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>
22 #include <trace/events/f2fs.h>
24 #define on_build_free_nids(nmi) mutex_is_locked(&nm_i->build_lock)
26 static struct kmem_cache *nat_entry_slab;
27 static struct kmem_cache *free_nid_slab;
28 static struct kmem_cache *nat_entry_set_slab;
30 bool available_free_memory(struct f2fs_sb_info *sbi, int type)
32 struct f2fs_nm_info *nm_i = NM_I(sbi);
34 unsigned long avail_ram;
35 unsigned long mem_size = 0;
40 /* only uses low memory */
41 avail_ram = val.totalram - val.totalhigh;
43 /* give 25%, 25%, 50%, 50% memory for each components respectively */
44 if (type == FREE_NIDS) {
45 mem_size = (nm_i->fcnt * sizeof(struct free_nid)) >>
47 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 2);
48 } else if (type == NAT_ENTRIES) {
49 mem_size = (nm_i->nat_cnt * sizeof(struct nat_entry)) >>
51 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 2);
52 } else if (type == DIRTY_DENTS) {
53 if (sbi->sb->s_bdi->dirty_exceeded)
55 mem_size = get_pages(sbi, F2FS_DIRTY_DENTS);
56 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 1);
57 } else if (type == INO_ENTRIES) {
60 if (sbi->sb->s_bdi->dirty_exceeded)
62 for (i = 0; i <= UPDATE_INO; i++)
63 mem_size += (sbi->im[i].ino_num *
64 sizeof(struct ino_entry)) >> PAGE_CACHE_SHIFT;
65 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 1);
70 static void clear_node_page_dirty(struct page *page)
72 struct address_space *mapping = page->mapping;
73 unsigned int long flags;
75 if (PageDirty(page)) {
76 spin_lock_irqsave(&mapping->tree_lock, flags);
77 radix_tree_tag_clear(&mapping->page_tree,
80 spin_unlock_irqrestore(&mapping->tree_lock, flags);
82 clear_page_dirty_for_io(page);
83 dec_page_count(F2FS_M_SB(mapping), F2FS_DIRTY_NODES);
85 ClearPageUptodate(page);
88 static struct page *get_current_nat_page(struct f2fs_sb_info *sbi, nid_t nid)
90 pgoff_t index = current_nat_addr(sbi, nid);
91 return get_meta_page(sbi, index);
94 static struct page *get_next_nat_page(struct f2fs_sb_info *sbi, nid_t nid)
96 struct page *src_page;
97 struct page *dst_page;
102 struct f2fs_nm_info *nm_i = NM_I(sbi);
104 src_off = current_nat_addr(sbi, nid);
105 dst_off = next_nat_addr(sbi, src_off);
107 /* get current nat block page with lock */
108 src_page = get_meta_page(sbi, src_off);
109 dst_page = grab_meta_page(sbi, dst_off);
110 f2fs_bug_on(sbi, PageDirty(src_page));
112 src_addr = page_address(src_page);
113 dst_addr = page_address(dst_page);
114 memcpy(dst_addr, src_addr, PAGE_CACHE_SIZE);
115 set_page_dirty(dst_page);
116 f2fs_put_page(src_page, 1);
118 set_to_next_nat(nm_i, nid);
123 static struct nat_entry *__lookup_nat_cache(struct f2fs_nm_info *nm_i, nid_t n)
125 return radix_tree_lookup(&nm_i->nat_root, n);
128 static unsigned int __gang_lookup_nat_cache(struct f2fs_nm_info *nm_i,
129 nid_t start, unsigned int nr, struct nat_entry **ep)
131 return radix_tree_gang_lookup(&nm_i->nat_root, (void **)ep, start, nr);
134 static void __del_from_nat_cache(struct f2fs_nm_info *nm_i, struct nat_entry *e)
137 radix_tree_delete(&nm_i->nat_root, nat_get_nid(e));
139 kmem_cache_free(nat_entry_slab, e);
142 static void __set_nat_cache_dirty(struct f2fs_nm_info *nm_i,
143 struct nat_entry *ne)
145 nid_t set = NAT_BLOCK_OFFSET(ne->ni.nid);
146 struct nat_entry_set *head;
148 if (get_nat_flag(ne, IS_DIRTY))
151 head = radix_tree_lookup(&nm_i->nat_set_root, set);
153 head = f2fs_kmem_cache_alloc(nat_entry_set_slab, GFP_ATOMIC);
155 INIT_LIST_HEAD(&head->entry_list);
156 INIT_LIST_HEAD(&head->set_list);
159 f2fs_radix_tree_insert(&nm_i->nat_set_root, set, head);
161 list_move_tail(&ne->list, &head->entry_list);
162 nm_i->dirty_nat_cnt++;
164 set_nat_flag(ne, IS_DIRTY, true);
167 static void __clear_nat_cache_dirty(struct f2fs_nm_info *nm_i,
168 struct nat_entry *ne)
170 nid_t set = NAT_BLOCK_OFFSET(ne->ni.nid);
171 struct nat_entry_set *head;
173 head = radix_tree_lookup(&nm_i->nat_set_root, set);
175 list_move_tail(&ne->list, &nm_i->nat_entries);
176 set_nat_flag(ne, IS_DIRTY, false);
178 nm_i->dirty_nat_cnt--;
182 static unsigned int __gang_lookup_nat_set(struct f2fs_nm_info *nm_i,
183 nid_t start, unsigned int nr, struct nat_entry_set **ep)
185 return radix_tree_gang_lookup(&nm_i->nat_set_root, (void **)ep,
189 bool is_checkpointed_node(struct f2fs_sb_info *sbi, nid_t nid)
191 struct f2fs_nm_info *nm_i = NM_I(sbi);
195 down_read(&nm_i->nat_tree_lock);
196 e = __lookup_nat_cache(nm_i, nid);
197 if (e && !get_nat_flag(e, IS_CHECKPOINTED))
199 up_read(&nm_i->nat_tree_lock);
203 bool has_fsynced_inode(struct f2fs_sb_info *sbi, nid_t ino)
205 struct f2fs_nm_info *nm_i = NM_I(sbi);
207 bool fsynced = false;
209 down_read(&nm_i->nat_tree_lock);
210 e = __lookup_nat_cache(nm_i, ino);
211 if (e && get_nat_flag(e, HAS_FSYNCED_INODE))
213 up_read(&nm_i->nat_tree_lock);
217 bool need_inode_block_update(struct f2fs_sb_info *sbi, nid_t ino)
219 struct f2fs_nm_info *nm_i = NM_I(sbi);
221 bool need_update = true;
223 down_read(&nm_i->nat_tree_lock);
224 e = __lookup_nat_cache(nm_i, ino);
225 if (e && get_nat_flag(e, HAS_LAST_FSYNC) &&
226 (get_nat_flag(e, IS_CHECKPOINTED) ||
227 get_nat_flag(e, HAS_FSYNCED_INODE)))
229 up_read(&nm_i->nat_tree_lock);
233 static struct nat_entry *grab_nat_entry(struct f2fs_nm_info *nm_i, nid_t nid)
235 struct nat_entry *new;
237 new = f2fs_kmem_cache_alloc(nat_entry_slab, GFP_ATOMIC);
238 f2fs_radix_tree_insert(&nm_i->nat_root, nid, new);
239 memset(new, 0, sizeof(struct nat_entry));
240 nat_set_nid(new, nid);
242 list_add_tail(&new->list, &nm_i->nat_entries);
247 static void cache_nat_entry(struct f2fs_nm_info *nm_i, nid_t nid,
248 struct f2fs_nat_entry *ne)
252 down_write(&nm_i->nat_tree_lock);
253 e = __lookup_nat_cache(nm_i, nid);
255 e = grab_nat_entry(nm_i, nid);
256 node_info_from_raw_nat(&e->ni, ne);
258 up_write(&nm_i->nat_tree_lock);
261 static void set_node_addr(struct f2fs_sb_info *sbi, struct node_info *ni,
262 block_t new_blkaddr, bool fsync_done)
264 struct f2fs_nm_info *nm_i = NM_I(sbi);
267 down_write(&nm_i->nat_tree_lock);
268 e = __lookup_nat_cache(nm_i, ni->nid);
270 e = grab_nat_entry(nm_i, ni->nid);
272 f2fs_bug_on(sbi, ni->blk_addr == NEW_ADDR);
273 } else if (new_blkaddr == NEW_ADDR) {
275 * when nid is reallocated,
276 * previous nat entry can be remained in nat cache.
