kmem_cache_free(free_nid_slab, i);
2409 * f2fs_alloc_nid() should be called prior to this function.
2411 void f2fs_alloc_nid_failed(struct f2fs_sb_info *sbi, nid_t nid)
2413 struct f2fs_nm_info *nm_i = NM_I(sbi);
2415 bool need_free = false;
2420 spin_lock(&nm_i->nid_list_lock);
2421 i = __lookup_free_nid_list(nm_i, nid);
2422 f2fs_bug_on(sbi, !i);
2424 if (!f2fs_available_free_memory(sbi, FREE_NIDS)) {
2425 __remove_free_nid(sbi, i, PREALLOC_NID);
2428 __move_free_nid(sbi, i, PREALLOC_NID, FREE_NID);
2431 nm_i->available_nids++;
2433 update_free_nid_bitmap(sbi, nid, true, false);
2435 spin_unlock(&nm_i->nid_list_lock);
2438 kmem_cache_free(free_nid_slab, i);
2441 int f2fs_try_to_free_nids(struct f2fs_sb_info *sbi, int nr_shrink)
2443 struct f2fs_nm_info *nm_i = NM_I(sbi);
2444 struct free_nid *i, *next;
2447 if (nm_i->nid_cnt[FREE_NID] <= MAX_FREE_NIDS)
2450 if (!mutex_trylock(&nm_i->build_lock))
2453 spin_lock(&nm_i->nid_list_lock);
2454 list_for_each_entry_safe(i, next, &nm_i->free_nid_list, list) {
2455 if (nr_shrink <= 0 ||
2456 nm_i->nid_cnt[FREE_NID] <= MAX_FREE_NIDS)
2459 __remove_free_nid(sbi, i, FREE_NID);
2460 kmem_cache_free(free_nid_slab, i);
2463 spin_unlock(&nm_i->nid_list_lock);
2464 mutex_unlock(&nm_i->build_lock);
2466 return nr - nr_shrink;
2469 void f2fs_recover_inline_xattr(struct inode *inode, struct page *page)
2471 void *src_addr, *dst_addr;
2474 struct f2fs_inode *ri;
2476 ipage = f2fs_get_node_page(F2FS_I_SB(inode), inode->i_ino);
2477 f2fs_bug_on(F2FS_I_SB(inode), IS_ERR(ipage));
2479 ri = F2FS_INODE(page);
2480 if (ri->i_inline & F2FS_INLINE_XATTR) {
2481 set_inode_flag(inode, FI_INLINE_XATTR);
2483 clear_inode_flag(inode, FI_INLINE_XATTR);
2487 dst_addr = inline_xattr_addr(inode, ipage);
2488 src_addr = inline_xattr_addr(inode, page);
2489 inline_size = inline_xattr_size(inode);
2491 f2fs_wait_on_page_writeback(ipage, NODE, true, true);
2492 memcpy(dst_addr, src_addr, inline_size);
2494 f2fs_update_inode(inode, ipage);
2495 f2fs_put_page(ipage, 1);
2498 int f2fs_recover_xattr_data(struct inode *inode, struct page *page)
2500 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
2501 nid_t prev_xnid = F2FS_I(inode)->i_xattr_nid;
2503 struct dnode_of_data dn;
2504 struct node_info ni;
2511 /* 1: invalidate the previous xattr nid */
2512 err = f2fs_get_node_info(sbi, prev_xnid, &ni);
2516 f2fs_invalidate_blocks(sbi, ni.blk_addr);
2517 dec_valid_node_count(sbi, inode, false);
2518 set_node_addr(sbi, &ni, NULL_ADDR, false);
2521 /* 2: update xattr nid in inode */
2522 if (!f2fs_alloc_nid(sbi, &new_xnid))
2525 set_new_dnode(&dn, inode, NULL, NULL, new_xnid);
2526 xpage = f2fs_new_node_page(&dn, XATTR_NODE_OFFSET);
2527 if (IS_ERR(xpage)) {
2528 f2fs_alloc_nid_failed(sbi, new_xnid);
2529 return PTR_ERR(xpage);
2532 f2fs_alloc_nid_done(sbi, new_xnid);
