671: fix socket feature check r=Dirbaio a=M1cha

it requires you to have at least one socket type enabled but the feature
`socket-dhcp` does not exist because it's name is `socket-dhcpv4`. So if
that's only socket type you want enabled you weren't able to do that due
to this broken check.

Co-authored-by: Michael Zimmermann <[email protected]>
Co-authored-by: Dario Nieuwenhuis <[email protected]>

Previously, error handling was performed in the closure and after the
closure as well. Now, error handling is performed in one place.

667: Change egress error handling r=Dirbaio a=thvdveld

Previously, error handling was performed in the closure and after the
closure as well. Now, error handling is performed in one place.

Co-authored-by: Thibaut Vandervelden <[email protected]>

RFC 6762 Section 5.1 specifies a one-shot multicast DNS query. This
query has minimal differences from a standard DNS query, mostly just
using a multicast address and a different port (5353 vs 53).

A fully standards compliant mDNS implementation would use UDP source
port 5353 as well to issue queries, however we MUST NOT use that port
and continue using an ephemeral port until features such as service
discovery are implemented.

This change also allows specifying what kind of DNS query we wish to
perform.

https://www.rfc-editor.org/rfc/rfc6762#section-5.1

Per RFC 2132 section 9.11 the server can manually specify a renewal (T1)
time different from the default of half the lease time through option
code 58. This PR updates the behavior of the dhcp client to use that
value, if provided, and only if not provided does it default to half of
the lease duration.

Since the current state of smoltcp does not seem to follow the REBINDING
state, I also made it look for a value in option code 59 (which should
be the rebinding (T2) interval) and use that if no T1 interval interval
is provided. This behavior seems sensible to me, given that we're not
following the REBINDING part of the spec, but I can change it to ignore
option code 59, or any other handling, if that is preferred.

683: Use renewal time from DHCP server ACK, if given r=Dirbaio a=JarredAllen

# Description

Per RFC 2132 section 9.11 the server can manually specify a renewal (T1) time different from the default value (half the lease time) through option code 58. This PR updates the behavior of the dhcp client to use that value, if provided, and only if not provided does it default to half of the lease duration.

Since smoltcp seems to ignore the REBINDING state, I also made it look for a provided rebinding (T2) time provided by the server (dhcp option 59) and made it use that value as the renewal time if no renewal time was provided and the rebinding time is less than the default. This behavior seems sensible to me, given that we're not following the REBINDING part of the spec, but I can change it to ignore option code 59, or any other handling, if that is preferred.

# Verification

I realized that this functionality was missing when I changed my configuration to set a 10 second renew time on a lease which lasts for a very long time, and observed that my devices (which use this library) weren't attempting to renew. To verify that this PR works, I ran it in my existing setup and confirmed that my devices now renew their leases at approximately 10 second intervals. I think this, alongside the tests in CI, should be good enough.

Co-authored-by: Jarred Allen <[email protected]>

684: Fix 6LoWPAN fragmentation r=thvdveld a=thvdveld



Co-authored-by: Thibaut Vandervelden <[email protected]>

Swaps the bool in the raw query API to an enum that can statically
prevent mdns usage without the feature flag enabled.

669: Adds one-shot mDNS resolution r=Dirbaio a=benbrittain

RFC 6762 Section 5.1 specifies a one-shot multicast DNS query. This
query has minimal differences from a standard DNS query, mostly just
using a multicast address and a different port (5353 vs 53).

A fully standards compliant mDNS implementation would use UDP source
port 5353 as well to issue queries, however we MUST NOT use that port
and continue using an ephemeral port until features such as service
discovery are implemented.

This change also allows specifying what kind of DNS query we wish to
perform.

https://www.rfc-editor.org/rfc/rfc6762#section-5.1

Co-authored-by: Benjamin Brittain <[email protected]>
Co-authored-by: Thibaut Vandervelden <[email protected]>

\# Description

If the user provides a buffer in which to store the packets, then the
contents of the received packet will be buffered and included in the
returned Config when the DHCP connection is made. However, it isn't
included in the Config struct until the value is returned to the user,
so the equality check for whether to call `config_changed` disregards
any additional information in the buffer beyond what this library
parses.

For my purposes, I need access to the contents of the buffer when they
change as a result of a new packet, even if everything else is the same.
Since two packets will almost certainly not be the same thanks to the
magic cookie (unless the packet gets duplicated on the network, which is
an acceptably low risk for my use of smoltcp), an acceptable option for
my uses is to just always return the new configuration when a packet is
received (gated on whether a buffer is provided to return the packet
into).

\# Alternatives

While this approach is the easiest and best for my uses, I can think of
the following alternatives which would also work and might be prefered
for other use-cases:
 * Allow the user to specify whether they wish to receive all packets
   instead of opting all users who provide a buffer into this behavior
 * Allow the user to provide a closure which compares the old and new
   packets and returns true if this represents a new config which should
   be returned.
 * Compare the old packet to the new packet (either byte-by-byte or
   looking at the provided options) and only return a new config if
   differences are found.
\# Verification

In my setup, I was seeing bugs that were caused by smoltcp not exposing
the config when the only changes were in the additional options that I
want to use but which smoltcp doesn't use directly. Using this branch
instead of the main release fixed those bugs and I was able to verify
that it behaves the way I expected. I think this verification, along
with CI tests passing, should be sufficient for verifying this PR.

686: Changes egress functions to pass up Err(Exhausted) r=thvdveld a=benbrittain

Currently the poll functions will return `Ok(true)` instead of `Err(Exhausted)` despite logging about the failed transmission

Co-authored-by: Benjamin Brittain <[email protected]>
Co-authored-by: Thibaut Vandervelden <[email protected]>

687: Add address context information for resolving 6LoWPAN addresses r=thvdveld a=thvdveld

Implements [3.1.2. Context Identifier Extension](https://www.rfc-editor.org/rfc/rfc6282#section-3.1.2)

Co-authored-by: Thibaut Vandervelden <[email protected]>

681: Refactor 6LoWPAN function r=thvdveld a=thvdveld

Now, handling UDP packets is done using `process_udp`. Also removed `cfg_if!` in `interface` since that messed with formatting of the code.

It also contains a fix for the fragmentation.

Co-authored-by: Thibaut Vandervelden <[email protected]>

690: Increase version number r=thvdveld a=thvdveld

This makes testing local changes with patches work in other crates.

Co-authored-by: Thibaut Vandervelden <[email protected]>

691: Make Public Ipv6RoutingType r=Dirbaio a=ngc0202

Of the four `Type` enums, for some reason the one from `ipv6routing` was missing. This pull request makes it public under the name `Ipv6RoutingType`

Co-authored-by: Nicholas Cyprus <[email protected]>

692: use same version number as released r=thvdveld a=thvdveld

This reverts the bump to `0.9.0` and just makes the version the same as the released one.

Co-authored-by: Thibaut Vandervelden <[email protected]>

699: Fix missing `#[test]` r=Dirbaio a=Dirbaio

fixes #696 

bors r+

Co-authored-by: Dario Nieuwenhuis <[email protected]>

698: Split interface into multiple files r=Dirbaio a=thvdveld

Currently, `src/iface/interface.rs` is 5178 lines. I started splitting the interface because I think it was becoming to big when working on RPL. I think splitting it the way I did is currently most logical, however, I'm sure I missed some things. What do you all think?

Co-authored-by: Thibaut Vandervelden <[email protected]>

693: Constification of the wire module r=Dirbaio a=ngc0202

A huge portion of the `wire` module is constifyable, as shown in this commit. Making these `const fn`s, especially the functions set aside in the `field` modules of wire, would be greatly beneficial to me as a consumer of this crate, and I imagine many others. Many of the structs in my own crate's "wire" module for a custom protocol are more constrained versions of the packets and reprs provided by smoltcp. Due to being more constrained, many fields which are variable in the generic reprs are now constant in mine, which then lend themselves to being constants in my crate, however these functions and methods not being marked `const` prevents me from using smoltcp's tools to calculate these values. **Most** of the functions which I marked `const` in this commit are by their very nature/semantics guaranteed to be constifiable anyway, however a handful aren't necessarily, therefore:

**Important note:** While the majority of the functions are marked `const` are very safe to do so, not all of them are guaranteed to be so, and the reviewer of this PR should take caution and feel free to unmark something as const if they're not comfortable making that guarantee in the public API.

Another note: There are various other functions which can't be made `const fn` as written, but could be with minor tweaks. I did not include these tweaks but I think it should be considered as further work for this PR or a potential follow-up PR. Furthermore, plenty outside of `wire` is constifyable as well but I did not include those in the scope of this PR.

Co-authored-by: Nicholas Cyprus <[email protected]>

689: fix boundary case of assembler::remove_contig_at r=Dirbaio a=hikaricai

Found a bug when learning source code of smoltcp.

Co-authored-by: caiyuanhao <[email protected]>

This functionality is completely unused, and was panicky with overflow checking (see #694)
Fixes #694

700: assembler: do not return whether there was an overlap. r=Dirbaio a=Dirbaio

This functionality is completely unused, and was panicky with overflow checking (see #694).

Fixes #694

bors r+

Co-authored-by: Dario Nieuwenhuis <[email protected]>

685: DHCP indicate new config if there's a packet buffer provided r=Dirbaio a=JarredAllen

# Description

If the user provides a buffer in which to store the packets, then the contents of the received packet will be buffered and included in the returned Config when the DHCP connection is made. However, it isn't included in the Config struct until the value is returned to the user, so the equality check for whether to call `config_changed` disregards any additional information in the buffer beyond what this library parses.

For my purposes, I need access to the contents of the buffer when they change as a result of a new packet, even if everything else is the same. Since two packets will almost certainly not be the same thanks to the magic cookie (unless the packet gets duplicated on the network, which is an acceptably low risk for my use of smoltcp), an acceptable option for my uses is to just always return the new configuration when a packet is received (gated on whether a buffer is provided to return the packet into).

# Alternatives

While this approach is the easiest and best for my uses, I can think of the following alternatives which would also work and might be preferred for other use-cases:
 * Allow the user to specify whether they wish to receive all packets instead of opting all users who provide a buffer into this behavior
 * Allow the user to provide a closure which compares the old and new packets and returns true if this represents a new config which should be returned.
 * Compare the old packet to the new packet (either byte-by-byte or looking at the provided options) and only return a new config if differences are found.

# Verification

In my setup, I was seeing bugs that were caused by smoltcp not exposing the config when the only changes were in the additional options that I want to use but which smoltcp doesn't use directly. Using this branch instead of the main release fixed those bugs and I was able to verify that it behaves the way I expected. I think this verification, along with CI tests passing, should be sufficient for verifying this PR.

Co-authored-by: Jarred Allen <[email protected]>

Instant was displayed incorrectly. For example,
`Instant::from_millis(74)` would have been displayed as "0.74s" instead
of the correct "0.074s".

701: Fix how `Instant` is displayed r=thvdveld a=thvdveld

Instant was displayed incorrectly. For example,
`Instant::from_millis(74)` would have been displayed as "0.74s" instead of the correct "0.074s".

bors r+

Co-authored-by: Thibaut Vandervelden <[email protected]>

The current `'a` lifetime in the `Device` trait is essentially a workaround for lack of GATs.
I'm just experimenting how this would look like, it'll have to wait until GATs are stable to go in.

The main benefit is structs implementing `Device` can now borrow stuff. This wasn't possible
before because the `for<'d> T: Device<'d>` bounds would essentially imply `T: 'static`.

572: GAT-based Device trait. r=Dirbaio a=Dirbaio

The current `'a` lifetime in the `Device` trait is essentially a workaround for lack of GATs.
I'm just experimenting how this would look like, it'll have to wait until GATs are stable to go in.

The main benefit is structs implementing `Device` can now borrow stuff. This wasn't possible
before because the `for<'d> T: Device<'d>` bounds would essentially imply `T: 'static`.

Co-authored-by: Dario Nieuwenhuis <[email protected]>

706: Fix medium-ip not compiling complaining of needing EthernetAddress r=Dirbaio a=Dirbaio

#703 rebased

cc `@gopakumarce`

Co-authored-by: Gopa Kumar <[email protected]>

682: 6lowpan tests r=Dirbaio a=thvdveld

Add tests for 6LoWPAN:
- [x] Incoming ICMP packets
- [x] Incoming UDP packets
- [x] Outgoing ICMP packets
- [x] Outgoing UDP packets

~~Should be merged after #681 .~~

Co-authored-by: Thibaut Vandervelden <[email protected]>

This allows simplifying all cfg's to just check alloc.

There should be no downside to using `extern crate alloc` in targets with `std`,
since std support implies alloc support.

709: Make `std` imply `alloc` feature. r=Dirbaio a=Dirbaio

This allows simplifying all cfg's to just check alloc.

There should be no downside to using `extern crate alloc` in targets with `std`, since std support implies alloc support.

bors r+

Co-authored-by: Dario Nieuwenhuis <[email protected]>

711: Addvprinting of the needed size of fragmentation buffer. r=thvdveld a=KOLANICH



Co-authored-by: KOLANICH <[email protected]>

712: add more tests in CI r=Dirbaio a=thvdveld

IPv4 fragmentation was not covered in CI, neither was 6LoWPAN without fragmentation.

Co-authored-by: Thibaut Vandervelden <[email protected]>

710: Fix panic when using IPv4 fragmentation with Layer 3 networks. r=Dirbaio a=KOLANICH

```console
7:     0x7fe44617f300 - smoltcp::iface::interface::InterfaceInner::lookup_hardware_addr::h640efcd865d49106
						at ~/sio-smoltcp/smoltcp/src/iface/interface/mod.rs:2007:15
18:     0x7fe44618039a - smoltcp::iface::interface::InterfaceInner::dispatch_ip::h477a4410ea082342
						at ~/sio-smoltcp/smoltcp/src/iface/interface/mod.rs:2133:19
19:     0x7fe44617cd0f - smoltcp::iface::interface::Interface::socket_egress::{{closure}}::h0b25e0ce461b54e7
						at ~/sio-smoltcp/smoltcp/src/iface/interface/mod.rs:1241:17
20:     0x7fe44617d346 - smoltcp::iface::interface::Interface::socket_egress::{{closure}}::h3e55a2ce63c16877
						at ~/sio-smoltcp/smoltcp/src/iface/interface/mod.rs:1274:21
21:     0x7fe4461a24b4 - smoltcp::socket::udp::Socket::dispatch::{{closure}}::h52b2f159c6047eb1
						at ~/sio-smoltcp/smoltcp/src/socket/udp.rs:465:13
22:     0x7fe4461a4fa8 - smoltcp::storage::packet_buffer::PacketBuffer<H>::dequeue_with::{{closure}}::{{closure}}::ha53ebbf3d6d0e373
						at ~/sio-smoltcp/smoltcp/src/storage/packet_buffer.rs:200:27
23:     0x7fe446185b41 - smoltcp::storage::ring_buffer::RingBuffer<T>::dequeue_many_with::h4769dc1f2ccd50aa
						at ~/sio-smoltcp/smoltcp/src/storage/ring_buffer.rs:245:30
24:     0x7fe4461a4b16 - smoltcp::storage::packet_buffer::PacketBuffer<H>::dequeue_with::{{closure}}::h9789743e8966cfbe
						at ~/sio-smoltcp/smoltcp/src/storage/packet_buffer.rs:196:13
25:     0x7fe4461851cd - smoltcp::storage::ring_buffer::RingBuffer<T>::dequeue_one_with::h1201370f614614d2
						at ~/sio-smoltcp/smoltcp/src/storage/ring_buffer.rs:154:19
26:     0x7fe4461a494c - smoltcp::storage::packet_buffer::PacketBuffer<H>::dequeue_with::h62d428db6c52466d
						at ~/sio-smoltcp/smoltcp/src/storage/packet_buffer.rs:190:9
27:     0x7fe4461a1ca6 - smoltcp::socket::udp::Socket::dispatch::h3345cb3f2e0a28c5
						at ~/sio-smoltcp/smoltcp/src/socket/udp.rs:431:19
28:     0x7fe44617c75e - smoltcp::iface::interface::Interface::socket_egress::h4832c80b5e44260b
						at ~/sio-smoltcp/smoltcp/src/iface/interface/mod.rs:1273:40
29:     0x7fe44617b8b4 - smoltcp::iface::interface::Interface::poll::h3576f55b9ff8b575
						at ~/sio-smoltcp/smoltcp/src/iface/interface/mod.rs:1088:31
```

Co-authored-by: KOLANICH <[email protected]>

714: Fix PacketBuffer contig_window check when empty r=Dirbaio a=mkeeter

Fixes #713 (which also includes a more detailed explanation)

Co-authored-by: Matt Keeter <[email protected]>

fix reversed udp server example in README

Signed-off-by: Klim Tsoutsman <[email protected]>

… failing)

