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 } else if (type == DIRTY_DENTS) {
58 if (sbi->sb->s_bdi->wb.dirty_exceeded)
60 mem_size = get_pages(sbi, F2FS_DIRTY_DENTS);
61 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 1);
62 } else if (type == INO_ENTRIES) {
65 for (i = 0; i <= UPDATE_INO; i++)
66 mem_size += (sbi->im[i].ino_num *
67 sizeof(struct ino_entry)) >> PAGE_SHIFT;
68 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 1);
69 } else if (type == EXTENT_CACHE) {
70 mem_size = (atomic_read(&sbi->total_ext_tree) *
71 sizeof(struct extent_tree) +
72 atomic_read(&sbi->total_ext_node) *
73 sizeof(struct extent_node)) >> PAGE_SHIFT;
74 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 1);
76 if (!sbi->sb->s_bdi->wb.dirty_exceeded)
82 static void clear_node_page_dirty(struct page *page)
84 struct address_space *mapping = page->mapping;
85 unsigned int long flags;
87 if (PageDirty(page)) {
88 spin_lock_irqsave(&mapping->tree_lock, flags);
89 radix_tree_tag_clear(&mapping->page_tree,
92 spin_unlock_irqrestore(&mapping->tree_lock, flags);
94 clear_page_dirty_for_io(page);
95 dec_page_count(F2FS_M_SB(mapping), F2FS_DIRTY_NODES);
97 ClearPageUptodate(page);
100 static struct page *get_current_nat_page(struct f2fs_sb_info *sbi, nid_t nid)
102 pgoff_t index = current_nat_addr(sbi, nid);
103 return get_meta_page(sbi, index);
106 static struct page *get_next_nat_page(struct f2fs_sb_info *sbi, nid_t nid)
108 struct page *src_page;
109 struct page *dst_page;
114 struct f2fs_nm_info *nm_i = NM_I(sbi);
116 src_off = current_nat_addr(sbi, nid);
117 dst_off = next_nat_addr(sbi, src_off);
119 /* get current nat block page with lock */
120 src_page = get_meta_page(sbi, src_off);
121 dst_page = grab_meta_page(sbi, dst_off);
122 f2fs_bug_on(sbi, PageDirty(src_page));
124 src_addr = page_address(src_page);
125 dst_addr = page_address(dst_page);
126 memcpy(dst_addr, src_addr, PAGE_SIZE);
127 set_page_dirty(dst_page);
128 f2fs_put_page(src_page, 1);
130 set_to_next_nat(nm_i, nid);
135 static struct nat_entry *__lookup_nat_cache(struct f2fs_nm_info *nm_i, nid_t n)
137 return radix_tree_lookup(&nm_i->nat_root, n);
140 static unsigned int __gang_lookup_nat_cache(struct f2fs_nm_info *nm_i,
141 nid_t start, unsigned int nr, struct nat_entry **ep)
143 return radix_tree_gang_lookup(&nm_i->nat_root, (void **)ep, start, nr);
146 static void __del_from_nat_cache(struct f2fs_nm_info *nm_i, struct nat_entry *e)
149 radix_tree_delete(&nm_i->nat_root, nat_get_nid(e));
151 kmem_cache_free(nat_entry_slab, e);
154 static void __set_nat_cache_dirty(struct f2fs_nm_info *nm_i,
155 struct nat_entry *ne)
157 nid_t set = NAT_BLOCK_OFFSET(ne->ni.nid);
158 struct nat_entry_set *head;
160 if (get_nat_flag(ne, IS_DIRTY))
163 head = radix_tree_lookup(&nm_i->nat_set_root, set);
165 head = f2fs_kmem_cache_alloc(nat_entry_set_slab, GFP_NOFS);
167 INIT_LIST_HEAD(&head->entry_list);
168 INIT_LIST_HEAD(&head->set_list);
171 f2fs_radix_tree_insert(&nm_i->nat_set_root, set, head);
173 list_move_tail(&ne->list, &head->entry_list);
174 nm_i->dirty_nat_cnt++;
176 set_nat_flag(ne, IS_DIRTY, true);
179 static void __clear_nat_cache_dirty(struct f2fs_nm_info *nm_i,
180 struct nat_entry *ne)
182 nid_t set = NAT_BLOCK_OFFSET(ne->ni.nid);
183 struct nat_entry_set *head;
185 head = radix_tree_lookup(&nm_i->nat_set_root, set);
187 list_move_tail(&ne->list, &nm_i->nat_entries);
188 set_nat_flag(ne, IS_DIRTY, false);
190 nm_i->dirty_nat_cnt--;
194 static unsigned int __gang_lookup_nat_set(struct f2fs_nm_info *nm_i,
195 nid_t start, unsigned int nr, struct nat_entry_set **ep)
197 return radix_tree_gang_lookup(&nm_i->nat_set_root, (void **)ep,
201 int need_dentry_mark(struct f2fs_sb_info *sbi, nid_t nid)
203 struct f2fs_nm_info *nm_i = NM_I(sbi);
207 down_read(&nm_i->nat_tree_lock);
208 e = __lookup_nat_cache(nm_i, nid);
210 if (!get_nat_flag(e, IS_CHECKPOINTED) &&
211 !get_nat_flag(e, HAS_FSYNCED_INODE))
214 up_read(&nm_i->nat_tree_lock);
218 bool is_checkpointed_node(struct f2fs_sb_info *sbi, nid_t nid)
220 struct f2fs_nm_info *nm_i = NM_I(sbi);
224 down_read(&nm_i->nat_tree_lock);
225 e = __lookup_nat_cache(nm_i, nid);
226 if (e && !get_nat_flag(e, IS_CHECKPOINTED))
228 up_read(&nm_i->nat_tree_lock);
232 bool need_inode_block_update(struct f2fs_sb_info *sbi, nid_t ino)
234 struct f2fs_nm_info *nm_i = NM_I(sbi);
236 bool need_update = true;
238 down_read(&nm_i->nat_tree_lock);
239 e = __lookup_nat_cache(nm_i, ino);
240 if (e && get_nat_flag(e, HAS_LAST_FSYNC) &&
241 (get_nat_flag(e, IS_CHECKPOINTED) ||
242 get_nat_flag(e, HAS_FSYNCED_INODE)))
244 up_read(&nm_i->nat_tree_lock);
248 static struct nat_entry *grab_nat_entry(struct f2fs_nm_info *nm_i, nid_t nid)
250 struct nat_entry *new;
252 new = f2fs_kmem_cache_alloc(nat_entry_slab, GFP_NOFS);
253 f2fs_radix_tree_insert(&nm_i->nat_root, nid, new);
254 memset(new, 0, sizeof(struct nat_entry));
255 nat_set_nid(new, nid);
257 list_add_tail(&new->list, &nm_i->nat_entries);
262 static void cache_nat_entry(struct f2fs_sb_info *sbi, nid_t nid,
263 struct f2fs_nat_entry *ne)
265 struct f2fs_nm_info *nm_i = NM_I(sbi);
268 e = __lookup_nat_cache(nm_i, nid);
270 e = grab_nat_entry(nm_i, nid);
271 node_info_from_raw_nat(&e->ni, ne);
273 f2fs_bug_on(sbi, nat_get_ino(e) != ne->ino ||
274 nat_get_blkaddr(e) != ne->block_addr ||
275 nat_get_version(e) != ne->version);
279 static void set_node_addr(struct f2fs_sb_info *sbi, struct node_info *ni,
280 block_t new_blkaddr, bool fsync_done)
282 struct f2fs_nm_info *nm_i = NM_I(sbi);
285 down_write(&nm_i->nat_tree_lock);
286 e = __lookup_nat_cache(nm_i, ni->nid);
288 e = grab_nat_entry(nm_i, ni->nid);
289 copy_node_info(&e->ni, ni);
290 f2fs_bug_on(sbi, ni->blk_addr == NEW_ADDR);
291 } else if (new_blkaddr == NEW_ADDR) {
293 * when nid is reallocated,
294 * previous nat entry can be remained in nat cache.
295 * So, reinitialize it with new information.
