When QA testing XFS 1.2 we observed in-memory corruption under extreme
load (fsx, usemem & bash-shared-mapping) when using block size < pagesizes.
In addition to some bugs inside XFS Russell Cattelan found a problem in
end_buffer_io_async.
The problem is that end_buffer_io_async sets the page uptodate as soon
as there are no more async or locked buffers, which is wrong if only
parts of the page are submitted for I/O (i.e. writes not on the page
boundary, etc..)
--- linux-2.4/fs/buffer.c Fri Oct 4 18:14:59 2002
+++ linux/fs/buffer.c Thu Nov 7 16:09:53 2002
@@ -749,6 +749,7 @@
unsigned long flags;
struct buffer_head *tmp;
struct page *page;
+ int fullup = 1;
mark_buffer_uptodate(bh, uptodate);
@@ -775,8 +876,11 @@
unlock_buffer(bh);
tmp = bh->b_this_page;
while (tmp != bh) {
- if (buffer_async(tmp) && buffer_locked(tmp))
- goto still_busy;
+ if (buffer_locked(tmp)) {
+ if (buffer_async(tmp))
+ goto still_busy;
+ } else if (!buffer_uptodate(tmp))
+ fullup = 0;
tmp = tmp->b_this_page;
}
@@ -784,10 +824,10 @@
spin_unlock_irqrestore(&page_uptodate_lock, flags);
/*
- * if none of the buffers had errors then we can set the
- * page uptodate:
+ * If none of the buffers had errors and all were uptodate
+ * then we can set the page uptodate:
*/
- if (!PageError(page))
+ if (fullup && !PageError(page))
SetPageUptodate(page);
UnlockPage(page);
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/
This archive was generated by hypermail 2b29 : Thu Nov 07 2002 - 22:00:49 EST