Michael Brown [Fri, 29 Jun 2012 15:07:12 +0000 (16:07 +0100)]
[iobuf] Relax alignment requirement for small I/O buffers
iPXE currently aligns all I/O buffers on a 2kB boundary. This is
overkill for transmitted packets, which are typically much smaller
than 2kB.
Align I/O buffers on their own size. This reduces the alignment
requirement for small buffers, while preserving the guarantee that I/O
buffers will never cross boundaries that might cause problems for some
DMA engines.
Michael Brown [Fri, 29 Jun 2012 14:28:15 +0000 (15:28 +0100)]
[tls] Request a maximum fragment length of 2048 bytes
The default maximum plaintext fragment length for TLS is 16kB, which
is a substantial amount of memory for iPXE to have to allocate for a
temporary decryption buffer.
Reduce the memory footprint of TLS connections by requesting a maximum
fragment length of 2kB.
Michael Brown [Fri, 29 Jun 2012 13:30:18 +0000 (14:30 +0100)]
[tcp] Add support for TCP window scaling
The maximum unscaled TCP window (64kB) implies a maximum bandwidth of
around 300kB/s on a WAN link with an RTT of 200ms. Add support for
the TCP window scaling option to remove this upper limit.
Michael Brown [Fri, 29 Jun 2012 00:03:38 +0000 (01:03 +0100)]
[undi] Align the received frame payload for faster processing
The undinet driver always has to make a copy of the received frame
into an I/O buffer. Align this copy sensibly so that subsequent
operations are as fast as possible.
Michael Brown [Thu, 28 Jun 2012 11:27:43 +0000 (12:27 +0100)]
[monojob] Check for keypresses only once per timer tick
Checking for keypresses takes a non-negligible amount of time, and
measurably affects our RTT. Minimise the impact by checking for
keypresses only once per timer tick.
Michael Brown [Tue, 26 Jun 2012 16:19:18 +0000 (17:19 +0100)]
[tcpip] Add faster algorithm for calculating the TCP/IP checksum
The generic TCP/IP checksum implementation requires approximately 10
CPU clocks per byte (as measured using the TSC). Improve this to
approximately 0.5 CPU clocks per byte by using "lodsl ; adcl" in an
unrolled loop.
Michael Brown [Tue, 26 Jun 2012 11:42:24 +0000 (12:42 +0100)]
[libc] Simplify memcpy() implementation
The "rep" prefix can be used with an iteration count of zero, which
allows the variable-length memcpy() to be implemented without using
any conditional jumps.
Michael Brown [Tue, 26 Jun 2012 11:42:06 +0000 (12:42 +0100)]
[cmdline] Increase resolution of "time" command
A reasonably large (512MB) file transferred via HTTP over Gigabit
Ethernet should complete in around 4.6 seconds. Increase the
resolution of the "time" command to tenths of a second, to allow such
transfers to be meaningfully measured.
Michael Brown [Wed, 20 Jun 2012 11:15:42 +0000 (12:15 +0100)]
[crypto] Allow an error margin on X.509 certificate validity periods
iPXE has no concept of the local time zone, mainly because there is no
viable way to obtain time zone information in the absence of local
state. This causes potential problems with newly-issued certificates
and certificates that are about to expire.
Avoid such problems by allowing an error margin of around 12 hours on
certificate validity periods, similar to the error margin already
allowed for OCSP response timestamps.
Michael Brown [Wed, 20 Jun 2012 10:56:43 +0000 (11:56 +0100)]
[dhcp] Request broadcast responses when we already have an IPv4 address
FCoE requires the use of multiple local unicast link-layer addresses.
To avoid the complexity of managing multiple addresses, iPXE operates
in promiscuous mode. As a consequence, any unicast packets with
non-matching IPv4 addresses are rejected at the IPv4 layer (rather
than at the link layer).
This can cause problems when issuing a second DHCP request: if the
address chosen by the DHCP server does not match the existing address,
then the DHCP response will itself be rejected.
Fix by requesting a broadcast response from the DHCP server if the
network interface already has any IPv4 addresses.
