ocfs2: Block signals for mkdir/link/symlink/O_CREAT.
authorJoel Becker <joel.becker@oracle.com>
Mon, 10 May 2010 18:56:52 +0000 (11:56 -0700)
committerJoel Becker <joel.becker@oracle.com>
Mon, 10 May 2010 18:56:52 +0000 (11:56 -0700)
Once file or link creation gets going, it can't be interrupted by a
signal.  They're not idempotent.

This blocks signals in ocfs2_mknod(), ocfs2_link(), and ocfs2_symlink()
once we start actually changing things.  ocfs2_mknod() covers mknod(),
creat(), mkdir(), and open(O_CREAT).

Signed-off-by: Joel Becker <joel.becker@oracle.com>
fs/ocfs2/namei.c

index 21d4a33..607084b 100644 (file)
@@ -239,6 +239,8 @@ static int ocfs2_mknod(struct inode *dir,
        };
        int did_quota_inode = 0;
        struct ocfs2_dir_lookup_result lookup = { NULL, };
+       sigset_t oldset;
+       int did_block_signals = 0;
 
        mlog_entry("(0x%p, 0x%p, %d, %lu, '%.*s')\n", dir, dentry, mode,
                   (unsigned long)dev, dentry->d_name.len,
@@ -350,6 +352,10 @@ static int ocfs2_mknod(struct inode *dir,
                goto leave;
        }
 
+       /* Starting to change things, restart is no longer possible. */
+       ocfs2_block_signals(&oldset);
+       did_block_signals = 1;
+
        status = dquot_alloc_inode(inode);
        if (status)
                goto leave;
@@ -430,6 +436,8 @@ leave:
                ocfs2_commit_trans(osb, handle);
 
        ocfs2_inode_unlock(dir, 1);
+       if (did_block_signals)
+               ocfs2_unblock_signals(&oldset);
 
        if (status == -ENOSPC)
                mlog(0, "Disk is full\n");
@@ -618,6 +626,7 @@ static int ocfs2_link(struct dentry *old_dentry,
        struct ocfs2_dinode *fe = NULL;
        struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
        struct ocfs2_dir_lookup_result lookup = { NULL, };
+       sigset_t oldset;
 
        mlog_entry("(inode=%lu, old='%.*s' new='%.*s')\n", inode->i_ino,
                   old_dentry->d_name.len, old_dentry->d_name.name,
@@ -674,6 +683,9 @@ static int ocfs2_link(struct dentry *old_dentry,
                goto out_unlock_inode;
        }
 
+       /* Starting to change things, restart is no longer possible. */
+       ocfs2_block_signals(&oldset);
+
        err = ocfs2_journal_access_di(handle, INODE_CACHE(inode), fe_bh,
                                      OCFS2_JOURNAL_ACCESS_WRITE);
        if (err < 0) {
@@ -710,6 +722,7 @@ static int ocfs2_link(struct dentry *old_dentry,
 
 out_commit:
        ocfs2_commit_trans(osb, handle);
+       ocfs2_unblock_signals(&oldset);
 out_unlock_inode:
        ocfs2_inode_unlock(inode, 1);
 
@@ -1568,6 +1581,8 @@ static int ocfs2_symlink(struct inode *dir,
        };
        int did_quota = 0, did_quota_inode = 0;
        struct ocfs2_dir_lookup_result lookup = { NULL, };
+       sigset_t oldset;
+       int did_block_signals = 0;
 
        mlog_entry("(0x%p, 0x%p, symname='%s' actual='%.*s')\n", dir,
                   dentry, symname, dentry->d_name.len, dentry->d_name.name);
@@ -1663,6 +1678,10 @@ static int ocfs2_symlink(struct inode *dir,
                goto bail;
        }
 
+       /* Starting to change things, restart is no longer possible. */
+       ocfs2_block_signals(&oldset);
+       did_block_signals = 1;
+
        status = dquot_alloc_inode(inode);
        if (status)
                goto bail;
@@ -1766,6 +1785,8 @@ bail:
                ocfs2_commit_trans(osb, handle);
 
        ocfs2_inode_unlock(dir, 1);
+       if (did_block_signals)
+               ocfs2_unblock_signals(&oldset);
 
        brelse(new_fe_bh);
        brelse(parent_fe_bh);