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_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_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_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_sb_info *sbi, nid_t nid,
261 struct f2fs_nat_entry *ne)
263 struct f2fs_nm_info *nm_i = NM_I(sbi);
266 e = __lookup_nat_cache(nm_i, nid);
268 e = grab_nat_entry(nm_i, nid);
269 node_info_from_raw_nat(&e->ni, ne);
271 f2fs_bug_on(sbi, nat_get_ino(e) != ne->ino ||
272 nat_get_blkaddr(e) != ne->block_addr ||
273 nat_get_version(e) != ne->version);
277 static void set_node_addr(struct f2fs_sb_info *sbi, struct node_info *ni,
278 block_t new_blkaddr, bool fsync_done)
280 struct f2fs_nm_info *nm_i = NM_I(sbi);
283 down_write(&nm_i->nat_tree_lock);
284 e = __lookup_nat_cache(nm_i, ni->nid);
286 e = grab_nat_entry(nm_i, ni->nid);
287 copy_node_info(&e->ni, ni);
288 f2fs_bug_on(sbi, ni->blk_addr == NEW_ADDR);
289 } else if (new_blkaddr == NEW_ADDR) {
291 * when nid is reallocated,
292 * previous nat entry can be remained in nat cache.
293 * So, reinitialize it with new information.
295 copy_node_info(&e->ni, ni);
296 f2fs_bug_on(sbi, ni->blk_addr != NULL_ADDR);
300 f2fs_bug_on(sbi, nat_get_blkaddr(e) != ni->blk_addr);
301 f2fs_bug_on(sbi, nat_get_blkaddr(e) == NULL_ADDR &&
302 new_blkaddr == NULL_ADDR);
303 f2fs_bug_on(sbi, nat_get_blkaddr(e) == NEW_ADDR &&
304 new_blkaddr == NEW_ADDR);
305 f2fs_bug_on(sbi, nat_get_blkaddr(e) != NEW_ADDR &&
306 nat_get_blkaddr(e) != NULL_ADDR &&
307 new_blkaddr == NEW_ADDR);
309 /* increment version no as node is removed */
310 if (nat_get_blkaddr(e) != NEW_ADDR && new_blkaddr == NULL_ADDR) {
311 unsigned char version = nat_get_version(e);
312 nat_set_version(e, inc_node_version(version));
314 /* in order to reuse the nid */
315 if (nm_i->next_scan_nid > ni->nid)
316 nm_i->next_scan_nid = ni->nid;
320 nat_set_blkaddr(e, new_blkaddr);
321 if (new_blkaddr == NEW_ADDR || new_blkaddr == NULL_ADDR)
322 set_nat_flag(e, IS_CHECKPOINTED, false);
323 __set_nat_cache_dirty(nm_i, e);
325 /* update fsync_mark if its inode nat entry is still alive */
326 if (ni->nid != ni->ino)
327 e = __lookup_nat_cache(nm_i, ni->ino);
329 if (fsync_done && ni->nid == ni->ino)
330 set_nat_flag(e, HAS_FSYNCED_INODE, true);
331 set_nat_flag(e, HAS_LAST_FSYNC, fsync_done);
333 up_write(&nm_i->nat_tree_lock);
336 int try_to_free_nats(struct f2fs_sb_info *sbi, int nr_shrink)
338 struct f2fs_nm_info *nm_i = NM_I(sbi);
341 if (!down_write_trylock(&nm_i->nat_tree_lock))
344 while (nr_shrink && !list_empty(&nm_i->nat_entries)) {
345 struct nat_entry *ne;
346 ne = list_first_entry(&nm_i->nat_entries,
347 struct nat_entry, list);
348 __del_from_nat_cache(nm_i, ne);
351 up_write(&nm_i->nat_tree_lock);
352 return nr - nr_shrink;
356 * This function always returns success
358 void get_node_info(struct f2fs_sb_info *sbi, nid_t nid, struct node_info *ni)
360 struct f2fs_nm_info *nm_i = NM_I(sbi);
361 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
362 struct f2fs_journal *journal = curseg->journal;
363 nid_t start_nid = START_NID(nid);
364 struct f2fs_nat_block *nat_blk;
365 struct page *page = NULL;
366 struct f2fs_nat_entry ne;
372 /* Check nat cache */
373 down_read(&nm_i->nat_tree_lock);
374 e = __lookup_nat_cache(nm_i, nid);
376 ni->ino = nat_get_ino(e);
377 ni->blk_addr = nat_get_blkaddr(e);
378 ni->version = nat_get_version(e);
379 up_read(&nm_i->nat_tree_lock);
383 memset(&ne, 0, sizeof(struct f2fs_nat_entry));
385 /* Check current segment summary */
386 down_read(&curseg->journal_rwsem);
387 i = lookup_journal_in_cursum(journal, NAT_JOURNAL, nid, 0);
389 ne = nat_in_journal(journal, i);
390 node_info_from_raw_nat(ni, &ne);
392 up_read(&curseg->journal_rwsem);
396 /* Fill node_info from nat page */
397 page = get_current_nat_page(sbi, start_nid);
398 nat_blk = (struct f2fs_nat_block *)page_address(page);
399 ne = nat_blk->entries[nid - start_nid];
400 node_info_from_raw_nat(ni, &ne);
401 f2fs_put_page(page, 1);
403 up_read(&nm_i->nat_tree_lock);
404 /* cache nat entry */
405 down_write(&nm_i->nat_tree_lock);
406 cache_nat_entry(sbi, nid, &ne);
407 up_write(&nm_i->nat_tree_lock);
411 * readahead MAX_RA_NODE number of node pages.
413 static void ra_node_pages(struct page *parent, int start, int n)
415 struct f2fs_sb_info *sbi = F2FS_P_SB(parent);
416 struct blk_plug plug;
420 blk_start_plug(&plug);
422 /* Then, try readahead for siblings of the desired node */
424 end = min(end, NIDS_PER_BLOCK);
425 for (i = start; i < end; i++) {
426 nid = get_nid(parent, i, false);
427 ra_node_page(sbi, nid);
430 blk_finish_plug(&plug);
433 pgoff_t get_next_page_offset(struct dnode_of_data *dn, pgoff_t pgofs)
435 const long direct_index = ADDRS_PER_INODE(dn->inode);
436 const long direct_blks = ADDRS_PER_BLOCK;
437 const long indirect_blks = ADDRS_PER_BLOCK * NIDS_PER_BLOCK;
438 unsigned int skipped_unit = ADDRS_PER_BLOCK;
439 int cur_level = dn->cur_level;
440 int max_level = dn->max_level;
446 while (max_level-- > cur_level)
447 skipped_unit *= NIDS_PER_BLOCK;
449 switch (dn->max_level) {
451 base += 2 * indirect_blks;
453 base += 2 * direct_blks;
455 base += direct_index;
458 f2fs_bug_on(F2FS_I_SB(dn->inode), 1);
461 return ((pgofs - base) / skipped_unit + 1) * skipped_unit + base;
465 * The maximum depth is four.
466 * Offset[0] will have raw inode offset.
