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 } else if (type == DIRTY_DENTS) {
56 if (sbi->sb->s_bdi->wb.dirty_exceeded)
58 mem_size = get_pages(sbi, F2FS_DIRTY_DENTS);
59 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 1);
60 } else if (type == INO_ENTRIES) {
63 for (i = 0; i <= UPDATE_INO; i++)
64 mem_size += (sbi->im[i].ino_num *
65 sizeof(struct ino_entry)) >> PAGE_CACHE_SHIFT;
66 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 1);
67 } else if (type == EXTENT_CACHE) {
68 mem_size = (atomic_read(&sbi->total_ext_tree) *
69 sizeof(struct extent_tree) +
70 atomic_read(&sbi->total_ext_node) *
71 sizeof(struct extent_node)) >> PAGE_CACHE_SHIFT;
72 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 1);
74 if (!sbi->sb->s_bdi->wb.dirty_exceeded)
80 static void clear_node_page_dirty(struct page *page)
82 struct address_space *mapping = page->mapping;
83 unsigned int long flags;
85 if (PageDirty(page)) {
86 spin_lock_irqsave(&mapping->tree_lock, flags);
87 radix_tree_tag_clear(&mapping->page_tree,
90 spin_unlock_irqrestore(&mapping->tree_lock, flags);
92 clear_page_dirty_for_io(page);
93 dec_page_count(F2FS_M_SB(mapping), F2FS_DIRTY_NODES);
95 ClearPageUptodate(page);
98 static struct page *get_current_nat_page(struct f2fs_sb_info *sbi, nid_t nid)
100 pgoff_t index = current_nat_addr(sbi, nid);
101 return get_meta_page(sbi, index);
104 static struct page *get_next_nat_page(struct f2fs_sb_info *sbi, nid_t nid)
106 struct page *src_page;
107 struct page *dst_page;
112 struct f2fs_nm_info *nm_i = NM_I(sbi);
114 src_off = current_nat_addr(sbi, nid);
115 dst_off = next_nat_addr(sbi, src_off);
117 /* get current nat block page with lock */
118 src_page = get_meta_page(sbi, src_off);
119 dst_page = grab_meta_page(sbi, dst_off);
120 f2fs_bug_on(sbi, PageDirty(src_page));
122 src_addr = page_address(src_page);
123 dst_addr = page_address(dst_page);
124 memcpy(dst_addr, src_addr, PAGE_CACHE_SIZE);
125 set_page_dirty(dst_page);
126 f2fs_put_page(src_page, 1);
128 set_to_next_nat(nm_i, nid);
133 static struct nat_entry *__lookup_nat_cache(struct f2fs_nm_info *nm_i, nid_t n)
135 return radix_tree_lookup(&nm_i->nat_root, n);
138 static unsigned int __gang_lookup_nat_cache(struct f2fs_nm_info *nm_i,
139 nid_t start, unsigned int nr, struct nat_entry **ep)
141 return radix_tree_gang_lookup(&nm_i->nat_root, (void **)ep, start, nr);
144 static void __del_from_nat_cache(struct f2fs_nm_info *nm_i, struct nat_entry *e)
147 radix_tree_delete(&nm_i->nat_root, nat_get_nid(e));
149 kmem_cache_free(nat_entry_slab, e);
152 static void __set_nat_cache_dirty(struct f2fs_nm_info *nm_i,
153 struct nat_entry *ne)
155 nid_t set = NAT_BLOCK_OFFSET(ne->ni.nid);
156 struct nat_entry_set *head;
158 if (get_nat_flag(ne, IS_DIRTY))
161 head = radix_tree_lookup(&nm_i->nat_set_root, set);
163 head = f2fs_kmem_cache_alloc(nat_entry_set_slab, GFP_NOFS);
165 INIT_LIST_HEAD(&head->entry_list);
166 INIT_LIST_HEAD(&head->set_list);
169 f2fs_radix_tree_insert(&nm_i->nat_set_root, set, head);
171 list_move_tail(&ne->list, &head->entry_list);
172 nm_i->dirty_nat_cnt++;
174 set_nat_flag(ne, IS_DIRTY, true);
177 static void __clear_nat_cache_dirty(struct f2fs_nm_info *nm_i,
178 struct nat_entry *ne)
180 nid_t set = NAT_BLOCK_OFFSET(ne->ni.nid);
181 struct nat_entry_set *head;
183 head = radix_tree_lookup(&nm_i->nat_set_root, set);
185 list_move_tail(&ne->list, &nm_i->nat_entries);
186 set_nat_flag(ne, IS_DIRTY, false);
188 nm_i->dirty_nat_cnt--;
192 static unsigned int __gang_lookup_nat_set(struct f2fs_nm_info *nm_i,
193 nid_t start, unsigned int nr, struct nat_entry_set **ep)
195 return radix_tree_gang_lookup(&nm_i->nat_set_root, (void **)ep,
199 int need_dentry_mark(struct f2fs_sb_info *sbi, nid_t nid)
201 struct f2fs_nm_info *nm_i = NM_I(sbi);
205 down_read(&nm_i->nat_tree_lock);
206 e = __lookup_nat_cache(nm_i, nid);
208 if (!get_nat_flag(e, IS_CHECKPOINTED) &&
209 !get_nat_flag(e, HAS_FSYNCED_INODE))
212 up_read(&nm_i->nat_tree_lock);
216 bool is_checkpointed_node(struct f2fs_sb_info *sbi, nid_t nid)
218 struct f2fs_nm_info *nm_i = NM_I(sbi);
222 down_read(&nm_i->nat_tree_lock);
223 e = __lookup_nat_cache(nm_i, nid);
224 if (e && !get_nat_flag(e, IS_CHECKPOINTED))
226 up_read(&nm_i->nat_tree_lock);
230 bool need_inode_block_update(struct f2fs_sb_info *sbi, nid_t ino)
232 struct f2fs_nm_info *nm_i = NM_I(sbi);
234 bool need_update = true;
236 down_read(&nm_i->nat_tree_lock);
237 e = __lookup_nat_cache(nm_i, ino);
238 if (e && get_nat_flag(e, HAS_LAST_FSYNC) &&
239 (get_nat_flag(e, IS_CHECKPOINTED) ||
240 get_nat_flag(e, HAS_FSYNCED_INODE)))
242 up_read(&nm_i->nat_tree_lock);
246 static struct nat_entry *grab_nat_entry(struct f2fs_nm_info *nm_i, nid_t nid)
248 struct nat_entry *new;
250 new = f2fs_kmem_cache_alloc(nat_entry_slab, GFP_NOFS);
251 f2fs_radix_tree_insert(&nm_i->nat_root, nid, new);
252 memset(new, 0, sizeof(struct nat_entry));
253 nat_set_nid(new, nid);
255 list_add_tail(&new->list, &nm_i->nat_entries);
260 static void cache_nat_entry(struct f2fs_nm_info *nm_i, nid_t nid,
261 struct f2fs_nat_entry *ne)
265 e = __lookup_nat_cache(nm_i, nid);
267 e = grab_nat_entry(nm_i, nid);
268 node_info_from_raw_nat(&e->ni, ne);
272 static void set_node_addr(struct f2fs_sb_info *sbi, struct node_info *ni,
273 block_t new_blkaddr, bool fsync_done)
275 struct f2fs_nm_info *nm_i = NM_I(sbi);
278 down_write(&nm_i->nat_tree_lock);
279 e = __lookup_nat_cache(nm_i, ni->nid);
281 e = grab_nat_entry(nm_i, ni->nid);
282 copy_node_info(&e->ni, ni);
283 f2fs_bug_on(sbi, ni->blk_addr == NEW_ADDR);
284 } else if (new_blkaddr == NEW_ADDR) {
286 * when nid is reallocated,
287 * previous nat entry can be remained in nat cache.
288 * So, reinitialize it with new information.
