f2fs: add node operations
authorJaegeuk Kim <jaegeuk.kim@samsung.com>
Fri, 2 Nov 2012 08:08:50 +0000 (17:08 +0900)
committerJaegeuk Kim <jaegeuk.kim@samsung.com>
Tue, 11 Dec 2012 04:43:40 +0000 (13:43 +0900)
This adds specific functions to manage NAT pages, a cache for NAT entries, free
nids, direct/indirect node blocks for indexing data, and address space for node
pages.

- The key information of an NAT entry consists of a node id and a block address.

- An NAT page is composed of block addresses covered by a certain range of NAT
  entries, which is maintained by the address space of meta_inode.

- A radix tree structure is used to cache NAT entries. The index for the tree
  is a node id.

- When there is no free nid, F2FS should scan NAT entries to find new one. In
  order to avoid scanning frequently, F2FS manages a list containing a number of
  free nids in memory. Only when free nids in the list are exhausted, scanning
  process, build_free_nids(), is triggered.

- F2FS has direct and indirect node blocks for indexing data. This patch adds
  fuctions related to the node block management such as getting, allocating, and
  truncating node blocks to index data.

- In order to cache node blocks in memory, F2FS has a node_inode with an address
  space for node pages. This patch also adds the address space operations for
  node_inode.

Signed-off-by: Jaegeuk Kim <jaegeuk.kim@samsung.com>
fs/f2fs/node.c [new file with mode: 0644]

diff --git a/fs/f2fs/node.c b/fs/f2fs/node.c
new file mode 100644 (file)
index 0000000..216f04d
--- /dev/null
@@ -0,0 +1,1763 @@
+/**
+ * fs/f2fs/node.c
+ *
+ * Copyright (c) 2012 Samsung Electronics Co., Ltd.
+ *             http://www.samsung.com/
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 as
+ * published by the Free Software Foundation.
+ */
+#include <linux/fs.h>
+#include <linux/f2fs_fs.h>
+#include <linux/mpage.h>
+#include <linux/backing-dev.h>
+#include <linux/blkdev.h>
+#include <linux/pagevec.h>
+#include <linux/swap.h>
+
+#include "f2fs.h"
+#include "node.h"
+#include "segment.h"
+
+static struct kmem_cache *nat_entry_slab;
+static struct kmem_cache *free_nid_slab;
+
+static void clear_node_page_dirty(struct page *page)
+{
+       struct address_space *mapping = page->mapping;
+       struct f2fs_sb_info *sbi = F2FS_SB(mapping->host->i_sb);
+       unsigned int long flags;
+
+       if (PageDirty(page)) {
+               spin_lock_irqsave(&mapping->tree_lock, flags);
+               radix_tree_tag_clear(&mapping->page_tree,
+                               page_index(page),
+                               PAGECACHE_TAG_DIRTY);
+               spin_unlock_irqrestore(&mapping->tree_lock, flags);
+
+               clear_page_dirty_for_io(page);
+               dec_page_count(sbi, F2FS_DIRTY_NODES);
+       }
+       ClearPageUptodate(page);
+}
+
+static struct page *get_current_nat_page(struct f2fs_sb_info *sbi, nid_t nid)
+{
+       pgoff_t index = current_nat_addr(sbi, nid);
+       return get_meta_page(sbi, index);
+}
+
+static struct page *get_next_nat_page(struct f2fs_sb_info *sbi, nid_t nid)
+{
+       struct page *src_page;
+       struct page *dst_page;
+       pgoff_t src_off;
+       pgoff_t dst_off;
+       void *src_addr;
+       void *dst_addr;
+       struct f2fs_nm_info *nm_i = NM_I(sbi);
+
+       src_off = current_nat_addr(sbi, nid);
+       dst_off = next_nat_addr(sbi, src_off);
+
+       /* get current nat block page with lock */
+       src_page = get_meta_page(sbi, src_off);
+
+       /* Dirty src_page means that it is already the new target NAT page. */
+       if (PageDirty(src_page))
+               return src_page;
+
+       dst_page = grab_meta_page(sbi, dst_off);
+
+       src_addr = page_address(src_page);
+       dst_addr = page_address(dst_page);
+       memcpy(dst_addr, src_addr, PAGE_CACHE_SIZE);
+       set_page_dirty(dst_page);
+       f2fs_put_page(src_page, 1);
+
+       set_to_next_nat(nm_i, nid);
+
+       return dst_page;
+}
+
+/**
+ * Readahead NAT pages
+ */
+static void ra_nat_pages(struct f2fs_sb_info *sbi, int nid)
+{
+       struct address_space *mapping = sbi->meta_inode->i_mapping;
+       struct f2fs_nm_info *nm_i = NM_I(sbi);
+       struct page *page;
+       pgoff_t index;
+       int i;
+
+       for (i = 0; i < FREE_NID_PAGES; i++, nid += NAT_ENTRY_PER_BLOCK) {
+               if (nid >= nm_i->max_nid)
+                       nid = 0;
+               index = current_nat_addr(sbi, nid);
+
+               page = grab_cache_page(mapping, index);
+               if (!page)
+                       continue;
+               if (f2fs_readpage(sbi, page, index, READ)) {
+                       f2fs_put_page(page, 1);
+                       continue;
+               }
+               page_cache_release(page);
+       }
+}
+
+static struct nat_entry *__lookup_nat_cache(struct f2fs_nm_info *nm_i, nid_t n)
+{
+       return radix_tree_lookup(&nm_i->nat_root, n);
+}
+
+static unsigned int __gang_lookup_nat_cache(struct f2fs_nm_info *nm_i,
+               nid_t start, unsigned int nr, struct nat_entry **ep)
+{
+       return radix_tree_gang_lookup(&nm_i->nat_root, (void **)ep, start, nr);
+}
+
+static void __del_from_nat_cache(struct f2fs_nm_info *nm_i, struct nat_entry *e)
+{
+       list_del(&e->list);
+       radix_tree_delete(&nm_i->nat_root, nat_get_nid(e));
+       nm_i->nat_cnt--;
+       kmem_cache_free(nat_entry_slab, e);
+}
+
+int is_checkpointed_node(struct f2fs_sb_info *sbi, nid_t nid)
+{
+       struct f2fs_nm_info *nm_i = NM_I(sbi);
+       struct nat_entry *e;
+       int is_cp = 1;
+
+       read_lock(&nm_i->nat_tree_lock);
+       e = __lookup_nat_cache(nm_i, nid);
+       if (e && !e->checkpointed)
+               is_cp = 0;
+       read_unlock(&nm_i->nat_tree_lock);
+       return is_cp;
+}
+
+static struct nat_entry *grab_nat_entry(struct f2fs_nm_info *nm_i, nid_t nid)
+{
+       struct nat_entry *new;
+
+       new = kmem_cache_alloc(nat_entry_slab, GFP_ATOMIC);
+       if (!new)
+               return NULL;
+       if (radix_tree_insert(&nm_i->nat_root, nid, new)) {
+               kmem_cache_free(nat_entry_slab, new);
+               return NULL;
+       }
+       memset(new, 0, sizeof(struct nat_entry));
+       nat_set_nid(new, nid);
+       list_add_tail(&new->list, &nm_i->nat_entries);
+       nm_i->nat_cnt++;
+       return new;
+}
+
+static void cache_nat_entry(struct f2fs_nm_info *nm_i, nid_t nid,
+                                               struct f2fs_nat_entry *ne)
+{
+       struct nat_entry *e;
+retry:
+       write_lock(&nm_i->nat_tree_lock);
+       e = __lookup_nat_cache(nm_i, nid);
+       if (!e) {
+               e = grab_nat_entry(nm_i, nid);
+               if (!e) {
+                       write_unlock(&nm_i->nat_tree_lock);
+                       goto retry;
+               }
+               nat_set_blkaddr(e, le32_to_cpu(ne->block_addr));
+               nat_set_ino(e, le32_to_cpu(ne->ino));
+               nat_set_version(e, ne->version);
+               e->checkpointed = true;
+       }
+       write_unlock(&nm_i->nat_tree_lock);
+}
+
+static void set_node_addr(struct f2fs_sb_info *sbi, struct node_info *ni,
+                       block_t new_blkaddr)
+{
+       struct f2fs_nm_info *nm_i = NM_I(sbi);
+       struct nat_entry *e;
+retry:
+       write_lock(&nm_i->nat_tree_lock);
+       e = __lookup_nat_cache(nm_i, ni->nid);
+       if (!e) {
+               e = grab_nat_entry(nm_i, ni->nid);
+               if (!e) {
+                       write_unlock(&nm_i->nat_tree_lock);
+                       goto retry;
+               }
+               e->ni = *ni;
+               e->checkpointed = true;
+               BUG_ON(ni->blk_addr == NEW_ADDR);
+       } else if (new_blkaddr == NEW_ADDR) {
+               /*
+                * when nid is reallocated,
+                * previous nat entry can be remained in nat cache.