468 static int get_node_path(struct inode *inode, long block,
469 int offset[4], unsigned int noffset[4])
471 const long direct_index = ADDRS_PER_INODE(inode);
472 const long direct_blks = ADDRS_PER_BLOCK;
473 const long dptrs_per_blk = NIDS_PER_BLOCK;
474 const long indirect_blks = ADDRS_PER_BLOCK * NIDS_PER_BLOCK;
475 const long dindirect_blks = indirect_blks * NIDS_PER_BLOCK;
481 if (block < direct_index) {
485 block -= direct_index;
486 if (block < direct_blks) {
487 offset[n++] = NODE_DIR1_BLOCK;
493 block -= direct_blks;
494 if (block < direct_blks) {
495 offset[n++] = NODE_DIR2_BLOCK;
501 block -= direct_blks;
502 if (block < indirect_blks) {
503 offset[n++] = NODE_IND1_BLOCK;
505 offset[n++] = block / direct_blks;
506 noffset[n] = 4 + offset[n - 1];
507 offset[n] = block % direct_blks;
511 block -= indirect_blks;
512 if (block < indirect_blks) {
513 offset[n++] = NODE_IND2_BLOCK;
514 noffset[n] = 4 + dptrs_per_blk;
515 offset[n++] = block / direct_blks;
516 noffset[n] = 5 + dptrs_per_blk + offset[n - 1];
517 offset[n] = block % direct_blks;
521 block -= indirect_blks;
522 if (block < dindirect_blks) {
523 offset[n++] = NODE_DIND_BLOCK;
524 noffset[n] = 5 + (dptrs_per_blk * 2);
525 offset[n++] = block / indirect_blks;
526 noffset[n] = 6 + (dptrs_per_blk * 2) +
527 offset[n - 1] * (dptrs_per_blk + 1);
528 offset[n++] = (block / direct_blks) % dptrs_per_blk;
529 noffset[n] = 7 + (dptrs_per_blk * 2) +
530 offset[n - 2] * (dptrs_per_blk + 1) +
532 offset[n] = block % direct_blks;
543 * Caller should call f2fs_put_dnode(dn).
544 * Also, it should grab and release a rwsem by calling f2fs_lock_op() and
545 * f2fs_unlock_op() only if ro is not set RDONLY_NODE.
546 * In the case of RDONLY_NODE, we don't need to care about mutex.
548 int get_dnode_of_data(struct dnode_of_data *dn, pgoff_t index, int mode)
550 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
551 struct page *npage[4];
552 struct page *parent = NULL;
554 unsigned int noffset[4];
559 level = get_node_path(dn->inode, index, offset, noffset);
561 nids[0] = dn->inode->i_ino;
562 npage[0] = dn->inode_page;
565 npage[0] = get_node_page(sbi, nids[0]);
566 if (IS_ERR(npage[0]))
567 return PTR_ERR(npage[0]);
570 /* if inline_data is set, should not report any block indices */
571 if (f2fs_has_inline_data(dn->inode) && index) {
573 f2fs_put_page(npage[0], 1);
579 nids[1] = get_nid(parent, offset[0], true);
580 dn->inode_page = npage[0];
581 dn->inode_page_locked = true;
583 /* get indirect or direct nodes */
584 for (i = 1; i <= level; i++) {
587 if (!nids[i] && mode == ALLOC_NODE) {
589 if (!alloc_nid(sbi, &(nids[i]))) {
595 npage[i] = new_node_page(dn, noffset[i], NULL);
596 if (IS_ERR(npage[i])) {
597 alloc_nid_failed(sbi, nids[i]);
598 err = PTR_ERR(npage[i]);
602 set_nid(parent, offset[i - 1], nids[i], i == 1);
603 alloc_nid_done(sbi, nids[i]);
605 } else if (mode == LOOKUP_NODE_RA && i == level && level > 1) {
606 npage[i] = get_node_page_ra(parent, offset[i - 1]);
607 if (IS_ERR(npage[i])) {
608 err = PTR_ERR(npage[i]);
614 dn->inode_page_locked = false;
617 f2fs_put_page(parent, 1);
621 npage[i] = get_node_page(sbi, nids[i]);
622 if (IS_ERR(npage[i])) {
623 err = PTR_ERR(npage[i]);
624 f2fs_put_page(npage[0], 0);
630 nids[i + 1] = get_nid(parent, offset[i], false);
633 dn->nid = nids[level];
634 dn->ofs_in_node = offset[level];
635 dn->node_page = npage[level];
636 dn->data_blkaddr = datablock_addr(dn->node_page, dn->ofs_in_node);
640 f2fs_put_page(parent, 1);
642 f2fs_put_page(npage[0], 0);
644 dn->inode_page = NULL;
645 dn->node_page = NULL;
646 if (err == -ENOENT) {
648 dn->max_level = level;
653 static void truncate_node(struct dnode_of_data *dn)
655 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
658 get_node_info(sbi, dn->nid, &ni);
659 if (dn->inode->i_blocks == 0) {
660 f2fs_bug_on(sbi, ni.blk_addr != NULL_ADDR);
663 f2fs_bug_on(sbi, ni.blk_addr == NULL_ADDR);
665 /* Deallocate node address */
666 invalidate_blocks(sbi, ni.blk_addr);
667 dec_valid_node_count(sbi, dn->inode);
668 set_node_addr(sbi, &ni, NULL_ADDR, false);
670 if (dn->nid == dn->inode->i_ino) {
671 remove_orphan_inode(sbi, dn->nid);
672 dec_valid_inode_count(sbi);
677 clear_node_page_dirty(dn->node_page);
678 set_sbi_flag(sbi, SBI_IS_DIRTY);
680 f2fs_put_page(dn->node_page, 1);
682 invalidate_mapping_pages(NODE_MAPPING(sbi),
683 dn->node_page->index, dn->node_page->index);
685 dn->node_page = NULL;
686 trace_f2fs_truncate_node(dn->inode, dn->nid, ni.blk_addr);
689 static int truncate_dnode(struct dnode_of_data *dn)
696 /* get direct node */
697 page = get_node_page(F2FS_I_SB(dn->inode), dn->nid);
698 if (IS_ERR(page) && PTR_ERR(page) == -ENOENT)
700 else if (IS_ERR(page))
701 return PTR_ERR(page);
703 /* Make dnode_of_data for parameter */
704 dn->node_page = page;
706 truncate_data_blocks(dn);
711 static int truncate_nodes(struct dnode_of_data *dn, unsigned int nofs,
714 struct dnode_of_data rdn = *dn;
716 struct f2fs_node *rn;
718 unsigned int child_nofs;
723 return NIDS_PER_BLOCK + 1;
725 trace_f2fs_truncate_nodes_enter(dn->inode, dn->nid, dn->data_blkaddr);
727 page = get_node_page(F2FS_I_SB(dn->inode), dn->nid);
729 trace_f2fs_truncate_nodes_exit(dn->inode, PTR_ERR(page));
730 return PTR_ERR(page);
733 ra_node_pages(page, ofs, NIDS_PER_BLOCK);
735 rn = F2FS_NODE(page);
737 for (i = ofs; i < NIDS_PER_BLOCK; i++, freed++) {
738 child_nid = le32_to_cpu(rn->in.nid[i]);
742 ret = truncate_dnode(&rdn);
745 if (set_nid(page, i, 0, false))
746 dn->node_changed = true;
749 child_nofs = nofs + ofs * (NIDS_PER_BLOCK + 1) + 1;
750 for (i = ofs; i < NIDS_PER_BLOCK; i++) {
751 child_nid = le32_to_cpu(rn->in.nid[i]);
752 if (child_nid == 0) {
753 child_nofs += NIDS_PER_BLOCK + 1;
757 ret = truncate_nodes(&rdn, child_nofs, 0, depth - 1);
758 if (ret == (NIDS_PER_BLOCK + 1)) {
759 if (set_nid(page, i, 0, false))
760 dn->node_changed = true;
762 } else if (ret < 0 && ret != -ENOENT) {
770 /* remove current indirect node */
771 dn->node_page = page;
775 f2fs_put_page(page, 1);
777 trace_f2fs_truncate_nodes_exit(dn->inode, freed);
781 f2fs_put_page(page, 1);
782 trace_f2fs_truncate_nodes_exit(dn->inode, ret);
786 static int truncate_partial_nodes(struct dnode_of_data *dn,
787 struct f2fs_inode *ri, int *offset, int depth)
789 struct page *pages[2];
796 nid[0] = le32_to_cpu(ri->i_nid[offset[0] - NODE_DIR1_BLOCK]);
800 /* get indirect nodes in the path */
801 for (i = 0; i < idx + 1; i++) {
802 /* reference count'll be increased */
803 pages[i] = get_node_page(F2FS_I_SB(dn->inode), nid[i]);
804 if (IS_ERR(pages[i])) {
805 err = PTR_ERR(pages[i]);
809 nid[i + 1] = get_nid(pages[i], offset[i + 1], false);
812 ra_node_pages(pages[idx], offset[idx + 1], NIDS_PER_BLOCK);
814 /* free direct nodes linked to a partial indirect node */
815 for (i = offset[idx + 1]; i < NIDS_PER_BLOCK; i++) {
816 child_nid = get_nid(pages[idx], i, false);
820 err = truncate_dnode(dn);
823 if (set_nid(pages[idx], i, 0, false))
824 dn->node_changed = true;
827 if (offset[idx + 1] == 0) {
828 dn->node_page = pages[idx];
832 f2fs_put_page(pages[idx], 1);
838 for (i = idx; i >= 0; i--)
839 f2fs_put_page(pages[i], 1);
841 trace_f2fs_truncate_partial_nodes(dn->inode, nid, depth, err);
847 * All the block addresses of data and nodes should be nullified.
