Record the speed of a USB device based on the port's speed at the time
that the device was enabled. This allows us to remember the device's
speed even after the device has been disconnected (and so the port's
current speed has changed).
In particular, this allows us to correctly identify the transaction
translator for a low-speed or full-speed device after the device has
been disconnected.
Signed-off-by: Michael Brown <mcb30@ipxe.org>
Provide BIT_QWORD_PTR() to allow for easy extraction of non-endian
fields (e.g. Infiniband GUIDs) without unnecessary byte swapping.
Signed-off-by: Michael Brown <mcb30@ipxe.org>
Tested using QEMU and usbredir to expose the LAN9512 chip present on a
Raspberry Pi.
There is a known issue with the LAN9512: an extra two bytes are
appended to every transmitted packet. These two bytes comprise:
{ 0x00, 0x08 } if packet length == 0 (mod 8)
{ CRC[0], 0x00 } if packet length == 7 (mod 8)
{ CRC[0], CRC[1] } otherwise
The extra bytes are appended whether the Ethernet CRC is generated
manually or added automatically by the hardware. The issue occurs
with the Linux kernel driver as well as the iPXE driver. It appears
to be an undocumented hardware errata.
TCP/IP traffic is not affected, since the IP header length field
causes the extraneous bytes to be discarded by the receiver. However,
protocols that rely on the length of the Ethernet frame (such as FCoE
or iPXE's "lotest" protocol) will be unusable on this hardware.
Signed-off-by: Michael Brown <mcb30@ipxe.org>
Allow the UEFI platform firmware to provide drivers for unrecognised
devices, by exposing our own implementation of EFI_USB_IO_PROTOCOL.
Signed-off-by: Michael Brown <mcb30@ipxe.org>
Make the class ID a property of the USB driver (rather than a property
of the USB device ID), and allow USB drivers to specify a wildcard ID
for any of the three component IDs (class, subclass, or protocol).
Signed-off-by: Michael Brown <mcb30@ipxe.org>
Generate a score for each possible USB device configuration based on
the available driver support, and select the configuration with the
highest score. This will allow us to prefer ECM over RNDIS (for
devices which support both) and will allow us to meaningfully select a
configuration even when we have drivers available for all functions
(e.g. when exposing unused functions via EFI_USB_IO_PROTOCOL).
Signed-off-by: Michael Brown <mcb30@ipxe.org>
The decision on whether or not a zero-length packet needs to be
transmitted is independent of the host controller and belongs in the
USB core.
Signed-off-by: Michael Brown <mcb30@ipxe.org>
TCP/IP checksum fields are one's complement values and therefore have
two possible representations of zero: positive zero (0x0000) and
negative zero (0xffff).
In RFC768, UDP over IPv4 exploits this redundancy to repurpose the
positive representation of zero (0x0000) to mean "no checksum
calculated"; checksums are optional for UDP over IPv4.
In RFC2460, checksums are made mandatory for UDP over IPv4. The
wording of the RFC is such that the UDP header is mandated to use only
the negative representation of zero (0xffff), rather than simply
requiring the checksum to be correct but allowing for either
representation of zero to be used.
In RFC1071, an example algorithm is given for calculating the TCP/IP
checksum. This algorithm happens to produce only the positive
representation of zero (0x0000); this is an artifact of the way that
unsigned arithmetic is used to calculate a signed one's complement
sum (and its final negation).
A common misconception has developed (exemplified in RFC1624) that
this artifact is part of the specification. Many people have assumed
that the checksum field should never contain the negative
representation of zero (0xffff).
A sensible receiver will calculate the checksum over the whole packet
and verify that the result is zero (in whichever representation of
zero happens to be generated by the receiver's algorithm). Such a
receiver will not care which representation of zero happens to be used
in the checksum field.
However, there are receivers in existence which will verify the
received checksum the hard way: by calculating the checksum over the
remainder of the packet and comparing the result against the checksum
field. If the representation of zero used by the receiver's algorithm
does not match the representation of zero used by the transmitter (and
so placed in the checksum field), and if the receiver does not
explicitly allow for both representations to compare as equal, then
the receiver may reject packets with a valid checksum.
