4 * Copyright (c) 2012 Samsung Electronics Co., Ltd.
5 * http://www.samsung.com/
7 * This program is free software; you can redistribute it and/or modify
8 * it under the terms of the GNU General Public License version 2 as
9 * published by the Free Software Foundation.
12 #include <linux/f2fs_fs.h>
13 #include <linux/mpage.h>
14 #include <linux/backing-dev.h>
15 #include <linux/blkdev.h>
16 #include <linux/pagevec.h>
17 #include <linux/swap.h>
24 #include <trace/events/f2fs.h>
26 #define on_f2fs_build_free_nids(nmi) mutex_is_locked(&(nm_i)->build_lock)
28 static struct kmem_cache *nat_entry_slab;
29 static struct kmem_cache *free_nid_slab;
30 static struct kmem_cache *nat_entry_set_slab;
31 static struct kmem_cache *fsync_node_entry_slab;
34 * Check whether the given nid is within node id range.
36 int f2fs_check_nid_range(struct f2fs_sb_info *sbi, nid_t nid)
38 if (unlikely(nid < F2FS_ROOT_INO(sbi) || nid >= NM_I(sbi)->max_nid)) {
39 set_sbi_flag(sbi, SBI_NEED_FSCK);
40 f2fs_msg(sbi->sb, KERN_WARNING,
41 "%s: out-of-range nid=%x, run fsck to fix.",
48 bool f2fs_available_free_memory(struct f2fs_sb_info *sbi, int type)
50 struct f2fs_nm_info *nm_i = NM_I(sbi);
52 unsigned long avail_ram;
53 unsigned long mem_size = 0;
58 /* only uses low memory */
59 avail_ram = val.totalram - val.totalhigh;
62 * give 25%, 25%, 50%, 50%, 50% memory for each components respectively
64 if (type == FREE_NIDS) {
65 mem_size = (nm_i->nid_cnt[FREE_NID] *
66 sizeof(struct free_nid)) >> PAGE_SHIFT;
67 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 2);
68 } else if (type == NAT_ENTRIES) {
69 mem_size = (nm_i->nat_cnt * sizeof(struct nat_entry)) >>
71 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 2);
72 if (excess_cached_nats(sbi))
74 } else if (type == DIRTY_DENTS) {
75 if (sbi->sb->s_bdi->wb.dirty_exceeded)
77 mem_size = get_pages(sbi, F2FS_DIRTY_DENTS);
78 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 1);
79 } else if (type == INO_ENTRIES) {
82 for (i = 0; i < MAX_INO_ENTRY; i++)
83 mem_size += sbi->im[i].ino_num *
84 sizeof(struct ino_entry);
85 mem_size >>= PAGE_SHIFT;
86 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 1);
87 } else if (type == EXTENT_CACHE) {
88 mem_size = (atomic_read(&sbi->total_ext_tree) *
89 sizeof(struct extent_tree) +
90 atomic_read(&sbi->total_ext_node) *
91 sizeof(struct extent_node)) >> PAGE_SHIFT;
92 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 1);
93 } else if (type == INMEM_PAGES) {
94 /* it allows 20% / total_ram for inmemory pages */
95 mem_size = get_pages(sbi, F2FS_INMEM_PAGES);
96 res = mem_size < (val.totalram / 5);
98 if (!sbi->sb->s_bdi->wb.dirty_exceeded)
104 static void clear_node_page_dirty(struct page *page)
106 if (PageDirty(page)) {
107 f2fs_clear_radix_tree_dirty_tag(page);
108 clear_page_dirty_for_io(page);
109 dec_page_count(F2FS_P_SB(page), F2FS_DIRTY_NODES);
111 ClearPageUptodate(page);
114 static struct page *get_current_nat_page(struct f2fs_sb_info *sbi, nid_t nid)
116 return f2fs_get_meta_page_nofail(sbi, current_nat_addr(sbi, nid));
119 static struct page *get_next_nat_page(struct f2fs_sb_info *sbi, nid_t nid)
121 struct page *src_page;
122 struct page *dst_page;
126 struct f2fs_nm_info *nm_i = NM_I(sbi);
128 dst_off = next_nat_addr(sbi, current_nat_addr(sbi, nid));
130 /* get current nat block page with lock */
131 src_page = get_current_nat_page(sbi, nid);
132 dst_page = f2fs_grab_meta_page(sbi, dst_off);
133 f2fs_bug_on(sbi, PageDirty(src_page));
135 src_addr = page_address(src_page);
136 dst_addr = page_address(dst_page);
137 memcpy(dst_addr, src_addr, PAGE_SIZE);
138 set_page_dirty(dst_page);
139 f2fs_put_page(src_page, 1);
141 set_to_next_nat(nm_i, nid);
146 static struct nat_entry *__alloc_nat_entry(nid_t nid, bool no_fail)
148 struct nat_entry *new;
151 new = f2fs_kmem_cache_alloc(nat_entry_slab, GFP_F2FS_ZERO);
153 new = kmem_cache_alloc(nat_entry_slab, GFP_F2FS_ZERO);
155 nat_set_nid(new, nid);
161 static void __free_nat_entry(struct nat_entry *e)
163 kmem_cache_free(nat_entry_slab, e);
166 /* must be locked by nat_tree_lock */
167 static struct nat_entry *__init_nat_entry(struct f2fs_nm_info *nm_i,
168 struct nat_entry *ne, struct f2fs_nat_entry *raw_ne, bool no_fail)
171 f2fs_radix_tree_insert(&nm_i->nat_root, nat_get_nid(ne), ne);
172 else if (radix_tree_insert(&nm_i->nat_root, nat_get_nid(ne), ne))
176 node_info_from_raw_nat(&ne->ni, raw_ne);
178 spin_lock(&nm_i->nat_list_lock);
179 list_add_tail(&ne->list, &nm_i->nat_entries);
180 spin_unlock(&nm_i->nat_list_lock);
186 static struct nat_entry *__lookup_nat_cache(struct f2fs_nm_info *nm_i, nid_t n)
188 struct nat_entry *ne;
190 ne = radix_tree_lookup(&nm_i->nat_root, n);
192 /* for recent accessed nat entry, move it to tail of lru list */
193 if (ne && !get_nat_flag(ne, IS_DIRTY)) {
194 spin_lock(&nm_i->nat_list_lock);
195 if (!list_empty(&ne->list))
196 list_move_tail(&ne->list, &nm_i->nat_entries);
197 spin_unlock(&nm_i->nat_list_lock);
203 static unsigned int __gang_lookup_nat_cache(struct f2fs_nm_info *nm_i,
204 nid_t start, unsigned int nr, struct nat_entry **ep)
206 return radix_tree_gang_lookup(&nm_i->nat_root, (void **)ep, start, nr);
209 static void __del_from_nat_cache(struct f2fs_nm_info *nm_i, struct nat_entry *e)
211 radix_tree_delete(&nm_i->nat_root, nat_get_nid(e));
216 static struct nat_entry_set *__grab_nat_entry_set(struct f2fs_nm_info *nm_i,
217 struct nat_entry *ne)
219 nid_t set = NAT_BLOCK_OFFSET(ne->ni.nid);
220 struct nat_entry_set *head;
222 head = radix_tree_lookup(&nm_i->nat_set_root, set);
224 head = f2fs_kmem_cache_alloc(nat_entry_set_slab, GFP_NOFS);
226 INIT_LIST_HEAD(&head->entry_list);
227 INIT_LIST_HEAD(&head->set_list);
230 f2fs_radix_tree_insert(&nm_i->nat_set_root, set, head);
235 static void __set_nat_cache_dirty(struct f2fs_nm_info *nm_i,
236 struct nat_entry *ne)
238 struct nat_entry_set *head;
239 bool new_ne = nat_get_blkaddr(ne) == NEW_ADDR;
242 head = __grab_nat_entry_set(nm_i, ne);
245 * update entry_cnt in below condition:
246 * 1. update NEW_ADDR to valid block address;
247 * 2. update old block address to new one;
249 if (!new_ne && (get_nat_flag(ne, IS_PREALLOC) ||
250 !get_nat_flag(ne, IS_DIRTY)))
253 set_nat_flag(ne, IS_PREALLOC, new_ne);
255 if (get_nat_flag(ne, IS_DIRTY))
258 nm_i->dirty_nat_cnt++;
259 set_nat_flag(ne, IS_DIRTY, true);
261 spin_lock(&nm_i->nat_list_lock);
263 list_del_init(&ne->list);
265 list_move_tail(&ne->list, &head->entry_list);
266 spin_unlock(&nm_i->nat_list_lock);
269 static void __clear_nat_cache_dirty(struct f2fs_nm_info *nm_i,
270 struct nat_entry_set *set, struct nat_entry *ne)
272 spin_lock(&nm_i->nat_list_lock);
273 list_move_tail(&ne->list, &nm_i->nat_entries);
274 spin_unlock(&nm_i->nat_list_lock);
276 set_nat_flag(ne, IS_DIRTY, false);
278 nm_i->dirty_nat_cnt--;
281 static unsigned int __gang_lookup_nat_set(struct f2fs_nm_info *nm_i,
282 nid_t start, unsigned int nr, struct nat_entry_set **ep)
284 return radix_tree_gang_lookup(&nm_i->nat_set_root, (void **)ep,
288 bool f2fs_in_warm_node_list(struct f2fs_sb_info *sbi, struct page *page)
290 return NODE_MAPPING(sbi) == page->mapping &&
291 IS_DNODE(page) && is_cold_node(page);
294 void f2fs_init_fsync_node_info(struct f2fs_sb_info *sbi)
296 spin_lock_init(&sbi->fsync_node_lock);
297 INIT_LIST_HEAD(&sbi->fsync_node_list);
298 sbi->fsync_seg_id = 0;
299 sbi->fsync_node_num = 0;
302 static unsigned int f2fs_add_fsync_node_entry(struct f2fs_sb_info *sbi,
305 struct fsync_node_entry *fn;
309 fn = f2fs_kmem_cache_alloc(fsync_node_entry_slab, GFP_NOFS);
313 INIT_LIST_HEAD(&fn->list);
315 spin_lock_irqsave(&sbi->fsync_node_lock, flags);
316 list_add_tail(&fn->list, &sbi->fsync_node_list);
317 fn->seq_id = sbi->fsync_seg_id++;
319 sbi->fsync_node_num++;
320 spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
325 void f2fs_del_fsync_node_entry(struct f2fs_sb_info *sbi, struct page *page)
327 struct fsync_node_entry *fn;
330 spin_lock_irqsave(&sbi->fsync_node_lock, flags);
331 list_for_each_entry(fn, &sbi->fsync_node_list, list) {
332 if (fn->page == page) {
334 sbi->fsync_node_num--;
335 spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
336 kmem_cache_free(fsync_node_entry_slab, fn);
341 spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
345 void f2fs_reset_fsync_node_info(struct f2fs_sb_info *sbi)
349 spin_lock_irqsave(&sbi->fsync_node_lock, flags);
350 sbi->fsync_seg_id = 0;
351 spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
354 int f2fs_need_dentry_mark(struct f2fs_sb_info *sbi, nid_t nid)
356 struct f2fs_nm_info *nm_i = NM_I(sbi);
360 down_read(&nm_i->nat_tree_lock);
361 e = __lookup_nat_cache(nm_i, nid);
363 if (!get_nat_flag(e, IS_CHECKPOINTED) &&
364 !get_nat_flag(e, HAS_FSYNCED_INODE))
367 up_read(&nm_i->nat_tree_lock);
371 bool f2fs_is_checkpointed_node(struct f2fs_sb_info *sbi, nid_t nid)
373 struct f2fs_nm_info *nm_i = NM_I(sbi);
377 down_read(&nm_i->nat_tree_lock);
378 e = __lookup_nat_cache(nm_i, nid);
379 if (e && !get_nat_flag(e, IS_CHECKPOINTED))
381 up_read(&nm_i->nat_tree_lock);
385 bool f2fs_need_inode_block_update(struct f2fs_sb_info *sbi, nid_t ino)
387 struct f2fs_nm_info *nm_i = NM_I(sbi);
389 bool need_update = true;
391 down_read(&nm_i->nat_tree_lock);
392 e = __lookup_nat_cache(nm_i, ino);
393 if (e && get_nat_flag(e, HAS_LAST_FSYNC) &&
394 (get_nat_flag(e, IS_CHECKPOINTED) ||
395 get_nat_flag(e, HAS_FSYNCED_INODE)))
397 up_read(&nm_i->nat_tree_lock);
401 /* must be locked by nat_tree_lock */
402 static void cache_nat_entry(struct f2fs_sb_info *sbi, nid_t nid,
403 struct f2fs_nat_entry *ne)
405 struct f2fs_nm_info *nm_i = NM_I(sbi);
406 struct nat_entry *new, *e;
408 new = __alloc_nat_entry(nid, false);
412 down_write(&nm_i->nat_tree_lock);
413 e = __lookup_nat_cache(nm_i, nid);
415 e = __init_nat_entry(nm_i, new, ne, false);
417 f2fs_bug_on(sbi, nat_get_ino(e) != le32_to_cpu(ne->ino) ||
418 nat_get_blkaddr(e) !=
419 le32_to_cpu(ne->block_addr) ||
420 nat_get_version(e) != ne->version);
421 up_write(&nm_i->nat_tree_lock);
423 __free_nat_entry(new);
426 static void set_node_addr(struct f2fs_sb_info *sbi, struct node_info *ni,
427 block_t new_blkaddr, bool fsync_done)
429 struct f2fs_nm_info *nm_i = NM_I(sbi);
431 struct nat_entry *new = __alloc_nat_entry(ni->nid, true);
433 down_write(&nm_i->nat_tree_lock);
434 e = __lookup_nat_cache(nm_i, ni->nid);
436 e = __init_nat_entry(nm_i, new, NULL, true);
437 copy_node_info(&e->ni, ni);
438 f2fs_bug_on(sbi, ni->blk_addr == NEW_ADDR);
439 } else if (new_blkaddr == NEW_ADDR) {
441 * when nid is reallocated,
442 * previous nat entry can be remained in nat cache.
