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;
44 /* give 25%, 25%, 50%, 50% memory for each components respectively */
45 if (type == FREE_NIDS) {
46 mem_size = (nm_i->fcnt * sizeof(struct free_nid)) >>
48 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 2);
49 } else if (type == NAT_ENTRIES) {
50 mem_size = (nm_i->nat_cnt * sizeof(struct nat_entry)) >>
52 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 2);
53 } else if (type == DIRTY_DENTS) {
54 if (sbi->sb->s_bdi->dirty_exceeded)
56 mem_size = get_pages(sbi, F2FS_DIRTY_DENTS);
57 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 1);
58 } else if (type == INO_ENTRIES) {
61 for (i = 0; i <= UPDATE_INO; i++)
62 mem_size += (sbi->im[i].ino_num *
63 sizeof(struct ino_entry)) >> PAGE_CACHE_SHIFT;
64 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 1);
66 if (sbi->sb->s_bdi->dirty_exceeded)
72 static void clear_node_page_dirty(struct page *page)
74 struct address_space *mapping = page->mapping;
75 unsigned int long flags;
77 if (PageDirty(page)) {
78 spin_lock_irqsave(&mapping->tree_lock, flags);
79 radix_tree_tag_clear(&mapping->page_tree,
82 spin_unlock_irqrestore(&mapping->tree_lock, flags);
84 clear_page_dirty_for_io(page);
85 dec_page_count(F2FS_M_SB(mapping), F2FS_DIRTY_NODES);
87 ClearPageUptodate(page);
90 static struct page *get_current_nat_page(struct f2fs_sb_info *sbi, nid_t nid)
92 pgoff_t index = current_nat_addr(sbi, nid);
93 return get_meta_page(sbi, index);
96 static struct page *get_next_nat_page(struct f2fs_sb_info *sbi, nid_t nid)
98 struct page *src_page;
99 struct page *dst_page;
104 struct f2fs_nm_info *nm_i = NM_I(sbi);
106 src_off = current_nat_addr(sbi, nid);
107 dst_off = next_nat_addr(sbi, src_off);
109 /* get current nat block page with lock */
110 src_page = get_meta_page(sbi, src_off);
111 dst_page = grab_meta_page(sbi, dst_off);
112 f2fs_bug_on(sbi, PageDirty(src_page));
114 src_addr = page_address(src_page);
115 dst_addr = page_address(dst_page);
116 memcpy(dst_addr, src_addr, PAGE_CACHE_SIZE);
117 set_page_dirty(dst_page);
118 f2fs_put_page(src_page, 1);
120 set_to_next_nat(nm_i, nid);
125 static struct nat_entry *__lookup_nat_cache(struct f2fs_nm_info *nm_i, nid_t n)
127 return radix_tree_lookup(&nm_i->nat_root, n);
130 static unsigned int __gang_lookup_nat_cache(struct f2fs_nm_info *nm_i,
131 nid_t start, unsigned int nr, struct nat_entry **ep)
133 return radix_tree_gang_lookup(&nm_i->nat_root, (void **)ep, start, nr);
136 static void __del_from_nat_cache(struct f2fs_nm_info *nm_i, struct nat_entry *e)
139 radix_tree_delete(&nm_i->nat_root, nat_get_nid(e));
141 kmem_cache_free(nat_entry_slab, e);
144 static void __set_nat_cache_dirty(struct f2fs_nm_info *nm_i,
145 struct nat_entry *ne)
147 nid_t set = NAT_BLOCK_OFFSET(ne->ni.nid);
148 struct nat_entry_set *head;
150 if (get_nat_flag(ne, IS_DIRTY))
153 head = radix_tree_lookup(&nm_i->nat_set_root, set);
155 head = f2fs_kmem_cache_alloc(nat_entry_set_slab, GFP_ATOMIC);
157 INIT_LIST_HEAD(&head->entry_list);
158 INIT_LIST_HEAD(&head->set_list);
161 f2fs_radix_tree_insert(&nm_i->nat_set_root, set, head);
163 list_move_tail(&ne->list, &head->entry_list);
164 nm_i->dirty_nat_cnt++;
166 set_nat_flag(ne, IS_DIRTY, true);
169 static void __clear_nat_cache_dirty(struct f2fs_nm_info *nm_i,
170 struct nat_entry *ne)
172 nid_t set = NAT_BLOCK_OFFSET(ne->ni.nid);
173 struct nat_entry_set *head;
175 head = radix_tree_lookup(&nm_i->nat_set_root, set);
177 list_move_tail(&ne->list, &nm_i->nat_entries);
178 set_nat_flag(ne, IS_DIRTY, false);
180 nm_i->dirty_nat_cnt--;
184 static unsigned int __gang_lookup_nat_set(struct f2fs_nm_info *nm_i,
185 nid_t start, unsigned int nr, struct nat_entry_set **ep)
187 return radix_tree_gang_lookup(&nm_i->nat_set_root, (void **)ep,
191 bool is_checkpointed_node(struct f2fs_sb_info *sbi, nid_t nid)
193 struct f2fs_nm_info *nm_i = NM_I(sbi);
197 down_read(&nm_i->nat_tree_lock);
198 e = __lookup_nat_cache(nm_i, nid);
199 if (e && !get_nat_flag(e, IS_CHECKPOINTED))
201 up_read(&nm_i->nat_tree_lock);
205 bool has_fsynced_inode(struct f2fs_sb_info *sbi, nid_t ino)
207 struct f2fs_nm_info *nm_i = NM_I(sbi);
209 bool fsynced = false;
211 down_read(&nm_i->nat_tree_lock);
212 e = __lookup_nat_cache(nm_i, ino);
213 if (e && get_nat_flag(e, HAS_FSYNCED_INODE))
215 up_read(&nm_i->nat_tree_lock);
219 bool need_inode_block_update(struct f2fs_sb_info *sbi, nid_t ino)
221 struct f2fs_nm_info *nm_i = NM_I(sbi);
223 bool need_update = true;
225 down_read(&nm_i->nat_tree_lock);
226 e = __lookup_nat_cache(nm_i, ino);
227 if (e && get_nat_flag(e, HAS_LAST_FSYNC) &&
228 (get_nat_flag(e, IS_CHECKPOINTED) ||
229 get_nat_flag(e, HAS_FSYNCED_INODE)))
231 up_read(&nm_i->nat_tree_lock);
235 static struct nat_entry *grab_nat_entry(struct f2fs_nm_info *nm_i, nid_t nid)
237 struct nat_entry *new;
239 new = f2fs_kmem_cache_alloc(nat_entry_slab, GFP_ATOMIC);
240 f2fs_radix_tree_insert(&nm_i->nat_root, nid, new);
241 memset(new, 0, sizeof(struct nat_entry));
242 nat_set_nid(new, nid);
244 list_add_tail(&new->list, &nm_i->nat_entries);
249 static void cache_nat_entry(struct f2fs_nm_info *nm_i, nid_t nid,
250 struct f2fs_nat_entry *ne)
254 down_write(&nm_i->nat_tree_lock);
255 e = __lookup_nat_cache(nm_i, nid);
257 e = grab_nat_entry(nm_i, nid);
258 node_info_from_raw_nat(&e->ni, ne);
260 up_write(&nm_i->nat_tree_lock);
263 static void set_node_addr(struct f2fs_sb_info *sbi, struct node_info *ni,
264 block_t new_blkaddr, bool fsync_done)
266 struct f2fs_nm_info *nm_i = NM_I(sbi);
269 down_write(&nm_i->nat_tree_lock);
270 e = __lookup_nat_cache(nm_i, ni->nid);
272 e = grab_nat_entry(nm_i, ni->nid);
273 copy_node_info(&e->ni, ni);
274 f2fs_bug_on(sbi, ni->blk_addr == NEW_ADDR);
275 } else if (new_blkaddr == NEW_ADDR) {
277 * when nid is reallocated,
278 * previous nat entry can be remained in nat cache.