For UDP, the combined RFCs effectively mandate that we should generate
only the negative representation of zero in the checksum field.
For IP, TCP and ICMP, the RFCs do not mandate which representation of
zero should be used, but the misconceptions which have grown up around
RFC1071 and RFC1624 suggest that it would be least surprising to
generate only the positive representation of zero in the checksum
field.
Fix by ensuring that all of our checksum algorithms generate only the
positive representation of zero, and explicitly inverting this in the
case of transmitted UDP packets.
Reported-by: Wissam Shoukair <wissams@mellanox.com>
Tested-by: Wissam Shoukair <wissams@mellanox.com>
Signed-off-by: Michael Brown <mcb30@ipxe.org>
Allow iPXE to coexist with other USB device drivers, by attaching to
the EFI_USB_IO_PROTOCOL instances provided by the UEFI platform
firmware.
The EFI_USB_IO_PROTOCOL is an unsurprisingly badly designed
abstraction of a USB device. The poor design choices intrinsic in the
UEFI specification prevent efficient operation as a network device,
with the result that devices operated using the EFI_USB_IO_PROTOCOL
operate approximately two orders of magnitude slower than devices
operated using our native EHCI or xHCI host controller drivers.
Since the performance is so abysmally slow, and since the underlying
problems are due to fundamental architectural mistakes in the UEFI
specification, support for the EFI_USB_IO_PROTOCOL host controller
driver is left as disabled by default. Users are advised to use the
native iPXE host controller drivers instead.
Signed-off-by: Michael Brown <mcb30@ipxe.org>
Many UEFI NBPs expect to find an EFI_PXE_BASE_CODE_PROTOCOL installed
in addition to the EFI_SIMPLE_NETWORK_PROTOCOL. Most NBPs use the
EFI_PXE_BASE_CODE_PROTOCOL only to retrieve the cached DHCP packets.
This implementation has been tested with grub.efi, shim.efi,
syslinux.efi, and wdsmgfw.efi. Some methods (such as Discover() and
Arp()) are not used by any known NBP and so have not (yet) been
implemented.
Usage notes for the tested bootstraps are:
- grub.efi uses EFI_PXE_BASE_CODE_PROTOCOL only to retrieve the
cached DHCP packet, and uses no other methods.
- shim.efi uses EFI_PXE_BASE_CODE_PROTOCOL to retrieve the cached
DHCP packet and to retrieve the next NBP via the Mtftp() method.
If shim.efi was downloaded via HTTP (or other non-TFTP protocol)
then shim.efi will blindly call Mtftp() with an HTTP URI as the
filename: this allows the next NBP (e.g. grubx64.efi) to also be
transparently retrieved by HTTP.
shim.efi can also use the EFI_SIMPLE_FILE_SYSTEM_PROTOCOL to
retrieve files previously loaded by "imgfetch" or similar commands
in iPXE. The current implementation of shim.efi will use the
EFI_SIMPLE_FILE_SYSTEM_PROTOCOL only if it does not find an
EFI_PXE_BASE_CODE_PROTOCOL; this patch therefore prevents this
usage of our EFI_SIMPLE_FILE_SYSTEM_PROTOCOL. This logic could be
trivially reversed in shim.efi if needed.
- syslinux.efi uses EFI_PXE_BASE_CODE_PROTOCOL only to retrieve the
cached DHCP packet. Versions 6.03 and earlier have a bug which
may cause syslinux.efi to attach to the wrong NIC if there are
multiple NICs in the system (or if the UEFI firmware supports
IPv6).
- wdsmgfw.efi (ab)uses EFI_PXE_BASE_CODE_PROTOCOL to retrieve the
cached DHCP packets, and to send and retrieve UDP packets via the
UdpWrite() and UdpRead() methods. (This was presumably done in
order to minimise the amount of benefit obtainable by switching to
UEFI, by replicating all of the design mistakes present in the
original PXE specification.)
