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>
23 #include <trace/events/f2fs.h>
25 #define on_build_free_nids(nmi) mutex_is_locked(&nm_i->build_lock)
27 static struct kmem_cache *nat_entry_slab;
28 static struct kmem_cache *free_nid_slab;
29 static struct kmem_cache *nat_entry_set_slab;
31 bool available_free_memory(struct f2fs_sb_info *sbi, int type)
33 struct f2fs_nm_info *nm_i = NM_I(sbi);
35 unsigned long avail_ram;
36 unsigned long mem_size = 0;
41 /* only uses low memory */
42 avail_ram = val.totalram - val.totalhigh;
45 * give 25%, 25%, 50%, 50%, 50% memory for each components respectively
47 if (type == FREE_NIDS) {
48 mem_size = (nm_i->fcnt * sizeof(struct free_nid)) >>
50 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 2);
51 } else if (type == NAT_ENTRIES) {
52 mem_size = (nm_i->nat_cnt * sizeof(struct nat_entry)) >>
54 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 2);
55 if (excess_cached_nats(sbi))
57 if (nm_i->nat_cnt > DEF_NAT_CACHE_THRESHOLD)
59 } else if (type == DIRTY_DENTS) {
60 if (sbi->sb->s_bdi->wb.dirty_exceeded)
62 mem_size = get_pages(sbi, F2FS_DIRTY_DENTS);
63 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 1);
64 } else if (type == INO_ENTRIES) {
67 for (i = 0; i <= UPDATE_INO; i++)
68 mem_size += (sbi->im[i].ino_num *
69 sizeof(struct ino_entry)) >> PAGE_SHIFT;
70 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 1);
71 } else if (type == EXTENT_CACHE) {
72 mem_size = (atomic_read(&sbi->total_ext_tree) *
73 sizeof(struct extent_tree) +
74 atomic_read(&sbi->total_ext_node) *
75 sizeof(struct extent_node)) >> PAGE_SHIFT;
76 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 1);
78 if (!sbi->sb->s_bdi->wb.dirty_exceeded)
84 static void clear_node_page_dirty(struct page *page)
86 struct address_space *mapping = page->mapping;
87 unsigned int long flags;
89 if (PageDirty(page)) {
90 spin_lock_irqsave(&mapping->tree_lock, flags);
91 radix_tree_tag_clear(&mapping->page_tree,
94 spin_unlock_irqrestore(&mapping->tree_lock, flags);
96 clear_page_dirty_for_io(page);
97 dec_page_count(F2FS_M_SB(mapping), F2FS_DIRTY_NODES);
99 ClearPageUptodate(page);
102 static struct page *get_current_nat_page(struct f2fs_sb_info *sbi, nid_t nid)
104 pgoff_t index = current_nat_addr(sbi, nid);
105 return get_meta_page(sbi, index);
108 static struct page *get_next_nat_page(struct f2fs_sb_info *sbi, nid_t nid)
110 struct page *src_page;
111 struct page *dst_page;
116 struct f2fs_nm_info *nm_i = NM_I(sbi);
118 src_off = current_nat_addr(sbi, nid);
119 dst_off = next_nat_addr(sbi, src_off);
121 /* get current nat block page with lock */
122 src_page = get_meta_page(sbi, src_off);
123 dst_page = grab_meta_page(sbi, dst_off);
124 f2fs_bug_on(sbi, PageDirty(src_page));
126 src_addr = page_address(src_page);
127 dst_addr = page_address(dst_page);
128 memcpy(dst_addr, src_addr, PAGE_SIZE);
129 set_page_dirty(dst_page);
130 f2fs_put_page(src_page, 1);
132 set_to_next_nat(nm_i, nid);
137 static struct nat_entry *__lookup_nat_cache(struct f2fs_nm_info *nm_i, nid_t n)
139 return radix_tree_lookup(&nm_i->nat_root, n);
142 static unsigned int __gang_lookup_nat_cache(struct f2fs_nm_info *nm_i,
143 nid_t start, unsigned int nr, struct nat_entry **ep)
145 return radix_tree_gang_lookup(&nm_i->nat_root, (void **)ep, start, nr);
148 static void __del_from_nat_cache(struct f2fs_nm_info *nm_i, struct nat_entry *e)
151 radix_tree_delete(&nm_i->nat_root, nat_get_nid(e));
153 kmem_cache_free(nat_entry_slab, e);
156 static void __set_nat_cache_dirty(struct f2fs_nm_info *nm_i,
157 struct nat_entry *ne)
159 nid_t set = NAT_BLOCK_OFFSET(ne->ni.nid);
160 struct nat_entry_set *head;
162 if (get_nat_flag(ne, IS_DIRTY))
165 head = radix_tree_lookup(&nm_i->nat_set_root, set);
167 head = f2fs_kmem_cache_alloc(nat_entry_set_slab, GFP_NOFS);
169 INIT_LIST_HEAD(&head->entry_list);
170 INIT_LIST_HEAD(&head->set_list);
173 f2fs_radix_tree_insert(&nm_i->nat_set_root, set, head);
175 list_move_tail(&ne->list, &head->entry_list);
176 nm_i->dirty_nat_cnt++;
178 set_nat_flag(ne, IS_DIRTY, true);
181 static void __clear_nat_cache_dirty(struct f2fs_nm_info *nm_i,
182 struct nat_entry *ne)
184 nid_t set = NAT_BLOCK_OFFSET(ne->ni.nid);
185 struct nat_entry_set *head;
187 head = radix_tree_lookup(&nm_i->nat_set_root, set);
189 list_move_tail(&ne->list, &nm_i->nat_entries);
190 set_nat_flag(ne, IS_DIRTY, false);
192 nm_i->dirty_nat_cnt--;
196 static unsigned int __gang_lookup_nat_set(struct f2fs_nm_info *nm_i,
197 nid_t start, unsigned int nr, struct nat_entry_set **ep)
199 return radix_tree_gang_lookup(&nm_i->nat_set_root, (void **)ep,
203 int need_dentry_mark(struct f2fs_sb_info *sbi, nid_t nid)
205 struct f2fs_nm_info *nm_i = NM_I(sbi);
209 down_read(&nm_i->nat_tree_lock);
210 e = __lookup_nat_cache(nm_i, nid);
212 if (!get_nat_flag(e, IS_CHECKPOINTED) &&
213 !get_nat_flag(e, HAS_FSYNCED_INODE))
216 up_read(&nm_i->nat_tree_lock);
220 bool is_checkpointed_node(struct f2fs_sb_info *sbi, nid_t nid)
222 struct f2fs_nm_info *nm_i = NM_I(sbi);
226 down_read(&nm_i->nat_tree_lock);
227 e = __lookup_nat_cache(nm_i, nid);
228 if (e && !get_nat_flag(e, IS_CHECKPOINTED))
230 up_read(&nm_i->nat_tree_lock);
234 bool need_inode_block_update(struct f2fs_sb_info *sbi, nid_t ino)
236 struct f2fs_nm_info *nm_i = NM_I(sbi);
238 bool need_update = true;
240 down_read(&nm_i->nat_tree_lock);
241 e = __lookup_nat_cache(nm_i, ino);
242 if (e && get_nat_flag(e, HAS_LAST_FSYNC) &&
243 (get_nat_flag(e, IS_CHECKPOINTED) ||
244 get_nat_flag(e, HAS_FSYNCED_INODE)))
246 up_read(&nm_i->nat_tree_lock);
250 static struct nat_entry *grab_nat_entry(struct f2fs_nm_info *nm_i, nid_t nid)
252 struct nat_entry *new;
254 new = f2fs_kmem_cache_alloc(nat_entry_slab, GFP_NOFS);
255 f2fs_radix_tree_insert(&nm_i->nat_root, nid, new);
256 memset(new, 0, sizeof(struct nat_entry));
257 nat_set_nid(new, nid);
259 list_add_tail(&new->list, &nm_i->nat_entries);
264 static void cache_nat_entry(struct f2fs_sb_info *sbi, nid_t nid,
265 struct f2fs_nat_entry *ne)
267 struct f2fs_nm_info *nm_i = NM_I(sbi);
270 e = __lookup_nat_cache(nm_i, nid);
272 e = grab_nat_entry(nm_i, nid);
273 node_info_from_raw_nat(&e->ni, ne);
275 f2fs_bug_on(sbi, nat_get_ino(e) != ne->ino ||
276 nat_get_blkaddr(e) != ne->block_addr ||
277 nat_get_version(e) != ne->version);
281 static void set_node_addr(struct f2fs_sb_info *sbi, struct node_info *ni,
282 block_t new_blkaddr, bool fsync_done)
284 struct f2fs_nm_info *nm_i = NM_I(sbi);
287 down_write(&nm_i->nat_tree_lock);
288 e = __lookup_nat_cache(nm_i, ni->nid);
290 e = grab_nat_entry(nm_i, ni->nid);
291 copy_node_info(&e->ni, ni);
292 f2fs_bug_on(sbi, ni->blk_addr == NEW_ADDR);
293 } else if (new_blkaddr == NEW_ADDR) {
295 * when nid is reallocated,
296 * previous nat entry can be remained in nat cache.