443 * So, reinitialize it with new information.
445 copy_node_info(&e->ni, ni);
446 f2fs_bug_on(sbi, ni->blk_addr != NULL_ADDR);
448 /* let's free early to reduce memory consumption */
450 __free_nat_entry(new);
453 f2fs_bug_on(sbi, nat_get_blkaddr(e) != ni->blk_addr);
454 f2fs_bug_on(sbi, nat_get_blkaddr(e) == NULL_ADDR &&
455 new_blkaddr == NULL_ADDR);
456 f2fs_bug_on(sbi, nat_get_blkaddr(e) == NEW_ADDR &&
457 new_blkaddr == NEW_ADDR);
458 f2fs_bug_on(sbi, is_valid_data_blkaddr(sbi, nat_get_blkaddr(e)) &&
459 new_blkaddr == NEW_ADDR);
461 /* increment version no as node is removed */
462 if (nat_get_blkaddr(e) != NEW_ADDR && new_blkaddr == NULL_ADDR) {
463 unsigned char version = nat_get_version(e);
464 nat_set_version(e, inc_node_version(version));
468 nat_set_blkaddr(e, new_blkaddr);
469 if (!is_valid_data_blkaddr(sbi, new_blkaddr))
470 set_nat_flag(e, IS_CHECKPOINTED, false);
471 __set_nat_cache_dirty(nm_i, e);
473 /* update fsync_mark if its inode nat entry is still alive */
474 if (ni->nid != ni->ino)
475 e = __lookup_nat_cache(nm_i, ni->ino);
477 if (fsync_done && ni->nid == ni->ino)
478 set_nat_flag(e, HAS_FSYNCED_INODE, true);
479 set_nat_flag(e, HAS_LAST_FSYNC, fsync_done);
481 up_write(&nm_i->nat_tree_lock);
484 int f2fs_try_to_free_nats(struct f2fs_sb_info *sbi, int nr_shrink)
486 struct f2fs_nm_info *nm_i = NM_I(sbi);
489 if (!down_write_trylock(&nm_i->nat_tree_lock))
492 spin_lock(&nm_i->nat_list_lock);
494 struct nat_entry *ne;
496 if (list_empty(&nm_i->nat_entries))
499 ne = list_first_entry(&nm_i->nat_entries,
500 struct nat_entry, list);
502 spin_unlock(&nm_i->nat_list_lock);
504 __del_from_nat_cache(nm_i, ne);
507 spin_lock(&nm_i->nat_list_lock);
509 spin_unlock(&nm_i->nat_list_lock);
511 up_write(&nm_i->nat_tree_lock);
512 return nr - nr_shrink;
516 * This function always returns success
518 int f2fs_get_node_info(struct f2fs_sb_info *sbi, nid_t nid,
519 struct node_info *ni)
521 struct f2fs_nm_info *nm_i = NM_I(sbi);
522 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
523 struct f2fs_journal *journal = curseg->journal;
524 nid_t start_nid = START_NID(nid);
525 struct f2fs_nat_block *nat_blk;
526 struct page *page = NULL;
527 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 /* cache nat entry */
574 cache_nat_entry(sbi, nid, &ne);
579 * readahead MAX_RA_NODE number of node pages.
581 static void f2fs_ra_node_pages(struct page *parent, int start, int n)
583 struct f2fs_sb_info *sbi = F2FS_P_SB(parent);
584 struct blk_plug plug;
588 blk_start_plug(&plug);
590 /* Then, try readahead for siblings of the desired node */
592 end = min(end, NIDS_PER_BLOCK);
593 for (i = start; i < end; i++) {
594 nid = get_nid(parent, i, false);
595 f2fs_ra_node_page(sbi, nid);
598 blk_finish_plug(&plug);
601 pgoff_t f2fs_get_next_page_offset(struct dnode_of_data *dn, pgoff_t pgofs)
603 const long direct_index = ADDRS_PER_INODE(dn->inode);
604 const long direct_blks = ADDRS_PER_BLOCK;
605 const long indirect_blks = ADDRS_PER_BLOCK * NIDS_PER_BLOCK;
606 unsigned int skipped_unit = ADDRS_PER_BLOCK;
607 int cur_level = dn->cur_level;
608 int max_level = dn->max_level;
614 while (max_level-- > cur_level)
615 skipped_unit *= NIDS_PER_BLOCK;
617 switch (dn->max_level) {
619 base += 2 * indirect_blks;
621 base += 2 * direct_blks;
623 base += direct_index;
626 f2fs_bug_on(F2FS_I_SB(dn->inode), 1);
629 return ((pgofs - base) / skipped_unit + 1) * skipped_unit + base;
633 * The maximum depth is four.
634 * Offset[0] will have raw inode offset.
636 static int get_node_path(struct inode *inode, long block,
637 int offset[4], unsigned int noffset[4])
639 const long direct_index = ADDRS_PER_INODE(inode);
640 const long direct_blks = ADDRS_PER_BLOCK;
641 const long dptrs_per_blk = NIDS_PER_BLOCK;
642 const long indirect_blks = ADDRS_PER_BLOCK * NIDS_PER_BLOCK;
643 const long dindirect_blks = indirect_blks * NIDS_PER_BLOCK;
649 if (block < direct_index) {
653 block -= direct_index;
654 if (block < direct_blks) {
655 offset[n++] = NODE_DIR1_BLOCK;
661 block -= direct_blks;
662 if (block < direct_blks) {
663 offset[n++] = NODE_DIR2_BLOCK;
669 block -= direct_blks;
670 if (block < indirect_blks) {
671 offset[n++] = NODE_IND1_BLOCK;
673 offset[n++] = block / direct_blks;
674 noffset[n] = 4 + offset[n - 1];
675 offset[n] = block % direct_blks;
679 block -= indirect_blks;
680 if (block < indirect_blks) {
681 offset[n++] = NODE_IND2_BLOCK;
682 noffset[n] = 4 + dptrs_per_blk;
683 offset[n++] = block / direct_blks;
684 noffset[n] = 5 + dptrs_per_blk + offset[n - 1];
685 offset[n] = block % direct_blks;
689 block -= indirect_blks;
690 if (block < dindirect_blks) {
691 offset[n++] = NODE_DIND_BLOCK;
692 noffset[n] = 5 + (dptrs_per_blk * 2);
693 offset[n++] = block / indirect_blks;
694 noffset[n] = 6 + (dptrs_per_blk * 2) +
695 offset[n - 1] * (dptrs_per_blk + 1);
696 offset[n++] = (block / direct_blks) % dptrs_per_blk;
697 noffset[n] = 7 + (dptrs_per_blk * 2) +
698 offset[n - 2] * (dptrs_per_blk + 1) +
700 offset[n] = block % direct_blks;
711 * Caller should call f2fs_put_dnode(dn).
712 * Also, it should grab and release a rwsem by calling f2fs_lock_op() and
713 * f2fs_unlock_op() only if ro is not set RDONLY_NODE.
714 * In the case of RDONLY_NODE, we don't need to care about mutex.
