Release Notes
QLogic uIP Linux Driver
Version 0.7.8.6
06/27/2019
QLogic Corporation
26650 Aliso Viejo Pkwy,
Aliso Viejo, CA 92656
Copyright (c) 2004 - 2013 Broadcom Corporation
Copyright (c) 2014, QLogic Corporation
All rights reserved
uIP v0.7.8.6 (Jun 27, 2019)
=======================================================
Fixes:
-------
1. Problem: OS fails to boot after one path is
disconnected from iSCSI MPIO config.
Change: In the event of DHCP failure, killing of enable_nic_thread did
not process any iscsid requests leading to error,
iscsistart: Could not broadcast to uIP after 5 tries
and login failure of active path.
Added fix to not kill enable_nic_thread and allow further
processing of iscsid requests and performing login
to next active path.
Impact: All
uIP v0.7.8.5 (Nov 20, 2018)
=======================================================
Fixes:
-------
1. Problem: CQ102578: observing ISCSI initiator IP ping drop
Change: 1. Do not flush tx queue on each uio interrupt
2. Use UIO BD index instead on buffer index.
3. Set buf_size in case of ICMP and ARP packet
Impact: QL41xxx adapters
2. Problem: CQ103034 - Unable to boot iSCSI BFS in IPv6 DHCP config
Change: Limit retries of performing dhcpv6 before declaring dhcp failure
Impact: All
3. Problem: CQ102438: I/O fails to resume on multipath LUN during port toggle.
Change: lib/cnic, lib/qedi, Release xmit_mutex in error code path and
during clear tx queue.
Impact: QL84xx adapters
4. Problem: Netlink buffer corruption when more than one host
try to xmit packet at the same time
Change: Add inter-host mutex while doing xmit
Impact: All
uIP v0.7.8.4 (Feb 22, 2018)
=======================================================
Fixes:
-------
1. Problem: CQ95605: iSCSI BFS in DHCP config intermittently fails to boot
into the OS when source and destination addresses are in
different networks.
Change: Allow ARP for non-matching source and destination addresses.
For source and destination IP addresses in different networks,
continue with the ARP retries and further login process
instead of assuming abrupt failure. iSCSI offload adapters
may not rely on netmask information for successful iSCSI
target login.
Impact: All
uIP v0.7.8.3 (May 18, 2017)
=======================================================
Fixes:
-------
1. Problem: CQ93985: iscsiuio seg faults if discovery done to not
reachable target
Change: Serialize xmit_mutex lock to prevent iscsiuio seg fault.
Impact: All
2. Problem: CQ91497 - Initiator fails to acquire IPv6 DHCP address
from the DHCP server
Change: Initialize the transaction-id within the dhcpv6 packet with
correct byte order, to fix the trans-id mismatch error.
Impact: All
3. Problem: Missing qedi ping transport hook
Change: Add qedi ping transport hook
Impact: 10/25/40/50GGbE Controller (iSCSI)
uIP v0.7.8.3 (Sept 28, 2016)
=======================================================
Enhancements
------------
1. Change: Add support for the new qedi transport
Impact: 10/25/40/50GGbE Controller (iSCSI)
uIP v0.7.8.2 (Dec 10, 2013)
=======================================================
Fixes
-----
1. Problem: Cont00072053 - Some hardware iSCSI paths fail during test
Cause: The test exercised a corner case where the ARP cache flush
mechanism didn't work properly
Change: Fixed the ARP cache flush mechanism
Impact: All
Enhancements
------------
1. Change: Added a new tx doorbell field in the uio path to work with
the new bnx2x/cnic drivers that supports VF_RSS
Impact: 10G only
2. Change: Fixed the iface.subnet_mask decoding for IPv6
Impact: IPv6
uIP v0.7.8.1b (May 01, 2013)
=======================================================
Enhancements
------------
1. Change: Performance optimization by caching the page size
Impact: All
2. Change: Fixed a bug in the tx completion interrupt handler
Impact: 10G only
uIP v0.7.6.1g (Jan 14, 2013)
=======================================================
Fixes
-----
1. Problem: Cont00067316 - IPv6 address prefix length < 32
bits fails to connect
Cause: CIDR notation has an order bug in the IPv6 section
whenever the prefix length specified is < 32
Change: Fixed the network order bug
Impact: IPv6 only
uIP v0.7.6.1f (Nov 14, 2012)
=======================================================
Fixes
-----
1. Problem: Cont00065768 - RHEL5.X iscsiuio segfault possible
if there is a specific 1024 byte size broadcast
packet
Cause: This is another corner case where the packet size
is also exactly 1024 bytes + padding that exceeded
the DMA rx buffer. The previous fix was not
sufficient
Change: Ensure that the packet size + padding do not
exceed this limit.
Impact: 10G only. 1G already has the guard against it.
uIP v0.7.6.1e (Nov 07, 2012)
=======================================================
Fixes
-----
1. Problem: Cont00066397 - Unable to connect to iSCSI target
with NPAR enabled on 57840
Cause: The PCI device ID for 57840_MF has been changed from
0x16ab to 0x16a4
Change: Updated the PCI id table to match exactly what the
bnx2x 1.76 indicates
Impact: 57840 MF
uIP v0.7.6.1d (Oct 31, 2012)
=======================================================
Enhancements
------------
1. Change: Added support for open-iscsi-2.0.873
Impact: All
uIP v0.7.6.1c (Oct 15, 2012)
=======================================================
Enhancements
------------
1. Change: Added support for 10G 57840 4x10 and 2x20
Impact: 10G 57840
uIP v0.7.6.1b (Oct 09, 2012)
=======================================================
Fixes
-----
1. Problem: Cont00065690 - Vconfig method of connecting over
tagged vlan with IPv6 failed
Cause: The new net param support changes has prevented
the old vconfig method from execising the IPv6
acquisition engine properly
Change: Ensure that this old vconfig method to run the IPv6
acquisition engine properly and to its entirety
Impact: IPv6 + VLAN using the network VLAN configuration
method
2. Problem: Cont00065768 - RHEL5.X iscsiuio segfault possible
if there is a specific 1024 byte size broadcast
packet
Cause: This is a corner case where the packet size is
exactly 1024 bytes + padding that exceeded the
DMA rx buffer. This has been there since day 1.
Change: Ensure that the packet size + padding do not
exceed this limit.
Impact: 10G only. 1G already has the guard against it.
Enhancements
------------
1. Change: Source optimization - backported source code fixes
as reported from the upstream submission patch
Impact: ALL
uIP v0.7.4.2k (Aug 10, 2012)
=======================================================
Enhancements
------------
1. Change: Enable HP SD mode
Impact: 577XX/578XX
uIP v0.7.4.2j (Jul 18, 2012)
=======================================================
Fixes
-----
1. Problem: Cont00064665 - Linux iSCSI connects via gateway address
on the wrong subnet
Cause: The gateway address used was not checked against the
subnet mask specified before the ARP requests. Since
this behavior deters from how L2 operates, therefore,
a change was made to correct this.
Change: Added check of the gateway specified against the subnet
specified.
Impact: Static IPv4 operation
2. Problem: Cont00064722 - Linux iSCSI unable to force IPv6 LL
override (advanced iface parameters)
Cause: The override LL address was not being populated to the
IPv6 address database correctly
Change: Added this correctly to the IPv6 initialization
Impact: Static/DHCP IPv6 LL address override only
uIP v0.7.4.2i (Jul 11, 2012)
=======================================================
Fixes
-----
1. Problem: Cont00064604 - Fails to connect to routed IPv6 target
via RA
Cause: The default router IPv6 address was not being retrieved
correctly.