290 copy_node_info(&e->ni, ni);
291 f2fs_bug_on(sbi, ni->blk_addr != NULL_ADDR);
295 f2fs_bug_on(sbi, nat_get_blkaddr(e) != ni->blk_addr);
296 f2fs_bug_on(sbi, nat_get_blkaddr(e) == NULL_ADDR &&
297 new_blkaddr == NULL_ADDR);
298 f2fs_bug_on(sbi, nat_get_blkaddr(e) == NEW_ADDR &&
299 new_blkaddr == NEW_ADDR);
300 f2fs_bug_on(sbi, nat_get_blkaddr(e) != NEW_ADDR &&
301 nat_get_blkaddr(e) != NULL_ADDR &&
302 new_blkaddr == NEW_ADDR);
304 /* increment version no as node is removed */
305 if (nat_get_blkaddr(e) != NEW_ADDR && new_blkaddr == NULL_ADDR) {
306 unsigned char version = nat_get_version(e);
307 nat_set_version(e, inc_node_version(version));
309 /* in order to reuse the nid */
310 if (nm_i->next_scan_nid > ni->nid)
311 nm_i->next_scan_nid = ni->nid;
315 nat_set_blkaddr(e, new_blkaddr);
316 if (new_blkaddr == NEW_ADDR || new_blkaddr == NULL_ADDR)
317 set_nat_flag(e, IS_CHECKPOINTED, false);
318 __set_nat_cache_dirty(nm_i, e);
320 /* update fsync_mark if its inode nat entry is still alive */
321 if (ni->nid != ni->ino)
322 e = __lookup_nat_cache(nm_i, ni->ino);
324 if (fsync_done && ni->nid == ni->ino)
325 set_nat_flag(e, HAS_FSYNCED_INODE, true);
326 set_nat_flag(e, HAS_LAST_FSYNC, fsync_done);
328 up_write(&nm_i->nat_tree_lock);
331 int try_to_free_nats(struct f2fs_sb_info *sbi, int nr_shrink)
333 struct f2fs_nm_info *nm_i = NM_I(sbi);
336 if (!down_write_trylock(&nm_i->nat_tree_lock))
339 while (nr_shrink && !list_empty(&nm_i->nat_entries)) {
340 struct nat_entry *ne;
341 ne = list_first_entry(&nm_i->nat_entries,
342 struct nat_entry, list);
343 __del_from_nat_cache(nm_i, ne);
346 up_write(&nm_i->nat_tree_lock);
347 return nr - nr_shrink;
351 * This function always returns success
353 void get_node_info(struct f2fs_sb_info *sbi, nid_t nid, struct node_info *ni)
355 struct f2fs_nm_info *nm_i = NM_I(sbi);
356 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
357 struct f2fs_summary_block *sum = curseg->sum_blk;
358 nid_t start_nid = START_NID(nid);
359 struct f2fs_nat_block *nat_blk;
360 struct page *page = NULL;
361 struct f2fs_nat_entry ne;
367 /* Check nat cache */
368 down_read(&nm_i->nat_tree_lock);
369 e = __lookup_nat_cache(nm_i, nid);
371 ni->ino = nat_get_ino(e);
372 ni->blk_addr = nat_get_blkaddr(e);
373 ni->version = nat_get_version(e);
375 up_read(&nm_i->nat_tree_lock);
379 memset(&ne, 0, sizeof(struct f2fs_nat_entry));
381 down_write(&nm_i->nat_tree_lock);
383 /* Check current segment summary */
384 mutex_lock(&curseg->curseg_mutex);
385 i = lookup_journal_in_cursum(sum, NAT_JOURNAL, nid, 0);
387 ne = nat_in_journal(sum, i);
388 node_info_from_raw_nat(ni, &ne);
390 mutex_unlock(&curseg->curseg_mutex);
394 /* Fill node_info from nat page */
395 page = get_current_nat_page(sbi, start_nid);
396 nat_blk = (struct f2fs_nat_block *)page_address(page);
397 ne = nat_blk->entries[nid - start_nid];
398 node_info_from_raw_nat(ni, &ne);
399 f2fs_put_page(page, 1);
401 /* cache nat entry */
402 cache_nat_entry(NM_I(sbi), nid, &ne);
403 up_write(&nm_i->nat_tree_lock);
406 pgoff_t get_next_page_offset(struct dnode_of_data *dn, pgoff_t pgofs)
408 const long direct_index = ADDRS_PER_INODE(dn->inode);
409 const long direct_blks = ADDRS_PER_BLOCK;
410 const long indirect_blks = ADDRS_PER_BLOCK * NIDS_PER_BLOCK;
411 unsigned int skipped_unit = ADDRS_PER_BLOCK;
412 int cur_level = dn->cur_level;
413 int max_level = dn->max_level;
419 while (max_level-- > cur_level)
420 skipped_unit *= NIDS_PER_BLOCK;
422 switch (dn->max_level) {
424 base += 2 * indirect_blks;
426 base += 2 * direct_blks;
428 base += direct_index;
431 f2fs_bug_on(F2FS_I_SB(dn->inode), 1);
434 return ((pgofs - base) / skipped_unit + 1) * skipped_unit + base;
438 * The maximum depth is four.
439 * Offset[0] will have raw inode offset.
441 static int get_node_path(struct inode *inode, long block,
442 int offset[4], unsigned int noffset[4])
444 const long direct_index = ADDRS_PER_INODE(inode);
445 const long direct_blks = ADDRS_PER_BLOCK;
446 const long dptrs_per_blk = NIDS_PER_BLOCK;
447 const long indirect_blks = ADDRS_PER_BLOCK * NIDS_PER_BLOCK;
448 const long dindirect_blks = indirect_blks * NIDS_PER_BLOCK;
454 if (block < direct_index) {
458 block -= direct_index;
459 if (block < direct_blks) {
460 offset[n++] = NODE_DIR1_BLOCK;
466 block -= direct_blks;
467 if (block < direct_blks) {
468 offset[n++] = NODE_DIR2_BLOCK;
474 block -= direct_blks;
475 if (block < indirect_blks) {
476 offset[n++] = NODE_IND1_BLOCK;
478 offset[n++] = block / direct_blks;
479 noffset[n] = 4 + offset[n - 1];
480 offset[n] = block % direct_blks;
484 block -= indirect_blks;
485 if (block < indirect_blks) {
486 offset[n++] = NODE_IND2_BLOCK;
487 noffset[n] = 4 + dptrs_per_blk;
488 offset[n++] = block / direct_blks;
489 noffset[n] = 5 + dptrs_per_blk + offset[n - 1];
490 offset[n] = block % direct_blks;
494 block -= indirect_blks;
495 if (block < dindirect_blks) {
496 offset[n++] = NODE_DIND_BLOCK;
497 noffset[n] = 5 + (dptrs_per_blk * 2);
498 offset[n++] = block / indirect_blks;
499 noffset[n] = 6 + (dptrs_per_blk * 2) +
500 offset[n - 1] * (dptrs_per_blk + 1);
501 offset[n++] = (block / direct_blks) % dptrs_per_blk;
502 noffset[n] = 7 + (dptrs_per_blk * 2) +
503 offset[n - 2] * (dptrs_per_blk + 1) +
505 offset[n] = block % direct_blks;
516 * Caller should call f2fs_put_dnode(dn).
517 * Also, it should grab and release a rwsem by calling f2fs_lock_op() and
518 * f2fs_unlock_op() only if ro is not set RDONLY_NODE.
519 * In the case of RDONLY_NODE, we don't need to care about mutex.