279 * So, reinitialize it with new information.
281 copy_node_info(&e->ni, ni);
282 f2fs_bug_on(sbi, ni->blk_addr != NULL_ADDR);
286 f2fs_bug_on(sbi, nat_get_blkaddr(e) != ni->blk_addr);
287 f2fs_bug_on(sbi, nat_get_blkaddr(e) == NULL_ADDR &&
288 new_blkaddr == NULL_ADDR);
289 f2fs_bug_on(sbi, nat_get_blkaddr(e) == NEW_ADDR &&
290 new_blkaddr == NEW_ADDR);
291 f2fs_bug_on(sbi, nat_get_blkaddr(e) != NEW_ADDR &&
292 nat_get_blkaddr(e) != NULL_ADDR &&
293 new_blkaddr == NEW_ADDR);
295 /* increment version no as node is removed */
296 if (nat_get_blkaddr(e) != NEW_ADDR && new_blkaddr == NULL_ADDR) {
297 unsigned char version = nat_get_version(e);
298 nat_set_version(e, inc_node_version(version));
302 nat_set_blkaddr(e, new_blkaddr);
303 if (new_blkaddr == NEW_ADDR || new_blkaddr == NULL_ADDR)
304 set_nat_flag(e, IS_CHECKPOINTED, false);
305 __set_nat_cache_dirty(nm_i, e);
307 /* update fsync_mark if its inode nat entry is still alive */
308 e = __lookup_nat_cache(nm_i, ni->ino);
310 if (fsync_done && ni->nid == ni->ino)
311 set_nat_flag(e, HAS_FSYNCED_INODE, true);
312 set_nat_flag(e, HAS_LAST_FSYNC, fsync_done);
314 up_write(&nm_i->nat_tree_lock);
317 int try_to_free_nats(struct f2fs_sb_info *sbi, int nr_shrink)
319 struct f2fs_nm_info *nm_i = NM_I(sbi);
321 if (available_free_memory(sbi, NAT_ENTRIES))
324 down_write(&nm_i->nat_tree_lock);
325 while (nr_shrink && !list_empty(&nm_i->nat_entries)) {
326 struct nat_entry *ne;
327 ne = list_first_entry(&nm_i->nat_entries,
328 struct nat_entry, list);
329 __del_from_nat_cache(nm_i, ne);
332 up_write(&nm_i->nat_tree_lock);
337 * This function always returns success
339 void get_node_info(struct f2fs_sb_info *sbi, nid_t nid, struct node_info *ni)
341 struct f2fs_nm_info *nm_i = NM_I(sbi);
342 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
343 struct f2fs_summary_block *sum = curseg->sum_blk;
344 nid_t start_nid = START_NID(nid);
345 struct f2fs_nat_block *nat_blk;
346 struct page *page = NULL;
347 struct f2fs_nat_entry ne;
353 /* Check nat cache */
354 down_read(&nm_i->nat_tree_lock);
355 e = __lookup_nat_cache(nm_i, nid);
357 ni->ino = nat_get_ino(e);
358 ni->blk_addr = nat_get_blkaddr(e);
359 ni->version = nat_get_version(e);
361 up_read(&nm_i->nat_tree_lock);
365 memset(&ne, 0, sizeof(struct f2fs_nat_entry));
367 /* Check current segment summary */
368 mutex_lock(&curseg->curseg_mutex);
369 i = lookup_journal_in_cursum(sum, NAT_JOURNAL, nid, 0);
371 ne = nat_in_journal(sum, i);
372 node_info_from_raw_nat(ni, &ne);
374 mutex_unlock(&curseg->curseg_mutex);
378 /* Fill node_info from nat page */
379 page = get_current_nat_page(sbi, start_nid);
380 nat_blk = (struct f2fs_nat_block *)page_address(page);
381 ne = nat_blk->entries[nid - start_nid];
382 node_info_from_raw_nat(ni, &ne);
383 f2fs_put_page(page, 1);
385 /* cache nat entry */
386 cache_nat_entry(NM_I(sbi), nid, &ne);
390 * The maximum depth is four.
391 * Offset[0] will have raw inode offset.
393 static int get_node_path(struct f2fs_inode_info *fi, long block,
394 int offset[4], unsigned int noffset[4])
396 const long direct_index = ADDRS_PER_INODE(fi);
397 const long direct_blks = ADDRS_PER_BLOCK;
398 const long dptrs_per_blk = NIDS_PER_BLOCK;
399 const long indirect_blks = ADDRS_PER_BLOCK * NIDS_PER_BLOCK;
400 const long dindirect_blks = indirect_blks * NIDS_PER_BLOCK;
406 if (block < direct_index) {
410 block -= direct_index;
411 if (block < direct_blks) {
412 offset[n++] = NODE_DIR1_BLOCK;
418 block -= direct_blks;
419 if (block < direct_blks) {
420 offset[n++] = NODE_DIR2_BLOCK;
426 block -= direct_blks;
427 if (block < indirect_blks) {
428 offset[n++] = NODE_IND1_BLOCK;
430 offset[n++] = block / direct_blks;
431 noffset[n] = 4 + offset[n - 1];
432 offset[n] = block % direct_blks;
436 block -= indirect_blks;
437 if (block < indirect_blks) {
438 offset[n++] = NODE_IND2_BLOCK;
439 noffset[n] = 4 + dptrs_per_blk;
440 offset[n++] = block / direct_blks;
441 noffset[n] = 5 + dptrs_per_blk + offset[n - 1];
442 offset[n] = block % direct_blks;
446 block -= indirect_blks;
447 if (block < dindirect_blks) {
448 offset[n++] = NODE_DIND_BLOCK;
449 noffset[n] = 5 + (dptrs_per_blk * 2);
450 offset[n++] = block / indirect_blks;
451 noffset[n] = 6 + (dptrs_per_blk * 2) +
452 offset[n - 1] * (dptrs_per_blk + 1);
453 offset[n++] = (block / direct_blks) % dptrs_per_blk;
454 noffset[n] = 7 + (dptrs_per_blk * 2) +
455 offset[n - 2] * (dptrs_per_blk + 1) +
457 offset[n] = block % direct_blks;
468 * Caller should call f2fs_put_dnode(dn).
469 * Also, it should grab and release a rwsem by calling f2fs_lock_op() and
470 * f2fs_unlock_op() only if ro is not set RDONLY_NODE.
471 * In the case of RDONLY_NODE, we don't need to care about mutex.