2533 f2fs_update_inode_page(inode);
2535 /* 3: update and set xattr node page dirty */
2536 memcpy(F2FS_NODE(xpage), F2FS_NODE(page), VALID_XATTR_BLOCK_SIZE);
2538 set_page_dirty(xpage);
2539 f2fs_put_page(xpage, 1);
2544 int f2fs_recover_inode_page(struct f2fs_sb_info *sbi, struct page *page)
2546 struct f2fs_inode *src, *dst;
2547 nid_t ino = ino_of_node(page);
2548 struct node_info old_ni, new_ni;
2552 err = f2fs_get_node_info(sbi, ino, &old_ni);
2556 if (unlikely(old_ni.blk_addr != NULL_ADDR))
2559 ipage = f2fs_grab_cache_page(NODE_MAPPING(sbi), ino, false);
2561 congestion_wait(BLK_RW_ASYNC, HZ/50);
2565 /* Should not use this inode from free nid list */
2566 remove_free_nid(sbi, ino);
2568 if (!PageUptodate(ipage))
2569 SetPageUptodate(ipage);
2570 fill_node_footer(ipage, ino, ino, 0, true);
2571 set_cold_node(ipage, false);
2573 src = F2FS_INODE(page);
2574 dst = F2FS_INODE(ipage);
2576 memcpy(dst, src, (unsigned long)&src->i_ext - (unsigned long)src);
2578 dst->i_blocks = cpu_to_le64(1);
2579 dst->i_links = cpu_to_le32(1);
2580 dst->i_xattr_nid = 0;
2581 dst->i_inline = src->i_inline & (F2FS_INLINE_XATTR | F2FS_EXTRA_ATTR);
2582 if (dst->i_inline & F2FS_EXTRA_ATTR) {
2583 dst->i_extra_isize = src->i_extra_isize;
2585 if (f2fs_sb_has_flexible_inline_xattr(sbi) &&
2586 F2FS_FITS_IN_INODE(src, le16_to_cpu(src->i_extra_isize),
2587 i_inline_xattr_size))
2588 dst->i_inline_xattr_size = src->i_inline_xattr_size;
2590 if (f2fs_sb_has_project_quota(sbi) &&
2591 F2FS_FITS_IN_INODE(src, le16_to_cpu(src->i_extra_isize),
2593 dst->i_projid = src->i_projid;
2595 if (f2fs_sb_has_inode_crtime(sbi) &&
2596 F2FS_FITS_IN_INODE(src, le16_to_cpu(src->i_extra_isize),
2598 dst->i_crtime = src->i_crtime;
2599 dst->i_crtime_nsec = src->i_crtime_nsec;
2606 if (unlikely(inc_valid_node_count(sbi, NULL, true)))
2608 set_node_addr(sbi, &new_ni, NEW_ADDR, false);
2609 inc_valid_inode_count(sbi);
2610 set_page_dirty(ipage);
2611 f2fs_put_page(ipage, 1);
2615 int f2fs_restore_node_summary(struct f2fs_sb_info *sbi,
2616 unsigned int segno, struct f2fs_summary_block *sum)
2618 struct f2fs_node *rn;
2619 struct f2fs_summary *sum_entry;
2621 int i, idx, last_offset, nrpages;
2623 /* scan the node segment */
2624 last_offset = sbi->blocks_per_seg;
2625 addr = START_BLOCK(sbi, segno);
2626 sum_entry = &sum->entries[0];
2628 for (i = 0; i < last_offset; i += nrpages, addr += nrpages) {
2629 nrpages = min(last_offset - i, BIO_MAX_PAGES);
2631 /* readahead node pages */
2632 f2fs_ra_meta_pages(sbi, addr, nrpages, META_POR, true);
2634 for (idx = addr; idx < addr + nrpages; idx++) {
2635 struct page *page = f2fs_get_tmp_page(sbi, idx);
2638 return PTR_ERR(page);
2640 rn = F2FS_NODE(page);
2641 sum_entry->nid = rn->footer.nid;
2642 sum_entry->version = 0;
2643 sum_entry->ofs_in_node = 0;