521 int get_dnode_of_data(struct dnode_of_data *dn, pgoff_t index, int mode)
523 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
524 struct page *npage[4];
525 struct page *parent = NULL;
527 unsigned int noffset[4];
532 level = get_node_path(dn->inode, index, offset, noffset);
534 nids[0] = dn->inode->i_ino;
535 npage[0] = dn->inode_page;
538 npage[0] = get_node_page(sbi, nids[0]);
539 if (IS_ERR(npage[0]))
540 return PTR_ERR(npage[0]);
543 /* if inline_data is set, should not report any block indices */
544 if (f2fs_has_inline_data(dn->inode) && index) {
546 f2fs_put_page(npage[0], 1);
552 nids[1] = get_nid(parent, offset[0], true);
553 dn->inode_page = npage[0];
554 dn->inode_page_locked = true;
556 /* get indirect or direct nodes */
557 for (i = 1; i <= level; i++) {
560 if (!nids[i] && mode == ALLOC_NODE) {
562 if (!alloc_nid(sbi, &(nids[i]))) {
568 npage[i] = new_node_page(dn, noffset[i], NULL);
569 if (IS_ERR(npage[i])) {
570 alloc_nid_failed(sbi, nids[i]);
571 err = PTR_ERR(npage[i]);
575 set_nid(parent, offset[i - 1], nids[i], i == 1);
576 alloc_nid_done(sbi, nids[i]);
578 } else if (mode == LOOKUP_NODE_RA && i == level && level > 1) {
579 npage[i] = get_node_page_ra(parent, offset[i - 1]);
580 if (IS_ERR(npage[i])) {
581 err = PTR_ERR(npage[i]);
587 dn->inode_page_locked = false;
590 f2fs_put_page(parent, 1);
594 npage[i] = get_node_page(sbi, nids[i]);
595 if (IS_ERR(npage[i])) {
596 err = PTR_ERR(npage[i]);
597 f2fs_put_page(npage[0], 0);
603 nids[i + 1] = get_nid(parent, offset[i], false);
606 dn->nid = nids[level];
607 dn->ofs_in_node = offset[level];
608 dn->node_page = npage[level];
609 dn->data_blkaddr = datablock_addr(dn->node_page, dn->ofs_in_node);
613 f2fs_put_page(parent, 1);
615 f2fs_put_page(npage[0], 0);
617 dn->inode_page = NULL;
618 dn->node_page = NULL;
619 if (err == -ENOENT) {
621 dn->max_level = level;
626 static void truncate_node(struct dnode_of_data *dn)
628 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
631 get_node_info(sbi, dn->nid, &ni);
632 if (dn->inode->i_blocks == 0) {
633 f2fs_bug_on(sbi, ni.blk_addr != NULL_ADDR);
636 f2fs_bug_on(sbi, ni.blk_addr == NULL_ADDR);
638 /* Deallocate node address */
639 invalidate_blocks(sbi, ni.blk_addr);
640 dec_valid_node_count(sbi, dn->inode);
641 set_node_addr(sbi, &ni, NULL_ADDR, false);
643 if (dn->nid == dn->inode->i_ino) {
644 remove_orphan_inode(sbi, dn->nid);
645 dec_valid_inode_count(sbi);
650 clear_node_page_dirty(dn->node_page);
651 set_sbi_flag(sbi, SBI_IS_DIRTY);
653 f2fs_put_page(dn->node_page, 1);
655 invalidate_mapping_pages(NODE_MAPPING(sbi),
656 dn->node_page->index, dn->node_page->index);
658 dn->node_page = NULL;
659 trace_f2fs_truncate_node(dn->inode, dn->nid, ni.blk_addr);
662 static int truncate_dnode(struct dnode_of_data *dn)
669 /* get direct node */
670 page = get_node_page(F2FS_I_SB(dn->inode), dn->nid);
671 if (IS_ERR(page) && PTR_ERR(page) == -ENOENT)
673 else if (IS_ERR(page))
674 return PTR_ERR(page);
676 /* Make dnode_of_data for parameter */
677 dn->node_page = page;
679 truncate_data_blocks(dn);
684 static int truncate_nodes(struct dnode_of_data *dn, unsigned int nofs,
687 struct dnode_of_data rdn = *dn;
689 struct f2fs_node *rn;
691 unsigned int child_nofs;
696 return NIDS_PER_BLOCK + 1;
698 trace_f2fs_truncate_nodes_enter(dn->inode, dn->nid, dn->data_blkaddr);
700 page = get_node_page(F2FS_I_SB(dn->inode), dn->nid);
702 trace_f2fs_truncate_nodes_exit(dn->inode, PTR_ERR(page));
703 return PTR_ERR(page);
706 rn = F2FS_NODE(page);
708 for (i = ofs; i < NIDS_PER_BLOCK; i++, freed++) {
709 child_nid = le32_to_cpu(rn->in.nid[i]);
713 ret = truncate_dnode(&rdn);
716 if (set_nid(page, i, 0, false))
717 dn->node_changed = true;
720 child_nofs = nofs + ofs * (NIDS_PER_BLOCK + 1) + 1;
721 for (i = ofs; i < NIDS_PER_BLOCK; i++) {
722 child_nid = le32_to_cpu(rn->in.nid[i]);
723 if (child_nid == 0) {
724 child_nofs += NIDS_PER_BLOCK + 1;
728 ret = truncate_nodes(&rdn, child_nofs, 0, depth - 1);
729 if (ret == (NIDS_PER_BLOCK + 1)) {
730 if (set_nid(page, i, 0, false))
731 dn->node_changed = true;
733 } else if (ret < 0 && ret != -ENOENT) {
741 /* remove current indirect node */
742 dn->node_page = page;
746 f2fs_put_page(page, 1);
748 trace_f2fs_truncate_nodes_exit(dn->inode, freed);
752 f2fs_put_page(page, 1);
753 trace_f2fs_truncate_nodes_exit(dn->inode, ret);
757 static int truncate_partial_nodes(struct dnode_of_data *dn,
758 struct f2fs_inode *ri, int *offset, int depth)
760 struct page *pages[2];
767 nid[0] = le32_to_cpu(ri->i_nid[offset[0] - NODE_DIR1_BLOCK]);
771 /* get indirect nodes in the path */
772 for (i = 0; i < idx + 1; i++) {
773 /* reference count'll be increased */
774 pages[i] = get_node_page(F2FS_I_SB(dn->inode), nid[i]);
775 if (IS_ERR(pages[i])) {
776 err = PTR_ERR(pages[i]);
780 nid[i + 1] = get_nid(pages[i], offset[i + 1], false);
783 /* free direct nodes linked to a partial indirect node */
784 for (i = offset[idx + 1]; i < NIDS_PER_BLOCK; i++) {
785 child_nid = get_nid(pages[idx], i, false);
789 err = truncate_dnode(dn);
792 if (set_nid(pages[idx], i, 0, false))
793 dn->node_changed = true;
796 if (offset[idx + 1] == 0) {
797 dn->node_page = pages[idx];
801 f2fs_put_page(pages[idx], 1);
807 for (i = idx; i >= 0; i--)
808 f2fs_put_page(pages[i], 1);
810 trace_f2fs_truncate_partial_nodes(dn->inode, nid, depth, err);
816 * All the block addresses of data and nodes should be nullified.