849 int truncate_inode_blocks(struct inode *inode, pgoff_t from)
851 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
852 int err = 0, cont = 1;
853 int level, offset[4], noffset[4];
854 unsigned int nofs = 0;
855 struct f2fs_inode *ri;
856 struct dnode_of_data dn;
859 trace_f2fs_truncate_inode_blocks_enter(inode, from);
861 level = get_node_path(inode, from, offset, noffset);
863 page = get_node_page(sbi, inode->i_ino);
865 trace_f2fs_truncate_inode_blocks_exit(inode, PTR_ERR(page));
866 return PTR_ERR(page);
869 set_new_dnode(&dn, inode, page, NULL, 0);
872 ri = F2FS_INODE(page);
880 if (!offset[level - 1])
882 err = truncate_partial_nodes(&dn, ri, offset, level);
883 if (err < 0 && err != -ENOENT)
885 nofs += 1 + NIDS_PER_BLOCK;
888 nofs = 5 + 2 * NIDS_PER_BLOCK;
889 if (!offset[level - 1])
891 err = truncate_partial_nodes(&dn, ri, offset, level);
892 if (err < 0 && err != -ENOENT)
901 dn.nid = le32_to_cpu(ri->i_nid[offset[0] - NODE_DIR1_BLOCK]);
903 case NODE_DIR1_BLOCK:
904 case NODE_DIR2_BLOCK:
905 err = truncate_dnode(&dn);
908 case NODE_IND1_BLOCK:
909 case NODE_IND2_BLOCK:
910 err = truncate_nodes(&dn, nofs, offset[1], 2);
913 case NODE_DIND_BLOCK:
914 err = truncate_nodes(&dn, nofs, offset[1], 3);
921 if (err < 0 && err != -ENOENT)
923 if (offset[1] == 0 &&
924 ri->i_nid[offset[0] - NODE_DIR1_BLOCK]) {
926 BUG_ON(page->mapping != NODE_MAPPING(sbi));
927 f2fs_wait_on_page_writeback(page, NODE, true);
928 ri->i_nid[offset[0] - NODE_DIR1_BLOCK] = 0;
929 set_page_dirty(page);
937 f2fs_put_page(page, 0);
938 trace_f2fs_truncate_inode_blocks_exit(inode, err);
939 return err > 0 ? 0 : err;
942 int truncate_xattr_node(struct inode *inode, struct page *page)
944 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
945 nid_t nid = F2FS_I(inode)->i_xattr_nid;
946 struct dnode_of_data dn;
952 npage = get_node_page(sbi, nid);
954 return PTR_ERR(npage);
956 F2FS_I(inode)->i_xattr_nid = 0;
958 /* need to do checkpoint during fsync */
959 F2FS_I(inode)->xattr_ver = cur_cp_version(F2FS_CKPT(sbi));
961 set_new_dnode(&dn, inode, page, npage, nid);
964 dn.inode_page_locked = true;
970 * Caller should grab and release a rwsem by calling f2fs_lock_op() and
973 int remove_inode_page(struct inode *inode)
975 struct dnode_of_data dn;
978 set_new_dnode(&dn, inode, NULL, NULL, inode->i_ino);
979 err = get_dnode_of_data(&dn, 0, LOOKUP_NODE);
983 err = truncate_xattr_node(inode, dn.inode_page);
989 /* remove potential inline_data blocks */
990 if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
991 S_ISLNK(inode->i_mode))
992 truncate_data_blocks_range(&dn, 1);
994 /* 0 is possible, after f2fs_new_inode() has failed */
995 f2fs_bug_on(F2FS_I_SB(inode),
996 inode->i_blocks != 0 && inode->i_blocks != 1);
998 /* will put inode & node pages */
1003 struct page *new_inode_page(struct inode *inode)
1005 struct dnode_of_data dn;
1007 /* allocate inode page for new inode */
1008 set_new_dnode(&dn, inode, NULL, NULL, inode->i_ino);
1010 /* caller should f2fs_put_page(page, 1); */
1011 return new_node_page(&dn, 0, NULL);
1014 struct page *new_node_page(struct dnode_of_data *dn,
1015 unsigned int ofs, struct page *ipage)
1017 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
1018 struct node_info old_ni, new_ni;
1022 if (unlikely(is_inode_flag_set(F2FS_I(dn->inode), FI_NO_ALLOC)))
1023 return ERR_PTR(-EPERM);
1025 page = f2fs_grab_cache_page(NODE_MAPPING(sbi), dn->nid, false);
1027 return ERR_PTR(-ENOMEM);
1029 if (unlikely(!inc_valid_node_count(sbi, dn->inode))) {
1034 get_node_info(sbi, dn->nid, &old_ni);
1036 /* Reinitialize old_ni with new node page */
1037 f2fs_bug_on(sbi, old_ni.blk_addr != NULL_ADDR);
1039 new_ni.ino = dn->inode->i_ino;
1040 set_node_addr(sbi, &new_ni, NEW_ADDR, false);
1042 f2fs_wait_on_page_writeback(page, NODE, true);
1043 fill_node_footer(page, dn->nid, dn->inode->i_ino, ofs, true);
1044 set_cold_node(dn->inode, page);
1045 SetPageUptodate(page);
1046 if (set_page_dirty(page))
1047 dn->node_changed = true;
1049 if (f2fs_has_xattr_block(ofs))
1050 F2FS_I(dn->inode)->i_xattr_nid = dn->nid;
1052 dn->node_page = page;
1054 update_inode(dn->inode, ipage);
1056 sync_inode_page(dn);
1058 inc_valid_inode_count(sbi);
1063 clear_node_page_dirty(page);
1064 f2fs_put_page(page, 1);
1065 return ERR_PTR(err);
1069 * Caller should do after getting the following values.
