gPXE currently overwrites the filename stored in the cached DHCP
packets when a call to PXENV_TFTP_READ_FILE or PXENV_RESTART_TFTP is
made. This code has existed for many years as a workaround for RIS,
which seemed to require that this be done.
pxe_set_cached_filename() causes problems with the Bootix NBP, and a
recent test demonstrates that RIS will complete successfully even with
pxe_set_cached_filename() removed. There have been many changes to
the DHCP and PXE logic since this code was first added, and it is
quite plausible that it was masking a bug that no longer exists.
Reported-by: Alex Zeffertt <alex.zeffertt@eu.citrix.com>
Debugged-by: Shao Miller <Shao.Miller@yrdsb.edu.on.ca>
Signed-off-by: Michael Brown <mcb30@etherboot.org>
Current gPXE code always returns "OURS" in response to
PXENV_UNDI_ISR:START. This is harmless for non-shared interrupt
lines, and avoids the complexity of trying to determine whether or not
we really did cause the interrupt. (This is a non-trivial
determination; some drivers don't have interrupt support and hook the
system timer interrupt instead, for example.)
A problem occurs when we have a shared interrupt line, the other
device asserts an interrupt, and the controlling ISR does not chain to
the other device's ISR when we return "OURS". Under these
circumstances, the other device's ISR never executes, and so the
interrupt remains asserted, causing an interrupt storm.
Work around this by returning "OURS" if and only if our net device's
interrupt is currently recorded as being enabled. Since we always
disable interrupts as a result of a call to PXENV_UNDI_ISR:START, this
guarantees that we will eventually (on the second call) return "NOT
OURS", allowing the other ISR to be called. Under normal operation,
including a non-shared interrupt situation, this change will make no
difference since PXENV_UNDI_ISR:START would be called only when
interrupts were enabled anyway.
Signed-off-by: Michael Brown <mcb30@etherboot.org>
The interrupt control mechanism on Phantom cards has changed
substantially since the driver was initially written. This updates
the code to match the mechanism used in production firmware.
This is sufficient to allow DOS wget to function successfully using
the 3Com UNDI/NDIS, Intel UNDI/NDIS, and UNDIPD.COM UNDI/PD stacks.
Signed-off-by: Michael Brown <mcb30@etherboot.org>
Some BIOSes (observed with an AMI BIOS on a SunFire X2200) seem to
reset the BIOS drive counter at 40:75 after a failed boot attempt.
This causes problems when attempting a Windows direct-to-iSCSI
installation: bootmgr.exe calls INT 13,0800 and gets told that there
are no hard disks, so never bothers to read the MBR in order to obtain
the boot disk signature. The Windows iSCSI initiator will detect the
iBFT and connect to the target, and everything will appear to work
except for the error message "This computer's hardware may not support
booting to this disk. Ensure that the disk's controller is enabled in
the computer's BIOS menu."
Fix by checking the BIOS drive counter on every INT 13 call, and
updating it whenever necessary.
The case of an unsupported SAN protocol will currently not result in
any error message. Fix by printing the error message at the top level
using strerror(), rather than using hard-coded error messages in the
error paths.
The latest RTL-generated register lists include (mostly redundant)
xxx_MSB values alongside xxx_LSB and xxx_RMASK, and also include
default register values.
Some subnet managers expect the GetResponse from a SetPortInfo MAD to
contain the new link state. The transition is not immediate, so we
often end up returning the previous link state. This can cause the SM
to fail to activate the port.
Fix by waiting for up to 20us for the link state transition to take
effect.
The first byte of the IPoIB MAC address is used for flags indicating
support for "connected mode". Strip out the non-QPN bits of the first
dword when constructing the address vector for transmitted IPoIB
packets, so as not to end up passing an invalid QPN in the BTH.
IBA section 14.2.5.2 states that "the contents of the NodeDescription
attribute are the same for all ports on a node". Satisfy this by
using the HCA GUID rather than the port GUID to form the node
description string.
We do not discard routing table entries when closing an interface. It
is plausible that multiple interfaces may be on the same physical
network; if so, then we may end up in a situation whereby outbound
packets attempt to route via a closed interface.
Fix by ignoring non-open net devices in ipv4_route().
ipv4.c calculates the default subnet mask before calling
fetch_ipv4_setting() to retrieve the configured subnet mask (if any).