+                * So, reinitialize it with new information.
+                */
+               e->ni = *ni;
+               BUG_ON(ni->blk_addr != NULL_ADDR);
+       }
+
+       if (new_blkaddr == NEW_ADDR)
+               e->checkpointed = false;
+
+       /* sanity check */
+       BUG_ON(nat_get_blkaddr(e) != ni->blk_addr);
+       BUG_ON(nat_get_blkaddr(e) == NULL_ADDR &&
+                       new_blkaddr == NULL_ADDR);
+       BUG_ON(nat_get_blkaddr(e) == NEW_ADDR &&
+                       new_blkaddr == NEW_ADDR);
+       BUG_ON(nat_get_blkaddr(e) != NEW_ADDR &&
+                       nat_get_blkaddr(e) != NULL_ADDR &&
+                       new_blkaddr == NEW_ADDR);
+
+       /* increament version no as node is removed */
+       if (nat_get_blkaddr(e) != NEW_ADDR && new_blkaddr == NULL_ADDR) {
+               unsigned char version = nat_get_version(e);
+               nat_set_version(e, inc_node_version(version));
+       }
+
+       /* change address */
+       nat_set_blkaddr(e, new_blkaddr);
+       __set_nat_cache_dirty(nm_i, e);
+       write_unlock(&nm_i->nat_tree_lock);
+}
+
+static int try_to_free_nats(struct f2fs_sb_info *sbi, int nr_shrink)
+{
+       struct f2fs_nm_info *nm_i = NM_I(sbi);
+
+       if (nm_i->nat_cnt < 2 * NM_WOUT_THRESHOLD)
+               return 0;
+
+       write_lock(&nm_i->nat_tree_lock);
+       while (nr_shrink && !list_empty(&nm_i->nat_entries)) {
+               struct nat_entry *ne;
+               ne = list_first_entry(&nm_i->nat_entries,
+                                       struct nat_entry, list);
+               __del_from_nat_cache(nm_i, ne);
+               nr_shrink--;
+       }
+       write_unlock(&nm_i->nat_tree_lock);
+       return nr_shrink;
+}
+
+/**
+ * This function returns always success
+ */
+void get_node_info(struct f2fs_sb_info *sbi, nid_t nid, struct node_info *ni)
+{
+       struct f2fs_nm_info *nm_i = NM_I(sbi);
+       struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
+       struct f2fs_summary_block *sum = curseg->sum_blk;
+       nid_t start_nid = START_NID(nid);
+       struct f2fs_nat_block *nat_blk;
+       struct page *page = NULL;
+       struct f2fs_nat_entry ne;
+       struct nat_entry *e;
+       int i;
+
+       ni->nid = nid;
+
+       /* Check nat cache */
+       read_lock(&nm_i->nat_tree_lock);
+       e = __lookup_nat_cache(nm_i, nid);
+       if (e) {
+               ni->ino = nat_get_ino(e);
+               ni->blk_addr = nat_get_blkaddr(e);
+               ni->version = nat_get_version(e);
+       }
+       read_unlock(&nm_i->nat_tree_lock);
+       if (e)
+               return;
+
+       /* Check current segment summary */
+       mutex_lock(&curseg->curseg_mutex);
+       i = lookup_journal_in_cursum(sum, NAT_JOURNAL, nid, 0);
+       if (i >= 0) {
+               ne = nat_in_journal(sum, i);
+               node_info_from_raw_nat(ni, &ne);
+       }
+       mutex_unlock(&curseg->curseg_mutex);
+       if (i >= 0)
+               goto cache;
+
+       /* Fill node_info from nat page */
+       page = get_current_nat_page(sbi, start_nid);
+       nat_blk = (struct f2fs_nat_block *)page_address(page);
+       ne = nat_blk->entries[nid - start_nid];
+       node_info_from_raw_nat(ni, &ne);
+       f2fs_put_page(page, 1);
+cache:
+       /* cache nat entry */
+       cache_nat_entry(NM_I(sbi), nid, &ne);
+}
+
+/**
+ * The maximum depth is four.
+ * Offset[0] will have raw inode offset.
+ */
+static int get_node_path(long block, int offset[4], unsigned int noffset[4])
+{
+       const long direct_index = ADDRS_PER_INODE;
+       const long direct_blks = ADDRS_PER_BLOCK;
+       const long dptrs_per_blk = NIDS_PER_BLOCK;
+       const long indirect_blks = ADDRS_PER_BLOCK * NIDS_PER_BLOCK;
+       const long dindirect_blks = indirect_blks * NIDS_PER_BLOCK;
+       int n = 0;
+       int level = 0;
+
+       noffset[0] = 0;
+
+       if (block < direct_index) {
+               offset[n++] = block;
+               level = 0;
+               goto got;
+       }
+       block -= direct_index;
+       if (block < direct_blks) {
+               offset[n++] = NODE_DIR1_BLOCK;
+               noffset[n] = 1;
+               offset[n++] = block;
+               level = 1;
+               goto got;
+       }
+       block -= direct_blks;
+       if (block < direct_blks) {
+               offset[n++] = NODE_DIR2_BLOCK;
+               noffset[n] = 2;
+               offset[n++] = block;
+               level = 1;
+               goto got;
+       }
+       block -= direct_blks;
+       if (block < indirect_blks) {
+               offset[n++] = NODE_IND1_BLOCK;
+               noffset[n] = 3;
+               offset[n++] = block / direct_blks;
+               noffset[n] = 4 + offset[n - 1];
+               offset[n++] = block % direct_blks;
+               level = 2;
+               goto got;
+       }
+       block -= indirect_blks;
+       if (block < indirect_blks) {
+               offset[n++] = NODE_IND2_BLOCK;
+               noffset[n] = 4 + dptrs_per_blk;
+               offset[n++] = block / direct_blks;
+               noffset[n] = 5 + dptrs_per_blk + offset[n - 1];
+               offset[n++] = block % direct_blks;
+               level = 2;
+               goto got;
+       }
+       block -= indirect_blks;
+       if (block < dindirect_blks) {
+               offset[n++] = NODE_DIND_BLOCK;
+               noffset[n] = 5 + (dptrs_per_blk * 2);
+               offset[n++] = block / indirect_blks;
+               noffset[n] = 6 + (dptrs_per_blk * 2) +
+                             offset[n - 1] * (dptrs_per_blk + 1);
+               offset[n++] = (block / direct_blks) % dptrs_per_blk;
+               noffset[n] = 7 + (dptrs_per_blk * 2) +
+                             offset[n - 2] * (dptrs_per_blk + 1) +
+                             offset[n - 1];
+               offset[n++] = block % direct_blks;
+               level = 3;
+               goto got;
+       } else {
+               BUG();
+       }
+got:
+       return level;
+}
+
+/*
+ * Caller should call f2fs_put_dnode(dn).