716 int f2fs_get_dnode_of_data(struct dnode_of_data *dn, pgoff_t index, int mode)
718 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
719 struct page *npage[4];
720 struct page *parent = NULL;
722 unsigned int noffset[4];
727 level = get_node_path(dn->inode, index, offset, noffset);
731 nids[0] = dn->inode->i_ino;
732 npage[0] = dn->inode_page;
735 npage[0] = f2fs_get_node_page(sbi, nids[0]);
736 if (IS_ERR(npage[0]))
737 return PTR_ERR(npage[0]);
740 /* if inline_data is set, should not report any block indices */
741 if (f2fs_has_inline_data(dn->inode) && index) {
743 f2fs_put_page(npage[0], 1);
749 nids[1] = get_nid(parent, offset[0], true);
750 dn->inode_page = npage[0];
751 dn->inode_page_locked = true;
753 /* get indirect or direct nodes */
754 for (i = 1; i <= level; i++) {
757 if (!nids[i] && mode == ALLOC_NODE) {
759 if (!f2fs_alloc_nid(sbi, &(nids[i]))) {
765 npage[i] = f2fs_new_node_page(dn, noffset[i]);
766 if (IS_ERR(npage[i])) {
767 f2fs_alloc_nid_failed(sbi, nids[i]);
768 err = PTR_ERR(npage[i]);
772 set_nid(parent, offset[i - 1], nids[i], i == 1);
773 f2fs_alloc_nid_done(sbi, nids[i]);
775 } else if (mode == LOOKUP_NODE_RA && i == level && level > 1) {
776 npage[i] = f2fs_get_node_page_ra(parent, offset[i - 1]);
777 if (IS_ERR(npage[i])) {
778 err = PTR_ERR(npage[i]);
784 dn->inode_page_locked = false;
787 f2fs_put_page(parent, 1);
791 npage[i] = f2fs_get_node_page(sbi, nids[i]);
792 if (IS_ERR(npage[i])) {
793 err = PTR_ERR(npage[i]);
794 f2fs_put_page(npage[0], 0);
800 nids[i + 1] = get_nid(parent, offset[i], false);
803 dn->nid = nids[level];
804 dn->ofs_in_node = offset[level];
805 dn->node_page = npage[level];
806 dn->data_blkaddr = datablock_addr(dn->inode,
807 dn->node_page, dn->ofs_in_node);
811 f2fs_put_page(parent, 1);
813 f2fs_put_page(npage[0], 0);
815 dn->inode_page = NULL;
816 dn->node_page = NULL;
817 if (err == -ENOENT) {
819 dn->max_level = level;
820 dn->ofs_in_node = offset[level];
825 static int truncate_node(struct dnode_of_data *dn)
827 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
832 err = f2fs_get_node_info(sbi, dn->nid, &ni);
836 /* Deallocate node address */
837 f2fs_invalidate_blocks(sbi, ni.blk_addr);
838 dec_valid_node_count(sbi, dn->inode, dn->nid == dn->inode->i_ino);
839 set_node_addr(sbi, &ni, NULL_ADDR, false);
841 if (dn->nid == dn->inode->i_ino) {
842 f2fs_remove_orphan_inode(sbi, dn->nid);
843 dec_valid_inode_count(sbi);
844 f2fs_inode_synced(dn->inode);
847 clear_node_page_dirty(dn->node_page);
848 set_sbi_flag(sbi, SBI_IS_DIRTY);
850 index = dn->node_page->index;
851 f2fs_put_page(dn->node_page, 1);
853 invalidate_mapping_pages(NODE_MAPPING(sbi),
856 dn->node_page = NULL;
857 trace_f2fs_truncate_node(dn->inode, dn->nid, ni.blk_addr);
862 static int truncate_dnode(struct dnode_of_data *dn)
870 /* get direct node */
871 page = f2fs_get_node_page(F2FS_I_SB(dn->inode), dn->nid);
872 if (IS_ERR(page) && PTR_ERR(page) == -ENOENT)
874 else if (IS_ERR(page))
875 return PTR_ERR(page);
877 /* Make dnode_of_data for parameter */
878 dn->node_page = page;
880 f2fs_truncate_data_blocks(dn);
881 err = truncate_node(dn);
888 static int truncate_nodes(struct dnode_of_data *dn, unsigned int nofs,
891 struct dnode_of_data rdn = *dn;
893 struct f2fs_node *rn;
895 unsigned int child_nofs;
900 return NIDS_PER_BLOCK + 1;
902 trace_f2fs_truncate_nodes_enter(dn->inode, dn->nid, dn->data_blkaddr);
904 page = f2fs_get_node_page(F2FS_I_SB(dn->inode), dn->nid);
906 trace_f2fs_truncate_nodes_exit(dn->inode, PTR_ERR(page));
907 return PTR_ERR(page);
910 f2fs_ra_node_pages(page, ofs, NIDS_PER_BLOCK);
912 rn = F2FS_NODE(page);
914 for (i = ofs; i < NIDS_PER_BLOCK; i++, freed++) {
915 child_nid = le32_to_cpu(rn->in.nid[i]);
919 ret = truncate_dnode(&rdn);
922 if (set_nid(page, i, 0, false))
923 dn->node_changed = true;
926 child_nofs = nofs + ofs * (NIDS_PER_BLOCK + 1) + 1;
927 for (i = ofs; i < NIDS_PER_BLOCK; i++) {
928 child_nid = le32_to_cpu(rn->in.nid[i]);
929 if (child_nid == 0) {
930 child_nofs += NIDS_PER_BLOCK + 1;
934 ret = truncate_nodes(&rdn, child_nofs, 0, depth - 1);
935 if (ret == (NIDS_PER_BLOCK + 1)) {
936 if (set_nid(page, i, 0, false))
937 dn->node_changed = true;
939 } else if (ret < 0 && ret != -ENOENT) {
947 /* remove current indirect node */
948 dn->node_page = page;
949 ret = truncate_node(dn);
954 f2fs_put_page(page, 1);
956 trace_f2fs_truncate_nodes_exit(dn->inode, freed);
960 f2fs_put_page(page, 1);
961 trace_f2fs_truncate_nodes_exit(dn->inode, ret);
965 static int truncate_partial_nodes(struct dnode_of_data *dn,
966 struct f2fs_inode *ri, int *offset, int depth)
968 struct page *pages[2];
975 nid[0] = le32_to_cpu(ri->i_nid[offset[0] - NODE_DIR1_BLOCK]);
979 /* get indirect nodes in the path */
980 for (i = 0; i < idx + 1; i++) {
981 /* reference count'll be increased */
982 pages[i] = f2fs_get_node_page(F2FS_I_SB(dn->inode), nid[i]);
983 if (IS_ERR(pages[i])) {
984 err = PTR_ERR(pages[i]);
988 nid[i + 1] = get_nid(pages[i], offset[i + 1], false);
991 f2fs_ra_node_pages(pages[idx], offset[idx + 1], NIDS_PER_BLOCK);
993 /* free direct nodes linked to a partial indirect node */
994 for (i = offset[idx + 1]; i < NIDS_PER_BLOCK; i++) {
995 child_nid = get_nid(pages[idx], i, false);
999 err = truncate_dnode(dn);
1002 if (set_nid(pages[idx], i, 0, false))
1003 dn->node_changed = true;
1006 if (offset[idx + 1] == 0) {
1007 dn->node_page = pages[idx];
1009 err = truncate_node(dn);
1013 f2fs_put_page(pages[idx], 1);
1016 offset[idx + 1] = 0;
1019 for (i = idx; i >= 0; i--)
1020 f2fs_put_page(pages[i], 1);
1022 trace_f2fs_truncate_partial_nodes(dn->inode, nid, depth, err);
1028 * All the block addresses of data and nodes should be nullified.
1030 int f2fs_truncate_inode_blocks(struct inode *inode, pgoff_t from)
1032 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1033 int err = 0, cont = 1;
1034 int level, offset[4], noffset[4];
1035 unsigned int nofs = 0;
1036 struct f2fs_inode *ri;
1037 struct dnode_of_data dn;
1040 trace_f2fs_truncate_inode_blocks_enter(inode, from);
1042 level = get_node_path(inode, from, offset, noffset);
1046 page = f2fs_get_node_page(sbi, inode->i_ino);
1048 trace_f2fs_truncate_inode_blocks_exit(inode, PTR_ERR(page));
1049 return PTR_ERR(page);
1052 set_new_dnode(&dn, inode, page, NULL, 0);
1055 ri = F2FS_INODE(page);
1063 if (!offset[level - 1])
1065 err = truncate_partial_nodes(&dn, ri, offset, level);
1066 if (err < 0 && err != -ENOENT)
1068 nofs += 1 + NIDS_PER_BLOCK;
1071 nofs = 5 + 2 * NIDS_PER_BLOCK;
1072 if (!offset[level - 1])
1074 err = truncate_partial_nodes(&dn, ri, offset, level);
1075 if (err < 0 && err != -ENOENT)
1084 dn.nid = le32_to_cpu(ri->i_nid[offset[0] - NODE_DIR1_BLOCK]);
1085 switch (offset[0]) {
1086 case NODE_DIR1_BLOCK:
1087 case NODE_DIR2_BLOCK:
1088 err = truncate_dnode(&dn);
1091 case NODE_IND1_BLOCK:
1092 case NODE_IND2_BLOCK:
1093 err = truncate_nodes(&dn, nofs, offset[1], 2);
1096 case NODE_DIND_BLOCK:
1097 err = truncate_nodes(&dn, nofs, offset[1], 3);
1104 if (err < 0 && err != -ENOENT)
1106 if (offset[1] == 0 &&
1107 ri->i_nid[offset[0] - NODE_DIR1_BLOCK]) {
1109 BUG_ON(page->mapping != NODE_MAPPING(sbi));
1110 f2fs_wait_on_page_writeback(page, NODE, true);
1111 ri->i_nid[offset[0] - NODE_DIR1_BLOCK] = 0;
1112 set_page_dirty(page);
1120 f2fs_put_page(page, 0);
1121 trace_f2fs_truncate_inode_blocks_exit(inode, err);
1122 return err > 0 ? 0 : err;
1125 /* caller must lock inode page */
1126 int f2fs_truncate_xattr_node(struct inode *inode)
1128 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1129 nid_t nid = F2FS_I(inode)->i_xattr_nid;
1130 struct dnode_of_data dn;
1137 npage = f2fs_get_node_page(sbi, nid);
1139 return PTR_ERR(npage);
1141 set_new_dnode(&dn, inode, NULL, npage, nid);
1142 err = truncate_node(&dn);
1144 f2fs_put_page(npage, 1);
1148 f2fs_i_xnid_write(inode, 0);
1154 * Caller should grab and release a rwsem by calling f2fs_lock_op() and
1157 int f2fs_remove_inode_page(struct inode *inode)
1159 struct dnode_of_data dn;
1162 set_new_dnode(&dn, inode, NULL, NULL, inode->i_ino);
1163 err = f2fs_get_dnode_of_data(&dn, 0, LOOKUP_NODE);
1167 err = f2fs_truncate_xattr_node(inode);
1169 f2fs_put_dnode(&dn);
1173 /* remove potential inline_data blocks */
1174 if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
1175 S_ISLNK(inode->i_mode))
1176 f2fs_truncate_data_blocks_range(&dn, 1);
1178 /* 0 is possible, after f2fs_new_inode() has failed */
1179 if (unlikely(f2fs_cp_error(F2FS_I_SB(inode)))) {
1180 f2fs_put_dnode(&dn);
1184 if (unlikely(inode->i_blocks != 0 && inode->i_blocks != 8)) {
1185 f2fs_msg(F2FS_I_SB(inode)->sb, KERN_WARNING,
1186 "Inconsistent i_blocks, ino:%lu, iblocks:%llu",
1188 (unsigned long long)inode->i_blocks);
1189 set_sbi_flag(F2FS_I_SB(inode), SBI_NEED_FSCK);
1192 /* will put inode & node pages */
1193 err = truncate_node(&dn);
1195 f2fs_put_dnode(&dn);
1201 struct page *f2fs_new_inode_page(struct inode *inode)
1203 struct dnode_of_data dn;
1205 /* allocate inode page for new inode */
1206 set_new_dnode(&dn, inode, NULL, NULL, inode->i_ino);
1208 /* caller should f2fs_put_page(page, 1); */
1209 return f2fs_new_node_page(&dn, 0);
1212 struct page *f2fs_new_node_page(struct dnode_of_data *dn, unsigned int ofs)
1214 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
1215 struct node_info new_ni;
1219 if (unlikely(is_inode_flag_set(dn->inode, FI_NO_ALLOC)))
1220 return ERR_PTR(-EPERM);
1222 page = f2fs_grab_cache_page(NODE_MAPPING(sbi), dn->nid, false);
1224 return ERR_PTR(-ENOMEM);
1226 if (unlikely((err = inc_valid_node_count(sbi, dn->inode, !ofs))))
1229 #ifdef CONFIG_F2FS_CHECK_FS
1230 err = f2fs_get_node_info(sbi, dn->nid, &new_ni);
1232 dec_valid_node_count(sbi, dn->inode, !ofs);
1235 f2fs_bug_on(sbi, new_ni.blk_addr != NULL_ADDR);
1237 new_ni.nid = dn->nid;
1238 new_ni.ino = dn->inode->i_ino;
1239 new_ni.blk_addr = NULL_ADDR;
1242 set_node_addr(sbi, &new_ni, NEW_ADDR, false);
1244 f2fs_wait_on_page_writeback(page, NODE, true);
1245 fill_node_footer(page, dn->nid, dn->inode->i_ino, ofs, true);
1246 set_cold_node(page, S_ISDIR(dn->inode->i_mode));
1247 if (!PageUptodate(page))
1248 SetPageUptodate(page);
1249 if (set_page_dirty(page))
1250 dn->node_changed = true;
1252 if (f2fs_has_xattr_block(ofs))
1253 f2fs_i_xnid_write(dn->inode, dn->nid);
1256 inc_valid_inode_count(sbi);
1260 clear_node_page_dirty(page);
1261 f2fs_put_page(page, 1);
1262 return ERR_PTR(err);
1266 * Caller should do after getting the following values.
