From: Adhemerval Zanella <adhemerval.zanella@linaro.org>
To: libc-alpha@sourceware.org, Florian Weimer <fweimer@redhat.com>
Subject: Re: [PATCH] Linux: Simplify __opensock and fix race condition [BZ #28353]
Date: Mon, 27 Sep 2021 12:33:34 -0300 [thread overview]
Message-ID: <48158338-3d06-72aa-5e1b-0cb34e7325cc@linaro.org> (raw)
In-Reply-To: <875yumghzs.fsf@oldenburg.str.redhat.com>
On 27/09/2021 12:23, Florian Weimer via Libc-alpha wrote:
> AF_NETLINK support is not quite optional on modern Linux systems
> anymore, so it is likely that the first attempt will always succeed.
> Consequently, there is no need to cache the result. Keep AF_UNIX
> and the Internet address families as a fallback, for the rare case
> that AF_NETLINK is fixing. The other address families previously
> probed are totally obsolete be now, so remove them.
>
> ---
> Tested on i686-linux-gnu and x86_64-linux-gnu (albeit with all four
> address families present). Initial build-many-glibcs.py results look
> okay as well.
>
> sysdeps/unix/sysv/linux/opensock.c | 103 ++++++-------------------------------
> 1 file changed, 16 insertions(+), 87 deletions(-)
>
> diff --git a/sysdeps/unix/sysv/linux/opensock.c b/sysdeps/unix/sysv/linux/opensock.c
> index e87d6e58b0..47464d3a30 100644
> --- a/sysdeps/unix/sysv/linux/opensock.c
> +++ b/sysdeps/unix/sysv/linux/opensock.c
> @@ -15,11 +15,7 @@
> License along with the GNU C Library; if not, see
> <https://www.gnu.org/licenses/>. */
>
> -#include <assert.h>
> #include <errno.h>
> -#include <stdio.h>
> -#include <string.h>
> -#include <unistd.h>
> #include <sys/socket.h>
>
> /* Return a socket of any type. The socket can be used in subsequent
> @@ -27,88 +23,21 @@
> int
> __opensock (void)
> {
> - static int last_family; /* Available socket family we will use. */
> - static int last_type;
> - static const struct
> - {
> - int family;
> - const char procname[15];
> - } afs[] =
> - {
> - { AF_UNIX, "net/unix" },
> - { AF_INET, "" },
> - { AF_INET6, "net/if_inet6" },
> - { AF_AX25, "net/ax25" },
> - { AF_NETROM, "net/nr" },
> - { AF_ROSE, "net/rose" },
> - { AF_IPX, "net/ipx" },
> - { AF_APPLETALK, "net/appletalk" },
> - { AF_ECONET, "sys/net/econet" },
> - { AF_ASH, "sys/net/ash" },
> - { AF_X25, "net/x25" },
> -#ifdef NEED_AF_IUCV
> - { AF_IUCV, "net/iucv" }
> -#endif
With this removal, there is no need to keep the s390 [1].
[1] sysdeps/unix/sysv/linux/s390/opensock.c
> - };
> -#define nafs (sizeof (afs) / sizeof (afs[0]))
> - char fname[sizeof "/proc/" + 14];
> - int result;
> - int has_proc;
> - size_t cnt;
> -
> - /* We already know which family to use from the last call. Use it
> - again. */
> - if (last_family != 0)
> - {
> - assert (last_type != 0);
> -
> - result = __socket (last_family, last_type | SOCK_CLOEXEC, 0);
> - if (result != -1 || errno != EAFNOSUPPORT)
> - /* Maybe the socket type isn't supported anymore (module is
> - unloaded). In this case again try to find the type. */
> - return result;
> -
> - /* Reset the values. They seem not valid anymore. */
> - last_family = 0;
> - last_type = 0;
> - }
> -
> - /* Check whether the /proc filesystem is available. */
> - has_proc = __access ("/proc/net", R_OK) != -1;
> - strcpy (fname, "/proc/");
> -
> - /* Iterate over the interface families and find one which is
> - available. */
> - for (cnt = 0; cnt < nafs; ++cnt)
> - {
> - int type = SOCK_DGRAM;
> -
> - if (has_proc && afs[cnt].procname[0] != '\0')
> - {
> - strcpy (fname + 6, afs[cnt].procname);
> - if (__access (fname, R_OK) == -1)
> - /* The /proc entry is not available. I.e., we cannot
> - create a socket of this type (without loading the
> - module). Don't look for it since this might trigger
> - loading the module. */
> - continue;
> - }
> -
> - if (afs[cnt].family == AF_NETROM || afs[cnt].family == AF_X25)
> - type = SOCK_SEQPACKET;
> -
> - result = __socket (afs[cnt].family, type | SOCK_CLOEXEC, 0);
> - if (result != -1)
> - {
> - /* Found an available family. */
> - last_type = type;
> - last_family = afs[cnt].family;
> - return result;
> - }
> - }
> -
> - /* None of the protocol families is available. It is unclear what kind
> - of error is returned. ENOENT seems like a reasonable choice. */
> + /* SOCK_DGRAM is supported by all address families. (Netlink does
> + not support SOCK_STREAM.) */
> + int type = SOCK_DGRAM | SOCK_CLOEXEC;
> + int fd = __socket (AF_NETLINK, type, 0);
> + if (fd >= 0)
> + return fd;
> + fd = __socket (AF_UNIX, type, 0);
> + if (fd >= 0)
> + return fd;
> + fd = __socket (AF_INET, type, 0);
> + if (fd >= 0)
> + return fd;
> + fd = __socket (AF_INET6, type, 0);
> + if (fd >= 0)
> + return fd;
> __set_errno (ENOENT);
> - return -1;
> + return fd;
> }
>
Wouldn't be better to move it as the generic implementation (and also cleanup
it by removing unsualy families like IPX, AX25, and APPLETALK)?
next prev parent reply other threads:[~2021-09-27 15:33 UTC|newest]
Thread overview: 8+ messages / expand[flat|nested] mbox.gz Atom feed top
2021-09-27 15:23 Florian Weimer
2021-09-27 15:33 ` Adhemerval Zanella [this message]
2021-09-27 16:10 ` Florian Weimer
2021-09-27 16:33 ` Adhemerval Zanella
2021-09-27 16:41 ` Florian Weimer
2021-09-27 18:29 Florian Weimer
2021-09-27 18:45 ` Andreas Schwab
2021-09-27 18:52 ` Florian Weimer
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=48158338-3d06-72aa-5e1b-0cb34e7325cc@linaro.org \
--to=adhemerval.zanella@linaro.org \
--cc=fweimer@redhat.com \
--cc=libc-alpha@sourceware.org \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for read-only IMAP folder(s) and NNTP newsgroup(s).