297 copy_node_info(&e->ni, ni);
298 f2fs_bug_on(sbi, ni->blk_addr != NULL_ADDR);
302 f2fs_bug_on(sbi, nat_get_blkaddr(e) != ni->blk_addr);
303 f2fs_bug_on(sbi, nat_get_blkaddr(e) == NULL_ADDR &&
304 new_blkaddr == NULL_ADDR);
305 f2fs_bug_on(sbi, nat_get_blkaddr(e) == NEW_ADDR &&
306 new_blkaddr == NEW_ADDR);
307 f2fs_bug_on(sbi, nat_get_blkaddr(e) != NEW_ADDR &&
308 nat_get_blkaddr(e) != NULL_ADDR &&
309 new_blkaddr == NEW_ADDR);
311 /* increment version no as node is removed */
312 if (nat_get_blkaddr(e) != NEW_ADDR && new_blkaddr == NULL_ADDR) {
313 unsigned char version = nat_get_version(e);
314 nat_set_version(e, inc_node_version(version));
316 /* in order to reuse the nid */
317 if (nm_i->next_scan_nid > ni->nid)
318 nm_i->next_scan_nid = ni->nid;
322 nat_set_blkaddr(e, new_blkaddr);
323 if (new_blkaddr == NEW_ADDR || new_blkaddr == NULL_ADDR)
324 set_nat_flag(e, IS_CHECKPOINTED, false);
325 __set_nat_cache_dirty(nm_i, e);
327 /* update fsync_mark if its inode nat entry is still alive */
328 if (ni->nid != ni->ino)
329 e = __lookup_nat_cache(nm_i, ni->ino);
331 if (fsync_done && ni->nid == ni->ino)
332 set_nat_flag(e, HAS_FSYNCED_INODE, true);
333 set_nat_flag(e, HAS_LAST_FSYNC, fsync_done);
335 up_write(&nm_i->nat_tree_lock);
338 int try_to_free_nats(struct f2fs_sb_info *sbi, int nr_shrink)
340 struct f2fs_nm_info *nm_i = NM_I(sbi);
343 if (!down_write_trylock(&nm_i->nat_tree_lock))
346 while (nr_shrink && !list_empty(&nm_i->nat_entries)) {
347 struct nat_entry *ne;
348 ne = list_first_entry(&nm_i->nat_entries,
349 struct nat_entry, list);
350 __del_from_nat_cache(nm_i, ne);
353 up_write(&nm_i->nat_tree_lock);
354 return nr - nr_shrink;
358 * This function always returns success
360 void get_node_info(struct f2fs_sb_info *sbi, nid_t nid, struct node_info *ni)
362 struct f2fs_nm_info *nm_i = NM_I(sbi);
363 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
364 struct f2fs_journal *journal = curseg->journal;
365 nid_t start_nid = START_NID(nid);
366 struct f2fs_nat_block *nat_blk;
367 struct page *page = NULL;
368 struct f2fs_nat_entry ne;
374 /* Check nat cache */
375 down_read(&nm_i->nat_tree_lock);
376 e = __lookup_nat_cache(nm_i, nid);
378 ni->ino = nat_get_ino(e);
379 ni->blk_addr = nat_get_blkaddr(e);
380 ni->version = nat_get_version(e);
381 up_read(&nm_i->nat_tree_lock);
385 memset(&ne, 0, sizeof(struct f2fs_nat_entry));
387 /* Check current segment summary */
388 down_read(&curseg->journal_rwsem);
389 i = lookup_journal_in_cursum(journal, NAT_JOURNAL, nid, 0);
391 ne = nat_in_journal(journal, i);
392 node_info_from_raw_nat(ni, &ne);
394 up_read(&curseg->journal_rwsem);
398 /* Fill node_info from nat page */
399 page = get_current_nat_page(sbi, start_nid);
400 nat_blk = (struct f2fs_nat_block *)page_address(page);
401 ne = nat_blk->entries[nid - start_nid];
402 node_info_from_raw_nat(ni, &ne);
403 f2fs_put_page(page, 1);
405 up_read(&nm_i->nat_tree_lock);
406 /* cache nat entry */
407 down_write(&nm_i->nat_tree_lock);
408 cache_nat_entry(sbi, nid, &ne);
409 up_write(&nm_i->nat_tree_lock);
413 * readahead MAX_RA_NODE number of node pages.
415 static void ra_node_pages(struct page *parent, int start, int n)
417 struct f2fs_sb_info *sbi = F2FS_P_SB(parent);
418 struct blk_plug plug;
422 blk_start_plug(&plug);
424 /* Then, try readahead for siblings of the desired node */
426 end = min(end, NIDS_PER_BLOCK);
427 for (i = start; i < end; i++) {
428 nid = get_nid(parent, i, false);
429 ra_node_page(sbi, nid);
432 blk_finish_plug(&plug);
435 pgoff_t get_next_page_offset(struct dnode_of_data *dn, pgoff_t pgofs)
437 const long direct_index = ADDRS_PER_INODE(dn->inode);
438 const long direct_blks = ADDRS_PER_BLOCK;
439 const long indirect_blks = ADDRS_PER_BLOCK * NIDS_PER_BLOCK;
440 unsigned int skipped_unit = ADDRS_PER_BLOCK;
441 int cur_level = dn->cur_level;
442 int max_level = dn->max_level;
448 while (max_level-- > cur_level)
449 skipped_unit *= NIDS_PER_BLOCK;
451 switch (dn->max_level) {
453 base += 2 * indirect_blks;
455 base += 2 * direct_blks;
457 base += direct_index;
460 f2fs_bug_on(F2FS_I_SB(dn->inode), 1);
463 return ((pgofs - base) / skipped_unit + 1) * skipped_unit + base;
467 * The maximum depth is four.
468 * Offset[0] will have raw inode offset.
470 static int get_node_path(struct inode *inode, long block,
471 int offset[4], unsigned int noffset[4])
473 const long direct_index = ADDRS_PER_INODE(inode);
474 const long direct_blks = ADDRS_PER_BLOCK;
475 const long dptrs_per_blk = NIDS_PER_BLOCK;
476 const long indirect_blks = ADDRS_PER_BLOCK * NIDS_PER_BLOCK;
477 const long dindirect_blks = indirect_blks * NIDS_PER_BLOCK;
483 if (block < direct_index) {
487 block -= direct_index;
488 if (block < direct_blks) {
489 offset[n++] = NODE_DIR1_BLOCK;
495 block -= direct_blks;
496 if (block < direct_blks) {
497 offset[n++] = NODE_DIR2_BLOCK;
503 block -= direct_blks;
504 if (block < indirect_blks) {
505 offset[n++] = NODE_IND1_BLOCK;
507 offset[n++] = block / direct_blks;
508 noffset[n] = 4 + offset[n - 1];
509 offset[n] = block % direct_blks;
513 block -= indirect_blks;
514 if (block < indirect_blks) {
515 offset[n++] = NODE_IND2_BLOCK;
516 noffset[n] = 4 + dptrs_per_blk;
517 offset[n++] = block / direct_blks;
518 noffset[n] = 5 + dptrs_per_blk + offset[n - 1];
519 offset[n] = block % direct_blks;
523 block -= indirect_blks;
524 if (block < dindirect_blks) {
525 offset[n++] = NODE_DIND_BLOCK;
526 noffset[n] = 5 + (dptrs_per_blk * 2);
527 offset[n++] = block / indirect_blks;
528 noffset[n] = 6 + (dptrs_per_blk * 2) +
529 offset[n - 1] * (dptrs_per_blk + 1);
530 offset[n++] = (block / direct_blks) % dptrs_per_blk;
531 noffset[n] = 7 + (dptrs_per_blk * 2) +
532 offset[n - 2] * (dptrs_per_blk + 1) +
534 offset[n] = block % direct_blks;
545 * Caller should call f2fs_put_dnode(dn).
546 * Also, it should grab and release a rwsem by calling f2fs_lock_op() and
547 * f2fs_unlock_op() only if ro is not set RDONLY_NODE.
548 * In the case of RDONLY_NODE, we don't need to care about mutex.