The EFI_DOWNGRADE_UX configuration option remains available for now,
until this implementation has received more widespread testing.
Signed-off-by: Michael Brown <mcb30@ipxe.org>
Merge the functionality of parse_next_server_and_filename() and
tftp_uri() into a single pxe_uri(), which takes a server address
(IPv4/IPv6/none) and a filename, and produces a URI using the rule:
- if the filename is a hierarchical absolute URI (i.e. includes a
scheme such as "http://" or "tftp://") then use that URI and ignore
the server address,
- otherwise, if the server address is recognised (according to
sa_family) then construct a TFTP URI based on the server address,
port, and filename
- otherwise fail.
Signed-off-by: Michael Brown <mcb30@ipxe.org>
Add definitions of protocols observed to be used by wdsmgfw.efi, and
add a handle name type for ConIn, ConOut, and StdErr.
Signed-off-by: Michael Brown <mcb30@ipxe.org>
Add debug wrappers for more boot services functions, and print
symbolic values rather than raw numbers where possible.
Signed-off-by: Michael Brown <mcb30@ipxe.org>
Commit 09b057c ("[settings] Remove "uristring" setting type") removed
support for URI-encoded settings via the "uristring" setting type, on
the basis that such encoding was no longer necessary to avoid problems
with the command line parser.
Other valid use cases for the "uristring" setting type do exist: for
example, a password containing a '/' character expanded via
chain http://username:${password:uristring}@server.name/boot.php
Restore the existence of the "uristring" setting, avoiding the
potentially large stack allocations that were used in the old code
prior to commit 09b057c ("[settings] Remove "uristring" setting
type").
Requested-by: Robin Smidsrød <robin@smidsrod.no>
Signed-off-by: Michael Brown <mcb30@ipxe.org>
The current usage pattern of image_probe() is a legacy from the time
before commit 34b6ecb ("[image] Simplify image management") when
loading an image to its executable location in memory was a separate
action from actually executing the image.
Call image_probe() as soon as an image is registered. This allows
"imgstat" to display image type information for all images and allows
image-consuming code to assume that image->type is already set
correctly.
Ignore failures if image_probe() does not recognise the image, since
we do expect to handle unrecognised images (initrds, modules, etc).
Unrecognised images will be left with a NULL image->type, which
image-consuming code can easily check.
Signed-off-by: Michael Brown <mcb30@ipxe.org>
Allow the return status from an embedded image to propagate out to the
eventual return status from main(). When running under Linux, this
allows the pass/fail result of unit tests to be observable without
having to visually inspect the console output.
Signed-off-by: Michael Brown <mcb30@ipxe.org>
Rewrite the HTTP core to allow for the addition of arbitrary content
encoding mechanisms, such as PeerDist and gzip.
The core now exposes http_open() which can be used to create requests
with an explicitly selected HTTP method, an optional requested content
range, and an optional request body. A simple wrapper provides the
preexisting behaviour of creating either a GET request or an
application/x-www-form-urlencoded POST request (if the URI includes
parameters).
The HTTP SAN interface is now implemented using the generic block
device translator. Individual blocks are requested using http_open()
to create a range request.
Server connections are now managed via a connection pool; this allows
for multiple requests to the same server (e.g. for SAN blocks) to be
completely unaware of each other. Repeated HTTPS connections to the
same server can reuse a pooled connection, avoiding the per-connection
overhead of establishing a TLS session (which can take several seconds
if using a client certificate).
Support for HTTP SAN booting and for the Basic and Digest
authentication schemes is now optional and can be controlled via the
SANBOOT_PROTO_HTTP, HTTP_AUTH_BASIC, and HTTP_AUTH_DIGEST build
configuration options in config/general.h.
Signed-off-by: Michael Brown <mcb30@ipxe.org>
UEFI platforms may provide a watchdog timer, which will reboot the
machine if an operating system takes more than five minutes to load.
