Pablo Neira Ayuso says:
====================
Netfilter fixes for net
The following patchset contains Netfilter fixes for net:
1) Use signed integer in ipv6_skip_exthdr() called from nf_confirm().
Reported by static analysis tooling, patch from Florian Westphal.
2) Missing set type checks in nf_tables: Validate that set declaration
matches the an existing set type, otherwise bail out with EEXIST.
Currently, nf_tables silently accepts the re-declaration with a
different type but it bails out later with EINVAL when the user adds
entries to the set. This fix is relatively large because it requires
two preparation patches that are included in this batch.
3) Do not ignore updates of timeout and gc_interval parameters in
existing sets.
4) Fix a hang when 0/0 subnets is added to a hash:net,port,net type of
ipset. Except hash:net,port,net and hash:net,iface, the set types don't
support 0/0 and the auxiliary functions rely on this fact. So 0/0 needs
a special handling in hash:net,port,net which was missing (hash:net,iface
was not affected by this bug), from Jozsef Kadlecsik.
5) When adding/deleting large number of elements in one step in ipset,
it can take a reasonable amount of time and can result in soft lockup
errors. This patch is a complete rework of the previous version in order
to use a smaller internal batch limit and at the same time removing
the external hard limit to add arbitrary number of elements in one step.
Also from Jozsef Kadlecsik.
Except for patch #1, which fixes a bug introduced in the previous net-next
development cycle, anything else has been broken for several releases.
====================
Signed-off-by: David S. Miller <davem@davemloft.net>
When adding/deleting large number of elements in one step in ipset, it can
take a reasonable amount of time and can result in soft lockup errors. The
patch 5f7b51bf09 ("netfilter: ipset: Limit the maximal range of
consecutive elements to add/delete") tried to fix it by limiting the max
elements to process at all. However it was not enough, it is still possible
that we get hung tasks. Lowering the limit is not reasonable, so the
approach in this patch is as follows: rely on the method used at resizing
sets and save the state when we reach a smaller internal batch limit,
unlock/lock and proceed from the saved state. Thus we can avoid long
continuous tasks and at the same time removed the limit to add/delete large
number of elements in one step.
The nfnl mutex is held during the whole operation which prevents one to
issue other ipset commands in parallel.
Fixes: 5f7b51bf09 ("netfilter: ipset: Limit the maximal range of consecutive elements to add/delete")
Reported-by: syzbot+9204e7399656300bf271@syzkaller.appspotmail.com
Signed-off-by: Jozsef Kadlecsik <kadlec@netfilter.org>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
The hash:net,port,net set type supports /0 subnets. However, the patch
commit 5f7b51bf09 titled "netfilter: ipset: Limit the maximal range
of consecutive elements to add/delete" did not take into account it and
resulted in an endless loop. The bug is actually older but the patch
5f7b51bf09 brings it out earlier.
Handle /0 subnets properly in hash:net,port,net set types.
Fixes: 5f7b51bf09 ("netfilter: ipset: Limit the maximal range of consecutive elements to add/delete")
Reported-by: Марк Коренберг <socketpair@gmail.com>
Signed-off-by: Jozsef Kadlecsik <kadlec@netfilter.org>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
There is an issue with the checking of the return value of
'of_get_mac_address', which returns 0 on success and negative value on
failure. The driver interpretated the result the opposite way. Therefore
if there was a MAC address defined in the DT, then the driver was
generating a random MAC address otherwise it would use address 0.
Fix this by checking correctly the return value of 'of_get_mac_address'
Fixes: b74ef9f9cb ("net: sparx5: Do not use mac_addr uninitialized in mchp_sparx5_probe()")
Signed-off-by: Horatiu Vultur <horatiu.vultur@microchip.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Fix W=1 kernel-doc warning:
net/sched/sch_htb.c:214: warning: expecting prototype for htb_classify(). Prototype was for HTB_DIRECT() instead
by moving the HTB_DIRECT() macro above the function.
Add kernel-doc notation for function parameters as well.
