linux-kernel - Re: [PATCH] um: add a generic "fd" vector transport

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Message-ID: <CAMxuvaz17pVzSQ--y1_zXNFeM-MkH8L8ayiM34eCN0OkEF21WQ@mail.gmail.com>
Date:   Wed, 8 Apr 2020 19:07:42 +0200
From:   Marc-André Lureau <marcandre.lureau@...hat.com>
To:     Anton Ivanov <anton.ivanov@...bridgegreys.com>
Cc:     linux-um <linux-um@...ts.infradead.org>,
        joerd.simons@...labora.co.uk, richard@....at, jdike@...toit.com,
        Linux Kernel Mailing List <linux-kernel@...r.kernel.org>,
        alex.dewar@....co.uk
Subject: Re: [PATCH] um: add a generic "fd" vector transport

Hi

On Tue, Apr 7, 2020 at 11:02 PM Anton Ivanov
<anton.ivanov@...bridgegreys.com> wrote:
>
> On 07/04/2020 21:28, Marc-André Lureau wrote:
> > Learn to take a pre-opened file-descriptor for vector IO.
> >
> > Instead of teaching the driver to open a FD in multiple ways, it can
> > rely on management layer to do it on its behalf. For example, this
> > allows inheriting a preconfigured device fd or a simple socketpair()
> > setup, without further arguments, privileges or system access by UML.
> >
> > Signed-off-by: Marc-André Lureau <marcandre.lureau@...hat.com>
> > ---
> >   arch/um/drivers/vector_user.c | 59 +++++++++++++++++++++++++++++++++++
> >   1 file changed, 59 insertions(+)
> >
> > diff --git a/arch/um/drivers/vector_user.c b/arch/um/drivers/vector_user.c
> > index 29fae0456ade..45c1550dbb37 100644
> > --- a/arch/um/drivers/vector_user.c
> > +++ b/arch/um/drivers/vector_user.c
> > @@ -29,6 +29,7 @@
> >   #include <netdb.h>
> >   #include <stdlib.h>
> >   #include <os.h>
> > +#include <limits.h>
> >   #include <um_malloc.h>
> >   #include "vector_user.h"
> >
> > @@ -42,6 +43,9 @@
> >   #define TRANS_RAW "raw"
> >   #define TRANS_RAW_LEN strlen(TRANS_RAW)
> >
> > +#define TRANS_FD "fd"
> > +#define TRANS_FD_LEN strlen(TRANS_FD)
> > +
> >   #define VNET_HDR_FAIL "could not enable vnet headers on fd %d"
> >   #define TUN_GET_F_FAIL "tapraw: TUNGETFEATURES failed: %s"
> >   #define L2TPV3_BIND_FAIL "l2tpv3_open : could not bind socket err=%i"
> > @@ -347,6 +351,59 @@ static struct vector_fds *user_init_unix_fds(struct arglist *ifspec, int id)
> >       return NULL;
> >   }
> >
> > +static int strtofd(const char *nptr)
> > +{
> > +     long fd;
> > +     char *endptr;
> > +
> > +     if (nptr == NULL)
> > +             return -1;
> > +
> > +     errno = 0;
> > +     fd = strtol(nptr, &endptr, 10);
> > +     if (nptr == endptr ||
> > +             errno != 0 ||
> > +             *endptr != '\0' ||
> > +             fd < 0 ||
> > +             fd > INT_MAX) {
> > +             return -1;
> > +     }
> > +     return fd;
> > +}
> > +
> > +static struct vector_fds *user_init_fd_fds(struct arglist *ifspec)
> > +{
> > +     int fd = -1;
> > +     char *fdarg = NULL;
> > +     struct vector_fds *result = NULL;
> > +
> > +     fdarg = uml_vector_fetch_arg(ifspec, "fd");
> > +     fd = strtofd(fdarg);
> > +     if (fd == -1) {
> > +             printk(UM_KERN_ERR "fd open: bad or missing fd argument");
> > +             goto fd_cleanup;
> > +     }
> > +
> > +     result = uml_kmalloc(sizeof(struct vector_fds), UM_GFP_KERNEL);
> > +     if (result == NULL) {
> > +             printk(UM_KERN_ERR "fd open: allocation failed");
> > +             goto fd_cleanup;
> > +     }
> > +
> > +     result->rx_fd = fd;
> > +     result->tx_fd = fd;
> > +     result->remote_addr_size = 0;
> > +     result->remote_addr = NULL;
> > +     return result;
> > +
> > +fd_cleanup:
> > +     if (fd >= 0)
> > +             os_close_file(fd);
> > +     if (result != NULL)
> > +             kfree(result);
> > +     return NULL;
> > +}
> > +
> >   static struct vector_fds *user_init_raw_fds(struct arglist *ifspec)
> >   {
> >       int rxfd = -1, txfd = -1;
> > @@ -578,6 +635,8 @@ struct vector_fds *uml_vector_user_open(
> >               return user_init_socket_fds(parsed, ID_L2TPV3);
> >       if (strncmp(transport, TRANS_BESS, TRANS_BESS_LEN) == 0)
> >               return user_init_unix_fds(parsed, ID_BESS);
> > +     if (strncmp(transport, TRANS_FD, TRANS_FD_LEN) == 0)
> > +             return user_init_fd_fds(parsed);
> >       return NULL;
> >   }
> >
> >
>
> We should also control enable/disable of recv/sendmmsg as an option here.
>
> It can be made generic and be fed into get_transport_options() in
> vector_kern.c as an override.

So actually, there seems to be a way to do that already:

vec0:transport=fd,fd=N,vec=0

>
> --
> Anton R. Ivanov
> Cambridgegreys Limited. Registered in England. Company Number 10273661
> https://www.cambridgegreys.com/
>