Eric Dumazet a écrit :

quoted text
> David Miller a écrit :

>> I'm not applying this until someone explains to me why

>> we should remove this test from the splice receive but

>> keep it in the tcp_recvmsg() code where it has been

>> essentially forever.

Reading again tcp_recvmsg(), I found it already is able to eat several skb

even in nonblocking mode.
setsockopt(5, SOL_SOCKET, SO_RCVLOWAT, [61440], 4) = 0

ioctl(5, FIONBIO, [1])                  = 0

poll([{fd=5, events=POLLIN, revents=POLLIN}], 1, -1) = 1

recv(5, "\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0"..., 65536, MSG_DONTWAIT) = 65536

write(3, "\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0"..., 65536) = 65536

poll([{fd=5, events=POLLIN, revents=POLLIN}], 1, -1) = 1

recv(5, "\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0"..., 65536, MSG_DONTWAIT) = 65536

write(3, "\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0"..., 65536) = 65536

poll([{fd=5, events=POLLIN, revents=POLLIN}], 1, -1) = 1

recv(5, "\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0"..., 65536, MSG_DONTWAIT) = 65536

write(3, "\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0"..., 65536) = 65536

poll([{fd=5, events=POLLIN, revents=POLLIN}], 1, -1) = 1

recv(5, "\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0"..., 65536, MSG_DONTWAIT) = 65536

write(3, "\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0"..., 65536) = 65536
David, if you referred to code at line 1374 of net/ipv4/tcp.c, I believe there is

no issue with it. We really want to break from this loop if !timeo .
Willy patch makes splice() behaving like tcp_recvmsg(), but we might call

tcp_cleanup_rbuf() several times, with copied=1460 (for each frame processed)
I wonder if the right fix should be done in tcp_read_sock() : this is the

one who should eat several skbs IMHO, if we want optimal ACK generation.
We break out of its loop at line 1246
if (!desc->count) /* this test is always true */

	break;
(__tcp_splice_read() set count to 0, right before calling tcp_read_sock())
So code at line 1246 (tcp_read_sock()) seems wrong, or pessimistic at least.
--

To unsubscribe from this list: send the line "unsubscribe netdev" in

the body of a message to majordomo@vger.kernel.org

More majordomo info at  http://vger.kernel.org/majordomo-info.html