This can cause long-lived iPXE downloads (or interactive shell
sessions) to unexpectedly reboot.
Fix by resetting the watchdog timer every ten seconds while the iPXE
main processing loop continues to run.
Reported-by: Bradley B Williams <bradleybwilliams@swbell.net>
Reported-by: John Clark <john.r.clark.3@gmail.com>
Reported-by: wdriever@gmail.com
Reported-by: Charlie Beima <cbeima@indiana.edu>
Signed-off-by: Michael Brown <mcb30@ipxe.org>
Add support for SHA-224, SHA-384, and SHA-512 as digest algorithms in
X.509 certificates, and allow the choice of public-key, cipher, and
digest algorithms to be configured at build time via config/crypto.h.
Originally-implemented-by: Tufan Karadere <tufank@gmail.com>
Signed-off-by: Michael Brown <mcb30@ipxe.org>
Check for existence of the UART in uart_select(), not just in
uart_init(). This allows uart_select() to refuse to set a non-working
address in uart->base, which in turns means that the serial console
code will not attempt to use a non-existent UART.
Reported-by: Torgeir Wulfsberg <Torgeir.Wulfsberg@kongsberg.com>
Reported-by: Ján ONDREJ (SAL) <ondrejj@salstar.sk>
Signed-off-by: Michael Brown <mcb30@ipxe.org>
We do not set up any kind of virtual addressing before invoking an
ELFBoot image. Reject if the image's program headers indicate that
virtual addresses are not equal to physical addresses.
This avoids problems when loading some RHEL5 kernels, which seem to
include ELFBoot headers using virtual addressing. With this change,
these kernels are no longer detected as ELFBoot, and so may be
(correctly) detected as bzImage instead.
Reported-by: Torgeir.Wulfsberg@kongsberg.com
Signed-off-by: Michael Brown <mcb30@ipxe.org>
Allow line buffer to accumulate multiple lines, with buffered_line()
returning each freshly-completed line as it is encountered. This
allows buffered lines to be subsequently processed as a group.
Signed-off-by: Michael Brown <mcb30@ipxe.org>
VLAN and 802.11 devices use a network device operations structure that
wraps an underlying structure. For example, the vlan_operations
structure wraps the network device operations structure of the
underlying trunk device. This can cause false positives from the
current implementation of netdev_irq_supported(), which will always
report that VLAN devices support interrupts since it has no visibility
into the support provided by the underlying trunk device.
Fix by allowing network devices to explicitly flag that interrupts are
not supported, despite the presence of an irq() method.
Originally-fixed-by: Wissam Shoukair <wissams@mellanox.com>
Signed-off-by: Michael Brown <mcb30@ipxe.org>
Extend the IPv6 concept of "scope ID" (indicating the network device
index) to IPv4 socket addresses, so that IPv4 multicast transmissions
may specify the transmitting network device.
The scope ID is not (currently) exposed via the string representation
of the socket address, since IPv4 does not use the IPv6 concept of
link-local addresses (which could legitimately be specified in a URI).
Signed-off-by: Michael Brown <mcb30@ipxe.org>
Redefine various IPv4 address constants and testing macros to avoid
unnecessary byte swapping at runtime, and slightly rename the macros
to prevent code from accidentally using the old definitions.
Signed-off-by: Michael Brown <mcb30@ipxe.org>
When an IPv6 socket address string specifies a link-local or multicast
address but does not specify the requisite network device name
(e.g. "fe80::69ff:fe50:5845" rather than "fe80::69ff:fe50:5845%net0"),
assume the use of "netX".
Signed-off-by: Michael Brown <mcb30@ipxe.org>
Replace the AES implementation from AXTLS with a dedicated iPXE
implementation which is slightly smaller and around 1000% faster.
This implementation has been verified using the existing self-tests
based on the NIST AES test vectors.
Signed-off-by: Michael Brown <mcb30@ipxe.org>
Reduce the cost of implementing object methods which convey no
information beyond the fact that the method has been called.