473 int get_dnode_of_data(struct dnode_of_data *dn, pgoff_t index, int mode)
475 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
476 struct page *npage[4];
477 struct page *parent = NULL;
479 unsigned int noffset[4];
484 level = get_node_path(F2FS_I(dn->inode), index, offset, noffset);
486 nids[0] = dn->inode->i_ino;
487 npage[0] = dn->inode_page;
490 npage[0] = get_node_page(sbi, nids[0]);
491 if (IS_ERR(npage[0]))
492 return PTR_ERR(npage[0]);
495 /* if inline_data is set, should not report any block indices */
496 if (f2fs_has_inline_data(dn->inode) && index) {
498 f2fs_put_page(npage[0], 1);
504 nids[1] = get_nid(parent, offset[0], true);
505 dn->inode_page = npage[0];
506 dn->inode_page_locked = true;
508 /* get indirect or direct nodes */
509 for (i = 1; i <= level; i++) {
512 if (!nids[i] && mode == ALLOC_NODE) {
514 if (!alloc_nid(sbi, &(nids[i]))) {
520 npage[i] = new_node_page(dn, noffset[i], NULL);
521 if (IS_ERR(npage[i])) {
522 alloc_nid_failed(sbi, nids[i]);
523 err = PTR_ERR(npage[i]);
527 set_nid(parent, offset[i - 1], nids[i], i == 1);
528 alloc_nid_done(sbi, nids[i]);
530 } else if (mode == LOOKUP_NODE_RA && i == level && level > 1) {
531 npage[i] = get_node_page_ra(parent, offset[i - 1]);
532 if (IS_ERR(npage[i])) {
533 err = PTR_ERR(npage[i]);
539 dn->inode_page_locked = false;
542 f2fs_put_page(parent, 1);
546 npage[i] = get_node_page(sbi, nids[i]);
547 if (IS_ERR(npage[i])) {
548 err = PTR_ERR(npage[i]);
549 f2fs_put_page(npage[0], 0);
555 nids[i + 1] = get_nid(parent, offset[i], false);
558 dn->nid = nids[level];
559 dn->ofs_in_node = offset[level];
560 dn->node_page = npage[level];
561 dn->data_blkaddr = datablock_addr(dn->node_page, dn->ofs_in_node);
565 f2fs_put_page(parent, 1);
567 f2fs_put_page(npage[0], 0);
569 dn->inode_page = NULL;
570 dn->node_page = NULL;
574 static void truncate_node(struct dnode_of_data *dn)
576 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
579 get_node_info(sbi, dn->nid, &ni);
580 if (dn->inode->i_blocks == 0) {
581 f2fs_bug_on(sbi, ni.blk_addr != NULL_ADDR);
584 f2fs_bug_on(sbi, ni.blk_addr == NULL_ADDR);
586 /* Deallocate node address */
587 invalidate_blocks(sbi, ni.blk_addr);
588 dec_valid_node_count(sbi, dn->inode);
589 set_node_addr(sbi, &ni, NULL_ADDR, false);
591 if (dn->nid == dn->inode->i_ino) {
592 remove_orphan_inode(sbi, dn->nid);
593 dec_valid_inode_count(sbi);
598 clear_node_page_dirty(dn->node_page);
599 set_sbi_flag(sbi, SBI_IS_DIRTY);
601 f2fs_put_page(dn->node_page, 1);
603 invalidate_mapping_pages(NODE_MAPPING(sbi),
604 dn->node_page->index, dn->node_page->index);
606 dn->node_page = NULL;
607 trace_f2fs_truncate_node(dn->inode, dn->nid, ni.blk_addr);
610 static int truncate_dnode(struct dnode_of_data *dn)
617 /* get direct node */
618 page = get_node_page(F2FS_I_SB(dn->inode), dn->nid);
619 if (IS_ERR(page) && PTR_ERR(page) == -ENOENT)
621 else if (IS_ERR(page))
622 return PTR_ERR(page);
624 /* Make dnode_of_data for parameter */
625 dn->node_page = page;
627 truncate_data_blocks(dn);
632 static int truncate_nodes(struct dnode_of_data *dn, unsigned int nofs,
635 struct dnode_of_data rdn = *dn;
637 struct f2fs_node *rn;
639 unsigned int child_nofs;
644 return NIDS_PER_BLOCK + 1;
646 trace_f2fs_truncate_nodes_enter(dn->inode, dn->nid, dn->data_blkaddr);
648 page = get_node_page(F2FS_I_SB(dn->inode), dn->nid);
650 trace_f2fs_truncate_nodes_exit(dn->inode, PTR_ERR(page));
651 return PTR_ERR(page);
654 rn = F2FS_NODE(page);
656 for (i = ofs; i < NIDS_PER_BLOCK; i++, freed++) {
657 child_nid = le32_to_cpu(rn->in.nid[i]);
661 ret = truncate_dnode(&rdn);
664 set_nid(page, i, 0, false);
667 child_nofs = nofs + ofs * (NIDS_PER_BLOCK + 1) + 1;
668 for (i = ofs; i < NIDS_PER_BLOCK; i++) {
669 child_nid = le32_to_cpu(rn->in.nid[i]);
670 if (child_nid == 0) {
671 child_nofs += NIDS_PER_BLOCK + 1;
675 ret = truncate_nodes(&rdn, child_nofs, 0, depth - 1);
676 if (ret == (NIDS_PER_BLOCK + 1)) {
677 set_nid(page, i, 0, false);
679 } else if (ret < 0 && ret != -ENOENT) {
687 /* remove current indirect node */
688 dn->node_page = page;
692 f2fs_put_page(page, 1);
694 trace_f2fs_truncate_nodes_exit(dn->inode, freed);
698 f2fs_put_page(page, 1);
699 trace_f2fs_truncate_nodes_exit(dn->inode, ret);
703 static int truncate_partial_nodes(struct dnode_of_data *dn,
704 struct f2fs_inode *ri, int *offset, int depth)
706 struct page *pages[2];
713 nid[0] = le32_to_cpu(ri->i_nid[offset[0] - NODE_DIR1_BLOCK]);
717 /* get indirect nodes in the path */
718 for (i = 0; i < idx + 1; i++) {
719 /* reference count'll be increased */
720 pages[i] = get_node_page(F2FS_I_SB(dn->inode), nid[i]);
721 if (IS_ERR(pages[i])) {
722 err = PTR_ERR(pages[i]);
726 nid[i + 1] = get_nid(pages[i], offset[i + 1], false);
729 /* free direct nodes linked to a partial indirect node */
730 for (i = offset[idx + 1]; i < NIDS_PER_BLOCK; i++) {
731 child_nid = get_nid(pages[idx], i, false);
735 err = truncate_dnode(dn);
738 set_nid(pages[idx], i, 0, false);
741 if (offset[idx + 1] == 0) {
742 dn->node_page = pages[idx];
746 f2fs_put_page(pages[idx], 1);
752 for (i = idx; i >= 0; i--)
753 f2fs_put_page(pages[i], 1);
755 trace_f2fs_truncate_partial_nodes(dn->inode, nid, depth, err);
761 * All the block addresses of data and nodes should be nullified.