277 * So, reinitialize it with new information.
280 f2fs_bug_on(sbi, ni->blk_addr != NULL_ADDR);
284 f2fs_bug_on(sbi, nat_get_blkaddr(e) != ni->blk_addr);
285 f2fs_bug_on(sbi, nat_get_blkaddr(e) == NULL_ADDR &&
286 new_blkaddr == NULL_ADDR);
287 f2fs_bug_on(sbi, nat_get_blkaddr(e) == NEW_ADDR &&
288 new_blkaddr == NEW_ADDR);
289 f2fs_bug_on(sbi, nat_get_blkaddr(e) != NEW_ADDR &&
290 nat_get_blkaddr(e) != NULL_ADDR &&
291 new_blkaddr == NEW_ADDR);
293 /* increment version no as node is removed */
294 if (nat_get_blkaddr(e) != NEW_ADDR && new_blkaddr == NULL_ADDR) {
295 unsigned char version = nat_get_version(e);
296 nat_set_version(e, inc_node_version(version));
300 nat_set_blkaddr(e, new_blkaddr);
301 if (new_blkaddr == NEW_ADDR || new_blkaddr == NULL_ADDR)
302 set_nat_flag(e, IS_CHECKPOINTED, false);
303 __set_nat_cache_dirty(nm_i, e);
305 /* update fsync_mark if its inode nat entry is still alive */
306 e = __lookup_nat_cache(nm_i, ni->ino);
308 if (fsync_done && ni->nid == ni->ino)
309 set_nat_flag(e, HAS_FSYNCED_INODE, true);
310 set_nat_flag(e, HAS_LAST_FSYNC, fsync_done);
312 up_write(&nm_i->nat_tree_lock);
315 int try_to_free_nats(struct f2fs_sb_info *sbi, int nr_shrink)
317 struct f2fs_nm_info *nm_i = NM_I(sbi);
319 if (available_free_memory(sbi, NAT_ENTRIES))
322 down_write(&nm_i->nat_tree_lock);
323 while (nr_shrink && !list_empty(&nm_i->nat_entries)) {
324 struct nat_entry *ne;
325 ne = list_first_entry(&nm_i->nat_entries,
326 struct nat_entry, list);
327 __del_from_nat_cache(nm_i, ne);
330 up_write(&nm_i->nat_tree_lock);
335 * This function always returns success
337 void get_node_info(struct f2fs_sb_info *sbi, nid_t nid, struct node_info *ni)
339 struct f2fs_nm_info *nm_i = NM_I(sbi);
340 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
341 struct f2fs_summary_block *sum = curseg->sum_blk;
342 nid_t start_nid = START_NID(nid);
343 struct f2fs_nat_block *nat_blk;
344 struct page *page = NULL;
345 struct f2fs_nat_entry ne;
349 memset(&ne, 0, sizeof(struct f2fs_nat_entry));
352 /* Check nat cache */
353 down_read(&nm_i->nat_tree_lock);
354 e = __lookup_nat_cache(nm_i, nid);
356 ni->ino = nat_get_ino(e);
357 ni->blk_addr = nat_get_blkaddr(e);
358 ni->version = nat_get_version(e);
360 up_read(&nm_i->nat_tree_lock);
364 /* Check current segment summary */
365 mutex_lock(&curseg->curseg_mutex);
366 i = lookup_journal_in_cursum(sum, NAT_JOURNAL, nid, 0);
368 ne = nat_in_journal(sum, i);
369 node_info_from_raw_nat(ni, &ne);
371 mutex_unlock(&curseg->curseg_mutex);
375 /* Fill node_info from nat page */
376 page = get_current_nat_page(sbi, start_nid);
377 nat_blk = (struct f2fs_nat_block *)page_address(page);
378 ne = nat_blk->entries[nid - start_nid];
379 node_info_from_raw_nat(ni, &ne);
380 f2fs_put_page(page, 1);
382 /* cache nat entry */
383 cache_nat_entry(NM_I(sbi), nid, &ne);
387 * The maximum depth is four.
388 * Offset[0] will have raw inode offset.
390 static int get_node_path(struct f2fs_inode_info *fi, long block,
391 int offset[4], unsigned int noffset[4])
393 const long direct_index = ADDRS_PER_INODE(fi);
394 const long direct_blks = ADDRS_PER_BLOCK;
395 const long dptrs_per_blk = NIDS_PER_BLOCK;
396 const long indirect_blks = ADDRS_PER_BLOCK * NIDS_PER_BLOCK;
397 const long dindirect_blks = indirect_blks * NIDS_PER_BLOCK;
403 if (block < direct_index) {
407 block -= direct_index;
408 if (block < direct_blks) {
409 offset[n++] = NODE_DIR1_BLOCK;
415 block -= direct_blks;
416 if (block < direct_blks) {
417 offset[n++] = NODE_DIR2_BLOCK;
423 block -= direct_blks;
424 if (block < indirect_blks) {
425 offset[n++] = NODE_IND1_BLOCK;
427 offset[n++] = block / direct_blks;
428 noffset[n] = 4 + offset[n - 1];
429 offset[n] = block % direct_blks;
433 block -= indirect_blks;
434 if (block < indirect_blks) {
435 offset[n++] = NODE_IND2_BLOCK;
436 noffset[n] = 4 + dptrs_per_blk;
437 offset[n++] = block / direct_blks;
438 noffset[n] = 5 + dptrs_per_blk + offset[n - 1];
439 offset[n] = block % direct_blks;
443 block -= indirect_blks;
444 if (block < dindirect_blks) {
445 offset[n++] = NODE_DIND_BLOCK;
446 noffset[n] = 5 + (dptrs_per_blk * 2);
447 offset[n++] = block / indirect_blks;
448 noffset[n] = 6 + (dptrs_per_blk * 2) +
449 offset[n - 1] * (dptrs_per_blk + 1);
450 offset[n++] = (block / direct_blks) % dptrs_per_blk;
451 noffset[n] = 7 + (dptrs_per_blk * 2) +
452 offset[n - 2] * (dptrs_per_blk + 1) +
454 offset[n] = block % direct_blks;
465 * Caller should call f2fs_put_dnode(dn).
466 * Also, it should grab and release a rwsem by calling f2fs_lock_op() and
467 * f2fs_unlock_op() only if ro is not set RDONLY_NODE.
468 * In the case of RDONLY_NODE, we don't need to care about mutex.