However, as of commit 612f4e7 "[settings] Avoid returning
uninitialised data on error in fetch_xxx_setting()",
fetch_ipv4_setting() will zero the IP address if the setting does not
exist, rather than leaving it unaltered.
Fix by fetching the setting first and calculating the default subnet
mask only if necessary.
ipv4.c uses a gateway address of INADDR_NONE to represent "no
gateway". It initialises the gateway address to INADDR_NONE before
calling fetch_ipv4_setting() to retrieve the configured gateway
address (if any).
However, as of commit 612f4e7 "[settings] Avoid returning
uninitialised data on error in fetch_xxx_setting()",
fetch_ipv4_setting() will zero the IP address if the setting does not
exist, rather than leaving it unaltered.
Fix by using a zero IP address to indicate "no gateway", so that a
non-existent gateway address setting will be treated as such.
The iBFT is Ethernet-centric in providing only six bytes for a MAC
address. This is most probably an indirect consequence of a similar
design flaw in the Windows NDIS stack. (The WinOF IPoIB stack
performs all sorts of contortions in order to pretend to the NDIS
layer that it is dealing with six-byte MAC addresses.)
There is no sensible way in which to extend the iBFT without breaking
compatibility with programs that expect to parse it. Add the notion
of an "Ethernet-compatible" MAC address to our link layer abstraction,
so that link layers can provide their own workarounds for this
limitation.
Some systems will retry their boot sequence in the event of a boot
failure. On these systems, the second and subsequent boot attempts
will fail to initialise the Hermon HCA.
Fix by resetting the HCA during probe(). This incurs a one-second
cost, but there seems to be no viable alternative.
Originally-fixed-by: Itay Gazit <itaygazit@gmail.com>
Some devices can only be reset via a mechanism that also resets the
card's PCI core, thus necessitating a backup and restore of all or
part of the PCI configuration space across a reset.
802.11 multicast hashing is the same as standard Ethernet hashing, so
just expose and use eth_mc_hash().
Signed-off-by: Joshua Oreman <oremanj@rwcr.net>
For IPoIB, the chaddr field is too small (16 bytes) to contain the
20-byte IPoIB link-layer address. RFC4390 mandates that we should
pass an empty chaddr field and rely on the DHCP client identifier
instead. This has many problems, not least of which is that a client
identifier containing an IPoIB link-layer address is not very useful
from the point of view of creating DHCP reservations, since the QPN
component is assigned at runtime and may vary between boots.
Leave the DHCP client identifier as-is, to avoid breaking existing
setups as far as possible, but expose the real hardware address (the
port GUID) via the DHCP chaddr field, using the broadcast flag to
instruct the DHCP server not to use this chaddr value as a link-layer
address.
This makes it possible (at least with ISC dhcpd) to create DHCP
reservations using host declarations such as:
host duckling {
fixed-address 10.252.252.99;
hardware unknown-32 00:02:c9:02:00:25:a1:b5;
}
IPoIB has a 20-byte link-layer address, of which only eight bytes
represent anything relating to a "hardware address".
The PXE and EFI SNP APIs expect the permanent address to be the same
size as the link-layer address, so fill in the "permanent address"
field with the initial link layer address (as generated by
register_netdev() based upon the real hardware address).
The hardware address is an intrinsic property of the hardware, while
the link-layer address can be changed at runtime. This separation is
exposed via APIs such as PXE and EFI, but is currently elided by gPXE.
Expose the hardware and link-layer addresses as separate properties
within a net device. Drivers should now fill in hw_addr, which will
be used to initialise ll_addr at the time of calling
register_netdev().
The SRP Boot Firmware Table serves a similar role to the iSCSI and AoE
Boot Firmware Tables; it provides information required by the loaded
OS in order to establish a connection back to the SRP boot device.
There is diagnostic value in being able to disambiguate between the
various reasons why an IB CM has rejected a connection attempt. In
particular, reason 8 "invalid service ID" can be used to identify an
incorrect SRP service_id root-path component, and reason 28 "consumer
reject" corresponds to a genuine SRP login rejection IU, which can be
passed up to the SRP layer.