Signed-off-by: Michael Brown <mcb30@ipxe.org>
Provide profile_custom() as a trivial wrapper around profile_update()
to allow for the use of the profiling infrastructure by code using
timers other than the default profile_timestamp() provider.
Signed-off-by: Michael Brown <mcb30@ipxe.org>
Provide an inject_corruption() function that can be used to randomly
corrupt data bytes with configurable probabilities.
Signed-off-by: Michael Brown <mcb30@ipxe.org>
Provide a generic inject_fault() function that can be used to inject
random faults with configurable probabilities.
Signed-off-by: Michael Brown <mcb30@ipxe.org>
Fix the TxBuf value filled in by GetStatus() to report the transmit
buffer address as required by the (now clarified) specification.
Simplify "interrupt" handling in GetStatus() to report only that one
or more packets have been transmitted or received; there is no need to
report one GetStatus() "interrupt" per packet.
Simplify receive handling to dequeue received packets immediately from
the network device into an internal list (thereby avoiding the hacks
previously used to determine when to report new packet arrivals).
Originally-fixed-by: Laszlo Ersek <lersek@redhat.com>
Signed-off-by: Michael Brown <mcb30@ipxe.org>
We currently do not wait for a received FIN before exiting to boot a
loaded OS. In the common case of booting from an HTTP server, this
means that the TCP connection is left consuming resources on the
server side: the server will retransmit the FIN several times before
giving up.
Fix by initiating a graceful close of all TCP connections and waiting
(for up to one second) for all connections to finish closing
gracefully (i.e. for the outgoing FIN to have been sent and ACKed, and
for the incoming FIN to have been received and ACKed at least once).
Signed-off-by: Michael Brown <mcb30@ipxe.org>
The only way to map an eIPoIB MAC address (REMAC) to an IPoIB MAC
address is to intercept an incoming ARP request or reply.
If we do not have an REMAC cache entry for a particular destination
MAC address, then we cannot transmit the packet. This can arise in at
least two situations:
- An external program (e.g. a PXE NBP using the UNDI API) may attempt
to transmit to a destination MAC address that has been obtained by
some method other than ARP.
- Memory pressure may have caused REMAC cache entries to be
discarded. This is fairly likely on a busy network, since REMAC
cache entries are created for all received (broadcast) ARP
requests. (We can't sensibly avoid creating these cache entries,
since they are required in order to send an ARP reply, and when we
are being used via the UNDI API we may have no knowledge of which
IP addresses are "ours".)
Attempt to ameliorate the situation by generating a semi-spurious ARP
request whenever we find a missing REMAC cache entry. This will
hopefully trigger an ARP reply, which would then provide us with the
information required to populate the REMAC cache.
Signed-off-by: Michael Brown <mcb30@ipxe.org>
A fairly common end-user problem is that the default configuration of
a switch may leave the port in a non-forwarding state for a
substantial length of time (tens of seconds) after link up. This can
cause iPXE to time out and give up attempting to boot.
We cannot force the switch to start forwarding packets sooner, since
any attempt to send a Spanning Tree Protocol bridge PDU may cause the
switch to disable our port (if the switch happens to have the Bridge
PDU Guard feature enabled for the port).
For non-ancient versions of the Spanning Tree Protocol, we can detect
whether or not the port is currently forwarding and use this to inform
the network device core that the link is currently blocked.
Signed-off-by: Michael Brown <mcb30@ipxe.org>
When Spanning Tree Protocol (STP) is used, there may be a substantial
delay (tens of seconds) from the time that the link goes up to the
time that the port starts forwarding packets.
Add a generic concept of a "blocked link" (i.e. a link which is up but
which is not expected to communicate successfully), and allow "ifstat"
to indicate when a link is blocked.
Signed-off-by: Michael Brown <mcb30@ipxe.org>
Virtual functions use a mailbox to communicate with the physical
function driver: this covers functionality such as obtaining the MAC
address.