470 int get_dnode_of_data(struct dnode_of_data *dn, pgoff_t index, int mode)
472 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
473 struct page *npage[4];
476 unsigned int noffset[4];
481 level = get_node_path(F2FS_I(dn->inode), index, offset, noffset);
483 nids[0] = dn->inode->i_ino;
484 npage[0] = dn->inode_page;
487 npage[0] = get_node_page(sbi, nids[0]);
488 if (IS_ERR(npage[0]))
489 return PTR_ERR(npage[0]);
493 nids[1] = get_nid(parent, offset[0], true);
494 dn->inode_page = npage[0];
495 dn->inode_page_locked = true;
497 /* get indirect or direct nodes */
498 for (i = 1; i <= level; i++) {
501 if (!nids[i] && mode == ALLOC_NODE) {
503 if (!alloc_nid(sbi, &(nids[i]))) {
509 npage[i] = new_node_page(dn, noffset[i], NULL);
510 if (IS_ERR(npage[i])) {
511 alloc_nid_failed(sbi, nids[i]);
512 err = PTR_ERR(npage[i]);
516 set_nid(parent, offset[i - 1], nids[i], i == 1);
517 alloc_nid_done(sbi, nids[i]);
519 } else if (mode == LOOKUP_NODE_RA && i == level && level > 1) {
520 npage[i] = get_node_page_ra(parent, offset[i - 1]);
521 if (IS_ERR(npage[i])) {
522 err = PTR_ERR(npage[i]);
528 dn->inode_page_locked = false;
531 f2fs_put_page(parent, 1);
535 npage[i] = get_node_page(sbi, nids[i]);
536 if (IS_ERR(npage[i])) {
537 err = PTR_ERR(npage[i]);
538 f2fs_put_page(npage[0], 0);
544 nids[i + 1] = get_nid(parent, offset[i], false);
547 dn->nid = nids[level];
548 dn->ofs_in_node = offset[level];
549 dn->node_page = npage[level];
550 dn->data_blkaddr = datablock_addr(dn->node_page, dn->ofs_in_node);
554 f2fs_put_page(parent, 1);
556 f2fs_put_page(npage[0], 0);
558 dn->inode_page = NULL;
559 dn->node_page = NULL;
563 static void truncate_node(struct dnode_of_data *dn)
565 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
568 get_node_info(sbi, dn->nid, &ni);
569 if (dn->inode->i_blocks == 0) {
570 f2fs_bug_on(sbi, ni.blk_addr != NULL_ADDR);
573 f2fs_bug_on(sbi, ni.blk_addr == NULL_ADDR);
575 /* Deallocate node address */
576 invalidate_blocks(sbi, ni.blk_addr);
577 dec_valid_node_count(sbi, dn->inode);
578 set_node_addr(sbi, &ni, NULL_ADDR, false);
580 if (dn->nid == dn->inode->i_ino) {
581 remove_orphan_inode(sbi, dn->nid);
582 dec_valid_inode_count(sbi);
587 clear_node_page_dirty(dn->node_page);
588 F2FS_SET_SB_DIRT(sbi);
590 f2fs_put_page(dn->node_page, 1);
592 invalidate_mapping_pages(NODE_MAPPING(sbi),
593 dn->node_page->index, dn->node_page->index);
595 dn->node_page = NULL;
596 trace_f2fs_truncate_node(dn->inode, dn->nid, ni.blk_addr);
599 static int truncate_dnode(struct dnode_of_data *dn)
606 /* get direct node */
607 page = get_node_page(F2FS_I_SB(dn->inode), dn->nid);
608 if (IS_ERR(page) && PTR_ERR(page) == -ENOENT)
610 else if (IS_ERR(page))
611 return PTR_ERR(page);
613 /* Make dnode_of_data for parameter */
614 dn->node_page = page;
616 truncate_data_blocks(dn);
621 static int truncate_nodes(struct dnode_of_data *dn, unsigned int nofs,
624 struct dnode_of_data rdn = *dn;
626 struct f2fs_node *rn;
628 unsigned int child_nofs;
633 return NIDS_PER_BLOCK + 1;
635 trace_f2fs_truncate_nodes_enter(dn->inode, dn->nid, dn->data_blkaddr);
637 page = get_node_page(F2FS_I_SB(dn->inode), dn->nid);
639 trace_f2fs_truncate_nodes_exit(dn->inode, PTR_ERR(page));
640 return PTR_ERR(page);
643 rn = F2FS_NODE(page);
645 for (i = ofs; i < NIDS_PER_BLOCK; i++, freed++) {
646 child_nid = le32_to_cpu(rn->in.nid[i]);
650 ret = truncate_dnode(&rdn);
653 set_nid(page, i, 0, false);
656 child_nofs = nofs + ofs * (NIDS_PER_BLOCK + 1) + 1;
657 for (i = ofs; i < NIDS_PER_BLOCK; i++) {
658 child_nid = le32_to_cpu(rn->in.nid[i]);
659 if (child_nid == 0) {
660 child_nofs += NIDS_PER_BLOCK + 1;
664 ret = truncate_nodes(&rdn, child_nofs, 0, depth - 1);
665 if (ret == (NIDS_PER_BLOCK + 1)) {
666 set_nid(page, i, 0, false);
668 } else if (ret < 0 && ret != -ENOENT) {
676 /* remove current indirect node */
677 dn->node_page = page;
681 f2fs_put_page(page, 1);
683 trace_f2fs_truncate_nodes_exit(dn->inode, freed);
687 f2fs_put_page(page, 1);
688 trace_f2fs_truncate_nodes_exit(dn->inode, ret);
692 static int truncate_partial_nodes(struct dnode_of_data *dn,
693 struct f2fs_inode *ri, int *offset, int depth)
695 struct page *pages[2];
702 nid[0] = le32_to_cpu(ri->i_nid[offset[0] - NODE_DIR1_BLOCK]);
706 /* get indirect nodes in the path */
707 for (i = 0; i < idx + 1; i++) {
708 /* reference count'll be increased */
709 pages[i] = get_node_page(F2FS_I_SB(dn->inode), nid[i]);
710 if (IS_ERR(pages[i])) {
711 err = PTR_ERR(pages[i]);
715 nid[i + 1] = get_nid(pages[i], offset[i + 1], false);
718 /* free direct nodes linked to a partial indirect node */
719 for (i = offset[idx + 1]; i < NIDS_PER_BLOCK; i++) {
720 child_nid = get_nid(pages[idx], i, false);
724 err = truncate_dnode(dn);
727 set_nid(pages[idx], i, 0, false);
730 if (offset[idx + 1] == 0) {
731 dn->node_page = pages[idx];
735 f2fs_put_page(pages[idx], 1);
741 for (i = idx; i >= 0; i--)
742 f2fs_put_page(pages[i], 1);
744 trace_f2fs_truncate_partial_nodes(dn->inode, nid, depth, err);
750 * All the block addresses of data and nodes should be nullified.