297 * So, reinitialize it with new information.
299 copy_node_info(&e->ni, ni);
300 f2fs_bug_on(sbi, ni->blk_addr != NULL_ADDR);
304 f2fs_bug_on(sbi, nat_get_blkaddr(e) != ni->blk_addr);
305 f2fs_bug_on(sbi, nat_get_blkaddr(e) == NULL_ADDR &&
306 new_blkaddr == NULL_ADDR);
307 f2fs_bug_on(sbi, nat_get_blkaddr(e) == NEW_ADDR &&
308 new_blkaddr == NEW_ADDR);
309 f2fs_bug_on(sbi, nat_get_blkaddr(e) != NEW_ADDR &&
310 nat_get_blkaddr(e) != NULL_ADDR &&
311 new_blkaddr == NEW_ADDR);
313 /* increment version no as node is removed */
314 if (nat_get_blkaddr(e) != NEW_ADDR && new_blkaddr == NULL_ADDR) {
315 unsigned char version = nat_get_version(e);
316 nat_set_version(e, inc_node_version(version));
318 /* in order to reuse the nid */
319 if (nm_i->next_scan_nid > ni->nid)
320 nm_i->next_scan_nid = ni->nid;
324 nat_set_blkaddr(e, new_blkaddr);
325 if (new_blkaddr == NEW_ADDR || new_blkaddr == NULL_ADDR)
326 set_nat_flag(e, IS_CHECKPOINTED, false);
327 __set_nat_cache_dirty(nm_i, e);
329 /* update fsync_mark if its inode nat entry is still alive */
330 if (ni->nid != ni->ino)
331 e = __lookup_nat_cache(nm_i, ni->ino);
333 if (fsync_done && ni->nid == ni->ino)
334 set_nat_flag(e, HAS_FSYNCED_INODE, true);
335 set_nat_flag(e, HAS_LAST_FSYNC, fsync_done);
337 up_write(&nm_i->nat_tree_lock);
340 int try_to_free_nats(struct f2fs_sb_info *sbi, int nr_shrink)
342 struct f2fs_nm_info *nm_i = NM_I(sbi);
345 if (!down_write_trylock(&nm_i->nat_tree_lock))
348 while (nr_shrink && !list_empty(&nm_i->nat_entries)) {
349 struct nat_entry *ne;
350 ne = list_first_entry(&nm_i->nat_entries,
351 struct nat_entry, list);
352 __del_from_nat_cache(nm_i, ne);
355 up_write(&nm_i->nat_tree_lock);
356 return nr - nr_shrink;
360 * This function always returns success
362 void get_node_info(struct f2fs_sb_info *sbi, nid_t nid, struct node_info *ni)
364 struct f2fs_nm_info *nm_i = NM_I(sbi);
365 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
366 struct f2fs_journal *journal = curseg->journal;
367 nid_t start_nid = START_NID(nid);
368 struct f2fs_nat_block *nat_blk;
369 struct page *page = NULL;
370 struct f2fs_nat_entry ne;
376 /* Check nat cache */
377 down_read(&nm_i->nat_tree_lock);
378 e = __lookup_nat_cache(nm_i, nid);
380 ni->ino = nat_get_ino(e);
381 ni->blk_addr = nat_get_blkaddr(e);
382 ni->version = nat_get_version(e);
383 up_read(&nm_i->nat_tree_lock);
387 memset(&ne, 0, sizeof(struct f2fs_nat_entry));
389 /* Check current segment summary */
390 down_read(&curseg->journal_rwsem);
391 i = lookup_journal_in_cursum(journal, NAT_JOURNAL, nid, 0);
393 ne = nat_in_journal(journal, i);
394 node_info_from_raw_nat(ni, &ne);
396 up_read(&curseg->journal_rwsem);
400 /* Fill node_info from nat page */
401 page = get_current_nat_page(sbi, start_nid);
402 nat_blk = (struct f2fs_nat_block *)page_address(page);
403 ne = nat_blk->entries[nid - start_nid];
404 node_info_from_raw_nat(ni, &ne);
405 f2fs_put_page(page, 1);
407 up_read(&nm_i->nat_tree_lock);
408 /* cache nat entry */
409 down_write(&nm_i->nat_tree_lock);
410 cache_nat_entry(sbi, nid, &ne);
411 up_write(&nm_i->nat_tree_lock);
415 * readahead MAX_RA_NODE number of node pages.
417 static void ra_node_pages(struct page *parent, int start, int n)
419 struct f2fs_sb_info *sbi = F2FS_P_SB(parent);
420 struct blk_plug plug;
424 blk_start_plug(&plug);
426 /* Then, try readahead for siblings of the desired node */
428 end = min(end, NIDS_PER_BLOCK);
429 for (i = start; i < end; i++) {
430 nid = get_nid(parent, i, false);
431 ra_node_page(sbi, nid);
434 blk_finish_plug(&plug);
437 pgoff_t get_next_page_offset(struct dnode_of_data *dn, pgoff_t pgofs)
439 const long direct_index = ADDRS_PER_INODE(dn->inode);
440 const long direct_blks = ADDRS_PER_BLOCK;
441 const long indirect_blks = ADDRS_PER_BLOCK * NIDS_PER_BLOCK;
442 unsigned int skipped_unit = ADDRS_PER_BLOCK;
443 int cur_level = dn->cur_level;
444 int max_level = dn->max_level;
450 while (max_level-- > cur_level)
451 skipped_unit *= NIDS_PER_BLOCK;
453 switch (dn->max_level) {
455 base += 2 * indirect_blks;
457 base += 2 * direct_blks;
459 base += direct_index;
462 f2fs_bug_on(F2FS_I_SB(dn->inode), 1);
465 return ((pgofs - base) / skipped_unit + 1) * skipped_unit + base;
469 * The maximum depth is four.
470 * Offset[0] will have raw inode offset.
472 static int get_node_path(struct inode *inode, long block,
473 int offset[4], unsigned int noffset[4])
475 const long direct_index = ADDRS_PER_INODE(inode);
476 const long direct_blks = ADDRS_PER_BLOCK;
477 const long dptrs_per_blk = NIDS_PER_BLOCK;
478 const long indirect_blks = ADDRS_PER_BLOCK * NIDS_PER_BLOCK;
479 const long dindirect_blks = indirect_blks * NIDS_PER_BLOCK;
485 if (block < direct_index) {
489 block -= direct_index;
490 if (block < direct_blks) {
491 offset[n++] = NODE_DIR1_BLOCK;
497 block -= direct_blks;
498 if (block < direct_blks) {
499 offset[n++] = NODE_DIR2_BLOCK;
505 block -= direct_blks;
506 if (block < indirect_blks) {
507 offset[n++] = NODE_IND1_BLOCK;
509 offset[n++] = block / direct_blks;
510 noffset[n] = 4 + offset[n - 1];
511 offset[n] = block % direct_blks;
515 block -= indirect_blks;
516 if (block < indirect_blks) {
517 offset[n++] = NODE_IND2_BLOCK;
518 noffset[n] = 4 + dptrs_per_blk;
519 offset[n++] = block / direct_blks;
520 noffset[n] = 5 + dptrs_per_blk + offset[n - 1];
521 offset[n] = block % direct_blks;
525 block -= indirect_blks;
526 if (block < dindirect_blks) {
527 offset[n++] = NODE_DIND_BLOCK;
528 noffset[n] = 5 + (dptrs_per_blk * 2);
529 offset[n++] = block / indirect_blks;
530 noffset[n] = 6 + (dptrs_per_blk * 2) +
531 offset[n - 1] * (dptrs_per_blk + 1);
532 offset[n++] = (block / direct_blks) % dptrs_per_blk;
533 noffset[n] = 7 + (dptrs_per_blk * 2) +
534 offset[n - 2] * (dptrs_per_blk + 1) +
536 offset[n] = block % direct_blks;
547 * Caller should call f2fs_put_dnode(dn).
548 * Also, it should grab and release a rwsem by calling f2fs_lock_op() and
549 * f2fs_unlock_op() only if ro is not set RDONLY_NODE.
