nilfs2: inode operations
authorRyusuke Konishi <konishi.ryusuke@lab.ntt.co.jp>
Tue, 7 Apr 2009 02:01:32 +0000 (19:01 -0700)
committerLinus Torvalds <torvalds@linux-foundation.org>
Tue, 7 Apr 2009 15:31:14 +0000 (08:31 -0700)
This adds inode level operations of the nilfs2 file system.

Signed-off-by: Ryusuke Konishi <konishi.ryusuke@lab.ntt.co.jp>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
fs/nilfs2/inode.c [new file with mode: 0644]

diff --git a/fs/nilfs2/inode.c b/fs/nilfs2/inode.c
new file mode 100644 (file)
index 0000000..b4697d9
--- /dev/null
@@ -0,0 +1,819 @@
+/*
+ * inode.c - NILFS inode operations.
+ *
+ * Copyright (C) 2005-2008 Nippon Telegraph and Telephone Corporation.
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License as published by
+ * the Free Software Foundation; either version 2 of the License, or
+ * (at your option) any later version.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+ * GNU General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with this program; if not, write to the Free Software
+ * Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
+ *
+ * Written by Ryusuke Konishi <ryusuke@osrg.net>
+ *
+ */
+
+#include <linux/buffer_head.h>
+#include <linux/mpage.h>
+#include <linux/writeback.h>
+#include "nilfs.h"
+#include "segment.h"
+#include "page.h"
+#include "mdt.h"
+#include "cpfile.h"
+#include "ifile.h"
+
+
+/**
+ * nilfs_get_block() - get a file block on the filesystem (callback function)
+ * @inode - inode struct of the target file
+ * @blkoff - file block number
+ * @bh_result - buffer head to be mapped on
+ * @create - indicate whether allocating the block or not when it has not
+ *      been allocated yet.
+ *
+ * This function does not issue actual read request of the specified data
+ * block. It is done by VFS.
+ * Bulk read for direct-io is not supported yet. (should be supported)
+ */
+int nilfs_get_block(struct inode *inode, sector_t blkoff,
+                   struct buffer_head *bh_result, int create)
+{
+       struct nilfs_inode_info *ii = NILFS_I(inode);
+       unsigned long blknum = 0;
+       int err = 0, ret;
+       struct inode *dat = nilfs_dat_inode(NILFS_I_NILFS(inode));
+
+       /* This exclusion control is a workaround; should be revised */
+       down_read(&NILFS_MDT(dat)->mi_sem);     /* XXX */
+       ret = nilfs_bmap_lookup(ii->i_bmap, (unsigned long)blkoff, &blknum);
+       up_read(&NILFS_MDT(dat)->mi_sem);       /* XXX */
+       if (ret == 0) { /* found */
+               map_bh(bh_result, inode->i_sb, blknum);
+               goto out;
+       }
+       if (unlikely(ret == 1)) {
+               printk(KERN_ERR "nilfs_get_block: bmap_lookup returns "
+                      "buffer_head pointer (blkoff=%llu, blknum=%lu)\n",
+                      (unsigned long long)blkoff, blknum);
+               BUG();
+       }
+       /* data block was not found */
+       if (ret == -ENOENT && create) {
+               struct nilfs_transaction_info ti;
+
+               bh_result->b_blocknr = 0;
+               err = nilfs_transaction_begin(inode->i_sb, &ti, 1);
+               if (unlikely(err))
+                       goto out;
+               err = nilfs_bmap_insert(ii->i_bmap, (unsigned long)blkoff,
+                                       (unsigned long)bh_result);
+               nilfs_transaction_end(inode->i_sb, !err);
+               if (unlikely(err != 0)) {
+                       if (err == -EEXIST) {
+                               /*
+                                * The get_block() function could be called
+                                * from multiple callers for an inode.
+                                * However, the page having this block must
+                                * be locked in this case.