For rejection reasons other than "consumer reject", we should not pass
through the private data, since it is most likely generated by the CM
without any protocol-specific knowledge.
With iSCSI, connection attempts are expensive; it may take many
seconds to determine that a connection will fail. SRP connection
attempts are much less expensive, so we may as well avoid the
"optimisation" of declaring a state of permanent failure after a
certain number of attempts. This allows a gPXE SRP initiator to
resume operations after an arbitrary amount of SRP target downtime.
Generate errors within individual MAD transaction consumers such as
ib_pathrec.c and ib_mcast.c, rather than within ib_mi.c. This allows
for more meaningful error messages to eventually be displayed to the
user.
SRP is the SCSI RDMA Protocol. It allows for a method of SAN booting
whereby the target is responsible for reading and writing data using
Remote DMA directly to the initiator's memory. The software initiator
merely sends and receives SCSI commands; it never has to touch the
actual data.
The minimal-surprise behaviour, when no explicit SRP initiator device
is specified, will probably be to use the most recently opened
Infiniband device. This matches our behaviour with using the most
recently opened net device for PXE, iSCSI, AoE, NBI, etc.
SRP over Infiniband uses a protocol whereby data is sent via a
combination of the CM private data fields and the RC queue pair
itself. This seems sufficiently generic that it's worth having
available as a separate protocol.
The ACK timeout determines how long we take to notice a failed
Reliable Connection. Reducing it from the arbitrary value of 19 down
to 14 reduces the individual ACK timeout from around 2.1s to 67ms;
this in turn reduces the time to tear down and re-establish a broken
SRP session from around 30s to around 1s.
The Infiniband Communication Manager will refuse to establish a
connection if it believes the connection is already established.
There is no immediately obvious way to ask it to tear down the
existing connection and replace it; to issue a DREP we would need to
know the local and remote communication IDs used for the previous
connection setup.
We can work around this by randomising the high-order bits of the
queue pair number; these have no significance to the hardware, but are
sufficient to convince the IB CM that this is a different connection.
We will terminate our transaction as soon as we receive the first CM
REP, since that provides all the state that we need. However, the
peer may resend the REP if it didn't see our RTU, and if we don't
respond with another RTU we risk being disconnected. (This protocol
appears not to handle retries gracefully.)
Fix by adding a management agent that will listen for these duplicate
REPs and send back an RTU.
A management interface is the component through which both local and
remote management agents are accessed.
This new implementation of a management interface allows for the user
to react to timed-out transactions, and also allows for cancellation
of in-progress transactions.
Some BIOSes support the BIOS Boot Specification (BBS) but fail to set
%es:%di correctly when calling the option ROM initialisation entry
point. This causes gPXE to identify the BIOS as non-PnP (and so
non-BBS), leaving the user unable to control the boot order.
Fix by scanning for the $PnP signature ourselves, rather than relying
on the BIOS having passed in %es:%di correctly.
Tested-by: Helmut Adrigan <helmut.adrigan@chello.at>
The IBA specification refers to management "interfaces" and "agents".
The interface is the component that connects to the queue pair and
sends and receives MADs; the agent is the component that constructs
the reply to the MAD.
Rename the IB_{QPN,QKEY,QPT} constants as a first step towards making
this separation in gPXE.
The function __intel_new_proc_init() (called implicitly when building
using icc) is marked with __attribute__((cdecl)). This breaks
building on x86_64, where cdecl is meaningless.
Fix by replacing with the existing __libgcc macro, which is already
defined to be "__attribute__((cdecl))" for i386 builds and empty for
x86_64 builds.
In several places, we currently use size_t to represent a difference
between TCP sequence numbers. This can cause compiler warnings
relating to printf format specifiers, since the result of
(uint32_t+size_t) may be an unsigned long on some compilers.
Fix by using uint32_t for all variables that represent a difference
between TCP sequence numbers.
Tested-by: Joshua Oreman <oremanj@xenon.get-linux.org>
The geniso, genliso and gensdsk scripts contain hard-coded temporary
directory names, and so could potentially collide with each other when
run as part of a concurrent build (e.g. "make -j 4").
Fix by using mktemp to generate suitable temporary directory names.
