From: Chuck Lever <chuck.lever@xxxxxxxxxx>
Jeff confirmed his original fix addressed his pynfs test failure,
but this same bug also impacted qemu: accessing qcow2 virtual disks
using direct I/O was failing. Jeff's fix missed that you have to
shorten the bio_vec element by the same amount as you increased
the page offset.
Reported-by: Maxim Levitsky <mlevitsky@xxxxxxxxxx>
Fixes: c96e2a695e00 ("sunrpc: set the bv_offset of first bvec in svc_tcp_sendmsg")
Tested-by: Maxim Levitsky <mlevitsky@xxxxxxxxxx>
Signed-off-by: Chuck Lever <chuck.lever@xxxxxxxxxx>
---
net/sunrpc/svcsock.c | 4 +++-
1 file changed, 3 insertions(+), 1 deletion(-)
The plan is to put this first in the nfsd-next pull request so that
it can back-ported immediately to 6.5.y.
diff --git a/net/sunrpc/svcsock.c b/net/sunrpc/svcsock.c
index 2eb8df44f894..589020ed909d 100644
--- a/net/sunrpc/svcsock.c
+++ b/net/sunrpc/svcsock.c
@@ -1244,8 +1244,10 @@ static int svc_tcp_sendmsg(struct socket *sock, struct xdr_buf *xdr,
if (ret != head->iov_len)
goto out;
- if (xdr_buf_pagecount(xdr))
+ if (xdr_buf_pagecount(xdr)) {
xdr->bvec[0].bv_offset = offset_in_page(xdr->page_base);
+ xdr->bvec[0].bv_len -= offset_in_page(xdr->page_base);
+ }
msg.msg_flags = MSG_SPLICE_PAGES;
iov_iter_bvec(&msg.msg_iter, ITER_SOURCE, xdr->bvec,