1267 * 0: f2fs_put_page(page, 0)
1268 * LOCKED_PAGE or error: f2fs_put_page(page, 1)
1270 static int read_node_page(struct page *page, int op_flags)
1272 struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1273 struct node_info ni;
1274 struct f2fs_io_info fio = {
1278 .op_flags = op_flags,
1280 .encrypted_page = NULL,
1284 if (PageUptodate(page)) {
1285 if (!f2fs_inode_chksum_verify(sbi, page)) {
1286 ClearPageUptodate(page);
1292 err = f2fs_get_node_info(sbi, page->index, &ni);
1296 if (unlikely(ni.blk_addr == NULL_ADDR) ||
1297 is_sbi_flag_set(sbi, SBI_IS_SHUTDOWN)) {
1298 ClearPageUptodate(page);
1302 fio.new_blkaddr = fio.old_blkaddr = ni.blk_addr;
1303 return f2fs_submit_page_bio(&fio);
1307 * Readahead a node page
1309 void f2fs_ra_node_page(struct f2fs_sb_info *sbi, nid_t nid)
1316 if (f2fs_check_nid_range(sbi, nid))
1320 apage = radix_tree_lookup(&NODE_MAPPING(sbi)->i_pages, nid);
1325 apage = f2fs_grab_cache_page(NODE_MAPPING(sbi), nid, false);
1329 err = read_node_page(apage, REQ_RAHEAD);
1330 f2fs_put_page(apage, err ? 1 : 0);
1333 static struct page *__get_node_page(struct f2fs_sb_info *sbi, pgoff_t nid,
1334 struct page *parent, int start)
1340 return ERR_PTR(-ENOENT);
1341 if (f2fs_check_nid_range(sbi, nid))
1342 return ERR_PTR(-EINVAL);
1344 page = f2fs_grab_cache_page(NODE_MAPPING(sbi), nid, false);
1346 return ERR_PTR(-ENOMEM);
1348 err = read_node_page(page, 0);
1350 f2fs_put_page(page, 1);
1351 return ERR_PTR(err);
1352 } else if (err == LOCKED_PAGE) {
1358 f2fs_ra_node_pages(parent, start + 1, MAX_RA_NODE);
1362 if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1363 f2fs_put_page(page, 1);
1367 if (unlikely(!PageUptodate(page))) {
1372 if (!f2fs_inode_chksum_verify(sbi, page)) {
1377 if(unlikely(nid != nid_of_node(page))) {
1378 f2fs_msg(sbi->sb, KERN_WARNING, "inconsistent node block, "
1379 "nid:%lu, node_footer[nid:%u,ino:%u,ofs:%u,cpver:%llu,blkaddr:%u]",
1380 nid, nid_of_node(page), ino_of_node(page),
1381 ofs_of_node(page), cpver_of_node(page),
1382 next_blkaddr_of_node(page));
1385 ClearPageUptodate(page);
1386 f2fs_put_page(page, 1);
1387 return ERR_PTR(err);
1392 struct page *f2fs_get_node_page(struct f2fs_sb_info *sbi, pgoff_t nid)
1394 return __get_node_page(sbi, nid, NULL, 0);
1397 struct page *f2fs_get_node_page_ra(struct page *parent, int start)
1399 struct f2fs_sb_info *sbi = F2FS_P_SB(parent);
1400 nid_t nid = get_nid(parent, start, false);
1402 return __get_node_page(sbi, nid, parent, start);
1405 static void flush_inline_data(struct f2fs_sb_info *sbi, nid_t ino)
1407 struct inode *inode;
1411 /* should flush inline_data before evict_inode */
1412 inode = ilookup(sbi->sb, ino);
1416 page = f2fs_pagecache_get_page(inode->i_mapping, 0,
1417 FGP_LOCK|FGP_NOWAIT, 0);
1421 if (!PageUptodate(page))
1424 if (!PageDirty(page))
1427 if (!clear_page_dirty_for_io(page))
1430 ret = f2fs_write_inline_data(inode, page);
1431 inode_dec_dirty_pages(inode);
1432 f2fs_remove_dirty_inode(inode);
1434 set_page_dirty(page);
1436 f2fs_put_page(page, 1);
1441 static struct page *last_fsync_dnode(struct f2fs_sb_info *sbi, nid_t ino)
1444 struct pagevec pvec;
1445 struct page *last_page = NULL;
1448 pagevec_init(&pvec);
1451 while ((nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1452 PAGECACHE_TAG_DIRTY))) {
1455 for (i = 0; i < nr_pages; i++) {
1456 struct page *page = pvec.pages[i];
1458 if (unlikely(f2fs_cp_error(sbi))) {
1459 f2fs_put_page(last_page, 0);
1460 pagevec_release(&pvec);
1461 return ERR_PTR(-EIO);
1464 if (!IS_DNODE(page) || !is_cold_node(page))
1466 if (ino_of_node(page) != ino)
1471 if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1476 if (ino_of_node(page) != ino)
1477 goto continue_unlock;
1479 if (!PageDirty(page)) {
1480 /* someone wrote it for us */
1481 goto continue_unlock;
1485 f2fs_put_page(last_page, 0);
1491 pagevec_release(&pvec);
1497 static int __write_node_page(struct page *page, bool atomic, bool *submitted,
1498 struct writeback_control *wbc, bool do_balance,
1499 enum iostat_type io_type, unsigned int *seq_id)
1501 struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1503 struct node_info ni;
1504 struct f2fs_io_info fio = {
1506 .ino = ino_of_node(page),
1509 .op_flags = wbc_to_write_flags(wbc),
1511 .encrypted_page = NULL,
1518 trace_f2fs_writepage(page, NODE);
1520 if (unlikely(f2fs_cp_error(sbi)))
1523 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
1526 if (wbc->sync_mode == WB_SYNC_NONE &&
1527 IS_DNODE(page) && is_cold_node(page))
1530 /* get old block addr of this node page */
1531 nid = nid_of_node(page);
1532 f2fs_bug_on(sbi, page->index != nid);
1534 if (f2fs_get_node_info(sbi, nid, &ni))
1537 if (wbc->for_reclaim) {
1538 if (!down_read_trylock(&sbi->node_write))
1541 down_read(&sbi->node_write);
1544 /* This page is already truncated */
1545 if (unlikely(ni.blk_addr == NULL_ADDR)) {
1546 ClearPageUptodate(page);
1547 dec_page_count(sbi, F2FS_DIRTY_NODES);
1548 up_read(&sbi->node_write);
1553 if (__is_valid_data_blkaddr(ni.blk_addr) &&
1554 !f2fs_is_valid_blkaddr(sbi, ni.blk_addr, DATA_GENERIC)) {
1555 up_read(&sbi->node_write);
1559 if (atomic && !test_opt(sbi, NOBARRIER))
1560 fio.op_flags |= REQ_PREFLUSH | REQ_FUA;
1562 /* should add to global list before clearing PAGECACHE status */
1563 if (f2fs_in_warm_node_list(sbi, page)) {
1564 seq = f2fs_add_fsync_node_entry(sbi, page);
1569 set_page_writeback(page);
1570 ClearPageError(page);
1572 fio.old_blkaddr = ni.blk_addr;
1573 f2fs_do_write_node_page(nid, &fio);
1574 set_node_addr(sbi, &ni, fio.new_blkaddr, is_fsync_dnode(page));
1575 dec_page_count(sbi, F2FS_DIRTY_NODES);
1576 up_read(&sbi->node_write);
1578 if (wbc->for_reclaim) {
1579 f2fs_submit_merged_write_cond(sbi, page->mapping->host, 0,
1586 if (unlikely(f2fs_cp_error(sbi))) {
1587 f2fs_submit_merged_write(sbi, NODE);
1591 *submitted = fio.submitted;
1594 f2fs_balance_fs(sbi, false);
1598 redirty_page_for_writepage(wbc, page);
1599 return AOP_WRITEPAGE_ACTIVATE;
1602 void f2fs_move_node_page(struct page *node_page, int gc_type)
1604 if (gc_type == FG_GC) {
1605 struct writeback_control wbc = {
1606 .sync_mode = WB_SYNC_ALL,
1611 set_page_dirty(node_page);
1612 f2fs_wait_on_page_writeback(node_page, NODE, true);
1614 f2fs_bug_on(F2FS_P_SB(node_page), PageWriteback(node_page));
1615 if (!clear_page_dirty_for_io(node_page))
1618 if (__write_node_page(node_page, false, NULL,
1619 &wbc, false, FS_GC_NODE_IO, NULL))
1620 unlock_page(node_page);
1623 /* set page dirty and write it */
1624 if (!PageWriteback(node_page))
1625 set_page_dirty(node_page);
1628 unlock_page(node_page);
1630 f2fs_put_page(node_page, 0);
1633 static int f2fs_write_node_page(struct page *page,
1634 struct writeback_control *wbc)
1636 return __write_node_page(page, false, NULL, wbc, false,
1640 int f2fs_fsync_node_pages(struct f2fs_sb_info *sbi, struct inode *inode,
1641 struct writeback_control *wbc, bool atomic,
1642 unsigned int *seq_id)
1645 pgoff_t last_idx = ULONG_MAX;
1646 struct pagevec pvec;
1648 struct page *last_page = NULL;
1649 bool marked = false;
1650 nid_t ino = inode->i_ino;
1654 last_page = last_fsync_dnode(sbi, ino);
1655 if (IS_ERR_OR_NULL(last_page))
1656 return PTR_ERR_OR_ZERO(last_page);
1659 pagevec_init(&pvec);
1662 while ((nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1663 PAGECACHE_TAG_DIRTY))) {
1666 for (i = 0; i < nr_pages; i++) {
1667 struct page *page = pvec.pages[i];
1668 bool submitted = false;
1670 if (unlikely(f2fs_cp_error(sbi))) {
1671 f2fs_put_page(last_page, 0);
1672 pagevec_release(&pvec);
1677 if (!IS_DNODE(page) || !is_cold_node(page))
1679 if (ino_of_node(page) != ino)
1684 if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1689 if (ino_of_node(page) != ino)
1690 goto continue_unlock;
1692 if (!PageDirty(page) && page != last_page) {
1693 /* someone wrote it for us */
1694 goto continue_unlock;
1697 f2fs_wait_on_page_writeback(page, NODE, true);
1698 BUG_ON(PageWriteback(page));
1700 set_fsync_mark(page, 0);
1701 set_dentry_mark(page, 0);
1703 if (!atomic || page == last_page) {
1704 set_fsync_mark(page, 1);
1705 if (IS_INODE(page)) {
1706 if (is_inode_flag_set(inode,
1708 f2fs_update_inode(inode, page);
1709 set_dentry_mark(page,
1710 f2fs_need_dentry_mark(sbi, ino));
1712 /* may be written by other thread */