763 int truncate_inode_blocks(struct inode *inode, pgoff_t from)
765 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
766 int err = 0, cont = 1;
767 int level, offset[4], noffset[4];
768 unsigned int nofs = 0;
769 struct f2fs_inode *ri;
770 struct dnode_of_data dn;
773 trace_f2fs_truncate_inode_blocks_enter(inode, from);
775 level = get_node_path(F2FS_I(inode), from, offset, noffset);
777 page = get_node_page(sbi, inode->i_ino);
779 trace_f2fs_truncate_inode_blocks_exit(inode, PTR_ERR(page));
780 return PTR_ERR(page);
783 set_new_dnode(&dn, inode, page, NULL, 0);
786 ri = F2FS_INODE(page);
794 if (!offset[level - 1])
796 err = truncate_partial_nodes(&dn, ri, offset, level);
797 if (err < 0 && err != -ENOENT)
799 nofs += 1 + NIDS_PER_BLOCK;
802 nofs = 5 + 2 * NIDS_PER_BLOCK;
803 if (!offset[level - 1])
805 err = truncate_partial_nodes(&dn, ri, offset, level);
806 if (err < 0 && err != -ENOENT)
815 dn.nid = le32_to_cpu(ri->i_nid[offset[0] - NODE_DIR1_BLOCK]);
817 case NODE_DIR1_BLOCK:
818 case NODE_DIR2_BLOCK:
819 err = truncate_dnode(&dn);
822 case NODE_IND1_BLOCK:
823 case NODE_IND2_BLOCK:
824 err = truncate_nodes(&dn, nofs, offset[1], 2);
827 case NODE_DIND_BLOCK:
828 err = truncate_nodes(&dn, nofs, offset[1], 3);
835 if (err < 0 && err != -ENOENT)
837 if (offset[1] == 0 &&
838 ri->i_nid[offset[0] - NODE_DIR1_BLOCK]) {
840 if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
841 f2fs_put_page(page, 1);
844 f2fs_wait_on_page_writeback(page, NODE);
845 ri->i_nid[offset[0] - NODE_DIR1_BLOCK] = 0;
846 set_page_dirty(page);
854 f2fs_put_page(page, 0);
855 trace_f2fs_truncate_inode_blocks_exit(inode, err);
856 return err > 0 ? 0 : err;
859 int truncate_xattr_node(struct inode *inode, struct page *page)
861 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
862 nid_t nid = F2FS_I(inode)->i_xattr_nid;
863 struct dnode_of_data dn;
869 npage = get_node_page(sbi, nid);
871 return PTR_ERR(npage);
873 F2FS_I(inode)->i_xattr_nid = 0;
875 /* need to do checkpoint during fsync */
876 F2FS_I(inode)->xattr_ver = cur_cp_version(F2FS_CKPT(sbi));
878 set_new_dnode(&dn, inode, page, npage, nid);
881 dn.inode_page_locked = true;
887 * Caller should grab and release a rwsem by calling f2fs_lock_op() and
890 void remove_inode_page(struct inode *inode)
892 struct dnode_of_data dn;
894 set_new_dnode(&dn, inode, NULL, NULL, inode->i_ino);
895 if (get_dnode_of_data(&dn, 0, LOOKUP_NODE))
898 if (truncate_xattr_node(inode, dn.inode_page)) {
903 /* remove potential inline_data blocks */
904 if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
905 S_ISLNK(inode->i_mode))
906 truncate_data_blocks_range(&dn, 1);
908 /* 0 is possible, after f2fs_new_inode() has failed */
909 f2fs_bug_on(F2FS_I_SB(inode),
910 inode->i_blocks != 0 && inode->i_blocks != 1);
912 /* will put inode & node pages */
916 struct page *new_inode_page(struct inode *inode)
918 struct dnode_of_data dn;
920 /* allocate inode page for new inode */
921 set_new_dnode(&dn, inode, NULL, NULL, inode->i_ino);
923 /* caller should f2fs_put_page(page, 1); */
924 return new_node_page(&dn, 0, NULL);
927 struct page *new_node_page(struct dnode_of_data *dn,
928 unsigned int ofs, struct page *ipage)
930 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
931 struct node_info old_ni, new_ni;
935 if (unlikely(is_inode_flag_set(F2FS_I(dn->inode), FI_NO_ALLOC)))
936 return ERR_PTR(-EPERM);
938 page = grab_cache_page(NODE_MAPPING(sbi), dn->nid);
940 return ERR_PTR(-ENOMEM);
942 if (unlikely(!inc_valid_node_count(sbi, dn->inode))) {
947 get_node_info(sbi, dn->nid, &old_ni);
949 /* Reinitialize old_ni with new node page */
950 f2fs_bug_on(sbi, old_ni.blk_addr != NULL_ADDR);
952 new_ni.ino = dn->inode->i_ino;
953 set_node_addr(sbi, &new_ni, NEW_ADDR, false);
955 f2fs_wait_on_page_writeback(page, NODE);
956 fill_node_footer(page, dn->nid, dn->inode->i_ino, ofs, true);
957 set_cold_node(dn->inode, page);
958 SetPageUptodate(page);
959 set_page_dirty(page);
961 if (f2fs_has_xattr_block(ofs))
962 F2FS_I(dn->inode)->i_xattr_nid = dn->nid;
964 dn->node_page = page;
966 update_inode(dn->inode, ipage);
970 inc_valid_inode_count(sbi);
975 clear_node_page_dirty(page);
976 f2fs_put_page(page, 1);
981 * Caller should do after getting the following values.
982 * 0: f2fs_put_page(page, 0)
983 * LOCKED_PAGE: f2fs_put_page(page, 1)
986 static int read_node_page(struct page *page, int rw)
988 struct f2fs_sb_info *sbi = F2FS_P_SB(page);
990 struct f2fs_io_info fio = {
995 get_node_info(sbi, page->index, &ni);
997 if (unlikely(ni.blk_addr == NULL_ADDR)) {
998 f2fs_put_page(page, 1);
1002 if (PageUptodate(page))
1005 fio.blk_addr = ni.blk_addr;
1006 return f2fs_submit_page_bio(sbi, page, &fio);
1010 * Readahead a node page
1012 void ra_node_page(struct f2fs_sb_info *sbi, nid_t nid)
1017 apage = find_get_page(NODE_MAPPING(sbi), nid);
1018 if (apage && PageUptodate(apage)) {
1019 f2fs_put_page(apage, 0);
1022 f2fs_put_page(apage, 0);
1024 apage = grab_cache_page(NODE_MAPPING(sbi), nid);
1028 err = read_node_page(apage, READA);
1030 f2fs_put_page(apage, 0);
1031 else if (err == LOCKED_PAGE)
1032 f2fs_put_page(apage, 1);
1035 struct page *get_node_page(struct f2fs_sb_info *sbi, pgoff_t nid)
1040 page = grab_cache_page(NODE_MAPPING(sbi), nid);
1042 return ERR_PTR(-ENOMEM);
1044 err = read_node_page(page, READ_SYNC);
1046 return ERR_PTR(err);
1047 else if (err != LOCKED_PAGE)
1050 if (unlikely(!PageUptodate(page) || nid != nid_of_node(page))) {
1051 ClearPageUptodate(page);
1052 f2fs_put_page(page, 1);
1053 return ERR_PTR(-EIO);
1055 if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1056 f2fs_put_page(page, 1);
1063 * Return a locked page for the desired node page.
1064 * And, readahead MAX_RA_NODE number of node pages.