818 int truncate_inode_blocks(struct inode *inode, pgoff_t from)
820 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
821 int err = 0, cont = 1;
822 int level, offset[4], noffset[4];
823 unsigned int nofs = 0;
824 struct f2fs_inode *ri;
825 struct dnode_of_data dn;
828 trace_f2fs_truncate_inode_blocks_enter(inode, from);
830 level = get_node_path(inode, from, offset, noffset);
832 page = get_node_page(sbi, inode->i_ino);
834 trace_f2fs_truncate_inode_blocks_exit(inode, PTR_ERR(page));
835 return PTR_ERR(page);
838 set_new_dnode(&dn, inode, page, NULL, 0);
841 ri = F2FS_INODE(page);
849 if (!offset[level - 1])
851 err = truncate_partial_nodes(&dn, ri, offset, level);
852 if (err < 0 && err != -ENOENT)
854 nofs += 1 + NIDS_PER_BLOCK;
857 nofs = 5 + 2 * NIDS_PER_BLOCK;
858 if (!offset[level - 1])
860 err = truncate_partial_nodes(&dn, ri, offset, level);
861 if (err < 0 && err != -ENOENT)
870 dn.nid = le32_to_cpu(ri->i_nid[offset[0] - NODE_DIR1_BLOCK]);
872 case NODE_DIR1_BLOCK:
873 case NODE_DIR2_BLOCK:
874 err = truncate_dnode(&dn);
877 case NODE_IND1_BLOCK:
878 case NODE_IND2_BLOCK:
879 err = truncate_nodes(&dn, nofs, offset[1], 2);
882 case NODE_DIND_BLOCK:
883 err = truncate_nodes(&dn, nofs, offset[1], 3);
890 if (err < 0 && err != -ENOENT)
892 if (offset[1] == 0 &&
893 ri->i_nid[offset[0] - NODE_DIR1_BLOCK]) {
895 if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
896 f2fs_put_page(page, 1);
899 f2fs_wait_on_page_writeback(page, NODE, true);
900 ri->i_nid[offset[0] - NODE_DIR1_BLOCK] = 0;
901 set_page_dirty(page);
909 f2fs_put_page(page, 0);
910 trace_f2fs_truncate_inode_blocks_exit(inode, err);
911 return err > 0 ? 0 : err;
914 int truncate_xattr_node(struct inode *inode, struct page *page)
916 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
917 nid_t nid = F2FS_I(inode)->i_xattr_nid;
918 struct dnode_of_data dn;
924 npage = get_node_page(sbi, nid);
926 return PTR_ERR(npage);
928 F2FS_I(inode)->i_xattr_nid = 0;
930 /* need to do checkpoint during fsync */
931 F2FS_I(inode)->xattr_ver = cur_cp_version(F2FS_CKPT(sbi));
933 set_new_dnode(&dn, inode, page, npage, nid);
936 dn.inode_page_locked = true;
942 * Caller should grab and release a rwsem by calling f2fs_lock_op() and
945 int remove_inode_page(struct inode *inode)
947 struct dnode_of_data dn;
950 set_new_dnode(&dn, inode, NULL, NULL, inode->i_ino);
951 err = get_dnode_of_data(&dn, 0, LOOKUP_NODE);
955 err = truncate_xattr_node(inode, dn.inode_page);
961 /* remove potential inline_data blocks */
962 if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
963 S_ISLNK(inode->i_mode))
964 truncate_data_blocks_range(&dn, 1);
966 /* 0 is possible, after f2fs_new_inode() has failed */
967 f2fs_bug_on(F2FS_I_SB(inode),
968 inode->i_blocks != 0 && inode->i_blocks != 1);
970 /* will put inode & node pages */
975 struct page *new_inode_page(struct inode *inode)
977 struct dnode_of_data dn;
979 /* allocate inode page for new inode */
980 set_new_dnode(&dn, inode, NULL, NULL, inode->i_ino);
982 /* caller should f2fs_put_page(page, 1); */
983 return new_node_page(&dn, 0, NULL);
986 struct page *new_node_page(struct dnode_of_data *dn,
987 unsigned int ofs, struct page *ipage)
989 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
990 struct node_info old_ni, new_ni;
994 if (unlikely(is_inode_flag_set(F2FS_I(dn->inode), FI_NO_ALLOC)))
995 return ERR_PTR(-EPERM);
997 page = grab_cache_page(NODE_MAPPING(sbi), dn->nid);
999 return ERR_PTR(-ENOMEM);
1001 if (unlikely(!inc_valid_node_count(sbi, dn->inode))) {
1006 get_node_info(sbi, dn->nid, &old_ni);
1008 /* Reinitialize old_ni with new node page */
1009 f2fs_bug_on(sbi, old_ni.blk_addr != NULL_ADDR);
1011 new_ni.ino = dn->inode->i_ino;
1012 set_node_addr(sbi, &new_ni, NEW_ADDR, false);
1014 f2fs_wait_on_page_writeback(page, NODE, true);
1015 fill_node_footer(page, dn->nid, dn->inode->i_ino, ofs, true);
1016 set_cold_node(dn->inode, page);
1017 SetPageUptodate(page);
1018 if (set_page_dirty(page))
1019 dn->node_changed = true;
1021 if (f2fs_has_xattr_block(ofs))
1022 F2FS_I(dn->inode)->i_xattr_nid = dn->nid;
1024 dn->node_page = page;
1026 update_inode(dn->inode, ipage);
1028 sync_inode_page(dn);
1030 inc_valid_inode_count(sbi);
1035 clear_node_page_dirty(page);
1036 f2fs_put_page(page, 1);
1037 return ERR_PTR(err);
1041 * Caller should do after getting the following values.
1042 * 0: f2fs_put_page(page, 0)
1043 * LOCKED_PAGE or error: f2fs_put_page(page, 1)
1045 static int read_node_page(struct page *page, int rw)
1047 struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1048 struct node_info ni;
1049 struct f2fs_io_info fio = {
1054 .encrypted_page = NULL,
1057 get_node_info(sbi, page->index, &ni);
1059 if (unlikely(ni.blk_addr == NULL_ADDR)) {
1060 ClearPageUptodate(page);
1064 if (PageUptodate(page))
1067 fio.blk_addr = ni.blk_addr;
1068 return f2fs_submit_page_bio(&fio);
1072 * Readahead a node page
1074 void ra_node_page(struct f2fs_sb_info *sbi, nid_t nid)
1081 f2fs_bug_on(sbi, check_nid_range(sbi, nid));
1083 apage = find_get_page(NODE_MAPPING(sbi), nid);
1084 if (apage && PageUptodate(apage)) {
1085 f2fs_put_page(apage, 0);
1088 f2fs_put_page(apage, 0);
1090 apage = grab_cache_page(NODE_MAPPING(sbi), nid);
1094 err = read_node_page(apage, READA);
1095 f2fs_put_page(apage, err ? 1 : 0);
1099 * readahead MAX_RA_NODE number of node pages.
1101 void ra_node_pages(struct page *parent, int start)
1103 struct f2fs_sb_info *sbi = F2FS_P_SB(parent);
1104 struct blk_plug plug;
1108 blk_start_plug(&plug);
1110 /* Then, try readahead for siblings of the desired node */
1111 end = start + MAX_RA_NODE;
1112 end = min(end, NIDS_PER_BLOCK);
1113 for (i = start; i < end; i++) {
1114 nid = get_nid(parent, i, false);
1115 ra_node_page(sbi, nid);
1118 blk_finish_plug(&plug);
1121 struct page *__get_node_page(struct f2fs_sb_info *sbi, pgoff_t nid,
1122 struct page *parent, int start)
1128 return ERR_PTR(-ENOENT);
1129 f2fs_bug_on(sbi, check_nid_range(sbi, nid));
1131 page = grab_cache_page(NODE_MAPPING(sbi), nid);
1133 return ERR_PTR(-ENOMEM);
1135 err = read_node_page(page, READ_SYNC);
1137 f2fs_put_page(page, 1);
1138 return ERR_PTR(err);
1139 } else if (err == LOCKED_PAGE) {
1144 ra_node_pages(parent, start + 1);
1148 if (unlikely(!PageUptodate(page))) {
1149 f2fs_put_page(page, 1);
1150 return ERR_PTR(-EIO);
1152 if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1153 f2fs_put_page(page, 1);
1157 f2fs_bug_on(sbi, nid != nid_of_node(page));
1161 struct page *get_node_page(struct f2fs_sb_info *sbi, pgoff_t nid)
1163 return __get_node_page(sbi, nid, NULL, 0);
1166 struct page *get_node_page_ra(struct page *parent, int start)
1168 struct f2fs_sb_info *sbi = F2FS_P_SB(parent);
1169 nid_t nid = get_nid(parent, start, false);
1171 return __get_node_page(sbi, nid, parent, start);
1174 void sync_inode_page(struct dnode_of_data *dn)
1178 if (IS_INODE(dn->node_page) || dn->inode_page == dn->node_page) {
1179 ret = update_inode(dn->inode, dn->node_page);
1180 } else if (dn->inode_page) {
1181 if (!dn->inode_page_locked)
1182 lock_page(dn->inode_page);
1183 ret = update_inode(dn->inode, dn->inode_page);
1184 if (!dn->inode_page_locked)
1185 unlock_page(dn->inode_page);
1187 ret = update_inode_page(dn->inode);
1189 dn->node_changed = ret ? true: false;
1192 static void flush_inline_data(struct f2fs_sb_info *sbi, nid_t ino)
1194 struct inode *inode;
1197 /* should flush inline_data before evict_inode */
1198 inode = ilookup(sbi->sb, ino);
1202 page = pagecache_get_page(inode->i_mapping, 0, FGP_LOCK|FGP_NOWAIT, 0);
1206 if (!PageUptodate(page))
1209 if (!PageDirty(page))
1212 if (!clear_page_dirty_for_io(page))
1215 if (!f2fs_write_inline_data(inode, page))
1216 inode_dec_dirty_pages(inode);
1218 set_page_dirty(page);
1220 f2fs_put_page(page, 1);
1225 int sync_node_pages(struct f2fs_sb_info *sbi, nid_t ino,
1226 struct writeback_control *wbc)
1229 struct pagevec pvec;
1230 int step = ino ? 2 : 0;
1231 int nwritten = 0, wrote = 0;
1233 pagevec_init(&pvec, 0);
1239 while (index <= end) {
1241 nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1242 PAGECACHE_TAG_DIRTY,
1243 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
1247 for (i = 0; i < nr_pages; i++) {
1248 struct page *page = pvec.pages[i];
1250 if (unlikely(f2fs_cp_error(sbi))) {
1251 pagevec_release(&pvec);
1256 * flushing sequence with step:
1261 if (step == 0 && IS_DNODE(page))
1263 if (step == 1 && (!IS_DNODE(page) ||
1264 is_cold_node(page)))
1266 if (step == 2 && (!IS_DNODE(page) ||
1267 !is_cold_node(page)))
1272 * we should not skip writing node pages.
