9p: fix enodata when reading growing file

Reading from a file that was just extended by a write, but the write had
not yet reached the server would return ENODATA as illustrated by this
command:
$ xfs_io -c 'open -ft test' -c 'w 4096 1000' -c 'r 0 1000'
wrote 1000/1000 bytes at offset 4096
1000.000000 bytes, 1 ops; 0.0001 sec (5.610 MiB/sec and 5882.3529 ops/sec)
pread: No data available

Fix this case by having netfs assume zeroes when reads from server come
short like AFS and CEPH do

Link: https://lkml.kernel.org/r/20220110111444.926753-1-asmadeus@codewreck.org
Cc: stable@vger.kernel.org
Fixes: eb497943fa21 ("9p: Convert to using the netfs helper lib to do reads and caching")
Co-authored-by: David Howells <dhowells@redhat.com>
Reviewed-by: David Howells <dhowells@redhat.com>
Tested-by: David Howells <dhowells@redhat.com>
Signed-off-by: Dominique Martinet <asmadeus@codewreck.org>

diff --git a/fs/9p/vfs_addr.c b/fs/9p/vfs_addr.c
index fac918c..1d554d0 100644 (file)
--- a/fs/9p/vfs_addr.c
+++ b/fs/9p/vfs_addr.c
@@ -42,6 +42,11 @@ static void v9fs_req_issue_op(struct netfs_read_subrequest *subreq)
        iov_iter_xarray(&to, READ, &rreq->mapping->i_pages, pos, len);
 
        total = p9_client_read(fid, pos, &to, &err);
+
+       /* if we just extended the file size, any portion not in
+        * cache won't be on server and is zeroes */
+       __set_bit(NETFS_SREQ_CLEAR_TAIL, &subreq->flags);
+
        netfs_subreq_terminated(subreq, err ?: total, false);
 }