550 * In the case of RDONLY_NODE, we don't need to care about mutex.
552 int get_dnode_of_data(struct dnode_of_data *dn, pgoff_t index, int mode)
554 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
555 struct page *npage[4];
556 struct page *parent = NULL;
558 unsigned int noffset[4];
563 level = get_node_path(dn->inode, index, offset, noffset);
565 nids[0] = dn->inode->i_ino;
566 npage[0] = dn->inode_page;
569 npage[0] = get_node_page(sbi, nids[0]);
570 if (IS_ERR(npage[0]))
571 return PTR_ERR(npage[0]);
574 /* if inline_data is set, should not report any block indices */
575 if (f2fs_has_inline_data(dn->inode) && index) {
577 f2fs_put_page(npage[0], 1);
583 nids[1] = get_nid(parent, offset[0], true);
584 dn->inode_page = npage[0];
585 dn->inode_page_locked = true;
587 /* get indirect or direct nodes */
588 for (i = 1; i <= level; i++) {
591 if (!nids[i] && mode == ALLOC_NODE) {
593 if (!alloc_nid(sbi, &(nids[i]))) {
599 npage[i] = new_node_page(dn, noffset[i], NULL);
600 if (IS_ERR(npage[i])) {
601 alloc_nid_failed(sbi, nids[i]);
602 err = PTR_ERR(npage[i]);
606 set_nid(parent, offset[i - 1], nids[i], i == 1);
607 alloc_nid_done(sbi, nids[i]);
609 } else if (mode == LOOKUP_NODE_RA && i == level && level > 1) {
610 npage[i] = get_node_page_ra(parent, offset[i - 1]);
611 if (IS_ERR(npage[i])) {
612 err = PTR_ERR(npage[i]);
618 dn->inode_page_locked = false;
621 f2fs_put_page(parent, 1);
625 npage[i] = get_node_page(sbi, nids[i]);
626 if (IS_ERR(npage[i])) {
627 err = PTR_ERR(npage[i]);
628 f2fs_put_page(npage[0], 0);
634 nids[i + 1] = get_nid(parent, offset[i], false);
637 dn->nid = nids[level];
638 dn->ofs_in_node = offset[level];
639 dn->node_page = npage[level];
640 dn->data_blkaddr = datablock_addr(dn->node_page, dn->ofs_in_node);
644 f2fs_put_page(parent, 1);
646 f2fs_put_page(npage[0], 0);
648 dn->inode_page = NULL;
649 dn->node_page = NULL;
650 if (err == -ENOENT) {
652 dn->max_level = level;
653 dn->ofs_in_node = offset[level];
658 static void truncate_node(struct dnode_of_data *dn)
660 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
663 get_node_info(sbi, dn->nid, &ni);
664 if (dn->inode->i_blocks == 0) {
665 f2fs_bug_on(sbi, ni.blk_addr != NULL_ADDR);
668 f2fs_bug_on(sbi, ni.blk_addr == NULL_ADDR);
670 /* Deallocate node address */
671 invalidate_blocks(sbi, ni.blk_addr);
672 dec_valid_node_count(sbi, dn->inode);
673 set_node_addr(sbi, &ni, NULL_ADDR, false);
675 if (dn->nid == dn->inode->i_ino) {
676 remove_orphan_inode(sbi, dn->nid);
677 dec_valid_inode_count(sbi);
678 f2fs_inode_synced(dn->inode);
681 clear_node_page_dirty(dn->node_page);
682 set_sbi_flag(sbi, SBI_IS_DIRTY);
684 f2fs_put_page(dn->node_page, 1);
686 invalidate_mapping_pages(NODE_MAPPING(sbi),
687 dn->node_page->index, dn->node_page->index);
689 dn->node_page = NULL;
690 trace_f2fs_truncate_node(dn->inode, dn->nid, ni.blk_addr);
693 static int truncate_dnode(struct dnode_of_data *dn)
700 /* get direct node */
701 page = get_node_page(F2FS_I_SB(dn->inode), dn->nid);
702 if (IS_ERR(page) && PTR_ERR(page) == -ENOENT)
704 else if (IS_ERR(page))
705 return PTR_ERR(page);
707 /* Make dnode_of_data for parameter */
708 dn->node_page = page;
710 truncate_data_blocks(dn);
715 static int truncate_nodes(struct dnode_of_data *dn, unsigned int nofs,
718 struct dnode_of_data rdn = *dn;
720 struct f2fs_node *rn;
722 unsigned int child_nofs;
727 return NIDS_PER_BLOCK + 1;
729 trace_f2fs_truncate_nodes_enter(dn->inode, dn->nid, dn->data_blkaddr);
731 page = get_node_page(F2FS_I_SB(dn->inode), dn->nid);
733 trace_f2fs_truncate_nodes_exit(dn->inode, PTR_ERR(page));
734 return PTR_ERR(page);
737 ra_node_pages(page, ofs, NIDS_PER_BLOCK);
739 rn = F2FS_NODE(page);
741 for (i = ofs; i < NIDS_PER_BLOCK; i++, freed++) {
742 child_nid = le32_to_cpu(rn->in.nid[i]);
746 ret = truncate_dnode(&rdn);
749 if (set_nid(page, i, 0, false))
750 dn->node_changed = true;
753 child_nofs = nofs + ofs * (NIDS_PER_BLOCK + 1) + 1;
754 for (i = ofs; i < NIDS_PER_BLOCK; i++) {
755 child_nid = le32_to_cpu(rn->in.nid[i]);
756 if (child_nid == 0) {
757 child_nofs += NIDS_PER_BLOCK + 1;
761 ret = truncate_nodes(&rdn, child_nofs, 0, depth - 1);
762 if (ret == (NIDS_PER_BLOCK + 1)) {
763 if (set_nid(page, i, 0, false))
764 dn->node_changed = true;
766 } else if (ret < 0 && ret != -ENOENT) {
774 /* remove current indirect node */
775 dn->node_page = page;
779 f2fs_put_page(page, 1);
781 trace_f2fs_truncate_nodes_exit(dn->inode, freed);
785 f2fs_put_page(page, 1);
786 trace_f2fs_truncate_nodes_exit(dn->inode, ret);
790 static int truncate_partial_nodes(struct dnode_of_data *dn,
791 struct f2fs_inode *ri, int *offset, int depth)
793 struct page *pages[2];
800 nid[0] = le32_to_cpu(ri->i_nid[offset[0] - NODE_DIR1_BLOCK]);
804 /* get indirect nodes in the path */
805 for (i = 0; i < idx + 1; i++) {
806 /* reference count'll be increased */
807 pages[i] = get_node_page(F2FS_I_SB(dn->inode), nid[i]);
808 if (IS_ERR(pages[i])) {
809 err = PTR_ERR(pages[i]);
813 nid[i + 1] = get_nid(pages[i], offset[i + 1], false);
816 ra_node_pages(pages[idx], offset[idx + 1], NIDS_PER_BLOCK);
818 /* free direct nodes linked to a partial indirect node */
819 for (i = offset[idx + 1]; i < NIDS_PER_BLOCK; i++) {
820 child_nid = get_nid(pages[idx], i, false);
824 err = truncate_dnode(dn);
827 if (set_nid(pages[idx], i, 0, false))
828 dn->node_changed = true;
831 if (offset[idx + 1] == 0) {
832 dn->node_page = pages[idx];
836 f2fs_put_page(pages[idx], 1);
842 for (i = idx; i >= 0; i--)
843 f2fs_put_page(pages[i], 1);
845 trace_f2fs_truncate_partial_nodes(dn->inode, nid, depth, err);
851 * All the block addresses of data and nodes should be nullified.