1274 if (ino && ino_of_node(page) == ino)
1276 else if (!trylock_page(page))
1279 if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1284 if (ino && ino_of_node(page) != ino)
1285 goto continue_unlock;
1287 if (!PageDirty(page)) {
1288 /* someone wrote it for us */
1289 goto continue_unlock;
1292 /* flush inline_data */
1293 if (!ino && is_inline_node(page)) {
1294 clear_inline_node(page);
1296 flush_inline_data(sbi, ino_of_node(page));
1300 f2fs_wait_on_page_writeback(page, NODE, true);
1302 BUG_ON(PageWriteback(page));
1303 if (!clear_page_dirty_for_io(page))
1304 goto continue_unlock;
1306 /* called by fsync() */
1307 if (ino && IS_DNODE(page)) {
1308 set_fsync_mark(page, 1);
1310 set_dentry_mark(page,
1311 need_dentry_mark(sbi, ino));
1314 set_fsync_mark(page, 0);
1315 set_dentry_mark(page, 0);
1318 if (NODE_MAPPING(sbi)->a_ops->writepage(page, wbc))
1323 if (--wbc->nr_to_write == 0)
1326 pagevec_release(&pvec);
1329 if (wbc->nr_to_write == 0) {
1342 f2fs_submit_merged_bio_cond(sbi, NULL, NULL,
1345 f2fs_submit_merged_bio(sbi, NODE, WRITE);
1350 int wait_on_node_pages_writeback(struct f2fs_sb_info *sbi, nid_t ino)
1352 pgoff_t index = 0, end = LONG_MAX;
1353 struct pagevec pvec;
1354 int ret2 = 0, ret = 0;
1356 pagevec_init(&pvec, 0);
1358 while (index <= end) {
1360 nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1361 PAGECACHE_TAG_WRITEBACK,
1362 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
1366 for (i = 0; i < nr_pages; i++) {
1367 struct page *page = pvec.pages[i];
1369 /* until radix tree lookup accepts end_index */
1370 if (unlikely(page->index > end))
1373 if (ino && ino_of_node(page) == ino) {
1374 f2fs_wait_on_page_writeback(page, NODE, true);
1375 if (TestClearPageError(page))
1379 pagevec_release(&pvec);
1383 if (unlikely(test_and_clear_bit(AS_ENOSPC, &NODE_MAPPING(sbi)->flags)))
1385 if (unlikely(test_and_clear_bit(AS_EIO, &NODE_MAPPING(sbi)->flags)))
1392 static int f2fs_write_node_page(struct page *page,
1393 struct writeback_control *wbc)
1395 struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1397 struct node_info ni;
1398 struct f2fs_io_info fio = {
1401 .rw = (wbc->sync_mode == WB_SYNC_ALL) ? WRITE_SYNC : WRITE,
1403 .encrypted_page = NULL,
1406 trace_f2fs_writepage(page, NODE);
1408 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
1410 if (unlikely(f2fs_cp_error(sbi)))
1413 /* get old block addr of this node page */
1414 nid = nid_of_node(page);
1415 f2fs_bug_on(sbi, page->index != nid);
1417 if (wbc->for_reclaim) {
1418 if (!down_read_trylock(&sbi->node_write))
1421 down_read(&sbi->node_write);
1424 get_node_info(sbi, nid, &ni);
1426 /* This page is already truncated */
1427 if (unlikely(ni.blk_addr == NULL_ADDR)) {
1428 ClearPageUptodate(page);
1429 dec_page_count(sbi, F2FS_DIRTY_NODES);
1430 up_read(&sbi->node_write);
1435 set_page_writeback(page);
1436 fio.blk_addr = ni.blk_addr;
1437 write_node_page(nid, &fio);
1438 set_node_addr(sbi, &ni, fio.blk_addr, is_fsync_dnode(page));
1439 dec_page_count(sbi, F2FS_DIRTY_NODES);
1440 up_read(&sbi->node_write);
1442 if (wbc->for_reclaim)
1443 f2fs_submit_merged_bio_cond(sbi, NULL, page, 0, NODE, WRITE);
1447 if (unlikely(f2fs_cp_error(sbi)))
1448 f2fs_submit_merged_bio(sbi, NODE, WRITE);
1453 redirty_page_for_writepage(wbc, page);
1454 return AOP_WRITEPAGE_ACTIVATE;
1457 static int f2fs_write_node_pages(struct address_space *mapping,
1458 struct writeback_control *wbc)
1460 struct f2fs_sb_info *sbi = F2FS_M_SB(mapping);
1463 /* balancing f2fs's metadata in background */
1464 f2fs_balance_fs_bg(sbi);
1466 /* collect a number of dirty node pages and write together */
1467 if (get_pages(sbi, F2FS_DIRTY_NODES) < nr_pages_to_skip(sbi, NODE))
1470 trace_f2fs_writepages(mapping->host, wbc, NODE);
1472 diff = nr_pages_to_write(sbi, NODE, wbc);
1473 wbc->sync_mode = WB_SYNC_NONE;
1474 sync_node_pages(sbi, 0, wbc);
1475 wbc->nr_to_write = max((long)0, wbc->nr_to_write - diff);
1479 wbc->pages_skipped += get_pages(sbi, F2FS_DIRTY_NODES);
1480 trace_f2fs_writepages(mapping->host, wbc, NODE);
1484 static int f2fs_set_node_page_dirty(struct page *page)
1486 trace_f2fs_set_page_dirty(page, NODE);
1488 SetPageUptodate(page);
1489 if (!PageDirty(page)) {
1490 __set_page_dirty_nobuffers(page);
1491 inc_page_count(F2FS_P_SB(page), F2FS_DIRTY_NODES);
1492 SetPagePrivate(page);
1493 f2fs_trace_pid(page);
1500 * Structure of the f2fs node operations
1502 const struct address_space_operations f2fs_node_aops = {
1503 .writepage = f2fs_write_node_page,
1504 .writepages = f2fs_write_node_pages,
1505 .set_page_dirty = f2fs_set_node_page_dirty,
1506 .invalidatepage = f2fs_invalidate_page,
1507 .releasepage = f2fs_release_page,
1510 static struct free_nid *__lookup_free_nid_list(struct f2fs_nm_info *nm_i,
1513 return radix_tree_lookup(&nm_i->free_nid_root, n);
1516 static void __del_from_free_nid_list(struct f2fs_nm_info *nm_i,