550 int get_dnode_of_data(struct dnode_of_data *dn, pgoff_t index, int mode)
552 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
553 struct page *npage[4];
554 struct page *parent = NULL;
556 unsigned int noffset[4];
561 level = get_node_path(dn->inode, index, offset, noffset);
563 nids[0] = dn->inode->i_ino;
564 npage[0] = dn->inode_page;
567 npage[0] = get_node_page(sbi, nids[0]);
568 if (IS_ERR(npage[0]))
569 return PTR_ERR(npage[0]);
572 /* if inline_data is set, should not report any block indices */
573 if (f2fs_has_inline_data(dn->inode) && index) {
575 f2fs_put_page(npage[0], 1);
581 nids[1] = get_nid(parent, offset[0], true);
582 dn->inode_page = npage[0];
583 dn->inode_page_locked = true;
585 /* get indirect or direct nodes */
586 for (i = 1; i <= level; i++) {
589 if (!nids[i] && mode == ALLOC_NODE) {
591 if (!alloc_nid(sbi, &(nids[i]))) {
597 npage[i] = new_node_page(dn, noffset[i], NULL);
598 if (IS_ERR(npage[i])) {
599 alloc_nid_failed(sbi, nids[i]);
600 err = PTR_ERR(npage[i]);
604 set_nid(parent, offset[i - 1], nids[i], i == 1);
605 alloc_nid_done(sbi, nids[i]);
607 } else if (mode == LOOKUP_NODE_RA && i == level && level > 1) {
608 npage[i] = get_node_page_ra(parent, offset[i - 1]);
609 if (IS_ERR(npage[i])) {
610 err = PTR_ERR(npage[i]);
616 dn->inode_page_locked = false;
619 f2fs_put_page(parent, 1);
623 npage[i] = get_node_page(sbi, nids[i]);
624 if (IS_ERR(npage[i])) {
625 err = PTR_ERR(npage[i]);
626 f2fs_put_page(npage[0], 0);
632 nids[i + 1] = get_nid(parent, offset[i], false);
635 dn->nid = nids[level];
636 dn->ofs_in_node = offset[level];
637 dn->node_page = npage[level];
638 dn->data_blkaddr = datablock_addr(dn->node_page, dn->ofs_in_node);
642 f2fs_put_page(parent, 1);
644 f2fs_put_page(npage[0], 0);
646 dn->inode_page = NULL;
647 dn->node_page = NULL;
648 if (err == -ENOENT) {
650 dn->max_level = level;
651 dn->ofs_in_node = offset[level];
656 static void truncate_node(struct dnode_of_data *dn)
658 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
661 get_node_info(sbi, dn->nid, &ni);
662 if (dn->inode->i_blocks == 0) {
663 f2fs_bug_on(sbi, ni.blk_addr != NULL_ADDR);
666 f2fs_bug_on(sbi, ni.blk_addr == NULL_ADDR);
668 /* Deallocate node address */
669 invalidate_blocks(sbi, ni.blk_addr);
670 dec_valid_node_count(sbi, dn->inode);
671 set_node_addr(sbi, &ni, NULL_ADDR, false);
673 if (dn->nid == dn->inode->i_ino) {
674 remove_orphan_inode(sbi, dn->nid);
675 dec_valid_inode_count(sbi);
676 f2fs_inode_synced(dn->inode);
679 clear_node_page_dirty(dn->node_page);
680 set_sbi_flag(sbi, SBI_IS_DIRTY);
682 f2fs_put_page(dn->node_page, 1);
684 invalidate_mapping_pages(NODE_MAPPING(sbi),
685 dn->node_page->index, dn->node_page->index);
687 dn->node_page = NULL;
688 trace_f2fs_truncate_node(dn->inode, dn->nid, ni.blk_addr);
691 static int truncate_dnode(struct dnode_of_data *dn)
698 /* get direct node */
699 page = get_node_page(F2FS_I_SB(dn->inode), dn->nid);
700 if (IS_ERR(page) && PTR_ERR(page) == -ENOENT)
702 else if (IS_ERR(page))
703 return PTR_ERR(page);
705 /* Make dnode_of_data for parameter */
706 dn->node_page = page;
708 truncate_data_blocks(dn);
713 static int truncate_nodes(struct dnode_of_data *dn, unsigned int nofs,
716 struct dnode_of_data rdn = *dn;
718 struct f2fs_node *rn;
720 unsigned int child_nofs;
725 return NIDS_PER_BLOCK + 1;
727 trace_f2fs_truncate_nodes_enter(dn->inode, dn->nid, dn->data_blkaddr);
729 page = get_node_page(F2FS_I_SB(dn->inode), dn->nid);
731 trace_f2fs_truncate_nodes_exit(dn->inode, PTR_ERR(page));
732 return PTR_ERR(page);
735 ra_node_pages(page, ofs, NIDS_PER_BLOCK);
737 rn = F2FS_NODE(page);
739 for (i = ofs; i < NIDS_PER_BLOCK; i++, freed++) {
740 child_nid = le32_to_cpu(rn->in.nid[i]);
744 ret = truncate_dnode(&rdn);
747 if (set_nid(page, i, 0, false))
748 dn->node_changed = true;
751 child_nofs = nofs + ofs * (NIDS_PER_BLOCK + 1) + 1;
752 for (i = ofs; i < NIDS_PER_BLOCK; i++) {
753 child_nid = le32_to_cpu(rn->in.nid[i]);
754 if (child_nid == 0) {
755 child_nofs += NIDS_PER_BLOCK + 1;
759 ret = truncate_nodes(&rdn, child_nofs, 0, depth - 1);
760 if (ret == (NIDS_PER_BLOCK + 1)) {
761 if (set_nid(page, i, 0, false))
762 dn->node_changed = true;
764 } else if (ret < 0 && ret != -ENOENT) {
772 /* remove current indirect node */
773 dn->node_page = page;
777 f2fs_put_page(page, 1);
779 trace_f2fs_truncate_nodes_exit(dn->inode, freed);
783 f2fs_put_page(page, 1);
784 trace_f2fs_truncate_nodes_exit(dn->inode, ret);
788 static int truncate_partial_nodes(struct dnode_of_data *dn,
789 struct f2fs_inode *ri, int *offset, int depth)
791 struct page *pages[2];
798 nid[0] = le32_to_cpu(ri->i_nid[offset[0] - NODE_DIR1_BLOCK]);
802 /* get indirect nodes in the path */
803 for (i = 0; i < idx + 1; i++) {
804 /* reference count'll be increased */
805 pages[i] = get_node_page(F2FS_I_SB(dn->inode), nid[i]);
806 if (IS_ERR(pages[i])) {
807 err = PTR_ERR(pages[i]);
811 nid[i + 1] = get_nid(pages[i], offset[i + 1], false);
814 ra_node_pages(pages[idx], offset[idx + 1], NIDS_PER_BLOCK);
816 /* free direct nodes linked to a partial indirect node */
817 for (i = offset[idx + 1]; i < NIDS_PER_BLOCK; i++) {
818 child_nid = get_nid(pages[idx], i, false);
822 err = truncate_dnode(dn);
825 if (set_nid(pages[idx], i, 0, false))
826 dn->node_changed = true;
829 if (offset[idx + 1] == 0) {
830 dn->node_page = pages[idx];
834 f2fs_put_page(pages[idx], 1);
840 for (i = idx; i >= 0; i--)
841 f2fs_put_page(pages[i], 1);
843 trace_f2fs_truncate_partial_nodes(dn->inode, nid, depth, err);
849 * All the block addresses of data and nodes should be nullified.
