[block] Retry reopening indefinitely for multipath devices
For multipath SAN devices, verify that the device is capable of being
opened (i.e. that all URIs are parseable and that at least one path is
alive) and thereafter retry indefinitely to reopen the device as
needed.
Signed-off-by: Michael Brown <mcb30@ipxe.org>
[block] Add a small delay between attempts to reopen SAN targets
When all SAN targets are completely unreachable, there will be a
natural delay between reopening attempts due to the network connection
timeout on the unreachable targets.
However, some SAN targets may accept connections instantly and report
a temporary unavailability by e.g. failing the TEST UNIT READY
command. If all targets are behaving this way then there will be no
natural delay, and we will attempt to saturate the network with
connection attempts.
Fix by introducing a small delay between attempts.
Signed-off-by: Michael Brown <mcb30@ipxe.org>
Allow the SAN retry count to be configured via the ${san-retry}
setting, defaulting to the current value of 10 retries if not
specified.
Note that setting a retry count of zero is inadvisable, since iSCSI
targets in particular will often report spurious errors such as "power
on occurred" for the first few commands.
Signed-off-by: Michael Brown <mcb30@ipxe.org>
[int13con] Avoid overwriting random portions of SAN boot disks
The INT13 console type (CONSOLE_INT13) autodetects at initialisation
time a magic partition to be used for logging iPXE console output. If
the INT13 drive number mapping is subsequently changed (e.g. because
iPXE was used to perform a SAN boot), then the console logging output
will be written to the incorrect disk.
Fix by recording the INT13 vector at initialisation time, and using
this original vector to emulate INT13 calls for all subsequent
accesses. This should be robust against drive remapping performed
either by ourselves or by another bootloader (e.g. a chainloaded
undionly.kpxe which then performs a SAN boot).
Signed-off-by: Michael Brown <mcb30@ipxe.org>
[int13] Improve geometry guessing for unaligned partitions
Some partition tables have partitions that are not aligned to a
cylinder boundary, which confuses the current geometry guessing logic.
Enhance the existing logic to ensure that we never reduce our guesses
for the number of heads or sectors per track, and add extra logic to
calculate the exact number of sectors per track if we find a partition
that starts within cylinder zero.
Signed-off-by: Michael Brown <mcb30@ipxe.org>
Add basic support for multipath block devices. The "sanboot" and
"sanhook" commands now accept a list of SAN URIs. We open all URIs
concurrently. The first connection to become available for issuing
block device commands is marked as the active path and used for all
subsequent commands; all other connections are then closed. Whenever
the active path fails, we reopen all URIs and repeat the process.
Signed-off-by: Michael Brown <mcb30@ipxe.org>
Add a dummy SAN device which allows the "sanhook" command to be tested
even when no SAN booting capability is present on the platform. This
allows substantial portions of the SAN boot code to be run in Linux
under Valgrind.
Signed-off-by: Michael Brown <mcb30@ipxe.org>
[scsi] Avoid duplicate call to scsicmd_close() on TEST UNIT READY failure
When the TEST UNIT READY command receives an error response, the
shutdown of the command's block data interface will result in
scsidev_ready() closing the SCSI device. This will subsequently
result in a duplicate call to scsicmd_close(), leading to an assertion
failure when list_del() is called for the second time.
Fix by removing the command from the list of outstanding commands
before shutting down the command's interfaces.
Signed-off-by: Michael Brown <mcb30@ipxe.org>
[iobuf] Increase minimum I/O buffer size to 128 bytes
The eIPoIB translation layer needs to translate outbound ARP packets
from Ethernet to IPoIB. A 64-byte buffer (starting with the Ethernet
header) does not provide enough tailroom to expand to hold the two
20-byte IPoIB MAC addresses. The result is that an UNDI API user will
be unable to send ARP packets.
We could potentially shuffle the packet contents to reuse the space
occupied by the stripped Ethernet link-layer header, but this would
add complexity. Instead, fix by increasing the minimum allocation
size to 128 bytes.
Signed-off-by: Michael Brown <mcb30@ipxe.org>
B0_CTST is a 24bit register according to the vendor driver (sk98lin).
A 16bit read on B0_CTST will always return 0 for Y2_VAUX_AVAIL
(1<<16), so use a 32bit read when testing Y2_VAUX_AVAIL.
[This patch is copied directly from the Linux kernel tree.]
Signed-off-by: Mike McCormack <mikem@ring3k.org>
Signed-off-by: Michael Brown <mcb30@ipxe.org>
The value of ( ( x & 0x0c00 ) | 0x0c00 ) is always 0x0c00 regardless
of the value of x, and so the read_csr() is redundant. (There are no
read side effects for this register, according to the datasheet.)
This line of code originated in Linux kernel 2.3.19pre1 as
a->write_csr(ioaddr, 80, a->read_csr(ioaddr, 80) | 0x0c00);
and was modified in kernel 2.3.41pre4 to read
a->write_csr(ioaddr, 80, (a->read_csr(ioaddr, 80) & 0x0C00) | 0x0c00);
In the absence of commit messages, the intention of the code is
unclear. However, the logic resulting in a fixed value of 0x0c00 has
remained unaltered for over 17 years, and can probably be assumed to
have the correct overall result.
Signed-off-by: Michael Brown <mcb30@ipxe.org>
Track the current and maximum heap usage, and display the maximum
during shutdown when DEBUG=malloc is enabled.
Signed-off-by: Michael Brown <mcb30@ipxe.org>
Our asprintf() implementation guarantees that strp will be NULL on
allocation failure, but this is not standard behaviour. Detect errors
by checking for a negative return value instead of a NULL pointer.
Signed-off-by: Michael Brown <mcb30@ipxe.org>
Avoid potential division by zero when performing the check against
multiplication overflow. (Note that if the width is zero then there
can be no overflow anyway, so it is then safe to bypass the check.)
Signed-off-by: Michael Brown <mcb30@ipxe.org>
[infiniband] Return status code from ib_create_cq() and ib_create_qp()
Any underlying errors arising during ib_create_cq() or ib_create_qp()
are lost since the functions simply return NULL on error. This makes
debugging harder, since a debug-enabled build is required to discover
the root cause of the error.
Fix by returning a status code from these functions, thereby allowing
any underlying errors to be propagated.
Signed-off-by: Michael Brown <mcb30@ipxe.org>
[build] Avoid confusing sparse in single-argument DBG() macros
For visual consistency with surrounding lines, the definitions of
DBG_MORE(), DBG_PAUSE(), etc include an unnecessary ##__VA_ARGS__
argument which is always elided. This confuses sparse, which
complains about DBG_MORE_IF() being called with more than one
argument.
Work around this problem by adding an unused variable argument list to
the single-argument macros DBG_MORE_IF() and DBG_PAUSE_IF().
Signed-off-by: Michael Brown <mcb30@ipxe.org>
[eoib] Avoid passing a NULL I/O buffer to netdev_tx_complete_err()
Report errors in eoib_duplicate() via netdev_tx_err() rather than
netdev_tx_complete_err(), since netdev_tx_complete_err() accepts only
valid I/O buffers that are currently in the network device's transmit
queue.
Signed-off-by: Michael Brown <mcb30@ipxe.org>
Michael Brown