1520 radix_tree_delete(&nm_i->free_nid_root, i->nid);
1523 static int add_free_nid(struct f2fs_sb_info *sbi, nid_t nid, bool build)
1525 struct f2fs_nm_info *nm_i = NM_I(sbi);
1527 struct nat_entry *ne;
1528 bool allocated = false;
1530 if (!available_free_memory(sbi, FREE_NIDS))
1533 /* 0 nid should not be used */
1534 if (unlikely(nid == 0))
1538 /* do not add allocated nids */
1539 ne = __lookup_nat_cache(nm_i, nid);
1540 if (ne && (!get_nat_flag(ne, IS_CHECKPOINTED) ||
1541 nat_get_blkaddr(ne) != NULL_ADDR))
1547 i = f2fs_kmem_cache_alloc(free_nid_slab, GFP_NOFS);
1551 if (radix_tree_preload(GFP_NOFS)) {
1552 kmem_cache_free(free_nid_slab, i);
1556 spin_lock(&nm_i->free_nid_list_lock);
1557 if (radix_tree_insert(&nm_i->free_nid_root, i->nid, i)) {
1558 spin_unlock(&nm_i->free_nid_list_lock);
1559 radix_tree_preload_end();
1560 kmem_cache_free(free_nid_slab, i);
1563 list_add_tail(&i->list, &nm_i->free_nid_list);
1565 spin_unlock(&nm_i->free_nid_list_lock);
1566 radix_tree_preload_end();
1570 static void remove_free_nid(struct f2fs_nm_info *nm_i, nid_t nid)
1573 bool need_free = false;
1575 spin_lock(&nm_i->free_nid_list_lock);
1576 i = __lookup_free_nid_list(nm_i, nid);
1577 if (i && i->state == NID_NEW) {
1578 __del_from_free_nid_list(nm_i, i);
1582 spin_unlock(&nm_i->free_nid_list_lock);
1585 kmem_cache_free(free_nid_slab, i);
1588 static void scan_nat_page(struct f2fs_sb_info *sbi,
1589 struct page *nat_page, nid_t start_nid)
1591 struct f2fs_nm_info *nm_i = NM_I(sbi);
1592 struct f2fs_nat_block *nat_blk = page_address(nat_page);
1596 i = start_nid % NAT_ENTRY_PER_BLOCK;
1598 for (; i < NAT_ENTRY_PER_BLOCK; i++, start_nid++) {
1600 if (unlikely(start_nid >= nm_i->max_nid))
1603 blk_addr = le32_to_cpu(nat_blk->entries[i].block_addr);
1604 f2fs_bug_on(sbi, blk_addr == NEW_ADDR);
1605 if (blk_addr == NULL_ADDR) {
1606 if (add_free_nid(sbi, start_nid, true) < 0)
1612 static void build_free_nids(struct f2fs_sb_info *sbi)
1614 struct f2fs_nm_info *nm_i = NM_I(sbi);
1615 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
1616 struct f2fs_summary_block *sum = curseg->sum_blk;
1618 nid_t nid = nm_i->next_scan_nid;
1620 /* Enough entries */
1621 if (nm_i->fcnt > NAT_ENTRY_PER_BLOCK)
1624 /* readahead nat pages to be scanned */
1625 ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nid), FREE_NID_PAGES,
1628 down_read(&nm_i->nat_tree_lock);
1631 struct page *page = get_current_nat_page(sbi, nid);
1633 scan_nat_page(sbi, page, nid);
1634 f2fs_put_page(page, 1);
1636 nid += (NAT_ENTRY_PER_BLOCK - (nid % NAT_ENTRY_PER_BLOCK));
1637 if (unlikely(nid >= nm_i->max_nid))
1640 if (++i >= FREE_NID_PAGES)
1644 /* go to the next free nat pages to find free nids abundantly */
1645 nm_i->next_scan_nid = nid;
1647 /* find free nids from current sum_pages */
1648 mutex_lock(&curseg->curseg_mutex);
1649 for (i = 0; i < nats_in_cursum(sum); i++) {
1650 block_t addr = le32_to_cpu(nat_in_journal(sum, i).block_addr);
1651 nid = le32_to_cpu(nid_in_journal(sum, i));
1652 if (addr == NULL_ADDR)
1653 add_free_nid(sbi, nid, true);
1655 remove_free_nid(nm_i, nid);
1657 mutex_unlock(&curseg->curseg_mutex);
1658 up_read(&nm_i->nat_tree_lock);
1660 ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nm_i->next_scan_nid),
1661 nm_i->ra_nid_pages, META_NAT, false);
1665 * If this function returns success, caller can obtain a new nid
1666 * from second parameter of this function.
1667 * The returned nid could be used ino as well as nid when inode is created.
1669 bool alloc_nid(struct f2fs_sb_info *sbi, nid_t *nid)
1671 struct f2fs_nm_info *nm_i = NM_I(sbi);
1672 struct free_nid *i = NULL;
1674 if (unlikely(sbi->total_valid_node_count + 1 > nm_i->available_nids))
1677 spin_lock(&nm_i->free_nid_list_lock);
1679 /* We should not use stale free nids created by build_free_nids */
1680 if (nm_i->fcnt && !on_build_free_nids(nm_i)) {
1681 f2fs_bug_on(sbi, list_empty(&nm_i->free_nid_list));
1682 list_for_each_entry(i, &nm_i->free_nid_list, list)
1683 if (i->state == NID_NEW)
1686 f2fs_bug_on(sbi, i->state != NID_NEW);
1688 i->state = NID_ALLOC;
1690 spin_unlock(&nm_i->free_nid_list_lock);
1693 spin_unlock(&nm_i->free_nid_list_lock);
1695 /* Let's scan nat pages and its caches to get free nids */
1696 mutex_lock(&nm_i->build_lock);
1697 build_free_nids(sbi);
1698 mutex_unlock(&nm_i->build_lock);
1703 * alloc_nid() should be called prior to this function.
1705 void alloc_nid_done(struct f2fs_sb_info *sbi, nid_t nid)
1707 struct f2fs_nm_info *nm_i = NM_I(sbi);
1710 spin_lock(&nm_i->free_nid_list_lock);
1711 i = __lookup_free_nid_list(nm_i, nid);
1712 f2fs_bug_on(sbi, !i || i->state != NID_ALLOC);
1713 __del_from_free_nid_list(nm_i, i);
1714 spin_unlock(&nm_i->free_nid_list_lock);
1716 kmem_cache_free(free_nid_slab, i);
1720 * alloc_nid() should be called prior to this function.