+ */
+int get_dnode_of_data(struct dnode_of_data *dn, pgoff_t index, int ro)
+{
+       struct f2fs_sb_info *sbi = F2FS_SB(dn->inode->i_sb);
+       struct page *npage[4];
+       struct page *parent;
+       int offset[4];
+       unsigned int noffset[4];
+       nid_t nids[4];
+       int level, i;
+       int err = 0;
+
+       level = get_node_path(index, offset, noffset);
+
+       nids[0] = dn->inode->i_ino;
+       npage[0] = get_node_page(sbi, nids[0]);
+       if (IS_ERR(npage[0]))
+               return PTR_ERR(npage[0]);
+
+       parent = npage[0];
+       nids[1] = get_nid(parent, offset[0], true);
+       dn->inode_page = npage[0];
+       dn->inode_page_locked = true;
+
+       /* get indirect or direct nodes */
+       for (i = 1; i <= level; i++) {
+               bool done = false;
+
+               if (!nids[i] && !ro) {
+                       mutex_lock_op(sbi, NODE_NEW);
+
+                       /* alloc new node */
+                       if (!alloc_nid(sbi, &(nids[i]))) {
+                               mutex_unlock_op(sbi, NODE_NEW);
+                               err = -ENOSPC;
+                               goto release_pages;
+                       }
+
+                       dn->nid = nids[i];
+                       npage[i] = new_node_page(dn, noffset[i]);
+                       if (IS_ERR(npage[i])) {
+                               alloc_nid_failed(sbi, nids[i]);
+                               mutex_unlock_op(sbi, NODE_NEW);
+                               err = PTR_ERR(npage[i]);
+                               goto release_pages;
+                       }
+
+                       set_nid(parent, offset[i - 1], nids[i], i == 1);
+                       alloc_nid_done(sbi, nids[i]);
+                       mutex_unlock_op(sbi, NODE_NEW);
+                       done = true;
+               } else if (ro && i == level && level > 1) {
+                       npage[i] = get_node_page_ra(parent, offset[i - 1]);
+                       if (IS_ERR(npage[i])) {
+                               err = PTR_ERR(npage[i]);
+                               goto release_pages;
+                       }
+                       done = true;
+               }
+               if (i == 1) {
+                       dn->inode_page_locked = false;
+                       unlock_page(parent);
+               } else {
+                       f2fs_put_page(parent, 1);
+               }
+
+               if (!done) {
+                       npage[i] = get_node_page(sbi, nids[i]);
+                       if (IS_ERR(npage[i])) {
+                               err = PTR_ERR(npage[i]);
+                               f2fs_put_page(npage[0], 0);
+                               goto release_out;
+                       }
+               }
+               if (i < level) {
+                       parent = npage[i];
+                       nids[i + 1] = get_nid(parent, offset[i], false);
+               }
+       }
+       dn->nid = nids[level];
+       dn->ofs_in_node = offset[level];
+       dn->node_page = npage[level];
+       dn->data_blkaddr = datablock_addr(dn->node_page, dn->ofs_in_node);
+       return 0;
+
+release_pages:
+       f2fs_put_page(parent, 1);
+       if (i > 1)
+               f2fs_put_page(npage[0], 0);
+release_out:
+       dn->inode_page = NULL;
+       dn->node_page = NULL;
+       return err;
+}
+
+static void truncate_node(struct dnode_of_data *dn)
+{
+       struct f2fs_sb_info *sbi = F2FS_SB(dn->inode->i_sb);
+       struct node_info ni;
+
+       get_node_info(sbi, dn->nid, &ni);
+       BUG_ON(ni.blk_addr == NULL_ADDR);
+
+       if (ni.blk_addr != NULL_ADDR)
+               invalidate_blocks(sbi, ni.blk_addr);
+
+       /* Deallocate node address */
+       dec_valid_node_count(sbi, dn->inode, 1);
+       set_node_addr(sbi, &ni, NULL_ADDR);
+
+       if (dn->nid == dn->inode->i_ino) {
+               remove_orphan_inode(sbi, dn->nid);
+               dec_valid_inode_count(sbi);
+       } else {
+               sync_inode_page(dn);
+       }
+
+       clear_node_page_dirty(dn->node_page);
+       F2FS_SET_SB_DIRT(sbi);
+
+       f2fs_put_page(dn->node_page, 1);
+       dn->node_page = NULL;
+}
+
+static int truncate_dnode(struct dnode_of_data *dn)
+{
+       struct f2fs_sb_info *sbi = F2FS_SB(dn->inode->i_sb);
+       struct page *page;
+
+       if (dn->nid == 0)
+               return 1;
+
+       /* get direct node */
+       page = get_node_page(sbi, dn->nid);
+       if (IS_ERR(page) && PTR_ERR(page) == -ENOENT)
+               return 1;
+       else if (IS_ERR(page))
+               return PTR_ERR(page);
+
+       /* Make dnode_of_data for parameter */
+       dn->node_page = page;
+       dn->ofs_in_node = 0;
+       truncate_data_blocks(dn);
+       truncate_node(dn);
+       return 1;
+}
+
+static int truncate_nodes(struct dnode_of_data *dn, unsigned int nofs,
+                                               int ofs, int depth)
+{
+       struct f2fs_sb_info *sbi = F2FS_SB(dn->inode->i_sb);
+       struct dnode_of_data rdn = *dn;
+       struct page *page;
+       struct f2fs_node *rn;
+       nid_t child_nid;
+       unsigned int child_nofs;
+       int freed = 0;
+       int i, ret;
+
+       if (dn->nid == 0)
+               return NIDS_PER_BLOCK + 1;
+
+       page = get_node_page(sbi, dn->nid);
+       if (IS_ERR(page))
+               return PTR_ERR(page);
+
+       rn = (struct f2fs_node *)page_address(page);
+       if (depth < 3) {
+               for (i = ofs; i < NIDS_PER_BLOCK; i++, freed++) {
+                       child_nid = le32_to_cpu(rn->in.nid[i]);
+                       if (child_nid == 0)
+                               continue;
+                       rdn.nid = child_nid;
+                       ret = truncate_dnode(&rdn);
+                       if (ret < 0)
+                               goto out_err;
+                       set_nid(page, i, 0, false);
+               }
+       } else {
+               child_nofs = nofs + ofs * (NIDS_PER_BLOCK + 1) + 1;
+               for (i = ofs; i < NIDS_PER_BLOCK; i++) {
+                       child_nid = le32_to_cpu(rn->in.nid[i]);
+                       if (child_nid == 0) {
+                               child_nofs += NIDS_PER_BLOCK + 1;
+                               continue;
+                       }
+                       rdn.nid = child_nid;
+                       ret = truncate_nodes(&rdn, child_nofs, 0, depth - 1);
+                       if (ret == (NIDS_PER_BLOCK + 1)) {
+                               set_nid(page, i, 0, false);
+                               child_nofs += ret;
+                       } else if (ret < 0 && ret != -ENOENT) {
+                               goto out_err;
+                       }
+               }
+               freed = child_nofs;
+       }
+
+       if (!ofs) {
+               /* remove current indirect node */
+               dn->node_page = page;
+               truncate_node(dn);
+               freed++;
+       } else {
+               f2fs_put_page(page, 1);
+       }
+       return freed;
+
+out_err:
+       f2fs_put_page(page, 1);
+       return ret;
+}
+
+static int truncate_partial_nodes(struct dnode_of_data *dn,
+                       struct f2fs_inode *ri, int *offset, int depth)
+{
+       struct f2fs_sb_info *sbi = F2FS_SB(dn->inode->i_sb);
+       struct page *pages[2];
+       nid_t nid[3];
+       nid_t child_nid;
+       int err = 0;
+       int i;
+       int idx = depth - 2;
+
+       nid[0] = le32_to_cpu(ri->i_nid[offset[0] - NODE_DIR1_BLOCK]);
+       if (!nid[0])
+               return 0;
+
+       /* get indirect nodes in the path */
+       for (i = 0; i < depth - 1; i++) {
+               /* refernece count'll be increased */
+               pages[i] = get_node_page(sbi, nid[i]);
+               if (IS_ERR(pages[i])) {
+                       depth = i + 1;
+                       err = PTR_ERR(pages[i]);
+                       goto fail;
+               }
+               nid[i + 1] = get_nid(pages[i], offset[i + 1], false);
+       }
+
+       /* free direct nodes linked to a partial indirect node */
+       for (i = offset[depth - 1]; i < NIDS_PER_BLOCK; i++) {
+               child_nid = get_nid(pages[idx], i, false);
+               if (!child_nid)
+                       continue;
+               dn->nid = child_nid;
+               err = truncate_dnode(dn);
+               if (err < 0)
+                       goto fail;
+               set_nid(pages[idx], i, 0, false);
+       }
+
+       if (offset[depth - 1] == 0) {
+               dn->node_page = pages[idx];
+               dn->nid = nid[idx];
+               truncate_node(dn);
+       } else {
+               f2fs_put_page(pages[idx], 1);
+       }
+       offset[idx]++;
+       offset[depth - 1] = 0;
+fail:
+       for (i = depth - 3; i >= 0; i--)
+               f2fs_put_page(pages[i], 1);
+       return err;
+}
+
+/**
+ * All the block addresses of data and nodes should be nullified.
