On Tue, 7 May 2013 13:38:35 -0700, Andrew Morton wrote:
> On Wed, 01 May 2013 16:39:30 +0900 (JST) Ryusuke Konishi <konishi.ryusuke@xxxxxxxxxxxxx> wrote:
>
>> DESCRIPTION:
>> There are use-cases when NILFS2 file system (formatted with block size
>> lesser than 4 KB) can be remounted in RO mode because of encountering
>> of "broken bmap" issue.
>>
>> ...
>>
>> --- a/fs/nilfs2/inode.c
>> +++ b/fs/nilfs2/inode.c
>> @@ -219,13 +219,25 @@ static int nilfs_writepage(struct page *page, struct writeback_control *wbc)
>>
>> static int nilfs_set_page_dirty(struct page *page)
>> {
>> - int ret = __set_page_dirty_buffers(page);
>> + int ret = __set_page_dirty_nobuffers(page);
>>
>> - if (ret) {
>> + if (page_has_buffers(page)) {
>> struct inode *inode = page->mapping->host;
>> - unsigned nr_dirty = 1 << (PAGE_SHIFT - inode->i_blkbits);
>> + unsigned nr_dirty = 0;
>> + struct buffer_head *bh, *head;
>>
>> - nilfs_set_file_dirty(inode, nr_dirty);
>> + bh = head = page_buffers(page);
>> + do {
>> + /* Do not mark hole blocks dirty */
>> + if (!buffer_dirty(bh) || !buffer_mapped(bh))
>> + continue;
>> +
>> + set_buffer_dirty(bh);
>
> This doesn't seem right. The only time we will run set_buffer_dirty()
> is if the buffer was dirty and mapped. What's the point in running
> set_buffer_dirty() against an already dirty buffer?
>
> I suspect you intended (buffer_dirty || !buffer_mapped)
Yes, sorry, this is my mistake.
I'll correct it and redo tests.
>> + nr_dirty++;
>> + } while (bh = bh->b_this_page, bh != head);
>> +
>> + if (nr_dirty)
>> + nilfs_set_file_dirty(inode, nr_dirty);
>
> Rather secondarily: the above code appears to assume that no other
> thread can come in and mark a buffer dirty while the loop is executing.
> This *should* be OK - the page is locked (or it should be). Can you
> confirm that this is the case? There are no other places where a
> buffer can be concurrently dirtied against an unlocked page?
Yes, I confirmed no other thread concurrently marks the above buffer
dirty.
The nilfs_page_set_dirty function is called only for data blocks of
regular files, directory data, or symlinks.
And, nilfs only dirties this type of buffers through routines in
fs/buffer.c. Every routine in fs/buffer.c that can be called from
nilfs, was protecting call sites of mark_buffer_dirty() by page lock.
I'll add a comment to brief this assumption.
Thanks,
Ryusuke Konishi
>> }
>> return ret;
>> }
>
> --
> To unsubscribe from this list: send the line "unsubscribe linux-fsdevel" in
> the body of a message to majordomo@xxxxxxxxxxxxxxx
> More majordomo info at http://vger.kernel.org/majordomo-info.html
--
To unsubscribe from this list: send the line "unsubscribe linux-fsdevel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