752 int truncate_inode_blocks(struct inode *inode, pgoff_t from)
754 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
755 int err = 0, cont = 1;
756 int level, offset[4], noffset[4];
757 unsigned int nofs = 0;
758 struct f2fs_inode *ri;
759 struct dnode_of_data dn;
762 trace_f2fs_truncate_inode_blocks_enter(inode, from);
764 level = get_node_path(F2FS_I(inode), from, offset, noffset);
766 page = get_node_page(sbi, inode->i_ino);
768 trace_f2fs_truncate_inode_blocks_exit(inode, PTR_ERR(page));
769 return PTR_ERR(page);
772 set_new_dnode(&dn, inode, page, NULL, 0);
775 ri = F2FS_INODE(page);
783 if (!offset[level - 1])
785 err = truncate_partial_nodes(&dn, ri, offset, level);
786 if (err < 0 && err != -ENOENT)
788 nofs += 1 + NIDS_PER_BLOCK;
791 nofs = 5 + 2 * NIDS_PER_BLOCK;
792 if (!offset[level - 1])
794 err = truncate_partial_nodes(&dn, ri, offset, level);
795 if (err < 0 && err != -ENOENT)
804 dn.nid = le32_to_cpu(ri->i_nid[offset[0] - NODE_DIR1_BLOCK]);
806 case NODE_DIR1_BLOCK:
807 case NODE_DIR2_BLOCK:
808 err = truncate_dnode(&dn);
811 case NODE_IND1_BLOCK:
812 case NODE_IND2_BLOCK:
813 err = truncate_nodes(&dn, nofs, offset[1], 2);
816 case NODE_DIND_BLOCK:
817 err = truncate_nodes(&dn, nofs, offset[1], 3);
824 if (err < 0 && err != -ENOENT)
826 if (offset[1] == 0 &&
827 ri->i_nid[offset[0] - NODE_DIR1_BLOCK]) {
829 if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
830 f2fs_put_page(page, 1);
833 f2fs_wait_on_page_writeback(page, NODE);
834 ri->i_nid[offset[0] - NODE_DIR1_BLOCK] = 0;
835 set_page_dirty(page);
843 f2fs_put_page(page, 0);
844 trace_f2fs_truncate_inode_blocks_exit(inode, err);
845 return err > 0 ? 0 : err;
848 int truncate_xattr_node(struct inode *inode, struct page *page)
850 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
851 nid_t nid = F2FS_I(inode)->i_xattr_nid;
852 struct dnode_of_data dn;
858 npage = get_node_page(sbi, nid);
860 return PTR_ERR(npage);
862 F2FS_I(inode)->i_xattr_nid = 0;
864 /* need to do checkpoint during fsync */
865 F2FS_I(inode)->xattr_ver = cur_cp_version(F2FS_CKPT(sbi));
867 set_new_dnode(&dn, inode, page, npage, nid);
870 dn.inode_page_locked = true;
876 * Caller should grab and release a rwsem by calling f2fs_lock_op() and
879 void remove_inode_page(struct inode *inode)
881 struct dnode_of_data dn;
883 set_new_dnode(&dn, inode, NULL, NULL, inode->i_ino);
884 if (get_dnode_of_data(&dn, 0, LOOKUP_NODE))
887 if (truncate_xattr_node(inode, dn.inode_page)) {
892 /* remove potential inline_data blocks */
893 if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
894 S_ISLNK(inode->i_mode))
895 truncate_data_blocks_range(&dn, 1);
897 /* 0 is possible, after f2fs_new_inode() has failed */
898 f2fs_bug_on(F2FS_I_SB(inode),
899 inode->i_blocks != 0 && inode->i_blocks != 1);
901 /* will put inode & node pages */
905 struct page *new_inode_page(struct inode *inode)
907 struct dnode_of_data dn;
909 /* allocate inode page for new inode */
910 set_new_dnode(&dn, inode, NULL, NULL, inode->i_ino);
912 /* caller should f2fs_put_page(page, 1); */
913 return new_node_page(&dn, 0, NULL);
916 struct page *new_node_page(struct dnode_of_data *dn,
917 unsigned int ofs, struct page *ipage)
919 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
920 struct node_info old_ni, new_ni;
924 if (unlikely(is_inode_flag_set(F2FS_I(dn->inode), FI_NO_ALLOC)))
925 return ERR_PTR(-EPERM);
927 page = grab_cache_page(NODE_MAPPING(sbi), dn->nid);
929 return ERR_PTR(-ENOMEM);
931 if (unlikely(!inc_valid_node_count(sbi, dn->inode))) {
936 get_node_info(sbi, dn->nid, &old_ni);
938 /* Reinitialize old_ni with new node page */
939 f2fs_bug_on(sbi, old_ni.blk_addr != NULL_ADDR);
941 new_ni.ino = dn->inode->i_ino;
942 set_node_addr(sbi, &new_ni, NEW_ADDR, false);
944 f2fs_wait_on_page_writeback(page, NODE);
945 fill_node_footer(page, dn->nid, dn->inode->i_ino, ofs, true);
946 set_cold_node(dn->inode, page);
947 SetPageUptodate(page);
948 set_page_dirty(page);
950 if (f2fs_has_xattr_block(ofs))
951 F2FS_I(dn->inode)->i_xattr_nid = dn->nid;
953 dn->node_page = page;
955 update_inode(dn->inode, ipage);
959 inc_valid_inode_count(sbi);
964 clear_node_page_dirty(page);
965 f2fs_put_page(page, 1);
970 * Caller should do after getting the following values.
971 * 0: f2fs_put_page(page, 0)
972 * LOCKED_PAGE: f2fs_put_page(page, 1)
975 static int read_node_page(struct page *page, int rw)
977 struct f2fs_sb_info *sbi = F2FS_P_SB(page);
980 get_node_info(sbi, page->index, &ni);
982 if (unlikely(ni.blk_addr == NULL_ADDR)) {
983 f2fs_put_page(page, 1);
987 if (PageUptodate(page))
990 return f2fs_submit_page_bio(sbi, page, ni.blk_addr, rw);
994 * Readahead a node page
996 void ra_node_page(struct f2fs_sb_info *sbi, nid_t nid)
1001 apage = find_get_page(NODE_MAPPING(sbi), nid);
1002 if (apage && PageUptodate(apage)) {
1003 f2fs_put_page(apage, 0);
1006 f2fs_put_page(apage, 0);
1008 apage = grab_cache_page(NODE_MAPPING(sbi), nid);
1012 err = read_node_page(apage, READA);
1014 f2fs_put_page(apage, 0);
1015 else if (err == LOCKED_PAGE)
1016 f2fs_put_page(apage, 1);
1019 struct page *get_node_page(struct f2fs_sb_info *sbi, pgoff_t nid)
1024 page = grab_cache_page(NODE_MAPPING(sbi), nid);
1026 return ERR_PTR(-ENOMEM);
1028 err = read_node_page(page, READ_SYNC);
1030 return ERR_PTR(err);
1031 else if (err == LOCKED_PAGE)
1035 if (unlikely(!PageUptodate(page) || nid != nid_of_node(page))) {
1036 f2fs_put_page(page, 1);
1037 return ERR_PTR(-EIO);
1039 if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1040 f2fs_put_page(page, 1);
1048 * Return a locked page for the desired node page.
1049 * And, readahead MAX_RA_NODE number of node pages.
1051 struct page *get_node_page_ra(struct page *parent, int start)
1053 struct f2fs_sb_info *sbi = F2FS_P_SB(parent);
1054 struct blk_plug plug;
1059 /* First, try getting the desired direct node. */
1060 nid = get_nid(parent, start, false);
1062 return ERR_PTR(-ENOENT);
1064 page = grab_cache_page(NODE_MAPPING(sbi), nid);
1066 return ERR_PTR(-ENOMEM);
1068 err = read_node_page(page, READ_SYNC);
1070 return ERR_PTR(err);
1071 else if (err == LOCKED_PAGE)
1074 blk_start_plug(&plug);
1076 /* Then, try readahead for siblings of the desired node */
1077 end = start + MAX_RA_NODE;
1078 end = min(end, NIDS_PER_BLOCK);
1079 for (i = start + 1; i < end; i++) {
1080 nid = get_nid(parent, i, false);
1083 ra_node_page(sbi, nid);
1086 blk_finish_plug(&plug);
1089 if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1090 f2fs_put_page(page, 1);
1094 if (unlikely(!PageUptodate(page))) {
1095 f2fs_put_page(page, 1);
1096 return ERR_PTR(-EIO);
1101 void sync_inode_page(struct dnode_of_data *dn)
1103 if (IS_INODE(dn->node_page) || dn->inode_page == dn->node_page) {
1104 update_inode(dn->inode, dn->node_page);
1105 } else if (dn->inode_page) {
1106 if (!dn->inode_page_locked)
1107 lock_page(dn->inode_page);
1108 update_inode(dn->inode, dn->inode_page);
1109 if (!dn->inode_page_locked)
1110 unlock_page(dn->inode_page);
1112 update_inode_page(dn->inode);
1116 int sync_node_pages(struct f2fs_sb_info *sbi, nid_t ino,
1117 struct writeback_control *wbc)
1120 struct pagevec pvec;
1121 int step = ino ? 2 : 0;
1122 int nwritten = 0, wrote = 0;
1124 pagevec_init(&pvec, 0);
1130 while (index <= end) {
1132 nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1133 PAGECACHE_TAG_DIRTY,
1134 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
1138 for (i = 0; i < nr_pages; i++) {
1139 struct page *page = pvec.pages[i];
1142 * flushing sequence with step:
1147 if (step == 0 && IS_DNODE(page))
1149 if (step == 1 && (!IS_DNODE(page) ||
1150 is_cold_node(page)))
1152 if (step == 2 && (!IS_DNODE(page) ||
1153 !is_cold_node(page)))
1158 * we should not skip writing node pages.