+ */
+int truncate_inode_blocks(struct inode *inode, pgoff_t from)
+{
+       struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
+       int err = 0, cont = 1;
+       int level, offset[4], noffset[4];
+       unsigned int nofs;
+       struct f2fs_node *rn;
+       struct dnode_of_data dn;
+       struct page *page;
+
+       level = get_node_path(from, offset, noffset);
+
+       page = get_node_page(sbi, inode->i_ino);
+       if (IS_ERR(page))
+               return PTR_ERR(page);
+
+       set_new_dnode(&dn, inode, page, NULL, 0);
+       unlock_page(page);
+
+       rn = page_address(page);
+       switch (level) {
+       case 0:
+       case 1:
+               nofs = noffset[1];
+               break;
+       case 2:
+               nofs = noffset[1];
+               if (!offset[level - 1])
+                       goto skip_partial;
+               err = truncate_partial_nodes(&dn, &rn->i, offset, level);
+               if (err < 0 && err != -ENOENT)
+                       goto fail;
+               nofs += 1 + NIDS_PER_BLOCK;
+               break;
+       case 3:
+               nofs = 5 + 2 * NIDS_PER_BLOCK;
+               if (!offset[level - 1])
+                       goto skip_partial;
+               err = truncate_partial_nodes(&dn, &rn->i, offset, level);
+               if (err < 0 && err != -ENOENT)
+                       goto fail;
+               break;
+       default:
+               BUG();
+       }
+
+skip_partial:
+       while (cont) {
+               dn.nid = le32_to_cpu(rn->i.i_nid[offset[0] - NODE_DIR1_BLOCK]);
+               switch (offset[0]) {
+               case NODE_DIR1_BLOCK:
+               case NODE_DIR2_BLOCK:
+                       err = truncate_dnode(&dn);
+                       break;
+
+               case NODE_IND1_BLOCK:
+               case NODE_IND2_BLOCK:
+                       err = truncate_nodes(&dn, nofs, offset[1], 2);
+                       break;
+
+               case NODE_DIND_BLOCK:
+                       err = truncate_nodes(&dn, nofs, offset[1], 3);
+                       cont = 0;
+                       break;
+
+               default:
+                       BUG();
+               }
+               if (err < 0 && err != -ENOENT)
+                       goto fail;
+               if (offset[1] == 0 &&
+                               rn->i.i_nid[offset[0] - NODE_DIR1_BLOCK]) {
+                       lock_page(page);
+                       wait_on_page_writeback(page);
+                       rn->i.i_nid[offset[0] - NODE_DIR1_BLOCK] = 0;
+                       set_page_dirty(page);
+                       unlock_page(page);
+               }
+               offset[1] = 0;
+               offset[0]++;
+               nofs += err;
+       }
+fail:
+       f2fs_put_page(page, 0);
+       return err > 0 ? 0 : err;
+}
+
+int remove_inode_page(struct inode *inode)
+{
+       struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
+       struct page *page;
+       nid_t ino = inode->i_ino;
+       struct dnode_of_data dn;
+
+       mutex_lock_op(sbi, NODE_TRUNC);
+       page = get_node_page(sbi, ino);
+       if (IS_ERR(page)) {
+               mutex_unlock_op(sbi, NODE_TRUNC);
+               return PTR_ERR(page);
+       }
+
+       if (F2FS_I(inode)->i_xattr_nid) {
+               nid_t nid = F2FS_I(inode)->i_xattr_nid;
+               struct page *npage = get_node_page(sbi, nid);
+
+               if (IS_ERR(npage)) {
+                       mutex_unlock_op(sbi, NODE_TRUNC);
+                       return PTR_ERR(npage);
+               }
+
+               F2FS_I(inode)->i_xattr_nid = 0;
+               set_new_dnode(&dn, inode, page, npage, nid);
+               dn.inode_page_locked = 1;
+               truncate_node(&dn);
+       }
+       if (inode->i_blocks == 1) {
+               /* inernally call f2fs_put_page() */
+               set_new_dnode(&dn, inode, page, page, ino);
+               truncate_node(&dn);
+       } else if (inode->i_blocks == 0) {
+               struct node_info ni;
+               get_node_info(sbi, inode->i_ino, &ni);
+
+               /* called after f2fs_new_inode() is failed */
+               BUG_ON(ni.blk_addr != NULL_ADDR);
+               f2fs_put_page(page, 1);
+       } else {
+               BUG();
+       }
+       mutex_unlock_op(sbi, NODE_TRUNC);
+       return 0;
+}
+
+int new_inode_page(struct inode *inode, struct dentry *dentry)
+{
+       struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
+       struct page *page;
+       struct dnode_of_data dn;
+
+       /* allocate inode page for new inode */
+       set_new_dnode(&dn, inode, NULL, NULL, inode->i_ino);
+       mutex_lock_op(sbi, NODE_NEW);
+       page = new_node_page(&dn, 0);
+       init_dent_inode(dentry, page);
+       mutex_unlock_op(sbi, NODE_NEW);
+       if (IS_ERR(page))
+               return PTR_ERR(page);
+       f2fs_put_page(page, 1);
+       return 0;
+}
+
+struct page *new_node_page(struct dnode_of_data *dn, unsigned int ofs)
+{
+       struct f2fs_sb_info *sbi = F2FS_SB(dn->inode->i_sb);
+       struct address_space *mapping = sbi->node_inode->i_mapping;
+       struct node_info old_ni, new_ni;
+       struct page *page;
+       int err;
+
+       if (is_inode_flag_set(F2FS_I(dn->inode), FI_NO_ALLOC))
+               return ERR_PTR(-EPERM);
+
+       page = grab_cache_page(mapping, dn->nid);
+       if (!page)
+               return ERR_PTR(-ENOMEM);
+
+       get_node_info(sbi, dn->nid, &old_ni);
+
+       SetPageUptodate(page);
+       fill_node_footer(page, dn->nid, dn->inode->i_ino, ofs, true);
+
+       /* Reinitialize old_ni with new node page */
+       BUG_ON(old_ni.blk_addr != NULL_ADDR);
+       new_ni = old_ni;
+       new_ni.ino = dn->inode->i_ino;
+
+       if (!inc_valid_node_count(sbi, dn->inode, 1)) {
+               err = -ENOSPC;
+               goto fail;
+       }
+       set_node_addr(sbi, &new_ni, NEW_ADDR);
+
+       dn->node_page = page;
+       sync_inode_page(dn);
+       set_page_dirty(page);
+       set_cold_node(dn->inode, page);
+       if (ofs == 0)
+               inc_valid_inode_count(sbi);
+
+       return page;
+
+fail:
+       f2fs_put_page(page, 1);
+       return ERR_PTR(err);
+}
+
+static int read_node_page(struct page *page, int type)
+{
+       struct f2fs_sb_info *sbi = F2FS_SB(page->mapping->host->i_sb);
+       struct node_info ni;
+
+       get_node_info(sbi, page->index, &ni);
+
+       if (ni.blk_addr == NULL_ADDR)
+               return -ENOENT;
+       return f2fs_readpage(sbi, page, ni.blk_addr, type);
+}
+
+/**
+ * Readahead a node page
+ */
+void ra_node_page(struct f2fs_sb_info *sbi, nid_t nid)
+{
+       struct address_space *mapping = sbi->node_inode->i_mapping;
+       struct page *apage;
+
+       apage = find_get_page(mapping, nid);
+       if (apage && PageUptodate(apage))
+               goto release_out;
+       f2fs_put_page(apage, 0);
+
+       apage = grab_cache_page(mapping, nid);
+       if (!apage)
+               return;
+
+       if (read_node_page(apage, READA))
+               goto unlock_out;
+
+       page_cache_release(apage);
+       return;
+
+unlock_out:
+       unlock_page(apage);
+release_out:
+       page_cache_release(apage);
+}
+
+struct page *get_node_page(struct f2fs_sb_info *sbi, pgoff_t nid)
+{
+       int err;
+       struct page *page;
+       struct address_space *mapping = sbi->node_inode->i_mapping;
+
+       page = grab_cache_page(mapping, nid);
+       if (!page)
+               return ERR_PTR(-ENOMEM);
+
+       err = read_node_page(page, READ_SYNC);
+       if (err) {
+               f2fs_put_page(page, 1);
+               return ERR_PTR(err);
+       }
+
+       BUG_ON(nid != nid_of_node(page));
+       mark_page_accessed(page);
+       return page;
+}
+
+/**
+ * Return a locked page for the desired node page.