Change: Fixed the default router IPv6 address read
Impact: All
uIP v0.7.4.2h (Jun 15, 2012)
=======================================================
Fixes
-----
1. Problem: Cont00063863 - can't boot into offload image
when VLAN is enabled
Cause: During the iSCSI login exchange, certain iSCSI targets
will send an ARP request even though the TCP connection
has been made. The bug was in this ARP reply where
the local MAC was corrupted when VLAN is enabled.
Change: Fixed the ARP reply packet
Impact: All
uIP v0.7.4.2g (Jun 08, 2012)
=======================================================
Fixes
-----
1. Problem: Cont00063816 - The initiator is not able to connect
to the iSCSI targets over VLAN
Cause: The process packet routine did not consider the PCP
of the VLAN tag to be non-zero. This created a
mismatch when this VLAN tag was compared against the
nic_iface->vlan_id which doesn't include the PCP.
Change: Added the consideration of non-zero PCP
Impact: All
uIP v0.7.4.2f (Jun 04, 2012)
=======================================================
Fixes
-----
1. Problem: Cont00063626 - Static IPv6 does not connect when
the prefix len is not set explicitly
Cause: The IPv6 prefix length was not set correctly
for Static IPv6 operation when CIDR notation is
not specified
Change: Fixed the default prefix length
Impact: Static IPv6
2. Problem: Cont00063651 - Cannot connect to iSCSI targets
HP PTM/SF
Cause: Switch-Dependent mode + invalid Outer VLAN was
not supported
Change: Allow SD+invalid OV to fallback to SF operation mode
Impact: 5771X/578XX
uIP v0.7.4.2e (May 30, 2012)
=======================================================
Fixes
-----
1. Problem: Cont00063443 - Compilation error on SLES11sp1
Cause: The iface_num field was not defined
Change: Fixed all references to iface_num
Impact: SLES11sp1
2. Problem: Cont00063518 - HBA fails to connect across router
using iface.gateway address
Cause: The gateway override code did not populate the
address into the lower level engine
Change: Fixed the gateway override code
Impact: IPv4 Static IP operation
3. Problem: Cont00063567 - IPv6 LL and RA override does not work
Cause: The IPv6 LL/RA override addresses were overwritten
by the NDP engine
Change: Fixed the LL/RA override code
Impact: IPv6 operation
Enhancements
------------
1. Added support for jumbo MTU (independent from the L2 MTU)
uIP v0.7.4.2d (May 21, 2012)
=======================================================
Fixes
-----
1. Problem: Cont00063421 - Static IPv6 cannot connect via RA/LL
Cause: The router advertise and the linklocal address
were corrupted due to the override capabilities
added for the newer open-iscsi util
Change: Fixed the address override code
Impact: Static IPv6
Enhancements
------------
1. Allow VLAN tag = 1 (router management) to connect offload
uIP v0.7.4.2c (May 09, 2012)
=======================================================
Fixes
-----
1. Problem: RHEL BZ 734010/804580 - issues found by the Coverity
scan
Cause: 10 code issues were flagged for revision
Change: Fixed all area of concern
Impact: All
2. Problem: Cont00063177 - IPv4 DHCP with VLAN specification in
iface file gets wrong address
Cause: The DHCPv4 handler was not discriminating the VLAN tag
associated with the DHCP offers from multiple DHCP
servers
Change: Changed the DHCPv4 handler to drop DHCP offer packets
that doesn't match the VLAN tag of the intended DHCP
discovery packet
Impact: DHCPv4 operation
uIP v0.7.4.2b (May 01, 2012)
=======================================================
Fixes
-----
1. Problem: Cont00062993 - IPv6 DHCP with VLAN specification in
iface file gets wrong address
Cause: The DHCPv6 request was using the same DUID as always
so the non-VLAN DHCP server responded to our broadcast
instead
Change: Changed the DHCPv6 request DUID to link address + time
instead of link address alone
Impact: DHCPv6 operation
uIP v0.7.4.1j (Apr 24, 2012)
=======================================================
Fixes
-----
1. Problem: Cont00062805 - Cannot login to iSCSI targets on RHEL6.3
Cause: The problem was caused by a change made to the iface_rec
structure in the RHEL6.3 inbox open-iscsi util
Change: The new changes is now incorporated
Impact: All
uIP v0.7.4.1i (Apr 16, 2012)
=======================================================
Fixes
-----
1. Problem: Cont00062660 - Unable to login with VLAN iscsiuio
on RHEL6.2
Cause: The open-iscsi util in RHEL6.2 has a bug which
does not pass the correct iface_num to iscsiuio
Change: Added workaround to fall back to do the legacy
VLAN support if iface_num and vlan_id = 0
Impact: RHEL6.2
uIP v0.7.4.1h (Apr 13, 2012)
=======================================================
Enhancements
------------
1. Added support for the new iface_num field in the iscsi_uevent
path
2. Fixed bug in the nic_iface search engine based on iface_num
uIP v0.7.4.1g (Mar 22, 2012)
=======================================================
Fixes
-----
1. Problem: Cont00061869 - Unable to setup an offload iSCSI
connection with FLR/NPAR under ESX5.0:PDA
Cause: The physical function ID was previously extracted
from the sysfs of the VM which might not be consistent
to the actual physical setup due to the function
remapping in the hypervisor
Change: Read the physical function ID directly from the BAR0
ME register
Impact: All
2. Problem: Cont00062170 - IPv6 login/logout stress fails
Cause: The packet interrupt was lost after running the test
for a much longer period of time. A bug in the
packet processing routine was found to exit prematurely
Change: Fixed the packet processing routine to process all
packets before exiting
Impact: All
uIP v0.7.4.1f (Mar 19, 2012)
=======================================================
Fixes
-----
1. Problem: Cont00062170 - IPv6 login/logout stress fails
Cause: The packet buffer routine for IPv6 did not take
network order <-> host order into consideration
Change: Added a htons call to compensate for the ntohs pair
Impact: All
uIP v0.7.4.1e (Mar 08, 2012)
=======================================================
Fixes
-----
1. Problem: Cont00061978 - Load/unload stress test fails
Cause: The bnx2x open request was failing due to the module
request procedure. However, the open failure was
not being handled correctly.
Change: Fixed the device open error handling
Impact: 5771X/578XX
uIP v0.7.4.1d (Mar 02, 2012)
=======================================================
Fixes
-----
1. Problem: Cont00061708 - Unable to log into target after running
driver load/unload
Cause: A bug was introduced in the previous bug fix (CQ61459)
where a pthread_cond_broadcast call was erroneously
enabled
Change: Restored this back
Impact: All
uIP v0.7.4.1c (Feb 16, 2012)
=======================================================
Fixes
-----
1. Problem: Cont00061529 - Unable to connect to target after an
initial failed login attempt until iscsi service is
restarted
Cause: Upon a failed DHCPv4 acquisition due to the wrong VLAN
tag in the initial iface setup, any iscsid connect request
from the same NIC will get dropped due to a bug.
Change: Fixed the bug which prevented new iscsid connect requests
from getting honored
Impact: All
Enhancements
------------
1. Updated README
uIP v0.7.4.1b (Feb 08, 2012)
=======================================================
Fixes
-----
1. Problem: Cont00061513 - Unable to connect to target over VLAN
interface
Cause: The VLAN id was not properly passed back to the CNIC
driver for the offload request
Change: Fixed the VLAN id being passed back to the CNIC driver
Impact: All
uIP v0.7.4.1a (Feb 01, 2012)
=======================================================
Fixes
-----
1. Problem: Cont00049383 - No mechanism in iface file to support
gateway/routing
Change: Added support for the additional network parameters
as passed from the newer iscsi-util.