1713 if (!PageDirty(page))
1714 set_page_dirty(page);
1717 if (!clear_page_dirty_for_io(page))
1718 goto continue_unlock;
1720 ret = __write_node_page(page, atomic &&
1722 &submitted, wbc, true,
1723 FS_NODE_IO, seq_id);
1726 f2fs_put_page(last_page, 0);
1728 } else if (submitted) {
1729 last_idx = page->index;
1732 if (page == last_page) {
1733 f2fs_put_page(page, 0);
1738 pagevec_release(&pvec);
1744 if (!ret && atomic && !marked) {
1745 f2fs_msg(sbi->sb, KERN_DEBUG,
1746 "Retry to write fsync mark: ino=%u, idx=%lx",
1747 ino, last_page->index);
1748 lock_page(last_page);
1749 f2fs_wait_on_page_writeback(last_page, NODE, true);
1750 set_page_dirty(last_page);
1751 unlock_page(last_page);
1755 if (last_idx != ULONG_MAX)
1756 f2fs_submit_merged_write_cond(sbi, NULL, ino, last_idx, NODE);
1757 return ret ? -EIO: 0;
1760 int f2fs_sync_node_pages(struct f2fs_sb_info *sbi,
1761 struct writeback_control *wbc,
1762 bool do_balance, enum iostat_type io_type)
1765 struct pagevec pvec;
1769 int nr_pages, done = 0;
1771 pagevec_init(&pvec);
1776 while (!done && (nr_pages = pagevec_lookup_tag(&pvec,
1777 NODE_MAPPING(sbi), &index, PAGECACHE_TAG_DIRTY))) {
1780 for (i = 0; i < nr_pages; i++) {
1781 struct page *page = pvec.pages[i];
1782 bool submitted = false;
1784 /* give a priority to WB_SYNC threads */
1785 if (atomic_read(&sbi->wb_sync_req[NODE]) &&
1786 wbc->sync_mode == WB_SYNC_NONE) {
1792 * flushing sequence with step:
1797 if (step == 0 && IS_DNODE(page))
1799 if (step == 1 && (!IS_DNODE(page) ||
1800 is_cold_node(page)))
1802 if (step == 2 && (!IS_DNODE(page) ||
1803 !is_cold_node(page)))
1806 if (wbc->sync_mode == WB_SYNC_ALL)
1808 else if (!trylock_page(page))
1811 if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1817 if (!PageDirty(page)) {
1818 /* someone wrote it for us */
1819 goto continue_unlock;
1822 /* flush inline_data */
1823 if (is_inline_node(page)) {
1824 clear_inline_node(page);
1826 flush_inline_data(sbi, ino_of_node(page));
1830 f2fs_wait_on_page_writeback(page, NODE, true);
1832 BUG_ON(PageWriteback(page));
1833 if (!clear_page_dirty_for_io(page))
1834 goto continue_unlock;
1836 set_fsync_mark(page, 0);
1837 set_dentry_mark(page, 0);
1839 ret = __write_node_page(page, false, &submitted,
1840 wbc, do_balance, io_type, NULL);
1846 if (--wbc->nr_to_write == 0)
1849 pagevec_release(&pvec);
1852 if (wbc->nr_to_write == 0) {
1859 if (wbc->sync_mode == WB_SYNC_NONE && step == 1)
1866 f2fs_submit_merged_write(sbi, NODE);
1868 if (unlikely(f2fs_cp_error(sbi)))
1873 int f2fs_wait_on_node_pages_writeback(struct f2fs_sb_info *sbi,
1874 unsigned int seq_id)
1876 struct fsync_node_entry *fn;
1878 struct list_head *head = &sbi->fsync_node_list;
1879 unsigned long flags;
1880 unsigned int cur_seq_id = 0;
1883 while (seq_id && cur_seq_id < seq_id) {
1884 spin_lock_irqsave(&sbi->fsync_node_lock, flags);
1885 if (list_empty(head)) {
1886 spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
1889 fn = list_first_entry(head, struct fsync_node_entry, list);
1890 if (fn->seq_id > seq_id) {
1891 spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
1894 cur_seq_id = fn->seq_id;
1897 spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
1899 f2fs_wait_on_page_writeback(page, NODE, true);
1900 if (TestClearPageError(page))
1909 ret2 = filemap_check_errors(NODE_MAPPING(sbi));
1916 static int f2fs_write_node_pages(struct address_space *mapping,
1917 struct writeback_control *wbc)
1919 struct f2fs_sb_info *sbi = F2FS_M_SB(mapping);
1920 struct blk_plug plug;
1923 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
1926 /* balancing f2fs's metadata in background */
1927 f2fs_balance_fs_bg(sbi);
1929 /* collect a number of dirty node pages and write together */
1930 if (get_pages(sbi, F2FS_DIRTY_NODES) < nr_pages_to_skip(sbi, NODE))
1933 if (wbc->sync_mode == WB_SYNC_ALL)
1934 atomic_inc(&sbi->wb_sync_req[NODE]);
1935 else if (atomic_read(&sbi->wb_sync_req[NODE]))
1938 trace_f2fs_writepages(mapping->host, wbc, NODE);
1940 diff = nr_pages_to_write(sbi, NODE, wbc);
1941 blk_start_plug(&plug);
1942 f2fs_sync_node_pages(sbi, wbc, true, FS_NODE_IO);
1943 blk_finish_plug(&plug);
1944 wbc->nr_to_write = max((long)0, wbc->nr_to_write - diff);
1946 if (wbc->sync_mode == WB_SYNC_ALL)
1947 atomic_dec(&sbi->wb_sync_req[NODE]);
1951 wbc->pages_skipped += get_pages(sbi, F2FS_DIRTY_NODES);
1952 trace_f2fs_writepages(mapping->host, wbc, NODE);
1956 static int f2fs_set_node_page_dirty(struct page *page)
1958 trace_f2fs_set_page_dirty(page, NODE);
1960 if (!PageUptodate(page))
1961 SetPageUptodate(page);
1962 #ifdef CONFIG_F2FS_CHECK_FS
1964 f2fs_inode_chksum_set(F2FS_P_SB(page), page);
1966 if (!PageDirty(page)) {
1967 __set_page_dirty_nobuffers(page);
1968 inc_page_count(F2FS_P_SB(page), F2FS_DIRTY_NODES);
1969 SetPagePrivate(page);
1970 f2fs_trace_pid(page);
1977 * Structure of the f2fs node operations
1979 const struct address_space_operations f2fs_node_aops = {
1980 .writepage = f2fs_write_node_page,
1981 .writepages = f2fs_write_node_pages,
1982 .set_page_dirty = f2fs_set_node_page_dirty,
1983 .invalidatepage = f2fs_invalidate_page,
1984 .releasepage = f2fs_release_page,
1985 #ifdef CONFIG_MIGRATION
1986 .migratepage = f2fs_migrate_page,
1990 static struct free_nid *__lookup_free_nid_list(struct f2fs_nm_info *nm_i,
1993 return radix_tree_lookup(&nm_i->free_nid_root, n);
1996 static int __insert_free_nid(struct f2fs_sb_info *sbi,
1997 struct free_nid *i, enum nid_state state)
1999 struct f2fs_nm_info *nm_i = NM_I(sbi);
2001 int err = radix_tree_insert(&nm_i->free_nid_root, i->nid, i);
2005 f2fs_bug_on(sbi, state != i->state);
2006 nm_i->nid_cnt[state]++;
2007 if (state == FREE_NID)
2008 list_add_tail(&i->list, &nm_i->free_nid_list);
2012 static void __remove_free_nid(struct f2fs_sb_info *sbi,
2013 struct free_nid *i, enum nid_state state)
2015 struct f2fs_nm_info *nm_i = NM_I(sbi);
2017 f2fs_bug_on(sbi, state != i->state);
2018 nm_i->nid_cnt[state]--;
2019 if (state == FREE_NID)
2021 radix_tree_delete(&nm_i->free_nid_root, i->nid);
2024 static void __move_free_nid(struct f2fs_sb_info *sbi, struct free_nid *i,
2025 enum nid_state org_state, enum nid_state dst_state)
2027 struct f2fs_nm_info *nm_i = NM_I(sbi);
2029 f2fs_bug_on(sbi, org_state != i->state);
2030 i->state = dst_state;
2031 nm_i->nid_cnt[org_state]--;
2032 nm_i->nid_cnt[dst_state]++;
2034 switch (dst_state) {
2039 list_add_tail(&i->list, &nm_i->free_nid_list);
2046 static void update_free_nid_bitmap(struct f2fs_sb_info *sbi, nid_t nid,
2047 bool set, bool build)
2049 struct f2fs_nm_info *nm_i = NM_I(sbi);
2050 unsigned int nat_ofs = NAT_BLOCK_OFFSET(nid);
2051 unsigned int nid_ofs = nid - START_NID(nid);
2053 if (!test_bit_le(nat_ofs, nm_i->nat_block_bitmap))
2057 if (test_bit_le(nid_ofs, nm_i->free_nid_bitmap[nat_ofs]))
2059 __set_bit_le(nid_ofs, nm_i->free_nid_bitmap[nat_ofs]);
2060 nm_i->free_nid_count[nat_ofs]++;
2062 if (!test_bit_le(nid_ofs, nm_i->free_nid_bitmap[nat_ofs]))
2064 __clear_bit_le(nid_ofs, nm_i->free_nid_bitmap[nat_ofs]);
2066 nm_i->free_nid_count[nat_ofs]--;
2070 /* return if the nid is recognized as free */
2071 static bool add_free_nid(struct f2fs_sb_info *sbi,
2072 nid_t nid, bool build, bool update)
2074 struct f2fs_nm_info *nm_i = NM_I(sbi);
2075 struct free_nid *i, *e;
2076 struct nat_entry *ne;
2080 /* 0 nid should not be used */
2081 if (unlikely(nid == 0))
2084 if (unlikely(f2fs_check_nid_range(sbi, nid)))
2087 i = f2fs_kmem_cache_alloc(free_nid_slab, GFP_NOFS);
2089 i->state = FREE_NID;
2091 radix_tree_preload(GFP_NOFS | __GFP_NOFAIL);
2093 spin_lock(&nm_i->nid_list_lock);
2101 * - __insert_nid_to_list(PREALLOC_NID)
2102 * - f2fs_balance_fs_bg
2103 * - f2fs_build_free_nids
2104 * - __f2fs_build_free_nids
2107 * - __lookup_nat_cache
2109 * - f2fs_init_inode_metadata
2110 * - f2fs_new_inode_page
2111 * - f2fs_new_node_page
2113 * - f2fs_alloc_nid_done
2114 * - __remove_nid_from_list(PREALLOC_NID)
2115 * - __insert_nid_to_list(FREE_NID)
2117 ne = __lookup_nat_cache(nm_i, nid);
2118 if (ne && (!get_nat_flag(ne, IS_CHECKPOINTED) ||
2119 nat_get_blkaddr(ne) != NULL_ADDR))
2122 e = __lookup_free_nid_list(nm_i, nid);
2124 if (e->state == FREE_NID)
2130 err = __insert_free_nid(sbi, i, FREE_NID);
2133 update_free_nid_bitmap(sbi, nid, ret, build);