851 int truncate_inode_blocks(struct inode *inode, pgoff_t from)
853 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
854 int err = 0, cont = 1;
855 int level, offset[4], noffset[4];
856 unsigned int nofs = 0;
857 struct f2fs_inode *ri;
858 struct dnode_of_data dn;
861 trace_f2fs_truncate_inode_blocks_enter(inode, from);
863 level = get_node_path(inode, from, offset, noffset);
865 page = get_node_page(sbi, inode->i_ino);
867 trace_f2fs_truncate_inode_blocks_exit(inode, PTR_ERR(page));
868 return PTR_ERR(page);
871 set_new_dnode(&dn, inode, page, NULL, 0);
874 ri = F2FS_INODE(page);
882 if (!offset[level - 1])
884 err = truncate_partial_nodes(&dn, ri, offset, level);
885 if (err < 0 && err != -ENOENT)
887 nofs += 1 + NIDS_PER_BLOCK;
890 nofs = 5 + 2 * NIDS_PER_BLOCK;
891 if (!offset[level - 1])
893 err = truncate_partial_nodes(&dn, ri, offset, level);
894 if (err < 0 && err != -ENOENT)
903 dn.nid = le32_to_cpu(ri->i_nid[offset[0] - NODE_DIR1_BLOCK]);
905 case NODE_DIR1_BLOCK:
906 case NODE_DIR2_BLOCK:
907 err = truncate_dnode(&dn);
910 case NODE_IND1_BLOCK:
911 case NODE_IND2_BLOCK:
912 err = truncate_nodes(&dn, nofs, offset[1], 2);
915 case NODE_DIND_BLOCK:
916 err = truncate_nodes(&dn, nofs, offset[1], 3);
923 if (err < 0 && err != -ENOENT)
925 if (offset[1] == 0 &&
926 ri->i_nid[offset[0] - NODE_DIR1_BLOCK]) {
928 BUG_ON(page->mapping != NODE_MAPPING(sbi));
929 f2fs_wait_on_page_writeback(page, NODE, true);
930 ri->i_nid[offset[0] - NODE_DIR1_BLOCK] = 0;
931 set_page_dirty(page);
939 f2fs_put_page(page, 0);
940 trace_f2fs_truncate_inode_blocks_exit(inode, err);
941 return err > 0 ? 0 : err;
944 int truncate_xattr_node(struct inode *inode, struct page *page)
946 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
947 nid_t nid = F2FS_I(inode)->i_xattr_nid;
948 struct dnode_of_data dn;
954 npage = get_node_page(sbi, nid);
956 return PTR_ERR(npage);
958 f2fs_i_xnid_write(inode, 0);
960 /* need to do checkpoint during fsync */
961 F2FS_I(inode)->xattr_ver = cur_cp_version(F2FS_CKPT(sbi));
963 set_new_dnode(&dn, inode, page, npage, nid);
966 dn.inode_page_locked = true;
972 * Caller should grab and release a rwsem by calling f2fs_lock_op() and
975 int remove_inode_page(struct inode *inode)
977 struct dnode_of_data dn;
980 set_new_dnode(&dn, inode, NULL, NULL, inode->i_ino);
981 err = get_dnode_of_data(&dn, 0, LOOKUP_NODE);
985 err = truncate_xattr_node(inode, dn.inode_page);
991 /* remove potential inline_data blocks */
992 if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
993 S_ISLNK(inode->i_mode))
994 truncate_data_blocks_range(&dn, 1);
996 /* 0 is possible, after f2fs_new_inode() has failed */
997 f2fs_bug_on(F2FS_I_SB(inode),
998 inode->i_blocks != 0 && inode->i_blocks != 1);
1000 /* will put inode & node pages */
1005 struct page *new_inode_page(struct inode *inode)
1007 struct dnode_of_data dn;
1009 /* allocate inode page for new inode */
1010 set_new_dnode(&dn, inode, NULL, NULL, inode->i_ino);
1012 /* caller should f2fs_put_page(page, 1); */
1013 return new_node_page(&dn, 0, NULL);
1016 struct page *new_node_page(struct dnode_of_data *dn,
1017 unsigned int ofs, struct page *ipage)
1019 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
1020 struct node_info old_ni, new_ni;
1024 if (unlikely(is_inode_flag_set(dn->inode, FI_NO_ALLOC)))
1025 return ERR_PTR(-EPERM);
1027 page = f2fs_grab_cache_page(NODE_MAPPING(sbi), dn->nid, false);
1029 return ERR_PTR(-ENOMEM);
1031 if (unlikely(!inc_valid_node_count(sbi, dn->inode))) {
1036 get_node_info(sbi, dn->nid, &old_ni);
1038 /* Reinitialize old_ni with new node page */
1039 f2fs_bug_on(sbi, old_ni.blk_addr != NULL_ADDR);
1041 new_ni.ino = dn->inode->i_ino;
1042 set_node_addr(sbi, &new_ni, NEW_ADDR, false);
1044 f2fs_wait_on_page_writeback(page, NODE, true);
1045 fill_node_footer(page, dn->nid, dn->inode->i_ino, ofs, true);
1046 set_cold_node(dn->inode, page);
1047 if (!PageUptodate(page))
1048 SetPageUptodate(page);
1049 if (set_page_dirty(page))
1050 dn->node_changed = true;
1052 if (f2fs_has_xattr_block(ofs))
1053 f2fs_i_xnid_write(dn->inode, dn->nid);
1056 inc_valid_inode_count(sbi);
1060 clear_node_page_dirty(page);
1061 f2fs_put_page(page, 1);
1062 return ERR_PTR(err);
1066 * Caller should do after getting the following values.
1067 * 0: f2fs_put_page(page, 0)
1068 * LOCKED_PAGE or error: f2fs_put_page(page, 1)
1070 static int read_node_page(struct page *page, int op_flags)
1072 struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1073 struct node_info ni;
1074 struct f2fs_io_info fio = {
1078 .op_flags = op_flags,
1080 .encrypted_page = NULL,
1083 if (PageUptodate(page))
1086 get_node_info(sbi, page->index, &ni);
1088 if (unlikely(ni.blk_addr == NULL_ADDR)) {
1089 ClearPageUptodate(page);
1093 fio.new_blkaddr = fio.old_blkaddr = ni.blk_addr;
1094 return f2fs_submit_page_bio(&fio);
1098 * Readahead a node page
1100 void ra_node_page(struct f2fs_sb_info *sbi, nid_t nid)
1107 f2fs_bug_on(sbi, check_nid_range(sbi, nid));
1110 apage = radix_tree_lookup(&NODE_MAPPING(sbi)->page_tree, nid);
1115 apage = f2fs_grab_cache_page(NODE_MAPPING(sbi), nid, false);
1119 err = read_node_page(apage, REQ_RAHEAD);
1120 f2fs_put_page(apage, err ? 1 : 0);
1123 static struct page *__get_node_page(struct f2fs_sb_info *sbi, pgoff_t nid,
1124 struct page *parent, int start)
1130 return ERR_PTR(-ENOENT);
1131 f2fs_bug_on(sbi, check_nid_range(sbi, nid));
1133 page = f2fs_grab_cache_page(NODE_MAPPING(sbi), nid, false);
1135 return ERR_PTR(-ENOMEM);
1137 err = read_node_page(page, READ_SYNC);
1139 f2fs_put_page(page, 1);
1140 return ERR_PTR(err);
1141 } else if (err == LOCKED_PAGE) {
1146 ra_node_pages(parent, start + 1, MAX_RA_NODE);
1150 if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1151 f2fs_put_page(page, 1);
1155 if (unlikely(!PageUptodate(page)))
1158 if(unlikely(nid != nid_of_node(page))) {
1159 f2fs_bug_on(sbi, 1);
1160 ClearPageUptodate(page);
1162 f2fs_put_page(page, 1);
1163 return ERR_PTR(-EIO);
1168 struct page *get_node_page(struct f2fs_sb_info *sbi, pgoff_t nid)
1170 return __get_node_page(sbi, nid, NULL, 0);
1173 struct page *get_node_page_ra(struct page *parent, int start)
1175 struct f2fs_sb_info *sbi = F2FS_P_SB(parent);
1176 nid_t nid = get_nid(parent, start, false);
1178 return __get_node_page(sbi, nid, parent, start);
1181 static void flush_inline_data(struct f2fs_sb_info *sbi, nid_t ino)
1183 struct inode *inode;
1187 /* should flush inline_data before evict_inode */
1188 inode = ilookup(sbi->sb, ino);
1192 page = pagecache_get_page(inode->i_mapping, 0, FGP_LOCK|FGP_NOWAIT, 0);
1196 if (!PageUptodate(page))
1199 if (!PageDirty(page))
1202 if (!clear_page_dirty_for_io(page))
1205 ret = f2fs_write_inline_data(inode, page);
1206 inode_dec_dirty_pages(inode);
1208 set_page_dirty(page);
1210 f2fs_put_page(page, 1);
1215 void move_node_page(struct page *node_page, int gc_type)