+                                */
+                               printk(KERN_ERR
+                                      "nilfs_get_block: a race condition "
+                                      "while inserting a data block. "
+                                      "(inode number=%lu, file block "
+                                      "offset=%llu)\n",
+                                      inode->i_ino,
+                                      (unsigned long long)blkoff);
+                               BUG();
+                       } else if (err == -EINVAL) {
+                               nilfs_error(inode->i_sb, __func__,
+                                           "broken bmap (inode=%lu)\n",
+                                           inode->i_ino);
+                               err = -EIO;
+                       }
+                       goto out;
+               }
+               /* Error handling should be detailed */
+               set_buffer_new(bh_result);
+               map_bh(bh_result, inode->i_sb, 0); /* dbn must be changed
+                                                     to proper value */
+       } else if (ret == -ENOENT) {
+               /* not found is not error (e.g. hole); must return without
+                  the mapped state flag. */
+               ;
+       } else {
+               err = ret;
+       }
+
+ out:
+       return err;
+}
+
+/**
+ * nilfs_readpage() - implement readpage() method of nilfs_aops {}
+ * address_space_operations.
+ * @file - file struct of the file to be read
+ * @page - the page to be read
+ */
+static int nilfs_readpage(struct file *file, struct page *page)
+{
+       return mpage_readpage(page, nilfs_get_block);
+}
+
+/**
+ * nilfs_readpages() - implement readpages() method of nilfs_aops {}
+ * address_space_operations.
+ * @file - file struct of the file to be read
+ * @mapping - address_space struct used for reading multiple pages
+ * @pages - the pages to be read
+ * @nr_pages - number of pages to be read
+ */
+static int nilfs_readpages(struct file *file, struct address_space *mapping,
+                          struct list_head *pages, unsigned nr_pages)
+{
+       return mpage_readpages(mapping, pages, nr_pages, nilfs_get_block);
+}
+
+static int nilfs_writepages(struct address_space *mapping,
+                           struct writeback_control *wbc)
+{
+       /* This empty method is required not to call generic_writepages() */
+       return 0;
+}
+
+static int nilfs_writepage(struct page *page, struct writeback_control *wbc)
+{
+       struct inode *inode = page->mapping->host;
+       int err;
+
+       redirty_page_for_writepage(wbc, page);
+       unlock_page(page);
+
+       if (wbc->sync_mode == WB_SYNC_ALL) {
+               err = nilfs_construct_segment(inode->i_sb);
+               if (unlikely(err))
+                       return err;
+       } else if (wbc->for_reclaim)
+               nilfs_flush_segment(inode->i_sb, inode->i_ino);
+
+       return 0;
+}
+
+static int nilfs_set_page_dirty(struct page *page)
+{
+       int ret = __set_page_dirty_buffers(page);
+
+       if (ret) {
+               struct inode *inode = page->mapping->host;
+               struct nilfs_sb_info *sbi = NILFS_SB(inode->i_sb);
+               unsigned nr_dirty = 1 << (PAGE_SHIFT - inode->i_blkbits);
+
+               nilfs_set_file_dirty(sbi, inode, nr_dirty);
+       }
+       return ret;
+}
+
+static int nilfs_write_begin(struct file *file, struct address_space *mapping,
+                            loff_t pos, unsigned len, unsigned flags,
+                            struct page **pagep, void **fsdata)
+
+{
+       struct inode *inode = mapping->host;
+       int err = nilfs_transaction_begin(inode->i_sb, NULL, 1);
+
+       if (unlikely(err))
+               return err;
+
+       *pagep = NULL;