Signed-off-by: Randy Dunlap <rdunlap@infradead.org>
Cc: Jamal Hadi Salim <jhs@mojatatu.com>
Cc: Cong Wang <xiyou.wangcong@gmail.com>
Cc: Jiri Pirko <jiri@resnulli.us>
Cc: "David S. Miller" <davem@davemloft.net>
Cc: Eric Dumazet <edumazet@google.com>
Cc: Jakub Kicinski <kuba@kernel.org>
Cc: Paolo Abeni <pabeni@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Jamal Hadi Salim says:
====================
net: dont intepret cls results when asked to drop
It is possible that an error in processing may occur in tcf_classify() which
will result in res.classid being some garbage value. Example of such a code path
is when the classifier goes into a loop due to bad policy. See patch 1/2
for a sample splat.
While the core code reacts correctly and asks the caller to drop the packet
(by returning TC_ACT_SHOT) some callers first intepret the res.class as
a pointer to memory and end up dropping the packet only after some activity with
the pointer. There is likelihood of this resulting in an exploit. So lets fix
all the known qdiscs that behave this way.
====================
Signed-off-by: David S. Miller <davem@davemloft.net>
If asked to drop a packet via TC_ACT_SHOT it is unsafe to assume
res.class contains a valid pointer
Fixes: b0188d4dbe ("[NET_SCHED]: sch_atm: Lindent")
Signed-off-by: Jamal Hadi Salim <jhs@mojatatu.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
As stated in marvell-orion-mdio.txt deleted in commit 0781434af8
("dt-bindings: net: orion-mdio: Convert to JSON schema") if
'interrupts' property is present, width of 'reg' should be 0x84.
Otherwise, width of 'reg' should be 0x4. Fix 'examples:' and add
constraints checking whether 'interrupts' property is present
and validate it against fixed values in reg.
Signed-off-by: Michał Grzelak <mig@semihalf.com>
Reviewed-by: Krzysztof Kozlowski <krzysztof.kozlowski@linaro.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
This property has always been supported by the Linux driver; see
commit 9f93ac8d40 ("net-next: stmmac: Add dwmac-sun8i"). In fact, the
original driver submission includes the phy-supply code but no mention
of it in the binding, so the omission appears to be accidental. In
addition, the property is documented in the binding for the previous
hardware generation, allwinner,sun7i-a20-gmac.
Document phy-supply in the binding to fix devicetree validation for the
25+ boards that already use this property.
Fixes: 0441bde003 ("dt-bindings: net-next: Add DT bindings documentation for Allwinner dwmac-sun8i")
Acked-by: Rob Herring <robh@kernel.org>
Reviewed-by: Andre Przywara <andre.przywara@arm.com>
Signed-off-by: Samuel Holland <samuel@sholland.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
It is now possible for a system to have more than 32 endpoints. As
a result, registers related to endpoint suspend are parameterized,
with 32 endpoints represented in one more registers.
In ipa_interrupt_suspend_control(), the IPA_SUSPEND_EN register
offset is determined properly, but the bit mask used still assumes
the number of enpoints won't exceed 32. This is a bug. Fix it.
Fixes: f298ba785e ("net: ipa: add a parameter to suspend registers")
Signed-off-by: Alex Elder <elder@linaro.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
Po-Hsu Lin says:
====================
selftests: net: fix for arp_ndisc_evict_nocarrier test
This patchset will fix a false-positive issue caused by the command in
cleanup_v6() of the arp_ndisc_evict_nocarrier test.
Also, it will make the test to return a non-zero value for any failure
reported in the test for us to avoid false-negative results.
====================
Signed-off-by: David S. Miller <davem@davemloft.net>
Return non-zero return value if there is any failure reported in this
script during the test. Otherwise it can only reflect the status of
the last command.