Michael Brown [Tue, 19 Jun 2012 17:54:22 +0000 (18:54 +0100)]
[romprefix] Treat 0xffffffff as an error return from PMM
PMM defines the return code 0xffffffff as meaning "unsupported
function". It's hard to imagine a PMM BIOS that doesn't support
pmmAllocate(), but apparently such things do exist.
Michael Brown [Sun, 10 Jun 2012 17:25:26 +0000 (18:25 +0100)]
[romprefix] Add a dummy ROM header to cover the .mrom payload
The header of a .mrom image declares its length to be only a few
kilobytes; the remainder is accessed via a sideband mechanism. This
makes it difficult to append an additional ROM image, such as an EFI
ROM.
Add a second, dummy ROM header covering the payload portion of the
.mrom image, allowing consumers to locate any appended ROM images in
the usual way.
Michael Brown [Sat, 9 Jun 2012 17:48:09 +0000 (18:48 +0100)]
[pending] Add concept of "pending operations"
iPXE is fundamentally asynchronous in operation: some operations
continue in the background even after the foreground has continued to
a new task. For example, the closing FIN/ACK exchanges of a TCP
connection will take place in the background after an HTTP download
has completed.
Michael Brown [Tue, 22 May 2012 11:23:34 +0000 (12:23 +0100)]
[http] Defer processing response code until after receiving all headers
Some headers can modify the meaning of the response code. For
example, a WWW-Authenticate header can change the interpretation of a
401 Unauthorized response from "Access denied" to "Please
authenticate".
Michael Brown [Tue, 22 May 2012 00:09:34 +0000 (01:09 +0100)]
[crypto] Rename KEY= to PRIVKEY= and "key" to "privkey"
The setting name "key" conflicts with the setting name "key" already
in use by the 802.11 code. Resolve the conflict by renaming the newer
setting to "privkey".
Michael Brown [Wed, 16 May 2012 18:04:33 +0000 (19:04 +0100)]
[efi] Work around platforms which choke on EFI_PCI_DEVICE_ENABLE
EFI_PCI_DEVICE_ENABLE is a list of the standard attributes that must
be enabled for a PCI device to function: I/O cycles, memory cycles,
and bus-mastering. We currently call EFI_PCI_IO_PROTOCOL::Attribute()
with the parameter EFI_PCI_DEVICE_ENABLE to enable a PCI device. This
should translate to a single write to PCI configuration space.
Simplicity is not a virtue within the UEFI world. Some platforms will
'helpfully' report an error if EFI_PCI_DEVICE_ENABLE is used on a
device that doesn't actually support all three of the relevant
attributes. For example, if a PCI device provides only memory-mapped
accesses (and so hardwires the I/O enable bit to zero), then using
EFI_PCI_DEVICE_ENABLE on such a platform will result in an
EFI_UNSUPPORTED error.
There is no plausible use case in which it is useful for the platform
to return an error in this way, and doing so makes it impossible to
distinguish genuine errors from noise.
Work around this broken behaviour by attempting to enable the three
attributes individually, and ignoring any errors.
Michael Brown [Wed, 16 May 2012 09:50:56 +0000 (10:50 +0100)]
[iscsi] Report SCSI response only when applicable
iSCSI generally includes a full SCSI response only when an error
occurs. iscsi_scsi_done() currently passes the NULL response through
to scsi_response(), which ends up causing scsicmd_response() to
dereference a NULL pointer.
Fix by calling scsi_response() only if we have a non-NULL response.
Reported-by: Brendon Walsh <brendonwalsh@niamu.com> Tested-by: Brendon Walsh <brendonwalsh@niamu.com> Signed-off-by: Michael Brown <mcb30@ipxe.org>
Michael Brown [Wed, 9 May 2012 09:01:14 +0000 (10:01 +0100)]
[crypto] Reduce standard debugging output
X.509 certificate processing currently produces an overwhelming amount
of debugging information. Move some of this from DBGLVL_LOG to
DBGLVL_EXTRA, to make the output more manageable.
Automatically attempt to download any required cross-signing
certificates from http://ca.ipxe.org/auto, in order to enable the use
of standard SSL certificates issued by public CAs.