1217 if (gc_type == FG_GC) {
1218 struct f2fs_sb_info *sbi = F2FS_P_SB(node_page);
1219 struct writeback_control wbc = {
1220 .sync_mode = WB_SYNC_ALL,
1225 set_page_dirty(node_page);
1226 f2fs_wait_on_page_writeback(node_page, NODE, true);
1228 f2fs_bug_on(sbi, PageWriteback(node_page));
1229 if (!clear_page_dirty_for_io(node_page))
1232 if (NODE_MAPPING(sbi)->a_ops->writepage(node_page, &wbc))
1233 unlock_page(node_page);
1236 /* set page dirty and write it */
1237 if (!PageWriteback(node_page))
1238 set_page_dirty(node_page);
1241 unlock_page(node_page);
1243 f2fs_put_page(node_page, 0);
1246 static struct page *last_fsync_dnode(struct f2fs_sb_info *sbi, nid_t ino)
1249 struct pagevec pvec;
1250 struct page *last_page = NULL;
1252 pagevec_init(&pvec, 0);
1256 while (index <= end) {
1258 nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1259 PAGECACHE_TAG_DIRTY,
1260 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
1264 for (i = 0; i < nr_pages; i++) {
1265 struct page *page = pvec.pages[i];
1267 if (unlikely(f2fs_cp_error(sbi))) {
1268 f2fs_put_page(last_page, 0);
1269 pagevec_release(&pvec);
1270 return ERR_PTR(-EIO);
1273 if (!IS_DNODE(page) || !is_cold_node(page))
1275 if (ino_of_node(page) != ino)
1280 if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1285 if (ino_of_node(page) != ino)
1286 goto continue_unlock;
1288 if (!PageDirty(page)) {
1289 /* someone wrote it for us */
1290 goto continue_unlock;
1294 f2fs_put_page(last_page, 0);
1300 pagevec_release(&pvec);
1306 int fsync_node_pages(struct f2fs_sb_info *sbi, struct inode *inode,
1307 struct writeback_control *wbc, bool atomic)
1310 struct pagevec pvec;
1312 struct page *last_page = NULL;
1313 bool marked = false;
1314 nid_t ino = inode->i_ino;
1317 last_page = last_fsync_dnode(sbi, ino);
1318 if (IS_ERR_OR_NULL(last_page))
1319 return PTR_ERR_OR_ZERO(last_page);
1322 pagevec_init(&pvec, 0);
1326 while (index <= end) {
1328 nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1329 PAGECACHE_TAG_DIRTY,
1330 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
1334 for (i = 0; i < nr_pages; i++) {
1335 struct page *page = pvec.pages[i];
1337 if (unlikely(f2fs_cp_error(sbi))) {
1338 f2fs_put_page(last_page, 0);
1339 pagevec_release(&pvec);
1343 if (!IS_DNODE(page) || !is_cold_node(page))
1345 if (ino_of_node(page) != ino)
1350 if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1355 if (ino_of_node(page) != ino)
1356 goto continue_unlock;
1358 if (!PageDirty(page) && page != last_page) {
1359 /* someone wrote it for us */
1360 goto continue_unlock;
1363 f2fs_wait_on_page_writeback(page, NODE, true);
1364 BUG_ON(PageWriteback(page));
1366 if (!atomic || page == last_page) {
1367 set_fsync_mark(page, 1);
1368 if (IS_INODE(page)) {
1369 if (is_inode_flag_set(inode,
1371 update_inode(inode, page);
1372 set_dentry_mark(page,
1373 need_dentry_mark(sbi, ino));
1375 /* may be written by other thread */
1376 if (!PageDirty(page))
1377 set_page_dirty(page);
1380 if (!clear_page_dirty_for_io(page))
1381 goto continue_unlock;
1383 ret = NODE_MAPPING(sbi)->a_ops->writepage(page, wbc);
1386 f2fs_put_page(last_page, 0);
1389 if (page == last_page) {
1390 f2fs_put_page(page, 0);
1395 pagevec_release(&pvec);
1401 if (!ret && atomic && !marked) {
1402 f2fs_msg(sbi->sb, KERN_DEBUG,
1403 "Retry to write fsync mark: ino=%u, idx=%lx",
1404 ino, last_page->index);
1405 lock_page(last_page);
1406 set_page_dirty(last_page);
1407 unlock_page(last_page);
1410 return ret ? -EIO: 0;
1413 int sync_node_pages(struct f2fs_sb_info *sbi, struct writeback_control *wbc)
1416 struct pagevec pvec;
1420 pagevec_init(&pvec, 0);
1426 while (index <= end) {
1428 nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1429 PAGECACHE_TAG_DIRTY,
1430 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
1434 for (i = 0; i < nr_pages; i++) {
1435 struct page *page = pvec.pages[i];
1437 if (unlikely(f2fs_cp_error(sbi))) {
1438 pagevec_release(&pvec);
1443 * flushing sequence with step:
1448 if (step == 0 && IS_DNODE(page))
1450 if (step == 1 && (!IS_DNODE(page) ||
1451 is_cold_node(page)))
1453 if (step == 2 && (!IS_DNODE(page) ||
1454 !is_cold_node(page)))
1457 if (!trylock_page(page))
1460 if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1466 if (!PageDirty(page)) {
1467 /* someone wrote it for us */
1468 goto continue_unlock;
1471 /* flush inline_data */
1472 if (is_inline_node(page)) {
1473 clear_inline_node(page);
1475 flush_inline_data(sbi, ino_of_node(page));
1479 f2fs_wait_on_page_writeback(page, NODE, true);
1481 BUG_ON(PageWriteback(page));
1482 if (!clear_page_dirty_for_io(page))
1483 goto continue_unlock;
1485 set_fsync_mark(page, 0);
1486 set_dentry_mark(page, 0);
1488 if (NODE_MAPPING(sbi)->a_ops->writepage(page, wbc))
1491 if (--wbc->nr_to_write == 0)
1494 pagevec_release(&pvec);
1497 if (wbc->nr_to_write == 0) {
1510 int wait_on_node_pages_writeback(struct f2fs_sb_info *sbi, nid_t ino)
1512 pgoff_t index = 0, end = ULONG_MAX;
1513 struct pagevec pvec;
1514 int ret2 = 0, ret = 0;
1516 pagevec_init(&pvec, 0);
1518 while (index <= end) {
1520 nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1521 PAGECACHE_TAG_WRITEBACK,
1522 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
1526 for (i = 0; i < nr_pages; i++) {
1527 struct page *page = pvec.pages[i];
1529 /* until radix tree lookup accepts end_index */
1530 if (unlikely(page->index > end))
1533 if (ino && ino_of_node(page) == ino) {
1534 f2fs_wait_on_page_writeback(page, NODE, true);
1535 if (TestClearPageError(page))
1539 pagevec_release(&pvec);
1543 if (unlikely(test_and_clear_bit(AS_ENOSPC, &NODE_MAPPING(sbi)->flags)))
1545 if (unlikely(test_and_clear_bit(AS_EIO, &NODE_MAPPING(sbi)->flags)))
1552 static int f2fs_write_node_page(struct page *page,
1553 struct writeback_control *wbc)
1555 struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1557 struct node_info ni;
1558 struct f2fs_io_info fio = {
1562 .op_flags = (wbc->sync_mode == WB_SYNC_ALL) ? WRITE_SYNC : 0,
1564 .encrypted_page = NULL,
1567 trace_f2fs_writepage(page, NODE);
1569 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
1571 if (unlikely(f2fs_cp_error(sbi)))
1574 /* get old block addr of this node page */
1575 nid = nid_of_node(page);
1576 f2fs_bug_on(sbi, page->index != nid);
1578 if (wbc->for_reclaim) {
1579 if (!down_read_trylock(&sbi->node_write))
1582 down_read(&sbi->node_write);
1585 get_node_info(sbi, nid, &ni);
1587 /* This page is already truncated */
1588 if (unlikely(ni.blk_addr == NULL_ADDR)) {
1589 ClearPageUptodate(page);
1590 dec_page_count(sbi, F2FS_DIRTY_NODES);
1591 up_read(&sbi->node_write);
1596 set_page_writeback(page);
1597 fio.old_blkaddr = ni.blk_addr;
1598 write_node_page(nid, &fio);
1599 set_node_addr(sbi, &ni, fio.new_blkaddr, is_fsync_dnode(page));
1600 dec_page_count(sbi, F2FS_DIRTY_NODES);
1601 up_read(&sbi->node_write);