+       err = block_write_begin(file, mapping, pos, len, flags, pagep,
+                               fsdata, nilfs_get_block);
+       if (unlikely(err))
+               nilfs_transaction_end(inode->i_sb, 0);
+       return err;
+}
+
+static int nilfs_write_end(struct file *file, struct address_space *mapping,
+                          loff_t pos, unsigned len, unsigned copied,
+                          struct page *page, void *fsdata)
+{
+       struct inode *inode = mapping->host;
+       unsigned start = pos & (PAGE_CACHE_SIZE - 1);
+       unsigned nr_dirty;
+       int err;
+
+       nr_dirty = nilfs_page_count_clean_buffers(page, start,
+                                                 start + copied);
+       copied = generic_write_end(file, mapping, pos, len, copied, page,
+                                  fsdata);
+       nilfs_set_file_dirty(NILFS_SB(inode->i_sb), inode, nr_dirty);
+       err = nilfs_transaction_end(inode->i_sb, 1);
+       return err ? : copied;
+}
+
+static ssize_t
+nilfs_direct_IO(int rw, struct kiocb *iocb, const struct iovec *iov,
+               loff_t offset, unsigned long nr_segs)
+{
+       struct file *file = iocb->ki_filp;
+       struct inode *inode = file->f_mapping->host;
+       ssize_t size;
+       int err;
+
+       err = nilfs_construct_dsync_segment(inode->i_sb, inode);
+       if (unlikely(err))
+               return err;
+
+       if (rw == WRITE)
+               return 0;
+
+       /* Needs synchronization with the cleaner */
+       size = blockdev_direct_IO(rw, iocb, inode, inode->i_sb->s_bdev, iov,
+                                 offset, nr_segs, nilfs_get_block, NULL);
+       return size;
+}
+
+struct address_space_operations nilfs_aops = {
+       .writepage              = nilfs_writepage,
+       .readpage               = nilfs_readpage,
+       /* .sync_page           = nilfs_sync_page, */
+       .writepages             = nilfs_writepages,
+       .set_page_dirty         = nilfs_set_page_dirty,
+       .readpages              = nilfs_readpages,
+       .write_begin            = nilfs_write_begin,
+       .write_end              = nilfs_write_end,
+       /* .releasepage         = nilfs_releasepage, */
+       .invalidatepage         = block_invalidatepage,
+       .direct_IO              = nilfs_direct_IO,
+};
+
+struct inode *nilfs_new_inode(struct inode *dir, int mode)
+{
+       struct super_block *sb = dir->i_sb;
+       struct nilfs_sb_info *sbi = NILFS_SB(sb);
+       struct inode *inode;
+       struct nilfs_inode_info *ii;
+       int err = -ENOMEM;
+       ino_t ino;
+
+       inode = new_inode(sb);
+       if (unlikely(!inode))
+               goto failed;
+
+       mapping_set_gfp_mask(inode->i_mapping,
+                            mapping_gfp_mask(inode->i_mapping) & ~__GFP_FS);
+
+       ii = NILFS_I(inode);
+       ii->i_state = 1 << NILFS_I_NEW;
+
+       err = nilfs_ifile_create_inode(sbi->s_ifile, &ino, &ii->i_bh);
+       if (unlikely(err))
+               goto failed_ifile_create_inode;
+       /* reference count of i_bh inherits from nilfs_mdt_read_block() */
+
+       atomic_inc(&sbi->s_inodes_count);
+
+       inode->i_uid = current_fsuid();
+       if (dir->i_mode & S_ISGID) {
+               inode->i_gid = dir->i_gid;
+               if (S_ISDIR(mode))
+                       mode |= S_ISGID;
+       } else
+               inode->i_gid = current_fsgid();
+
+       inode->i_mode = mode;
+       inode->i_ino = ino;
+       inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
+
+       if (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)) {
+               err = nilfs_bmap_read(ii->i_bmap, NULL);
+               if (err < 0)
+                       goto failed_bmap;
+
+               set_bit(NILFS_I_BMAP, &ii->i_state);
+               /* No lock is needed; iget() ensures it. */
