Commit a6b31d18 authored by Trond Myklebust's avatar Trond Myklebust

SUNRPC: Fix a data corruption issue when retransmitting RPC calls

The following scenario can cause silent data corruption when doing
NFS writes. It has mainly been observed when doing database writes
using O_DIRECT.

1) The RPC client uses sendpage() to do zero-copy of the page data.
2) Due to networking issues, the reply from the server is delayed,
   and so the RPC client times out.

3) The client issues a second sendpage of the page data as part of
   an RPC call retransmission.

4) The reply to the first transmission arrives from the server
   _before_ the client hardware has emptied the TCP socket send
   buffer.
5) After processing the reply, the RPC state machine rules that
   the call to be done, and triggers the completion callbacks.
6) The application notices the RPC call is done, and reuses the
   pages to store something else (e.g. a new write).

7) The client NIC drains the TCP socket send buffer. Since the
   page data has now changed, it reads a corrupted version of the
   initial RPC call, and puts it on the wire.

This patch fixes the problem in the following manner:

The ordering guarantees of TCP ensure that when the server sends a
reply, then we know that the _first_ transmission has completed. Using
zero-copy in that situation is therefore safe.
If a time out occurs, we then send the retransmission using sendmsg()
(i.e. no zero-copy), We then know that the socket contains a full copy of
the data, and so it will retransmit a faithful reproduction even if the
RPC call completes, and the application reuses the O_DIRECT buffer in
the meantime.
Signed-off-by: default avatarTrond Myklebust <Trond.Myklebust@netapp.com>
Cc: stable@vger.kernel.org
parent fab99ebe
...@@ -393,8 +393,10 @@ static int xs_send_kvec(struct socket *sock, struct sockaddr *addr, int addrlen, ...@@ -393,8 +393,10 @@ static int xs_send_kvec(struct socket *sock, struct sockaddr *addr, int addrlen,
return kernel_sendmsg(sock, &msg, NULL, 0, 0); return kernel_sendmsg(sock, &msg, NULL, 0, 0);
} }
static int xs_send_pagedata(struct socket *sock, struct xdr_buf *xdr, unsigned int base, int more) static int xs_send_pagedata(struct socket *sock, struct xdr_buf *xdr, unsigned int base, int more, bool zerocopy)
{ {
ssize_t (*do_sendpage)(struct socket *sock, struct page *page,
int offset, size_t size, int flags);
struct page **ppage; struct page **ppage;
unsigned int remainder; unsigned int remainder;
int err, sent = 0; int err, sent = 0;
...@@ -403,6 +405,9 @@ static int xs_send_pagedata(struct socket *sock, struct xdr_buf *xdr, unsigned i ...@@ -403,6 +405,9 @@ static int xs_send_pagedata(struct socket *sock, struct xdr_buf *xdr, unsigned i
base += xdr->page_base; base += xdr->page_base;
ppage = xdr->pages + (base >> PAGE_SHIFT); ppage = xdr->pages + (base >> PAGE_SHIFT);
base &= ~PAGE_MASK; base &= ~PAGE_MASK;
do_sendpage = sock->ops->sendpage;
if (!zerocopy)
do_sendpage = sock_no_sendpage;
for(;;) { for(;;) {
unsigned int len = min_t(unsigned int, PAGE_SIZE - base, remainder); unsigned int len = min_t(unsigned int, PAGE_SIZE - base, remainder);
int flags = XS_SENDMSG_FLAGS; int flags = XS_SENDMSG_FLAGS;
...@@ -410,7 +415,7 @@ static int xs_send_pagedata(struct socket *sock, struct xdr_buf *xdr, unsigned i ...@@ -410,7 +415,7 @@ static int xs_send_pagedata(struct socket *sock, struct xdr_buf *xdr, unsigned i
remainder -= len; remainder -= len;
if (remainder != 0 || more) if (remainder != 0 || more)
flags |= MSG_MORE; flags |= MSG_MORE;
err = sock->ops->sendpage(sock, *ppage, base, len, flags); err = do_sendpage(sock, *ppage, base, len, flags);
if (remainder == 0 || err != len) if (remainder == 0 || err != len)
break; break;
sent += err; sent += err;
...@@ -431,9 +436,10 @@ static int xs_send_pagedata(struct socket *sock, struct xdr_buf *xdr, unsigned i ...@@ -431,9 +436,10 @@ static int xs_send_pagedata(struct socket *sock, struct xdr_buf *xdr, unsigned i
* @addrlen: UDP only -- length of destination address * @addrlen: UDP only -- length of destination address