These parameters include:
IPv4: subnet_mask, gateway
IPv6: ipv6_linklocal, ipv6_router,
ipv6_autocfg, linklocal_autocfg, router_autocfg
VLAN: vlan_id, vlan_priority, vlan_state
Other: mtu, port
Impact: All
2. Problem: Cont00060806 - Unable to connect target using DHCP over
tagged VLAN
Change: DHCP+VLAN is a new feature enhancement that was added
alongside all other new iface parameters.
Impact: All
Enhancements
------------
1. Lock iscsid's connect request with path_req so connect requests
with DHCP/Static will no longer override each other
2. Fixed the if_down handler from global to nic specific
3. Fixed various synchronization issues
uIP v0.7.2.1e (Jan 05, 2012)
=======================================================
Fixes
-----
1. Problem: Cont00060734 - ifupdown-mtu change stress with active
session causes iscsiuio to fail
Change: Fixed a race condition between the nic enable thread
and when DHCP fails
Impact: All
uIP v0.7.2.1d (Dec 28, 2011)
=======================================================
Fixes
-----
1. Problem: Cont00060368 - segfault observed after failing both
mpio paths
Change: Various memory leaks were identified and resolved in
the nic cleanup path
Impact: All
uIP v0.7.2.1c (Dec 16, 2011)
=======================================================
Enhancements
------------
1. Change: Disable HP SD mode
uIP v0.7.2.1b (Dec 14, 2011)
=======================================================
Enhancements
------------
1. Change: Default iscsiuio logging to off. Use the '-d'
option to enable
uIP v0.7.0.14g (Oct 25, 2011)
=======================================================
Enhancements
------------
1. Change: Fixed the compilation under RHEL6.2
2. Change: Added oom_adjust call to prevent OOM Killer from killing
iscsiuio when memory is low
3. Change: Added mlockall setting to prevent page swap
uIP v0.7.0.14f (Oct 20, 2011)
=======================================================
Fixes
-----
1. Problem: Cont00058994 - DOS vulnerability in uip during UDP flood
Cause: The warning messages from the UDP handler was logging
at a rate faster than the log file logrotate rate
Therefore, the system's OOM eventually got kicked in to
start terminating running processes which includes iscsiuio
Change: Moved several UDP warning messages from the default log
level to the debug log level
Impact: All (minor)
2. Problem: Cont00059288 - Show segfault w/ SLES11 SP1 Xen kernel
Cause: The bnx2x chip_id was not read correctly from the PCIe BAR1
under the Xen kernel. The error was in the mmap area.
Change: Corrected the mmapping of the PCI MMIO space.
Impact: Xen kernels
Enhancements
------------
1. Change: Changed the log file open error to a warning and let
the daemon progress. This was only observed under iSCSI boot
uIP v0.7.0.14e (Sep 19, 2011)
=======================================================
Fixes
-----
1. Problem: Cont00058678 - Can not iboot target from ipv6 path
using VLAN
Cause: A bug was found in the path request path where the vlan
iface's protocol family was not used correctly in the
iface search
Change: This has been corrected
uIP v0.7.0.14d (Sep 16, 2011)
=======================================================
Fixes
-----
1. Problem: Cont00058602 - Can't iboot using IPv6 offload path
Cause: The bug was exposed by a fix in 0.7.0.14c where the
IPv6 router solicitation timeout exceeded the nic
enable thread timeout.
Change: The IPv6 router solicitation timeout has been adjusted
uIP v0.7.0.14c (Sep 01, 2011)
=======================================================
Fixes
-----
1. Problem: Cont00058256 - Sessions fail after loginstress to via
simultaneous ipv4 and ipv6 dhcp
Cause: Switching between DHCPv4/v6 coupled with VLAN exposed
a drawback in our nic_iface architecture design where
VLAN is not specified by iscsid.
Change: The code was optimized and improved the performance when
switching between DHCPv4/v6+VLAN. However, the ultimate
fix is to make use of the net config parameters introduced
in the newer open-iscsi util which will identify the
specific VLAN nic_iface to use.
Enhancements
------------
1. Change: Added support for bnx2x-1.71.00
uIP v0.7.0.14b (Aug 23, 2011)
=======================================================
Fixes
-----
1. Problem: Cont00057840 - RHEL6.2 inbox: Unable to connect to
targets with 5709
Cause: For cases when the bnx2/bnx2x driver gets removed, the
uio database that was built by cnic would have the device
->net reference removed. This has caused an unnecessary
timeout of 5s for each stale uio entry in the database.
Change: Adjusted the routine which seeks the device->net entry
to include more logic instead of hard waiting for 5s.
Enhancements
------------
1. Change: Added support for RHEL6.2 for out-of-box release
2. Change: Updated the man page with -h and -p info
3. Change: Updated the -h info
uIP v0.7.0.13 (Aug 10, 2011)
=======================================================
Fixes
-----
1. Problem: Cont00057768 - iscsiuio logrotate causes daemon failure
Cause: The logrotate script will send a SIGUSR1 signal to notify
the iscsiuio daemon of such action. However, the daemon
wasn't programmed to catch this signal.
Change: Restored the catching of this signal
uIP v0.7.0.12 (Aug 04, 2011)
=======================================================
Fixes
-----
1. Problem: Cont00050634 - brcm_iscsiuio Tainted: running IoZone,
Iometer and receiving a UDP flood on 3260
Cause: Upon iscsiuio termination, because of the UDP flood,
the nic thread will be busy servicing those UDP packets
while the signal handling thread will free up all nic
resources. The two threads were not in sync.
Change: Added a nic_remove_all routine to destroy all nic threads
before the nic resources get freed.
Enhancements
------------
1. Change: Fixed all warnings as reported by RHELS' Coverity testing.
uIP v0.7.0.11 (Aug 02, 2011)
=======================================================
Fixes
-----
1. Problem: Erroneous VLAN tag was being passed by iscsid for connect
request
Cause: The iscsid's iface_rec_t ipc message does not contain this
vlan field. This field was added in uIP for future vlan
support. Since the buffer allocated to receive such message
in uIP didn't get initialized, therefore, garbled up VLAN
tag was getting used.
Change: Added the initialization of this buffer.
uIP v0.7.0.10 (Jul 26, 2011)
=======================================================
Fixes
-----
1. Problem: Can't offload when switching from Static to DHCP then back to
Static IPv4 when connecting through a VLAN interface
Cause: The VLAN processing code did not reinstall the IP address
from the default nic_iface to the associated VLAN nic_iface.
This was only done on the very first time when the VLAN
interface was created and not on subsequent instances.
Change: Added code to mirror the default nic_iface IP/netmask/ip_config
on the VLAN nic_iface on every new connection request.
uIP v0.7.0.9 (Jul 19, 2011)
=======================================================
Fixes
-----
1. Problem: Can't offload to 57810 NPAR NIC
Cause: The MF/VF variant of the PCI IDs were not supported previously
Change: Added support for the MF/VF variants for 57800/57810/57840
uIP v0.7.0.8 (Jun 30, 2011)
=======================================================
Fixes
-----
1. Problem: Cont00056522 - Unable to connect to iSCSI target using
netxtreme2 package 7.0.9
Cause: The iSCSI L2 ring's CID has changed from 17 to 49
Change: The code now gets L2 iSCSI ring CID from the l2_buf directly.