1722 void alloc_nid_failed(struct f2fs_sb_info *sbi, nid_t nid)
1724 struct f2fs_nm_info *nm_i = NM_I(sbi);
1726 bool need_free = false;
1731 spin_lock(&nm_i->free_nid_list_lock);
1732 i = __lookup_free_nid_list(nm_i, nid);
1733 f2fs_bug_on(sbi, !i || i->state != NID_ALLOC);
1734 if (!available_free_memory(sbi, FREE_NIDS)) {
1735 __del_from_free_nid_list(nm_i, i);
1741 spin_unlock(&nm_i->free_nid_list_lock);
1744 kmem_cache_free(free_nid_slab, i);
1747 int try_to_free_nids(struct f2fs_sb_info *sbi, int nr_shrink)
1749 struct f2fs_nm_info *nm_i = NM_I(sbi);
1750 struct free_nid *i, *next;
1753 if (!mutex_trylock(&nm_i->build_lock))
1756 spin_lock(&nm_i->free_nid_list_lock);
1757 list_for_each_entry_safe(i, next, &nm_i->free_nid_list, list) {
1758 if (nr_shrink <= 0 || nm_i->fcnt <= NAT_ENTRY_PER_BLOCK)
1760 if (i->state == NID_ALLOC)
1762 __del_from_free_nid_list(nm_i, i);
1763 kmem_cache_free(free_nid_slab, i);
1767 spin_unlock(&nm_i->free_nid_list_lock);
1768 mutex_unlock(&nm_i->build_lock);
1770 return nr - nr_shrink;
1773 void recover_inline_xattr(struct inode *inode, struct page *page)
1775 void *src_addr, *dst_addr;
1778 struct f2fs_inode *ri;
1780 ipage = get_node_page(F2FS_I_SB(inode), inode->i_ino);
1781 f2fs_bug_on(F2FS_I_SB(inode), IS_ERR(ipage));
1783 ri = F2FS_INODE(page);
1784 if (!(ri->i_inline & F2FS_INLINE_XATTR)) {
1785 clear_inode_flag(F2FS_I(inode), FI_INLINE_XATTR);
1789 dst_addr = inline_xattr_addr(ipage);
1790 src_addr = inline_xattr_addr(page);
1791 inline_size = inline_xattr_size(inode);
1793 f2fs_wait_on_page_writeback(ipage, NODE, true);
1794 memcpy(dst_addr, src_addr, inline_size);
1796 update_inode(inode, ipage);
1797 f2fs_put_page(ipage, 1);
1800 void recover_xattr_data(struct inode *inode, struct page *page, block_t blkaddr)
1802 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1803 nid_t prev_xnid = F2FS_I(inode)->i_xattr_nid;
1804 nid_t new_xnid = nid_of_node(page);
1805 struct node_info ni;
1807 /* 1: invalidate the previous xattr nid */
1811 /* Deallocate node address */
1812 get_node_info(sbi, prev_xnid, &ni);
1813 f2fs_bug_on(sbi, ni.blk_addr == NULL_ADDR);
1814 invalidate_blocks(sbi, ni.blk_addr);
1815 dec_valid_node_count(sbi, inode);
1816 set_node_addr(sbi, &ni, NULL_ADDR, false);
1819 /* 2: allocate new xattr nid */
1820 if (unlikely(!inc_valid_node_count(sbi, inode)))
1821 f2fs_bug_on(sbi, 1);
1823 remove_free_nid(NM_I(sbi), new_xnid);
1824 get_node_info(sbi, new_xnid, &ni);
1825 ni.ino = inode->i_ino;
1826 set_node_addr(sbi, &ni, NEW_ADDR, false);
1827 F2FS_I(inode)->i_xattr_nid = new_xnid;
1829 /* 3: update xattr blkaddr */
1830 refresh_sit_entry(sbi, NEW_ADDR, blkaddr);
1831 set_node_addr(sbi, &ni, blkaddr, false);
1833 update_inode_page(inode);
1836 int recover_inode_page(struct f2fs_sb_info *sbi, struct page *page)
1838 struct f2fs_inode *src, *dst;
1839 nid_t ino = ino_of_node(page);
1840 struct node_info old_ni, new_ni;
1843 get_node_info(sbi, ino, &old_ni);
1845 if (unlikely(old_ni.blk_addr != NULL_ADDR))
1848 ipage = grab_cache_page(NODE_MAPPING(sbi), ino);
1852 /* Should not use this inode from free nid list */
1853 remove_free_nid(NM_I(sbi), ino);
1855 SetPageUptodate(ipage);
1856 fill_node_footer(ipage, ino, ino, 0, true);
1858 src = F2FS_INODE(page);
1859 dst = F2FS_INODE(ipage);
1861 memcpy(dst, src, (unsigned long)&src->i_ext - (unsigned long)src);
1863 dst->i_blocks = cpu_to_le64(1);
1864 dst->i_links = cpu_to_le32(1);
1865 dst->i_xattr_nid = 0;
1866 dst->i_inline = src->i_inline & F2FS_INLINE_XATTR;
1871 if (unlikely(!inc_valid_node_count(sbi, NULL)))
1873 set_node_addr(sbi, &new_ni, NEW_ADDR, false);
1874 inc_valid_inode_count(sbi);
1875 set_page_dirty(ipage);
1876 f2fs_put_page(ipage, 1);
1880 int restore_node_summary(struct f2fs_sb_info *sbi,
1881 unsigned int segno, struct f2fs_summary_block *sum)
1883 struct f2fs_node *rn;
1884 struct f2fs_summary *sum_entry;
1886 int bio_blocks = MAX_BIO_BLOCKS(sbi);
1887 int i, idx, last_offset, nrpages;
1889 /* scan the node segment */
1890 last_offset = sbi->blocks_per_seg;
1891 addr = START_BLOCK(sbi, segno);
1892 sum_entry = &sum->entries[0];
1894 for (i = 0; i < last_offset; i += nrpages, addr += nrpages) {
1895 nrpages = min(last_offset - i, bio_blocks);
1897 /* readahead node pages */
1898 ra_meta_pages(sbi, addr, nrpages, META_POR, true);
1900 for (idx = addr; idx < addr + nrpages; idx++) {
1901 struct page *page = get_tmp_page(sbi, idx);
1903 rn = F2FS_NODE(page);
1904 sum_entry->nid = rn->footer.nid;
1905 sum_entry->version = 0;
1906 sum_entry->ofs_in_node = 0;
1908 f2fs_put_page(page, 1);
1911 invalidate_mapping_pages(META_MAPPING(sbi), addr,
1917 static void remove_nats_in_journal(struct f2fs_sb_info *sbi)
1919 struct f2fs_nm_info *nm_i = NM_I(sbi);
1920 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
1921 struct f2fs_summary_block *sum = curseg->sum_blk;
1924 mutex_lock(&curseg->curseg_mutex);
1925 for (i = 0; i < nats_in_cursum(sum); i++) {
1926 struct nat_entry *ne;
1927 struct f2fs_nat_entry raw_ne;
1928 nid_t nid = le32_to_cpu(nid_in_journal(sum, i));
1930 raw_ne = nat_in_journal(sum, i);
1932 ne = __lookup_nat_cache(nm_i, nid);
1934 ne = grab_nat_entry(nm_i, nid);
1935 node_info_from_raw_nat(&ne->ni, &raw_ne);
1937 __set_nat_cache_dirty(nm_i, ne);
1939 update_nats_in_cursum(sum, -i);
1940 mutex_unlock(&curseg->curseg_mutex);
1943 static void __adjust_nat_entry_set(struct nat_entry_set *nes,
1944 struct list_head *head, int max)
1946 struct nat_entry_set *cur;
1948 if (nes->entry_cnt >= max)
1951 list_for_each_entry(cur, head, set_list) {
1952 if (cur->entry_cnt >= nes->entry_cnt) {
1953 list_add(&nes->set_list, cur->set_list.prev);
1958 list_add_tail(&nes->set_list, head);
1961 static void __flush_nat_entry_set(struct f2fs_sb_info *sbi,
1962 struct nat_entry_set *set)
1964 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
1965 struct f2fs_summary_block *sum = curseg->sum_blk;
1966 nid_t start_nid = set->set * NAT_ENTRY_PER_BLOCK;
1967 bool to_journal = true;
1968 struct f2fs_nat_block *nat_blk;
1969 struct nat_entry *ne, *cur;
1970 struct page *page = NULL;
1973 * there are two steps to flush nat entries:
1974 * #1, flush nat entries to journal in current hot data summary block.
1975 * #2, flush nat entries to nat page.
1977 if (!__has_cursum_space(sum, set->entry_cnt, NAT_JOURNAL))
1981 mutex_lock(&curseg->curseg_mutex);
1983 page = get_next_nat_page(sbi, start_nid);
1984 nat_blk = page_address(page);
1985 f2fs_bug_on(sbi, !nat_blk);
1988 /* flush dirty nats in nat entry set */
1989 list_for_each_entry_safe(ne, cur, &set->entry_list, list) {
1990 struct f2fs_nat_entry *raw_ne;
1991 nid_t nid = nat_get_nid(ne);
1994 if (nat_get_blkaddr(ne) == NEW_ADDR)
1998 offset = lookup_journal_in_cursum(sum,
1999 NAT_JOURNAL, nid, 1);
2000 f2fs_bug_on(sbi, offset < 0);
2001 raw_ne = &nat_in_journal(sum, offset);
2002 nid_in_journal(sum, offset) = cpu_to_le32(nid);
2004 raw_ne = &nat_blk->entries[nid - start_nid];
2006 raw_nat_from_node_info(raw_ne, &ne->ni);
2008 __clear_nat_cache_dirty(NM_I(sbi), ne);
2009 if (nat_get_blkaddr(ne) == NULL_ADDR)
2010 add_free_nid(sbi, nid, false);
2014 mutex_unlock(&curseg->curseg_mutex);
2016 f2fs_put_page(page, 1);
2018 f2fs_bug_on(sbi, set->entry_cnt);
2020 radix_tree_delete(&NM_I(sbi)->nat_set_root, set->set);
2021 kmem_cache_free(nat_entry_set_slab, set);
2025 * This function is called during the checkpointing process.
