Hi Stephan,
>>
This brings me to another proposal for read buffer sizing: AF_ALG akcipher
can guarantee that partial reads (where the read buffer is shorter than
the output of the crypto op) will work using the same semantics as
SOCK_DGRAM/SOCK_SEQPACKET. With those sockets, as much data as will fit is
copied in to the read buffer and the remainder is discarded.
I realize there's a performance and memory tradeoff, since the crypto
algorithm needs a sufficiently large output buffer that would have to be
created by AF_ALG akcipher. The user could manage that tradeoff by
providing a larger buffer (typically key_size?) if it wants to avoid
allocating and copying intermediate buffers inside the kernel.
How shall the user know that something got truncated or that the kernel
created memory?
To the former point, recall the signature of recv:
ssize_t recv(int sockfd, void *buf, size_t len, int flags);
Traditionally, userspace apps can know that the buffer provided to recv
was too small in two ways:
The return value from recv / recvmsg was >= len.
In the case of recvmsg, the MSG_TRUNC flag is set.
To quote man recv:
"All three calls return the length of the message on successful comple‐
tion. If a message is too long to fit in the supplied buffer, excess
bytes may be discarded depending on the type of socket the message is
received from."
and
"MSG_TRUNC (since Linux 2.2)
For raw (AF_PACKET), Internet datagram (since Linux
2.4.27/2.6.8), netlink (since Linux 2.6.22), and UNIX datagram
(since Linux 3.4) sockets: return the real length of the packet
or datagram, even when it was longer than the passed buffer.
"
Regards,
-Denis
--
To unsubscribe from this list: send the line "unsubscribe linux-api" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