This will work with any version of the cnic driver because
the location is a zero before this change.
uIP v0.7.0.7 (Jun 23, 2011)
=======================================================
Fixes
-----
1. Problem: Cont00056460 - iSCSI Offload boot RHEL5u5 x64 dropped tagged
packets with iSCSI Offload Boot with untagged
Cause: The ICMP echo replies to the target was corrupted in both
1g and 10g mode
Change: The code will now handle both VLAN stripped and no VLAN stripped
incoming packets correctly. Also modified the transmit routine
to strip out any inline VLAN tag before setting up the hw to
insert VLAN tag.
uIP v0.7.0.6 (Jun 21, 2011)
=======================================================
Fixes
-----
1. Problem: Cont00056231 - DHCPv4 not working with iSCSI HBA w/
linux-nx2 v7.0.7
Cause: The 10g L2 FW HSI has been modified for PCIe performance
enhancement in the 7.0.7 package (FW 1.70.20) which uIP
has not adapted to.
Change: The eth_rx_cqe size has been increased from 32B to 64B.
Enhancements
------------
1. Change: The utility name has changed from brcm_iscsiuio to iscsiuio
as preparation for upstream submission.
2. Change: Updated README
uIP v0.7.0.5 (Jun 02, 2011)
=======================================================
Fixes
-----
1. Problem: Cont00055915 - iSCSI does not connect on 57800 in 4-port mode
Cause: The 4-port mode was not being determined correctly
Change: Fixed the PORT4MODE register offset and the QZONE_ID macros
uIP v0.7.0.4 (May 24, 2011)
=======================================================
Fixes
-----
1. Problem: Cont00055832 - linux iscsiboot can not login to target using
offload path (57800)
Cause: The device ID comparison routine did not take care of the case
when one device ID is bitwise superset of another.
Change: Fixed the device ID comparison routine.
uIP v0.7.0.3 (May. 19, 2011)
=======================================================
Enhancements
------------
1. Change: Updated all fixes to match the released uIP 0.6.4.17
2. Change: Modified source and Copyright info as preparation for upstream
submission
uIP v0.7.0.2 (May. 03, 2011)
=======================================================
Fixes
-----
1. Problem: Cont00048972 - brcm-iscsi.log has no max size and would grow
to consume all free space on hard disk
Cause: There was no mechanism to rotate the log
Change: Added logrotate entry and SIGUSR1 signal handling for log rotate
action
2. Problem: Cont00054996 - Multi-session, multi-protocol mtu stress
does not recover all sessions
Cause: A segfault was observed during the load/unload module. The
problem was caused by an illegal dereference of a pointer
when IPv6 couldn't find the longest match address from
the ARP (Neighbor) table.
Change: Fixed the dereferencing error
3. Problem: Cont00054900 - Linux uIP - Please add ability to connect
to routed target with static iface IPv6
Cause: Static IPv6 never runs the IPv6 NDP router sol/adv engine.
Change: IPv6 NDP router sol/adv has now been added to static IPv6
operation.
4. Problem: Cont00054996 - Multi-session, multi-protocol mtu stress
does not recover all sessions
Cause: Segfaults were observed caused by the accessing of the IPv6
NDP structure while the nic is undergoing a reset either
due to a DHCPv4 request from iscsid or the handling of
if_down due to the NL handler from CNIC.
Change: The fix involves the following:
- Fixed the handling of staggered IPv4/v6 DHCP/static requests
- Fixed memory leak due to reallocation of IPv4 and IPv6
DHCP structs
- Fixed the pthread join stuck problem in the handling
of the if_down NL message
5. Problem: Cont00054810 - Linux NMI - bnx2x_init_hw_common:PXP2 CFG
failed running iSCSI MTU stress test
Cause: This only happens in DHCPv4 mode. The problem was caused
by contention between the elongated window of performing
DHCP in the enable_nic thread while receiving the asynchronous
if_down NL message (from the MTU change event) from the
CNIC NL thread. The problem occurs when the enable_nic
thread tries to call bnx2x_open while the other thread
calls the bnx2x_close routine.
Change: Fixed mutex lock bugs for the enable_nic thread. Also
extended the nic_disable timeout to 10s to compensate for
the DHCP operation.
6. Problem: Cont00054818 - RH6.0 - Unable to logout of iSCSI session
after running PQA baseline scripts
Cause: This was caused by the call to cancel the enable_nic
thread when disabling the nic but failed to unlock the
nic mutex that the enable_nic thread held.
Change: Wake up the enable_nic thread and wait for it to complete
instead of canceling it in the nic_disable path.
7. Problem: Cont00054725 - Previous static HBA IP will be used after
a new static HBA IP has been created
Cause: There was an assumption in the code where if the same
nic_iface structure was found based on the nic/vlan pair,
the specified IP address would not be used. Instead, it
will continue to use the previous defined IP address.
Change: The previous IP address will now be compared against the
the specified IP address before finishing the parce
iface request from iscsid. If different, the current
nic will be disabled and then re-enabled with the newly
specified IP address.
8. Problem: Cont00054571 - Unable to connect to routed ipv6 target
with RA address and iface DHCPv6
Cause: The default router address was not being employed for
the IPv6 neighbor negotiation. Additionally, the return
address of our neighbor advertisement was incorrect as
it should use the best matched src address instead.
Change: Fixed both the IPv6 neighbor solicitation and advertisement
transmission and handling.
9. Problem: Cont00054510 - fails to login to 32 session with blanket
login IPv6
Cause: A bug was introduced in uIP 0.6.4.6 where the NIC_RUNNING
flag might not be set when entering the main loop under
certain situations depending on the nic bring up.
Change: A new NIC_STARTED_RUNNING flag is now defined to fix CQ53511.
10. Problem: Cont00053807 - RA and link local are unable to connect if DHCPv6
fails
Cause: The host link local address was not being searched as one of
the host address to be replied to CNIC for the connect request.
Change: The path reply now includes the search of host link local
address as well.
11. Problem: Cont00054236 - iSCSI service must be restarted before an IPv6
connection can be made to the Equalogic target
Cause: The problem was intermittent as it depends on which IPv6 address
the target was redirecting to. Since uIP was only extracting
the target's IPv6 address + MAC from the target's neighbor
advertisement packet itself and not from the ICMPv6 option, so
the wrong or no MAC address will get send down to CNIC for the
connection establishment; hence the no connect.
Change: Added the updating of the neighbor discovery table to also use
the Target IPv6 address + MAC specified in the incoming neighbor
advertisement's ICMPv6 option field.
12. Problem: Cont00053255 - bnx2x panic dump logging into multiple
discovered IPv6 nodes (Equalogic IPv6 target)
Cause: The bnx2x panic was fixed in the 10g fw 6.4.29.
A IPv6 connectivity issue was then found and led to different
kernel/uIP crashes. This was caused by the same IPv6
connectivity problem mentioned above.
Change: Same as above
13. Problem: Cont00053728 - Sessions never recover after doing initiator-side
cable pull test with IPv6 traffic against Equalogic targets
Cause: It was discovered that the Equalogic would send out periodic
neighbor solicitation to maintain the connection to the
initiator. Since uIP was responding with the assigned IPv6
link local address in the neighbor advertisement
unconditionally, the target was observed to stop transmitting on
the connection specified.
Change: The neighbor advertisement generated will now use the dst IPv6
address from the input neighbor solicitation packet instead of
the assigned IPv6 link local address for both the packet and the
ICMPv6 source IPv6 address.
14. Problem: Compile error under 32-bit OS
Cause: A bug was introduced in the previous release 0.6.4.6 which
caused a compilation error in 32-bit OS (64-bit compiles
fine)
Change: Fixed the bug
15. Problem: Cont00053807 - RA and Link local are unable to connect if dhcpv6
fails
Cause: There was a bug in the nl reply where the RA address will never
be sent back to CNIC for the connection request
Change: The best matched address to the dst will now be sent back to
CNIC in the path rsp.
Enhancements
------------
1. Change: Updated README to remove the 57713/E references
2. Change: Allow the ICMP option field in the IPv6 Neighbor Advertisement
response to be included without discrimination. This fixes
an issue connecting against the EQL via RA for DHCPv6.