1070 * 0: f2fs_put_page(page, 0)
1071 * LOCKED_PAGE or error: f2fs_put_page(page, 1)
1073 static int read_node_page(struct page *page, int rw)
1075 struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1076 struct node_info ni;
1077 struct f2fs_io_info fio = {
1082 .encrypted_page = NULL,
1085 get_node_info(sbi, page->index, &ni);
1087 if (unlikely(ni.blk_addr == NULL_ADDR)) {
1088 ClearPageUptodate(page);
1092 if (PageUptodate(page))
1095 fio.new_blkaddr = fio.old_blkaddr = ni.blk_addr;
1096 return f2fs_submit_page_bio(&fio);
1100 * Readahead a node page
1102 void ra_node_page(struct f2fs_sb_info *sbi, nid_t nid)
1109 f2fs_bug_on(sbi, check_nid_range(sbi, nid));
1112 apage = radix_tree_lookup(&NODE_MAPPING(sbi)->page_tree, nid);
1117 apage = f2fs_grab_cache_page(NODE_MAPPING(sbi), nid, false);
1121 err = read_node_page(apage, READA);
1122 f2fs_put_page(apage, err ? 1 : 0);
1125 static struct page *__get_node_page(struct f2fs_sb_info *sbi, pgoff_t nid,
1126 struct page *parent, int start)
1132 return ERR_PTR(-ENOENT);
1133 f2fs_bug_on(sbi, check_nid_range(sbi, nid));
1135 page = f2fs_grab_cache_page(NODE_MAPPING(sbi), nid, false);
1137 return ERR_PTR(-ENOMEM);
1139 err = read_node_page(page, READ_SYNC);
1141 f2fs_put_page(page, 1);
1142 return ERR_PTR(err);
1143 } else if (err == LOCKED_PAGE) {
1148 ra_node_pages(parent, start + 1, MAX_RA_NODE);
1152 if (unlikely(!PageUptodate(page))) {
1153 f2fs_put_page(page, 1);
1154 return ERR_PTR(-EIO);
1156 if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1157 f2fs_put_page(page, 1);
1161 f2fs_bug_on(sbi, nid != nid_of_node(page));
1165 struct page *get_node_page(struct f2fs_sb_info *sbi, pgoff_t nid)
1167 return __get_node_page(sbi, nid, NULL, 0);
1170 struct page *get_node_page_ra(struct page *parent, int start)
1172 struct f2fs_sb_info *sbi = F2FS_P_SB(parent);
1173 nid_t nid = get_nid(parent, start, false);
1175 return __get_node_page(sbi, nid, parent, start);
1178 void sync_inode_page(struct dnode_of_data *dn)
1182 if (IS_INODE(dn->node_page) || dn->inode_page == dn->node_page) {
1183 ret = update_inode(dn->inode, dn->node_page);
1184 } else if (dn->inode_page) {
1185 if (!dn->inode_page_locked)
1186 lock_page(dn->inode_page);
1187 ret = update_inode(dn->inode, dn->inode_page);
1188 if (!dn->inode_page_locked)
1189 unlock_page(dn->inode_page);
1191 ret = update_inode_page(dn->inode);
1193 dn->node_changed = ret ? true: false;
1196 static void flush_inline_data(struct f2fs_sb_info *sbi, nid_t ino)
1198 struct inode *inode;
1202 /* should flush inline_data before evict_inode */
1203 inode = ilookup(sbi->sb, ino);
1207 page = pagecache_get_page(inode->i_mapping, 0, FGP_LOCK|FGP_NOWAIT, 0);
1211 if (!PageUptodate(page))
1214 if (!PageDirty(page))
1217 if (!clear_page_dirty_for_io(page))
1220 ret = f2fs_write_inline_data(inode, page);
1221 inode_dec_dirty_pages(inode);
1223 set_page_dirty(page);
1225 f2fs_put_page(page, 1);
1230 void move_node_page(struct page *node_page, int gc_type)
1232 if (gc_type == FG_GC) {
1233 struct f2fs_sb_info *sbi = F2FS_P_SB(node_page);
1234 struct writeback_control wbc = {
1235 .sync_mode = WB_SYNC_ALL,
1240 set_page_dirty(node_page);
1241 f2fs_wait_on_page_writeback(node_page, NODE, true);
1243 f2fs_bug_on(sbi, PageWriteback(node_page));
1244 if (!clear_page_dirty_for_io(node_page))
1247 if (NODE_MAPPING(sbi)->a_ops->writepage(node_page, &wbc))
1248 unlock_page(node_page);
1251 /* set page dirty and write it */
1252 if (!PageWriteback(node_page))
1253 set_page_dirty(node_page);
1256 unlock_page(node_page);
1258 f2fs_put_page(node_page, 0);
1261 static struct page *last_fsync_dnode(struct f2fs_sb_info *sbi, nid_t ino)
1264 struct pagevec pvec;
1265 struct page *last_page = NULL;
1267 pagevec_init(&pvec, 0);
1271 while (index <= end) {
1273 nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1274 PAGECACHE_TAG_DIRTY,
1275 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
1279 for (i = 0; i < nr_pages; i++) {
1280 struct page *page = pvec.pages[i];
1282 if (unlikely(f2fs_cp_error(sbi))) {
1283 f2fs_put_page(last_page, 0);
1284 pagevec_release(&pvec);
1285 return ERR_PTR(-EIO);
1288 if (!IS_DNODE(page) || !is_cold_node(page))
1290 if (ino_of_node(page) != ino)
1295 if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1300 if (ino_of_node(page) != ino)
1301 goto continue_unlock;
1303 if (!PageDirty(page)) {
1304 /* someone wrote it for us */
1305 goto continue_unlock;
1309 f2fs_put_page(last_page, 0);
1315 pagevec_release(&pvec);
1321 int fsync_node_pages(struct f2fs_sb_info *sbi, nid_t ino,
1322 struct writeback_control *wbc, bool atomic)
1325 struct pagevec pvec;
1327 struct page *last_page = NULL;
1328 bool marked = false;
1331 last_page = last_fsync_dnode(sbi, ino);
1332 if (IS_ERR_OR_NULL(last_page))
1333 return PTR_ERR_OR_ZERO(last_page);
1336 pagevec_init(&pvec, 0);
1340 while (index <= end) {
1342 nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1343 PAGECACHE_TAG_DIRTY,
1344 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
1348 for (i = 0; i < nr_pages; i++) {
1349 struct page *page = pvec.pages[i];
1351 if (unlikely(f2fs_cp_error(sbi))) {
1352 f2fs_put_page(last_page, 0);
1353 pagevec_release(&pvec);
1357 if (!IS_DNODE(page) || !is_cold_node(page))
1359 if (ino_of_node(page) != ino)
1364 if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1369 if (ino_of_node(page) != ino)
1370 goto continue_unlock;
1372 if (!PageDirty(page) && page != last_page) {