1066 struct page *get_node_page_ra(struct page *parent, int start)
1068 struct f2fs_sb_info *sbi = F2FS_P_SB(parent);
1069 struct blk_plug plug;
1074 /* First, try getting the desired direct node. */
1075 nid = get_nid(parent, start, false);
1077 return ERR_PTR(-ENOENT);
1079 page = grab_cache_page(NODE_MAPPING(sbi), nid);
1081 return ERR_PTR(-ENOMEM);
1083 err = read_node_page(page, READ_SYNC);
1085 return ERR_PTR(err);
1086 else if (err == LOCKED_PAGE)
1089 blk_start_plug(&plug);
1091 /* Then, try readahead for siblings of the desired node */
1092 end = start + MAX_RA_NODE;
1093 end = min(end, NIDS_PER_BLOCK);
1094 for (i = start + 1; i < end; i++) {
1095 nid = get_nid(parent, i, false);
1098 ra_node_page(sbi, nid);
1101 blk_finish_plug(&plug);
1104 if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1105 f2fs_put_page(page, 1);
1109 if (unlikely(!PageUptodate(page))) {
1110 f2fs_put_page(page, 1);
1111 return ERR_PTR(-EIO);
1116 void sync_inode_page(struct dnode_of_data *dn)
1118 if (IS_INODE(dn->node_page) || dn->inode_page == dn->node_page) {
1119 update_inode(dn->inode, dn->node_page);
1120 } else if (dn->inode_page) {
1121 if (!dn->inode_page_locked)
1122 lock_page(dn->inode_page);
1123 update_inode(dn->inode, dn->inode_page);
1124 if (!dn->inode_page_locked)
1125 unlock_page(dn->inode_page);
1127 update_inode_page(dn->inode);
1131 int sync_node_pages(struct f2fs_sb_info *sbi, nid_t ino,
1132 struct writeback_control *wbc)
1135 struct pagevec pvec;
1136 int step = ino ? 2 : 0;
1137 int nwritten = 0, wrote = 0;
1139 pagevec_init(&pvec, 0);
1145 while (index <= end) {
1147 nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1148 PAGECACHE_TAG_DIRTY,
1149 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
1153 for (i = 0; i < nr_pages; i++) {
1154 struct page *page = pvec.pages[i];
1157 * flushing sequence with step:
1162 if (step == 0 && IS_DNODE(page))
1164 if (step == 1 && (!IS_DNODE(page) ||
1165 is_cold_node(page)))
1167 if (step == 2 && (!IS_DNODE(page) ||
1168 !is_cold_node(page)))
1173 * we should not skip writing node pages.
1175 if (ino && ino_of_node(page) == ino)
1177 else if (!trylock_page(page))
1180 if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1185 if (ino && ino_of_node(page) != ino)
1186 goto continue_unlock;
1188 if (!PageDirty(page)) {
1189 /* someone wrote it for us */
1190 goto continue_unlock;
1193 if (!clear_page_dirty_for_io(page))
1194 goto continue_unlock;
1196 /* called by fsync() */
1197 if (ino && IS_DNODE(page)) {
1198 set_fsync_mark(page, 1);
1199 if (IS_INODE(page)) {
1200 if (!is_checkpointed_node(sbi, ino) &&
1201 !has_fsynced_inode(sbi, ino))
1202 set_dentry_mark(page, 1);
1204 set_dentry_mark(page, 0);
1208 set_fsync_mark(page, 0);
1209 set_dentry_mark(page, 0);
1212 if (NODE_MAPPING(sbi)->a_ops->writepage(page, wbc))
1217 if (--wbc->nr_to_write == 0)
1220 pagevec_release(&pvec);
1223 if (wbc->nr_to_write == 0) {
1235 f2fs_submit_merged_bio(sbi, NODE, WRITE);
1239 int wait_on_node_pages_writeback(struct f2fs_sb_info *sbi, nid_t ino)
1241 pgoff_t index = 0, end = LONG_MAX;
1242 struct pagevec pvec;
1243 int ret2 = 0, ret = 0;
1245 pagevec_init(&pvec, 0);
1247 while (index <= end) {
1249 nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1250 PAGECACHE_TAG_WRITEBACK,
1251 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
1255 for (i = 0; i < nr_pages; i++) {
1256 struct page *page = pvec.pages[i];
1258 /* until radix tree lookup accepts end_index */
1259 if (unlikely(page->index > end))
1262 if (ino && ino_of_node(page) == ino) {
1263 f2fs_wait_on_page_writeback(page, NODE);
1264 if (TestClearPageError(page))
1268 pagevec_release(&pvec);
1272 if (unlikely(test_and_clear_bit(AS_ENOSPC, &NODE_MAPPING(sbi)->flags)))
1274 if (unlikely(test_and_clear_bit(AS_EIO, &NODE_MAPPING(sbi)->flags)))
1281 static int f2fs_write_node_page(struct page *page,
1282 struct writeback_control *wbc)
1284 struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1286 struct node_info ni;
1287 struct f2fs_io_info fio = {
1289 .rw = (wbc->sync_mode == WB_SYNC_ALL) ? WRITE_SYNC : WRITE,
1292 trace_f2fs_writepage(page, NODE);
1294 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
1296 if (unlikely(f2fs_cp_error(sbi)))
1299 f2fs_wait_on_page_writeback(page, NODE);
1301 /* get old block addr of this node page */
1302 nid = nid_of_node(page);
1303 f2fs_bug_on(sbi, page->index != nid);
1305 get_node_info(sbi, nid, &ni);
1307 /* This page is already truncated */
1308 if (unlikely(ni.blk_addr == NULL_ADDR)) {
1309 dec_page_count(sbi, F2FS_DIRTY_NODES);
1314 if (wbc->for_reclaim) {
1315 if (!down_read_trylock(&sbi->node_write))
1318 down_read(&sbi->node_write);
1321 set_page_writeback(page);
1322 fio.blk_addr = ni.blk_addr;
1323 write_node_page(sbi, page, nid, &fio);
1324 set_node_addr(sbi, &ni, fio.blk_addr, is_fsync_dnode(page));
1325 dec_page_count(sbi, F2FS_DIRTY_NODES);
1326 up_read(&sbi->node_write);
1329 if (wbc->for_reclaim)
1330 f2fs_submit_merged_bio(sbi, NODE, WRITE);
1335 redirty_page_for_writepage(wbc, page);
1336 return AOP_WRITEPAGE_ACTIVATE;
1339 static int f2fs_write_node_pages(struct address_space *mapping,
1340 struct writeback_control *wbc)
1342 struct f2fs_sb_info *sbi = F2FS_M_SB(mapping);
1345 trace_f2fs_writepages(mapping->host, wbc, NODE);
1347 /* balancing f2fs's metadata in background */
1348 f2fs_balance_fs_bg(sbi);
1350 /* collect a number of dirty node pages and write together */
1351 if (get_pages(sbi, F2FS_DIRTY_NODES) < nr_pages_to_skip(sbi, NODE))
1354 diff = nr_pages_to_write(sbi, NODE, wbc);
1355 wbc->sync_mode = WB_SYNC_NONE;
1356 sync_node_pages(sbi, 0, wbc);