853 int truncate_inode_blocks(struct inode *inode, pgoff_t from)
855 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
856 int err = 0, cont = 1;
857 int level, offset[4], noffset[4];
858 unsigned int nofs = 0;
859 struct f2fs_inode *ri;
860 struct dnode_of_data dn;
863 trace_f2fs_truncate_inode_blocks_enter(inode, from);
865 level = get_node_path(inode, from, offset, noffset);
867 page = get_node_page(sbi, inode->i_ino);
869 trace_f2fs_truncate_inode_blocks_exit(inode, PTR_ERR(page));
870 return PTR_ERR(page);
873 set_new_dnode(&dn, inode, page, NULL, 0);
876 ri = F2FS_INODE(page);
884 if (!offset[level - 1])
886 err = truncate_partial_nodes(&dn, ri, offset, level);
887 if (err < 0 && err != -ENOENT)
889 nofs += 1 + NIDS_PER_BLOCK;
892 nofs = 5 + 2 * NIDS_PER_BLOCK;
893 if (!offset[level - 1])
895 err = truncate_partial_nodes(&dn, ri, offset, level);
896 if (err < 0 && err != -ENOENT)
905 dn.nid = le32_to_cpu(ri->i_nid[offset[0] - NODE_DIR1_BLOCK]);
907 case NODE_DIR1_BLOCK:
908 case NODE_DIR2_BLOCK:
909 err = truncate_dnode(&dn);
912 case NODE_IND1_BLOCK:
913 case NODE_IND2_BLOCK:
914 err = truncate_nodes(&dn, nofs, offset[1], 2);
917 case NODE_DIND_BLOCK:
918 err = truncate_nodes(&dn, nofs, offset[1], 3);
925 if (err < 0 && err != -ENOENT)
927 if (offset[1] == 0 &&
928 ri->i_nid[offset[0] - NODE_DIR1_BLOCK]) {
930 BUG_ON(page->mapping != NODE_MAPPING(sbi));
931 f2fs_wait_on_page_writeback(page, NODE, true);
932 ri->i_nid[offset[0] - NODE_DIR1_BLOCK] = 0;
933 set_page_dirty(page);
941 f2fs_put_page(page, 0);
942 trace_f2fs_truncate_inode_blocks_exit(inode, err);
943 return err > 0 ? 0 : err;
946 int truncate_xattr_node(struct inode *inode, struct page *page)
948 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
949 nid_t nid = F2FS_I(inode)->i_xattr_nid;
950 struct dnode_of_data dn;
956 npage = get_node_page(sbi, nid);
958 return PTR_ERR(npage);
960 f2fs_i_xnid_write(inode, 0);
962 /* need to do checkpoint during fsync */
963 F2FS_I(inode)->xattr_ver = cur_cp_version(F2FS_CKPT(sbi));
965 set_new_dnode(&dn, inode, page, npage, nid);
968 dn.inode_page_locked = true;
974 * Caller should grab and release a rwsem by calling f2fs_lock_op() and
977 int remove_inode_page(struct inode *inode)
979 struct dnode_of_data dn;
982 set_new_dnode(&dn, inode, NULL, NULL, inode->i_ino);
983 err = get_dnode_of_data(&dn, 0, LOOKUP_NODE);
987 err = truncate_xattr_node(inode, dn.inode_page);
993 /* remove potential inline_data blocks */
994 if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
995 S_ISLNK(inode->i_mode))
996 truncate_data_blocks_range(&dn, 1);
998 /* 0 is possible, after f2fs_new_inode() has failed */
999 f2fs_bug_on(F2FS_I_SB(inode),
1000 inode->i_blocks != 0 && inode->i_blocks != 1);
1002 /* will put inode & node pages */
1007 struct page *new_inode_page(struct inode *inode)
1009 struct dnode_of_data dn;
1011 /* allocate inode page for new inode */
1012 set_new_dnode(&dn, inode, NULL, NULL, inode->i_ino);
1014 /* caller should f2fs_put_page(page, 1); */
1015 return new_node_page(&dn, 0, NULL);
1018 struct page *new_node_page(struct dnode_of_data *dn,
1019 unsigned int ofs, struct page *ipage)
1021 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
1022 struct node_info old_ni, new_ni;
1026 if (unlikely(is_inode_flag_set(dn->inode, FI_NO_ALLOC)))
1027 return ERR_PTR(-EPERM);
1029 page = f2fs_grab_cache_page(NODE_MAPPING(sbi), dn->nid, false);
1031 return ERR_PTR(-ENOMEM);
1033 if (unlikely(!inc_valid_node_count(sbi, dn->inode))) {
1038 get_node_info(sbi, dn->nid, &old_ni);
1040 /* Reinitialize old_ni with new node page */
1041 f2fs_bug_on(sbi, old_ni.blk_addr != NULL_ADDR);
1043 new_ni.ino = dn->inode->i_ino;
1044 set_node_addr(sbi, &new_ni, NEW_ADDR, false);
1046 f2fs_wait_on_page_writeback(page, NODE, true);
1047 fill_node_footer(page, dn->nid, dn->inode->i_ino, ofs, true);
1048 set_cold_node(dn->inode, page);
1049 if (!PageUptodate(page))
1050 SetPageUptodate(page);
1051 if (set_page_dirty(page))
1052 dn->node_changed = true;
1054 if (f2fs_has_xattr_block(ofs))
1055 f2fs_i_xnid_write(dn->inode, dn->nid);
1058 inc_valid_inode_count(sbi);
1062 clear_node_page_dirty(page);
1063 f2fs_put_page(page, 1);
1064 return ERR_PTR(err);
1068 * Caller should do after getting the following values.
1069 * 0: f2fs_put_page(page, 0)
1070 * LOCKED_PAGE or error: f2fs_put_page(page, 1)
1072 static int read_node_page(struct page *page, int op_flags)
1074 struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1075 struct node_info ni;
1076 struct f2fs_io_info fio = {
1080 .op_flags = op_flags,
1082 .encrypted_page = NULL,
1085 if (PageUptodate(page))
1088 get_node_info(sbi, page->index, &ni);
1090 if (unlikely(ni.blk_addr == NULL_ADDR)) {
1091 ClearPageUptodate(page);
1095 fio.new_blkaddr = fio.old_blkaddr = ni.blk_addr;
1096 return f2fs_submit_page_bio(&fio);
1100 * Readahead a node page
1102 void ra_node_page(struct f2fs_sb_info *sbi, nid_t nid)
1109 f2fs_bug_on(sbi, check_nid_range(sbi, nid));
1112 apage = radix_tree_lookup(&NODE_MAPPING(sbi)->page_tree, nid);
1117 apage = f2fs_grab_cache_page(NODE_MAPPING(sbi), nid, false);
1121 err = read_node_page(apage, REQ_RAHEAD);
1122 f2fs_put_page(apage, err ? 1 : 0);
1125 static struct page *__get_node_page(struct f2fs_sb_info *sbi, pgoff_t nid,
1126 struct page *parent, int start)
1132 return ERR_PTR(-ENOENT);
1133 f2fs_bug_on(sbi, check_nid_range(sbi, nid));
1135 page = f2fs_grab_cache_page(NODE_MAPPING(sbi), nid, false);
1137 return ERR_PTR(-ENOMEM);
1139 err = read_node_page(page, READ_SYNC);
1141 f2fs_put_page(page, 1);
1142 return ERR_PTR(err);
1143 } else if (err == LOCKED_PAGE) {
1148 ra_node_pages(parent, start + 1, MAX_RA_NODE);
1152 if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1153 f2fs_put_page(page, 1);
1157 if (unlikely(!PageUptodate(page)))
1160 if(unlikely(nid != nid_of_node(page))) {
1161 f2fs_bug_on(sbi, 1);
1162 ClearPageUptodate(page);
1164 f2fs_put_page(page, 1);
1165 return ERR_PTR(-EIO);
1170 struct page *get_node_page(struct f2fs_sb_info *sbi, pgoff_t nid)
1172 return __get_node_page(sbi, nid, NULL, 0);
1175 struct page *get_node_page_ra(struct page *parent, int start)
1177 struct f2fs_sb_info *sbi = F2FS_P_SB(parent);
1178 nid_t nid = get_nid(parent, start, false);
1180 return __get_node_page(sbi, nid, parent, start);
1183 static void flush_inline_data(struct f2fs_sb_info *sbi, nid_t ino)
1185 struct inode *inode;
1189 /* should flush inline_data before evict_inode */
1190 inode = ilookup(sbi->sb, ino);
1194 page = pagecache_get_page(inode->i_mapping, 0, FGP_LOCK|FGP_NOWAIT, 0);
1198 if (!PageUptodate(page))
1201 if (!PageDirty(page))
1204 if (!clear_page_dirty_for_io(page))
1207 ret = f2fs_write_inline_data(inode, page);
1208 inode_dec_dirty_pages(inode);
1210 set_page_dirty(page);
1212 f2fs_put_page(page, 1);