Signed-off-by: Michael Brown <mcb30@ipxe.org>
When USB network card drivers are used, the BIOS' legacy USB
capability is necessarily disabled since there is no way to share the
host controller between the BIOS and iPXE. This currently results in
USB keyboards becoming non-functional in USB-enabled builds of iPXE.
Fix by adding basic support for USB keyboards, enabled by default in
iPXE builds which include USB support.
Signed-off-by: Michael Brown <mcb30@ipxe.org>
When an EHCI hotplug action results in the controller disowning the
port, it will result in a hotplug action on the corresponding UHCI or
OHCI controller. Allow such hotplug actions to be carried out as part
of the same call to usb_step() or usb_register_bus(), by maintaining a
single central list of changed ports.
Signed-off-by: Michael Brown <mcb30@ipxe.org>
The USB core will currently fail to detect disconnections if a new
device has attached by the time the port is examined in
usb_hotplug().
Fix by recording the fact that a disconnection has taken place
whenever the "connection status changed" (CSC) bit is observed to be
set. (Whether the change represents a disconnection or a
reconnection, it indicates that the port has experienced some time of
being disconnected.)
Note that the time at which a disconnection can be detected varies by
hub type. In particular: root hubs can observe the CSC bit when
polling, and so will record the disconnection before calling
usb_port_changed(), but USB hubs read the port status (and hence the
CSC bit) only during the call to hub_speed(), long after the call to
usb_port_changed().
Signed-off-by: Michael Brown <mcb30@ipxe.org>
Rename PCI_CLASS() (which constructs a struct pci_class_id) to
PCI_CLASS_ID(), and provide PCI_CLASS() as a macro which constructs
the 24-bit scalar value of a PCI class code.
Signed-off-by: Michael Brown <mcb30@ipxe.org>
The USB API currently assumes that host controllers will have
immediate data buffer space available in which to store the setup
packet. This is true for xHCI, partially true for EHCI (which happens
to have 12 bytes of padding in each transfer descriptor due to
alignment requirements), and not true at all for UHCI.
Include the setup packet within the I/O buffer passed to the host
controller's message() method, thereby eliminating the requirement for
host controllers to provide immediate data buffers.
Signed-off-by: Michael Brown <mcb30@ipxe.org>
The current API for Base16 (and Base64) encoding requires the caller
to always provide sufficient buffer space. This prevents the use of
the generic encoding/decoding functionality in some situations, such
as in formatting the hex setting types.
Implement a generic hex_encode() (based on the existing
format_hex_setting()), implement base16_encode() and base16_decode()
in terms of the more generic hex_encode() and hex_decode(), and update
all callers to provide the additional buffer length parameter.
Signed-off-by: Michael Brown <mcb30@ipxe.org>
Entropy gathering via timer ticks is slow under UEFI (of the order of
20-30 seconds on some machines). Use the EFI_RNG_PROTOCOL if
available, to speed up the process of entropy gathering.
Note that some implementations (including EDK2) will fail if we
request fewer than 32 random bytes at a time, and that the RNG
protocol provides no guarantees about the amount of entropy provided
by a call to GetRNG(). We take the (hopefully pessimistic) view that
a 32-byte block returned by GetRNG() will contain at least the 1.3
bits of entropy claimed by min_entropy_per_sample().
Signed-off-by: Michael Brown <mcb30@ipxe.org>
SHA-512/224 is almost identical to SHA-512, with differing initial
hash values and a truncated output length.
This implementation has been verified using the NIST SHA-512/224 test
vectors.
Signed-off-by: Michael Brown <mcb30@ipxe.org>
SHA-512/256 is almost identical to SHA-512, with differing initial
hash values and a truncated output length.
This implementation has been verified using the NIST SHA-512/256 test
vectors.
Signed-off-by: Michael Brown <mcb30@ipxe.org>
SHA-384 is almost identical to SHA-512, with differing initial hash
values and a truncated output length.
This implementation has been verified using the NIST SHA-384 test
vectors.
Signed-off-by: Michael Brown <mcb30@ipxe.org>
SHA-224 is almost identical to SHA-256, with differing initial hash
values and a truncated output length.