2135 nm_i->available_nids++;
2137 spin_unlock(&nm_i->nid_list_lock);
2138 radix_tree_preload_end();
2141 kmem_cache_free(free_nid_slab, i);
2145 static void remove_free_nid(struct f2fs_sb_info *sbi, nid_t nid)
2147 struct f2fs_nm_info *nm_i = NM_I(sbi);
2149 bool need_free = false;
2151 spin_lock(&nm_i->nid_list_lock);
2152 i = __lookup_free_nid_list(nm_i, nid);
2153 if (i && i->state == FREE_NID) {
2154 __remove_free_nid(sbi, i, FREE_NID);
2157 spin_unlock(&nm_i->nid_list_lock);
2160 kmem_cache_free(free_nid_slab, i);
2163 static int scan_nat_page(struct f2fs_sb_info *sbi,
2164 struct page *nat_page, nid_t start_nid)
2166 struct f2fs_nm_info *nm_i = NM_I(sbi);
2167 struct f2fs_nat_block *nat_blk = page_address(nat_page);
2169 unsigned int nat_ofs = NAT_BLOCK_OFFSET(start_nid);
2172 __set_bit_le(nat_ofs, nm_i->nat_block_bitmap);
2174 i = start_nid % NAT_ENTRY_PER_BLOCK;
2176 for (; i < NAT_ENTRY_PER_BLOCK; i++, start_nid++) {
2177 if (unlikely(start_nid >= nm_i->max_nid))
2180 blk_addr = le32_to_cpu(nat_blk->entries[i].block_addr);
2182 if (blk_addr == NEW_ADDR)
2185 if (blk_addr == NULL_ADDR) {
2186 add_free_nid(sbi, start_nid, true, true);
2188 spin_lock(&NM_I(sbi)->nid_list_lock);
2189 update_free_nid_bitmap(sbi, start_nid, false, true);
2190 spin_unlock(&NM_I(sbi)->nid_list_lock);
2197 static void scan_curseg_cache(struct f2fs_sb_info *sbi)
2199 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
2200 struct f2fs_journal *journal = curseg->journal;
2203 down_read(&curseg->journal_rwsem);
2204 for (i = 0; i < nats_in_cursum(journal); i++) {
2208 addr = le32_to_cpu(nat_in_journal(journal, i).block_addr);
2209 nid = le32_to_cpu(nid_in_journal(journal, i));
2210 if (addr == NULL_ADDR)
2211 add_free_nid(sbi, nid, true, false);
2213 remove_free_nid(sbi, nid);
2215 up_read(&curseg->journal_rwsem);
2218 static void scan_free_nid_bits(struct f2fs_sb_info *sbi)
2220 struct f2fs_nm_info *nm_i = NM_I(sbi);
2221 unsigned int i, idx;
2224 down_read(&nm_i->nat_tree_lock);
2226 for (i = 0; i < nm_i->nat_blocks; i++) {
2227 if (!test_bit_le(i, nm_i->nat_block_bitmap))
2229 if (!nm_i->free_nid_count[i])
2231 for (idx = 0; idx < NAT_ENTRY_PER_BLOCK; idx++) {
2232 idx = find_next_bit_le(nm_i->free_nid_bitmap[i],
2233 NAT_ENTRY_PER_BLOCK, idx);
2234 if (idx >= NAT_ENTRY_PER_BLOCK)
2237 nid = i * NAT_ENTRY_PER_BLOCK + idx;
2238 add_free_nid(sbi, nid, true, false);
2240 if (nm_i->nid_cnt[FREE_NID] >= MAX_FREE_NIDS)
2245 scan_curseg_cache(sbi);
2247 up_read(&nm_i->nat_tree_lock);
2250 static int __f2fs_build_free_nids(struct f2fs_sb_info *sbi,
2251 bool sync, bool mount)
2253 struct f2fs_nm_info *nm_i = NM_I(sbi);
2255 nid_t nid = nm_i->next_scan_nid;
2257 if (unlikely(nid >= nm_i->max_nid))
2260 if (unlikely(nid % NAT_ENTRY_PER_BLOCK))
2261 nid = NAT_BLOCK_OFFSET(nid) * NAT_ENTRY_PER_BLOCK;
2263 /* Enough entries */
2264 if (nm_i->nid_cnt[FREE_NID] >= NAT_ENTRY_PER_BLOCK)
2267 if (!sync && !f2fs_available_free_memory(sbi, FREE_NIDS))
2271 /* try to find free nids in free_nid_bitmap */
2272 scan_free_nid_bits(sbi);
2274 if (nm_i->nid_cnt[FREE_NID] >= NAT_ENTRY_PER_BLOCK)
2278 /* readahead nat pages to be scanned */
2279 f2fs_ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nid), FREE_NID_PAGES,
2282 down_read(&nm_i->nat_tree_lock);
2285 if (!test_bit_le(NAT_BLOCK_OFFSET(nid),
2286 nm_i->nat_block_bitmap)) {
2287 struct page *page = get_current_nat_page(sbi, nid);
2289 ret = scan_nat_page(sbi, page, nid);
2290 f2fs_put_page(page, 1);
2293 up_read(&nm_i->nat_tree_lock);
2294 f2fs_bug_on(sbi, !mount);
2295 f2fs_msg(sbi->sb, KERN_ERR,
2296 "NAT is corrupt, run fsck to fix it");
2301 nid += (NAT_ENTRY_PER_BLOCK - (nid % NAT_ENTRY_PER_BLOCK));
2302 if (unlikely(nid >= nm_i->max_nid))
2305 if (++i >= FREE_NID_PAGES)
2309 /* go to the next free nat pages to find free nids abundantly */
2310 nm_i->next_scan_nid = nid;
2312 /* find free nids from current sum_pages */
2313 scan_curseg_cache(sbi);
2315 up_read(&nm_i->nat_tree_lock);
2317 f2fs_ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nm_i->next_scan_nid),
2318 nm_i->ra_nid_pages, META_NAT, false);
2323 int f2fs_build_free_nids(struct f2fs_sb_info *sbi, bool sync, bool mount)
2327 mutex_lock(&NM_I(sbi)->build_lock);
2328 ret = __f2fs_build_free_nids(sbi, sync, mount);
2329 mutex_unlock(&NM_I(sbi)->build_lock);
2335 * If this function returns success, caller can obtain a new nid
2336 * from second parameter of this function.
2337 * The returned nid could be used ino as well as nid when inode is created.
2339 bool f2fs_alloc_nid(struct f2fs_sb_info *sbi, nid_t *nid)
2341 struct f2fs_nm_info *nm_i = NM_I(sbi);
2342 struct free_nid *i = NULL;
2344 if (time_to_inject(sbi, FAULT_ALLOC_NID)) {
2345 f2fs_show_injection_info(FAULT_ALLOC_NID);
2349 spin_lock(&nm_i->nid_list_lock);
2351 if (unlikely(nm_i->available_nids == 0)) {
2352 spin_unlock(&nm_i->nid_list_lock);
2356 /* We should not use stale free nids created by f2fs_build_free_nids */
2357 if (nm_i->nid_cnt[FREE_NID] && !on_f2fs_build_free_nids(nm_i)) {
2358 f2fs_bug_on(sbi, list_empty(&nm_i->free_nid_list));
2359 i = list_first_entry(&nm_i->free_nid_list,
2360 struct free_nid, list);
2363 __move_free_nid(sbi, i, FREE_NID, PREALLOC_NID);
2364 nm_i->available_nids--;
2366 update_free_nid_bitmap(sbi, *nid, false, false);
2368 spin_unlock(&nm_i->nid_list_lock);
2371 spin_unlock(&nm_i->nid_list_lock);
2373 /* Let's scan nat pages and its caches to get free nids */
2374 if (!f2fs_build_free_nids(sbi, true, false))
2380 * f2fs_alloc_nid() should be called prior to this function.
2382 void f2fs_alloc_nid_done(struct f2fs_sb_info *sbi, nid_t nid)
2384 struct f2fs_nm_info *nm_i = NM_I(sbi);
2387 spin_lock(&nm_i->nid_list_lock);
2388 i = __lookup_free_nid_list(nm_i, nid);
2389 f2fs_bug_on(sbi, !i);
2390 __remove_free_nid(sbi, i, PREALLOC_NID);
2391 spin_unlock(&nm_i->nid_list_lock);
2393 kmem_cache_free(free_nid_slab, i);
2397 * f2fs_alloc_nid() should be called prior to this function.
2399 void f2fs_alloc_nid_failed(struct f2fs_sb_info *sbi, nid_t nid)
2401 struct f2fs_nm_info *nm_i = NM_I(sbi);
2403 bool need_free = false;
2408 spin_lock(&nm_i->nid_list_lock);
2409 i = __lookup_free_nid_list(nm_i, nid);
2410 f2fs_bug_on(sbi, !i);
2412 if (!f2fs_available_free_memory(sbi, FREE_NIDS)) {
2413 __remove_free_nid(sbi, i, PREALLOC_NID);
2416 __move_free_nid(sbi, i, PREALLOC_NID, FREE_NID);
2419 nm_i->available_nids++;
2421 update_free_nid_bitmap(sbi, nid, true, false);
2423 spin_unlock(&nm_i->nid_list_lock);
2426 kmem_cache_free(free_nid_slab, i);
2429 int f2fs_try_to_free_nids(struct f2fs_sb_info *sbi, int nr_shrink)
2431 struct f2fs_nm_info *nm_i = NM_I(sbi);
2432 struct free_nid *i, *next;
2435 if (nm_i->nid_cnt[FREE_NID] <= MAX_FREE_NIDS)
2438 if (!mutex_trylock(&nm_i->build_lock))
2441 spin_lock(&nm_i->nid_list_lock);
2442 list_for_each_entry_safe(i, next, &nm_i->free_nid_list, list) {
2443 if (nr_shrink <= 0 ||
2444 nm_i->nid_cnt[FREE_NID] <= MAX_FREE_NIDS)
2447 __remove_free_nid(sbi, i, FREE_NID);
2448 kmem_cache_free(free_nid_slab, i);
2451 spin_unlock(&nm_i->nid_list_lock);
2452 mutex_unlock(&nm_i->build_lock);
2454 return nr - nr_shrink;
2457 int f2fs_recover_inline_xattr(struct inode *inode, struct page *page)
2459 void *src_addr, *dst_addr;
2462 struct f2fs_inode *ri;
2464 ipage = f2fs_get_node_page(F2FS_I_SB(inode), inode->i_ino);
2466 return PTR_ERR(ipage);
2468 ri = F2FS_INODE(page);
2469 if (ri->i_inline & F2FS_INLINE_XATTR) {
2470 set_inode_flag(inode, FI_INLINE_XATTR);
2472 clear_inode_flag(inode, FI_INLINE_XATTR);
2476 dst_addr = inline_xattr_addr(inode, ipage);
2477 src_addr = inline_xattr_addr(inode, page);
2478 inline_size = inline_xattr_size(inode);
2480 f2fs_wait_on_page_writeback(ipage, NODE, true);
2481 memcpy(dst_addr, src_addr, inline_size);
2483 f2fs_update_inode(inode, ipage);
2484 f2fs_put_page(ipage, 1);
2488 int f2fs_recover_xattr_data(struct inode *inode, struct page *page)
2490 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
2491 nid_t prev_xnid = F2FS_I(inode)->i_xattr_nid;
2493 struct dnode_of_data dn;
2494 struct node_info ni;
2501 /* 1: invalidate the previous xattr nid */
2502 err = f2fs_get_node_info(sbi, prev_xnid, &ni);
2506 f2fs_invalidate_blocks(sbi, ni.blk_addr);
2507 dec_valid_node_count(sbi, inode, false);