1373 /* someone wrote it for us */
1374 goto continue_unlock;
1377 f2fs_wait_on_page_writeback(page, NODE, true);
1378 BUG_ON(PageWriteback(page));
1380 if (!atomic || page == last_page) {
1381 set_fsync_mark(page, 1);
1383 set_dentry_mark(page,
1384 need_dentry_mark(sbi, ino));
1385 /* may be written by other thread */
1386 if (!PageDirty(page))
1387 set_page_dirty(page);
1390 if (!clear_page_dirty_for_io(page))
1391 goto continue_unlock;
1393 ret = NODE_MAPPING(sbi)->a_ops->writepage(page, wbc);
1396 f2fs_put_page(last_page, 0);
1399 if (page == last_page) {
1400 f2fs_put_page(page, 0);
1405 pagevec_release(&pvec);
1411 if (!ret && atomic && !marked) {
1412 f2fs_msg(sbi->sb, KERN_DEBUG,
1413 "Retry to write fsync mark: ino=%u, idx=%lx",
1414 ino, last_page->index);
1415 lock_page(last_page);
1416 set_page_dirty(last_page);
1417 unlock_page(last_page);
1420 return ret ? -EIO: 0;
1423 int sync_node_pages(struct f2fs_sb_info *sbi, struct writeback_control *wbc)
1426 struct pagevec pvec;
1430 pagevec_init(&pvec, 0);
1436 while (index <= end) {
1438 nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1439 PAGECACHE_TAG_DIRTY,
1440 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
1444 for (i = 0; i < nr_pages; i++) {
1445 struct page *page = pvec.pages[i];
1447 if (unlikely(f2fs_cp_error(sbi))) {
1448 pagevec_release(&pvec);
1453 * flushing sequence with step:
1458 if (step == 0 && IS_DNODE(page))
1460 if (step == 1 && (!IS_DNODE(page) ||
1461 is_cold_node(page)))
1463 if (step == 2 && (!IS_DNODE(page) ||
1464 !is_cold_node(page)))
1467 if (!trylock_page(page))
1470 if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1476 if (!PageDirty(page)) {
1477 /* someone wrote it for us */
1478 goto continue_unlock;
1481 /* flush inline_data */
1482 if (is_inline_node(page)) {
1483 clear_inline_node(page);
1485 flush_inline_data(sbi, ino_of_node(page));
1489 f2fs_wait_on_page_writeback(page, NODE, true);
1491 BUG_ON(PageWriteback(page));
1492 if (!clear_page_dirty_for_io(page))
1493 goto continue_unlock;
1495 set_fsync_mark(page, 0);
1496 set_dentry_mark(page, 0);
1498 if (NODE_MAPPING(sbi)->a_ops->writepage(page, wbc))
1501 if (--wbc->nr_to_write == 0)
1504 pagevec_release(&pvec);
1507 if (wbc->nr_to_write == 0) {
1520 int wait_on_node_pages_writeback(struct f2fs_sb_info *sbi, nid_t ino)
1522 pgoff_t index = 0, end = ULONG_MAX;
1523 struct pagevec pvec;
1524 int ret2 = 0, ret = 0;
1526 pagevec_init(&pvec, 0);
1528 while (index <= end) {
1530 nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1531 PAGECACHE_TAG_WRITEBACK,
1532 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
1536 for (i = 0; i < nr_pages; i++) {
1537 struct page *page = pvec.pages[i];
1539 /* until radix tree lookup accepts end_index */
1540 if (unlikely(page->index > end))
1543 if (ino && ino_of_node(page) == ino) {
1544 f2fs_wait_on_page_writeback(page, NODE, true);
1545 if (TestClearPageError(page))
1549 pagevec_release(&pvec);
1553 if (unlikely(test_and_clear_bit(AS_ENOSPC, &NODE_MAPPING(sbi)->flags)))
1555 if (unlikely(test_and_clear_bit(AS_EIO, &NODE_MAPPING(sbi)->flags)))
1562 static int f2fs_write_node_page(struct page *page,
1563 struct writeback_control *wbc)
1565 struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1567 struct node_info ni;
1568 struct f2fs_io_info fio = {
1571 .rw = (wbc->sync_mode == WB_SYNC_ALL) ? WRITE_SYNC : WRITE,
1573 .encrypted_page = NULL,
1576 trace_f2fs_writepage(page, NODE);
1578 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
1580 if (unlikely(f2fs_cp_error(sbi)))
1583 /* get old block addr of this node page */
1584 nid = nid_of_node(page);
1585 f2fs_bug_on(sbi, page->index != nid);
1587 if (wbc->for_reclaim) {
1588 if (!down_read_trylock(&sbi->node_write))
1591 down_read(&sbi->node_write);
1594 get_node_info(sbi, nid, &ni);
1596 /* This page is already truncated */
1597 if (unlikely(ni.blk_addr == NULL_ADDR)) {
1598 ClearPageUptodate(page);
1599 dec_page_count(sbi, F2FS_DIRTY_NODES);
1600 up_read(&sbi->node_write);
1605 set_page_writeback(page);
1606 fio.old_blkaddr = ni.blk_addr;
1607 write_node_page(nid, &fio);
1608 set_node_addr(sbi, &ni, fio.new_blkaddr, is_fsync_dnode(page));
1609 dec_page_count(sbi, F2FS_DIRTY_NODES);
1610 up_read(&sbi->node_write);
1612 if (wbc->for_reclaim)
1613 f2fs_submit_merged_bio_cond(sbi, NULL, page, 0, NODE, WRITE);
1617 if (unlikely(f2fs_cp_error(sbi)))
1618 f2fs_submit_merged_bio(sbi, NODE, WRITE);
1623 redirty_page_for_writepage(wbc, page);
1624 return AOP_WRITEPAGE_ACTIVATE;
1627 static int f2fs_write_node_pages(struct address_space *mapping,
1628 struct writeback_control *wbc)
1630 struct f2fs_sb_info *sbi = F2FS_M_SB(mapping);
1633 /* balancing f2fs's metadata in background */
1634 f2fs_balance_fs_bg(sbi);
1636 /* collect a number of dirty node pages and write together */
1637 if (get_pages(sbi, F2FS_DIRTY_NODES) < nr_pages_to_skip(sbi, NODE))
1640 trace_f2fs_writepages(mapping->host, wbc, NODE);
1642 diff = nr_pages_to_write(sbi, NODE, wbc);
1643 wbc->sync_mode = WB_SYNC_NONE;
1644 sync_node_pages(sbi, wbc);
1645 wbc->nr_to_write = max((long)0, wbc->nr_to_write - diff);
1649 wbc->pages_skipped += get_pages(sbi, F2FS_DIRTY_NODES);
1650 trace_f2fs_writepages(mapping->host, wbc, NODE);
1654 static int f2fs_set_node_page_dirty(struct page *page)
1656 trace_f2fs_set_page_dirty(page, NODE);
1658 SetPageUptodate(page);
1659 if (!PageDirty(page)) {
1660 __set_page_dirty_nobuffers(page);
1661 inc_page_count(F2FS_P_SB(page), F2FS_DIRTY_NODES);
