Re: [PATCH] retries in ext3_prepare_write() violate ordering requirements

[Kirill Korotaev]

Andrew,

>>in journal=ordered or journal=data mode retry in ext3_prepare_write()
>>breaks the requirements of journaling of data with respect to metadata.
>>The fix is to call commit_write to commit allocated zero blocks before
>>retry.
>>
> 
> 
> How was this problem detected? (ie: why was block_prepare_write() failing?)
purely theoretically while hunting for other bugs related to ext3 and quota.
block_prepare_write() can fail e.g. if quota returns -EDQUOT in ext3_alloc_blocks().

> How was the patch tested?
1. it was tested as part of OpenVZ kernel
2. there were ext3 stress test done with lots of disk activity by Dmitry Monakhov.

> Was nobh-mode also tested?
I will ask to perform some more tests 100% triggering ext3_prepare_failure()
and with NOBH mode.

Thanks,
Kirill


>>--- ./fs/ext3/inode.c.ext3pw	2006-11-08 17:44:14.000000000 +0300
>>+++ ./fs/ext3/inode.c	2006-11-08 17:48:59.000000000 +0300
>>@@ -1148,37 +1148,89 @@ static int do_journal_get_write_access(h
>> 	return ext3_journal_get_write_access(handle, bh);
>> }
>> 
>>+/*
>>+ * The idea of this helper function is following:
>>+ * if prepare_write has allocated some blocks, but not all of them, the
>>+ * transaction must include the content of the newly allocated blocks.
>>+ * This content is expected to be set to zeroes by block_prepare_write().
>>+ * 2006/10/14  SAW
>>+ */
>>+static int ext3_prepare_failure(struct file *file, struct page *page,
>>+				unsigned from, unsigned to)
>>+{
>>+	struct address_space *mapping;
>>+	struct buffer_head *bh, *head, *next;
>>+	unsigned block_start, block_end;
>>+	unsigned blocksize;
>>+
>>+	mapping = page->mapping;
>>+	if (ext3_should_writeback_data(mapping->host)) {
>>+		/* optimization: no constraints about data */
>>+skip:
>>+		ext3_journal_stop(ext3_journal_current_handle());
>>+		return 0;
> 
> 
> Should this be `return ext3_journal_stop(...);'?
> 
> 
>>+	}
>>+
>>+	head = page_buffers(page);
>>+	blocksize = head->b_size;
>>+	for (	bh = head, block_start = 0;
>>+		bh != head || !block_start;
>>+	    	block_start = block_end, bh = next)
>>+	{
>>+		next = bh->b_this_page;
>>+		block_end = block_start + blocksize;
>>+		if (block_end <= from)
>>+			continue;
>>+		if (block_start >= to) {
>>+			block_start = to;
>>+			break;
>>+		}
>>+		if (!buffer_mapped(bh))
>>+			break;
> 
> 
> What is the significance of buffer_mapped() here?  Outside EOF or into a
> hole?  If so, then block_start >= to, and we can't get here??
> 
> 
>>+	}
>>+	if (block_start <= from)
>>+		goto skip;
>>+
>>+	/* commit allocated and zeroed buffers */
>>+	return mapping->a_ops->commit_write(file, page, from, block_start);
>>+}
>>+
>> static int ext3_prepare_write(struct file *file, struct page *page,
>> 			      unsigned from, unsigned to)
>> {
>> 	struct inode *inode = page->mapping->host;
>>-	int ret, needed_blocks = ext3_writepage_trans_blocks(inode);
>>+	int ret, ret2;
>>+	int needed_blocks = ext3_writepage_trans_blocks(inode);
>> 	handle_t *handle;
>> 	int retries = 0;
>> 
>> retry:
>> 	handle = ext3_journal_start(inode, needed_blocks);
>>-	if (IS_ERR(handle)) {
>>-		ret = PTR_ERR(handle);
>>-		goto out;
>>-	}
>>+	if (IS_ERR(handle))
>>+		return PTR_ERR(handle);
>> 	if (test_opt(inode->i_sb, NOBH) && ext3_should_writeback_data(inode))
>> 		ret = nobh_prepare_write(page, from, to, ext3_get_block);
>> 	else
>> 		ret = block_prepare_write(page, from, to, ext3_get_block);
>> 	if (ret)
>>-		goto prepare_write_failed;
>>+		goto failure;
>> 
>> 	if (ext3_should_journal_data(inode)) {
>> 		ret = walk_page_buffers(handle, page_buffers(page),
>> 				from, to, NULL, do_journal_get_write_access);
>>+		if (ret)
>>+			/* fatal error, just put the handle and return */
>>+			journal_stop(handle);
>> 	}
>>-prepare_write_failed:
>>-	if (ret)
>>-		ext3_journal_stop(handle);
>>+	return ret;
>>+
>>+failure:
>>+	ret2 = ext3_prepare_failure(file, page, from, to);
>>+	if (ret2 < 0)
>>+		return ret2;
>> 	if (ret == -ENOSPC && ext3_should_retry_alloc(inode->i_sb, &retries))
>> 		goto retry;
>>-out:
>>+	/* retry number exceeded, or other error like -EDQUOT */
>> 	return ret;
>> }
>> 
> 
> 

-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/