1357 wbc->nr_to_write = max((long)0, wbc->nr_to_write - diff);
1361 wbc->pages_skipped += get_pages(sbi, F2FS_DIRTY_NODES);
1365 static int f2fs_set_node_page_dirty(struct page *page)
1367 trace_f2fs_set_page_dirty(page, NODE);
1369 SetPageUptodate(page);
1370 if (!PageDirty(page)) {
1371 __set_page_dirty_nobuffers(page);
1372 inc_page_count(F2FS_P_SB(page), F2FS_DIRTY_NODES);
1373 SetPagePrivate(page);
1374 f2fs_trace_pid(page);
1381 * Structure of the f2fs node operations
1383 const struct address_space_operations f2fs_node_aops = {
1384 .writepage = f2fs_write_node_page,
1385 .writepages = f2fs_write_node_pages,
1386 .set_page_dirty = f2fs_set_node_page_dirty,
1387 .invalidatepage = f2fs_invalidate_page,
1388 .releasepage = f2fs_release_page,
1391 static struct free_nid *__lookup_free_nid_list(struct f2fs_nm_info *nm_i,
1394 return radix_tree_lookup(&nm_i->free_nid_root, n);
1397 static void __del_from_free_nid_list(struct f2fs_nm_info *nm_i,
1401 radix_tree_delete(&nm_i->free_nid_root, i->nid);
1404 static int add_free_nid(struct f2fs_sb_info *sbi, nid_t nid, bool build)
1406 struct f2fs_nm_info *nm_i = NM_I(sbi);
1408 struct nat_entry *ne;
1409 bool allocated = false;
1411 if (!available_free_memory(sbi, FREE_NIDS))
1414 /* 0 nid should not be used */
1415 if (unlikely(nid == 0))
1419 /* do not add allocated nids */
1420 down_read(&nm_i->nat_tree_lock);
1421 ne = __lookup_nat_cache(nm_i, nid);
1423 (!get_nat_flag(ne, IS_CHECKPOINTED) ||
1424 nat_get_blkaddr(ne) != NULL_ADDR))
1426 up_read(&nm_i->nat_tree_lock);
1431 i = f2fs_kmem_cache_alloc(free_nid_slab, GFP_NOFS);
1435 if (radix_tree_preload(GFP_NOFS)) {
1436 kmem_cache_free(free_nid_slab, i);
1440 spin_lock(&nm_i->free_nid_list_lock);
1441 if (radix_tree_insert(&nm_i->free_nid_root, i->nid, i)) {
1442 spin_unlock(&nm_i->free_nid_list_lock);
1443 radix_tree_preload_end();
1444 kmem_cache_free(free_nid_slab, i);
1447 list_add_tail(&i->list, &nm_i->free_nid_list);
1449 spin_unlock(&nm_i->free_nid_list_lock);
1450 radix_tree_preload_end();
1454 static void remove_free_nid(struct f2fs_nm_info *nm_i, nid_t nid)
1457 bool need_free = false;
1459 spin_lock(&nm_i->free_nid_list_lock);
1460 i = __lookup_free_nid_list(nm_i, nid);
1461 if (i && i->state == NID_NEW) {
1462 __del_from_free_nid_list(nm_i, i);
1466 spin_unlock(&nm_i->free_nid_list_lock);
1469 kmem_cache_free(free_nid_slab, i);
1472 static void scan_nat_page(struct f2fs_sb_info *sbi,
1473 struct page *nat_page, nid_t start_nid)
1475 struct f2fs_nm_info *nm_i = NM_I(sbi);
1476 struct f2fs_nat_block *nat_blk = page_address(nat_page);
1480 i = start_nid % NAT_ENTRY_PER_BLOCK;
1482 for (; i < NAT_ENTRY_PER_BLOCK; i++, start_nid++) {
1484 if (unlikely(start_nid >= nm_i->max_nid))
1487 blk_addr = le32_to_cpu(nat_blk->entries[i].block_addr);
1488 f2fs_bug_on(sbi, blk_addr == NEW_ADDR);
1489 if (blk_addr == NULL_ADDR) {
1490 if (add_free_nid(sbi, start_nid, true) < 0)
1496 static void build_free_nids(struct f2fs_sb_info *sbi)
1498 struct f2fs_nm_info *nm_i = NM_I(sbi);
1499 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
1500 struct f2fs_summary_block *sum = curseg->sum_blk;
1502 nid_t nid = nm_i->next_scan_nid;
1504 /* Enough entries */
1505 if (nm_i->fcnt > NAT_ENTRY_PER_BLOCK)
1508 /* readahead nat pages to be scanned */
1509 ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nid), FREE_NID_PAGES, META_NAT);
1512 struct page *page = get_current_nat_page(sbi, nid);
1514 scan_nat_page(sbi, page, nid);
1515 f2fs_put_page(page, 1);
1517 nid += (NAT_ENTRY_PER_BLOCK - (nid % NAT_ENTRY_PER_BLOCK));
1518 if (unlikely(nid >= nm_i->max_nid))
1521 if (i++ == FREE_NID_PAGES)
1525 /* go to the next free nat pages to find free nids abundantly */
1526 nm_i->next_scan_nid = nid;
1528 /* find free nids from current sum_pages */
1529 mutex_lock(&curseg->curseg_mutex);
1530 for (i = 0; i < nats_in_cursum(sum); i++) {
1531 block_t addr = le32_to_cpu(nat_in_journal(sum, i).block_addr);
1532 nid = le32_to_cpu(nid_in_journal(sum, i));
1533 if (addr == NULL_ADDR)
1534 add_free_nid(sbi, nid, true);
1536 remove_free_nid(nm_i, nid);
1538 mutex_unlock(&curseg->curseg_mutex);
1542 * If this function returns success, caller can obtain a new nid
1543 * from second parameter of this function.
1544 * The returned nid could be used ino as well as nid when inode is created.
1546 bool alloc_nid(struct f2fs_sb_info *sbi, nid_t *nid)
1548 struct f2fs_nm_info *nm_i = NM_I(sbi);
1549 struct free_nid *i = NULL;
1551 if (unlikely(sbi->total_valid_node_count + 1 > nm_i->available_nids))
1554 spin_lock(&nm_i->free_nid_list_lock);
1556 /* We should not use stale free nids created by build_free_nids */
1557 if (nm_i->fcnt && !on_build_free_nids(nm_i)) {
1558 f2fs_bug_on(sbi, list_empty(&nm_i->free_nid_list));
1559 list_for_each_entry(i, &nm_i->free_nid_list, list)
1560 if (i->state == NID_NEW)
1563 f2fs_bug_on(sbi, i->state != NID_NEW);
1565 i->state = NID_ALLOC;
1567 spin_unlock(&nm_i->free_nid_list_lock);
1570 spin_unlock(&nm_i->free_nid_list_lock);
1572 /* Let's scan nat pages and its caches to get free nids */
1573 mutex_lock(&nm_i->build_lock);
1574 build_free_nids(sbi);
1575 mutex_unlock(&nm_i->build_lock);
1580 * alloc_nid() should be called prior to this function.
1582 void alloc_nid_done(struct f2fs_sb_info *sbi, nid_t nid)
1584 struct f2fs_nm_info *nm_i = NM_I(sbi);
1587 spin_lock(&nm_i->free_nid_list_lock);
1588 i = __lookup_free_nid_list(nm_i, nid);
1589 f2fs_bug_on(sbi, !i || i->state != NID_ALLOC);
1590 __del_from_free_nid_list(nm_i, i);
1591 spin_unlock(&nm_i->free_nid_list_lock);
1593 kmem_cache_free(free_nid_slab, i);
1597 * alloc_nid() should be called prior to this function.