1160 if (ino && ino_of_node(page) == ino)
1162 else if (!trylock_page(page))
1165 if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1170 if (ino && ino_of_node(page) != ino)
1171 goto continue_unlock;
1173 if (!PageDirty(page)) {
1174 /* someone wrote it for us */
1175 goto continue_unlock;
1178 if (!clear_page_dirty_for_io(page))
1179 goto continue_unlock;
1181 /* called by fsync() */
1182 if (ino && IS_DNODE(page)) {
1183 set_fsync_mark(page, 1);
1184 if (IS_INODE(page)) {
1185 if (!is_checkpointed_node(sbi, ino) &&
1186 !has_fsynced_inode(sbi, ino))
1187 set_dentry_mark(page, 1);
1189 set_dentry_mark(page, 0);
1193 set_fsync_mark(page, 0);
1194 set_dentry_mark(page, 0);
1197 if (NODE_MAPPING(sbi)->a_ops->writepage(page, wbc))
1202 if (--wbc->nr_to_write == 0)
1205 pagevec_release(&pvec);
1208 if (wbc->nr_to_write == 0) {
1220 f2fs_submit_merged_bio(sbi, NODE, WRITE);
1224 int wait_on_node_pages_writeback(struct f2fs_sb_info *sbi, nid_t ino)
1226 pgoff_t index = 0, end = LONG_MAX;
1227 struct pagevec pvec;
1228 int ret2 = 0, ret = 0;
1230 pagevec_init(&pvec, 0);
1232 while (index <= end) {
1234 nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1235 PAGECACHE_TAG_WRITEBACK,
1236 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
1240 for (i = 0; i < nr_pages; i++) {
1241 struct page *page = pvec.pages[i];
1243 /* until radix tree lookup accepts end_index */
1244 if (unlikely(page->index > end))
1247 if (ino && ino_of_node(page) == ino) {
1248 f2fs_wait_on_page_writeback(page, NODE);
1249 if (TestClearPageError(page))
1253 pagevec_release(&pvec);
1257 if (unlikely(test_and_clear_bit(AS_ENOSPC, &NODE_MAPPING(sbi)->flags)))
1259 if (unlikely(test_and_clear_bit(AS_EIO, &NODE_MAPPING(sbi)->flags)))
1266 static int f2fs_write_node_page(struct page *page,
1267 struct writeback_control *wbc)
1269 struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1272 struct node_info ni;
1273 struct f2fs_io_info fio = {
1275 .rw = (wbc->sync_mode == WB_SYNC_ALL) ? WRITE_SYNC : WRITE,
1278 trace_f2fs_writepage(page, NODE);
1280 if (unlikely(sbi->por_doing))
1282 if (unlikely(f2fs_cp_error(sbi)))
1285 f2fs_wait_on_page_writeback(page, NODE);
1287 /* get old block addr of this node page */
1288 nid = nid_of_node(page);
1289 f2fs_bug_on(sbi, page->index != nid);
1291 get_node_info(sbi, nid, &ni);
1293 /* This page is already truncated */
1294 if (unlikely(ni.blk_addr == NULL_ADDR)) {
1295 dec_page_count(sbi, F2FS_DIRTY_NODES);
1300 if (wbc->for_reclaim) {
1301 if (!down_read_trylock(&sbi->node_write))
1304 down_read(&sbi->node_write);
1306 set_page_writeback(page);
1307 write_node_page(sbi, page, &fio, nid, ni.blk_addr, &new_addr);
1308 set_node_addr(sbi, &ni, new_addr, is_fsync_dnode(page));
1309 dec_page_count(sbi, F2FS_DIRTY_NODES);
1310 up_read(&sbi->node_write);
1313 if (wbc->for_reclaim)
1314 f2fs_submit_merged_bio(sbi, NODE, WRITE);
1319 redirty_page_for_writepage(wbc, page);
1320 return AOP_WRITEPAGE_ACTIVATE;
1323 static int f2fs_write_node_pages(struct address_space *mapping,
1324 struct writeback_control *wbc)
1326 struct f2fs_sb_info *sbi = F2FS_M_SB(mapping);
1329 trace_f2fs_writepages(mapping->host, wbc, NODE);
1331 /* balancing f2fs's metadata in background */
1332 f2fs_balance_fs_bg(sbi);
1334 /* collect a number of dirty node pages and write together */
1335 if (get_pages(sbi, F2FS_DIRTY_NODES) < nr_pages_to_skip(sbi, NODE))
1338 diff = nr_pages_to_write(sbi, NODE, wbc);
1339 wbc->sync_mode = WB_SYNC_NONE;
1340 sync_node_pages(sbi, 0, wbc);
1341 wbc->nr_to_write = max((long)0, wbc->nr_to_write - diff);
1345 wbc->pages_skipped += get_pages(sbi, F2FS_DIRTY_NODES);
1349 static int f2fs_set_node_page_dirty(struct page *page)
1351 trace_f2fs_set_page_dirty(page, NODE);
1353 SetPageUptodate(page);
1354 if (!PageDirty(page)) {
1355 __set_page_dirty_nobuffers(page);
1356 inc_page_count(F2FS_P_SB(page), F2FS_DIRTY_NODES);
1357 SetPagePrivate(page);
1363 static void f2fs_invalidate_node_page(struct page *page, unsigned int offset,
1364 unsigned int length)
1366 struct inode *inode = page->mapping->host;
1367 if (PageDirty(page))
1368 dec_page_count(F2FS_I_SB(inode), F2FS_DIRTY_NODES);
1369 ClearPagePrivate(page);
1372 static int f2fs_release_node_page(struct page *page, gfp_t wait)
1374 ClearPagePrivate(page);
1379 * Structure of the f2fs node operations
1381 const struct address_space_operations f2fs_node_aops = {
1382 .writepage = f2fs_write_node_page,
1383 .writepages = f2fs_write_node_pages,
1384 .set_page_dirty = f2fs_set_node_page_dirty,
1385 .invalidatepage = f2fs_invalidate_node_page,
1386 .releasepage = f2fs_release_node_page,
1389 static struct free_nid *__lookup_free_nid_list(struct f2fs_nm_info *nm_i,
1392 return radix_tree_lookup(&nm_i->free_nid_root, n);
