CIFS: Fix splice read for non-cached files

commit 9c25702cee1405099f982894c865c163de7909a8 upstream.

Currently we call copy_page_to_iter() for uncached reading into a pipe.
This is wrong because it treats pages as VFS cache pages and copies references
rather than actual data. When we are trying to read from the pipe we end up
calling page_cache_pipe_buf_confirm() which returns -ENODATA. This error
is translated into 0 which is returned to a user.

This issue is reproduced by running xfs-tests suite (generic test #249)
against mount points with "cache=none". Fix it by mapping pages manually
and calling copy_to_iter() that copies data into the pipe.

Signed-off-by: Pavel Shilovsky <pshilov@microsoft.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>

diff --git a/fs/cifs/file.c b/fs/cifs/file.c
index 18a1e1d6671fc0825b2f265280a96e065e6c6aae..1cd0e2eefc667aa8dbf22aa84516c1d527d774ac 100644 (file)
--- a/fs/cifs/file.c
+++ b/fs/cifs/file.c
@@ -2884,7 +2884,15 @@ cifs_readdata_to_iov(struct cifs_readdata *rdata, struct iov_iter *iter)
        for (i = 0; i < rdata->nr_pages; i++) {
                struct page *page = rdata->pages[i];
                size_t copy = min_t(size_t, remaining, PAGE_SIZE);
-               size_t written = copy_page_to_iter(page, 0, copy, iter);
+               size_t written;
+
+               if (unlikely(iter->type & ITER_PIPE)) {
+                       void *addr = kmap_atomic(page);
+
+                       written = copy_to_iter(addr, copy, iter);
+                       kunmap_atomic(addr);
+               } else
+                       written = copy_page_to_iter(page, 0, copy, iter);
                remaining -= written;
                if (written < copy && iov_iter_count(iter) > 0)
                        break;