Fixes: f86ca07eb5 ("selftests: net: add arp_ndisc_evict_nocarrier")
Signed-off-by: Po-Hsu Lin <po-hsu.lin@canonical.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
The cleanup_v6() will cause the arp_ndisc_evict_nocarrier script exit
with 255 (No such file or directory), even the tests are good:
# selftests: net: arp_ndisc_evict_nocarrier.sh
# run arp_evict_nocarrier=1 test
# RTNETLINK answers: File exists
# ok
# run arp_evict_nocarrier=0 test
# RTNETLINK answers: File exists
# ok
# run all.arp_evict_nocarrier=0 test
# RTNETLINK answers: File exists
# ok
# run ndisc_evict_nocarrier=1 test
# ok
# run ndisc_evict_nocarrier=0 test
# ok
# run all.ndisc_evict_nocarrier=0 test
# ok
not ok 1 selftests: net: arp_ndisc_evict_nocarrier.sh # exit=255
This is because it's trying to modify the parameter for ipv4 instead.
Also, tests for ipv6 (run_ndisc_evict_nocarrier_enabled() and
run_ndisc_evict_nocarrier_disabled() are working on veth1, reflect
this fact in cleanup_v6().
Fixes: f86ca07eb5 ("selftests: net: add arp_ndisc_evict_nocarrier")
Signed-off-by: Po-Hsu Lin <po-hsu.lin@canonical.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Now that phylink no longer calls phy_get_rate_matching with
PHY_INTERFACE_MODE_NA, phys no longer need to support it. Remove the
documentation mandating support.
Fixes: 7642cc28fd ("net: phylink: fix PHY validation with rate adaption")
Signed-off-by: Sean Anderson <sean.anderson@seco.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Christian Marangi says:
====================
net: dsa: qca8k: multiple fix on mdio read/write
Due to some problems in reading the Documentation and elaborating it
some wrong assumption were done. The error was reported and notice only
now due to how things are setup in the code flow.
First 2 patch fix mgmt eth where the lenght calculation is very
confusing and in step of word size. (the related commit description have
an extensive description about how this mess works)
Last 3 patch revert the broken mdio cache and apply a correct version
that should still save some extra mdio in phy poll secnario.
These 5 patch fix each related problem and apply what the Documentation
actually say.
Changes v2:
- Add cover letter
- Fix typo in revert patch
====================
Signed-off-by: David S. Miller <davem@davemloft.net>
Improve mdio master read/write by using singe mii read/write lo/hi.
In a read and write we need to poll the mdio master regs in a busy loop
to check for a specific bit present in the upper half of the reg. We can
ignore the other half since it won't contain useful data. This will save
an additional useless read for each read and write operation.
In a read operation the returned data is present in the mdio master reg
lower half. We can ignore the other half since it won't contain useful
data. This will save an additional useless read for each read operation.
In a read operation it's needed to just set the hi half of the mdio
master reg as the lo half will be replaced by the result. This will save
an additional useless write for each read operation.
Tested-by: Ronald Wahl <ronald.wahl@raritan.com>
Signed-off-by: Christian Marangi <ansuelsmth@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
It may be useful to read/write just the lo or hi half of a reg.
This is especially useful for phy poll with the use of mdio master.
The mdio master reg is composed by the first 16 bit related to setup and
the other half with the returned data or data to write.
Refactor the mii function to permit single mii read/write of lo or hi
half of the reg.
Tested-by: Ronald Wahl <ronald.wahl@raritan.com>
Signed-off-by: Christian Marangi <ansuelsmth@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
This reverts commit 2481d206fa.
The Documentation is very confusing about the topic.
The cache logic for hi and lo is wrong and actually miss some regs to be
actually written.
What the Documentation actually intended was that it's possible to skip
writing hi OR lo if half of the reg is not needed to be written or read.
Revert the change in favor of a better and correct implementation.
Reported-by: Ronald Wahl <ronald.wahl@raritan.com>
Signed-off-by: Christian Marangi <ansuelsmth@gmail.com>
Cc: stable@vger.kernel.org # v5.18+
Signed-off-by: David S. Miller <davem@davemloft.net>
It was discovered that MGMT_DATA2 can contain up to 28 bytes of data
instead of the 12 bytes written in the Documentation by accounting the
limit of 16 bytes declared in Documentation subtracting the first 4 byte
in the packet header.
Update the define with the real world value.