1217 void move_node_page(struct page *node_page, int gc_type)
1219 if (gc_type == FG_GC) {
1220 struct f2fs_sb_info *sbi = F2FS_P_SB(node_page);
1221 struct writeback_control wbc = {
1222 .sync_mode = WB_SYNC_ALL,
1227 set_page_dirty(node_page);
1228 f2fs_wait_on_page_writeback(node_page, NODE, true);
1230 f2fs_bug_on(sbi, PageWriteback(node_page));
1231 if (!clear_page_dirty_for_io(node_page))
1234 if (NODE_MAPPING(sbi)->a_ops->writepage(node_page, &wbc))
1235 unlock_page(node_page);
1238 /* set page dirty and write it */
1239 if (!PageWriteback(node_page))
1240 set_page_dirty(node_page);
1243 unlock_page(node_page);
1245 f2fs_put_page(node_page, 0);
1248 static struct page *last_fsync_dnode(struct f2fs_sb_info *sbi, nid_t ino)
1251 struct pagevec pvec;
1252 struct page *last_page = NULL;
1254 pagevec_init(&pvec, 0);
1258 while (index <= end) {
1260 nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1261 PAGECACHE_TAG_DIRTY,
1262 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
1266 for (i = 0; i < nr_pages; i++) {
1267 struct page *page = pvec.pages[i];
1269 if (unlikely(f2fs_cp_error(sbi))) {
1270 f2fs_put_page(last_page, 0);
1271 pagevec_release(&pvec);
1272 return ERR_PTR(-EIO);
1275 if (!IS_DNODE(page) || !is_cold_node(page))
1277 if (ino_of_node(page) != ino)
1282 if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1287 if (ino_of_node(page) != ino)
1288 goto continue_unlock;
1290 if (!PageDirty(page)) {
1291 /* someone wrote it for us */
1292 goto continue_unlock;
1296 f2fs_put_page(last_page, 0);
1302 pagevec_release(&pvec);
1308 int fsync_node_pages(struct f2fs_sb_info *sbi, struct inode *inode,
1309 struct writeback_control *wbc, bool atomic)
1312 struct pagevec pvec;
1314 struct page *last_page = NULL;
1315 bool marked = false;
1316 nid_t ino = inode->i_ino;
1319 last_page = last_fsync_dnode(sbi, ino);
1320 if (IS_ERR_OR_NULL(last_page))
1321 return PTR_ERR_OR_ZERO(last_page);
1324 pagevec_init(&pvec, 0);
1328 while (index <= end) {
1330 nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1331 PAGECACHE_TAG_DIRTY,
1332 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
1336 for (i = 0; i < nr_pages; i++) {
1337 struct page *page = pvec.pages[i];
1339 if (unlikely(f2fs_cp_error(sbi))) {
1340 f2fs_put_page(last_page, 0);
1341 pagevec_release(&pvec);
1345 if (!IS_DNODE(page) || !is_cold_node(page))
1347 if (ino_of_node(page) != ino)
1352 if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1357 if (ino_of_node(page) != ino)
1358 goto continue_unlock;
1360 if (!PageDirty(page) && page != last_page) {
1361 /* someone wrote it for us */
1362 goto continue_unlock;
1365 f2fs_wait_on_page_writeback(page, NODE, true);
1366 BUG_ON(PageWriteback(page));
1368 if (!atomic || page == last_page) {
1369 set_fsync_mark(page, 1);
1370 if (IS_INODE(page)) {
1371 if (is_inode_flag_set(inode,
1373 update_inode(inode, page);
1374 set_dentry_mark(page,
1375 need_dentry_mark(sbi, ino));
1377 /* may be written by other thread */
1378 if (!PageDirty(page))
1379 set_page_dirty(page);
1382 if (!clear_page_dirty_for_io(page))
1383 goto continue_unlock;
1385 ret = NODE_MAPPING(sbi)->a_ops->writepage(page, wbc);
1388 f2fs_put_page(last_page, 0);
1391 if (page == last_page) {
1392 f2fs_put_page(page, 0);
1397 pagevec_release(&pvec);
1403 if (!ret && atomic && !marked) {
1404 f2fs_msg(sbi->sb, KERN_DEBUG,
1405 "Retry to write fsync mark: ino=%u, idx=%lx",
1406 ino, last_page->index);
1407 lock_page(last_page);
1408 set_page_dirty(last_page);
1409 unlock_page(last_page);
1412 return ret ? -EIO: 0;
1415 int sync_node_pages(struct f2fs_sb_info *sbi, struct writeback_control *wbc)
1418 struct pagevec pvec;
1422 pagevec_init(&pvec, 0);
1428 while (index <= end) {
1430 nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1431 PAGECACHE_TAG_DIRTY,
1432 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
1436 for (i = 0; i < nr_pages; i++) {
1437 struct page *page = pvec.pages[i];
1439 if (unlikely(f2fs_cp_error(sbi))) {
1440 pagevec_release(&pvec);
1445 * flushing sequence with step:
1450 if (step == 0 && IS_DNODE(page))
1452 if (step == 1 && (!IS_DNODE(page) ||
1453 is_cold_node(page)))
1455 if (step == 2 && (!IS_DNODE(page) ||
1456 !is_cold_node(page)))
1459 if (!trylock_page(page))
1462 if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1468 if (!PageDirty(page)) {
1469 /* someone wrote it for us */
1470 goto continue_unlock;
1473 /* flush inline_data */
1474 if (is_inline_node(page)) {
1475 clear_inline_node(page);
1477 flush_inline_data(sbi, ino_of_node(page));
1481 f2fs_wait_on_page_writeback(page, NODE, true);
1483 BUG_ON(PageWriteback(page));
1484 if (!clear_page_dirty_for_io(page))
1485 goto continue_unlock;
1487 set_fsync_mark(page, 0);
1488 set_dentry_mark(page, 0);
1490 if (NODE_MAPPING(sbi)->a_ops->writepage(page, wbc))
1493 if (--wbc->nr_to_write == 0)
1496 pagevec_release(&pvec);
1499 if (wbc->nr_to_write == 0) {
1512 int wait_on_node_pages_writeback(struct f2fs_sb_info *sbi, nid_t ino)
1514 pgoff_t index = 0, end = ULONG_MAX;
1515 struct pagevec pvec;
1518 pagevec_init(&pvec, 0);
1520 while (index <= end) {
1522 nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1523 PAGECACHE_TAG_WRITEBACK,
1524 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
1528 for (i = 0; i < nr_pages; i++) {
1529 struct page *page = pvec.pages[i];
1531 /* until radix tree lookup accepts end_index */
1532 if (unlikely(page->index > end))
1535 if (ino && ino_of_node(page) == ino) {
1536 f2fs_wait_on_page_writeback(page, NODE, true);
1537 if (TestClearPageError(page))
1541 pagevec_release(&pvec);
1545 ret2 = filemap_check_errors(NODE_MAPPING(sbi));
1551 static int f2fs_write_node_page(struct page *page,
1552 struct writeback_control *wbc)
1554 struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1556 struct node_info ni;
1557 struct f2fs_io_info fio = {
1561 .op_flags = (wbc->sync_mode == WB_SYNC_ALL) ? WRITE_SYNC : 0,
1563 .encrypted_page = NULL,
1566 trace_f2fs_writepage(page, NODE);
1568 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
1570 if (unlikely(f2fs_cp_error(sbi)))
1573 /* get old block addr of this node page */
1574 nid = nid_of_node(page);
1575 f2fs_bug_on(sbi, page->index != nid);
1577 if (wbc->for_reclaim) {
1578 if (!down_read_trylock(&sbi->node_write))
1581 down_read(&sbi->node_write);
1584 get_node_info(sbi, nid, &ni);
1586 /* This page is already truncated */
1587 if (unlikely(ni.blk_addr == NULL_ADDR)) {
1588 ClearPageUptodate(page);
1589 dec_page_count(sbi, F2FS_DIRTY_NODES);
1590 up_read(&sbi->node_write);
1595 set_page_writeback(page);
1596 fio.old_blkaddr = ni.blk_addr;
1597 write_node_page(nid, &fio);
1598 set_node_addr(sbi, &ni, fio.new_blkaddr, is_fsync_dnode(page));
1599 dec_page_count(sbi, F2FS_DIRTY_NODES);
1600 up_read(&sbi->node_write);