2645 f2fs_put_page(page, 1);
2648 invalidate_mapping_pages(META_MAPPING(sbi), addr,
2654 static void remove_nats_in_journal(struct f2fs_sb_info *sbi)
2656 struct f2fs_nm_info *nm_i = NM_I(sbi);
2657 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
2658 struct f2fs_journal *journal = curseg->journal;
2661 down_write(&curseg->journal_rwsem);
2662 for (i = 0; i < nats_in_cursum(journal); i++) {
2663 struct nat_entry *ne;
2664 struct f2fs_nat_entry raw_ne;
2665 nid_t nid = le32_to_cpu(nid_in_journal(journal, i));
2667 raw_ne = nat_in_journal(journal, i);
2669 ne = __lookup_nat_cache(nm_i, nid);
2671 ne = __alloc_nat_entry(nid, true);
2672 __init_nat_entry(nm_i, ne, &raw_ne, true);
2676 * if a free nat in journal has not been used after last
2677 * checkpoint, we should remove it from available nids,
2678 * since later we will add it again.
2680 if (!get_nat_flag(ne, IS_DIRTY) &&
2681 le32_to_cpu(raw_ne.block_addr) == NULL_ADDR) {
2682 spin_lock(&nm_i->nid_list_lock);
2683 nm_i->available_nids--;
2684 spin_unlock(&nm_i->nid_list_lock);
2687 __set_nat_cache_dirty(nm_i, ne);
2689 update_nats_in_cursum(journal, -i);
2690 up_write(&curseg->journal_rwsem);
2693 static void __adjust_nat_entry_set(struct nat_entry_set *nes,
2694 struct list_head *head, int max)
2696 struct nat_entry_set *cur;
2698 if (nes->entry_cnt >= max)
2701 list_for_each_entry(cur, head, set_list) {
2702 if (cur->entry_cnt >= nes->entry_cnt) {
2703 list_add(&nes->set_list, cur->set_list.prev);
2708 list_add_tail(&nes->set_list, head);
2711 static void __update_nat_bits(struct f2fs_sb_info *sbi, nid_t start_nid,
2714 struct f2fs_nm_info *nm_i = NM_I(sbi);
2715 unsigned int nat_index = start_nid / NAT_ENTRY_PER_BLOCK;
2716 struct f2fs_nat_block *nat_blk = page_address(page);
2720 if (!enabled_nat_bits(sbi, NULL))
2723 if (nat_index == 0) {
2727 for (; i < NAT_ENTRY_PER_BLOCK; i++) {
2728 if (nat_blk->entries[i].block_addr != NULL_ADDR)
2732 __set_bit_le(nat_index, nm_i->empty_nat_bits);
2733 __clear_bit_le(nat_index, nm_i->full_nat_bits);
2737 __clear_bit_le(nat_index, nm_i->empty_nat_bits);
2738 if (valid == NAT_ENTRY_PER_BLOCK)
2739 __set_bit_le(nat_index, nm_i->full_nat_bits);
2741 __clear_bit_le(nat_index, nm_i->full_nat_bits);
2744 static int __flush_nat_entry_set(struct f2fs_sb_info *sbi,
2745 struct nat_entry_set *set, struct cp_control *cpc)
2747 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
2748 struct f2fs_journal *journal = curseg->journal;
2749 nid_t start_nid = set->set * NAT_ENTRY_PER_BLOCK;
2750 bool to_journal = true;
2751 struct f2fs_nat_block *nat_blk;
2752 struct nat_entry *ne, *cur;
2753 struct page *page = NULL;
2756 * there are two steps to flush nat entries:
2757 * #1, flush nat entries to journal in current hot data summary block.
2758 * #2, flush nat entries to nat page.