1395 static void __del_from_free_nid_list(struct f2fs_nm_info *nm_i,
1399 radix_tree_delete(&nm_i->free_nid_root, i->nid);
1402 static int add_free_nid(struct f2fs_sb_info *sbi, nid_t nid, bool build)
1404 struct f2fs_nm_info *nm_i = NM_I(sbi);
1406 struct nat_entry *ne;
1407 bool allocated = false;
1409 if (!available_free_memory(sbi, FREE_NIDS))
1412 /* 0 nid should not be used */
1413 if (unlikely(nid == 0))
1417 /* do not add allocated nids */
1418 down_read(&nm_i->nat_tree_lock);
1419 ne = __lookup_nat_cache(nm_i, nid);
1421 (!get_nat_flag(ne, IS_CHECKPOINTED) ||
1422 nat_get_blkaddr(ne) != NULL_ADDR))
1424 up_read(&nm_i->nat_tree_lock);
1429 i = f2fs_kmem_cache_alloc(free_nid_slab, GFP_NOFS);
1433 if (radix_tree_preload(GFP_NOFS)) {
1434 kmem_cache_free(free_nid_slab, i);
1438 spin_lock(&nm_i->free_nid_list_lock);
1439 if (radix_tree_insert(&nm_i->free_nid_root, i->nid, i)) {
1440 spin_unlock(&nm_i->free_nid_list_lock);
1441 radix_tree_preload_end();
1442 kmem_cache_free(free_nid_slab, i);
1445 list_add_tail(&i->list, &nm_i->free_nid_list);
1447 spin_unlock(&nm_i->free_nid_list_lock);
1448 radix_tree_preload_end();
1452 static void remove_free_nid(struct f2fs_nm_info *nm_i, nid_t nid)
1455 bool need_free = false;
1457 spin_lock(&nm_i->free_nid_list_lock);
1458 i = __lookup_free_nid_list(nm_i, nid);
1459 if (i && i->state == NID_NEW) {
1460 __del_from_free_nid_list(nm_i, i);
1464 spin_unlock(&nm_i->free_nid_list_lock);
1467 kmem_cache_free(free_nid_slab, i);
1470 static void scan_nat_page(struct f2fs_sb_info *sbi,
1471 struct page *nat_page, nid_t start_nid)
1473 struct f2fs_nm_info *nm_i = NM_I(sbi);
1474 struct f2fs_nat_block *nat_blk = page_address(nat_page);
1478 i = start_nid % NAT_ENTRY_PER_BLOCK;
1480 for (; i < NAT_ENTRY_PER_BLOCK; i++, start_nid++) {
1482 if (unlikely(start_nid >= nm_i->max_nid))
1485 blk_addr = le32_to_cpu(nat_blk->entries[i].block_addr);
1486 f2fs_bug_on(sbi, blk_addr == NEW_ADDR);
1487 if (blk_addr == NULL_ADDR) {
1488 if (add_free_nid(sbi, start_nid, true) < 0)
1494 static void build_free_nids(struct f2fs_sb_info *sbi)
1496 struct f2fs_nm_info *nm_i = NM_I(sbi);
1497 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
1498 struct f2fs_summary_block *sum = curseg->sum_blk;
1500 nid_t nid = nm_i->next_scan_nid;
1502 /* Enough entries */
1503 if (nm_i->fcnt > NAT_ENTRY_PER_BLOCK)
1506 /* readahead nat pages to be scanned */
1507 ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nid), FREE_NID_PAGES, META_NAT);
1510 struct page *page = get_current_nat_page(sbi, nid);
1512 scan_nat_page(sbi, page, nid);
1513 f2fs_put_page(page, 1);
1515 nid += (NAT_ENTRY_PER_BLOCK - (nid % NAT_ENTRY_PER_BLOCK));
1516 if (unlikely(nid >= nm_i->max_nid))
1519 if (i++ == FREE_NID_PAGES)
1523 /* go to the next free nat pages to find free nids abundantly */
1524 nm_i->next_scan_nid = nid;
1526 /* find free nids from current sum_pages */
1527 mutex_lock(&curseg->curseg_mutex);
1528 for (i = 0; i < nats_in_cursum(sum); i++) {
1529 block_t addr = le32_to_cpu(nat_in_journal(sum, i).block_addr);
1530 nid = le32_to_cpu(nid_in_journal(sum, i));
1531 if (addr == NULL_ADDR)
1532 add_free_nid(sbi, nid, true);
1534 remove_free_nid(nm_i, nid);
1536 mutex_unlock(&curseg->curseg_mutex);
1540 * If this function returns success, caller can obtain a new nid
1541 * from second parameter of this function.
1542 * The returned nid could be used ino as well as nid when inode is created.
1544 bool alloc_nid(struct f2fs_sb_info *sbi, nid_t *nid)
1546 struct f2fs_nm_info *nm_i = NM_I(sbi);
1547 struct free_nid *i = NULL;
1549 if (unlikely(sbi->total_valid_node_count + 1 > nm_i->available_nids))
1552 spin_lock(&nm_i->free_nid_list_lock);
1554 /* We should not use stale free nids created by build_free_nids */
1555 if (nm_i->fcnt && !on_build_free_nids(nm_i)) {
1556 f2fs_bug_on(sbi, list_empty(&nm_i->free_nid_list));
1557 list_for_each_entry(i, &nm_i->free_nid_list, list)
1558 if (i->state == NID_NEW)
1561 f2fs_bug_on(sbi, i->state != NID_NEW);
1563 i->state = NID_ALLOC;
1565 spin_unlock(&nm_i->free_nid_list_lock);
1568 spin_unlock(&nm_i->free_nid_list_lock);
1570 /* Let's scan nat pages and its caches to get free nids */
1571 mutex_lock(&nm_i->build_lock);
1572 build_free_nids(sbi);
1573 mutex_unlock(&nm_i->build_lock);
1578 * alloc_nid() should be called prior to this function.
1580 void alloc_nid_done(struct f2fs_sb_info *sbi, nid_t nid)
1582 struct f2fs_nm_info *nm_i = NM_I(sbi);
1585 spin_lock(&nm_i->free_nid_list_lock);
1586 i = __lookup_free_nid_list(nm_i, nid);
1587 f2fs_bug_on(sbi, !i || i->state != NID_ALLOC);
1588 __del_from_free_nid_list(nm_i, i);
1589 spin_unlock(&nm_i->free_nid_list_lock);
1591 kmem_cache_free(free_nid_slab, i);
1595 * alloc_nid() should be called prior to this function.