This implementation has been verified using the NIST SHA-224 test
vectors.
Signed-off-by: Michael Brown <mcb30@ipxe.org>
xHCI provides a somewhat convoluted mechanism for specifying details
of a transaction translator. Hubs must be marked as such in the
device slot context. The only opportunity to do so is as part of a
Configure Endpoint command, which can be executed only when opening
the hub's interrupt endpoint.
We add a mechanism for host controllers to intercept the opening of
hub devices, providing xHCI with an opportunity to update the internal
device slot structure for the corresponding USB device to indicate
that the device is a hub. We then include the hub-specific details in
the input context whenever any Configure Endpoint command is issued.
When a device is opened, we record the device slot and port for its
transaction translator (if any), and supply these as part of the
Address Device command.
Signed-off-by: Michael Brown <mcb30@ipxe.org>
The current endpoint reset logic defers the reset until the caller
attempts to enqueue a new transfer to that endpoint. This is
insufficient when dealing with endpoints behind a transaction
translator, since the transaction translator is a resource shared
between multiple endpoints.
We cannot reset the endpoint as part of the completion handling, since
that would introduce recursive calls to usb_poll(). Instead, we
add the endpoint to a list of halted endpoints, and perform the reset
on the next call to usb_step().
Signed-off-by: Michael Brown <mcb30@ipxe.org>
Several of the USB timeouts were chosen on the principle of "pick an
arbitrary but ridiculously large value, just to be safe". It turns
out that some of the timeouts permitted by the USB specification are
even larger: for example, control transactions are allowed to take up
to five seconds to complete.
Fix up these USB timeout values to match those found in the USB2
specification.
Debugged-by: Robin Smidsrød <robin@smidsrod.no>
Tested-by: Robin Smidsrød <robin@smidsrod.no>
Signed-off-by: Michael Brown <mcb30@ipxe.org>
The TCP Selective Acknowledgement option (specified in RFC2018)
provides a mechanism for the receiver to indicate packets that have
been received out of order (e.g. due to earlier dropped packets).
iPXE often operates in environments in which there is a high
probability of packet loss. For example, the legacy USB keyboard
emulation in some BIOSes involves polling the USB bus from within a
system management interrupt: this introduces an invisible delay of
around 500us which is long enough for around 40 full-length packets to
be dropped. Similarly, almost all 1Gbps USB2 devices will eventually
end up dropping packets because the USB2 bus does not provide enough
bandwidth to sustain a 1Gbps stream, and most devices will not provide
enough internal buffering to hold a full TCP window's worth of
received packets.
Add support for sending TCP Selective Acknowledgements. This provides
the sender with more detailed information about which packets have
been lost, and so allows for a more efficient retransmission strategy.
We include a SACK-permitted option in our SYN packet, since
experimentation shows that at least Linux peers will not include a
SACK-permitted option in the SYN-ACK packet if one was not present in
the initial SYN. (RFC2018 does not seem to mandate this behaviour,
but it is consistent with the approach taken in RFC1323.) We ignore
any received SACK options; this is safe to do since SACK is only ever
advisory and we never have to send non-trivial amounts of data.
Since our TCP receive queue is a candidate for cache discarding under
low memory conditions, we may end up discarding data that has been
reported as received via a SACK option. This is permitted by RFC2018.
We follow the stricture that SACK blocks must not report data which is
no longer held by the receiver: previously-reported blocks are
validated against the current receive queue before being included
within the current SACK block list.
Experiments in a qemu VM using forced packet drops (by setting
NETDEV_DISCARD_RATE to 32) show that implementing SACK improves
throughput by around 400%.
Experiments with a USB2 NIC (an SMSC7500) show that implementing SACK
improves throughput by around 700%, increasing the download rate from
35Mbps up to 250Mbps (which is approximately the usable bandwidth
limit for USB2).