2760 if (enabled_nat_bits(sbi, cpc) ||
2761 !__has_cursum_space(journal, set->entry_cnt, NAT_JOURNAL))
2765 down_write(&curseg->journal_rwsem);
2767 page = get_next_nat_page(sbi, start_nid);
2769 return PTR_ERR(page);
2771 nat_blk = page_address(page);
2772 f2fs_bug_on(sbi, !nat_blk);
2775 /* flush dirty nats in nat entry set */
2776 list_for_each_entry_safe(ne, cur, &set->entry_list, list) {
2777 struct f2fs_nat_entry *raw_ne;
2778 nid_t nid = nat_get_nid(ne);
2781 f2fs_bug_on(sbi, nat_get_blkaddr(ne) == NEW_ADDR);
2784 offset = f2fs_lookup_journal_in_cursum(journal,
2785 NAT_JOURNAL, nid, 1);
2786 f2fs_bug_on(sbi, offset < 0);
2787 raw_ne = &nat_in_journal(journal, offset);
2788 nid_in_journal(journal, offset) = cpu_to_le32(nid);
2790 raw_ne = &nat_blk->entries[nid - start_nid];
2792 raw_nat_from_node_info(raw_ne, &ne->ni);
2794 __clear_nat_cache_dirty(NM_I(sbi), set, ne);
2795 if (nat_get_blkaddr(ne) == NULL_ADDR) {
2796 add_free_nid(sbi, nid, false, true);
2798 spin_lock(&NM_I(sbi)->nid_list_lock);
2799 update_free_nid_bitmap(sbi, nid, false, false);
2800 spin_unlock(&NM_I(sbi)->nid_list_lock);
2805 up_write(&curseg->journal_rwsem);
2807 __update_nat_bits(sbi, start_nid, page);
2808 f2fs_put_page(page, 1);
2811 /* Allow dirty nats by node block allocation in write_begin */
2812 if (!set->entry_cnt) {
2813 radix_tree_delete(&NM_I(sbi)->nat_set_root, set->set);
2814 kmem_cache_free(nat_entry_set_slab, set);
2820 * This function is called during the checkpointing process.
2822 int f2fs_flush_nat_entries(struct f2fs_sb_info *sbi, struct cp_control *cpc)
2824 struct f2fs_nm_info *nm_i = NM_I(sbi);
2825 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
2826 struct f2fs_journal *journal = curseg->journal;
2827 struct nat_entry_set *setvec[SETVEC_SIZE];
2828 struct nat_entry_set *set, *tmp;
2834 /* during unmount, let's flush nat_bits before checking dirty_nat_cnt */
2835 if (enabled_nat_bits(sbi, cpc)) {
2836 down_write(&nm_i->nat_tree_lock);
2837 remove_nats_in_journal(sbi);
2838 up_write(&nm_i->nat_tree_lock);
2841 if (!nm_i->dirty_nat_cnt)
2844 down_write(&nm_i->nat_tree_lock);
2847 * if there are no enough space in journal to store dirty nat
2848 * entries, remove all entries from journal and merge them
2849 * into nat entry set.