1597 void alloc_nid_failed(struct f2fs_sb_info *sbi, nid_t nid)
1599 struct f2fs_nm_info *nm_i = NM_I(sbi);
1601 bool need_free = false;
1606 spin_lock(&nm_i->free_nid_list_lock);
1607 i = __lookup_free_nid_list(nm_i, nid);
1608 f2fs_bug_on(sbi, !i || i->state != NID_ALLOC);
1609 if (!available_free_memory(sbi, FREE_NIDS)) {
1610 __del_from_free_nid_list(nm_i, i);
1616 spin_unlock(&nm_i->free_nid_list_lock);
1619 kmem_cache_free(free_nid_slab, i);
1622 void recover_inline_xattr(struct inode *inode, struct page *page)
1624 void *src_addr, *dst_addr;
1627 struct f2fs_inode *ri;
1629 ipage = get_node_page(F2FS_I_SB(inode), inode->i_ino);
1630 f2fs_bug_on(F2FS_I_SB(inode), IS_ERR(ipage));
1632 ri = F2FS_INODE(page);
1633 if (!(ri->i_inline & F2FS_INLINE_XATTR)) {
1634 clear_inode_flag(F2FS_I(inode), FI_INLINE_XATTR);
1638 dst_addr = inline_xattr_addr(ipage);
1639 src_addr = inline_xattr_addr(page);
1640 inline_size = inline_xattr_size(inode);
1642 f2fs_wait_on_page_writeback(ipage, NODE);
1643 memcpy(dst_addr, src_addr, inline_size);
1645 update_inode(inode, ipage);
1646 f2fs_put_page(ipage, 1);
1649 void recover_xattr_data(struct inode *inode, struct page *page, block_t blkaddr)
1651 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1652 nid_t prev_xnid = F2FS_I(inode)->i_xattr_nid;
1653 nid_t new_xnid = nid_of_node(page);
1654 struct node_info ni;
1656 /* 1: invalidate the previous xattr nid */
1660 /* Deallocate node address */
1661 get_node_info(sbi, prev_xnid, &ni);
1662 f2fs_bug_on(sbi, ni.blk_addr == NULL_ADDR);
1663 invalidate_blocks(sbi, ni.blk_addr);
1664 dec_valid_node_count(sbi, inode);
1665 set_node_addr(sbi, &ni, NULL_ADDR, false);
1668 /* 2: allocate new xattr nid */
1669 if (unlikely(!inc_valid_node_count(sbi, inode)))
1670 f2fs_bug_on(sbi, 1);
1672 remove_free_nid(NM_I(sbi), new_xnid);
1673 get_node_info(sbi, new_xnid, &ni);
1674 ni.ino = inode->i_ino;
1675 set_node_addr(sbi, &ni, NEW_ADDR, false);
1676 F2FS_I(inode)->i_xattr_nid = new_xnid;
1678 /* 3: update xattr blkaddr */
1679 refresh_sit_entry(sbi, NEW_ADDR, blkaddr);
1680 set_node_addr(sbi, &ni, blkaddr, false);
1682 update_inode_page(inode);
1685 int recover_inode_page(struct f2fs_sb_info *sbi, struct page *page)
1687 struct f2fs_inode *src, *dst;
1688 nid_t ino = ino_of_node(page);
1689 struct node_info old_ni, new_ni;
1692 get_node_info(sbi, ino, &old_ni);
1694 if (unlikely(old_ni.blk_addr != NULL_ADDR))
1697 ipage = grab_cache_page(NODE_MAPPING(sbi), ino);
1701 /* Should not use this inode from free nid list */
1702 remove_free_nid(NM_I(sbi), ino);
1704 SetPageUptodate(ipage);
1705 fill_node_footer(ipage, ino, ino, 0, true);
1707 src = F2FS_INODE(page);
1708 dst = F2FS_INODE(ipage);
1710 memcpy(dst, src, (unsigned long)&src->i_ext - (unsigned long)src);
1712 dst->i_blocks = cpu_to_le64(1);
1713 dst->i_links = cpu_to_le32(1);
1714 dst->i_xattr_nid = 0;
1715 dst->i_inline = src->i_inline & F2FS_INLINE_XATTR;
1720 if (unlikely(!inc_valid_node_count(sbi, NULL)))
1722 set_node_addr(sbi, &new_ni, NEW_ADDR, false);
1723 inc_valid_inode_count(sbi);
1724 set_page_dirty(ipage);
1725 f2fs_put_page(ipage, 1);
1730 * ra_sum_pages() merge contiguous pages into one bio and submit.
1731 * these pre-read pages are allocated in bd_inode's mapping tree.
1733 static int ra_sum_pages(struct f2fs_sb_info *sbi, struct page **pages,
1734 int start, int nrpages)
1736 struct inode *inode = sbi->sb->s_bdev->bd_inode;
1737 struct address_space *mapping = inode->i_mapping;
1738 int i, page_idx = start;
1739 struct f2fs_io_info fio = {
1741 .rw = READ_SYNC | REQ_META | REQ_PRIO
1744 for (i = 0; page_idx < start + nrpages; page_idx++, i++) {
1745 /* alloc page in bd_inode for reading node summary info */
1746 pages[i] = grab_cache_page(mapping, page_idx);
1749 f2fs_submit_page_mbio(sbi, pages[i], page_idx, &fio);
1752 f2fs_submit_merged_bio(sbi, META, READ);
1756 int restore_node_summary(struct f2fs_sb_info *sbi,
1757 unsigned int segno, struct f2fs_summary_block *sum)
1759 struct f2fs_node *rn;
1760 struct f2fs_summary *sum_entry;
1761 struct inode *inode = sbi->sb->s_bdev->bd_inode;
1763 int bio_blocks = MAX_BIO_BLOCKS(sbi);
1764 struct page *pages[bio_blocks];
1765 int i, idx, last_offset, nrpages, err = 0;
1767 /* scan the node segment */
1768 last_offset = sbi->blocks_per_seg;
1769 addr = START_BLOCK(sbi, segno);
1770 sum_entry = &sum->entries[0];
1772 for (i = 0; !err && i < last_offset; i += nrpages, addr += nrpages) {
1773 nrpages = min(last_offset - i, bio_blocks);
1775 /* readahead node pages */
1776 nrpages = ra_sum_pages(sbi, pages, addr, nrpages);
1780 for (idx = 0; idx < nrpages; idx++) {
1784 lock_page(pages[idx]);
1785 if (unlikely(!PageUptodate(pages[idx]))) {
1788 rn = F2FS_NODE(pages[idx]);
1789 sum_entry->nid = rn->footer.nid;
1790 sum_entry->version = 0;
1791 sum_entry->ofs_in_node = 0;
1794 unlock_page(pages[idx]);
1796 page_cache_release(pages[idx]);
1799 invalidate_mapping_pages(inode->i_mapping, addr,
1805 static void remove_nats_in_journal(struct f2fs_sb_info *sbi)
1807 struct f2fs_nm_info *nm_i = NM_I(sbi);
1808 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
1809 struct f2fs_summary_block *sum = curseg->sum_blk;
1812 mutex_lock(&curseg->curseg_mutex);
1813 for (i = 0; i < nats_in_cursum(sum); i++) {
1814 struct nat_entry *ne;
1815 struct f2fs_nat_entry raw_ne;
1816 nid_t nid = le32_to_cpu(nid_in_journal(sum, i));
1818 raw_ne = nat_in_journal(sum, i);
1820 down_write(&nm_i->nat_tree_lock);
1821 ne = __lookup_nat_cache(nm_i, nid);
1823 ne = grab_nat_entry(nm_i, nid);
1824 node_info_from_raw_nat(&ne->ni, &raw_ne);
1826 __set_nat_cache_dirty(nm_i, ne);
1827 up_write(&nm_i->nat_tree_lock);
1829 update_nats_in_cursum(sum, -i);
1830 mutex_unlock(&curseg->curseg_mutex);
1833 static void __adjust_nat_entry_set(struct nat_entry_set *nes,
1834 struct list_head *head, int max)
1836 struct nat_entry_set *cur;
1838 if (nes->entry_cnt >= max)
1841 list_for_each_entry(cur, head, set_list) {
1842 if (cur->entry_cnt >= nes->entry_cnt) {
1843 list_add(&nes->set_list, cur->set_list.prev);
1848 list_add_tail(&nes->set_list, head);
1851 static void __flush_nat_entry_set(struct f2fs_sb_info *sbi,
1852 struct nat_entry_set *set)
1854 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
1855 struct f2fs_summary_block *sum = curseg->sum_blk;
1856 nid_t start_nid = set->set * NAT_ENTRY_PER_BLOCK;
1857 bool to_journal = true;
1858 struct f2fs_nat_block *nat_blk;
1859 struct nat_entry *ne, *cur;
1860 struct page *page = NULL;
1863 * there are two steps to flush nat entries:
1864 * #1, flush nat entries to journal in current hot data summary block.
1865 * #2, flush nat entries to nat page.
