Follow @Openwall on Twitter for new release announcements and other news
[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Date: Sun,  3 Oct 2021 10:14:59 +0200 (CEST)
From: "J. Hanne" <>
Subject: Re: CMSG_LEN macro


thanks for your thoughts. The NLMSG macros also already gave me similar headache some time ago. I personally find both APIs counter-intuitive because they use the term "ALIGN", when they mean "PAD", so that the *following* item is aligned.

I would suggest the following patch now:

Do not use CMSG_ALIGN on struct cmsghdr, because:
- This has no effect on any architecture anyway, because sizeof(struct cmsghdr) == 16 on all archs
- Using it contradicts with CMSG_DATA, which does NOT apply any padding after struct cmsghdr
- This is consistent with the NLMSG_* macros
diff -uNr a/include/sys/socket.h b/include/sys/socket.h
--- a/include/sys/socket.h	2021-01-15 03:26:00.000000000 +0100
+++ b/include/sys/socket.h	2021-10-03 09:49:35.000000000 +0200
@@ -358,8 +358,8 @@
 #define CMSG_FIRSTHDR(mhdr) ((size_t) (mhdr)->msg_controllen >= sizeof (struct cmsghdr) ? (struct cmsghdr *) (mhdr)->msg_control : (struct cmsghdr *) 0)
 #define CMSG_ALIGN(len) (((len) + sizeof (size_t) - 1) & (size_t) ~(sizeof (size_t) - 1))
-#define CMSG_SPACE(len) (CMSG_ALIGN (len) + CMSG_ALIGN (sizeof (struct cmsghdr)))
-#define CMSG_LEN(len)   (CMSG_ALIGN (sizeof (struct cmsghdr)) + (len))
+#define CMSG_SPACE(len) (CMSG_ALIGN (len) + sizeof (struct cmsghdr))
+#define CMSG_LEN(len)   (sizeof (struct cmsghdr) + (len))
 #define SCM_RIGHTS      0x01
 #define SCM_CREDENTIALS 0x02

By the way, the question which led me to all this stuff is: How do I get the payload length of a received cmsg. Neither the man page nor an Internet search gave me any satisfactory answer. So my best guess was "do some arithmetic with CMSG_LEN":
payloadlen = cmsghdr->cmsg_len - CMSG_LEN(0);
However, when double-checking with musl source code, the CMSG_ALIGN on struct cmsghdr made me doubt my approach. Now, I will stick to it - as long as nobody else has a better idea?


Markus Wichmann schrieb am 05.09.2021 19:27 (GMT +02:00):
> On Fri, Sep 03, 2021 at 09:12:13AM -0400, Rich Felker wrote:
> > Anyone else have thoughts on this?
> >
> > Rich
> I noticed something similar about the NLMSG_* macros that allow for
> padding where there can be none (in the interface). struct nlmsghdr has
> alignment of 4, and the netlink message alignment is also 4, and that
> can never be changed on any existing arch since it would break binary
> compatibility. And for netlink, it is unlikely they would add
> architecture specific alignment in future, given that today it is
> arch-independent.
> I guess those are symptoms of overly general software design. The macros
> must exist, but I concur with your conclusion that they can be
> implemented without reference to CMSG_ALIGN.
> BTW, I just checked the implementation of the NLMSG_* macros in musl,
> and they do assume the alignment of struct nlmsghdr. So for consistency,
> we should probably do the same for the CMSG_* macros.
> Ciao,
> Markus

Powered by blists - more mailing lists

Confused about mailing lists and their use? Read about mailing lists on Wikipedia and check out these guidelines on proper formatting of your messages.