[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Date: Thu, 9 Feb 2023 19:26:01 -0500
From: Rich Felker <dalias@...c.org>
To: Colin Cross <ccross@...gle.com>
Cc: musl@...ts.openwall.com
Subject: Re: Increase sendmsg internal buffer to match kernel
SCM_MAX_FD limit
On Thu, Feb 09, 2023 at 03:08:50PM -0800, Colin Cross wrote:
> I came across a test at
> https://cs.android.com/android/platform/superproject/+/master:frameworks/native/libs/binder/tests/binderRpcTest.cpp;l=954;drc=68a556190553a4060babf4a4e5cb1bb16ae61ab2
> that verifies that some fd passing code can handle passing SCM_MAX_FD
> fds through a unix socket. SCM_MAX_FD is an arbitrary 253 fd limit
> imposed by the kernel since 2.6.38 (before that it was 255). An
> SCM_RIGHTS ancillary message contiang 253 fds is only slightly larger
> than the existing 1024 byte internal buffer in sendmsg, so this patch
> slightly increases the arbitrary limit in musl to match an arbitrary
> limit in the kernel.
> From 4a9c1a5b14fddd3924561e9cc5d126111ea881c4 Mon Sep 17 00:00:00 2001
> From: Colin Cross <ccross@...roid.com>
> Date: Thu, 9 Feb 2023 14:50:49 -0800
> Subject: [PATCH] Increase sendmsg internal buffer to support SCM_MAX_FD
>
> The kernel defines a limit on the number of fds that can be passed
> through an SCM_RIGHTS ancillary message as SCM_MAX_FD. The value
> has been 253 since kernel 2.6.38 (before that it was 255). On x86_64,
> and SCM_RIGHTS ancillary message with 253 fds requires 1032 bytes,
> slightly more than the current 1024 byte internal buffer in sendmsg.
> 1024 is an arbitrary size, so increase it to match the the arbitrary
> size limit in the kernel. This fixes tests that are verifying they
> support up to SCM_MAX_FD fds.
> ---
> src/network/sendmsg.c | 7 +++++--
> 1 file changed, 5 insertions(+), 2 deletions(-)
>
> diff --git a/src/network/sendmsg.c b/src/network/sendmsg.c
> index 80cc5f41..b5ce6629 100644
> --- a/src/network/sendmsg.c
> +++ b/src/network/sendmsg.c
> @@ -8,13 +8,16 @@ ssize_t sendmsg(int fd, const struct msghdr *msg, int flags)
> {
> #if LONG_MAX > INT_MAX
> struct msghdr h;
> - struct cmsghdr chbuf[1024/sizeof(struct cmsghdr)+1], *c;
> + /* Kernels since 2.6.38 set SCM_MAX_FD to 253, allocate enough
> + * space to support an SCM_RIGHTS ancillary message with 253 fds. */
> + const size_t chbufsize = CMSG_SPACE(253*sizeof(int));
> + struct cmsghdr chbuf[chbufsize/sizeof(struct cmsghdr)+1], *c;
> if (msg) {
> h = *msg;
> h.__pad1 = h.__pad2 = 0;
> msg = &h;
> if (h.msg_controllen) {
> - if (h.msg_controllen > 1024) {
> + if (h.msg_controllen > chbufsize) {
> errno = ENOMEM;
> return -1;
> }
> --
> 2.39.1.581.gbfd45094c4-goog
>
The concept of this seems fine, but if the limit was previously 255 on
supported kernel versions, why stop at 253? It doesn't really cost
anything to go up large enough that the 255 would work too.
As a technical detail, I'd probably also just put the full size
expression in the [], then use sizeof when you need it later. As
written, this change makes chbuf[] formally a VLA. The compiler
probably optimizes it to the same code as if it wasn't a VLA, but
there's no good reason for it to be a VLA.
Rich
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.