1599 void alloc_nid_failed(struct f2fs_sb_info *sbi, nid_t nid)
1601 struct f2fs_nm_info *nm_i = NM_I(sbi);
1603 bool need_free = false;
1608 spin_lock(&nm_i->free_nid_list_lock);
1609 i = __lookup_free_nid_list(nm_i, nid);
1610 f2fs_bug_on(sbi, !i || i->state != NID_ALLOC);
1611 if (!available_free_memory(sbi, FREE_NIDS)) {
1612 __del_from_free_nid_list(nm_i, i);
1618 spin_unlock(&nm_i->free_nid_list_lock);
1621 kmem_cache_free(free_nid_slab, i);
1624 void recover_inline_xattr(struct inode *inode, struct page *page)
1626 void *src_addr, *dst_addr;
1629 struct f2fs_inode *ri;
1631 ipage = get_node_page(F2FS_I_SB(inode), inode->i_ino);
1632 f2fs_bug_on(F2FS_I_SB(inode), IS_ERR(ipage));
1634 ri = F2FS_INODE(page);
1635 if (!(ri->i_inline & F2FS_INLINE_XATTR)) {
1636 clear_inode_flag(F2FS_I(inode), FI_INLINE_XATTR);
1640 dst_addr = inline_xattr_addr(ipage);
1641 src_addr = inline_xattr_addr(page);
1642 inline_size = inline_xattr_size(inode);
1644 f2fs_wait_on_page_writeback(ipage, NODE);
1645 memcpy(dst_addr, src_addr, inline_size);
1647 update_inode(inode, ipage);
1648 f2fs_put_page(ipage, 1);
1651 void recover_xattr_data(struct inode *inode, struct page *page, block_t blkaddr)
1653 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1654 nid_t prev_xnid = F2FS_I(inode)->i_xattr_nid;
1655 nid_t new_xnid = nid_of_node(page);
1656 struct node_info ni;
1658 /* 1: invalidate the previous xattr nid */
1662 /* Deallocate node address */
1663 get_node_info(sbi, prev_xnid, &ni);
1664 f2fs_bug_on(sbi, ni.blk_addr == NULL_ADDR);
1665 invalidate_blocks(sbi, ni.blk_addr);
1666 dec_valid_node_count(sbi, inode);
1667 set_node_addr(sbi, &ni, NULL_ADDR, false);
1670 /* 2: allocate new xattr nid */
1671 if (unlikely(!inc_valid_node_count(sbi, inode)))
1672 f2fs_bug_on(sbi, 1);
1674 remove_free_nid(NM_I(sbi), new_xnid);
1675 get_node_info(sbi, new_xnid, &ni);
1676 ni.ino = inode->i_ino;
1677 set_node_addr(sbi, &ni, NEW_ADDR, false);
1678 F2FS_I(inode)->i_xattr_nid = new_xnid;
1680 /* 3: update xattr blkaddr */
1681 refresh_sit_entry(sbi, NEW_ADDR, blkaddr);
1682 set_node_addr(sbi, &ni, blkaddr, false);
1684 update_inode_page(inode);
1687 int recover_inode_page(struct f2fs_sb_info *sbi, struct page *page)
1689 struct f2fs_inode *src, *dst;
1690 nid_t ino = ino_of_node(page);
1691 struct node_info old_ni, new_ni;
1694 get_node_info(sbi, ino, &old_ni);
1696 if (unlikely(old_ni.blk_addr != NULL_ADDR))
1699 ipage = grab_cache_page(NODE_MAPPING(sbi), ino);
1703 /* Should not use this inode from free nid list */
1704 remove_free_nid(NM_I(sbi), ino);
1706 SetPageUptodate(ipage);
1707 fill_node_footer(ipage, ino, ino, 0, true);
1709 src = F2FS_INODE(page);
1710 dst = F2FS_INODE(ipage);
1712 memcpy(dst, src, (unsigned long)&src->i_ext - (unsigned long)src);
1714 dst->i_blocks = cpu_to_le64(1);
1715 dst->i_links = cpu_to_le32(1);
1716 dst->i_xattr_nid = 0;
1717 dst->i_inline = src->i_inline & F2FS_INLINE_XATTR;
1722 if (unlikely(!inc_valid_node_count(sbi, NULL)))
1724 set_node_addr(sbi, &new_ni, NEW_ADDR, false);
1725 inc_valid_inode_count(sbi);
1726 set_page_dirty(ipage);
1727 f2fs_put_page(ipage, 1);
1731 int restore_node_summary(struct f2fs_sb_info *sbi,
1732 unsigned int segno, struct f2fs_summary_block *sum)
1734 struct f2fs_node *rn;
1735 struct f2fs_summary *sum_entry;
1737 int bio_blocks = MAX_BIO_BLOCKS(sbi);
1738 int i, idx, last_offset, nrpages;
1740 /* scan the node segment */
1741 last_offset = sbi->blocks_per_seg;
1742 addr = START_BLOCK(sbi, segno);
1743 sum_entry = &sum->entries[0];
1745 for (i = 0; i < last_offset; i += nrpages, addr += nrpages) {
1746 nrpages = min(last_offset - i, bio_blocks);
1748 /* readahead node pages */
1749 ra_meta_pages(sbi, addr, nrpages, META_POR);
1751 for (idx = addr; idx < addr + nrpages; idx++) {
1752 struct page *page = get_meta_page(sbi, idx);
1754 rn = F2FS_NODE(page);
1755 sum_entry->nid = rn->footer.nid;
1756 sum_entry->version = 0;
1757 sum_entry->ofs_in_node = 0;
1759 f2fs_put_page(page, 1);
1762 invalidate_mapping_pages(META_MAPPING(sbi), addr,
1768 static void remove_nats_in_journal(struct f2fs_sb_info *sbi)
1770 struct f2fs_nm_info *nm_i = NM_I(sbi);
1771 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
1772 struct f2fs_summary_block *sum = curseg->sum_blk;
1775 mutex_lock(&curseg->curseg_mutex);
1776 for (i = 0; i < nats_in_cursum(sum); i++) {
1777 struct nat_entry *ne;
1778 struct f2fs_nat_entry raw_ne;
1779 nid_t nid = le32_to_cpu(nid_in_journal(sum, i));
1781 raw_ne = nat_in_journal(sum, i);
1783 down_write(&nm_i->nat_tree_lock);
1784 ne = __lookup_nat_cache(nm_i, nid);
1786 ne = grab_nat_entry(nm_i, nid);
1787 node_info_from_raw_nat(&ne->ni, &raw_ne);
1789 __set_nat_cache_dirty(nm_i, ne);
1790 up_write(&nm_i->nat_tree_lock);
1792 update_nats_in_cursum(sum, -i);
1793 mutex_unlock(&curseg->curseg_mutex);
1796 static void __adjust_nat_entry_set(struct nat_entry_set *nes,
1797 struct list_head *head, int max)
1799 struct nat_entry_set *cur;
1801 if (nes->entry_cnt >= max)
1804 list_for_each_entry(cur, head, set_list) {
1805 if (cur->entry_cnt >= nes->entry_cnt) {
1806 list_add(&nes->set_list, cur->set_list.prev);
1811 list_add_tail(&nes->set_list, head);
1814 static void __flush_nat_entry_set(struct f2fs_sb_info *sbi,
1815 struct nat_entry_set *set)
1817 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
1818 struct f2fs_summary_block *sum = curseg->sum_blk;
1819 nid_t start_nid = set->set * NAT_ENTRY_PER_BLOCK;
1820 bool to_journal = true;
1821 struct f2fs_nat_block *nat_blk;
1822 struct nat_entry *ne, *cur;
1823 struct page *page = NULL;
1826 * there are two steps to flush nat entries:
1827 * #1, flush nat entries to journal in current hot data summary block.
1828 * #2, flush nat entries to nat page.
1830 if (!__has_cursum_space(sum, set->entry_cnt, NAT_JOURNAL))
1834 mutex_lock(&curseg->curseg_mutex);
1836 page = get_next_nat_page(sbi, start_nid);
1837 nat_blk = page_address(page);
1838 f2fs_bug_on(sbi, !nat_blk);
1841 /* flush dirty nats in nat entry set */
1842 list_for_each_entry_safe(ne, cur, &set->entry_list, list) {
1843 struct f2fs_nat_entry *raw_ne;
1844 nid_t nid = nat_get_nid(ne);
1847 if (nat_get_blkaddr(ne) == NEW_ADDR)
1851 offset = lookup_journal_in_cursum(sum,
1852 NAT_JOURNAL, nid, 1);
1853 f2fs_bug_on(sbi, offset < 0);
1854 raw_ne = &nat_in_journal(sum, offset);
1855 nid_in_journal(sum, offset) = cpu_to_le32(nid);
1857 raw_ne = &nat_blk->entries[nid - start_nid];
1859 raw_nat_from_node_info(raw_ne, &ne->ni);
1861 down_write(&NM_I(sbi)->nat_tree_lock);
1863 __clear_nat_cache_dirty(NM_I(sbi), ne);
1864 up_write(&NM_I(sbi)->nat_tree_lock);
1866 if (nat_get_blkaddr(ne) == NULL_ADDR)
1867 add_free_nid(sbi, nid, false);
1871 mutex_unlock(&curseg->curseg_mutex);
1873 f2fs_put_page(page, 1);
1875 f2fs_bug_on(sbi, set->entry_cnt);
1877 radix_tree_delete(&NM_I(sbi)->nat_set_root, set->set);
1878 kmem_cache_free(nat_entry_set_slab, set);
1882 * This function is called during the checkpointing process.