Signed-off-by: Michael Brown <mcb30@ipxe.org>
This driver is functional but any downloads via a TCP-based protocol
tend to perform poorly. The 1Gbps Ethernet line rate is substantially
higher than the 480Mbps (in practice around 280Mbps) provided by USB2,
and the device has only 32kB of internal buffer memory. Our 256kB TCP
receive window therefore rapidly overflows the RX FIFO, leading to
multiple dropped packets (usually within the same TCP window) and
hence a low overall throughput.
Reducing the TCP window size so that the RX FIFO does not overflow
greatly increases throughput, but is not a general-purpose solution.
Further investigation is required to determine how other OSes
(e.g. Linux) cope with this scenario. It is possible that
implementing TCP SACK would provide some benefit.
Signed-off-by: Michael Brown <mcb30@ipxe.org>
Most devices expose at least the link up/down status via a bit in a
MAC register, since the MAC generally already needs to know whether or
not the link is up. Some devices (e.g. the SMSC75xx USB NIC) expose
this information to software only via the MII registers.
Provide a generic mii_check_link() implementation to check the BMSR
and report the link status via netdev_link_{up,down}().
Signed-off-by: Michael Brown <mcb30@ipxe.org>
Relicense files with kind permission from
Stefan Hajnoczi <stefanha@redhat.com>
alongside the contributors who have already granted such relicensing
permission.
Signed-off-by: Michael Brown <mcb30@ipxe.org>
Rewrite (and relicense) the header files which are included in all
builds of iPXE (including non-Linux builds).
Signed-off-by: Michael Brown <mcb30@ipxe.org>
At some point in the past few years, binutils became more aggressive
at removing unused symbols. To function as a symbol requirement, a
relocation record must now be in a section marked with @progbits and
must not be in a section which gets discarded during the link (either
via --gc-sections or via /DISCARD/).
Update REQUIRE_SYMBOL() to generate relocation records meeting these
criteria. To minimise the impact upon the final binary size, we use
existing symbols (specified via the REQUIRING_SYMBOL() macro) as the
relocation targets where possible. We use R_386_NONE or R_X86_64_NONE
relocation types to prevent any actual unwanted relocation taking
place. Where no suitable symbol exists for REQUIRING_SYMBOL() (such
as in config.c), the macro PROVIDE_REQUIRING_SYMBOL() can be used to
generate a one-byte-long symbol to act as the relocation target.
If there are versions of binutils for which this approach fails, then
the fallback will probably involve killing off REQUEST_SYMBOL(),
redefining REQUIRE_SYMBOL() to use the current definition of
REQUEST_SYMBOL(), and postprocessing the linked ELF file with
something along the lines of "nm -u | wc -l" to check that there are
no undefined symbols remaining.
Signed-off-by: Michael Brown <mcb30@ipxe.org>
The valgrind headers are not x86-specific; they detect the CPU
architecture and contain inline assembly for multiple architectures.
Signed-off-by: Michael Brown <mcb30@ipxe.org>
The code in list.h was originally taken from the Linux kernel many
years ago, but has been rewritten to the point that no original code
remains, and may therefore be relicensed.
The functions and data structures remain largely API-compatible, to
facilitate the conversion of Linux network drivers to iPXE.
Signed-off-by: Michael Brown <mcb30@ipxe.org>
These files cannot be automatically relicensed by util/relicense.pl
since they either contain unusual but trivial contributions (such as
the addition of __nonnull function attributes), or contain lines
dating back to the initial git revision (and so require manual
knowledge of the code's origin).
Signed-off-by: Michael Brown <mcb30@ipxe.org>
Relicence files with kind permission from the following contributors:
Alex Williamson <alex.williamson@redhat.com>
Eduardo Habkost <ehabkost@redhat.com>
Greg Jednaszewski <jednaszewski@gmail.com>
H. Peter Anvin <hpa@zytor.com>
Marin Hannache <git@mareo.fr>
Robin Smidsrød <robin@smidsrod.no>
Shao Miller <sha0.miller@gmail.com>
Thomas Horsten <thomas@horsten.com>
Signed-off-by: Michael Brown <mcb30@ipxe.org>