2851 if (enabled_nat_bits(sbi, cpc) ||
2852 !__has_cursum_space(journal, nm_i->dirty_nat_cnt, NAT_JOURNAL))
2853 remove_nats_in_journal(sbi);
2855 while ((found = __gang_lookup_nat_set(nm_i,
2856 set_idx, SETVEC_SIZE, setvec))) {
2858 set_idx = setvec[found - 1]->set + 1;
2859 for (idx = 0; idx < found; idx++)
2860 __adjust_nat_entry_set(setvec[idx], &sets,
2861 MAX_NAT_JENTRIES(journal));
2864 /* flush dirty nats in nat entry set */
2865 list_for_each_entry_safe(set, tmp, &sets, set_list) {
2866 err = __flush_nat_entry_set(sbi, set, cpc);
2871 up_write(&nm_i->nat_tree_lock);
2872 /* Allow dirty nats by node block allocation in write_begin */
2877 static int __get_nat_bitmaps(struct f2fs_sb_info *sbi)
2879 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
2880 struct f2fs_nm_info *nm_i = NM_I(sbi);
2881 unsigned int nat_bits_bytes = nm_i->nat_blocks / BITS_PER_BYTE;
2883 __u64 cp_ver = cur_cp_version(ckpt);
2884 block_t nat_bits_addr;
2886 if (!enabled_nat_bits(sbi, NULL))
2889 nm_i->nat_bits_blocks = F2FS_BLK_ALIGN((nat_bits_bytes << 1) + 8);
2890 nm_i->nat_bits = f2fs_kzalloc(sbi,
2891 nm_i->nat_bits_blocks << F2FS_BLKSIZE_BITS, GFP_KERNEL);
2892 if (!nm_i->nat_bits)
2895 nat_bits_addr = __start_cp_addr(sbi) + sbi->blocks_per_seg -
2896 nm_i->nat_bits_blocks;
2897 for (i = 0; i < nm_i->nat_bits_blocks; i++) {
2900 page = f2fs_get_meta_page(sbi, nat_bits_addr++);
2902 return PTR_ERR(page);
2904 memcpy(nm_i->nat_bits + (i << F2FS_BLKSIZE_BITS),
2905 page_address(page), F2FS_BLKSIZE);
2906 f2fs_put_page(page, 1);
2909 cp_ver |= (cur_cp_crc(ckpt) << 32);
2910 if (cpu_to_le64(cp_ver) != *(__le64 *)nm_i->nat_bits) {
2911 disable_nat_bits(sbi, true);
2915 nm_i->full_nat_bits = nm_i->nat_bits + 8;
2916 nm_i->empty_nat_bits = nm_i->full_nat_bits + nat_bits_bytes;
2918 f2fs_msg(sbi->sb, KERN_NOTICE, "Found nat_bits in checkpoint");
2922 static inline void load_free_nid_bitmap(struct f2fs_sb_info *sbi)
2924 struct f2fs_nm_info *nm_i = NM_I(sbi);
2926 nid_t nid, last_nid;
2928 if (!enabled_nat_bits(sbi, NULL))
2931 for (i = 0; i < nm_i->nat_blocks; i++) {
2932 i = find_next_bit_le(nm_i->empty_nat_bits, nm_i->nat_blocks, i);
2933 if (i >= nm_i->nat_blocks)
2936 __set_bit_le(i, nm_i->nat_block_bitmap);
2938 nid = i * NAT_ENTRY_PER_BLOCK;
2939 last_nid = nid + NAT_ENTRY_PER_BLOCK;
2941 spin_lock(&NM_I(sbi)->nid_list_lock);
2942 for (; nid < last_nid; nid++)
2943 update_free_nid_bitmap(sbi, nid, true, true);
2944 spin_unlock(&NM_I(sbi)->nid_list_lock);
2947 for (i = 0; i < nm_i->nat_blocks; i++) {
2948 i = find_next_bit_le(nm_i->full_nat_bits, nm_i->nat_blocks, i);
2949 if (i >= nm_i->nat_blocks)
2952 __set_bit_le(i, nm_i->nat_block_bitmap);
2956 static int init_node_manager(struct f2fs_sb_info *sbi)
2958 struct f2fs_super_block *sb_raw = F2FS_RAW_SUPER(sbi);
2959 struct f2fs_nm_info *nm_i = NM_I(sbi);
