On Sat, Feb 18, 2017 at 12:02:14AM +0000, Al Viro wrote:
> On Fri, Feb 17, 2017 at 05:03:15PM +0000, Al Viro wrote:
> > On Fri, Feb 17, 2017 at 10:54:20AM -0500, David Miller wrote:
> > > From: Al Viro <v...@zeniv.linux.org.uk>
> > > Date: Tue, 14 Feb 2017 01:33:06 +0000
> > >
> > > > OK... Remaining interesting question is whether it adds a noticable
> > > > overhead. Could somebody try it on assorted benchmarks and see if
> > > > it slows the things down? The patch in question follows:
> > >
> > > That's about a 40 byte copy onto the stack for each invocation of this
> > > thing. You can benchmark all you want, but it's clear that this is
> > > non-trivial amount of work and will take some operations from fitting
> > > in the cache to not doing so for sure.
> >
> > In principle, that could be reduced a bit (32 bytes - ->type is never
> > changed, so we don't need to restore it), but that's not much of
> > improvement...
>
> Actually, I've a better solution. Namely, analogue of iov_iter_advance()
> for going backwards. The restriction is that you should never unroll
> further than where you've initially started *or* have the iovec, etc.
> array modified under you. For iovec/kvec/bio_vec it's trivial, for pipe -
> a bit more convoluted, but still doable. Then net/core/datagram.c stuff
> could simply use iov_iter_unroll() in case of error - all we need to keep
> track of is how much had we copied and that's easy to do.
>
> The patch below is completely untested, but if it works it should avoid
> buggering the fast paths at all, still giving the same semantics re
> reverting ->msg_iter both on EINVAL and EFAULT. Comments?
Incidentally, at least 4 other places in net/* can stop messing with
copying iov_iter just in case we'll need to revert and I'm fairly sure
there's more of the same in the tree. FWIW, net/rds and net/tipc parts
(on top of the previous patch, also completely untested) follow:
diff --git a/net/rds/recv.c b/net/rds/recv.c
index 9d0666e5fe35..ed407156b3b4 100644
--- a/net/rds/recv.c
+++ b/net/rds/recv.c
@@ -564,7 +564,6 @@ int rds_recvmsg(struct socket *sock, struct msghdr *msg,
size_t size,
goto out;
while (1) {
- struct iov_iter save;
/* If there are pending notifications, do those - and nothing
else */
if (!list_empty(&rs->rs_notify_queue)) {
ret = rds_notify_queue_get(rs, msg);
@@ -600,7 +599,6 @@ int rds_recvmsg(struct socket *sock, struct msghdr *msg,
size_t size,
rdsdebug("copying inc %p from %pI4:%u to user\n", inc,
&inc->i_conn->c_faddr,
ntohs(inc->i_hdr.h_sport));
- save = msg->msg_iter;
ret = inc->i_conn->c_trans->inc_copy_to_user(inc,
&msg->msg_iter);
if (ret < 0)
break;
@@ -614,7 +612,7 @@ int rds_recvmsg(struct socket *sock, struct msghdr *msg,
size_t size,
rds_inc_put(inc);
inc = NULL;
rds_stats_inc(s_recv_deliver_raced);
- msg->msg_iter = save;
+ iov_iter_unroll(&msg->msg_iter, ret);
continue;
}
diff --git a/net/tipc/msg.c b/net/tipc/msg.c
index a22be502f1bd..5bfd5d641ba2 100644
--- a/net/tipc/msg.c
+++ b/net/tipc/msg.c
@@ -316,6 +316,7 @@ int tipc_msg_build(struct tipc_msg *mhdr, struct msghdr *m,
skb = tipc_buf_acquire(pktsz);
if (!skb) {
rc = -ENOMEM;
+ iov_iter_unroll(&m->msg_iter, dsz - drem);
goto error;
}
skb_orphan(skb);
diff --git a/net/tipc/socket.c b/net/tipc/socket.c
index 333c5dae0072..1fa220a2fa00 100644
--- a/net/tipc/socket.c
+++ b/net/tipc/socket.c
@@ -688,7 +688,6 @@ static int tipc_sendmcast(struct socket *sock, struct
tipc_name_seq *seq,
struct net *net = sock_net(sk);
struct tipc_msg *mhdr = &tsk->phdr;
struct sk_buff_head pktchain;
- struct iov_iter save = msg->msg_iter;
uint mtu;
int rc;
@@ -725,7 +724,7 @@ static int tipc_sendmcast(struct socket *sock, struct
tipc_name_seq *seq,
}
__skb_queue_purge(&pktchain);
if (rc == -EMSGSIZE) {
- msg->msg_iter = save;
+ iov_iter_unroll(&msg->msg_iter, dsz);
goto new_mtu;
}
break;
@@ -891,7 +890,7 @@ static int __tipc_sendmsg(struct socket *sock, struct
msghdr *m, size_t dsz)
bool is_connectionless = tipc_sk_type_connectionless(sk);
struct sk_buff *skb;
struct tipc_name_seq *seq;
- struct iov_iter save;
+ int size;
u32 mtu;
long timeo;
int rc;
@@ -950,10 +949,9 @@ static int __tipc_sendmsg(struct socket *sock, struct
msghdr *m, size_t dsz)
}
skb_queue_head_init(&pktchain);
- save = m->msg_iter;
new_mtu:
mtu = tipc_node_get_mtu(net, dnode, tsk->portid);
- rc = tipc_msg_build(mhdr, m, 0, dsz, mtu, &pktchain);
+ size = rc = tipc_msg_build(mhdr, m, 0, dsz, mtu, &pktchain);
if (rc < 0)
return rc;
@@ -974,7 +972,7 @@ static int __tipc_sendmsg(struct socket *sock, struct
msghdr *m, size_t dsz)
}
__skb_queue_purge(&pktchain);
if (rc == -EMSGSIZE) {
- m->msg_iter = save;
+ iov_iter_unroll(&m->msg_iter, size);
goto new_mtu;
}
break;
@@ -1049,7 +1047,7 @@ static int __tipc_send_stream(struct socket *sock, struct
msghdr *m, size_t dsz)
long timeo;
u32 dnode;
uint mtu, send, sent = 0;
- struct iov_iter save;
+ int size;
int hlen = MIN_H_SIZE;
/* Handle implied connection establishment */
@@ -1078,10 +1076,9 @@ static int __tipc_send_stream(struct socket *sock,
struct msghdr *m, size_t dsz)
skb_queue_head_init(&pktchain);
next:
- save = m->msg_iter;
mtu = tsk->max_pkt;
send = min_t(uint, dsz - sent, TIPC_MAX_USER_MSG_SIZE);
- rc = tipc_msg_build(mhdr, m, sent, send, mtu, &pktchain);
+ size = rc = tipc_msg_build(mhdr, m, sent, send, mtu, &pktchain);
if (unlikely(rc < 0))
return rc;
@@ -1099,7 +1096,7 @@ static int __tipc_send_stream(struct socket *sock, struct
msghdr *m, size_t dsz)
__skb_queue_purge(&pktchain);
tsk->max_pkt = tipc_node_get_mtu(net, dnode,
portid);
- m->msg_iter = save;
+ iov_iter_unroll(&m->msg_iter, size);
goto next;
}
if (rc != -ELINKCONG)