+       }
+
+       ii->i_flags = NILFS_I(dir)->i_flags;
+       if (S_ISLNK(mode))
+               ii->i_flags &= ~(NILFS_IMMUTABLE_FL | NILFS_APPEND_FL);
+       if (!S_ISDIR(mode))
+               ii->i_flags &= ~NILFS_DIRSYNC_FL;
+
+       /* ii->i_file_acl = 0; */
+       /* ii->i_dir_acl = 0; */
+       ii->i_dtime = 0;
+       ii->i_dir_start_lookup = 0;
+#ifdef CONFIG_NILFS_FS_POSIX_ACL
+       ii->i_acl = NULL;
+       ii->i_default_acl = NULL;
+#endif
+       ii->i_cno = 0;
+       nilfs_set_inode_flags(inode);
+       spin_lock(&sbi->s_next_gen_lock);
+       inode->i_generation = sbi->s_next_generation++;
+       spin_unlock(&sbi->s_next_gen_lock);
+       insert_inode_hash(inode);
+
+       err = nilfs_init_acl(inode, dir);
+       if (unlikely(err))
+               goto failed_acl; /* never occur. When supporting
+                                   nilfs_init_acl(), proper cancellation of
+                                   above jobs should be considered */
+
+       mark_inode_dirty(inode);
+       return inode;
+
+ failed_acl:
+ failed_bmap:
+       inode->i_nlink = 0;
+       iput(inode);  /* raw_inode will be deleted through
+                        generic_delete_inode() */
+       goto failed;
+
+ failed_ifile_create_inode:
+       make_bad_inode(inode);
+       iput(inode);  /* if i_nlink == 1, generic_forget_inode() will be
+                        called */
+ failed:
+       return ERR_PTR(err);
+}
+
+void nilfs_free_inode(struct inode *inode)
+{
+       struct super_block *sb = inode->i_sb;
+       struct nilfs_sb_info *sbi = NILFS_SB(sb);
+
+       clear_inode(inode);
+       /* XXX: check error code? Is there any thing I can do? */
+       (void) nilfs_ifile_delete_inode(sbi->s_ifile, inode->i_ino);
+       atomic_dec(&sbi->s_inodes_count);
+}
+
+void nilfs_set_inode_flags(struct inode *inode)
+{
+       unsigned int flags = NILFS_I(inode)->i_flags;
+
+       inode->i_flags &= ~(S_SYNC | S_APPEND | S_IMMUTABLE | S_NOATIME |
+                           S_DIRSYNC);
+       if (flags & NILFS_SYNC_FL)
+               inode->i_flags |= S_SYNC;
+       if (flags & NILFS_APPEND_FL)
+               inode->i_flags |= S_APPEND;
+       if (flags & NILFS_IMMUTABLE_FL)
+               inode->i_flags |= S_IMMUTABLE;
+#ifndef NILFS_ATIME_DISABLE
+       if (flags & NILFS_NOATIME_FL)
+#endif
+               inode->i_flags |= S_NOATIME;
+       if (flags & NILFS_DIRSYNC_FL)
+               inode->i_flags |= S_DIRSYNC;
+       mapping_set_gfp_mask(inode->i_mapping,
+                            mapping_gfp_mask(inode->i_mapping) & ~__GFP_FS);
+}
+
+int nilfs_read_inode_common(struct inode *inode,
+                           struct nilfs_inode *raw_inode)
+{
+       struct nilfs_inode_info *ii = NILFS_I(inode);
+       int err;
+
+       inode->i_mode = le16_to_cpu(raw_inode->i_mode);
+       inode->i_uid = (uid_t)le32_to_cpu(raw_inode->i_uid);
+       inode->i_gid = (gid_t)le32_to_cpu(raw_inode->i_gid);
+       inode->i_nlink = le16_to_cpu(raw_inode->i_links_count);
+       inode->i_size = le64_to_cpu(raw_inode->i_size);
+       inode->i_atime.tv_sec = le64_to_cpu(raw_inode->i_mtime);
+       inode->i_ctime.tv_sec = le64_to_cpu(raw_inode->i_ctime);
+       inode->i_mtime.tv_sec = le64_to_cpu(raw_inode->i_mtime);
+       inode->i_atime.tv_nsec = 0;
+       inode->i_ctime.tv_nsec = 0;
+       inode->i_mtime.tv_nsec = 0;
+       ii->i_dtime = le64_to_cpu(raw_inode->i_dtime);
+       if (inode->i_nlink == 0 && (inode->i_mode == 0 || ii->i_dtime))
+               return -EINVAL; /* this inode is deleted */
+
+       inode->i_blocks = le64_to_cpu(raw_inode->i_blocks);
+       ii->i_flags = le32_to_cpu(raw_inode->i_flags);
+#if 0
+       ii->i_file_acl = le32_to_cpu(raw_inode->i_file_acl);
+       ii->i_dir_acl = S_ISREG(inode->i_mode) ?