2027 void flush_nat_entries(struct f2fs_sb_info *sbi)
2029 struct f2fs_nm_info *nm_i = NM_I(sbi);
2030 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
2031 struct f2fs_summary_block *sum = curseg->sum_blk;
2032 struct nat_entry_set *setvec[SETVEC_SIZE];
2033 struct nat_entry_set *set, *tmp;
2038 if (!nm_i->dirty_nat_cnt)
2041 down_write(&nm_i->nat_tree_lock);
2044 * if there are no enough space in journal to store dirty nat
2045 * entries, remove all entries from journal and merge them
2046 * into nat entry set.
2048 if (!__has_cursum_space(sum, nm_i->dirty_nat_cnt, NAT_JOURNAL))
2049 remove_nats_in_journal(sbi);
2051 while ((found = __gang_lookup_nat_set(nm_i,
2052 set_idx, SETVEC_SIZE, setvec))) {
2054 set_idx = setvec[found - 1]->set + 1;
2055 for (idx = 0; idx < found; idx++)
2056 __adjust_nat_entry_set(setvec[idx], &sets,
2057 MAX_NAT_JENTRIES(sum));
2060 /* flush dirty nats in nat entry set */
2061 list_for_each_entry_safe(set, tmp, &sets, set_list)
2062 __flush_nat_entry_set(sbi, set);
2064 up_write(&nm_i->nat_tree_lock);
2066 f2fs_bug_on(sbi, nm_i->dirty_nat_cnt);
2069 static int init_node_manager(struct f2fs_sb_info *sbi)
2071 struct f2fs_super_block *sb_raw = F2FS_RAW_SUPER(sbi);
2072 struct f2fs_nm_info *nm_i = NM_I(sbi);
2073 unsigned char *version_bitmap;
2074 unsigned int nat_segs, nat_blocks;
2076 nm_i->nat_blkaddr = le32_to_cpu(sb_raw->nat_blkaddr);
2078 /* segment_count_nat includes pair segment so divide to 2. */
2079 nat_segs = le32_to_cpu(sb_raw->segment_count_nat) >> 1;
2080 nat_blocks = nat_segs << le32_to_cpu(sb_raw->log_blocks_per_seg);
2082 nm_i->max_nid = NAT_ENTRY_PER_BLOCK * nat_blocks;
2084 /* not used nids: 0, node, meta, (and root counted as valid node) */
2085 nm_i->available_nids = nm_i->max_nid - F2FS_RESERVED_NODE_NUM;
2088 nm_i->ram_thresh = DEF_RAM_THRESHOLD;
2089 nm_i->ra_nid_pages = DEF_RA_NID_PAGES;
2090 nm_i->dirty_nats_ratio = DEF_DIRTY_NAT_RATIO_THRESHOLD;
2092 INIT_RADIX_TREE(&nm_i->free_nid_root, GFP_ATOMIC);
2093 INIT_LIST_HEAD(&nm_i->free_nid_list);
2094 INIT_RADIX_TREE(&nm_i->nat_root, GFP_NOIO);
2095 INIT_RADIX_TREE(&nm_i->nat_set_root, GFP_NOIO);
2096 INIT_LIST_HEAD(&nm_i->nat_entries);
2098 mutex_init(&nm_i->build_lock);
2099 spin_lock_init(&nm_i->free_nid_list_lock);
2100 init_rwsem(&nm_i->nat_tree_lock);
2102 nm_i->next_scan_nid = le32_to_cpu(sbi->ckpt->next_free_nid);
2103 nm_i->bitmap_size = __bitmap_size(sbi, NAT_BITMAP);
2104 version_bitmap = __bitmap_ptr(sbi, NAT_BITMAP);
2105 if (!version_bitmap)
2108 nm_i->nat_bitmap = kmemdup(version_bitmap, nm_i->bitmap_size,
2110 if (!nm_i->nat_bitmap)
2115 int build_node_manager(struct f2fs_sb_info *sbi)
2119 sbi->nm_info = kzalloc(sizeof(struct f2fs_nm_info), GFP_KERNEL);
2123 err = init_node_manager(sbi);
2127 build_free_nids(sbi);
2131 void destroy_node_manager(struct f2fs_sb_info *sbi)
2133 struct f2fs_nm_info *nm_i = NM_I(sbi);
2134 struct free_nid *i, *next_i;
2135 struct nat_entry *natvec[NATVEC_SIZE];
2136 struct nat_entry_set *setvec[SETVEC_SIZE];
2143 /* destroy free nid list */
2144 spin_lock(&nm_i->free_nid_list_lock);
2145 list_for_each_entry_safe(i, next_i, &nm_i->free_nid_list, list) {
2146 f2fs_bug_on(sbi, i->state == NID_ALLOC);
2147 __del_from_free_nid_list(nm_i, i);
2149 spin_unlock(&nm_i->free_nid_list_lock);
2150 kmem_cache_free(free_nid_slab, i);
2151 spin_lock(&nm_i->free_nid_list_lock);
2153 f2fs_bug_on(sbi, nm_i->fcnt);
2154 spin_unlock(&nm_i->free_nid_list_lock);
2156 /* destroy nat cache */
2157 down_write(&nm_i->nat_tree_lock);
2158 while ((found = __gang_lookup_nat_cache(nm_i,
2159 nid, NATVEC_SIZE, natvec))) {
2162 nid = nat_get_nid(natvec[found - 1]) + 1;
2163 for (idx = 0; idx < found; idx++)
2164 __del_from_nat_cache(nm_i, natvec[idx]);
2166 f2fs_bug_on(sbi, nm_i->nat_cnt);
2168 /* destroy nat set cache */
2170 while ((found = __gang_lookup_nat_set(nm_i,
2171 nid, SETVEC_SIZE, setvec))) {
2174 nid = setvec[found - 1]->set + 1;
2175 for (idx = 0; idx < found; idx++) {
2176 /* entry_cnt is not zero, when cp_error was occurred */
2177 f2fs_bug_on(sbi, !list_empty(&setvec[idx]->entry_list));
2178 radix_tree_delete(&nm_i->nat_set_root, setvec[idx]->set);
2179 kmem_cache_free(nat_entry_set_slab, setvec[idx]);
2182 up_write(&nm_i->nat_tree_lock);
2184 kfree(nm_i->nat_bitmap);
2185 sbi->nm_info = NULL;
2189 int __init create_node_manager_caches(void)
2191 nat_entry_slab = f2fs_kmem_cache_create("nat_entry",
2192 sizeof(struct nat_entry));
2193 if (!nat_entry_slab)
2196 free_nid_slab = f2fs_kmem_cache_create("free_nid",
2197 sizeof(struct free_nid));
2199 goto destroy_nat_entry;
2201 nat_entry_set_slab = f2fs_kmem_cache_create("nat_entry_set",
2202 sizeof(struct nat_entry_set));
2203 if (!nat_entry_set_slab)
2204 goto destroy_free_nid;
2208 kmem_cache_destroy(free_nid_slab);
2210 kmem_cache_destroy(nat_entry_slab);
2215 void destroy_node_manager_caches(void)
2217 kmem_cache_destroy(nat_entry_set_slab);
2218 kmem_cache_destroy(free_nid_slab);
2219 kmem_cache_destroy(nat_entry_slab);