1603 if (wbc->for_reclaim)
1604 f2fs_submit_merged_bio_cond(sbi, NULL, page, 0, NODE, WRITE);
1608 if (unlikely(f2fs_cp_error(sbi)))
1609 f2fs_submit_merged_bio(sbi, NODE, WRITE);
1614 redirty_page_for_writepage(wbc, page);
1615 return AOP_WRITEPAGE_ACTIVATE;
1618 static int f2fs_write_node_pages(struct address_space *mapping,
1619 struct writeback_control *wbc)
1621 struct f2fs_sb_info *sbi = F2FS_M_SB(mapping);
1622 struct blk_plug plug;
1625 /* balancing f2fs's metadata in background */
1626 f2fs_balance_fs_bg(sbi);
1628 /* collect a number of dirty node pages and write together */
1629 if (get_pages(sbi, F2FS_DIRTY_NODES) < nr_pages_to_skip(sbi, NODE))
1632 trace_f2fs_writepages(mapping->host, wbc, NODE);
1634 diff = nr_pages_to_write(sbi, NODE, wbc);
1635 wbc->sync_mode = WB_SYNC_NONE;
1636 blk_start_plug(&plug);
1637 sync_node_pages(sbi, wbc);
1638 blk_finish_plug(&plug);
1639 wbc->nr_to_write = max((long)0, wbc->nr_to_write - diff);
1643 wbc->pages_skipped += get_pages(sbi, F2FS_DIRTY_NODES);
1644 trace_f2fs_writepages(mapping->host, wbc, NODE);
1648 static int f2fs_set_node_page_dirty(struct page *page)
1650 trace_f2fs_set_page_dirty(page, NODE);
1652 if (!PageUptodate(page))
1653 SetPageUptodate(page);
1654 if (!PageDirty(page)) {
1655 f2fs_set_page_dirty_nobuffers(page);
1656 inc_page_count(F2FS_P_SB(page), F2FS_DIRTY_NODES);
1657 SetPagePrivate(page);
1658 f2fs_trace_pid(page);
1665 * Structure of the f2fs node operations
1667 const struct address_space_operations f2fs_node_aops = {
1668 .writepage = f2fs_write_node_page,
1669 .writepages = f2fs_write_node_pages,
1670 .set_page_dirty = f2fs_set_node_page_dirty,
1671 .invalidatepage = f2fs_invalidate_page,
1672 .releasepage = f2fs_release_page,
1673 #ifdef CONFIG_MIGRATION
1674 .migratepage = f2fs_migrate_page,
1678 static struct free_nid *__lookup_free_nid_list(struct f2fs_nm_info *nm_i,
1681 return radix_tree_lookup(&nm_i->free_nid_root, n);
1684 static void __del_from_free_nid_list(struct f2fs_nm_info *nm_i,
1688 radix_tree_delete(&nm_i->free_nid_root, i->nid);
1691 static int add_free_nid(struct f2fs_sb_info *sbi, nid_t nid, bool build)
1693 struct f2fs_nm_info *nm_i = NM_I(sbi);
1695 struct nat_entry *ne;
1697 if (!available_free_memory(sbi, FREE_NIDS))
1700 /* 0 nid should not be used */
1701 if (unlikely(nid == 0))
1705 /* do not add allocated nids */
1706 ne = __lookup_nat_cache(nm_i, nid);
1707 if (ne && (!get_nat_flag(ne, IS_CHECKPOINTED) ||
1708 nat_get_blkaddr(ne) != NULL_ADDR))
1712 i = f2fs_kmem_cache_alloc(free_nid_slab, GFP_NOFS);
1716 if (radix_tree_preload(GFP_NOFS)) {
1717 kmem_cache_free(free_nid_slab, i);
1721 spin_lock(&nm_i->free_nid_list_lock);
1722 if (radix_tree_insert(&nm_i->free_nid_root, i->nid, i)) {
1723 spin_unlock(&nm_i->free_nid_list_lock);
1724 radix_tree_preload_end();
1725 kmem_cache_free(free_nid_slab, i);
1728 list_add_tail(&i->list, &nm_i->free_nid_list);
1730 spin_unlock(&nm_i->free_nid_list_lock);
1731 radix_tree_preload_end();
1735 static void remove_free_nid(struct f2fs_nm_info *nm_i, nid_t nid)
1738 bool need_free = false;
1740 spin_lock(&nm_i->free_nid_list_lock);
1741 i = __lookup_free_nid_list(nm_i, nid);
1742 if (i && i->state == NID_NEW) {
1743 __del_from_free_nid_list(nm_i, i);
1747 spin_unlock(&nm_i->free_nid_list_lock);
1750 kmem_cache_free(free_nid_slab, i);
1753 static void scan_nat_page(struct f2fs_sb_info *sbi,
1754 struct page *nat_page, nid_t start_nid)
1756 struct f2fs_nm_info *nm_i = NM_I(sbi);
1757 struct f2fs_nat_block *nat_blk = page_address(nat_page);
1761 i = start_nid % NAT_ENTRY_PER_BLOCK;
1763 for (; i < NAT_ENTRY_PER_BLOCK; i++, start_nid++) {
1765 if (unlikely(start_nid >= nm_i->max_nid))
1768 blk_addr = le32_to_cpu(nat_blk->entries[i].block_addr);
1769 f2fs_bug_on(sbi, blk_addr == NEW_ADDR);
1770 if (blk_addr == NULL_ADDR) {
1771 if (add_free_nid(sbi, start_nid, true) < 0)
1777 void build_free_nids(struct f2fs_sb_info *sbi)
1779 struct f2fs_nm_info *nm_i = NM_I(sbi);
1780 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
1781 struct f2fs_journal *journal = curseg->journal;
1783 nid_t nid = nm_i->next_scan_nid;
1785 /* Enough entries */
1786 if (nm_i->fcnt >= NAT_ENTRY_PER_BLOCK)
1789 /* readahead nat pages to be scanned */
1790 ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nid), FREE_NID_PAGES,
1793 down_read(&nm_i->nat_tree_lock);
1796 struct page *page = get_current_nat_page(sbi, nid);
1798 scan_nat_page(sbi, page, nid);
1799 f2fs_put_page(page, 1);
1801 nid += (NAT_ENTRY_PER_BLOCK - (nid % NAT_ENTRY_PER_BLOCK));
1802 if (unlikely(nid >= nm_i->max_nid))
1805 if (++i >= FREE_NID_PAGES)
1809 /* go to the next free nat pages to find free nids abundantly */
1810 nm_i->next_scan_nid = nid;
1812 /* find free nids from current sum_pages */
1813 down_read(&curseg->journal_rwsem);
1814 for (i = 0; i < nats_in_cursum(journal); i++) {
1817 addr = le32_to_cpu(nat_in_journal(journal, i).block_addr);
1818 nid = le32_to_cpu(nid_in_journal(journal, i));
1819 if (addr == NULL_ADDR)
1820 add_free_nid(sbi, nid, true);
1822 remove_free_nid(nm_i, nid);
1824 up_read(&curseg->journal_rwsem);
1825 up_read(&nm_i->nat_tree_lock);
1827 ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nm_i->next_scan_nid),
1828 nm_i->ra_nid_pages, META_NAT, false);
1832 * If this function returns success, caller can obtain a new nid
1833 * from second parameter of this function.
1834 * The returned nid could be used ino as well as nid when inode is created.
1836 bool alloc_nid(struct f2fs_sb_info *sbi, nid_t *nid)
1838 struct f2fs_nm_info *nm_i = NM_I(sbi);
1839 struct free_nid *i = NULL;
1841 #ifdef CONFIG_F2FS_FAULT_INJECTION
1842 if (time_to_inject(FAULT_ALLOC_NID))
1845 if (unlikely(sbi->total_valid_node_count + 1 > nm_i->available_nids))
1848 spin_lock(&nm_i->free_nid_list_lock);
1850 /* We should not use stale free nids created by build_free_nids */
1851 if (nm_i->fcnt && !on_build_free_nids(nm_i)) {
1852 f2fs_bug_on(sbi, list_empty(&nm_i->free_nid_list));
1853 list_for_each_entry(i, &nm_i->free_nid_list, list)
1854 if (i->state == NID_NEW)
1857 f2fs_bug_on(sbi, i->state != NID_NEW);
1859 i->state = NID_ALLOC;
1861 spin_unlock(&nm_i->free_nid_list_lock);
1864 spin_unlock(&nm_i->free_nid_list_lock);
1866 /* Let's scan nat pages and its caches to get free nids */
1867 mutex_lock(&nm_i->build_lock);
1868 build_free_nids(sbi);
1869 mutex_unlock(&nm_i->build_lock);
1874 * alloc_nid() should be called prior to this function.
1876 void alloc_nid_done(struct f2fs_sb_info *sbi, nid_t nid)
1878 struct f2fs_nm_info *nm_i = NM_I(sbi);
1881 spin_lock(&nm_i->free_nid_list_lock);
1882 i = __lookup_free_nid_list(nm_i, nid);
1883 f2fs_bug_on(sbi, !i || i->state != NID_ALLOC);
1884 __del_from_free_nid_list(nm_i, i);
1885 spin_unlock(&nm_i->free_nid_list_lock);
1887 kmem_cache_free(free_nid_slab, i);
1891 * alloc_nid() should be called prior to this function.