* @xdr: buffer containing this request * @xdr: buffer containing this request
* @base: starting position in the buffer * @base: starting position in the buffer
* @zerocopy: true if it is safe to use sendpage()
* *
*/ */
static int xs_sendpages(struct socket *sock, struct sockaddr *addr, int addrlen, struct xdr_buf *xdr, unsigned int base) static int xs_sendpages(struct socket *sock, struct sockaddr *addr, int addrlen, struct xdr_buf *xdr, unsigned int base, bool zerocopy)
{ {
unsigned int remainder = xdr->len - base; unsigned int remainder = xdr->len - base;
int err, sent = 0; int err, sent = 0;
...@@ -461,7 +467,7 @@ static int xs_sendpages(struct socket *sock, struct sockaddr *addr, int addrlen, ...@@ -461,7 +467,7 @@ static int xs_sendpages(struct socket *sock, struct sockaddr *addr, int addrlen,
if (base < xdr->page_len) { if (base < xdr->page_len) {
unsigned int len = xdr->page_len - base; unsigned int len = xdr->page_len - base;
remainder -= len; remainder -= len;
err = xs_send_pagedata(sock, xdr, base, remainder != 0); err = xs_send_pagedata(sock, xdr, base, remainder != 0, zerocopy);
if (remainder == 0 || err != len) if (remainder == 0 || err != len)
goto out; goto out;
sent += err; sent += err;
...@@ -564,7 +570,7 @@ static int xs_local_send_request(struct rpc_task *task) ...@@ -564,7 +570,7 @@ static int xs_local_send_request(struct rpc_task *task)
req->rq_svec->iov_base, req->rq_svec->iov_len); req->rq_svec->iov_base, req->rq_svec->iov_len);
status = xs_sendpages(transport->sock, NULL, 0, status = xs_sendpages(transport->sock, NULL, 0,
xdr, req->rq_bytes_sent); xdr, req->rq_bytes_sent, true);
dprintk("RPC: %s(%u) = %d\n", dprintk("RPC: %s(%u) = %d\n",
__func__, xdr->len - req->rq_bytes_sent, status); __func__, xdr->len - req->rq_bytes_sent, status);
if (likely(status >= 0)) { if (likely(status >= 0)) {
...@@ -620,7 +626,7 @@ static int xs_udp_send_request(struct rpc_task *task) ...@@ -620,7 +626,7 @@ static int xs_udp_send_request(struct rpc_task *task)
status = xs_sendpages(transport->sock, status = xs_sendpages(transport->sock,
xs_addr(xprt), xs_addr(xprt),
xprt->addrlen, xdr, xprt->addrlen, xdr,
req->rq_bytes_sent); req->rq_bytes_sent, true);
dprintk("RPC: xs_udp_send_request(%u) = %d\n", dprintk("RPC: xs_udp_send_request(%u) = %d\n",
xdr->len - req->rq_bytes_sent, status); xdr->len - req->rq_bytes_sent, status);
...@@ -693,6 +699,7 @@ static int xs_tcp_send_request(struct rpc_task *task) ...@@ -693,6 +699,7 @@ static int xs_tcp_send_request(struct rpc_task *task)
struct rpc_xprt *xprt = req->rq_xprt; struct rpc_xprt *xprt = req->rq_xprt;
struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
struct xdr_buf *xdr = &req->rq_snd_buf; struct xdr_buf *xdr = &req->rq_snd_buf;
bool zerocopy = true;
int status; int status;
xs_encode_stream_record_marker(&req->rq_snd_buf); xs_encode_stream_record_marker(&req->rq_snd_buf);
...@@ -700,13 +707,20 @@ static int xs_tcp_send_request(struct rpc_task *task) ...@@ -700,13 +707,20 @@ static int xs_tcp_send_request(struct rpc_task *task)
xs_pktdump("packet data:", xs_pktdump("packet data:",
req->rq_svec->iov_base, req->rq_svec->iov_base,
req->rq_svec->iov_len); req->rq_svec->iov_len);
/* Don't use zero copy if this is a resend. If the RPC call
* completes while the socket holds a reference to the pages,
* then we may end up resending corrupted data.
*/
if (task->tk_flags & RPC_TASK_SENT)
zerocopy = false;
/* Continue transmitting the packet/record. We must be careful /* Continue transmitting the packet/record. We must be careful
* to cope with writespace callbacks arriving _after_ we have * to cope with writespace callbacks arriving _after_ we have
* called sendmsg(). */ * called sendmsg(). */
while (1) { while (1) {
status = xs_sendpages(transport->sock, status = xs_sendpages(transport->sock,
NULL, 0, xdr, req->rq_bytes_sent); NULL, 0, xdr, req->rq_bytes_sent,
zerocopy);
dprintk("RPC: xs_tcp_send_request(%u) = %d\n", dprintk("RPC: xs_tcp_send_request(%u) = %d\n",
xdr->len - req->rq_bytes_sent, status); xdr->len - req->rq_bytes_sent, status);
......
Markdown is supported
0%
or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment