GitLab

ncsi: Add Mellanox Get MAC Address handler

See merge request !125

Attempted to mirror the upstream Linux driver[1] as closely as reasonably
possible.

[1] https://github.com/torvalds/linux/blob/42226c989789d8da4af1de0c31070c96726d990c/net/ncsi/ncsi-rsp.c#L614-L638

Signed-off-by: Peter Delevoryas <pdel@fb.com>

If a network card supports NC-SI, then it redirects all traffic with the
out-of-band (OOB) management controller's (MC) ethernet address to the
out-of-band management controller, usually over some sideband RMII
interface, not like the PCIe connection to the main host.

It's also pretty common for the network card to provision the out-of-band
management controller's ethernet address. At startup, the OOB MC asks the
network card what its MAC address is through OEM NC-SI commands. This
protocol is so common that it's going to be standardized soon in NC-SI 1.2.0
[1] as "Get MC MAC Address".

Note: At some point, the network card may provision *multiple* OOB ethernet
addresses, but right now everything just uses one.

[1] https://www.dmtf.org/sites/default/files/standards/documents/DSP0222_1.2.0WIP80.pdf

Signed-off-by: Peter Delevoryas <pdel@fb.com>

This commit just sets up the OEM command handler to respond with
"unsupported" for now, as verified in the test.
Signed-off-by: Peter Delevoryas <pdel@fb.com>

Signed-off-by: Peter Delevoryas <pdel@fb.com>

In the Linux NC-SI driver[1], each response's length is validated with a
statically declared payload length, _unless_ it's an OEM command or some
more complicated NC-SI packet that you can't determine the length of just
from the "type" field, in which case it just uses the length provided by the
response's header.

To support OEM response handlers without requiring too many modifications we
can make the default payload length use the value specified in the handler
table, and then allow OEM handlers to override the length by modifying the
"length" in the response header within the handler implementation.

[1] https://github.com/torvalds/linux/blob/ec7f49619d8ee13e108740c82f942cd401b989e9/net/ncsi/ncsi-rsp.c#L1215-L1220

Signed-off-by: Peter Delevoryas <pdel@fb.com>

This change passes the command header as an additional read-only parameter
to each response handler so that they can make more response handling
descisions based on the command header fields. This is especially useful for
handling OEM NC-SI commands, or any protocol that's encapsulated in an NC-SI
header.
Signed-off-by: Peter Delevoryas <pdel@fb.com>

Thomas Huth authored May 16, 2022 and Samuel Thibault committed May 17, 2022

Windows Vista is not supported by its vendor anymore. Additionally,
glib uses 0x0601 as setting for _WIN32_WINNT since version 2.53.6
already, so unless libslirp is used with a very old version of glib,
we are depending on Windows 7 anyway.
Signed-off-by: Thomas Huth <thuth@redhat.com>
Reviewed-by: Samuel Thibault <samuel.thibault@ens-lyon.org>
Message-Id: <20220516090410.39727-1-thuth@redhat.com>

ncsi: Add basic Get Version ID response

See merge request !122

Get Version ID is one of the first commands used in NC-SI, because BMC's
use a lot of OEM NC-SI extensions, and you need to query the device's
manufacturer through Get Version ID before you can decide which OEM
NC-SI extensions to use.

The response format is documented in the NC-SI spec[1]. We're just
setting the NC-SI version supported to 1.0.0 (BCD-encoded[2]) and
returning the manufacturer's ID in network byte-order.

[1] https://www.dmtf.org/sites/default/files/standards/documents/DSP0222_1.0.0.pdf
[2] https://en.wikipedia.org/wiki/Binary-coded_decimal

Signed-off-by: Peter Delevoryas <pdel@fb.com>

This will let us use Slirp fields to generate more interesting NC-SI
responses.
Signed-off-by: Peter Delevoryas <pdel@fb.com>

The manufacturer's ID is used in NC-SI commands such as "Get Version ID"
[1]. It is also essential to providing a path towards adding OEM
(non-standardized) NC-SI commands.

This field should be derived from the IANA Private Enterprise Numbers list,
per the NC-SI specification. It may be useful for things besides NC-SI, but
NC-SI responses for BMC's in QEMU are the main use case I have in mind.

Note: I did not add this attribute to slirp_init, since it is deprecated.

[1] https://www.dmtf.org/sites/default/files/standards/documents/DSP0222_1.0.0.pdf
[2] https://www.iana.org/assignments/enterprise-numbers/enterprise-numbers

Signed-off-by: Peter Delevoryas <pdel@fb.com>

Add support for Haiku to meson.build

See merge request !123

win32: declare some local functions as static

See merge request !121

Enable extra warnings based no what QEMU uses.
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>

These functions do not have a prototype and are not meant to be public.
Declare them as static.
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>

Include <sys/socket.h> and <arpa/inet.h> for AF_INET6 and inet_pton

Closes #58

See merge request !120

Fixes #58

Release v4.7.0

See merge request !119

Signed-off-by: Marc-André Lureau <marcandre.lureau@redhat.com>

Support for CFI

See merge request !117

Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>

The introduction of .timer_new_opaque adds an interesting conundrum.
The Slirp* needs to be stored in .timer_new_opaque so that it can be
passed back to slirp_handle_timer, but it is not returned by slirp_new
and slirp_init until after the first call to .timer_new_opaque (which
is in ip6_init).  This is a problem for programs that, like QEMU, use
more than one Slirp*.

Fix them by passing the Slirp* to a callback before slirp_new returns,
and initializing the timer afterwards.
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>

Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>

Add a new callback that supports CFI better by avoiding having function pointers
in the external libslirp API.  Instead, the new API passes an opaque integer
and requests the application to call a new libslirp function when the timer
expires.
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>

In order to support CFI, we would like to avoid passing function pointers out of
libslirp, and instead use opaque integers for the timer callbacks.  This patch
introduces an internal API for creating a timer that is based on such a timer
callback.

For now, it receives the id (SLIRP_TIMER_RA is the only one) and calls
slirp->cb->timer_new.  The prototype of ra_timer_handler is changed slightly
to fit better with the next patch which introduces a new libslirp function,
to be called on timer expiration.
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>

Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>

Handle ECONNABORTED from recv

See merge request !116

recv can return ECONNABORTED due to a time-out on the socket. The socket is no longer useable once it enters this state.

bootp: fix g_str_has_prefix warning/critical

See merge request !115

Fixes: commit 60535013

 ("bootp: add support for UEFI HTTP boot")
Signed-off-by: Marc-André Lureau <marcandre.lureau@redhat.com>

Brett Nash authored Feb 10, 2022 and Samuel Thibault committed Feb 11, 2022

When debug is enabled, we duplicate the packet in tcp_reass, but ti is still
pointing to the old buffer, resulting in a use after free.

This makes valgrind debugging a little trickier, but makes it
crash a lot less.
Signed-off-by: Brett Nash <nash@fb.com>

Brett Nash authored Feb 10, 2022 and Samuel Thibault committed Feb 11, 2022

Avoid a macro translation when not needed, and potentially doing really
weird things to headers.

mbuf: Use SLIRP_DEBUG to enable mbuf debugging instead of DEBUG

See merge request !114

DEBUG is set by a number of IDEs and development environments (*cough*
xcode *cough*).  This means we use mbuf duplication when we don't need
(or expect) it.  Change the name to SLIRP_DEBUG to enable this feature.
Signed-off-by: Pablo Fiori <pablofiori@fb.com>
Signed-off-by: Brett Nash <nash@fb.com>     

inet_ntoa() returns a static pointer which is subject to safety issues.
Use the recommended alternative.
Signed-off-by: Marc-André Lureau <marcandre.lureau@redhat.com>

Add VMS_END marker

See merge request !112

The migration VMSTATE_END_OF_LIST marker relies on a NULL field
to mark th eend of a list;  that's a bit delicate when the marker is
forgotten and the code hits a NULL by luck.

Use an explicit flag to mark the end as well.

See: https://lists.gnu.org/archive/html/qemu-devel/2022-01/msg03025.html

Signed-off-by: Dr. David Alan Gilbert <dgilbert@redhat.com>