Signed-off-by: Klim Tsoutsman <[email protected]>

718: Make `AnySocket` object safe r=Dirbaio a=tsoutsman

Hi, I'm using `smoltcp` in an operating system and need to keep a global list of sockets NICs can poll when they receive an interrupt. Each socket is stored in its own socket set as that needs ownership over the socket and is the only way it can be polled. When the socket is registered, we know the concrete type, and so my `Socket` struct looks like so:

```rust
#[repr(transparent)]
pub struct Socket<T>
where
    T: AnySocket<'static> + ?Sized,
{
    pub(crate) inner: SocketSet<'static>,
    phantom_data: PhantomData<T>,
}

impl<T> Socket<T>
where
    T: AnySocket<'static> + ?Sized,
{
    pub(crate) fn new(inner: SocketSet<'static>) -> Self {
        Self {
            inner,
            phantom_data: PhantomData,
        }
    }
}

impl<T> Deref for Socket<T>
where
    T: AnySocket<'static>,
{
    type Target = T;

    fn deref(&self) -> &Self::Target {
        AnySocket::downcast(self.inner.iter().next().expect("no socket in socket set").1)
            .expect("incorrect socket type")
    }
}

impl<T> DerefMut for Socket<T>
where
    T: AnySocket<'static>,
{
    fn deref_mut(&mut self) -> &mut Self::Target {
        AnySocket::downcast_mut(
            self.inner
                .iter_mut()
                .next()
                .expect("no socket in socket set")
                .1,
        )
        .expect("incorrect socket type")
    }
}
```

However, they must be type erased as they are stored in a vec:
```rust
static SOCKETS: Mutex<Vec<Arc<Mutex<Socket<dyn AnySocket<'static> + Send>>>>>;

pub fn add_socket<T>(&self, socket: T) -> Arc<Mutex<crate::Socket<T>>>
where
    T: AnySocket<'static> + Send,
{
    let mut socket_set = SocketSet::new([iface::SocketStorage::default(); 1]);
    socket_set.add(socket);
    let socket_arc = Arc::new(Mutex::new(crate::Socket::<T>::new(socket_set)));
    SOCKETS
        .lock()
        // SAFETY: Socket has a transparent representation and so the memory layout is the same
        // regardless of T. The Send bound on T ensures that transmuting to a type that
        // implemnts Send is sound.
        .push(unsafe { core::mem::transmute(socket_arc.clone()) });
    socket_arc
}

```

This is only possible if `AnySocket` is object safe.

Co-authored-by: Klim Tsoutsman <[email protected]>

719: Interface `ip_addrs` field from slice to `heapless::Vec` r=Dirbaio a=davidedellagiustina

As stated in the title.
One cargo test is still failing, needs to be looked into.

Moreover, for now the `heapless::Vec` used is fixed-size.

Co-authored-by: Davide Della Giustina <[email protected]>

Apologies, in #718, removing the sized bound in `downcast` and
`downcast_mut` implicitly made `AnySocket` object unsafe as they don't
have a `self` method. I had a bit of a brain fart.

Signed-off-by: Klim Tsoutsman <[email protected]>

721: Make `AnySocket` object safe r=Dirbaio a=tsoutsman

Apologies, in #718, removing the sized bound in `downcast` and `downcast_mut` implicitly made `AnySocket` object unsafe as they don't have a `self` method. I had a bit of a brain fart.

Signed-off-by: Klim Tsoutsman <[email protected]>

Co-authored-by: Klim Tsoutsman <[email protected]>

Lookup is O(n) now. However, it previously did 32 (or 128 for ipv6!)
map lookups. Since the route table typically doesn't have that many
routes, the new code is likely faster even if it's O(n).

725: Remove feature rust-1_28 r=Dirbaio a=Dirbaio

Rust 1.28 is wayyyyyy below our MSRV so this feature is effectively useless.

Co-authored-by: Dario Nieuwenhuis <[email protected]>

More heapless

This fixes #723.

724: Make doc-test `no_run` instead of `ignore` r=Dirbaio a=thvdveld

This fixes #723.

Co-authored-by: Thibaut Vandervelden <[email protected]>

728: Clippy fixes. r=Dirbaio a=Dirbaio

bors r+

Co-authored-by: Dario Nieuwenhuis <[email protected]>

729: Switch to Rust 2021 r=Dirbaio a=Dirbaio

bors r+

Co-authored-by: Dario Nieuwenhuis <[email protected]>

- Assembler no longer needs the total buffer size, added range indices can be as high as needed.
- Fixes quadratic run time when adding a range that overlaps with many existing ranges.