2960 unsigned char *version_bitmap;
2961 unsigned int nat_segs;
2964 nm_i->nat_blkaddr = le32_to_cpu(sb_raw->nat_blkaddr);
2966 /* segment_count_nat includes pair segment so divide to 2. */
2967 nat_segs = le32_to_cpu(sb_raw->segment_count_nat) >> 1;
2968 nm_i->nat_blocks = nat_segs << le32_to_cpu(sb_raw->log_blocks_per_seg);
2969 nm_i->max_nid = NAT_ENTRY_PER_BLOCK * nm_i->nat_blocks;
2971 /* not used nids: 0, node, meta, (and root counted as valid node) */
2972 nm_i->available_nids = nm_i->max_nid - sbi->total_valid_node_count -
2973 sbi->nquota_files - F2FS_RESERVED_NODE_NUM;
2974 nm_i->nid_cnt[FREE_NID] = 0;
2975 nm_i->nid_cnt[PREALLOC_NID] = 0;
2977 nm_i->ram_thresh = DEF_RAM_THRESHOLD;
2978 nm_i->ra_nid_pages = DEF_RA_NID_PAGES;
2979 nm_i->dirty_nats_ratio = DEF_DIRTY_NAT_RATIO_THRESHOLD;
2981 INIT_RADIX_TREE(&nm_i->free_nid_root, GFP_ATOMIC);
2982 INIT_LIST_HEAD(&nm_i->free_nid_list);
2983 INIT_RADIX_TREE(&nm_i->nat_root, GFP_NOIO);
2984 INIT_RADIX_TREE(&nm_i->nat_set_root, GFP_NOIO);
2985 INIT_LIST_HEAD(&nm_i->nat_entries);
2986 spin_lock_init(&nm_i->nat_list_lock);
2988 mutex_init(&nm_i->build_lock);
2989 spin_lock_init(&nm_i->nid_list_lock);
2990 init_rwsem(&nm_i->nat_tree_lock);
2992 nm_i->next_scan_nid = le32_to_cpu(sbi->ckpt->next_free_nid);
2993 nm_i->bitmap_size = __bitmap_size(sbi, NAT_BITMAP);
2994 version_bitmap = __bitmap_ptr(sbi, NAT_BITMAP);
2995 if (!version_bitmap)
2998 nm_i->nat_bitmap = kmemdup(version_bitmap, nm_i->bitmap_size,
3000 if (!nm_i->nat_bitmap)
3003 err = __get_nat_bitmaps(sbi);
3007 #ifdef CONFIG_F2FS_CHECK_FS
3008 nm_i->nat_bitmap_mir = kmemdup(version_bitmap, nm_i->bitmap_size,
3010 if (!nm_i->nat_bitmap_mir)
3017 static int init_free_nid_cache(struct f2fs_sb_info *sbi)
3019 struct f2fs_nm_info *nm_i = NM_I(sbi);
3022 nm_i->free_nid_bitmap =
3023 f2fs_kzalloc(sbi, array_size(sizeof(unsigned char *),
3026 if (!nm_i->free_nid_bitmap)
3029 for (i = 0; i < nm_i->nat_blocks; i++) {
3030 nm_i->free_nid_bitmap[i] = f2fs_kvzalloc(sbi,
3031 f2fs_bitmap_size(NAT_ENTRY_PER_BLOCK), GFP_KERNEL);
3032 if (!nm_i->free_nid_bitmap[i])
3036 nm_i->nat_block_bitmap = f2fs_kvzalloc(sbi, nm_i->nat_blocks / 8,
3038 if (!nm_i->nat_block_bitmap)
3041 nm_i->free_nid_count =
3042 f2fs_kvzalloc(sbi, array_size(sizeof(unsigned short),
3045 if (!nm_i->free_nid_count)
3050 int f2fs_build_node_manager(struct f2fs_sb_info *sbi)
3054 sbi->nm_info = f2fs_kzalloc(sbi, sizeof(struct f2fs_nm_info),
3059 err = init_node_manager(sbi);
3063 err = init_free_nid_cache(sbi);
3067 /* load free nid status from nat_bits table */
3068 load_free_nid_bitmap(sbi);
3070 return f2fs_build_free_nids(sbi, true, true);
3073 void f2fs_destroy_node_manager(struct f2fs_sb_info *sbi)
3075 struct f2fs_nm_info *nm_i = NM_I(sbi);
3076 struct free_nid *i, *next_i;
3077 struct nat_entry *natvec[NATVEC_SIZE];