1893 void alloc_nid_failed(struct f2fs_sb_info *sbi, nid_t nid)
1895 struct f2fs_nm_info *nm_i = NM_I(sbi);
1897 bool need_free = false;
1902 spin_lock(&nm_i->free_nid_list_lock);
1903 i = __lookup_free_nid_list(nm_i, nid);
1904 f2fs_bug_on(sbi, !i || i->state != NID_ALLOC);
1905 if (!available_free_memory(sbi, FREE_NIDS)) {
1906 __del_from_free_nid_list(nm_i, i);
1912 spin_unlock(&nm_i->free_nid_list_lock);
1915 kmem_cache_free(free_nid_slab, i);
1918 int try_to_free_nids(struct f2fs_sb_info *sbi, int nr_shrink)
1920 struct f2fs_nm_info *nm_i = NM_I(sbi);
1921 struct free_nid *i, *next;
1924 if (nm_i->fcnt <= MAX_FREE_NIDS)
1927 if (!mutex_trylock(&nm_i->build_lock))
1930 spin_lock(&nm_i->free_nid_list_lock);
1931 list_for_each_entry_safe(i, next, &nm_i->free_nid_list, list) {
1932 if (nr_shrink <= 0 || nm_i->fcnt <= MAX_FREE_NIDS)
1934 if (i->state == NID_ALLOC)
1936 __del_from_free_nid_list(nm_i, i);
1937 kmem_cache_free(free_nid_slab, i);
1941 spin_unlock(&nm_i->free_nid_list_lock);
1942 mutex_unlock(&nm_i->build_lock);
1944 return nr - nr_shrink;
1947 void recover_inline_xattr(struct inode *inode, struct page *page)
1949 void *src_addr, *dst_addr;
1952 struct f2fs_inode *ri;
1954 ipage = get_node_page(F2FS_I_SB(inode), inode->i_ino);
1955 f2fs_bug_on(F2FS_I_SB(inode), IS_ERR(ipage));
1957 ri = F2FS_INODE(page);
1958 if (!(ri->i_inline & F2FS_INLINE_XATTR)) {
1959 clear_inode_flag(inode, FI_INLINE_XATTR);
1963 dst_addr = inline_xattr_addr(ipage);
1964 src_addr = inline_xattr_addr(page);
1965 inline_size = inline_xattr_size(inode);
1967 f2fs_wait_on_page_writeback(ipage, NODE, true);
1968 memcpy(dst_addr, src_addr, inline_size);
1970 update_inode(inode, ipage);
1971 f2fs_put_page(ipage, 1);
1974 void recover_xattr_data(struct inode *inode, struct page *page, block_t blkaddr)
1976 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1977 nid_t prev_xnid = F2FS_I(inode)->i_xattr_nid;
1978 nid_t new_xnid = nid_of_node(page);
1979 struct node_info ni;
1981 /* 1: invalidate the previous xattr nid */
1985 /* Deallocate node address */
1986 get_node_info(sbi, prev_xnid, &ni);
1987 f2fs_bug_on(sbi, ni.blk_addr == NULL_ADDR);
1988 invalidate_blocks(sbi, ni.blk_addr);
1989 dec_valid_node_count(sbi, inode);
1990 set_node_addr(sbi, &ni, NULL_ADDR, false);
1993 /* 2: allocate new xattr nid */
1994 if (unlikely(!inc_valid_node_count(sbi, inode)))
1995 f2fs_bug_on(sbi, 1);
1997 remove_free_nid(NM_I(sbi), new_xnid);
1998 get_node_info(sbi, new_xnid, &ni);
1999 ni.ino = inode->i_ino;
2000 set_node_addr(sbi, &ni, NEW_ADDR, false);
2001 f2fs_i_xnid_write(inode, new_xnid);
2003 /* 3: update xattr blkaddr */
2004 refresh_sit_entry(sbi, NEW_ADDR, blkaddr);
2005 set_node_addr(sbi, &ni, blkaddr, false);
2008 int recover_inode_page(struct f2fs_sb_info *sbi, struct page *page)
2010 struct f2fs_inode *src, *dst;
2011 nid_t ino = ino_of_node(page);
2012 struct node_info old_ni, new_ni;
2015 get_node_info(sbi, ino, &old_ni);
2017 if (unlikely(old_ni.blk_addr != NULL_ADDR))
2020 ipage = f2fs_grab_cache_page(NODE_MAPPING(sbi), ino, false);
2022 congestion_wait(BLK_RW_ASYNC, HZ/50);
2026 /* Should not use this inode from free nid list */
2027 remove_free_nid(NM_I(sbi), ino);
2029 if (!PageUptodate(ipage))
2030 SetPageUptodate(ipage);
2031 fill_node_footer(ipage, ino, ino, 0, true);
2033 src = F2FS_INODE(page);
2034 dst = F2FS_INODE(ipage);
2036 memcpy(dst, src, (unsigned long)&src->i_ext - (unsigned long)src);
2038 dst->i_blocks = cpu_to_le64(1);
2039 dst->i_links = cpu_to_le32(1);
2040 dst->i_xattr_nid = 0;
2041 dst->i_inline = src->i_inline & F2FS_INLINE_XATTR;
2046 if (unlikely(!inc_valid_node_count(sbi, NULL)))
2048 set_node_addr(sbi, &new_ni, NEW_ADDR, false);
2049 inc_valid_inode_count(sbi);
2050 set_page_dirty(ipage);
2051 f2fs_put_page(ipage, 1);
2055 int restore_node_summary(struct f2fs_sb_info *sbi,
2056 unsigned int segno, struct f2fs_summary_block *sum)
2058 struct f2fs_node *rn;
2059 struct f2fs_summary *sum_entry;
2061 int bio_blocks = MAX_BIO_BLOCKS(sbi);
2062 int i, idx, last_offset, nrpages;
2064 /* scan the node segment */
2065 last_offset = sbi->blocks_per_seg;
2066 addr = START_BLOCK(sbi, segno);
2067 sum_entry = &sum->entries[0];
2069 for (i = 0; i < last_offset; i += nrpages, addr += nrpages) {
2070 nrpages = min(last_offset - i, bio_blocks);
2072 /* readahead node pages */
2073 ra_meta_pages(sbi, addr, nrpages, META_POR, true);
2075 for (idx = addr; idx < addr + nrpages; idx++) {
2076 struct page *page = get_tmp_page(sbi, idx);
2078 rn = F2FS_NODE(page);
2079 sum_entry->nid = rn->footer.nid;
2080 sum_entry->version = 0;
2081 sum_entry->ofs_in_node = 0;
2083 f2fs_put_page(page, 1);
2086 invalidate_mapping_pages(META_MAPPING(sbi), addr,
2092 static void remove_nats_in_journal(struct f2fs_sb_info *sbi)
2094 struct f2fs_nm_info *nm_i = NM_I(sbi);
2095 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
2096 struct f2fs_journal *journal = curseg->journal;
2099 down_write(&curseg->journal_rwsem);
2100 for (i = 0; i < nats_in_cursum(journal); i++) {
2101 struct nat_entry *ne;
2102 struct f2fs_nat_entry raw_ne;
2103 nid_t nid = le32_to_cpu(nid_in_journal(journal, i));
2105 raw_ne = nat_in_journal(journal, i);
2107 ne = __lookup_nat_cache(nm_i, nid);
2109 ne = grab_nat_entry(nm_i, nid);
2110 node_info_from_raw_nat(&ne->ni, &raw_ne);
2112 __set_nat_cache_dirty(nm_i, ne);
2114 update_nats_in_cursum(journal, -i);
2115 up_write(&curseg->journal_rwsem);
2118 static void __adjust_nat_entry_set(struct nat_entry_set *nes,
2119 struct list_head *head, int max)
2121 struct nat_entry_set *cur;
2123 if (nes->entry_cnt >= max)
2126 list_for_each_entry(cur, head, set_list) {
2127 if (cur->entry_cnt >= nes->entry_cnt) {
2128 list_add(&nes->set_list, cur->set_list.prev);
2133 list_add_tail(&nes->set_list, head);
2136 static void __flush_nat_entry_set(struct f2fs_sb_info *sbi,
2137 struct nat_entry_set *set)
2139 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
2140 struct f2fs_journal *journal = curseg->journal;
2141 nid_t start_nid = set->set * NAT_ENTRY_PER_BLOCK;
2142 bool to_journal = true;
2143 struct f2fs_nat_block *nat_blk;
2144 struct nat_entry *ne, *cur;
2145 struct page *page = NULL;
2148 * there are two steps to flush nat entries:
2149 * #1, flush nat entries to journal in current hot data summary block.
2150 * #2, flush nat entries to nat page.