Michael Brown [Tue, 8 May 2012 11:10:01 +0000 (12:10 +0100)]
[crypto] Add asynchronous certificate validator
To allow for automatic download of cross-signing certificates and for
OCSP, the validation of certificates must be an asynchronous process.
Create a stub validator which uses a job-control interface to report
the result of certificate validation.
Michael Brown [Tue, 8 May 2012 10:12:09 +0000 (11:12 +0100)]
[crypto] Add x509_auto_append()
CMS includes an unordered certificate set, from which certificates
must be extracted in order by matching up issuers with subjects. We
will use the same functionality as part of the automatic download of
cross-signing certificates. Generalise cms_find_subject() to
x509_find_subject(), and create x509_auto_append().
Michael Brown [Fri, 4 May 2012 16:12:32 +0000 (17:12 +0100)]
[crypto] Allow certificate chains to be long-lived data structures
At present, certificate chain validation is treated as an
instantaneous process that can be carried out using only data that is
already in memory. This model does not allow for validation to
include non-instantaneous steps, such as downloading a cross-signing
certificate, or determining certificate revocation status via OCSP.
Redesign the internal representation of certificate chains to allow
chains to outlive the scope of the original source of certificates
(such as a TLS Certificate record).
Allow for certificates to be cached, so that each certificate needs to
be validated only once.
Michael Brown [Fri, 4 May 2012 16:12:49 +0000 (17:12 +0100)]
[malloc] Allow Valgrind to be used when all assertions are enabled
The free-memory-block traversal code triggers multiple warnings from
Valgrind when assertions are enabled, since the list consistency
checks performed by list_check() end up accessing areas that have been
marked as inaccessible.
Fix by ensuring that any memory areas that will be accessed by
list_check() are marked as defined when necessary.
Michael Brown [Sat, 28 Apr 2012 14:47:46 +0000 (15:47 +0100)]
[realtek] Support RTL8139 cards within generic Realtek driver
RTL8139C+ cards use essentially the same datapath as RTL8169, which is
zerocopy and 64-bit capable. Older RTL8139 cards use a single receive
ring buffer rather than a descriptor ring, but still share substantial
amounts of functionality with RTL8169.
Include support for RTL8139 cards within the generic Realtek driver,
since there is no way to differentiate between RTL8139 and RTL8139C+
cards based on the PCI IDs alone.
Many thanks to all the people who worked on the rtl8139 driver over
the years.
Michael Brown [Tue, 24 Apr 2012 15:10:22 +0000 (16:10 +0100)]
[crypto] Do not allow build-time cryptography settings to be overridden
If a root certificate has been explicitly specified at build time
using TRUST=/path/to/cert then do not allow this to be overridden even
from a trustworthy settings source (such as VMware GuestInfo).
Similarly, if a client certificate (and private key) has been
explicitly specified at build time, then do not allow it to be
overridden at runtime.
Michael Brown [Tue, 24 Apr 2012 12:11:21 +0000 (13:11 +0100)]
[build] Use $(xxx_DEPS) for additional dependency information
Some objects (embedded.o, rootcert.o, and clientcert.o) define
additional dependencies on external files, using syntax such as:
$(BIN)/clientcert.o : $(CERT_LIST)
This dependency can be missed when using debug builds. For example,
if DEBUG=clientcert is used, then the relevant object is
$(BIN)/clientcert.dbg1.o rather than $(BIN)/clientcert.o.
Fix by adding dependencies to $(clientcert_DEPS) instead:
The WinCE, a.out and FreeBSD loaders are designed to be #included by
core/loader.c, which no longer exists. These old loaders are not
usable anymore and cause compilation failures when enabled in
config/general.h.
Signed-off-by: Marin Hannache <mareo@mareo.fr> Signed-off-by: Michael Brown <mcb30@ipxe.org>
Michael Brown [Mon, 23 Apr 2012 22:26:29 +0000 (23:26 +0100)]
[http] Avoid using stack-allocated memory in http_step()
http_step() allocates a potentially large block of storage (since the
URI can be arbitrarily long), and can be invoked as part of an already
deep call stack via xfer_window_changed().
projects / thirdparty / ipxe.git / log