3. Change: Updated README for the IPv6 operation, VLAN, and discovery.
uIP v0.7.0.1 (Mar. 29, 2011)
=======================================================
Fixes
-----
1. Problem: Cont00053511 - bnx2x panic dump during ifup/down stress with
iSCSI traffic
Cause: The panic dump was resolved by the driver's rq dbell size fix.
After that, uIP crashed due to the asynchronous if_down event
that took the chip resources away while the nic thread is still
continuing to try to send DHCP request.
Change: Added synchronization between the two threads so proper clean up
of the threads can occur.
Enhancements
------------
1. Change: Added support for E3 (57800, 57810, and 57840)
uIP v0.6.4.5 (Mar. 23, 2011)
=======================================================
Enhancements
------------
1. Change: Optimized the double VLAN fix of CQ53870 to match
what will be submitted for RHELS5.7 and RHELS6.1 inbox
uIP v0.6.4.4 (Mar. 17, 2011)
=======================================================
Fixes
-----
1. Problem: Cont00053870 - Unable to login to iSCSI target via offload
through a Nexus 5020 switch with DCBx enabled
Cause: Double VLAN tagging was observed due to DCBx enabled.
The chip actually adds a VLAN tag if the txbd does not have
VLAN tag enabled under the DCBx environment for PRI setting.
Since uIP does not make use of hw assisted VLAN tagging,
2 VLAN tag was observed in the data stream.
Change: Enabled hw assisted VLAN tagging in uIP for both 1g and 10g.
2. Problem: Cont00053792 - maxconnections intermittently fail and
recover using iface DHCPv4
Cause: The DHCPv4 engine erroneously keeps on requesting for a
new lease which tremendously hamper normal path_req
operation. The problem is that the lease time parameter
has overflowed when converted to ticks count.
Change: Expanded the lease timer ticks count parameter from 16 to
32 bits.
3. Problem: Cont00053807 - RA and link local are unable to connect if
DHCPv6 fails
Cause: The DHCPv6 engine does not have the failover to use RA
mechanism
Change: Expanded to use best match address instead regardless of
DHCPv6 success or not, or using static v6.
Enhancements
------------
1. Change: Cont00051823 - Added man page for brcm_iscsiuio
uIP v0.6.4.3 (Mar. 15, 2011)
=======================================================
Fixes
-----
1. Problem: Cont00053719 - intermittent logging into targets that
are not in the same subnet as defined in the iface
Cause: The default route was used erroneously due to a miscompare
Change: Fixed this comparison so if the requested dst is not in
in the same subnet, uIP would not even ARP out.
2. Problem: Cont00053580 - Unable to do iSCSI boot into Linux OS using
57710 adapters
Cause: The E1 iro USTORM_RX_PROD_OFFSET doesn't match the t6.4 fw
Change: This is now fixed
uIP v0.6.4.2 (Feb. 24, 2011)
=======================================================
Fixes
-----
1. Problem: Cont00050343 - HBA does not follow RFC2131 spec for IPv4
DHCP lease expiration
Cause: The dhcp engine did not have this feature implemented
Change: Added lease time tracking and renewal
2. Problem: Cont00050801 - Unable to connect to target after switching
between DHCPv4 to static v4
Cause: The configuration flags got corrupted when switching between
dhcp and static or vice versa.
Change: Fixed the flag handling. Also needed to zero out the static
ip address in the host memory when switching to dhcp.
Otherwise, the static ip address will get used mistakenly.
Enhancements
------------
1. Change: Cont00051936 - Added IPv6 NDP and DHCPv6 support.
uIP v0.6.4.1 (Jan. 27, 2011)
=======================================================
Fixes
-----
1. Problem: Cont00049766 - segfault seen while stopping iscsi service
Cause: The logger output routine was accessing the log resource
while another thread calls fini_logger to free the same
resources
Change: Added pthread mutex lock to the logger routine to exclude
the initializer, user, and finisher
Enhancements
------------
1. Change: Added new t6.4 HSI and 57713 support.
uIP v0.6.2.13 (Jan. 04, 2011)
=======================================================
Fixes
-----
1. Problem: Cont00049665 - iscsiboot:linux failed to boot into iscsi
boot image in offload path after 5 iterations
Cause: The hw consumer index for the uIP ring got out of sync
with the producer index. This has led to the xmit mutex
lock be held forever so subsequent ARP requests will not
get transmitted to the wire
Change: Added this out of sync detection and rescue the xmit mutex
lock
uIP v0.6.2.12 (Dec. 21, 2010)
=======================================================
Fixes
-----
1. Problem: Cont00051820 - Session fails to reconnect after gateway
fallback
Cause: Under the HSRP test scenario, it was found that an ARP
request from the SUT is required in order for the HSRP
router to begin sending packets downstream to the SUT.
The default ARP age was originally set to 20 minutes
before a new ARP request will get sent,
Change: Changed the ARP age default to Linux default at 5 minutes
uIP v0.6.2.11 (Dec. 17, 2010)
=======================================================
Fixes
-----
1. Problem: For IPv4, the gateway route was not being utilized
when the subnet mask given or calculated does not
match. This resulted in many unwanted connection
attempts.
Cause: A bug was found in the default gateway calculation
logic which prevented the gateway address from being
used.
Change: Fixed the default gateway logic
2. Problem: For IPv6, there are scenarios where it won't connect
Cause: The IPv6 subnet mask as extracted from the CIDR
format might contain garbage data. This garbage data
was then used as part of the subnet mask which would
prevent the correct address mask.
Change: Fixed the subnet mask
uIP v0.6.2.10 (Dec. 15, 2010)
=======================================================
Fixes
-----
1. Problem: IPv6 does not connect for non-CIDR iface.ipaddress
specification
Cause: A bug where all ones was used as the IPv6 netmask
instead of all zeroes. This prevented all IPv6
path requests from being honored
Change: Fixed the subnet mask used
uIP v0.6.2.9 (Dec. 14, 2010)
=======================================================
Enhancements
------------
1. Change: Added IP address CIDR notation support for the
iface.ipaddress field in the iface file.
This will allow subnet mask to be defined and used.
uIP v0.6.2.8 (Dec. 9, 2010)
=======================================================
Fixes
-----
1. Problem: ipv6 + ifup/down fails to reconnect
Cause: There were 2 problems found:
- the xmit_mutex lock was being held indefinitely
- the nl_process_if_down flag for 10g doorbell ringing
did not get reinitialized
Change: Fixed the xmit_mutex deadlock via trylock
Added nl_process_if_down initialization in the IF_DOWN
process
2. Problem: Added fix for the NPAR disabled for 57712
Cause: The mac address was not handled correctly
Change: Fixed the mac address handling. Also requires corresponding
kernel component for the complete fix
uIP v0.6.2.7 (Dec. 7, 2010)
=======================================================
Enhancements
------------
1. Change: Use the gateway address from the DHCP server the
destination IP address is not in the current subnet.
uIP v0.6.2.6 (Nov. 16, 2010)
=======================================================
Fixes
-----
1. Problem: Warning message seen in the kernel logs,
"uio uio2: uevent: unsupported action string"
Cause: The improper string was echo'ed into the UIO trigger
field. With an improper string, this message would
appear in the kernel logs.
Change: uIP will now write the string "online" to the UIO
trigger field. This is the string expected by the
Linux kernel base driver.
2. Problem: uIP would segfault during a heavily login/logout
iSCSI subsystem reset senario
Cause: A double free occurred in the logging portion of the
uIP code, but this was root cause to a double free when
manipulating the NetLink buffers.
Change: Properly look at the return code from the routine which
will read NetLink messages. Also only free buffers
if they are allocated.