The Linux IB Communication Manager will always send MADs to QP1,
rather than back to the originating QP. On Hermon, QP1 is by default
handled by the embedded firmware. We can change this, but the cost is
that we have to handle both QP0 and QP1 (i.e. we have to provide SMA
as well as GMA service in software), and we have to use MLX queues
rather than standard UD queues (i.e. we have to construct the UD
datagrams by hand).
There doesn't seem to be any viable way around this situation, ugly
though it is.
Queue pairs are now assumed to be created in the INIT state, with a
call to ib_modify_qp() required to bring the queue pair to the RTS
state.
ib_modify_qp() no longer takes a modification list; callers should
modify the relevant queue pair parameters (e.g. qkey) directly and
then call ib_modify_qp() to synchronise the changes to the hardware.
The packet sequence number is now a property of the queue pair, rather
than of the device.
Each queue pair may have an associated address vector. For RC queue
pairs, this is the address vector that will be programmed in to the
hardware as the remote address. For UD queue pairs, it will be used
as the default address vector if none is supplied to ib_post_send().
Now that MAD handlers no longer return a status code, we can allow
them to return a pointer to a MAD structure if and only if they want
to send a response. This provides a more natural and flexible
approach than using a "response method" field within the handler's
descriptor.
MAD handlers have to set the status fields within the MAD itself
anyway, in order to provide a meaningful response MAD; the additional
gPXE return status code is just noise.
Note that we probably don't need to ever explicitly set the status to
IB_MGMT_STATUS_OK, since it should already have this value from the
request. (By not explicitly setting the status in this way, we can
safely have ib_sma_set_xxx() call ib_sma_get_xxx() in order to
generate the GetResponse MAD without worrying that ib_sma_get_xxx()
will clear any error status set by ib_sma_set_xxx().)
Most IB hardware seems not to allow allocation of the genuine QPNs 0
and 1, so allow for the externally-visible QPN (as constructed and
parsed by ib_packet, where used) to differ from the real
hardware-allocated QPN.
The queue key is stored as a property of the queue pair, and so can
optionally be added by the Infiniband core at the time of calling
ib_post_send(), rather than always having to be specified by the
caller.
This allows IPoIB to avoid explicitly keeping track of the data queue
key.
Now that path record lookups are handled entirely via
ib_resolve_path(), the only role of the IPoIB peer cache is as a
lookup table for MAC addresses. Update the code structure and
comments to reflect this.
The IPoIB broadcast MAC address varies according to the partition key.
Now that the broadcast MAC address is a property of the network device
rather than of the link layer, we can expose this real MAC address
directly.
The broadcast LID is now identified via a path record lookup; this is
marginally inefficient (since it was present in the MCMemberRecord
GetResponse), but avoids the need to special-case broadcasts when
constructing the address vector in ipoib_transmit().
Generalise the subnet management agent into a general management agent
capable of sending and responding to MADs, including support for
retransmissions as necessary.
Currently, all Infiniband users must create a process for polling
their completion queues (or rely on a regular hook such as
netdev_poll() in ipoib.c).
Move instead to a model whereby the Infiniband core maintains a single
process calling ib_poll_eq(), and polling the event queue triggers
polls of the applicable completion queues. (At present, the
Infiniband core simply polls all of the device's completion queues.)
Polling a completion queue will now implicitly refill all attached
receive work queues; this is analogous to the way that netdev_poll()
implicitly refills the RX ring.
Infiniband users no longer need to create a process just to poll their
completion queues and refill their receive rings.
IPoIB and the SMA have separate constants for the packet size to be
used to I/O buffer allocations. Merge these into the single
IB_MAX_PAYLOAD_SIZE constant.
(Various other points in the Infiniband stack have hard-coded
assumptions of a 2048-byte payload; we don't currently support
variable MTUs.)
IPoIB has a link-layer broadcast address that varies according to the
partition key. We currently go through several contortions to pretend
that the link-layer address is a fixed constant; by making the
broadcast address a property of the network device rather than the
link-layer protocol it will be possible to simplify IPoIB's broadcast
handling.
Move the icky call to step() from aoe.c to ata.c; this takes it at
least one step further away from where it really doesn't belong.
Unfortunately, AoE has the ugly aoe_discover() mechanism which means
that we still have a step() loop in aoe.c for now; this needs to be
replaced at some future point.
Michael Brown