+               0 : le32_to_cpu(raw_inode->i_dir_acl);
+#endif
+       ii->i_cno = 0;
+       inode->i_generation = le32_to_cpu(raw_inode->i_generation);
+
+       if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
+           S_ISLNK(inode->i_mode)) {
+               err = nilfs_bmap_read(ii->i_bmap, raw_inode);
+               if (err < 0)
+                       return err;
+               set_bit(NILFS_I_BMAP, &ii->i_state);
+               /* No lock is needed; iget() ensures it. */
+       }
+       return 0;
+}
+
+static int nilfs_read_sketch_inode(struct inode *inode)
+{
+       struct nilfs_sb_info *sbi = NILFS_SB(inode->i_sb);
+       int err = 0;
+
+       if (sbi->s_snapshot_cno) {
+               struct the_nilfs *nilfs = sbi->s_nilfs;
+               struct buffer_head *bh_cp;
+               struct nilfs_checkpoint *raw_cp;
+
+               err = nilfs_cpfile_get_checkpoint(
+                       nilfs->ns_cpfile, sbi->s_snapshot_cno, 0, &raw_cp,
+                       &bh_cp);
+               if (likely(!err)) {
+                       if (!nilfs_checkpoint_sketch(raw_cp))
+                               inode->i_size = 0;
+                       nilfs_cpfile_put_checkpoint(
+                               nilfs->ns_cpfile, sbi->s_snapshot_cno, bh_cp);
+               }
+               inode->i_flags |= S_NOCMTIME;
+       }
+       return err;
+}
+
+static int __nilfs_read_inode(struct super_block *sb, unsigned long ino,
+                             struct inode *inode)
+{
+       struct nilfs_sb_info *sbi = NILFS_SB(sb);
+       struct inode *dat = nilfs_dat_inode(sbi->s_nilfs);
+       struct buffer_head *bh;
+       struct nilfs_inode *raw_inode;
+       int err;
+
+       down_read(&NILFS_MDT(dat)->mi_sem);     /* XXX */
+       err = nilfs_ifile_get_inode_block(sbi->s_ifile, ino, &bh);
+       if (unlikely(err))
+               goto bad_inode;
+
+       raw_inode = nilfs_ifile_map_inode(sbi->s_ifile, ino, bh);
+
+#ifdef CONFIG_NILFS_FS_POSIX_ACL
+       ii->i_acl = NILFS_ACL_NOT_CACHED;
+       ii->i_default_acl = NILFS_ACL_NOT_CACHED;
+#endif
+       if (nilfs_read_inode_common(inode, raw_inode))
+               goto failed_unmap;
+
+       if (S_ISREG(inode->i_mode)) {
+               inode->i_op = &nilfs_file_inode_operations;
+               inode->i_fop = &nilfs_file_operations;
+               inode->i_mapping->a_ops = &nilfs_aops;
+               if (unlikely(inode->i_ino == NILFS_SKETCH_INO)) {
+                       err = nilfs_read_sketch_inode(inode);
+                       if (unlikely(err))
+                               goto failed_unmap;
+               }
+       } else if (S_ISDIR(inode->i_mode)) {
+               inode->i_op = &nilfs_dir_inode_operations;
+               inode->i_fop = &nilfs_dir_operations;
+               inode->i_mapping->a_ops = &nilfs_aops;
+       } else if (S_ISLNK(inode->i_mode)) {
+               inode->i_op = &nilfs_symlink_inode_operations;
+               inode->i_mapping->a_ops = &nilfs_aops;
+       } else {
+               inode->i_op = &nilfs_special_inode_operations;
+               init_special_inode(
+                       inode, inode->i_mode,
+                       new_decode_dev(le64_to_cpu(raw_inode->i_device_code)));
+       }
+       nilfs_ifile_unmap_inode(sbi->s_ifile, ino, bh);
+       brelse(bh);
+       up_read(&NILFS_MDT(dat)->mi_sem);       /* XXX */
+       nilfs_set_inode_flags(inode);
+       return 0;
+
+ failed_unmap:
+       nilfs_ifile_unmap_inode(sbi->s_ifile, ino, bh);
+       brelse(bh);
+
+ bad_inode:
+       up_read(&NILFS_MDT(dat)->mi_sem);       /* XXX */
+       return err;
+}
+
+struct inode *nilfs_iget(struct super_block *sb, unsigned long ino)
