Linux splice() returning EINVAL (“Invalid argument

问题:

I'm trying to experiment with using splice (man 2 splice) to copy data from a UDP socket directly to a file. Unfortunately the first call to splice() returns EINVAL.

The man page states:

EINVAL Target file system doesn't support splicing; target file is opened in
       append mode; neither of the descriptors refers to a pipe; or offset
       given for nonseekable device.

However, I believe none of those conditions apply. I'm using Fedora 15 (kernel 2.6.40-4) so I believe splice() is supported on all filesystems. The target file should be irrelevant in the first call to splice, but for completeness I'm opening it via open(path, O_CREAT | O_WRONLY | O_TRUNC, S_IRUSR | S_IWUSR). Both calls use a pipe and neither call uses an offset besides NULL.

Here's my sample code:

int sz = splice(sock_fd, 0, mPipeFds[1], 0, 8192, SPLICE_F_MORE);
if (-1 == sz)
{
int err = errno;
LOG4CXX_ERROR(spLogger, "splice from: " << strerror(err));
return 0;
}

sz = splice(mPipeFds[0], 0, file_fd, 0, sz, SPLICE_F_MORE);
if (-1 == sz)
{
int err = errno;
LOG4CXX_ERROR(spLogger, "splice to: " << strerror(err));
}

return 0;

sock_fd is initialized by the following psuedocode:

int sock_fd = socket(AF_INET, SOCK_DGRAM, IPPROTO_UDP);
setsockopt(sock_fd, SOL_SOCKET, SO_REUSEADDR, &one, sizeof(one));
fcntl(sock_fd, F_SETFL, flags | O_NONBLOCK);
bind(sock_fd, ...);

Possibly related is that this code snippet is running inside a libevent loop. libevent is using epoll() to determine if the UDP socket is hot.

回答1:

Found my answer. tl;dr - UDP isn't supported on the inbound side.

After enough Googling I stumbled upon a forum discussion and some test code which prints out a table of in/out fd types and their support:

$ ./a.out 
in\out     pipe    reg     chr     unix    tcp    udp
pipe       yes     yes     yes     yes     yes    yes
reg        yes     no      no      no      no     no
chr        yes     no      no      no      no     no
unix       no      no      no      no      no     no
tcp        yes     no      no      no      no     no
udp        no      no      no      no      no     no

回答2:

Yeah, it is definitely not supported for reading from a UDP socket, even in the latest kernels. References to the kernel source follow.

splice invokes do_splice in the kernel, which calls do_splice_to, which calls the splice_read member in the file_operations structure for the file.

For sockets, that structure is defined as socket_file_ops in net/socket.c, which initializes the splice_read field to sock_splice_read.

That function, in turn, contains this line of code:

if (unlikely(!sock->ops->splice_read))
    return -EINVAL;

The ops field of the socket is a struct proto_ops. For an IPv4 UDP socket, it is initialized to inet_dgram_ops in net/ipv4/af_inet.c. Finally, that structure does not explicitly initialize the splice_read field of struct proto_ops; i.e., it initializes it to zero.

So sock_splice_read returns -EINVAL, and that propagates up.