Enhancements
------------
1. Change: Add ability to print kernel version and machine
architecture to further help debug problems.
2. Change: Apply the netmask from DHCP if provided.
uIP v0.6.2.5 (Nov. 10, 2010)
=======================================================
Fixes
-----
1. Problem: iscsid would try to conenct with unintended iSCSI
targets
Cause: uIP would blindly return the iSCSI target MAC address
regardless if the iSCSI target is reachable via the
given port.
Change: uIP will try to filter the requests coming from CNIC
by automatically generating a network mask based off
the configured IP addressed. Then this netmask is
masked with the destination IP address. If there is
a match, then the path_req is allowed through.
2. Problem: Problems reconnecting back to the target when running
MTU stress tests.
Cause: cnic/bnx2i and uIP could possibly get out of sync when
an if_down message is sent.
Change: uIP will now immediately react to the if_down message,
and flush all the path req's and then to process to
if_close.
Enhancements
------------
1. Change: Fix compile warnings for src/unix/nic_nl.c,
and src/unix/main.c
uIP v0.6.2.4 (Nov. 4, 2010)
=======================================================
Fixes
-----
1. Problem: iSCSI HBA: brcm_iscsiuio segfault during ifdown
with many active sessions
Cause: uIP will segfault when traversing the error path when
an iSCSI connection is starting but the sysfs entries
have not been created yet.
Change: Use the errno value rather then the one from the file
descriptor because the file descriptor will be NULL and
the NULL dereference will cause a segfault.
Enhancements
------------
1. Change: Added initial changes for iSCSI multi-function support for
10G NIC's.
2. Change: Add more detailed messages for error pathes in nic_utils
uIP v0.6.2.3 (October 28, 2010)
=======================================================
Enhancements
------------
1. Change: Add support for bnx2x-1.62.x drivers
uIP v0.6.2.2 (October 18, 2010)
=======================================================
Enhancements
------------
1. Change: Only allow iSCSI connections with known bnx2x HSI's.
uIP v0.6.2.1 (October 7, 2010)
=======================================================
Fixes
-----
1. Problem: After multiple MTU changes, the ethtool IOCTL used to
determine the bnx2x driver version fails and eventually
iSCSI connections would not reconnect.
Cause: The socket file descriptor used during the ethtool IOCTL
call was never closed and leaked.
Change: On the error path when calling the ethtool IOCTL, the
file descriptor is now properly closed.
uIP v0.5.39 (September 15, 2010)
=======================================================
Fixes
-----
1. Problem: Could not offload IPv4 VLAN connection when the target tries
to ARP the iSCSI initiator
Cause: In the ARP reply, the ether field was incorrect.
Change: Properly set the ether field to 802.1Q type (0x8100)
uIP v0.5.38 (September 14, 2010)
=======================================================
Fixes
-----
1. Problem: uIP would cause a panic dump when the NIC was going down
Cause: uIP and CNIC where not synchonized on NIC state
Change: Check if the RX BD's which are zero'ed by CNIC when the
NIC is going down. If the BD addresses are zero, then
uIP will drop the TX packets.
uIP v0.5.37 (August 21, 2010)
=======================================================
Fixes
-----
1. Problem: uIP would segfault on ifup/ifdown stress test when using
DHCP to determine local IP address.
Cause: The uIP would use a NULL buffer during data transmission.
Change: Drop packets when there are no buffer avaliable.
uIP v0.5.36 (August 21, 2010)
=======================================================
Fixes
-----
1. Problem: iSCSI boot would not completely login after the pivot
root operation.
Cause: The uIP would not properly start the NIC interface.
Change: uIP should only check the NIC state to determine whether
to start the NIC thread or not.
2. Problem: uIP would segfault during if'up if'down testing.
Cause: The uIP would improperly start 2 NIC threads for the
same NIC interface.
Change: uIP should properly lock the NIC list when disabling/removing
the NIC threads.
uIP v0.5.35 (August 20, 2010)
=======================================================
Fixes
-----
1. Problem: Sessions would hang with ethtool self-test
Cause: The uIP would hang because the socket layer was stuck
because there is much contention for that socket. This
would hang the CNIC thread.
Change: Remove any IOCTL calls in uIP which may colide with
the ethtool self test. The driver version is only
capture during uIP initialization.
2. Problem: There were session recovery issue when using DHCP
if up/down tests.
Cause: The uIP would hang because the DHCP requests would
timeout if the network interface is downed which would
hang all the other uIP threads.
Change: Ensure that the DHCP state machine had exit points
if the network interface was down'ed.
uIP v0.5.34 (August 18, 2010)
=======================================================
Fixes
-----
1. Problem: Sessions would not recover with ethtool self-test
Cause: The uIP would hang because either the NetLink buffer is
full or that any socket operations used to manipulate
multicast addresses would block.
Change: Ensure that the socket used for multicast addressing is
set to nonblocking. Drain the NetLink buffer without
using the eventing, but with a more aggressive poll routine.
2. Problem: Sessions would not recover with L2 driver load/unload on
RHEL 6.0 SS9
Cause: The uIP would close the NIC thread too early and would
deadlock on cloing the NIC thread.
Change: Ensure that the NIC thread is canceled/closed only in one
location, in the NIC remove routine.
uIP v0.5.33 (August 17, 2010)
=======================================================
Fixes
-----
1. Problem: Error message seen from the uIP stack for valid packets.
Cause: The uIP was incorrectly marking logging messages for valid
packets as errors because it didn't know how to parase them.
Change: Changed the following from error to debug message
ipv6: invalid version
ipv4: invalid version or header length.
icmpv6: unknown ICMP message.
ip: neither tcp nor icmp
Changed the following from error to warn message
udp: bad checksum
tcp: bad checksum
tcp: got reset, aborting connection.
2. Problem: After multiple iterations the loading and unloading of
the Broadcom Linux drivers with active connections
would not cause the sessions to recover on RHEL 6.0
snapshot 9.
Cause: There was a deadlock in the nic mutex
Change: Lock ordering for the nic mutex and nic list mutex must
be inforced.
3. Problem: After multiple iterations of running the ethtool selftest
the Broadcom Linux drivers with active connections
would not cause the sessions to recover on RHEL 5.5.
Cause: The Netlink buffer between uIP and CNIC would get full.
Change: Poll more regularly for packets in the Netlink buffer
from 4 times a second to 100 times a 1 second.
Drain packets during the PATH_REQ packet pull.
uIP v0.5.32 (August 14, 2010)
=======================================================
Fixes
-----
1. Problem: Error message 'nic eth0: Didn't find type 0xaa bb' seen.
Cause: Valid non-DIX Ethernet packets as being passed to the
uIP. uIP will drop these packets but should be logged
correctly.
Change: These packets are valid, and should only be logged for
debugging purposes.
2. Problem: Error message 'Dropped previous transmitted packet' seen.
Cause: The TX ring is full, and here uIP is trying to transmit a
packet which will be dropped. This is a valid state but
the log message is marked incorrectly
Change: These messages are not warnings and should be logging when
debugging is enabled.
3. Problem: Error message: "iscsi_ipc eth0 Transport name is not
equal expected: got: bnx2i" seen.
Cause: The iface_rec structure is different between iscsid version.
For RHEL 5.5, iscsid is versioned 871, for RHEL 6.0 is
versioned 872.
Change: Allow uIP to compile against a different version of iscsid.
uIP v0.5.31 (August 12, 2010)
=======================================================
Fixes
-----
1. Problem: Softlock would occur showing that the NetLink table
lock was taken but never released.
Cause: NetLink socket buffer would fill with constant PATH_REQ
messages preventing PATH_REQ response from libiscsi
Change: Now uIP will drain the NetLink buffer while looking for
a response.
