[PATCH V2] Btrfs: fix old data problem caused by aio vs dio

[PATCH V2] Btrfs: fix old data problem caused by aio vs dio

[Miao Xie]

The 209th case of xfstests failed because of the race between aio and dio. The
detail reason is following:
	Task1		Task2			Btrfs-worker
			invalidate pages
	read pages
			do direct io
			invalidate pages fail*
						finish ordered io
	read data from
	pages

* This step failed because the kernel found the ordered extent object that
covered the pages and thought the pages were still under busy. And then Task1
read the old data from those pages.

This patch fixes the above problem by updating the existed pages directly.

Signed-off-by: Miao Xie <miaox@cn.fujitsu.com>
---
 fs/btrfs/file.c  |   91 ++++++++++++++++++++++++++++++++++++++++++++++++++++++
 fs/btrfs/inode.c |    1 +
 2 files changed, 92 insertions(+), 0 deletions(-)

diff --git a/fs/btrfs/file.c b/fs/btrfs/file.c
index 876cddd..fc0f485 100644
--- a/fs/btrfs/file.c
+++ b/fs/btrfs/file.c
@@ -18,6 +18,7 @@
 
 #include <linux/fs.h>
 #include <linux/pagemap.h>
+#include <linux/pagevec.h>
 #include <linux/highmem.h>
 #include <linux/time.h>
 #include <linux/init.h>
@@ -1328,6 +1329,91 @@ static noinline ssize_t __btrfs_buffered_write(struct file *file,
 	return num_written ? num_written : ret;
 }
 
+static void btrfs_dio_update_existed_pages(struct inode *inode,
+					   struct iov_iter *it,
+					   loff_t pos, size_t size)
+{
+	struct address_space *mapping = inode->i_mapping;
+	struct pagevec pvec;
+	pgoff_t index;
+	pgoff_t end;
+	size_t copied;
+	loff_t copy_pos = pos;
+	int offset = pos & (PAGE_CACHE_SIZE - 1);
+	int i;
+
+	BUG_ON(pos & (PAGE_CACHE_SIZE - 1));
+	BUG_ON(size & (PAGE_CACHE_SIZE - 1));
+
+	pagevec_init(&pvec, 0);
+	index = pos >> PAGE_CACHE_SHIFT;
+	end = (pos + size - 1) >> PAGE_CACHE_SHIFT;
+
+	while (index <= end && pagevec_lookup(&pvec, mapping, index,
+			min(end - index, (pgoff_t)PAGEVEC_SIZE - 1) + 1)) {
+		i = 0;
+		while (i < pagevec_count(&pvec)) {
+			struct page *page = pvec.pages[i];
+			size_t count = min_t(size_t, PAGE_CACHE_SIZE - offset,
+					     size);
+
+			index = page->index;
+			if (index > end)
+				break;
+
+			lock_page(page);
+			WARN_ON(page->index != index);
+			BUG_ON(PageDirty(page));
+			BUG_ON(PageWriteback(page));
+
+			if (page->mapping != mapping) {
+				unlock_page(page);
+				i++;
+				continue;
+			}
+
+			if (!PageUptodate(page)) {
+				unlock_page(page);
+				i++;
+				continue;
+			}
+
+			if ((index << PAGE_CACHE_SHIFT) > copy_pos) {
+				copied = (index << PAGE_CACHE_SHIFT) - copy_pos;
+				iov_iter_advance(it, copied);
+				offset = 0;
+				size -= copied;
+				count = min_t(size_t, PAGE_CACHE_SIZE, size);
+			}
+
+			pagefault_disable();
+			copied = iov_iter_copy_from_user_atomic(page, it,
+								offset,
+								count);
+			pagefault_enable();
+			flush_dcache_page(page);
+
+			if (copied < count)
+				copied = 0;
+
+			iov_iter_advance(it, copied);
+			size -= copied;
+			copy_pos += copied;
+
+			if (unlikely(copied < PAGE_CACHE_SIZE - offset)) {
+				offset += copied;
+			} else {
+				offset = 0;
+				i++;
+			}
+			unlock_page(page);
+		}
+		pagevec_release(&pvec);
+		cond_resched();
+		index++;
+	}
+}
+
 static ssize_t __btrfs_direct_write(struct kiocb *iocb,
 				    const struct iovec *iov,
 				    unsigned long nr_segs, loff_t pos,
@@ -1356,6 +1442,11 @@ static ssize_t __btrfs_direct_write(struct kiocb *iocb,
 		mark_inode_dirty(inode);
 	}
 
+	if (written > 0) {
+		iov_iter_init(&i, iov, nr_segs, count, 0);
+		btrfs_dio_update_existed_pages(inode, &i, pos, written);
+	}
+
 	if (written < 0 || written == count)
 		return written;
 
diff --git a/fs/btrfs/inode.c b/fs/btrfs/inode.c
index 3f2c8cb..6de67a5 100644
--- a/fs/btrfs/inode.c
+++ b/fs/btrfs/inode.c
@@ -6350,6 +6350,7 @@ static ssize_t check_direct_IO(struct btrfs_root *root, int rw, struct kiocb *io
 out:
 	return retval;
 }
+
 static ssize_t btrfs_direct_IO(int rw, struct kiocb *iocb,
 			const struct iovec *iov, loff_t offset,
 			unsigned long nr_segs)
-- 
1.7.6.5

Re: [PATCH V2] Btrfs: fix old data problem caused by aio vs dio

[Josef Bacik]

On Tue, Jun 26, 2012 at 12:07:17AM -0600, Miao Xie wrote:
> The 209th case of xfstests failed because of the race between aio and dio. The
> detail reason is following:
> 	Task1		Task2			Btrfs-worker
> 			invalidate pages
> 	read pages
> 			do direct io
> 			invalidate pages fail*
> 						finish ordered io
> 	read data from
> 	pages
> 

This just papers over the problem and makes DIO touch page cache which it
shouldn't be doing if it's working properly, so NAK.  We need to figure out why
exactly my patch didn't work, since it should be working.  The write should be
doing

lock_extent
setup ordered extent
unlock_extent

and the read should be doing

lock_extent
check for ordered extent
	if there is one unlock and wait and then loop
do read
unlock_extent

there should be no room for races in here.  The patch I sent earlier should have
caught if we had done a read between the invalidate and the locking and should
be invalidating the range again and then checking.  If this isn't working then
something else is going sideways and we really need to figure out what it is
rather than just working around the issue, as it will likely bite us in a
different way later.  Thanks,

Josef