Tested-by: Ronald Wahl <ronald.wahl@raritan.com>
Fixes: c2ee8181fd ("net: dsa: tag_qca: add define for handling mgmt Ethernet packet")
Signed-off-by: Christian Marangi <ansuelsmth@gmail.com>
Cc: stable@vger.kernel.org # v5.18+
Signed-off-by: David S. Miller <davem@davemloft.net>
The assumption that Documentation was right about how this value work was
wrong. It was discovered that the length value of the mgmt header is in
step of word size.
As an example to process 4 byte of data the correct length to set is 2.
To process 8 byte 4, 12 byte 6, 16 byte 8...
Odd values will always return the next size on the ack packet.
(length of 3 (6 byte) will always return 8 bytes of data)
This means that a value of 15 (0xf) actually means reading/writing 32 bytes
of data instead of 16 bytes. This behaviour is totally absent and not
documented in the switch Documentation.
In fact from Documentation the max value that mgmt eth can process is
16 byte of data while in reality it can process 32 bytes at once.
To handle this we always round up the length after deviding it for word
size. We check if the result is odd and we round another time to align
to what the switch will provide in the ack packet.
The workaround for the length limit of 15 is still needed as the length
reg max value is 0xf(15)
Reported-by: Ronald Wahl <ronald.wahl@raritan.com>
Tested-by: Ronald Wahl <ronald.wahl@raritan.com>
Fixes: 90386223f4 ("net: dsa: qca8k: add support for larger read/write size with mgmt Ethernet")
Signed-off-by: Christian Marangi <ansuelsmth@gmail.com>
Cc: stable@vger.kernel.org # v5.18+
Signed-off-by: David S. Miller <davem@davemloft.net>
of_phy_find_device() return device node with refcount incremented.
Call put_device() to relese it when not needed anymore.
Fixes: ab4e6ee578 ("net: phy: xgmiitorgmii: Check phy_driver ready before accessing")
Signed-off-by: Miaoqian Lin <linmq006@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
The device supports a PCIe optimization hint, which indicates on
which NUMA the queue is currently processed. This hint is utilized
by PCIe in order to reduce its access time by accessing the
correct NUMA resources and maintaining cache coherence.
The driver calls the register update for the hint (called TPH -
TLP Processing Hint) during the NAPI loop.
Though the update is expected upon a NUMA change (when a queue
is moved from one NUMA to the other), the current logic performs
a register update when the queue is moved to a different CPU,
but the CPU is not necessarily in a different NUMA.
The changes include:
1. Performing the TPH update only when the queue has switched
a NUMA node.
2. Moving the TPH update call to be triggered only when NAPI was
scheduled from interrupt context, as opposed to a busy-polling loop.
This is due to the fact that during busy-polling, the frequency
of CPU switches for a particular queue is significantly higher,
thus, the likelihood to switch NUMA is much higher. Therefore,
providing the frequent updates to the device upon a NUMA update
are unlikely to be beneficial.
Fixes: 1738cd3ed3 ("net: ena: Add a driver for Amazon Elastic Network Adapters (ENA)")
Signed-off-by: David Arinzon <darinzon@amazon.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
RX ring can be NULL in XDP use cases where only TX queues
are configured. In this scenario, the RX interrupt moderation
value sent to the device remains in its default value of 0.
In this change, setting the default value of the RX interrupt
moderation to be the same as of the TX.
Fixes: 548c4940b9 ("net: ena: Implement XDP_TX action")
Signed-off-by: David Arinzon <darinzon@amazon.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Make the upper bound on rx_copybreak tighter, by
making sure it is smaller than the minimum of mtu and
ENA_PAGE_SIZE. With the current upper bound of mtu,
rx_copybreak can be larger than a page. Such large
rx_copybreak will not bring any performance benefit to
the user and therefore makes no sense.
In addition, the value update was only reflected in
the adapter structure, but not applied for each ring,
causing it to not take effect.
Fixes: 1738cd3ed3 ("net: ena: Add a driver for Amazon Elastic Network Adapters (ENA)")
Signed-off-by: Osama Abboud <osamaabb@amazon.com>
Signed-off-by: Arthur Kiyanovski <akiyano@amazon.com>
Signed-off-by: David Arinzon <darinzon@amazon.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Redirecting packets with XDP Redirect is done in two phases:
1. A packet is passed by the driver to the kernel using
xdp_do_redirect().