1602 if (wbc->for_reclaim)
1603 f2fs_submit_merged_bio_cond(sbi, NULL, page, 0, NODE, WRITE);
1607 if (unlikely(f2fs_cp_error(sbi)))
1608 f2fs_submit_merged_bio(sbi, NODE, WRITE);
1613 redirty_page_for_writepage(wbc, page);
1614 return AOP_WRITEPAGE_ACTIVATE;
1617 static int f2fs_write_node_pages(struct address_space *mapping,
1618 struct writeback_control *wbc)
1620 struct f2fs_sb_info *sbi = F2FS_M_SB(mapping);
1621 struct blk_plug plug;
1624 /* balancing f2fs's metadata in background */
1625 f2fs_balance_fs_bg(sbi);
1627 /* collect a number of dirty node pages and write together */
1628 if (get_pages(sbi, F2FS_DIRTY_NODES) < nr_pages_to_skip(sbi, NODE))
1631 trace_f2fs_writepages(mapping->host, wbc, NODE);
1633 diff = nr_pages_to_write(sbi, NODE, wbc);
1634 wbc->sync_mode = WB_SYNC_NONE;
1635 blk_start_plug(&plug);
1636 sync_node_pages(sbi, wbc);
1637 blk_finish_plug(&plug);
1638 wbc->nr_to_write = max((long)0, wbc->nr_to_write - diff);
1642 wbc->pages_skipped += get_pages(sbi, F2FS_DIRTY_NODES);
1643 trace_f2fs_writepages(mapping->host, wbc, NODE);
1647 static int f2fs_set_node_page_dirty(struct page *page)
1649 trace_f2fs_set_page_dirty(page, NODE);
1651 if (!PageUptodate(page))
1652 SetPageUptodate(page);
1653 if (!PageDirty(page)) {
1654 f2fs_set_page_dirty_nobuffers(page);
1655 inc_page_count(F2FS_P_SB(page), F2FS_DIRTY_NODES);
1656 SetPagePrivate(page);
1657 f2fs_trace_pid(page);
1664 * Structure of the f2fs node operations
1666 const struct address_space_operations f2fs_node_aops = {
1667 .writepage = f2fs_write_node_page,
1668 .writepages = f2fs_write_node_pages,
1669 .set_page_dirty = f2fs_set_node_page_dirty,
1670 .invalidatepage = f2fs_invalidate_page,
1671 .releasepage = f2fs_release_page,
1674 static struct free_nid *__lookup_free_nid_list(struct f2fs_nm_info *nm_i,
1677 return radix_tree_lookup(&nm_i->free_nid_root, n);
1680 static void __del_from_free_nid_list(struct f2fs_nm_info *nm_i,
1684 radix_tree_delete(&nm_i->free_nid_root, i->nid);
1687 static int add_free_nid(struct f2fs_sb_info *sbi, nid_t nid, bool build)
1689 struct f2fs_nm_info *nm_i = NM_I(sbi);
1691 struct nat_entry *ne;
1693 if (!available_free_memory(sbi, FREE_NIDS))
1696 /* 0 nid should not be used */
1697 if (unlikely(nid == 0))
1701 /* do not add allocated nids */
1702 ne = __lookup_nat_cache(nm_i, nid);
1703 if (ne && (!get_nat_flag(ne, IS_CHECKPOINTED) ||
1704 nat_get_blkaddr(ne) != NULL_ADDR))
1708 i = f2fs_kmem_cache_alloc(free_nid_slab, GFP_NOFS);
1712 if (radix_tree_preload(GFP_NOFS)) {
1713 kmem_cache_free(free_nid_slab, i);
1717 spin_lock(&nm_i->free_nid_list_lock);
1718 if (radix_tree_insert(&nm_i->free_nid_root, i->nid, i)) {
1719 spin_unlock(&nm_i->free_nid_list_lock);
1720 radix_tree_preload_end();
1721 kmem_cache_free(free_nid_slab, i);
1724 list_add_tail(&i->list, &nm_i->free_nid_list);
1726 spin_unlock(&nm_i->free_nid_list_lock);
1727 radix_tree_preload_end();
1731 static void remove_free_nid(struct f2fs_nm_info *nm_i, nid_t nid)
1734 bool need_free = false;
1736 spin_lock(&nm_i->free_nid_list_lock);
1737 i = __lookup_free_nid_list(nm_i, nid);
1738 if (i && i->state == NID_NEW) {
1739 __del_from_free_nid_list(nm_i, i);
1743 spin_unlock(&nm_i->free_nid_list_lock);
1746 kmem_cache_free(free_nid_slab, i);
1749 static void scan_nat_page(struct f2fs_sb_info *sbi,
1750 struct page *nat_page, nid_t start_nid)
1752 struct f2fs_nm_info *nm_i = NM_I(sbi);
1753 struct f2fs_nat_block *nat_blk = page_address(nat_page);
1757 i = start_nid % NAT_ENTRY_PER_BLOCK;
1759 for (; i < NAT_ENTRY_PER_BLOCK; i++, start_nid++) {
1761 if (unlikely(start_nid >= nm_i->max_nid))
1764 blk_addr = le32_to_cpu(nat_blk->entries[i].block_addr);
1765 f2fs_bug_on(sbi, blk_addr == NEW_ADDR);
1766 if (blk_addr == NULL_ADDR) {
1767 if (add_free_nid(sbi, start_nid, true) < 0)
1773 void build_free_nids(struct f2fs_sb_info *sbi)
1775 struct f2fs_nm_info *nm_i = NM_I(sbi);
1776 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
1777 struct f2fs_journal *journal = curseg->journal;
1779 nid_t nid = nm_i->next_scan_nid;
1781 /* Enough entries */
1782 if (nm_i->fcnt >= NAT_ENTRY_PER_BLOCK)
1785 /* readahead nat pages to be scanned */
1786 ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nid), FREE_NID_PAGES,
1789 down_read(&nm_i->nat_tree_lock);
1792 struct page *page = get_current_nat_page(sbi, nid);
1794 scan_nat_page(sbi, page, nid);
1795 f2fs_put_page(page, 1);
1797 nid += (NAT_ENTRY_PER_BLOCK - (nid % NAT_ENTRY_PER_BLOCK));
1798 if (unlikely(nid >= nm_i->max_nid))
1801 if (++i >= FREE_NID_PAGES)
1805 /* go to the next free nat pages to find free nids abundantly */
1806 nm_i->next_scan_nid = nid;
1808 /* find free nids from current sum_pages */
1809 down_read(&curseg->journal_rwsem);
1810 for (i = 0; i < nats_in_cursum(journal); i++) {
1813 addr = le32_to_cpu(nat_in_journal(journal, i).block_addr);
1814 nid = le32_to_cpu(nid_in_journal(journal, i));
1815 if (addr == NULL_ADDR)
1816 add_free_nid(sbi, nid, true);
1818 remove_free_nid(nm_i, nid);
1820 up_read(&curseg->journal_rwsem);
1821 up_read(&nm_i->nat_tree_lock);
1823 ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nm_i->next_scan_nid),
1824 nm_i->ra_nid_pages, META_NAT, false);
1828 * If this function returns success, caller can obtain a new nid
1829 * from second parameter of this function.
1830 * The returned nid could be used ino as well as nid when inode is created.
1832 bool alloc_nid(struct f2fs_sb_info *sbi, nid_t *nid)
1834 struct f2fs_nm_info *nm_i = NM_I(sbi);
1835 struct free_nid *i = NULL;
1837 #ifdef CONFIG_F2FS_FAULT_INJECTION
1838 if (time_to_inject(FAULT_ALLOC_NID))
1841 if (unlikely(sbi->total_valid_node_count + 1 > nm_i->available_nids))
1844 spin_lock(&nm_i->free_nid_list_lock);
1846 /* We should not use stale free nids created by build_free_nids */
1847 if (nm_i->fcnt && !on_build_free_nids(nm_i)) {
1848 f2fs_bug_on(sbi, list_empty(&nm_i->free_nid_list));
1849 list_for_each_entry(i, &nm_i->free_nid_list, list)
1850 if (i->state == NID_NEW)
1853 f2fs_bug_on(sbi, i->state != NID_NEW);
1855 i->state = NID_ALLOC;
1857 spin_unlock(&nm_i->free_nid_list_lock);
1860 spin_unlock(&nm_i->free_nid_list_lock);
1862 /* Let's scan nat pages and its caches to get free nids */
1863 mutex_lock(&nm_i->build_lock);
1864 build_free_nids(sbi);
1865 mutex_unlock(&nm_i->build_lock);
1870 * alloc_nid() should be called prior to this function.
1872 void alloc_nid_done(struct f2fs_sb_info *sbi, nid_t nid)
1874 struct f2fs_nm_info *nm_i = NM_I(sbi);
1877 spin_lock(&nm_i->free_nid_list_lock);
1878 i = __lookup_free_nid_list(nm_i, nid);
1879 f2fs_bug_on(sbi, !i || i->state != NID_ALLOC);
1880 __del_from_free_nid_list(nm_i, i);
1881 spin_unlock(&nm_i->free_nid_list_lock);
1883 kmem_cache_free(free_nid_slab, i);
1887 * alloc_nid() should be called prior to this function.