1662 SetPagePrivate(page);
1663 f2fs_trace_pid(page);
1670 * Structure of the f2fs node operations
1672 const struct address_space_operations f2fs_node_aops = {
1673 .writepage = f2fs_write_node_page,
1674 .writepages = f2fs_write_node_pages,
1675 .set_page_dirty = f2fs_set_node_page_dirty,
1676 .invalidatepage = f2fs_invalidate_page,
1677 .releasepage = f2fs_release_page,
1680 static struct free_nid *__lookup_free_nid_list(struct f2fs_nm_info *nm_i,
1683 return radix_tree_lookup(&nm_i->free_nid_root, n);
1686 static void __del_from_free_nid_list(struct f2fs_nm_info *nm_i,
1690 radix_tree_delete(&nm_i->free_nid_root, i->nid);
1693 static int add_free_nid(struct f2fs_sb_info *sbi, nid_t nid, bool build)
1695 struct f2fs_nm_info *nm_i = NM_I(sbi);
1697 struct nat_entry *ne;
1699 if (!available_free_memory(sbi, FREE_NIDS))
1702 /* 0 nid should not be used */
1703 if (unlikely(nid == 0))
1707 /* do not add allocated nids */
1708 ne = __lookup_nat_cache(nm_i, nid);
1709 if (ne && (!get_nat_flag(ne, IS_CHECKPOINTED) ||
1710 nat_get_blkaddr(ne) != NULL_ADDR))
1714 i = f2fs_kmem_cache_alloc(free_nid_slab, GFP_NOFS);
1718 if (radix_tree_preload(GFP_NOFS)) {
1719 kmem_cache_free(free_nid_slab, i);
1723 spin_lock(&nm_i->free_nid_list_lock);
1724 if (radix_tree_insert(&nm_i->free_nid_root, i->nid, i)) {
1725 spin_unlock(&nm_i->free_nid_list_lock);
1726 radix_tree_preload_end();
1727 kmem_cache_free(free_nid_slab, i);
1730 list_add_tail(&i->list, &nm_i->free_nid_list);
1732 spin_unlock(&nm_i->free_nid_list_lock);
1733 radix_tree_preload_end();
1737 static void remove_free_nid(struct f2fs_nm_info *nm_i, nid_t nid)
1740 bool need_free = false;
1742 spin_lock(&nm_i->free_nid_list_lock);
1743 i = __lookup_free_nid_list(nm_i, nid);
1744 if (i && i->state == NID_NEW) {
1745 __del_from_free_nid_list(nm_i, i);
1749 spin_unlock(&nm_i->free_nid_list_lock);
1752 kmem_cache_free(free_nid_slab, i);
1755 static void scan_nat_page(struct f2fs_sb_info *sbi,
1756 struct page *nat_page, nid_t start_nid)
1758 struct f2fs_nm_info *nm_i = NM_I(sbi);
1759 struct f2fs_nat_block *nat_blk = page_address(nat_page);
1763 i = start_nid % NAT_ENTRY_PER_BLOCK;
1765 for (; i < NAT_ENTRY_PER_BLOCK; i++, start_nid++) {
1767 if (unlikely(start_nid >= nm_i->max_nid))
1770 blk_addr = le32_to_cpu(nat_blk->entries[i].block_addr);
1771 f2fs_bug_on(sbi, blk_addr == NEW_ADDR);
1772 if (blk_addr == NULL_ADDR) {
1773 if (add_free_nid(sbi, start_nid, true) < 0)
1779 static void build_free_nids(struct f2fs_sb_info *sbi)
1781 struct f2fs_nm_info *nm_i = NM_I(sbi);
1782 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
1783 struct f2fs_journal *journal = curseg->journal;
1785 nid_t nid = nm_i->next_scan_nid;
1787 /* Enough entries */
1788 if (nm_i->fcnt > NAT_ENTRY_PER_BLOCK)
1791 /* readahead nat pages to be scanned */
1792 ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nid), FREE_NID_PAGES,
1795 down_read(&nm_i->nat_tree_lock);
1798 struct page *page = get_current_nat_page(sbi, nid);
1800 scan_nat_page(sbi, page, nid);
1801 f2fs_put_page(page, 1);
1803 nid += (NAT_ENTRY_PER_BLOCK - (nid % NAT_ENTRY_PER_BLOCK));
1804 if (unlikely(nid >= nm_i->max_nid))
1807 if (++i >= FREE_NID_PAGES)
1811 /* go to the next free nat pages to find free nids abundantly */
1812 nm_i->next_scan_nid = nid;
1814 /* find free nids from current sum_pages */
1815 down_read(&curseg->journal_rwsem);
1816 for (i = 0; i < nats_in_cursum(journal); i++) {
1819 addr = le32_to_cpu(nat_in_journal(journal, i).block_addr);
1820 nid = le32_to_cpu(nid_in_journal(journal, i));
1821 if (addr == NULL_ADDR)
1822 add_free_nid(sbi, nid, true);
1824 remove_free_nid(nm_i, nid);
1826 up_read(&curseg->journal_rwsem);
1827 up_read(&nm_i->nat_tree_lock);
1829 ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nm_i->next_scan_nid),
1830 nm_i->ra_nid_pages, META_NAT, false);
1834 * If this function returns success, caller can obtain a new nid
1835 * from second parameter of this function.
1836 * The returned nid could be used ino as well as nid when inode is created.
1838 bool alloc_nid(struct f2fs_sb_info *sbi, nid_t *nid)
1840 struct f2fs_nm_info *nm_i = NM_I(sbi);
1841 struct free_nid *i = NULL;
1843 #ifdef CONFIG_F2FS_FAULT_INJECTION
1844 if (time_to_inject(FAULT_ALLOC_NID))
1847 if (unlikely(sbi->total_valid_node_count + 1 > nm_i->available_nids))
1850 spin_lock(&nm_i->free_nid_list_lock);
1852 /* We should not use stale free nids created by build_free_nids */
1853 if (nm_i->fcnt && !on_build_free_nids(nm_i)) {
1854 f2fs_bug_on(sbi, list_empty(&nm_i->free_nid_list));
1855 list_for_each_entry(i, &nm_i->free_nid_list, list)
1856 if (i->state == NID_NEW)
1859 f2fs_bug_on(sbi, i->state != NID_NEW);
1861 i->state = NID_ALLOC;
1863 spin_unlock(&nm_i->free_nid_list_lock);
1866 spin_unlock(&nm_i->free_nid_list_lock);
1868 /* Let's scan nat pages and its caches to get free nids */
1869 mutex_lock(&nm_i->build_lock);
1870 build_free_nids(sbi);
1871 mutex_unlock(&nm_i->build_lock);
1876 * alloc_nid() should be called prior to this function.
1878 void alloc_nid_done(struct f2fs_sb_info *sbi, nid_t nid)
1880 struct f2fs_nm_info *nm_i = NM_I(sbi);
1883 spin_lock(&nm_i->free_nid_list_lock);
1884 i = __lookup_free_nid_list(nm_i, nid);
1885 f2fs_bug_on(sbi, !i || i->state != NID_ALLOC);
1886 __del_from_free_nid_list(nm_i, i);
1887 spin_unlock(&nm_i->free_nid_list_lock);
1889 kmem_cache_free(free_nid_slab, i);
1893 * alloc_nid() should be called prior to this function.