+ * And, readahead MAX_RA_NODE number of node pages.
+ */
+struct page *get_node_page_ra(struct page *parent, int start)
+{
+       struct f2fs_sb_info *sbi = F2FS_SB(parent->mapping->host->i_sb);
+       struct address_space *mapping = sbi->node_inode->i_mapping;
+       int i, end;
+       int err = 0;
+       nid_t nid;
+       struct page *page;
+
+       /* First, try getting the desired direct node. */
+       nid = get_nid(parent, start, false);
+       if (!nid)
+               return ERR_PTR(-ENOENT);
+
+       page = find_get_page(mapping, nid);
+       if (page && PageUptodate(page))
+               goto page_hit;
+       f2fs_put_page(page, 0);
+
+repeat:
+       page = grab_cache_page(mapping, nid);
+       if (!page)
+               return ERR_PTR(-ENOMEM);
+
+       err = read_node_page(page, READA);
+       if (err) {
+               f2fs_put_page(page, 1);
+               return ERR_PTR(err);
+       }
+
+       /* Then, try readahead for siblings of the desired node */
+       end = start + MAX_RA_NODE;
+       end = min(end, NIDS_PER_BLOCK);
+       for (i = start + 1; i < end; i++) {
+               nid = get_nid(parent, i, false);
+               if (!nid)
+                       continue;
+               ra_node_page(sbi, nid);
+       }
+
+page_hit:
+       lock_page(page);
+       if (PageError(page)) {
+               f2fs_put_page(page, 1);
+               return ERR_PTR(-EIO);
+       }
+
+       /* Has the page been truncated? */
+       if (page->mapping != mapping) {
+               f2fs_put_page(page, 1);
+               goto repeat;
+       }
+       return page;
+}
+
+void sync_inode_page(struct dnode_of_data *dn)
+{
+       if (IS_INODE(dn->node_page) || dn->inode_page == dn->node_page) {
+               update_inode(dn->inode, dn->node_page);
+       } else if (dn->inode_page) {
+               if (!dn->inode_page_locked)
+                       lock_page(dn->inode_page);
+               update_inode(dn->inode, dn->inode_page);
+               if (!dn->inode_page_locked)
+                       unlock_page(dn->inode_page);
+       } else {
+               f2fs_write_inode(dn->inode, NULL);
+       }
+}
+
+int sync_node_pages(struct f2fs_sb_info *sbi, nid_t ino,
+                                       struct writeback_control *wbc)
+{
+       struct address_space *mapping = sbi->node_inode->i_mapping;
+       pgoff_t index, end;
+       struct pagevec pvec;
+       int step = ino ? 2 : 0;
+       int nwritten = 0, wrote = 0;
+
+       pagevec_init(&pvec, 0);
+
+next_step:
+       index = 0;
+       end = LONG_MAX;
+
+       while (index <= end) {
+               int i, nr_pages;
+               nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
+                               PAGECACHE_TAG_DIRTY,
+                               min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
+               if (nr_pages == 0)
+                       break;
+
+               for (i = 0; i < nr_pages; i++) {
+                       struct page *page = pvec.pages[i];
+
+                       /*
+                        * flushing sequence with step:
+                        * 0. indirect nodes
+                        * 1. dentry dnodes
+                        * 2. file dnodes
+                        */
+                       if (step == 0 && IS_DNODE(page))
+                               continue;
+                       if (step == 1 && (!IS_DNODE(page) ||
+                                               is_cold_node(page)))
+                               continue;
+                       if (step == 2 && (!IS_DNODE(page) ||
+                                               !is_cold_node(page)))
+                               continue;
+
+                       /*
+                        * If an fsync mode,
+                        * we should not skip writing node pages.
+                        */
+                       if (ino && ino_of_node(page) == ino)
+                               lock_page(page);
+                       else if (!trylock_page(page))
+                               continue;
+
+                       if (unlikely(page->mapping != mapping)) {
+continue_unlock:
+                               unlock_page(page);
+                               continue;
+                       }
+                       if (ino && ino_of_node(page) != ino)
+                               goto continue_unlock;
+
+                       if (!PageDirty(page)) {
+                               /* someone wrote it for us */
+                               goto continue_unlock;
+                       }
+
+                       if (!clear_page_dirty_for_io(page))
+                               goto continue_unlock;
+
+                       /* called by fsync() */
+                       if (ino && IS_DNODE(page)) {
+                               int mark = !is_checkpointed_node(sbi, ino);
+                               set_fsync_mark(page, 1);
+                               if (IS_INODE(page))
+                                       set_dentry_mark(page, mark);
+                               nwritten++;
+                       } else {
+                               set_fsync_mark(page, 0);
+                               set_dentry_mark(page, 0);
+                       }
+                       mapping->a_ops->writepage(page, wbc);
+                       wrote++;
+
+                       if (--wbc->nr_to_write == 0)
+                               break;
+               }
+               pagevec_release(&pvec);
+               cond_resched();
+
+               if (wbc->nr_to_write == 0) {
+                       step = 2;
+                       break;
+               }
+       }
+
+       if (step < 2) {
+               step++;
+               goto next_step;
+       }
+
+       if (wrote)
+               f2fs_submit_bio(sbi, NODE, wbc->sync_mode == WB_SYNC_ALL);
+
+       return nwritten;
+}
+
+static int f2fs_write_node_page(struct page *page,
+                               struct writeback_control *wbc)
+{
+       struct f2fs_sb_info *sbi = F2FS_SB(page->mapping->host->i_sb);
+       nid_t nid;
+       unsigned int nofs;
+       block_t new_addr;
+       struct node_info ni;
+
+       if (wbc->for_reclaim) {
+               dec_page_count(sbi, F2FS_DIRTY_NODES);
+               wbc->pages_skipped++;
+               set_page_dirty(page);
+               return AOP_WRITEPAGE_ACTIVATE;
+       }
+
+       wait_on_page_writeback(page);
+
+       mutex_lock_op(sbi, NODE_WRITE);
+
+       /* get old block addr of this node page */
+       nid = nid_of_node(page);
+       nofs = ofs_of_node(page);
+       BUG_ON(page->index != nid);
+
+       get_node_info(sbi, nid, &ni);
+
+       /* This page is already truncated */
+       if (ni.blk_addr == NULL_ADDR)
+               return 0;
+
+       set_page_writeback(page);
+
+       /* insert node offset */
+       write_node_page(sbi, page, nid, ni.blk_addr, &new_addr);