3078 struct nat_entry_set *setvec[SETVEC_SIZE];
3085 /* destroy free nid list */
3086 spin_lock(&nm_i->nid_list_lock);
3087 list_for_each_entry_safe(i, next_i, &nm_i->free_nid_list, list) {
3088 __remove_free_nid(sbi, i, FREE_NID);
3089 spin_unlock(&nm_i->nid_list_lock);
3090 kmem_cache_free(free_nid_slab, i);
3091 spin_lock(&nm_i->nid_list_lock);
3093 f2fs_bug_on(sbi, nm_i->nid_cnt[FREE_NID]);
3094 f2fs_bug_on(sbi, nm_i->nid_cnt[PREALLOC_NID]);
3095 f2fs_bug_on(sbi, !list_empty(&nm_i->free_nid_list));
3096 spin_unlock(&nm_i->nid_list_lock);
3098 /* destroy nat cache */
3099 down_write(&nm_i->nat_tree_lock);
3100 while ((found = __gang_lookup_nat_cache(nm_i,
3101 nid, NATVEC_SIZE, natvec))) {
3104 nid = nat_get_nid(natvec[found - 1]) + 1;
3105 for (idx = 0; idx < found; idx++) {
3106 spin_lock(&nm_i->nat_list_lock);
3107 list_del(&natvec[idx]->list);
3108 spin_unlock(&nm_i->nat_list_lock);
3110 __del_from_nat_cache(nm_i, natvec[idx]);
3113 f2fs_bug_on(sbi, nm_i->nat_cnt);
3115 /* destroy nat set cache */
3117 while ((found = __gang_lookup_nat_set(nm_i,
3118 nid, SETVEC_SIZE, setvec))) {
3121 nid = setvec[found - 1]->set + 1;
3122 for (idx = 0; idx < found; idx++) {
3123 /* entry_cnt is not zero, when cp_error was occurred */
3124 f2fs_bug_on(sbi, !list_empty(&setvec[idx]->entry_list));
3125 radix_tree_delete(&nm_i->nat_set_root, setvec[idx]->set);
3126 kmem_cache_free(nat_entry_set_slab, setvec[idx]);
3129 up_write(&nm_i->nat_tree_lock);
3131 kvfree(nm_i->nat_block_bitmap);
3132 if (nm_i->free_nid_bitmap) {
3135 for (i = 0; i < nm_i->nat_blocks; i++)
3136 kvfree(nm_i->free_nid_bitmap[i]);
3137 kvfree(nm_i->free_nid_bitmap);
3139 kvfree(nm_i->free_nid_count);
3141 kvfree(nm_i->nat_bitmap);
3142 kvfree(nm_i->nat_bits);
3143 #ifdef CONFIG_F2FS_CHECK_FS
3144 kvfree(nm_i->nat_bitmap_mir);
3146 sbi->nm_info = NULL;
3150 int __init f2fs_create_node_manager_caches(void)
3152 nat_entry_slab = f2fs_kmem_cache_create("nat_entry",
3153 sizeof(struct nat_entry));
3154 if (!nat_entry_slab)
3157 free_nid_slab = f2fs_kmem_cache_create("free_nid",
3158 sizeof(struct free_nid));
3160 goto destroy_nat_entry;
3162 nat_entry_set_slab = f2fs_kmem_cache_create("nat_entry_set",
3163 sizeof(struct nat_entry_set));
3164 if (!nat_entry_set_slab)
3165 goto destroy_free_nid;
3167 fsync_node_entry_slab = f2fs_kmem_cache_create("fsync_node_entry",
3168 sizeof(struct fsync_node_entry));
3169 if (!fsync_node_entry_slab)
3170 goto destroy_nat_entry_set;
3173 destroy_nat_entry_set:
3174 kmem_cache_destroy(nat_entry_set_slab);
3176 kmem_cache_destroy(free_nid_slab);
3178 kmem_cache_destroy(nat_entry_slab);
3183 void f2fs_destroy_node_manager_caches(void)
3185 kmem_cache_destroy(fsync_node_entry_slab);
3186 kmem_cache_destroy(nat_entry_set_slab);
3187 kmem_cache_destroy(free_nid_slab);
3188 kmem_cache_destroy(nat_entry_slab);