2. After finishing polling for new packets the driver lets the kernel
know that it can now process the redirected packet using
xdp_do_flush_map().
The packets' redirection is handled in the napi context of the
queue that called xdp_do_redirect()
To avoid calling xdp_do_flush_map() each time the driver first checks
whether any packets were redirected, using
xdp_flags |= xdp_verdict;
and
if (xdp_flags & XDP_REDIRECT)
xdp_do_flush_map()
essentially treating XDP instructions as a bitmask, which isn't the case:
enum xdp_action {
XDP_ABORTED = 0,
XDP_DROP,
XDP_PASS,
XDP_TX,
XDP_REDIRECT,
};
Given the current possible values of xdp_action, the current design
doesn't have a bug (since XDP_REDIRECT = 100b), but it is still
flawed.
This patch makes the driver use a bitmask instead, to avoid future
issues.
Fixes: a318c70ad1 ("net: ena: introduce XDP redirect implementation")
Signed-off-by: Shay Agroskin <shayagr@amazon.com>
Signed-off-by: David Arinzon <darinzon@amazon.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
The size of packets that were forwarded or dropped by XDP wasn't added
to the total processed bytes statistic.
Fixes: 548c4940b9 ("net: ena: Implement XDP_TX action")
Signed-off-by: Shay Agroskin <shayagr@amazon.com>
Signed-off-by: David Arinzon <darinzon@amazon.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Since the queues aren't destroyed when we only exchange XDP programs,
there's no need to re-register them again.
Fixes: 548c4940b9 ("net: ena: Implement XDP_TX action")
Signed-off-by: Shay Agroskin <shayagr@amazon.com>
Signed-off-by: David Arinzon <darinzon@amazon.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
On driver initialization, RSS hash initial value is set to zero,
instead of the default value. This happens because we pass NULL as
the RSS key parameter, which caused us to never initialize
the RSS hash value.
This patch fixes it by making sure the initial value is set, no matter
what the value of the RSS key is.
Fixes: 91a65b7d3e ("net: ena: fix potential crash when rxfh key is NULL")
Signed-off-by: Nati Koler <nkoler@amazon.com>
Signed-off-by: David Arinzon <darinzon@amazon.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
This cmsg_so_mark.sh test will hang on non-amd64 systems because of the
infinity loop for argument parsing in cmsg_sender.
Variable "o" in cs_parse_args() for taking getopt() should be an int,
otherwise it will be 255 when getopt() returns -1 on non-amd64 system
and thus causing infinity loop.
Link: https://lore.kernel.org/lkml/CA+G9fYsM2k7mrF7W4V_TrZ-qDauWM394=8yEJ=-t1oUg8_40YA@mail.gmail.com/t/
Signed-off-by: Po-Hsu Lin <po-hsu.lin@canonical.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
The driver does not call tasklet_kill in several places.
Add the calls to fix it.
Fixes: 85b85c8534 ("amd-xgbe: Re-issue interrupt if interrupt status not cleared")
Signed-off-by: Jiguang Xiao <jiguang.xiao@windriver.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Currently, the PF check the VF alive by the KEEP_ALVE
mailbox from VF. VF keep sending the mailbox per 2
seconds. Once PF lost the mailbox for more than 8
seconds, it will regards the VF is abnormal, and stop
notifying the state change to VF, include link state,
vf mac, reset, even though it receives the KEEP_ALIVE
mailbox again. It's inreasonable.
This patch fixes it. PF will record the state change which
need to notify VF when lost the VF's KEEP_ALIVE mailbox.
And notify VF when receive the mailbox again. Introduce a
new flag HCLGE_VPORT_STATE_INITED, used to distinguish the
case whether VF driver loaded or not. For VF will query
these states when initializing, so it's unnecessary to
notify it in this case.
Fixes: aa5c4f175b ("net: hns3: add reset handling for VF when doing PF reset")
Signed-off-by: Jian Shen <shenjian15@huawei.com>
Signed-off-by: Hao Lan <lanhao@huawei.com>
Reported-by: kernel test robot <lkp@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
A remove callback just returning 0 is equivalent to no remove callback
at all. So drop the useless function.