+       set_node_addr(sbi, &ni, new_addr);
+       dec_page_count(sbi, F2FS_DIRTY_NODES);
+
+       mutex_unlock_op(sbi, NODE_WRITE);
+       unlock_page(page);
+       return 0;
+}
+
+static int f2fs_write_node_pages(struct address_space *mapping,
+                           struct writeback_control *wbc)
+{
+       struct f2fs_sb_info *sbi = F2FS_SB(mapping->host->i_sb);
+       struct block_device *bdev = sbi->sb->s_bdev;
+       long nr_to_write = wbc->nr_to_write;
+
+       if (wbc->for_kupdate)
+               return 0;
+
+       if (get_pages(sbi, F2FS_DIRTY_NODES) == 0)
+               return 0;
+
+       if (try_to_free_nats(sbi, NAT_ENTRY_PER_BLOCK)) {
+               write_checkpoint(sbi, false, false);
+               return 0;
+       }
+
+       /* if mounting is failed, skip writing node pages */
+       wbc->nr_to_write = bio_get_nr_vecs(bdev);
+       sync_node_pages(sbi, 0, wbc);
+       wbc->nr_to_write = nr_to_write -
+               (bio_get_nr_vecs(bdev) - wbc->nr_to_write);
+       return 0;
+}
+
+static int f2fs_set_node_page_dirty(struct page *page)
+{
+       struct address_space *mapping = page->mapping;
+       struct f2fs_sb_info *sbi = F2FS_SB(mapping->host->i_sb);
+
+       SetPageUptodate(page);
+       if (!PageDirty(page)) {
+               __set_page_dirty_nobuffers(page);
+               inc_page_count(sbi, F2FS_DIRTY_NODES);
+               SetPagePrivate(page);
+               return 1;
+       }
+       return 0;
+}
+
+static void f2fs_invalidate_node_page(struct page *page, unsigned long offset)
+{
+       struct inode *inode = page->mapping->host;
+       struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
+       if (PageDirty(page))
+               dec_page_count(sbi, F2FS_DIRTY_NODES);
+       ClearPagePrivate(page);
+}
+
+static int f2fs_release_node_page(struct page *page, gfp_t wait)
+{
+       ClearPagePrivate(page);
+       return 0;
+}
+
+/**
+ * Structure of the f2fs node operations
+ */
+const struct address_space_operations f2fs_node_aops = {
+       .writepage      = f2fs_write_node_page,
+       .writepages     = f2fs_write_node_pages,
+       .set_page_dirty = f2fs_set_node_page_dirty,
+       .invalidatepage = f2fs_invalidate_node_page,
+       .releasepage    = f2fs_release_node_page,
+};
+
+static struct free_nid *__lookup_free_nid_list(nid_t n, struct list_head *head)
+{
+       struct list_head *this;
+       struct free_nid *i = NULL;
+       list_for_each(this, head) {
+               i = list_entry(this, struct free_nid, list);
+               if (i->nid == n)
+                       break;
+               i = NULL;
+       }
+       return i;
+}
+
+static void __del_from_free_nid_list(struct free_nid *i)
+{
+       list_del(&i->list);
+       kmem_cache_free(free_nid_slab, i);
+}
+
+static int add_free_nid(struct f2fs_nm_info *nm_i, nid_t nid)
+{
+       struct free_nid *i;
+
+       if (nm_i->fcnt > 2 * MAX_FREE_NIDS)
+               return 0;
+retry:
+       i = kmem_cache_alloc(free_nid_slab, GFP_NOFS);
+       if (!i) {
+               cond_resched();
+               goto retry;
+       }
+       i->nid = nid;
+       i->state = NID_NEW;
+
+       spin_lock(&nm_i->free_nid_list_lock);
+       if (__lookup_free_nid_list(nid, &nm_i->free_nid_list)) {
+               spin_unlock(&nm_i->free_nid_list_lock);
+               kmem_cache_free(free_nid_slab, i);
+               return 0;
+       }
+       list_add_tail(&i->list, &nm_i->free_nid_list);
+       nm_i->fcnt++;
+       spin_unlock(&nm_i->free_nid_list_lock);
+       return 1;
+}
+
+static void remove_free_nid(struct f2fs_nm_info *nm_i, nid_t nid)
+{
+       struct free_nid *i;
+       spin_lock(&nm_i->free_nid_list_lock);
+       i = __lookup_free_nid_list(nid, &nm_i->free_nid_list);
+       if (i && i->state == NID_NEW) {
+               __del_from_free_nid_list(i);
+               nm_i->fcnt--;
+       }
+       spin_unlock(&nm_i->free_nid_list_lock);
+}
+
+static int scan_nat_page(struct f2fs_nm_info *nm_i,
+                       struct page *nat_page, nid_t start_nid)
+{
+       struct f2fs_nat_block *nat_blk = page_address(nat_page);
+       block_t blk_addr;
+       int fcnt = 0;
+       int i;
+
+       /* 0 nid should not be used */
+       if (start_nid == 0)
+               ++start_nid;
+
+       i = start_nid % NAT_ENTRY_PER_BLOCK;
+
+       for (; i < NAT_ENTRY_PER_BLOCK; i++, start_nid++) {
+               blk_addr  = le32_to_cpu(nat_blk->entries[i].block_addr);
+               BUG_ON(blk_addr == NEW_ADDR);
+               if (blk_addr == NULL_ADDR)
+                       fcnt += add_free_nid(nm_i, start_nid);
+       }
+       return fcnt;
+}
+
+static void build_free_nids(struct f2fs_sb_info *sbi)
+{
+       struct free_nid *fnid, *next_fnid;
+       struct f2fs_nm_info *nm_i = NM_I(sbi);
+       struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
+       struct f2fs_summary_block *sum = curseg->sum_blk;
+       nid_t nid = 0;
+       bool is_cycled = false;
+       int fcnt = 0;
+       int i;
+
+       nid = nm_i->next_scan_nid;
+       nm_i->init_scan_nid = nid;
+
+       ra_nat_pages(sbi, nid);
+
+       while (1) {
+               struct page *page = get_current_nat_page(sbi, nid);
+
+               fcnt += scan_nat_page(nm_i, page, nid);
+               f2fs_put_page(page, 1);
+
+               nid += (NAT_ENTRY_PER_BLOCK - (nid % NAT_ENTRY_PER_BLOCK));
+
+               if (nid >= nm_i->max_nid) {
+                       nid = 0;
+                       is_cycled = true;
+               }
+               if (fcnt > MAX_FREE_NIDS)
+                       break;
+               if (is_cycled && nm_i->init_scan_nid <= nid)
+                       break;
+       }
+
+       nm_i->next_scan_nid = nid;
+
+       /* find free nids from current sum_pages */
+       mutex_lock(&curseg->curseg_mutex);
+       for (i = 0; i < nats_in_cursum(sum); i++) {
+               block_t addr = le32_to_cpu(nat_in_journal(sum, i).block_addr);
+               nid = le32_to_cpu(nid_in_journal(sum, i));
+               if (addr == NULL_ADDR)
+                       add_free_nid(nm_i, nid);
+               else
+                       remove_free_nid(nm_i, nid);
+       }
+       mutex_unlock(&curseg->curseg_mutex);
+
+       /* remove the free nids from current allocated nids */
+       list_for_each_entry_safe(fnid, next_fnid, &nm_i->free_nid_list, list) {
+               struct nat_entry *ne;
+
+               read_lock(&nm_i->nat_tree_lock);
+               ne = __lookup_nat_cache(nm_i, fnid->nid);
+               if (ne && nat_get_blkaddr(ne) != NULL_ADDR)
+                       remove_free_nid(nm_i, fnid->nid);
+               read_unlock(&nm_i->nat_tree_lock);
+       }
+}
+
+/*
+ * If this function returns success, caller can obtain a new nid
+ * from second parameter of this function.
+ * The returned nid could be used ino as well as nid when inode is created.