2508 set_node_addr(sbi, &ni, NULL_ADDR, false);
2511 /* 2: update xattr nid in inode */
2512 if (!f2fs_alloc_nid(sbi, &new_xnid))
2515 set_new_dnode(&dn, inode, NULL, NULL, new_xnid);
2516 xpage = f2fs_new_node_page(&dn, XATTR_NODE_OFFSET);
2517 if (IS_ERR(xpage)) {
2518 f2fs_alloc_nid_failed(sbi, new_xnid);
2519 return PTR_ERR(xpage);
2522 f2fs_alloc_nid_done(sbi, new_xnid);
2523 f2fs_update_inode_page(inode);
2525 /* 3: update and set xattr node page dirty */
2526 memcpy(F2FS_NODE(xpage), F2FS_NODE(page), VALID_XATTR_BLOCK_SIZE);
2528 set_page_dirty(xpage);
2529 f2fs_put_page(xpage, 1);
2534 int f2fs_recover_inode_page(struct f2fs_sb_info *sbi, struct page *page)
2536 struct f2fs_inode *src, *dst;
2537 nid_t ino = ino_of_node(page);
2538 struct node_info old_ni, new_ni;
2542 err = f2fs_get_node_info(sbi, ino, &old_ni);
2546 if (unlikely(old_ni.blk_addr != NULL_ADDR))
2549 ipage = f2fs_grab_cache_page(NODE_MAPPING(sbi), ino, false);
2551 congestion_wait(BLK_RW_ASYNC, HZ/50);
2555 /* Should not use this inode from free nid list */
2556 remove_free_nid(sbi, ino);
2558 if (!PageUptodate(ipage))
2559 SetPageUptodate(ipage);
2560 fill_node_footer(ipage, ino, ino, 0, true);
2561 set_cold_node(ipage, false);
2563 src = F2FS_INODE(page);
2564 dst = F2FS_INODE(ipage);
2566 memcpy(dst, src, (unsigned long)&src->i_ext - (unsigned long)src);
2568 dst->i_blocks = cpu_to_le64(1);
2569 dst->i_links = cpu_to_le32(1);
2570 dst->i_xattr_nid = 0;
2571 dst->i_inline = src->i_inline & (F2FS_INLINE_XATTR | F2FS_EXTRA_ATTR);
2572 if (dst->i_inline & F2FS_EXTRA_ATTR) {
2573 dst->i_extra_isize = src->i_extra_isize;
2575 if (f2fs_sb_has_flexible_inline_xattr(sbi->sb) &&
2576 F2FS_FITS_IN_INODE(src, le16_to_cpu(src->i_extra_isize),
2577 i_inline_xattr_size))
2578 dst->i_inline_xattr_size = src->i_inline_xattr_size;
2580 if (f2fs_sb_has_project_quota(sbi->sb) &&
2581 F2FS_FITS_IN_INODE(src, le16_to_cpu(src->i_extra_isize),
2583 dst->i_projid = src->i_projid;
2585 if (f2fs_sb_has_inode_crtime(sbi->sb) &&
2586 F2FS_FITS_IN_INODE(src, le16_to_cpu(src->i_extra_isize),
2588 dst->i_crtime = src->i_crtime;
2589 dst->i_crtime_nsec = src->i_crtime_nsec;
2596 if (unlikely(inc_valid_node_count(sbi, NULL, true)))
2598 set_node_addr(sbi, &new_ni, NEW_ADDR, false);
2599 inc_valid_inode_count(sbi);
2600 set_page_dirty(ipage);
2601 f2fs_put_page(ipage, 1);
2605 int f2fs_restore_node_summary(struct f2fs_sb_info *sbi,
2606 unsigned int segno, struct f2fs_summary_block *sum)
2608 struct f2fs_node *rn;
2609 struct f2fs_summary *sum_entry;
2611 int i, idx, last_offset, nrpages;
2613 /* scan the node segment */
2614 last_offset = sbi->blocks_per_seg;
2615 addr = START_BLOCK(sbi, segno);
2616 sum_entry = &sum->entries[0];
2618 for (i = 0; i < last_offset; i += nrpages, addr += nrpages) {
2619 nrpages = min(last_offset - i, BIO_MAX_PAGES);
2621 /* readahead node pages */
2622 f2fs_ra_meta_pages(sbi, addr, nrpages, META_POR, true);
2624 for (idx = addr; idx < addr + nrpages; idx++) {
2625 struct page *page = f2fs_get_tmp_page(sbi, idx);
2628 return PTR_ERR(page);
2630 rn = F2FS_NODE(page);
2631 sum_entry->nid = rn->footer.nid;
2632 sum_entry->version = 0;
2633 sum_entry->ofs_in_node = 0;
2635 f2fs_put_page(page, 1);
2638 invalidate_mapping_pages(META_MAPPING(sbi), addr,
2644 static void remove_nats_in_journal(struct f2fs_sb_info *sbi)
2646 struct f2fs_nm_info *nm_i = NM_I(sbi);
2647 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
2648 struct f2fs_journal *journal = curseg->journal;
2651 down_write(&curseg->journal_rwsem);
2652 for (i = 0; i < nats_in_cursum(journal); i++) {
2653 struct nat_entry *ne;
2654 struct f2fs_nat_entry raw_ne;
2655 nid_t nid = le32_to_cpu(nid_in_journal(journal, i));
2657 if (f2fs_check_nid_range(sbi, nid))
2660 raw_ne = nat_in_journal(journal, i);
2662 ne = __lookup_nat_cache(nm_i, nid);
2664 ne = __alloc_nat_entry(nid, true);
2665 __init_nat_entry(nm_i, ne, &raw_ne, true);
2669 * if a free nat in journal has not been used after last
2670 * checkpoint, we should remove it from available nids,
2671 * since later we will add it again.
2673 if (!get_nat_flag(ne, IS_DIRTY) &&
2674 le32_to_cpu(raw_ne.block_addr) == NULL_ADDR) {
2675 spin_lock(&nm_i->nid_list_lock);
2676 nm_i->available_nids--;
2677 spin_unlock(&nm_i->nid_list_lock);
2680 __set_nat_cache_dirty(nm_i, ne);
2682 update_nats_in_cursum(journal, -i);
2683 up_write(&curseg->journal_rwsem);
2686 static void __adjust_nat_entry_set(struct nat_entry_set *nes,
2687 struct list_head *head, int max)
2689 struct nat_entry_set *cur;
2691 if (nes->entry_cnt >= max)
2694 list_for_each_entry(cur, head, set_list) {
2695 if (cur->entry_cnt >= nes->entry_cnt) {
2696 list_add(&nes->set_list, cur->set_list.prev);
2701 list_add_tail(&nes->set_list, head);
2704 static void __update_nat_bits(struct f2fs_sb_info *sbi, nid_t start_nid,
2707 struct f2fs_nm_info *nm_i = NM_I(sbi);
2708 unsigned int nat_index = start_nid / NAT_ENTRY_PER_BLOCK;
2709 struct f2fs_nat_block *nat_blk = page_address(page);
2713 if (!enabled_nat_bits(sbi, NULL))
2716 if (nat_index == 0) {
2720 for (; i < NAT_ENTRY_PER_BLOCK; i++) {
2721 if (nat_blk->entries[i].block_addr != NULL_ADDR)
2725 __set_bit_le(nat_index, nm_i->empty_nat_bits);
2726 __clear_bit_le(nat_index, nm_i->full_nat_bits);
2730 __clear_bit_le(nat_index, nm_i->empty_nat_bits);
2731 if (valid == NAT_ENTRY_PER_BLOCK)
2732 __set_bit_le(nat_index, nm_i->full_nat_bits);
2734 __clear_bit_le(nat_index, nm_i->full_nat_bits);
2737 static void __flush_nat_entry_set(struct f2fs_sb_info *sbi,
2738 struct nat_entry_set *set, struct cp_control *cpc)
2740 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
2741 struct f2fs_journal *journal = curseg->journal;
2742 nid_t start_nid = set->set * NAT_ENTRY_PER_BLOCK;
2743 bool to_journal = true;
2744 struct f2fs_nat_block *nat_blk;
2745 struct nat_entry *ne, *cur;
2746 struct page *page = NULL;
2749 * there are two steps to flush nat entries:
2750 * #1, flush nat entries to journal in current hot data summary block.
2751 * #2, flush nat entries to nat page.
2753 if (enabled_nat_bits(sbi, cpc) ||
2754 !__has_cursum_space(journal, set->entry_cnt, NAT_JOURNAL))
2758 down_write(&curseg->journal_rwsem);
2760 page = get_next_nat_page(sbi, start_nid);
2761 nat_blk = page_address(page);
2762 f2fs_bug_on(sbi, !nat_blk);
2765 /* flush dirty nats in nat entry set */
2766 list_for_each_entry_safe(ne, cur, &set->entry_list, list) {
2767 struct f2fs_nat_entry *raw_ne;
2768 nid_t nid = nat_get_nid(ne);
2771 f2fs_bug_on(sbi, nat_get_blkaddr(ne) == NEW_ADDR);
2774 offset = f2fs_lookup_journal_in_cursum(journal,
2775 NAT_JOURNAL, nid, 1);
2776 f2fs_bug_on(sbi, offset < 0);
2777 raw_ne = &nat_in_journal(journal, offset);
2778 nid_in_journal(journal, offset) = cpu_to_le32(nid);
2780 raw_ne = &nat_blk->entries[nid - start_nid];
2782 raw_nat_from_node_info(raw_ne, &ne->ni);
2784 __clear_nat_cache_dirty(NM_I(sbi), set, ne);
2785 if (nat_get_blkaddr(ne) == NULL_ADDR) {
2786 add_free_nid(sbi, nid, false, true);
2788 spin_lock(&NM_I(sbi)->nid_list_lock);
2789 update_free_nid_bitmap(sbi, nid, false, false);
2790 spin_unlock(&NM_I(sbi)->nid_list_lock);
2795 up_write(&curseg->journal_rwsem);
2797 __update_nat_bits(sbi, start_nid, page);
2798 f2fs_put_page(page, 1);
2801 /* Allow dirty nats by node block allocation in write_begin */
2802 if (!set->entry_cnt) {
2803 radix_tree_delete(&NM_I(sbi)->nat_set_root, set->set);
2804 kmem_cache_free(nat_entry_set_slab, set);
2809 * This function is called during the checkpointing process.
2811 void f2fs_flush_nat_entries(struct f2fs_sb_info *sbi, struct cp_control *cpc)
2813 struct f2fs_nm_info *nm_i = NM_I(sbi);
2814 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
2815 struct f2fs_journal *journal = curseg->journal;
2816 struct nat_entry_set *setvec[SETVEC_SIZE];
2817 struct nat_entry_set *set, *tmp;
2822 /* during unmount, let's flush nat_bits before checking dirty_nat_cnt */
2823 if (enabled_nat_bits(sbi, cpc)) {
2824 down_write(&nm_i->nat_tree_lock);
2825 remove_nats_in_journal(sbi);
2826 up_write(&nm_i->nat_tree_lock);
2829 if (!nm_i->dirty_nat_cnt)
2832 down_write(&nm_i->nat_tree_lock);
2835 * if there are no enough space in journal to store dirty nat
2836 * entries, remove all entries from journal and merge them
2837 * into nat entry set.