1889 void alloc_nid_failed(struct f2fs_sb_info *sbi, nid_t nid)
1891 struct f2fs_nm_info *nm_i = NM_I(sbi);
1893 bool need_free = false;
1898 spin_lock(&nm_i->free_nid_list_lock);
1899 i = __lookup_free_nid_list(nm_i, nid);
1900 f2fs_bug_on(sbi, !i || i->state != NID_ALLOC);
1901 if (!available_free_memory(sbi, FREE_NIDS)) {
1902 __del_from_free_nid_list(nm_i, i);
1908 spin_unlock(&nm_i->free_nid_list_lock);
1911 kmem_cache_free(free_nid_slab, i);
1914 int try_to_free_nids(struct f2fs_sb_info *sbi, int nr_shrink)
1916 struct f2fs_nm_info *nm_i = NM_I(sbi);
1917 struct free_nid *i, *next;
1920 if (nm_i->fcnt <= MAX_FREE_NIDS)
1923 if (!mutex_trylock(&nm_i->build_lock))
1926 spin_lock(&nm_i->free_nid_list_lock);
1927 list_for_each_entry_safe(i, next, &nm_i->free_nid_list, list) {
1928 if (nr_shrink <= 0 || nm_i->fcnt <= MAX_FREE_NIDS)
1930 if (i->state == NID_ALLOC)
1932 __del_from_free_nid_list(nm_i, i);
1933 kmem_cache_free(free_nid_slab, i);
1937 spin_unlock(&nm_i->free_nid_list_lock);
1938 mutex_unlock(&nm_i->build_lock);
1940 return nr - nr_shrink;
1943 void recover_inline_xattr(struct inode *inode, struct page *page)
1945 void *src_addr, *dst_addr;
1948 struct f2fs_inode *ri;
1950 ipage = get_node_page(F2FS_I_SB(inode), inode->i_ino);
1951 f2fs_bug_on(F2FS_I_SB(inode), IS_ERR(ipage));
1953 ri = F2FS_INODE(page);
1954 if (!(ri->i_inline & F2FS_INLINE_XATTR)) {
1955 clear_inode_flag(inode, FI_INLINE_XATTR);
1959 dst_addr = inline_xattr_addr(ipage);
1960 src_addr = inline_xattr_addr(page);
1961 inline_size = inline_xattr_size(inode);
1963 f2fs_wait_on_page_writeback(ipage, NODE, true);
1964 memcpy(dst_addr, src_addr, inline_size);
1966 update_inode(inode, ipage);
1967 f2fs_put_page(ipage, 1);
1970 void recover_xattr_data(struct inode *inode, struct page *page, block_t blkaddr)
1972 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1973 nid_t prev_xnid = F2FS_I(inode)->i_xattr_nid;
1974 nid_t new_xnid = nid_of_node(page);
1975 struct node_info ni;
1977 /* 1: invalidate the previous xattr nid */
1981 /* Deallocate node address */
1982 get_node_info(sbi, prev_xnid, &ni);
1983 f2fs_bug_on(sbi, ni.blk_addr == NULL_ADDR);
1984 invalidate_blocks(sbi, ni.blk_addr);
1985 dec_valid_node_count(sbi, inode);
1986 set_node_addr(sbi, &ni, NULL_ADDR, false);
1989 /* 2: allocate new xattr nid */
1990 if (unlikely(!inc_valid_node_count(sbi, inode)))
1991 f2fs_bug_on(sbi, 1);
1993 remove_free_nid(NM_I(sbi), new_xnid);
1994 get_node_info(sbi, new_xnid, &ni);
1995 ni.ino = inode->i_ino;
1996 set_node_addr(sbi, &ni, NEW_ADDR, false);
1997 f2fs_i_xnid_write(inode, new_xnid);
1999 /* 3: update xattr blkaddr */
2000 refresh_sit_entry(sbi, NEW_ADDR, blkaddr);
2001 set_node_addr(sbi, &ni, blkaddr, false);
2004 int recover_inode_page(struct f2fs_sb_info *sbi, struct page *page)
2006 struct f2fs_inode *src, *dst;
2007 nid_t ino = ino_of_node(page);
2008 struct node_info old_ni, new_ni;
2011 get_node_info(sbi, ino, &old_ni);
2013 if (unlikely(old_ni.blk_addr != NULL_ADDR))
2016 ipage = f2fs_grab_cache_page(NODE_MAPPING(sbi), ino, false);
2020 /* Should not use this inode from free nid list */
2021 remove_free_nid(NM_I(sbi), ino);
2023 if (!PageUptodate(ipage))
2024 SetPageUptodate(ipage);
2025 fill_node_footer(ipage, ino, ino, 0, true);
2027 src = F2FS_INODE(page);
2028 dst = F2FS_INODE(ipage);
2030 memcpy(dst, src, (unsigned long)&src->i_ext - (unsigned long)src);
2032 dst->i_blocks = cpu_to_le64(1);
2033 dst->i_links = cpu_to_le32(1);
2034 dst->i_xattr_nid = 0;
2035 dst->i_inline = src->i_inline & F2FS_INLINE_XATTR;
2040 if (unlikely(!inc_valid_node_count(sbi, NULL)))
2042 set_node_addr(sbi, &new_ni, NEW_ADDR, false);
2043 inc_valid_inode_count(sbi);
2044 set_page_dirty(ipage);
2045 f2fs_put_page(ipage, 1);
2049 int restore_node_summary(struct f2fs_sb_info *sbi,
2050 unsigned int segno, struct f2fs_summary_block *sum)
2052 struct f2fs_node *rn;
2053 struct f2fs_summary *sum_entry;
2055 int bio_blocks = MAX_BIO_BLOCKS(sbi);
2056 int i, idx, last_offset, nrpages;
2058 /* scan the node segment */
2059 last_offset = sbi->blocks_per_seg;
2060 addr = START_BLOCK(sbi, segno);
2061 sum_entry = &sum->entries[0];
2063 for (i = 0; i < last_offset; i += nrpages, addr += nrpages) {
2064 nrpages = min(last_offset - i, bio_blocks);
2066 /* readahead node pages */
2067 ra_meta_pages(sbi, addr, nrpages, META_POR, true);
2069 for (idx = addr; idx < addr + nrpages; idx++) {
2070 struct page *page = get_tmp_page(sbi, idx);
2072 rn = F2FS_NODE(page);
2073 sum_entry->nid = rn->footer.nid;
2074 sum_entry->version = 0;
2075 sum_entry->ofs_in_node = 0;
2077 f2fs_put_page(page, 1);
2080 invalidate_mapping_pages(META_MAPPING(sbi), addr,
2086 static void remove_nats_in_journal(struct f2fs_sb_info *sbi)
2088 struct f2fs_nm_info *nm_i = NM_I(sbi);
2089 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
2090 struct f2fs_journal *journal = curseg->journal;
2093 down_write(&curseg->journal_rwsem);
2094 for (i = 0; i < nats_in_cursum(journal); i++) {
2095 struct nat_entry *ne;
2096 struct f2fs_nat_entry raw_ne;
2097 nid_t nid = le32_to_cpu(nid_in_journal(journal, i));
2099 raw_ne = nat_in_journal(journal, i);
2101 ne = __lookup_nat_cache(nm_i, nid);
2103 ne = grab_nat_entry(nm_i, nid);
2104 node_info_from_raw_nat(&ne->ni, &raw_ne);
2106 __set_nat_cache_dirty(nm_i, ne);
2108 update_nats_in_cursum(journal, -i);
2109 up_write(&curseg->journal_rwsem);
2112 static void __adjust_nat_entry_set(struct nat_entry_set *nes,
2113 struct list_head *head, int max)
2115 struct nat_entry_set *cur;
2117 if (nes->entry_cnt >= max)
2120 list_for_each_entry(cur, head, set_list) {
2121 if (cur->entry_cnt >= nes->entry_cnt) {
2122 list_add(&nes->set_list, cur->set_list.prev);
2127 list_add_tail(&nes->set_list, head);
2130 static void __flush_nat_entry_set(struct f2fs_sb_info *sbi,
2131 struct nat_entry_set *set)
2133 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
2134 struct f2fs_journal *journal = curseg->journal;
2135 nid_t start_nid = set->set * NAT_ENTRY_PER_BLOCK;
2136 bool to_journal = true;
2137 struct f2fs_nat_block *nat_blk;
2138 struct nat_entry *ne, *cur;
2139 struct page *page = NULL;
2142 * there are two steps to flush nat entries:
2143 * #1, flush nat entries to journal in current hot data summary block.
2144 * #2, flush nat entries to nat page.
