ext4: retry allocation when inline->extent conversion failed
authorJan Kara <jack@suse.cz>
Mon, 6 Jan 2014 19:02:23 +0000 (14:02 -0500)
committerTheodore Ts'o <tytso@mit.edu>
Mon, 6 Jan 2014 19:02:23 +0000 (14:02 -0500)
Similarly as other ->write_begin functions in ext4, also
ext4_da_write_inline_data_begin() should retry allocation if the
conversion failed because of ENOSPC. This avoids returning ENOSPC
prematurely because of uncommitted block deletions.

Signed-off-by: Jan Kara <jack@suse.cz>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
fs/ext4/inline.c

index bae9875..ed6e71f 100644 (file)
@@ -849,11 +849,13 @@ int ext4_da_write_inline_data_begin(struct address_space *mapping,
        handle_t *handle;
        struct page *page;
        struct ext4_iloc iloc;
+       int retries;
 
        ret = ext4_get_inode_loc(inode, &iloc);
        if (ret)
                return ret;
 
+retry_journal:
        handle = ext4_journal_start(inode, EXT4_HT_INODE, 1);
        if (IS_ERR(handle)) {
                ret = PTR_ERR(handle);
@@ -875,6 +877,11 @@ int ext4_da_write_inline_data_begin(struct address_space *mapping,
                                                            inode,
                                                            flags,
                                                            fsdata);
+               ext4_journal_stop(handle);
+               handle = NULL;
+               if (ret == -ENOSPC &&
+                   ext4_should_retry_alloc(inode->i_sb, &retries))
+                       goto retry_journal;
                goto out;
        }