Enhancements
------------
1. Change: Add documentation for VLAN configuration and restrictions.
uIP v0.5.30 (August 6, 2010)
=======================================================
Fixes
-----
1. Problem: iscsid thread will stall if closing the uio files nodes
is stuck
Cause: uIP would indefinitely block waiting for the mutex shared
by the close routine.
Change: Now uIP will try and poll a bit for the mutex. If it can't
get this mutex in the iscsid thread then an error is return
rather then hold the thread.
2. Problem: IPv6 Unicast Neighbor Adveriserments would have the
ICMPv6 option header specifying a MAC.
Cause: uIP should use the source IPv6 address to detmine whether
to strip the option header or not and not the target address
in the ICMPv6 field.
Change: The uIP stack return a unicast IPv6 Neighbor Advertisement
without the ICMPv6 option as a response to unicast
IPv6 Neighbor Solicitations.
3. Problem: There would be TCP SYN packets with improper MAC address.
Cause: A zero'ed MAC address was not passed to CNIC to indicate an
error or if the IP address didn't resolve.
Change: The uIP stack will now return a zero'ed MAC address if it
can't find any entries.
uIP v0.5.29 (August 6, 2010)
=======================================================
Fixes
-----
1. Problem: "uip udp: no matching connection found: lport: 35072"
seen numerous times in the brcm_iscsiuio log file
Cause: This message was incorrectly marked as an error
Change: These messages are valid log entries especially if the
packet was a broadcast UDP packet not destined for the SUT
I will change the code to mark these logs entries as debug.
uIP v0.5.28 (August 5, 2010)
=======================================================
Fixes
-----
1. Problem: Can't login into a redirected Equilogic Target
Cause: The Equilogic Target uses a unicast IPv6 Neighbor
Solicitation to test if the host is up. The uIP stack
would return a Neighbor Advertisement with an unneeded
ICMPv6 option.
Change: Only have the uIP stack return a unicast IPv6 Neighbor
Advertisement without the ICMPv6 option.
2. Problem: With older bnx2/bnx2x/cnic/bnx2i driver combinations
uIP would segfault when these drivers were unloaded.
Cause: When the older drivers were removed, the underlying uio
instance was removed causing uIP to have a stale file handle.
When uIP finally closes using this stale file handle, either
uIP would segfault, or there would be an error in the
uio_release() path.
Change: Only have the uIP close if the UIO file node exists.
uIP v0.5.27 (July 31, 2010)
=======================================================
Fixes
-----
1. Problem: iSCSI HBA: Unable to use DHCP address for iSCSI interface
if a connection was previously made with a static address
on bnx2 devices.
Cause: Because the device is closed and reopen'ed the TX consumer
indexes were not persisted
Change: Only discard the TX consumer indexes only when the devices
will be discarded or closed
Enhancements
------------
1. Change: Change CNIC references to bnx2 in the bnx2 user space
driver.
uIP v0.5.26 (July 30, 2010)
=======================================================
Fixes
-----
1. Problem: iSCSI HBA: Unable to use DHCP address for iSCSI interface
if a connection was previously made with a static address on
bnx2x devices.
Cause: Because the device is closed and reopen'ed the TX consumer
indexes were not persisted
Change: Only discard the TX consumer indexes only when the devices
will be discarded
2. Problem: IPv6 using VLAN's didn't login
Cause: The uIP code used to determine if the packet was an IPv6
or not was not working. This VLAN packets for IPv6 were
being mis-interpreted.
Change: Make the function is_ipv6() VLAN aware
3. Problem: Persistant targets was not loggin in during boot
Cause: If udev was slow and the /dev/uio* were creatly slowly
uIP would fail.
Change: Poll uIP waiting for /dev/uio* file nodes.
uIP v0.5.25 (July 27, 2010)
=======================================================
Fixes
-----
1. Problem: When using IPv4 DHCP, there are no initial DHCP Discover
packets were not seen on the wire.
Cause: Packets generated from the app handler from the uIP stack
were not placed on the wire.
Change: Packets originating from the uIP stack are now always placed
on the wire.
uIP v0.5.24 (July 25, 2010)
=======================================================
Fixes
-----
1. Problem: One would see invalid packet packets flow through the
uIP stack, where the logs would indicate there is a packet
with an invalid length
Cause: The BD and CQE consumer indexes were not properly incremented
and masked.
Change: The BD index is now properly masked. The CQE index is not
incremented using the CQE index rather the mistaken BD index.
Impact: 10G only
2. Problem: uIP would segfault during the booting of the machine.
Cause: uIP was using a NULL data pointer because there was an
incorrect packet passed to the stack.
Change: Only allow uIP to process data if the packet exists.
3. Problem: uIP would stop processing packets
Cause: The uIP code would not properly drain the CQE ring causing
it to eventually be full
Change: Consume all the CQE elements even if they are ethernet types
or not.
Impact: 10G only
4. Problem: uIP would stop after if/down of the network interface.
Cause: uIP was not kick starting the NIC loop thread properly.
Change: Ensure that the NIC loop thread is started by when iscsid
request that the interface start the offload. Mark the NIC
only if the thread is truly canceled.
uIP v0.5.23 (July 20, 2010)
=======================================================
Fixes
-----
1. Problem: Segfault during brcm_iscsiuio initialization
Cause: uIP was using a NULL data pointer, because a different
thread re-initialized the uIP stack
Change: Properly synchronize the initialization of the stack
2. Problem: Deadlock during the printing of heavy debug messages
Cause: The variable macro structures would point to invalid
data
Change: With each invocation of va_copy() a corresponding
invocation of va_end() in the same function for the proper
cleanup
3. Problem: uIP would hang when the interface could go up/down
Cause: uIP would get out of sync with the state of the network
interface
Change: Instead of detriving state from the UIO file nodes, uIP
will take direction from iscsid on when interfaces will be
started.
uIP v0.5.22 (July 15, 2010)
=======================================================
Fixes
-----
1. Problem: Unable to reconnect via iSCSI offload after
ifup/ifdown
Cause: uIP was stuck on the thread when closing the NIC main
loop
Change: Properly synchronize the NetLink CNIC and uevent threads
2. Problem: uIP would crash during boot up.
Cause: uIP would overwrite a memory location which was already
freed during nic_remove().
Change: Since the NIC is freed there is no need to write to
update the NIC flags
Enhancements
------------
1. Change: Added IPv6 Link Local support
uIP v0.5.21 (July 5, 2010)
=======================================================
Fixes
-----
1. Problem: Unable to connect via iSCSI offload after
changing L2 address
Cause: uIP didn't notice the network inferface going down
Change: Allow uIP to persist the stack's IP address after
a reset
2. Problem: Unable to connect via IPv4 and IPv6 concurrently
Cause: uIP didn't notice the network inferface going down
Change: Allow uIP to persist the stack's IP address after
a reset and properly bring up the interface
3. Problem: Unable to connect via VLAN
Cause: IP address was no persisted after a device reset
Change: When CNIC requests a path request, uIP will use the
VLAN passed by the CNIC.
uIP v0.5.20 (June 24, 2010)
uIP v0.5.20 (June 24, 2010)
=======================================================
Fixes
-----
1. Problem: Certain IPv6 addresses are not repsonded to by
the target.
Cause: The MAC was generated from the target's IPv6
address not the deterived multicast IPv6 address.
Change: The destination MAC address should be deterived
from the packet's destination IPv6 address and
not the target.