Signed-off-by: Uwe Kleine-König <u.kleine-koenig@pengutronix.de>
Signed-off-by: David S. Miller <davem@davemloft.net>
A remove callback just returning 0 is equivalent to no remove callback
at all. So drop the useless function.
Signed-off-by: Uwe Kleine-König <u.kleine-koenig@pengutronix.de>
Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Kuniyuki Iwashima says:
===================
tcp: Fix bhash2 and TIME_WAIT regression.
We forgot to add twsk to bhash2. Therefore TIME_WAIT sockets cannot
prevent bind() to the same local address and port.
Changes:
v1:
* Patch 1:
* Add tw_bind2_node in inet_timewait_sock instead of
moving sk_bind2_node from struct sock to struct
sock_common.
====================
Signed-off-by: David S. Miller <davem@davemloft.net>
bhash2 split the bind() validation logic into wildcard and non-wildcard
cases. Let's add a test to catch future regression.
Before the previous patch:
# ./bind_timewait
TAP version 13
1..2
# Starting 2 tests from 3 test cases.
# RUN bind_timewait.localhost.1 ...
# bind_timewait.c:87:1:Expected ret (0) == -1 (-1)
# 1: Test terminated by assertion
# FAIL bind_timewait.localhost.1
not ok 1 bind_timewait.localhost.1
# RUN bind_timewait.addrany.1 ...
# OK bind_timewait.addrany.1
ok 2 bind_timewait.addrany.1
# FAILED: 1 / 2 tests passed.
# Totals: pass:1 fail:1 xfail:0 xpass:0 skip:0 error:0
After:
# ./bind_timewait
TAP version 13
1..2
# Starting 2 tests from 3 test cases.
# RUN bind_timewait.localhost.1 ...
# OK bind_timewait.localhost.1
ok 1 bind_timewait.localhost.1
# RUN bind_timewait.addrany.1 ...
# OK bind_timewait.addrany.1
ok 2 bind_timewait.addrany.1
# PASSED: 2 / 2 tests passed.
# Totals: pass:2 fail:0 xfail:0 xpass:0 skip:0 error:0
Signed-off-by: Kuniyuki Iwashima <kuniyu@amazon.com>
Acked-by: Joanne Koong <joannelkoong@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Jiri Slaby reported regression of bind() with a simple repro. [0]
The repro creates a TIME_WAIT socket and tries to bind() a new socket
with the same local address and port. Before commit 28044fc1d4 ("net:
Add a bhash2 table hashed by port and address"), the bind() failed with
-EADDRINUSE, but now it succeeds.
The cited commit should have put TIME_WAIT sockets into bhash2; otherwise,
inet_bhash2_conflict() misses TIME_WAIT sockets when validating bind()
requests if the address is not a wildcard one.
The straight option is to move sk_bind2_node from struct sock to struct
sock_common to add twsk to bhash2 as implemented as RFC. [1] However, the
binary layout change in the struct sock could affect performances moving
hot fields on different cachelines.
To avoid that, we add another TIME_WAIT list in inet_bind2_bucket and check
it while validating bind().
[0]: https://lore.kernel.org/netdev/6b971a4e-c7d8-411e-1f92-fda29b5b2fb9@kernel.org/
[1]: https://lore.kernel.org/netdev/20221221151258.25748-2-kuniyu@amazon.com/
Fixes: 28044fc1d4 ("net: Add a bhash2 table hashed by port and address")
Reported-by: Jiri Slaby <jirislaby@kernel.org>
Suggested-by: Paolo Abeni <pabeni@redhat.com>
Signed-off-by: Kuniyuki Iwashima <kuniyu@amazon.com>
Acked-by: Joanne Koong <joannelkoong@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
The cited patch added support of matching on geneve option by setting
geneve_tlv_option_0_data mask and key but didn't set geneve_tlv_option_0_exist
bit which is required on some HWs when matching geneve_tlv_option_0_data parameter,
this may cause in some cases for packets to wrongly match on rules with different
geneve option.