+ */
+bool alloc_nid(struct f2fs_sb_info *sbi, nid_t *nid)
+{
+       struct f2fs_nm_info *nm_i = NM_I(sbi);
+       struct free_nid *i = NULL;
+       struct list_head *this;
+retry:
+       mutex_lock(&nm_i->build_lock);
+       if (!nm_i->fcnt) {
+               /* scan NAT in order to build free nid list */
+               build_free_nids(sbi);
+               if (!nm_i->fcnt) {
+                       mutex_unlock(&nm_i->build_lock);
+                       return false;
+               }
+       }
+       mutex_unlock(&nm_i->build_lock);
+
+       /*
+        * We check fcnt again since previous check is racy as
+        * we didn't hold free_nid_list_lock. So other thread
+        * could consume all of free nids.
+        */
+       spin_lock(&nm_i->free_nid_list_lock);
+       if (!nm_i->fcnt) {
+               spin_unlock(&nm_i->free_nid_list_lock);
+               goto retry;
+       }
+
+       BUG_ON(list_empty(&nm_i->free_nid_list));
+       list_for_each(this, &nm_i->free_nid_list) {
+               i = list_entry(this, struct free_nid, list);
+               if (i->state == NID_NEW)
+                       break;
+       }
+
+       BUG_ON(i->state != NID_NEW);
+       *nid = i->nid;
+       i->state = NID_ALLOC;
+       nm_i->fcnt--;
+       spin_unlock(&nm_i->free_nid_list_lock);
+       return true;
+}
+
+/**
+ * alloc_nid() should be called prior to this function.
+ */
+void alloc_nid_done(struct f2fs_sb_info *sbi, nid_t nid)
+{
+       struct f2fs_nm_info *nm_i = NM_I(sbi);
+       struct free_nid *i;
+
+       spin_lock(&nm_i->free_nid_list_lock);
+       i = __lookup_free_nid_list(nid, &nm_i->free_nid_list);
+       if (i) {
+               BUG_ON(i->state != NID_ALLOC);
+               __del_from_free_nid_list(i);
+       }
+       spin_unlock(&nm_i->free_nid_list_lock);
+}
+
+/**
+ * alloc_nid() should be called prior to this function.
+ */
+void alloc_nid_failed(struct f2fs_sb_info *sbi, nid_t nid)
+{
+       alloc_nid_done(sbi, nid);
+       add_free_nid(NM_I(sbi), nid);
+}
+
+void recover_node_page(struct f2fs_sb_info *sbi, struct page *page,
+               struct f2fs_summary *sum, struct node_info *ni,
+               block_t new_blkaddr)
+{
+       rewrite_node_page(sbi, page, sum, ni->blk_addr, new_blkaddr);
+       set_node_addr(sbi, ni, new_blkaddr);
+       clear_node_page_dirty(page);
+}
+
+int recover_inode_page(struct f2fs_sb_info *sbi, struct page *page)
+{
+       struct address_space *mapping = sbi->node_inode->i_mapping;
+       struct f2fs_node *src, *dst;
+       nid_t ino = ino_of_node(page);
+       struct node_info old_ni, new_ni;
+       struct page *ipage;
+
+       ipage = grab_cache_page(mapping, ino);
+       if (!ipage)
+               return -ENOMEM;
+
+       /* Should not use this inode  from free nid list */
+       remove_free_nid(NM_I(sbi), ino);
+
+       get_node_info(sbi, ino, &old_ni);
+       SetPageUptodate(ipage);
+       fill_node_footer(ipage, ino, ino, 0, true);
+
+       src = (struct f2fs_node *)page_address(page);
+       dst = (struct f2fs_node *)page_address(ipage);
+
+       memcpy(dst, src, (unsigned long)&src->i.i_ext - (unsigned long)&src->i);
+       dst->i.i_size = 0;
+       dst->i.i_blocks = 1;
+       dst->i.i_links = 1;
+       dst->i.i_xattr_nid = 0;
+
+       new_ni = old_ni;
+       new_ni.ino = ino;
+
+       set_node_addr(sbi, &new_ni, NEW_ADDR);
+       inc_valid_inode_count(sbi);
+
+       f2fs_put_page(ipage, 1);
+       return 0;
+}
+
+int restore_node_summary(struct f2fs_sb_info *sbi,
+                       unsigned int segno, struct f2fs_summary_block *sum)
+{
+       struct f2fs_node *rn;
+       struct f2fs_summary *sum_entry;
+       struct page *page;
+       block_t addr;
+       int i, last_offset;
+
+       /* alloc temporal page for read node */
+       page = alloc_page(GFP_NOFS | __GFP_ZERO);
+       if (IS_ERR(page))
+               return PTR_ERR(page);
+       lock_page(page);
+
+       /* scan the node segment */
+       last_offset = sbi->blocks_per_seg;
+       addr = START_BLOCK(sbi, segno);
+       sum_entry = &sum->entries[0];
+
+       for (i = 0; i < last_offset; i++, sum_entry++) {
+               if (f2fs_readpage(sbi, page, addr, READ_SYNC))
+                       goto out;
+
+               rn = (struct f2fs_node *)page_address(page);
+               sum_entry->nid = rn->footer.nid;
+               sum_entry->version = 0;
+               sum_entry->ofs_in_node = 0;
+               addr++;
+
+               /*
+                * In order to read next node page,
+                * we must clear PageUptodate flag.
+                */
+               ClearPageUptodate(page);
+       }
+out:
+       unlock_page(page);
+       __free_pages(page, 0);
+       return 0;
+}
+
+static bool flush_nats_in_journal(struct f2fs_sb_info *sbi)
+{
+       struct f2fs_nm_info *nm_i = NM_I(sbi);
+       struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
+       struct f2fs_summary_block *sum = curseg->sum_blk;
+       int i;
+
+       mutex_lock(&curseg->curseg_mutex);
+
+       if (nats_in_cursum(sum) < NAT_JOURNAL_ENTRIES) {
+               mutex_unlock(&curseg->curseg_mutex);
+               return false;
+       }
+
+       for (i = 0; i < nats_in_cursum(sum); i++) {
+               struct nat_entry *ne;
+               struct f2fs_nat_entry raw_ne;
+               nid_t nid = le32_to_cpu(nid_in_journal(sum, i));
+
+               raw_ne = nat_in_journal(sum, i);
+retry:
+               write_lock(&nm_i->nat_tree_lock);
+               ne = __lookup_nat_cache(nm_i, nid);
+               if (ne) {
+                       __set_nat_cache_dirty(nm_i, ne);
+                       write_unlock(&nm_i->nat_tree_lock);
+                       continue;
+               }
+               ne = grab_nat_entry(nm_i, nid);
+               if (!ne) {
+                       write_unlock(&nm_i->nat_tree_lock);
+                       goto retry;
+               }
+               nat_set_blkaddr(ne, le32_to_cpu(raw_ne.block_addr));
+               nat_set_ino(ne, le32_to_cpu(raw_ne.ino));
+               nat_set_version(ne, raw_ne.version);
+               __set_nat_cache_dirty(nm_i, ne);
+               write_unlock(&nm_i->nat_tree_lock);
+       }
+       update_nats_in_cursum(sum, -i);
+       mutex_unlock(&curseg->curseg_mutex);
+       return true;
+}
+
+/**
+ * This function is called during the checkpointing process.