2. Problem: brcm_iscsiuio would segfault when L2 interface is
bought up and down after being logged into
Cause: The NIC thread was not stopped properly
Change: When the UIO device is remove and when the
cooresponding NIC tracked by brcm_iscsiuio, the
daemon would properly wait for the NIC thread to
stop.
uIP v0.5.19 (June 22, 2010)
=======================================================
Fixes
-----
1. Problem: Can't login after boot
Cause: If NIC interfaces are brough up and down quickly
uIP wait on an invalid NIC thread
Change: Only wait for the NIC thread if the NIC thread
exists.
uIP v0.5.18 (June 21, 2010)
=======================================================
Fixes
-----
1. Problem: Does not compile on SLES 11 SP1
Cause: Automake cached files were included as part of the
uIP-0.5.17 package
Change: Remove automake cached files, and allow these files
to be generated each time the source is compiled
2. Problem: Does not always receive multicast packets
Cause: Multicast bit was not set in SORT USER 2 register
Change: brcm_iscsiuio will now set the SORT USER 2 registers
with both the broadcast and multicast bits.
3. Problem: Existing iSCSI connections do not reconnect after
operations which require equivalent driver
load/unload operations
Cause: Multiple path requests would trample NIC configurations
Change: Allow only one path request at a time
uIP v0.5.17 (June 16, 2010)
=======================================================
Fixes
-----
1. Problem: IPv6 neighbor solicitations from brcm_iscsiuio could
not be responded to
Cause: The IPv6 neighbor solicitation packet had an invalid
multicast MAC address
Change: Properly set the MAC address multicast bit and OR
with the IPv6 destination address
2. Problem: NIC state was not properly synchronized and noticed
by Shyam Iyer <shiyer@redhat.com>
Change: Properly lock the NIC device when changing state
Enhancements
------------
1. Change: Listen for iscsid before daemonizing to close a timing
gap which might allow iscsid to start before uIP is
completely initialized.
uIP v0.5.16 (June 2, 2010)
=======================================================
Enhancements
------------
1. Change: Formally add IPv6 support. Only a static IPv6 address
is supported.
uIP v0.5.15 (May 20, 2010)
=======================================================
Fixes
-----
1. Problem: brcm_iscsiuio would echo packets off the wire
Cause: Stale packets from the uIP stack could potentially
make it onto the wire causing a network flood
Change: Only place on the wire packets uIP intended to place
on the wire. Drop all other packets.
uIP v0.5.14 (May 18, 2010)
=======================================================
Fixes
-----
1. Problem: brcm_iscsiuio would crash when offloading using a
bnx2x device /dev/mem could not be
opened, (ie. SE Linux enabled)
Cause: /dev/mem could not be opened, (ie. SE Linux enabled)
and then the NIC would be improperly initialized.
Change: If /dev/mem is not able to be opened, then the device
is closed
2. Problem: brcm_iscsiuio would crash when brcm_iscsiuio is
being shutdown
Cause: The NIC mutex was deferenced imporperly when the NIC
is being closed
Change: Take the NIC mutex lock only when the NIC is closed.
uIP v0.5.13 (May 16, 2010)
=======================================================
Fixes
-----
1. Problem: brcm_iscsiuio would crash with heavy traffic directed
at the iSCSI traffic
Cause: Packets which are sized between 1006-1024 bytes would
crash brcm_iscsiuio because brcm_iscsiuio is not sized
to handle such large packets
Change: Drop large packets, properly hold the NIC mutex lock
for the duration when NIC fields are being used.
uIP v0.5.12 (May 13, 2010)
=======================================================
Fixes
-----
1. Problem: brcm_iscsiuio could crash on when L2 interface is
ifdown'ed
Cause: The local NIC pointer was not initialized properly
in the routine parse_iface()
Change: Properly initialize the NIC pointer
2. Problem: Documentation referred to older admin_client which
doesn't exist any more because brcm_iscsiuio uses
the iscsid iface file
Change: Remove the stale references
uIP v0.5.11 (May 11, 2010)
=======================================================
Fixes
-----
1. Problem: brcm_iscsiuio could crash on invalid packet sizes
Cause: The hardware BD could be a large value because of a
hardware error
Change: Limit the size of the packet dumped to the MTU size
Enhancements
------------
1. Change: During the running of the configure script now
the script will check for ar and ranlib binaries
uIP v0.5.10 (May 03, 2010)
=======================================================
Fixes
-----
1. Problem: BCM57712 not recognized
Cause: The PCI ID's in the bnx2x file were missing.
Change: Added proper BCM57712, BCM57712E, BCM57713, BCM57713E
PCI ID's
2. Problem: (CQ 47481) brcm_iscsiuio not installed in correct location
Cause: Default install path for autoconf is /usr/local
Change: Change the default prefix to '/' so the brcm_iscsiuio
binary is installed to /sbin/
Enhancements
------------
1. Change: Remove dependency on Yacc and Lex
uIP v0.5.9 (April 28, 2010)
=======================================================
Fixes
-----
1. Problem: bnx2x T6.0 driver would not login
Cause: The bnx2x code was not using the T6.0 HSI offsets
Change: Determine to bnx2x driver version eariler to properly use the
T4.8 or T6.0 HSI
Enhancements
------------
1. Change: Collapse all the various locks to use the NIC lock to shrink
memory footprint
2. Change: Consolidate upper layer checksumming code
uIP v0.5.5 (March 02, 2010)
=======================================================
Enhancements
------------
1. Change: Add support for T6.0 bnx2x HSI and 57712.
2. Change: Initial support for IPv6
uIP v0.5.8 (April 22, 2010)
=======================================================
Enhancements
------------
1. Change: Add support for T6.0 bnx2x HSI and 57712.
2. Change: Initial support for IPv6
uIP v0.5.7 (March 17, 2010)
=======================================================
Enhancements
------------
1. Change: Add to documentation on discovering on a particular
iface before logging in
uIP v0.5.6 (Mar 05, 2009)
=======================================================
Fixes
-----
1. Problem: bnx2x panic dump would be seen when sending
traffic to uIP
Cause: The TX producer index was not properly
incrementing when the wrapping occured
Change: Do not skip the last TX producer index like the
TX BD's
Impact: None.
uIP v0.5.5 (March 02, 2010)
=======================================================
Initial release
Enhancements
------------
1. Change: Add to documentation on debugging/logging for uIP
uIP v0.5.4 (Feb 22, 2010)
=======================================================
Fixes
-----
1. Problem: Compile error where 'ETHERTYPE_VLAN' define
is missing
Cause: Certain distributions do not define 'ETHERTYPE_VLAN'
in the header file "net/ethernet.h".
Change: Added proper defines for ETHERTYPE_VLAN when necessary
Impact: None.
uIP v0.5.3 (Feb 18, 2010)
=======================================================
Fixes
-----
1. Problem: Using VLAN's on offloaded iSCSI connections
Cause: (CQ45983) VLAN tags were not being properly inserted
when sending the ARP request packets
Change: Added VLAN tags when sending ARP request packets
Impact: None.
uIP v0.5.2 (Dec 10, 2009)
=======================================================
Fixes
-----
1. Problem: Switching between 10G and 1G iSCSI offloaded
devices caused login connectivity problems
Cause: The NIC devices within uIP were not cleanup
properly.
Change: The NIC structure is not re-initialized and the
NIC thread is destroyed when the host network
interface is brought down.
Impact: None.
uIP v0.5.1 (Dec 9, 2009)
=======================================================
Fixes
-----
1. Problem: 10G devices behind PCI bridges would not collect
Cause: PCI bus:slot.func string was parsed incorrectly
because the bridge string was used
Change: Parse the proper PCI bus:slot.func string.
Impact: None.
uIP v0.5.0b (Nov 24, 2009)
=======================================================
Initial release
Enhancements
------------
1. Change: Add Broadcom 10G iSCSI offload support
Impact: Linux