+{
+       struct inode *inode;
+       int err;
+
+       inode = iget_locked(sb, ino);
+       if (unlikely(!inode))
+               return ERR_PTR(-ENOMEM);
+       if (!(inode->i_state & I_NEW))
+               return inode;
+
+       err = __nilfs_read_inode(sb, ino, inode);
+       if (unlikely(err)) {
+               iget_failed(inode);
+               return ERR_PTR(err);
+       }
+       unlock_new_inode(inode);
+       return inode;
+}
+
+void nilfs_write_inode_common(struct inode *inode,
+                             struct nilfs_inode *raw_inode, int has_bmap)
+{
+       struct nilfs_inode_info *ii = NILFS_I(inode);
+
+       raw_inode->i_mode = cpu_to_le16(inode->i_mode);
+       raw_inode->i_uid = cpu_to_le32(inode->i_uid);
+       raw_inode->i_gid = cpu_to_le32(inode->i_gid);
+       raw_inode->i_links_count = cpu_to_le16(inode->i_nlink);
+       raw_inode->i_size = cpu_to_le64(inode->i_size);
+       raw_inode->i_ctime = cpu_to_le64(inode->i_ctime.tv_sec);
+       raw_inode->i_mtime = cpu_to_le64(inode->i_mtime.tv_sec);
+       raw_inode->i_blocks = cpu_to_le64(inode->i_blocks);
+
+       raw_inode->i_dtime = cpu_to_le64(ii->i_dtime);
+       raw_inode->i_flags = cpu_to_le32(ii->i_flags);
+       raw_inode->i_generation = cpu_to_le32(inode->i_generation);
+
+       if (has_bmap)
+               nilfs_bmap_write(ii->i_bmap, raw_inode);
+       else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode))
+               raw_inode->i_device_code =
+                       cpu_to_le64(new_encode_dev(inode->i_rdev));
+       /* When extending inode, nilfs->ns_inode_size should be checked
+          for substitutions of appended fields */
+}
+
+void nilfs_update_inode(struct inode *inode, struct buffer_head *ibh)
+{
+       ino_t ino = inode->i_ino;
+       struct nilfs_inode_info *ii = NILFS_I(inode);
+       struct super_block *sb = inode->i_sb;
+       struct nilfs_sb_info *sbi = NILFS_SB(sb);
+       struct nilfs_inode *raw_inode;
+
+       raw_inode = nilfs_ifile_map_inode(sbi->s_ifile, ino, ibh);
+
+       /* The buffer is guarded with lock_buffer() by the caller */
+       if (test_and_clear_bit(NILFS_I_NEW, &ii->i_state))
+               memset(raw_inode, 0, NILFS_MDT(sbi->s_ifile)->mi_entry_size);
+       set_bit(NILFS_I_INODE_DIRTY, &ii->i_state);
+
+       nilfs_write_inode_common(inode, raw_inode, 0);
+               /* XXX: call with has_bmap = 0 is a workaround to avoid
+                  deadlock of bmap. This delays update of i_bmap to just
+                  before writing */
+       nilfs_ifile_unmap_inode(sbi->s_ifile, ino, ibh);
+}
+
+#define NILFS_MAX_TRUNCATE_BLOCKS      16384  /* 64MB for 4KB block */
+
+static void nilfs_truncate_bmap(struct nilfs_inode_info *ii,
+                               unsigned long from)
+{
+       unsigned long b;
+       int ret;
+
+       if (!test_bit(NILFS_I_BMAP, &ii->i_state))
+               return;
+ repeat:
+       ret = nilfs_bmap_last_key(ii->i_bmap, &b);
+       if (ret == -ENOENT)
+               return;
+       else if (ret < 0)
+               goto failed;
+
+       if (b < from)
+               return;
+
+       b -= min_t(unsigned long, NILFS_MAX_TRUNCATE_BLOCKS, b - from);
+       ret = nilfs_bmap_truncate(ii->i_bmap, b);
+       nilfs_relax_pressure_in_lock(ii->vfs_inode.i_sb);
+       if (!ret || (ret == -ENOMEM &&
+                    nilfs_bmap_truncate(ii->i_bmap, b) == 0))
+               goto repeat;
+
+ failed:
+       if (ret == -EINVAL)
+               nilfs_error(ii->vfs_inode.i_sb, __func__,
+                           "bmap is broken (ino=%lu)", ii->vfs_inode.i_ino);
+       else
+               nilfs_warning(ii->vfs_inode.i_sb, __func__,