+ */
+void flush_nat_entries(struct f2fs_sb_info *sbi)
+{
+       struct f2fs_nm_info *nm_i = NM_I(sbi);
+       struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
+       struct f2fs_summary_block *sum = curseg->sum_blk;
+       struct list_head *cur, *n;
+       struct page *page = NULL;
+       struct f2fs_nat_block *nat_blk = NULL;
+       nid_t start_nid = 0, end_nid = 0;
+       bool flushed;
+
+       flushed = flush_nats_in_journal(sbi);
+
+       if (!flushed)
+               mutex_lock(&curseg->curseg_mutex);
+
+       /* 1) flush dirty nat caches */
+       list_for_each_safe(cur, n, &nm_i->dirty_nat_entries) {
+               struct nat_entry *ne;
+               nid_t nid;
+               struct f2fs_nat_entry raw_ne;
+               int offset = -1;
+               block_t old_blkaddr, new_blkaddr;
+
+               ne = list_entry(cur, struct nat_entry, list);
+               nid = nat_get_nid(ne);
+
+               if (nat_get_blkaddr(ne) == NEW_ADDR)
+                       continue;
+               if (flushed)
+                       goto to_nat_page;
+
+               /* if there is room for nat enries in curseg->sumpage */
+               offset = lookup_journal_in_cursum(sum, NAT_JOURNAL, nid, 1);
+               if (offset >= 0) {
+                       raw_ne = nat_in_journal(sum, offset);
+                       old_blkaddr = le32_to_cpu(raw_ne.block_addr);
+                       goto flush_now;
+               }
+to_nat_page:
+               if (!page || (start_nid > nid || nid > end_nid)) {
+                       if (page) {
+                               f2fs_put_page(page, 1);
+                               page = NULL;
+                       }
+                       start_nid = START_NID(nid);
+                       end_nid = start_nid + NAT_ENTRY_PER_BLOCK - 1;
+
+                       /*
+                        * get nat block with dirty flag, increased reference
+                        * count, mapped and lock
+                        */
+                       page = get_next_nat_page(sbi, start_nid);
+                       nat_blk = page_address(page);
+               }
+
+               BUG_ON(!nat_blk);
+               raw_ne = nat_blk->entries[nid - start_nid];
+               old_blkaddr = le32_to_cpu(raw_ne.block_addr);
+flush_now:
+               new_blkaddr = nat_get_blkaddr(ne);
+
+               raw_ne.ino = cpu_to_le32(nat_get_ino(ne));
+               raw_ne.block_addr = cpu_to_le32(new_blkaddr);
+               raw_ne.version = nat_get_version(ne);
+
+               if (offset < 0) {
+                       nat_blk->entries[nid - start_nid] = raw_ne;
+               } else {
+                       nat_in_journal(sum, offset) = raw_ne;
+                       nid_in_journal(sum, offset) = cpu_to_le32(nid);
+               }
+
+               if (nat_get_blkaddr(ne) == NULL_ADDR) {
+                       write_lock(&nm_i->nat_tree_lock);
+                       __del_from_nat_cache(nm_i, ne);
+                       write_unlock(&nm_i->nat_tree_lock);
+
+                       /* We can reuse this freed nid at this point */
+                       add_free_nid(NM_I(sbi), nid);
+               } else {
+                       write_lock(&nm_i->nat_tree_lock);
+                       __clear_nat_cache_dirty(nm_i, ne);
+                       ne->checkpointed = true;
+                       write_unlock(&nm_i->nat_tree_lock);
+               }
+       }
+       if (!flushed)
+               mutex_unlock(&curseg->curseg_mutex);
+       f2fs_put_page(page, 1);
+
+       /* 2) shrink nat caches if necessary */
+       try_to_free_nats(sbi, nm_i->nat_cnt - NM_WOUT_THRESHOLD);
+}
+
+static int init_node_manager(struct f2fs_sb_info *sbi)
+{
+       struct f2fs_super_block *sb_raw = F2FS_RAW_SUPER(sbi);
+       struct f2fs_nm_info *nm_i = NM_I(sbi);
+       unsigned char *version_bitmap;
+       unsigned int nat_segs, nat_blocks;
+
+       nm_i->nat_blkaddr = le32_to_cpu(sb_raw->nat_blkaddr);
+
+       /* segment_count_nat includes pair segment so divide to 2. */
+       nat_segs = le32_to_cpu(sb_raw->segment_count_nat) >> 1;
+       nat_blocks = nat_segs << le32_to_cpu(sb_raw->log_blocks_per_seg);
+       nm_i->max_nid = NAT_ENTRY_PER_BLOCK * nat_blocks;
+       nm_i->fcnt = 0;
+       nm_i->nat_cnt = 0;
+
+       INIT_LIST_HEAD(&nm_i->free_nid_list);
+       INIT_RADIX_TREE(&nm_i->nat_root, GFP_ATOMIC);
+       INIT_LIST_HEAD(&nm_i->nat_entries);
+       INIT_LIST_HEAD(&nm_i->dirty_nat_entries);
+
+       mutex_init(&nm_i->build_lock);
+       spin_lock_init(&nm_i->free_nid_list_lock);
+       rwlock_init(&nm_i->nat_tree_lock);
+
+       nm_i->bitmap_size = __bitmap_size(sbi, NAT_BITMAP);
+       nm_i->init_scan_nid = le32_to_cpu(sbi->ckpt->next_free_nid);
+       nm_i->next_scan_nid = le32_to_cpu(sbi->ckpt->next_free_nid);
+
+       nm_i->nat_bitmap = kzalloc(nm_i->bitmap_size, GFP_KERNEL);
+       if (!nm_i->nat_bitmap)
+               return -ENOMEM;
+       version_bitmap = __bitmap_ptr(sbi, NAT_BITMAP);
+       if (!version_bitmap)
+               return -EFAULT;
+
+       /* copy version bitmap */
+       memcpy(nm_i->nat_bitmap, version_bitmap, nm_i->bitmap_size);
+       return 0;
+}
+
+int build_node_manager(struct f2fs_sb_info *sbi)
+{
+       int err;
+
+       sbi->nm_info = kzalloc(sizeof(struct f2fs_nm_info), GFP_KERNEL);
+       if (!sbi->nm_info)
+               return -ENOMEM;
+
+       err = init_node_manager(sbi);
+       if (err)
+               return err;
+
+       build_free_nids(sbi);
+       return 0;
+}
+
+void destroy_node_manager(struct f2fs_sb_info *sbi)
+{
+       struct f2fs_nm_info *nm_i = NM_I(sbi);
+       struct free_nid *i, *next_i;
+       struct nat_entry *natvec[NATVEC_SIZE];
+       nid_t nid = 0;
+       unsigned int found;
+
+       if (!nm_i)
+               return;
+
+       /* destroy free nid list */
+       spin_lock(&nm_i->free_nid_list_lock);
+       list_for_each_entry_safe(i, next_i, &nm_i->free_nid_list, list) {
+               BUG_ON(i->state == NID_ALLOC);
+               __del_from_free_nid_list(i);
+               nm_i->fcnt--;
+       }
+       BUG_ON(nm_i->fcnt);
+       spin_unlock(&nm_i->free_nid_list_lock);
+
+       /* destroy nat cache */
+       write_lock(&nm_i->nat_tree_lock);
+       while ((found = __gang_lookup_nat_cache(nm_i,
+                                       nid, NATVEC_SIZE, natvec))) {
+               unsigned idx;
+               for (idx = 0; idx < found; idx++) {
+                       struct nat_entry *e = natvec[idx];
+                       nid = nat_get_nid(e) + 1;
+                       __del_from_nat_cache(nm_i, e);
+               }
+       }
+       BUG_ON(nm_i->nat_cnt);
+       write_unlock(&nm_i->nat_tree_lock);
+
+       kfree(nm_i->nat_bitmap);
+       sbi->nm_info = NULL;
+       kfree(nm_i);
+}
+
+int create_node_manager_caches(void)
+{
+       nat_entry_slab = f2fs_kmem_cache_create("nat_entry",
+                       sizeof(struct nat_entry), NULL);
+       if (!nat_entry_slab)
+               return -ENOMEM;
+
+       free_nid_slab = f2fs_kmem_cache_create("free_nid",
+                       sizeof(struct free_nid), NULL);
+       if (!free_nid_slab) {
+               kmem_cache_destroy(nat_entry_slab);
+               return -ENOMEM;
+       }
+       return 0;
+}
+
+void destroy_node_manager_caches(void)
+{
+       kmem_cache_destroy(free_nid_slab);
+       kmem_cache_destroy(nat_entry_slab);
+}