1895 void alloc_nid_failed(struct f2fs_sb_info *sbi, nid_t nid)
1897 struct f2fs_nm_info *nm_i = NM_I(sbi);
1899 bool need_free = false;
1904 spin_lock(&nm_i->free_nid_list_lock);
1905 i = __lookup_free_nid_list(nm_i, nid);
1906 f2fs_bug_on(sbi, !i || i->state != NID_ALLOC);
1907 if (!available_free_memory(sbi, FREE_NIDS)) {
1908 __del_from_free_nid_list(nm_i, i);
1914 spin_unlock(&nm_i->free_nid_list_lock);
1917 kmem_cache_free(free_nid_slab, i);
1920 int try_to_free_nids(struct f2fs_sb_info *sbi, int nr_shrink)
1922 struct f2fs_nm_info *nm_i = NM_I(sbi);
1923 struct free_nid *i, *next;
1926 if (!mutex_trylock(&nm_i->build_lock))
1929 spin_lock(&nm_i->free_nid_list_lock);
1930 list_for_each_entry_safe(i, next, &nm_i->free_nid_list, list) {
1931 if (nr_shrink <= 0 || nm_i->fcnt <= NAT_ENTRY_PER_BLOCK)
1933 if (i->state == NID_ALLOC)
1935 __del_from_free_nid_list(nm_i, i);
1936 kmem_cache_free(free_nid_slab, i);
1940 spin_unlock(&nm_i->free_nid_list_lock);
1941 mutex_unlock(&nm_i->build_lock);
1943 return nr - nr_shrink;
1946 void recover_inline_xattr(struct inode *inode, struct page *page)
1948 void *src_addr, *dst_addr;
1951 struct f2fs_inode *ri;
1953 ipage = get_node_page(F2FS_I_SB(inode), inode->i_ino);
1954 f2fs_bug_on(F2FS_I_SB(inode), IS_ERR(ipage));
1956 ri = F2FS_INODE(page);
1957 if (!(ri->i_inline & F2FS_INLINE_XATTR)) {
1958 clear_inode_flag(F2FS_I(inode), FI_INLINE_XATTR);
1962 dst_addr = inline_xattr_addr(ipage);
1963 src_addr = inline_xattr_addr(page);
1964 inline_size = inline_xattr_size(inode);
1966 f2fs_wait_on_page_writeback(ipage, NODE, true);
1967 memcpy(dst_addr, src_addr, inline_size);
1969 update_inode(inode, ipage);
1970 f2fs_put_page(ipage, 1);
1973 void recover_xattr_data(struct inode *inode, struct page *page, block_t blkaddr)
1975 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1976 nid_t prev_xnid = F2FS_I(inode)->i_xattr_nid;
1977 nid_t new_xnid = nid_of_node(page);
1978 struct node_info ni;
1980 /* 1: invalidate the previous xattr nid */
1984 /* Deallocate node address */
1985 get_node_info(sbi, prev_xnid, &ni);
1986 f2fs_bug_on(sbi, ni.blk_addr == NULL_ADDR);
1987 invalidate_blocks(sbi, ni.blk_addr);
1988 dec_valid_node_count(sbi, inode);
1989 set_node_addr(sbi, &ni, NULL_ADDR, false);
1992 /* 2: allocate new xattr nid */
1993 if (unlikely(!inc_valid_node_count(sbi, inode)))
1994 f2fs_bug_on(sbi, 1);
1996 remove_free_nid(NM_I(sbi), new_xnid);
1997 get_node_info(sbi, new_xnid, &ni);
1998 ni.ino = inode->i_ino;
1999 set_node_addr(sbi, &ni, NEW_ADDR, false);
2000 F2FS_I(inode)->i_xattr_nid = new_xnid;
2002 /* 3: update xattr blkaddr */
2003 refresh_sit_entry(sbi, NEW_ADDR, blkaddr);
2004 set_node_addr(sbi, &ni, blkaddr, false);
2006 update_inode_page(inode);
2009 int recover_inode_page(struct f2fs_sb_info *sbi, struct page *page)
2011 struct f2fs_inode *src, *dst;
2012 nid_t ino = ino_of_node(page);
2013 struct node_info old_ni, new_ni;
2016 get_node_info(sbi, ino, &old_ni);
2018 if (unlikely(old_ni.blk_addr != NULL_ADDR))
2021 ipage = f2fs_grab_cache_page(NODE_MAPPING(sbi), ino, false);
2025 /* Should not use this inode from free nid list */
2026 remove_free_nid(NM_I(sbi), ino);
2028 SetPageUptodate(ipage);
2029 fill_node_footer(ipage, ino, ino, 0, true);
2031 src = F2FS_INODE(page);
2032 dst = F2FS_INODE(ipage);
2034 memcpy(dst, src, (unsigned long)&src->i_ext - (unsigned long)src);
2036 dst->i_blocks = cpu_to_le64(1);
2037 dst->i_links = cpu_to_le32(1);
2038 dst->i_xattr_nid = 0;
2039 dst->i_inline = src->i_inline & F2FS_INLINE_XATTR;
2044 if (unlikely(!inc_valid_node_count(sbi, NULL)))
2046 set_node_addr(sbi, &new_ni, NEW_ADDR, false);
2047 inc_valid_inode_count(sbi);
2048 set_page_dirty(ipage);
2049 f2fs_put_page(ipage, 1);
2053 int restore_node_summary(struct f2fs_sb_info *sbi,
2054 unsigned int segno, struct f2fs_summary_block *sum)
2056 struct f2fs_node *rn;
2057 struct f2fs_summary *sum_entry;
2059 int bio_blocks = MAX_BIO_BLOCKS(sbi);
2060 int i, idx, last_offset, nrpages;
2062 /* scan the node segment */
2063 last_offset = sbi->blocks_per_seg;
2064 addr = START_BLOCK(sbi, segno);
2065 sum_entry = &sum->entries[0];
2067 for (i = 0; i < last_offset; i += nrpages, addr += nrpages) {
2068 nrpages = min(last_offset - i, bio_blocks);
2070 /* readahead node pages */
2071 ra_meta_pages(sbi, addr, nrpages, META_POR, true);
2073 for (idx = addr; idx < addr + nrpages; idx++) {
2074 struct page *page = get_tmp_page(sbi, idx);
2076 rn = F2FS_NODE(page);
2077 sum_entry->nid = rn->footer.nid;
2078 sum_entry->version = 0;
2079 sum_entry->ofs_in_node = 0;
2081 f2fs_put_page(page, 1);
2084 invalidate_mapping_pages(META_MAPPING(sbi), addr,
2090 static void remove_nats_in_journal(struct f2fs_sb_info *sbi)
2092 struct f2fs_nm_info *nm_i = NM_I(sbi);
2093 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
2094 struct f2fs_journal *journal = curseg->journal;
2097 down_write(&curseg->journal_rwsem);
2098 for (i = 0; i < nats_in_cursum(journal); i++) {
2099 struct nat_entry *ne;
2100 struct f2fs_nat_entry raw_ne;
2101 nid_t nid = le32_to_cpu(nid_in_journal(journal, i));
2103 raw_ne = nat_in_journal(journal, i);
2105 ne = __lookup_nat_cache(nm_i, nid);
2107 ne = grab_nat_entry(nm_i, nid);
2108 node_info_from_raw_nat(&ne->ni, &raw_ne);
2110 __set_nat_cache_dirty(nm_i, ne);
2112 update_nats_in_cursum(journal, -i);
2113 up_write(&curseg->journal_rwsem);
2116 static void __adjust_nat_entry_set(struct nat_entry_set *nes,
2117 struct list_head *head, int max)
2119 struct nat_entry_set *cur;
2121 if (nes->entry_cnt >= max)
2124 list_for_each_entry(cur, head, set_list) {
2125 if (cur->entry_cnt >= nes->entry_cnt) {
2126 list_add(&nes->set_list, cur->set_list.prev);
2131 list_add_tail(&nes->set_list, head);
2134 static void __flush_nat_entry_set(struct f2fs_sb_info *sbi,
2135 struct nat_entry_set *set)
2137 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
2138 struct f2fs_journal *journal = curseg->journal;
2139 nid_t start_nid = set->set * NAT_ENTRY_PER_BLOCK;
2140 bool to_journal = true;
2141 struct f2fs_nat_block *nat_blk;
2142 struct nat_entry *ne, *cur;
2143 struct page *page = NULL;
2146 * there are two steps to flush nat entries:
2147 * #1, flush nat entries to journal in current hot data summary block.
2148 * #2, flush nat entries to nat page.
2150 if (!__has_cursum_space(journal, set->entry_cnt, NAT_JOURNAL))
2154 down_write(&curseg->journal_rwsem);
2156 page = get_next_nat_page(sbi, start_nid);
2157 nat_blk = page_address(page);
2158 f2fs_bug_on(sbi, !nat_blk);
2161 /* flush dirty nats in nat entry set */
2162 list_for_each_entry_safe(ne, cur, &set->entry_list, list) {
2163 struct f2fs_nat_entry *raw_ne;
2164 nid_t nid = nat_get_nid(ne);
2167 if (nat_get_blkaddr(ne) == NEW_ADDR)
2171 offset = lookup_journal_in_cursum(journal,
2172 NAT_JOURNAL, nid, 1);
2173 f2fs_bug_on(sbi, offset < 0);
2174 raw_ne = &nat_in_journal(journal, offset);
2175 nid_in_journal(journal, offset) = cpu_to_le32(nid);
2177 raw_ne = &nat_blk->entries[nid - start_nid];
2179 raw_nat_from_node_info(raw_ne, &ne->ni);
2181 __clear_nat_cache_dirty(NM_I(sbi), ne);
2182 if (nat_get_blkaddr(ne) == NULL_ADDR)
2183 add_free_nid(sbi, nid, false);
2187 up_write(&curseg->journal_rwsem);
2189 f2fs_put_page(page, 1);
2191 f2fs_bug_on(sbi, set->entry_cnt);
2193 radix_tree_delete(&NM_I(sbi)->nat_set_root, set->set);
2194 kmem_cache_free(nat_entry_set_slab, set);
2198 * This function is called during the checkpointing process.