+                             "failed to truncate bmap (ino=%lu, err=%d)",
+                             ii->vfs_inode.i_ino, ret);
+}
+
+void nilfs_truncate(struct inode *inode)
+{
+       unsigned long blkoff;
+       unsigned int blocksize;
+       struct nilfs_transaction_info ti;
+       struct super_block *sb = inode->i_sb;
+       struct nilfs_inode_info *ii = NILFS_I(inode);
+       int ret;
+
+       if (!test_bit(NILFS_I_BMAP, &ii->i_state))
+               return;
+       if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
+               return;
+
+       blocksize = sb->s_blocksize;
+       blkoff = (inode->i_size + blocksize - 1) >> sb->s_blocksize_bits;
+       ret = nilfs_transaction_begin(sb, &ti, 0);
+       BUG_ON(ret);
+
+       block_truncate_page(inode->i_mapping, inode->i_size, nilfs_get_block);
+
+       nilfs_truncate_bmap(ii, blkoff);
+
+       inode->i_mtime = inode->i_ctime = CURRENT_TIME;
+       if (IS_SYNC(inode))
+               nilfs_set_transaction_flag(NILFS_TI_SYNC);
+
+       nilfs_set_file_dirty(NILFS_SB(sb), inode, 0);
+       nilfs_transaction_end(sb, 1);
+       /* May construct a logical segment and may fail in sync mode.
+          But truncate has no return value. */
+}
+
+void nilfs_delete_inode(struct inode *inode)
+{
+       struct nilfs_transaction_info ti;
+       struct super_block *sb = inode->i_sb;
+       struct nilfs_inode_info *ii = NILFS_I(inode);
+       int err;
+
+       if (unlikely(is_bad_inode(inode))) {
+               if (inode->i_data.nrpages)
+                       truncate_inode_pages(&inode->i_data, 0);
+               clear_inode(inode);
+               return;
+       }
+       err = nilfs_transaction_begin(sb, &ti, 0);
+       BUG_ON(err);
+       if (inode->i_data.nrpages)
+               truncate_inode_pages(&inode->i_data, 0);
+
+       nilfs_truncate_bmap(ii, 0);
+       nilfs_free_inode(inode);
+       /* nilfs_free_inode() marks inode buffer dirty */
+       if (IS_SYNC(inode))
+               nilfs_set_transaction_flag(NILFS_TI_SYNC);
+       nilfs_transaction_end(sb, 1);
+       /* May construct a logical segment and may fail in sync mode.
+          But delete_inode has no return value. */
+}
+
+int nilfs_setattr(struct dentry *dentry, struct iattr *iattr)
+{
+       struct nilfs_transaction_info ti;
+       struct inode *inode = dentry->d_inode;
+       struct super_block *sb = inode->i_sb;
+       int err, err2;
+
+       err = inode_change_ok(inode, iattr);
+       if (err)
+               return err;
+
+       err = nilfs_transaction_begin(sb, &ti, 0);
+       if (unlikely(err))
+               return err;
+       err = inode_setattr(inode, iattr);
+       if (!err && (iattr->ia_valid & ATTR_MODE))
+               err = nilfs_acl_chmod(inode);
+       err2 = nilfs_transaction_end(sb, 1);
+       return err ? : err2;
+}
+
+int nilfs_load_inode_block(struct nilfs_sb_info *sbi, struct inode *inode,
+                          struct buffer_head **pbh)
+{
+       struct nilfs_inode_info *ii = NILFS_I(inode);
+       int err;
+
+       spin_lock(&sbi->s_inode_lock);
+       /* Caller of this function MUST lock s_inode_lock */
+       if (ii->i_bh == NULL) {
+               spin_unlock(&sbi->s_inode_lock);
+               err = nilfs_ifile_get_inode_block(sbi->s_ifile, inode->i_ino,
+                                                 pbh);
+               if (unlikely(err))
+                       return err;
+               spin_lock(&sbi->s_inode_lock);
+               if (ii->i_bh == NULL)
+                       ii->i_bh = *pbh;
+               else {
+                       brelse(*pbh);
+                       *pbh = ii->i_bh;
+               }
+       } else