1867 if (!__has_cursum_space(sum, set->entry_cnt, NAT_JOURNAL))
1871 mutex_lock(&curseg->curseg_mutex);
1873 page = get_next_nat_page(sbi, start_nid);
1874 nat_blk = page_address(page);
1875 f2fs_bug_on(sbi, !nat_blk);
1878 /* flush dirty nats in nat entry set */
1879 list_for_each_entry_safe(ne, cur, &set->entry_list, list) {
1880 struct f2fs_nat_entry *raw_ne;
1881 nid_t nid = nat_get_nid(ne);
1884 if (nat_get_blkaddr(ne) == NEW_ADDR)
1888 offset = lookup_journal_in_cursum(sum,
1889 NAT_JOURNAL, nid, 1);
1890 f2fs_bug_on(sbi, offset < 0);
1891 raw_ne = &nat_in_journal(sum, offset);
1892 nid_in_journal(sum, offset) = cpu_to_le32(nid);
1894 raw_ne = &nat_blk->entries[nid - start_nid];
1896 raw_nat_from_node_info(raw_ne, &ne->ni);
1898 down_write(&NM_I(sbi)->nat_tree_lock);
1900 __clear_nat_cache_dirty(NM_I(sbi), ne);
1901 up_write(&NM_I(sbi)->nat_tree_lock);
1903 if (nat_get_blkaddr(ne) == NULL_ADDR)
1904 add_free_nid(sbi, nid, false);
1908 mutex_unlock(&curseg->curseg_mutex);
1910 f2fs_put_page(page, 1);
1912 f2fs_bug_on(sbi, set->entry_cnt);
1914 radix_tree_delete(&NM_I(sbi)->nat_set_root, set->set);
1915 kmem_cache_free(nat_entry_set_slab, set);
1919 * This function is called during the checkpointing process.
1921 void flush_nat_entries(struct f2fs_sb_info *sbi)
1923 struct f2fs_nm_info *nm_i = NM_I(sbi);
1924 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
1925 struct f2fs_summary_block *sum = curseg->sum_blk;
1926 struct nat_entry_set *setvec[NATVEC_SIZE];
1927 struct nat_entry_set *set, *tmp;
1932 if (!nm_i->dirty_nat_cnt)
1935 * if there are no enough space in journal to store dirty nat
1936 * entries, remove all entries from journal and merge them
1937 * into nat entry set.
1939 if (!__has_cursum_space(sum, nm_i->dirty_nat_cnt, NAT_JOURNAL))
1940 remove_nats_in_journal(sbi);
1942 while ((found = __gang_lookup_nat_set(nm_i,
1943 set_idx, NATVEC_SIZE, setvec))) {
1945 set_idx = setvec[found - 1]->set + 1;
1946 for (idx = 0; idx < found; idx++)
1947 __adjust_nat_entry_set(setvec[idx], &sets,
1948 MAX_NAT_JENTRIES(sum));
1951 /* flush dirty nats in nat entry set */
1952 list_for_each_entry_safe(set, tmp, &sets, set_list)
1953 __flush_nat_entry_set(sbi, set);
1955 f2fs_bug_on(sbi, nm_i->dirty_nat_cnt);
1958 static int init_node_manager(struct f2fs_sb_info *sbi)
1960 struct f2fs_super_block *sb_raw = F2FS_RAW_SUPER(sbi);
1961 struct f2fs_nm_info *nm_i = NM_I(sbi);
1962 unsigned char *version_bitmap;
1963 unsigned int nat_segs, nat_blocks;
1965 nm_i->nat_blkaddr = le32_to_cpu(sb_raw->nat_blkaddr);
1967 /* segment_count_nat includes pair segment so divide to 2. */
1968 nat_segs = le32_to_cpu(sb_raw->segment_count_nat) >> 1;
1969 nat_blocks = nat_segs << le32_to_cpu(sb_raw->log_blocks_per_seg);
1971 nm_i->max_nid = NAT_ENTRY_PER_BLOCK * nat_blocks;
1973 /* not used nids: 0, node, meta, (and root counted as valid node) */
1974 nm_i->available_nids = nm_i->max_nid - F2FS_RESERVED_NODE_NUM;
1977 nm_i->ram_thresh = DEF_RAM_THRESHOLD;
1979 INIT_RADIX_TREE(&nm_i->free_nid_root, GFP_ATOMIC);
1980 INIT_LIST_HEAD(&nm_i->free_nid_list);
1981 INIT_RADIX_TREE(&nm_i->nat_root, GFP_NOIO);
1982 INIT_RADIX_TREE(&nm_i->nat_set_root, GFP_NOIO);
1983 INIT_LIST_HEAD(&nm_i->nat_entries);
1985 mutex_init(&nm_i->build_lock);
1986 spin_lock_init(&nm_i->free_nid_list_lock);
1987 init_rwsem(&nm_i->nat_tree_lock);
1989 nm_i->next_scan_nid = le32_to_cpu(sbi->ckpt->next_free_nid);
1990 nm_i->bitmap_size = __bitmap_size(sbi, NAT_BITMAP);
1991 version_bitmap = __bitmap_ptr(sbi, NAT_BITMAP);
1992 if (!version_bitmap)
1995 nm_i->nat_bitmap = kmemdup(version_bitmap, nm_i->bitmap_size,
1997 if (!nm_i->nat_bitmap)
2002 int build_node_manager(struct f2fs_sb_info *sbi)
2006 sbi->nm_info = kzalloc(sizeof(struct f2fs_nm_info), GFP_KERNEL);
2010 err = init_node_manager(sbi);
2014 build_free_nids(sbi);
2018 void destroy_node_manager(struct f2fs_sb_info *sbi)
2020 struct f2fs_nm_info *nm_i = NM_I(sbi);
2021 struct free_nid *i, *next_i;
2022 struct nat_entry *natvec[NATVEC_SIZE];
2029 /* destroy free nid list */
2030 spin_lock(&nm_i->free_nid_list_lock);
2031 list_for_each_entry_safe(i, next_i, &nm_i->free_nid_list, list) {
2032 f2fs_bug_on(sbi, i->state == NID_ALLOC);
2033 __del_from_free_nid_list(nm_i, i);
2035 spin_unlock(&nm_i->free_nid_list_lock);
2036 kmem_cache_free(free_nid_slab, i);
2037 spin_lock(&nm_i->free_nid_list_lock);
2039 f2fs_bug_on(sbi, nm_i->fcnt);
2040 spin_unlock(&nm_i->free_nid_list_lock);
2042 /* destroy nat cache */
2043 down_write(&nm_i->nat_tree_lock);
2044 while ((found = __gang_lookup_nat_cache(nm_i,
2045 nid, NATVEC_SIZE, natvec))) {
2047 nid = nat_get_nid(natvec[found - 1]) + 1;
2048 for (idx = 0; idx < found; idx++)
2049 __del_from_nat_cache(nm_i, natvec[idx]);
2051 f2fs_bug_on(sbi, nm_i->nat_cnt);
2052 up_write(&nm_i->nat_tree_lock);
2054 kfree(nm_i->nat_bitmap);
2055 sbi->nm_info = NULL;
2059 int __init create_node_manager_caches(void)
2061 nat_entry_slab = f2fs_kmem_cache_create("nat_entry",
2062 sizeof(struct nat_entry));
2063 if (!nat_entry_slab)
2066 free_nid_slab = f2fs_kmem_cache_create("free_nid",
2067 sizeof(struct free_nid));
2069 goto destroy_nat_entry;
2071 nat_entry_set_slab = f2fs_kmem_cache_create("nat_entry_set",
2072 sizeof(struct nat_entry_set));
2073 if (!nat_entry_set_slab)
2074 goto destroy_free_nid;
2078 kmem_cache_destroy(free_nid_slab);
2080 kmem_cache_destroy(nat_entry_slab);
2085 void destroy_node_manager_caches(void)
2087 kmem_cache_destroy(nat_entry_set_slab);
2088 kmem_cache_destroy(free_nid_slab);
2089 kmem_cache_destroy(nat_entry_slab);