2839 if (enabled_nat_bits(sbi, cpc) ||
2840 !__has_cursum_space(journal, nm_i->dirty_nat_cnt, NAT_JOURNAL))
2841 remove_nats_in_journal(sbi);
2843 while ((found = __gang_lookup_nat_set(nm_i,
2844 set_idx, SETVEC_SIZE, setvec))) {
2846 set_idx = setvec[found - 1]->set + 1;
2847 for (idx = 0; idx < found; idx++)
2848 __adjust_nat_entry_set(setvec[idx], &sets,
2849 MAX_NAT_JENTRIES(journal));
2852 /* flush dirty nats in nat entry set */
2853 list_for_each_entry_safe(set, tmp, &sets, set_list)
2854 __flush_nat_entry_set(sbi, set, cpc);
2856 up_write(&nm_i->nat_tree_lock);
2857 /* Allow dirty nats by node block allocation in write_begin */
2860 static int __get_nat_bitmaps(struct f2fs_sb_info *sbi)
2862 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
2863 struct f2fs_nm_info *nm_i = NM_I(sbi);
2864 unsigned int nat_bits_bytes = nm_i->nat_blocks / BITS_PER_BYTE;
2866 __u64 cp_ver = cur_cp_version(ckpt);
2867 block_t nat_bits_addr;
2869 if (!enabled_nat_bits(sbi, NULL))
2872 nm_i->nat_bits_blocks = F2FS_BLK_ALIGN((nat_bits_bytes << 1) + 8);
2873 nm_i->nat_bits = f2fs_kzalloc(sbi,
2874 nm_i->nat_bits_blocks << F2FS_BLKSIZE_BITS, GFP_KERNEL);
2875 if (!nm_i->nat_bits)
2878 nat_bits_addr = __start_cp_addr(sbi) + sbi->blocks_per_seg -
2879 nm_i->nat_bits_blocks;
2880 for (i = 0; i < nm_i->nat_bits_blocks; i++) {
2883 page = f2fs_get_meta_page(sbi, nat_bits_addr++);
2885 disable_nat_bits(sbi, true);
2886 return PTR_ERR(page);
2889 memcpy(nm_i->nat_bits + (i << F2FS_BLKSIZE_BITS),
2890 page_address(page), F2FS_BLKSIZE);
2891 f2fs_put_page(page, 1);
2894 cp_ver |= (cur_cp_crc(ckpt) << 32);
2895 if (cpu_to_le64(cp_ver) != *(__le64 *)nm_i->nat_bits) {
2896 disable_nat_bits(sbi, true);
2900 nm_i->full_nat_bits = nm_i->nat_bits + 8;
2901 nm_i->empty_nat_bits = nm_i->full_nat_bits + nat_bits_bytes;
2903 f2fs_msg(sbi->sb, KERN_NOTICE, "Found nat_bits in checkpoint");
2907 static inline void load_free_nid_bitmap(struct f2fs_sb_info *sbi)
2909 struct f2fs_nm_info *nm_i = NM_I(sbi);
2911 nid_t nid, last_nid;
2913 if (!enabled_nat_bits(sbi, NULL))
2916 for (i = 0; i < nm_i->nat_blocks; i++) {
2917 i = find_next_bit_le(nm_i->empty_nat_bits, nm_i->nat_blocks, i);
2918 if (i >= nm_i->nat_blocks)
2921 __set_bit_le(i, nm_i->nat_block_bitmap);
2923 nid = i * NAT_ENTRY_PER_BLOCK;
2924 last_nid = nid + NAT_ENTRY_PER_BLOCK;
2926 spin_lock(&NM_I(sbi)->nid_list_lock);
2927 for (; nid < last_nid; nid++)
2928 update_free_nid_bitmap(sbi, nid, true, true);
2929 spin_unlock(&NM_I(sbi)->nid_list_lock);
2932 for (i = 0; i < nm_i->nat_blocks; i++) {
2933 i = find_next_bit_le(nm_i->full_nat_bits, nm_i->nat_blocks, i);
2934 if (i >= nm_i->nat_blocks)
2937 __set_bit_le(i, nm_i->nat_block_bitmap);
2941 static int init_node_manager(struct f2fs_sb_info *sbi)
2943 struct f2fs_super_block *sb_raw = F2FS_RAW_SUPER(sbi);
2944 struct f2fs_nm_info *nm_i = NM_I(sbi);
2945 unsigned char *version_bitmap;
2946 unsigned int nat_segs;
2949 nm_i->nat_blkaddr = le32_to_cpu(sb_raw->nat_blkaddr);
2951 /* segment_count_nat includes pair segment so divide to 2. */
2952 nat_segs = le32_to_cpu(sb_raw->segment_count_nat) >> 1;
2953 nm_i->nat_blocks = nat_segs << le32_to_cpu(sb_raw->log_blocks_per_seg);
2954 nm_i->max_nid = NAT_ENTRY_PER_BLOCK * nm_i->nat_blocks;
2956 /* not used nids: 0, node, meta, (and root counted as valid node) */
2957 nm_i->available_nids = nm_i->max_nid - sbi->total_valid_node_count -
2958 sbi->nquota_files - F2FS_RESERVED_NODE_NUM;
2959 nm_i->nid_cnt[FREE_NID] = 0;
2960 nm_i->nid_cnt[PREALLOC_NID] = 0;
2962 nm_i->ram_thresh = DEF_RAM_THRESHOLD;
2963 nm_i->ra_nid_pages = DEF_RA_NID_PAGES;
2964 nm_i->dirty_nats_ratio = DEF_DIRTY_NAT_RATIO_THRESHOLD;
2966 INIT_RADIX_TREE(&nm_i->free_nid_root, GFP_ATOMIC);
2967 INIT_LIST_HEAD(&nm_i->free_nid_list);
2968 INIT_RADIX_TREE(&nm_i->nat_root, GFP_NOIO);
2969 INIT_RADIX_TREE(&nm_i->nat_set_root, GFP_NOIO);
2970 INIT_LIST_HEAD(&nm_i->nat_entries);
2971 spin_lock_init(&nm_i->nat_list_lock);
2973 mutex_init(&nm_i->build_lock);
2974 spin_lock_init(&nm_i->nid_list_lock);
2975 init_rwsem(&nm_i->nat_tree_lock);
2977 nm_i->next_scan_nid = le32_to_cpu(sbi->ckpt->next_free_nid);
2978 nm_i->bitmap_size = __bitmap_size(sbi, NAT_BITMAP);
2979 version_bitmap = __bitmap_ptr(sbi, NAT_BITMAP);
2980 if (!version_bitmap)
2983 nm_i->nat_bitmap = kmemdup(version_bitmap, nm_i->bitmap_size,
2985 if (!nm_i->nat_bitmap)
2988 err = __get_nat_bitmaps(sbi);
2992 #ifdef CONFIG_F2FS_CHECK_FS
2993 nm_i->nat_bitmap_mir = kmemdup(version_bitmap, nm_i->bitmap_size,
2995 if (!nm_i->nat_bitmap_mir)
3002 static int init_free_nid_cache(struct f2fs_sb_info *sbi)
3004 struct f2fs_nm_info *nm_i = NM_I(sbi);
3007 nm_i->free_nid_bitmap =
3008 f2fs_kzalloc(sbi, array_size(sizeof(unsigned char *),
3011 if (!nm_i->free_nid_bitmap)
3014 for (i = 0; i < nm_i->nat_blocks; i++) {
3015 nm_i->free_nid_bitmap[i] = f2fs_kvzalloc(sbi,
3016 f2fs_bitmap_size(NAT_ENTRY_PER_BLOCK), GFP_KERNEL);
3017 if (!nm_i->free_nid_bitmap[i])
3021 nm_i->nat_block_bitmap = f2fs_kvzalloc(sbi, nm_i->nat_blocks / 8,
3023 if (!nm_i->nat_block_bitmap)
3026 nm_i->free_nid_count =
3027 f2fs_kvzalloc(sbi, array_size(sizeof(unsigned short),
3030 if (!nm_i->free_nid_count)
3035 int f2fs_build_node_manager(struct f2fs_sb_info *sbi)
3039 sbi->nm_info = f2fs_kzalloc(sbi, sizeof(struct f2fs_nm_info),
3044 err = init_node_manager(sbi);
3048 err = init_free_nid_cache(sbi);
3052 /* load free nid status from nat_bits table */
3053 load_free_nid_bitmap(sbi);
3055 return f2fs_build_free_nids(sbi, true, true);
3058 void f2fs_destroy_node_manager(struct f2fs_sb_info *sbi)
3060 struct f2fs_nm_info *nm_i = NM_I(sbi);
3061 struct free_nid *i, *next_i;
3062 struct nat_entry *natvec[NATVEC_SIZE];
3063 struct nat_entry_set *setvec[SETVEC_SIZE];
3070 /* destroy free nid list */
3071 spin_lock(&nm_i->nid_list_lock);
3072 list_for_each_entry_safe(i, next_i, &nm_i->free_nid_list, list) {
3073 __remove_free_nid(sbi, i, FREE_NID);
3074 spin_unlock(&nm_i->nid_list_lock);
3075 kmem_cache_free(free_nid_slab, i);
3076 spin_lock(&nm_i->nid_list_lock);
3078 f2fs_bug_on(sbi, nm_i->nid_cnt[FREE_NID]);
3079 f2fs_bug_on(sbi, nm_i->nid_cnt[PREALLOC_NID]);
3080 f2fs_bug_on(sbi, !list_empty(&nm_i->free_nid_list));
3081 spin_unlock(&nm_i->nid_list_lock);
3083 /* destroy nat cache */
3084 down_write(&nm_i->nat_tree_lock);
3085 while ((found = __gang_lookup_nat_cache(nm_i,
3086 nid, NATVEC_SIZE, natvec))) {
3089 nid = nat_get_nid(natvec[found - 1]) + 1;
3090 for (idx = 0; idx < found; idx++) {
3091 spin_lock(&nm_i->nat_list_lock);
3092 list_del(&natvec[idx]->list);
3093 spin_unlock(&nm_i->nat_list_lock);
3095 __del_from_nat_cache(nm_i, natvec[idx]);
3098 f2fs_bug_on(sbi, nm_i->nat_cnt);
3100 /* destroy nat set cache */
3102 while ((found = __gang_lookup_nat_set(nm_i,
3103 nid, SETVEC_SIZE, setvec))) {
3106 nid = setvec[found - 1]->set + 1;
3107 for (idx = 0; idx < found; idx++) {
3108 /* entry_cnt is not zero, when cp_error was occurred */
3109 f2fs_bug_on(sbi, !list_empty(&setvec[idx]->entry_list));
3110 radix_tree_delete(&nm_i->nat_set_root, setvec[idx]->set);
3111 kmem_cache_free(nat_entry_set_slab, setvec[idx]);
3114 up_write(&nm_i->nat_tree_lock);
3116 kvfree(nm_i->nat_block_bitmap);
3117 if (nm_i->free_nid_bitmap) {
3120 for (i = 0; i < nm_i->nat_blocks; i++)
3121 kvfree(nm_i->free_nid_bitmap[i]);
3122 kfree(nm_i->free_nid_bitmap);
3124 kvfree(nm_i->free_nid_count);
3126 kfree(nm_i->nat_bitmap);
3127 kfree(nm_i->nat_bits);
3128 #ifdef CONFIG_F2FS_CHECK_FS
3129 kfree(nm_i->nat_bitmap_mir);
3131 sbi->nm_info = NULL;
3135 int __init f2fs_create_node_manager_caches(void)
3137 nat_entry_slab = f2fs_kmem_cache_create("nat_entry",
3138 sizeof(struct nat_entry));
3139 if (!nat_entry_slab)
3142 free_nid_slab = f2fs_kmem_cache_create("free_nid",
3143 sizeof(struct free_nid));
3145 goto destroy_nat_entry;
3147 nat_entry_set_slab = f2fs_kmem_cache_create("nat_entry_set",
3148 sizeof(struct nat_entry_set));
3149 if (!nat_entry_set_slab)
3150 goto destroy_free_nid;
3152 fsync_node_entry_slab = f2fs_kmem_cache_create("fsync_node_entry",
3153 sizeof(struct fsync_node_entry));
3154 if (!fsync_node_entry_slab)
3155 goto destroy_nat_entry_set;
3158 destroy_nat_entry_set:
3159 kmem_cache_destroy(nat_entry_set_slab);
3161 kmem_cache_destroy(free_nid_slab);
3163 kmem_cache_destroy(nat_entry_slab);
3168 void f2fs_destroy_node_manager_caches(void)
3170 kmem_cache_destroy(fsync_node_entry_slab);
3171 kmem_cache_destroy(nat_entry_set_slab);
3172 kmem_cache_destroy(free_nid_slab);
3173 kmem_cache_destroy(nat_entry_slab);