1884 void flush_nat_entries(struct f2fs_sb_info *sbi)
1886 struct f2fs_nm_info *nm_i = NM_I(sbi);
1887 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
1888 struct f2fs_summary_block *sum = curseg->sum_blk;
1889 struct nat_entry_set *setvec[SETVEC_SIZE];
1890 struct nat_entry_set *set, *tmp;
1895 if (!nm_i->dirty_nat_cnt)
1898 * if there are no enough space in journal to store dirty nat
1899 * entries, remove all entries from journal and merge them
1900 * into nat entry set.
1902 if (!__has_cursum_space(sum, nm_i->dirty_nat_cnt, NAT_JOURNAL))
1903 remove_nats_in_journal(sbi);
1905 while ((found = __gang_lookup_nat_set(nm_i,
1906 set_idx, SETVEC_SIZE, setvec))) {
1908 set_idx = setvec[found - 1]->set + 1;
1909 for (idx = 0; idx < found; idx++)
1910 __adjust_nat_entry_set(setvec[idx], &sets,
1911 MAX_NAT_JENTRIES(sum));
1914 /* flush dirty nats in nat entry set */
1915 list_for_each_entry_safe(set, tmp, &sets, set_list)
1916 __flush_nat_entry_set(sbi, set);
1918 f2fs_bug_on(sbi, nm_i->dirty_nat_cnt);
1921 static int init_node_manager(struct f2fs_sb_info *sbi)
1923 struct f2fs_super_block *sb_raw = F2FS_RAW_SUPER(sbi);
1924 struct f2fs_nm_info *nm_i = NM_I(sbi);
1925 unsigned char *version_bitmap;
1926 unsigned int nat_segs, nat_blocks;
1928 nm_i->nat_blkaddr = le32_to_cpu(sb_raw->nat_blkaddr);
1930 /* segment_count_nat includes pair segment so divide to 2. */
1931 nat_segs = le32_to_cpu(sb_raw->segment_count_nat) >> 1;
1932 nat_blocks = nat_segs << le32_to_cpu(sb_raw->log_blocks_per_seg);
1934 nm_i->max_nid = NAT_ENTRY_PER_BLOCK * nat_blocks;
1936 /* not used nids: 0, node, meta, (and root counted as valid node) */
1937 nm_i->available_nids = nm_i->max_nid - F2FS_RESERVED_NODE_NUM;
1940 nm_i->ram_thresh = DEF_RAM_THRESHOLD;
1942 INIT_RADIX_TREE(&nm_i->free_nid_root, GFP_ATOMIC);
1943 INIT_LIST_HEAD(&nm_i->free_nid_list);
1944 INIT_RADIX_TREE(&nm_i->nat_root, GFP_NOIO);
1945 INIT_RADIX_TREE(&nm_i->nat_set_root, GFP_NOIO);
1946 INIT_LIST_HEAD(&nm_i->nat_entries);
1948 mutex_init(&nm_i->build_lock);
1949 spin_lock_init(&nm_i->free_nid_list_lock);
1950 init_rwsem(&nm_i->nat_tree_lock);
1952 nm_i->next_scan_nid = le32_to_cpu(sbi->ckpt->next_free_nid);
1953 nm_i->bitmap_size = __bitmap_size(sbi, NAT_BITMAP);
1954 version_bitmap = __bitmap_ptr(sbi, NAT_BITMAP);
1955 if (!version_bitmap)
1958 nm_i->nat_bitmap = kmemdup(version_bitmap, nm_i->bitmap_size,
1960 if (!nm_i->nat_bitmap)
1965 int build_node_manager(struct f2fs_sb_info *sbi)
1969 sbi->nm_info = kzalloc(sizeof(struct f2fs_nm_info), GFP_KERNEL);
1973 err = init_node_manager(sbi);
1977 build_free_nids(sbi);
1981 void destroy_node_manager(struct f2fs_sb_info *sbi)
1983 struct f2fs_nm_info *nm_i = NM_I(sbi);
1984 struct free_nid *i, *next_i;
1985 struct nat_entry *natvec[NATVEC_SIZE];
1986 struct nat_entry_set *setvec[SETVEC_SIZE];
1993 /* destroy free nid list */
1994 spin_lock(&nm_i->free_nid_list_lock);
1995 list_for_each_entry_safe(i, next_i, &nm_i->free_nid_list, list) {
1996 f2fs_bug_on(sbi, i->state == NID_ALLOC);
1997 __del_from_free_nid_list(nm_i, i);
1999 spin_unlock(&nm_i->free_nid_list_lock);
2000 kmem_cache_free(free_nid_slab, i);
2001 spin_lock(&nm_i->free_nid_list_lock);
2003 f2fs_bug_on(sbi, nm_i->fcnt);
2004 spin_unlock(&nm_i->free_nid_list_lock);
2006 /* destroy nat cache */
2007 down_write(&nm_i->nat_tree_lock);
2008 while ((found = __gang_lookup_nat_cache(nm_i,
2009 nid, NATVEC_SIZE, natvec))) {
2012 nid = nat_get_nid(natvec[found - 1]) + 1;
2013 for (idx = 0; idx < found; idx++)
2014 __del_from_nat_cache(nm_i, natvec[idx]);
2016 f2fs_bug_on(sbi, nm_i->nat_cnt);
2018 /* destroy nat set cache */
2020 while ((found = __gang_lookup_nat_set(nm_i,
2021 nid, SETVEC_SIZE, setvec))) {
2024 nid = setvec[found - 1]->set + 1;
2025 for (idx = 0; idx < found; idx++) {
2026 /* entry_cnt is not zero, when cp_error was occurred */
2027 f2fs_bug_on(sbi, !list_empty(&setvec[idx]->entry_list));
2028 radix_tree_delete(&nm_i->nat_set_root, setvec[idx]->set);
2029 kmem_cache_free(nat_entry_set_slab, setvec[idx]);
2032 up_write(&nm_i->nat_tree_lock);
2034 kfree(nm_i->nat_bitmap);
2035 sbi->nm_info = NULL;
2039 int __init create_node_manager_caches(void)
2041 nat_entry_slab = f2fs_kmem_cache_create("nat_entry",
2042 sizeof(struct nat_entry));
2043 if (!nat_entry_slab)
2046 free_nid_slab = f2fs_kmem_cache_create("free_nid",
2047 sizeof(struct free_nid));
2049 goto destroy_nat_entry;
2051 nat_entry_set_slab = f2fs_kmem_cache_create("nat_entry_set",
2052 sizeof(struct nat_entry_set));
2053 if (!nat_entry_set_slab)
2054 goto destroy_free_nid;
2058 kmem_cache_destroy(free_nid_slab);
2060 kmem_cache_destroy(nat_entry_slab);
2065 void destroy_node_manager_caches(void)
2067 kmem_cache_destroy(nat_entry_set_slab);
2068 kmem_cache_destroy(free_nid_slab);
2069 kmem_cache_destroy(nat_entry_slab);