2146 if (!__has_cursum_space(journal, set->entry_cnt, NAT_JOURNAL))
2150 down_write(&curseg->journal_rwsem);
2152 page = get_next_nat_page(sbi, start_nid);
2153 nat_blk = page_address(page);
2154 f2fs_bug_on(sbi, !nat_blk);
2157 /* flush dirty nats in nat entry set */
2158 list_for_each_entry_safe(ne, cur, &set->entry_list, list) {
2159 struct f2fs_nat_entry *raw_ne;
2160 nid_t nid = nat_get_nid(ne);
2163 if (nat_get_blkaddr(ne) == NEW_ADDR)
2167 offset = lookup_journal_in_cursum(journal,
2168 NAT_JOURNAL, nid, 1);
2169 f2fs_bug_on(sbi, offset < 0);
2170 raw_ne = &nat_in_journal(journal, offset);
2171 nid_in_journal(journal, offset) = cpu_to_le32(nid);
2173 raw_ne = &nat_blk->entries[nid - start_nid];
2175 raw_nat_from_node_info(raw_ne, &ne->ni);
2177 __clear_nat_cache_dirty(NM_I(sbi), ne);
2178 if (nat_get_blkaddr(ne) == NULL_ADDR)
2179 add_free_nid(sbi, nid, false);
2183 up_write(&curseg->journal_rwsem);
2185 f2fs_put_page(page, 1);
2187 f2fs_bug_on(sbi, set->entry_cnt);
2189 radix_tree_delete(&NM_I(sbi)->nat_set_root, set->set);
2190 kmem_cache_free(nat_entry_set_slab, set);
2194 * This function is called during the checkpointing process.
2196 void flush_nat_entries(struct f2fs_sb_info *sbi)
2198 struct f2fs_nm_info *nm_i = NM_I(sbi);
2199 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
2200 struct f2fs_journal *journal = curseg->journal;
2201 struct nat_entry_set *setvec[SETVEC_SIZE];
2202 struct nat_entry_set *set, *tmp;
2207 if (!nm_i->dirty_nat_cnt)
2210 down_write(&nm_i->nat_tree_lock);
2213 * if there are no enough space in journal to store dirty nat
2214 * entries, remove all entries from journal and merge them
2215 * into nat entry set.
2217 if (!__has_cursum_space(journal, nm_i->dirty_nat_cnt, NAT_JOURNAL))
2218 remove_nats_in_journal(sbi);
2220 while ((found = __gang_lookup_nat_set(nm_i,
2221 set_idx, SETVEC_SIZE, setvec))) {
2223 set_idx = setvec[found - 1]->set + 1;
2224 for (idx = 0; idx < found; idx++)
2225 __adjust_nat_entry_set(setvec[idx], &sets,
2226 MAX_NAT_JENTRIES(journal));
2229 /* flush dirty nats in nat entry set */
2230 list_for_each_entry_safe(set, tmp, &sets, set_list)
2231 __flush_nat_entry_set(sbi, set);
2233 up_write(&nm_i->nat_tree_lock);
2235 f2fs_bug_on(sbi, nm_i->dirty_nat_cnt);
2238 static int init_node_manager(struct f2fs_sb_info *sbi)
2240 struct f2fs_super_block *sb_raw = F2FS_RAW_SUPER(sbi);
2241 struct f2fs_nm_info *nm_i = NM_I(sbi);
2242 unsigned char *version_bitmap;
2243 unsigned int nat_segs, nat_blocks;
2245 nm_i->nat_blkaddr = le32_to_cpu(sb_raw->nat_blkaddr);
2247 /* segment_count_nat includes pair segment so divide to 2. */
2248 nat_segs = le32_to_cpu(sb_raw->segment_count_nat) >> 1;
2249 nat_blocks = nat_segs << le32_to_cpu(sb_raw->log_blocks_per_seg);
2251 nm_i->max_nid = NAT_ENTRY_PER_BLOCK * nat_blocks;
2253 /* not used nids: 0, node, meta, (and root counted as valid node) */
2254 nm_i->available_nids = nm_i->max_nid - F2FS_RESERVED_NODE_NUM;
2257 nm_i->ram_thresh = DEF_RAM_THRESHOLD;
2258 nm_i->ra_nid_pages = DEF_RA_NID_PAGES;
2259 nm_i->dirty_nats_ratio = DEF_DIRTY_NAT_RATIO_THRESHOLD;
2261 INIT_RADIX_TREE(&nm_i->free_nid_root, GFP_ATOMIC);
2262 INIT_LIST_HEAD(&nm_i->free_nid_list);
2263 INIT_RADIX_TREE(&nm_i->nat_root, GFP_NOIO);
2264 INIT_RADIX_TREE(&nm_i->nat_set_root, GFP_NOIO);
2265 INIT_LIST_HEAD(&nm_i->nat_entries);
2267 mutex_init(&nm_i->build_lock);
2268 spin_lock_init(&nm_i->free_nid_list_lock);
2269 init_rwsem(&nm_i->nat_tree_lock);
2271 nm_i->next_scan_nid = le32_to_cpu(sbi->ckpt->next_free_nid);
2272 nm_i->bitmap_size = __bitmap_size(sbi, NAT_BITMAP);
2273 version_bitmap = __bitmap_ptr(sbi, NAT_BITMAP);
2274 if (!version_bitmap)
2277 nm_i->nat_bitmap = kmemdup(version_bitmap, nm_i->bitmap_size,
2279 if (!nm_i->nat_bitmap)
2284 int build_node_manager(struct f2fs_sb_info *sbi)
2288 sbi->nm_info = kzalloc(sizeof(struct f2fs_nm_info), GFP_KERNEL);
2292 err = init_node_manager(sbi);
2296 build_free_nids(sbi);
2300 void destroy_node_manager(struct f2fs_sb_info *sbi)
2302 struct f2fs_nm_info *nm_i = NM_I(sbi);
2303 struct free_nid *i, *next_i;
2304 struct nat_entry *natvec[NATVEC_SIZE];
2305 struct nat_entry_set *setvec[SETVEC_SIZE];
2312 /* destroy free nid list */
2313 spin_lock(&nm_i->free_nid_list_lock);
2314 list_for_each_entry_safe(i, next_i, &nm_i->free_nid_list, list) {
2315 f2fs_bug_on(sbi, i->state == NID_ALLOC);
2316 __del_from_free_nid_list(nm_i, i);
2318 spin_unlock(&nm_i->free_nid_list_lock);
2319 kmem_cache_free(free_nid_slab, i);
2320 spin_lock(&nm_i->free_nid_list_lock);
2322 f2fs_bug_on(sbi, nm_i->fcnt);
2323 spin_unlock(&nm_i->free_nid_list_lock);
2325 /* destroy nat cache */
2326 down_write(&nm_i->nat_tree_lock);
2327 while ((found = __gang_lookup_nat_cache(nm_i,
2328 nid, NATVEC_SIZE, natvec))) {
2331 nid = nat_get_nid(natvec[found - 1]) + 1;
2332 for (idx = 0; idx < found; idx++)
2333 __del_from_nat_cache(nm_i, natvec[idx]);
2335 f2fs_bug_on(sbi, nm_i->nat_cnt);
2337 /* destroy nat set cache */
2339 while ((found = __gang_lookup_nat_set(nm_i,
2340 nid, SETVEC_SIZE, setvec))) {
2343 nid = setvec[found - 1]->set + 1;
2344 for (idx = 0; idx < found; idx++) {
2345 /* entry_cnt is not zero, when cp_error was occurred */
2346 f2fs_bug_on(sbi, !list_empty(&setvec[idx]->entry_list));
2347 radix_tree_delete(&nm_i->nat_set_root, setvec[idx]->set);
2348 kmem_cache_free(nat_entry_set_slab, setvec[idx]);
2351 up_write(&nm_i->nat_tree_lock);
2353 kfree(nm_i->nat_bitmap);
2354 sbi->nm_info = NULL;
2358 int __init create_node_manager_caches(void)
2360 nat_entry_slab = f2fs_kmem_cache_create("nat_entry",
2361 sizeof(struct nat_entry));
2362 if (!nat_entry_slab)
2365 free_nid_slab = f2fs_kmem_cache_create("free_nid",
2366 sizeof(struct free_nid));
2368 goto destroy_nat_entry;
2370 nat_entry_set_slab = f2fs_kmem_cache_create("nat_entry_set",
2371 sizeof(struct nat_entry_set));
2372 if (!nat_entry_set_slab)
2373 goto destroy_free_nid;
2377 kmem_cache_destroy(free_nid_slab);
2379 kmem_cache_destroy(nat_entry_slab);
2384 void destroy_node_manager_caches(void)
2386 kmem_cache_destroy(nat_entry_set_slab);
2387 kmem_cache_destroy(free_nid_slab);
2388 kmem_cache_destroy(nat_entry_slab);