2200 void flush_nat_entries(struct f2fs_sb_info *sbi)
2202 struct f2fs_nm_info *nm_i = NM_I(sbi);
2203 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
2204 struct f2fs_journal *journal = curseg->journal;
2205 struct nat_entry_set *setvec[SETVEC_SIZE];
2206 struct nat_entry_set *set, *tmp;
2211 if (!nm_i->dirty_nat_cnt)
2214 down_write(&nm_i->nat_tree_lock);
2217 * if there are no enough space in journal to store dirty nat
2218 * entries, remove all entries from journal and merge them
2219 * into nat entry set.
2221 if (!__has_cursum_space(journal, nm_i->dirty_nat_cnt, NAT_JOURNAL))
2222 remove_nats_in_journal(sbi);
2224 while ((found = __gang_lookup_nat_set(nm_i,
2225 set_idx, SETVEC_SIZE, setvec))) {
2227 set_idx = setvec[found - 1]->set + 1;
2228 for (idx = 0; idx < found; idx++)
2229 __adjust_nat_entry_set(setvec[idx], &sets,
2230 MAX_NAT_JENTRIES(journal));
2233 /* flush dirty nats in nat entry set */
2234 list_for_each_entry_safe(set, tmp, &sets, set_list)
2235 __flush_nat_entry_set(sbi, set);
2237 up_write(&nm_i->nat_tree_lock);
2239 f2fs_bug_on(sbi, nm_i->dirty_nat_cnt);
2242 static int init_node_manager(struct f2fs_sb_info *sbi)
2244 struct f2fs_super_block *sb_raw = F2FS_RAW_SUPER(sbi);
2245 struct f2fs_nm_info *nm_i = NM_I(sbi);
2246 unsigned char *version_bitmap;
2247 unsigned int nat_segs, nat_blocks;
2249 nm_i->nat_blkaddr = le32_to_cpu(sb_raw->nat_blkaddr);
2251 /* segment_count_nat includes pair segment so divide to 2. */
2252 nat_segs = le32_to_cpu(sb_raw->segment_count_nat) >> 1;
2253 nat_blocks = nat_segs << le32_to_cpu(sb_raw->log_blocks_per_seg);
2255 nm_i->max_nid = NAT_ENTRY_PER_BLOCK * nat_blocks;
2257 /* not used nids: 0, node, meta, (and root counted as valid node) */
2258 nm_i->available_nids = nm_i->max_nid - F2FS_RESERVED_NODE_NUM;
2261 nm_i->ram_thresh = DEF_RAM_THRESHOLD;
2262 nm_i->ra_nid_pages = DEF_RA_NID_PAGES;
2263 nm_i->dirty_nats_ratio = DEF_DIRTY_NAT_RATIO_THRESHOLD;
2265 INIT_RADIX_TREE(&nm_i->free_nid_root, GFP_ATOMIC);
2266 INIT_LIST_HEAD(&nm_i->free_nid_list);
2267 INIT_RADIX_TREE(&nm_i->nat_root, GFP_NOIO);
2268 INIT_RADIX_TREE(&nm_i->nat_set_root, GFP_NOIO);
2269 INIT_LIST_HEAD(&nm_i->nat_entries);
2271 mutex_init(&nm_i->build_lock);
2272 spin_lock_init(&nm_i->free_nid_list_lock);
2273 init_rwsem(&nm_i->nat_tree_lock);
2275 nm_i->next_scan_nid = le32_to_cpu(sbi->ckpt->next_free_nid);
2276 nm_i->bitmap_size = __bitmap_size(sbi, NAT_BITMAP);
2277 version_bitmap = __bitmap_ptr(sbi, NAT_BITMAP);
2278 if (!version_bitmap)
2281 nm_i->nat_bitmap = kmemdup(version_bitmap, nm_i->bitmap_size,
2283 if (!nm_i->nat_bitmap)
2288 int build_node_manager(struct f2fs_sb_info *sbi)
2292 sbi->nm_info = kzalloc(sizeof(struct f2fs_nm_info), GFP_KERNEL);
2296 err = init_node_manager(sbi);
2300 build_free_nids(sbi);
2304 void destroy_node_manager(struct f2fs_sb_info *sbi)
2306 struct f2fs_nm_info *nm_i = NM_I(sbi);
2307 struct free_nid *i, *next_i;
2308 struct nat_entry *natvec[NATVEC_SIZE];
2309 struct nat_entry_set *setvec[SETVEC_SIZE];
2316 /* destroy free nid list */
2317 spin_lock(&nm_i->free_nid_list_lock);
2318 list_for_each_entry_safe(i, next_i, &nm_i->free_nid_list, list) {
2319 f2fs_bug_on(sbi, i->state == NID_ALLOC);
2320 __del_from_free_nid_list(nm_i, i);
2322 spin_unlock(&nm_i->free_nid_list_lock);
2323 kmem_cache_free(free_nid_slab, i);
2324 spin_lock(&nm_i->free_nid_list_lock);
2326 f2fs_bug_on(sbi, nm_i->fcnt);
2327 spin_unlock(&nm_i->free_nid_list_lock);
2329 /* destroy nat cache */
2330 down_write(&nm_i->nat_tree_lock);
2331 while ((found = __gang_lookup_nat_cache(nm_i,
2332 nid, NATVEC_SIZE, natvec))) {
2335 nid = nat_get_nid(natvec[found - 1]) + 1;
2336 for (idx = 0; idx < found; idx++)
2337 __del_from_nat_cache(nm_i, natvec[idx]);
2339 f2fs_bug_on(sbi, nm_i->nat_cnt);
2341 /* destroy nat set cache */
2343 while ((found = __gang_lookup_nat_set(nm_i,
2344 nid, SETVEC_SIZE, setvec))) {
2347 nid = setvec[found - 1]->set + 1;
2348 for (idx = 0; idx < found; idx++) {
2349 /* entry_cnt is not zero, when cp_error was occurred */
2350 f2fs_bug_on(sbi, !list_empty(&setvec[idx]->entry_list));
2351 radix_tree_delete(&nm_i->nat_set_root, setvec[idx]->set);
2352 kmem_cache_free(nat_entry_set_slab, setvec[idx]);
2355 up_write(&nm_i->nat_tree_lock);
2357 kfree(nm_i->nat_bitmap);
2358 sbi->nm_info = NULL;
2362 int __init create_node_manager_caches(void)
2364 nat_entry_slab = f2fs_kmem_cache_create("nat_entry",
2365 sizeof(struct nat_entry));
2366 if (!nat_entry_slab)
2369 free_nid_slab = f2fs_kmem_cache_create("free_nid",
2370 sizeof(struct free_nid));
2372 goto destroy_nat_entry;
2374 nat_entry_set_slab = f2fs_kmem_cache_create("nat_entry_set",
2375 sizeof(struct nat_entry_set));
2376 if (!nat_entry_set_slab)
2377 goto destroy_free_nid;
2381 kmem_cache_destroy(free_nid_slab);
2383 kmem_cache_destroy(nat_entry_slab);
2388 void destroy_node_manager_caches(void)
2390 kmem_cache_destroy(nat_entry_set_slab);
2391 kmem_cache_destroy(free_nid_slab);
2392 kmem_cache_destroy(nat_entry_slab);