+               *pbh = ii->i_bh;
+
+       get_bh(*pbh);
+       spin_unlock(&sbi->s_inode_lock);
+       return 0;
+}
+
+int nilfs_inode_dirty(struct inode *inode)
+{
+       struct nilfs_inode_info *ii = NILFS_I(inode);
+       struct nilfs_sb_info *sbi = NILFS_SB(inode->i_sb);
+       int ret = 0;
+
+       if (!list_empty(&ii->i_dirty)) {
+               spin_lock(&sbi->s_inode_lock);
+               ret = test_bit(NILFS_I_DIRTY, &ii->i_state) ||
+                       test_bit(NILFS_I_BUSY, &ii->i_state);
+               spin_unlock(&sbi->s_inode_lock);
+       }
+       return ret;
+}
+
+int nilfs_set_file_dirty(struct nilfs_sb_info *sbi, struct inode *inode,
+                        unsigned nr_dirty)
+{
+       struct nilfs_inode_info *ii = NILFS_I(inode);
+
+       atomic_add(nr_dirty, &sbi->s_nilfs->ns_ndirtyblks);
+
+       if (test_and_set_bit(NILFS_I_DIRTY, &ii->i_state) ||
+           unlikely(inode->i_ino == NILFS_SKETCH_INO))
+               return 0;
+
+       spin_lock(&sbi->s_inode_lock);
+       if (!test_bit(NILFS_I_QUEUED, &ii->i_state) &&
+           !test_bit(NILFS_I_BUSY, &ii->i_state)) {
+               /* Because this routine may race with nilfs_dispose_list(),
+                  we have to check NILFS_I_QUEUED here, too. */
+               if (list_empty(&ii->i_dirty) && igrab(inode) == NULL) {
+                       /* This will happen when somebody is freeing
+                          this inode. */
+                       nilfs_warning(sbi->s_super, __func__,
+                                     "cannot get inode (ino=%lu)\n",
+                                     inode->i_ino);
+                       spin_unlock(&sbi->s_inode_lock);
+                       return -EINVAL; /* NILFS_I_DIRTY may remain for
+                                          freeing inode */
+               }
+               list_del(&ii->i_dirty);
+               list_add_tail(&ii->i_dirty, &sbi->s_dirty_files);
+               set_bit(NILFS_I_QUEUED, &ii->i_state);
+       }
+       spin_unlock(&sbi->s_inode_lock);
+       return 0;
+}
+
+int nilfs_mark_inode_dirty(struct inode *inode)
+{
+       struct nilfs_sb_info *sbi = NILFS_SB(inode->i_sb);
+       struct buffer_head *ibh;
+       int err;
+
+       err = nilfs_load_inode_block(sbi, inode, &ibh);
+       if (unlikely(err)) {
+               nilfs_warning(inode->i_sb, __func__,
+                             "failed to reget inode block.\n");
+               return err;
+       }
+       lock_buffer(ibh);
+       nilfs_update_inode(inode, ibh);
+       unlock_buffer(ibh);
+       nilfs_mdt_mark_buffer_dirty(ibh);
+       nilfs_mdt_mark_dirty(sbi->s_ifile);
+       brelse(ibh);
+       return 0;
+}
+
+/**
+ * nilfs_dirty_inode - reflect changes on given inode to an inode block.
+ * @inode: inode of the file to be registered.
+ *
+ * nilfs_dirty_inode() loads a inode block containing the specified
+ * @inode and copies data from a nilfs_inode to a corresponding inode
+ * entry in the inode block. This operation is excluded from the segment
+ * construction. This function can be called both as a single operation
+ * and as a part of indivisible file operations.
+ */
+void nilfs_dirty_inode(struct inode *inode)
+{
+       struct nilfs_transaction_info ti;
+
+       if (is_bad_inode(inode)) {
+               nilfs_warning(inode->i_sb, __func__,
+                             "tried to mark bad_inode dirty. ignored.\n");
+               dump_stack();
+               return;
+       }
+       nilfs_transaction_begin(inode->i_sb, &ti, 0);
+       if (likely(inode->i_ino != NILFS_SKETCH_INO))
+               nilfs_mark_inode_dirty(inode);
+       nilfs_transaction_end(inode->i_sb, 1); /* never fails */
+}