2152 if (!__has_cursum_space(journal, set->entry_cnt, NAT_JOURNAL))
2156 down_write(&curseg->journal_rwsem);
2158 page = get_next_nat_page(sbi, start_nid);
2159 nat_blk = page_address(page);
2160 f2fs_bug_on(sbi, !nat_blk);
2163 /* flush dirty nats in nat entry set */
2164 list_for_each_entry_safe(ne, cur, &set->entry_list, list) {
2165 struct f2fs_nat_entry *raw_ne;
2166 nid_t nid = nat_get_nid(ne);
2169 if (nat_get_blkaddr(ne) == NEW_ADDR)
2173 offset = lookup_journal_in_cursum(journal,
2174 NAT_JOURNAL, nid, 1);
2175 f2fs_bug_on(sbi, offset < 0);
2176 raw_ne = &nat_in_journal(journal, offset);
2177 nid_in_journal(journal, offset) = cpu_to_le32(nid);
2179 raw_ne = &nat_blk->entries[nid - start_nid];
2181 raw_nat_from_node_info(raw_ne, &ne->ni);
2183 __clear_nat_cache_dirty(NM_I(sbi), ne);
2184 if (nat_get_blkaddr(ne) == NULL_ADDR)
2185 add_free_nid(sbi, nid, false);
2189 up_write(&curseg->journal_rwsem);
2191 f2fs_put_page(page, 1);
2193 f2fs_bug_on(sbi, set->entry_cnt);
2195 radix_tree_delete(&NM_I(sbi)->nat_set_root, set->set);
2196 kmem_cache_free(nat_entry_set_slab, set);
2200 * This function is called during the checkpointing process.
2202 void flush_nat_entries(struct f2fs_sb_info *sbi)
2204 struct f2fs_nm_info *nm_i = NM_I(sbi);
2205 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
2206 struct f2fs_journal *journal = curseg->journal;
2207 struct nat_entry_set *setvec[SETVEC_SIZE];
2208 struct nat_entry_set *set, *tmp;
2213 if (!nm_i->dirty_nat_cnt)
2216 down_write(&nm_i->nat_tree_lock);
2219 * if there are no enough space in journal to store dirty nat
2220 * entries, remove all entries from journal and merge them
2221 * into nat entry set.
2223 if (!__has_cursum_space(journal, nm_i->dirty_nat_cnt, NAT_JOURNAL))
2224 remove_nats_in_journal(sbi);
2226 while ((found = __gang_lookup_nat_set(nm_i,
2227 set_idx, SETVEC_SIZE, setvec))) {
2229 set_idx = setvec[found - 1]->set + 1;
2230 for (idx = 0; idx < found; idx++)
2231 __adjust_nat_entry_set(setvec[idx], &sets,
2232 MAX_NAT_JENTRIES(journal));
2235 /* flush dirty nats in nat entry set */
2236 list_for_each_entry_safe(set, tmp, &sets, set_list)
2237 __flush_nat_entry_set(sbi, set);
2239 up_write(&nm_i->nat_tree_lock);
2241 f2fs_bug_on(sbi, nm_i->dirty_nat_cnt);
2244 static int init_node_manager(struct f2fs_sb_info *sbi)
2246 struct f2fs_super_block *sb_raw = F2FS_RAW_SUPER(sbi);
2247 struct f2fs_nm_info *nm_i = NM_I(sbi);
2248 unsigned char *version_bitmap;
2249 unsigned int nat_segs, nat_blocks;
2251 nm_i->nat_blkaddr = le32_to_cpu(sb_raw->nat_blkaddr);
2253 /* segment_count_nat includes pair segment so divide to 2. */
2254 nat_segs = le32_to_cpu(sb_raw->segment_count_nat) >> 1;
2255 nat_blocks = nat_segs << le32_to_cpu(sb_raw->log_blocks_per_seg);
2257 nm_i->max_nid = NAT_ENTRY_PER_BLOCK * nat_blocks;
2259 /* not used nids: 0, node, meta, (and root counted as valid node) */
2260 nm_i->available_nids = nm_i->max_nid - F2FS_RESERVED_NODE_NUM;
2263 nm_i->ram_thresh = DEF_RAM_THRESHOLD;
2264 nm_i->ra_nid_pages = DEF_RA_NID_PAGES;
2265 nm_i->dirty_nats_ratio = DEF_DIRTY_NAT_RATIO_THRESHOLD;
2267 INIT_RADIX_TREE(&nm_i->free_nid_root, GFP_ATOMIC);
2268 INIT_LIST_HEAD(&nm_i->free_nid_list);
2269 INIT_RADIX_TREE(&nm_i->nat_root, GFP_NOIO);
2270 INIT_RADIX_TREE(&nm_i->nat_set_root, GFP_NOIO);
2271 INIT_LIST_HEAD(&nm_i->nat_entries);
2273 mutex_init(&nm_i->build_lock);
2274 spin_lock_init(&nm_i->free_nid_list_lock);
2275 init_rwsem(&nm_i->nat_tree_lock);
2277 nm_i->next_scan_nid = le32_to_cpu(sbi->ckpt->next_free_nid);
2278 nm_i->bitmap_size = __bitmap_size(sbi, NAT_BITMAP);
2279 version_bitmap = __bitmap_ptr(sbi, NAT_BITMAP);
2280 if (!version_bitmap)
2283 nm_i->nat_bitmap = kmemdup(version_bitmap, nm_i->bitmap_size,
2285 if (!nm_i->nat_bitmap)
2290 int build_node_manager(struct f2fs_sb_info *sbi)
2294 sbi->nm_info = kzalloc(sizeof(struct f2fs_nm_info), GFP_KERNEL);
2298 err = init_node_manager(sbi);
2302 build_free_nids(sbi);
2306 void destroy_node_manager(struct f2fs_sb_info *sbi)
2308 struct f2fs_nm_info *nm_i = NM_I(sbi);
2309 struct free_nid *i, *next_i;
2310 struct nat_entry *natvec[NATVEC_SIZE];
2311 struct nat_entry_set *setvec[SETVEC_SIZE];
2318 /* destroy free nid list */
2319 spin_lock(&nm_i->free_nid_list_lock);
2320 list_for_each_entry_safe(i, next_i, &nm_i->free_nid_list, list) {
2321 f2fs_bug_on(sbi, i->state == NID_ALLOC);
2322 __del_from_free_nid_list(nm_i, i);
2324 spin_unlock(&nm_i->free_nid_list_lock);
2325 kmem_cache_free(free_nid_slab, i);
2326 spin_lock(&nm_i->free_nid_list_lock);
2328 f2fs_bug_on(sbi, nm_i->fcnt);
2329 spin_unlock(&nm_i->free_nid_list_lock);
2331 /* destroy nat cache */
2332 down_write(&nm_i->nat_tree_lock);
2333 while ((found = __gang_lookup_nat_cache(nm_i,
2334 nid, NATVEC_SIZE, natvec))) {
2337 nid = nat_get_nid(natvec[found - 1]) + 1;
2338 for (idx = 0; idx < found; idx++)
2339 __del_from_nat_cache(nm_i, natvec[idx]);
2341 f2fs_bug_on(sbi, nm_i->nat_cnt);
2343 /* destroy nat set cache */
2345 while ((found = __gang_lookup_nat_set(nm_i,
2346 nid, SETVEC_SIZE, setvec))) {
2349 nid = setvec[found - 1]->set + 1;
2350 for (idx = 0; idx < found; idx++) {
2351 /* entry_cnt is not zero, when cp_error was occurred */
2352 f2fs_bug_on(sbi, !list_empty(&setvec[idx]->entry_list));
2353 radix_tree_delete(&nm_i->nat_set_root, setvec[idx]->set);
2354 kmem_cache_free(nat_entry_set_slab, setvec[idx]);
2357 up_write(&nm_i->nat_tree_lock);
2359 kfree(nm_i->nat_bitmap);
2360 sbi->nm_info = NULL;
2364 int __init create_node_manager_caches(void)
2366 nat_entry_slab = f2fs_kmem_cache_create("nat_entry",
2367 sizeof(struct nat_entry));
2368 if (!nat_entry_slab)
2371 free_nid_slab = f2fs_kmem_cache_create("free_nid",
2372 sizeof(struct free_nid));
2374 goto destroy_nat_entry;
2376 nat_entry_set_slab = f2fs_kmem_cache_create("nat_entry_set",
2377 sizeof(struct nat_entry_set));
2378 if (!nat_entry_set_slab)
2379 goto destroy_free_nid;
2383 kmem_cache_destroy(free_nid_slab);
2385 kmem_cache_destroy(nat_entry_slab);
2390 void destroy_node_manager_caches(void)
2392 kmem_cache_destroy(nat_entry_set_slab);
2393 kmem_cache_destroy(free_nid_slab);
2394 kmem_cache_destroy(nat_entry_slab);