Example of such case is packet with geneve_tlv_object class=789 and data=456
will wrongly match on rule with match geneve_tlv_object class=123 and data=456.
Fix it by setting geneve_tlv_option_0_exist bit when supported by the HW when matching
on geneve_tlv_option_0_data parameter.
Fixes: 9272e3df30 ("net/mlx5e: Geneve, Add support for encap/decap flows offload")
Signed-off-by: Maor Dickman <maord@nvidia.com>
Reviewed-by: Roi Dayan <roid@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
Current xdp xmit functions logic (mlx5e_xmit_xdp_frame_mpwqe or
mlx5e_xmit_xdp_frame), validates xdp packet length by comparing it to
hw mtu (configured at xdp sq allocation) before xmiting it. This check
does not account for ethernet fcs length (calculated and filled by the
nic). Hence, when we try sending packets with length > (hw-mtu -
ethernet-fcs-size), the device port drops it and tx_errors_phy is
incremented. Desired behavior is to catch these packets and drop them
by the driver.
Fix this behavior in XDP SQ allocation function (mlx5e_alloc_xdpsq) by
subtracting ethernet FCS header size (4 Bytes) from current hw mtu
value, since ethernet FCS is calculated and written to ethernet frames
by the nic.
Fixes: d8bec2b29a ("net/mlx5e: Support bpf_xdp_adjust_head()")
Signed-off-by: Adham Faris <afaris@nvidia.com>
Reviewed-by: Tariq Toukan <tariqt@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
The cited commit introduced a bug for multiple encapsulations flow.
If one dest encap becomes invalid, the flow is set slow path flag.
But when other dests encap become invalid, they are not cleared due
to slow path flag of the flow. When neigh-update-add is running, it
will use invalid encap.
Fix it by checking slow path flag after clearing dest encap.
Fixes: 9a5f9cc794 ("net/mlx5e: Fix possible use-after-free deleting fdb rule")
Signed-off-by: Chris Mi <cmi@nvidia.com>
Reviewed-by: Roi Dayan <roid@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
Need to use sprintf to build a string instead of sscanf. Otherwise
dirname is null and both "ct_nic" and "ct_fdb" won't be created.
But its redundant anyway as driver could be in switchdev mode but
still add nic rules. So use "ct" as folder name.
Fixes: 77422a8f6f ("net/mlx5e: CT: Add ct driver counters")
Signed-off-by: Chris Mi <cmi@nvidia.com>
Reviewed-by: Roi Dayan <roid@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
RX reporter mistakenly reads from the regular (inactive) RQ
when XSK RQ is active. Fix it here.
Fixes: 3db4c85cde ("net/mlx5e: xsk: Use queue indices starting from 0 for XSK queues")
Signed-off-by: Tariq Toukan <tariqt@nvidia.com>
Reviewed-by: Gal Pressman <gal@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
mlx5e_build_nic_params will turn CQE compression on if the hardware
capability is enabled and the slow_pci_heuristic condition is detected.
As IPoIB doesn't support CQE compression, make sure to disable the
feature in the IPoIB profile init.
Please note that the feature is not exposed to the user for IPoIB
interfaces, so it can't be subsequently turned on.
Fixes: b797a684b0 ("net/mlx5e: Enable CQE compression when PCI is slower than link")
Signed-off-by: Dragos Tatulea <dtatulea@nvidia.com>
Reviewed-by: Gal Pressman <gal@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
mlx5 PF can disable RoCE for its VFs and SFs. In such case RoCE is
marked as unsupported on those VFs/SFs.
The cited patch added an option for disable (and enable) RoCE at HCA
level. However, that commit didn't check whether RoCE is supported on
the HCA and enabled user to try and set RoCE to on.
Fix it by checking whether the HCA supports RoCE.
Fixes: fbfa97b4d7 ("net/mlx5: Disable roce at HCA level")
Signed-off-by: Shay Drory <shayd@nvidia.com>
Reviewed-by: Moshe Shemesh <moshe@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
Currently, recovery is done without considering whether the device is
still in probe flow.
This may lead to recovery before device have finished probed
successfully. e.g.: while mlx5_init_one() is running. Recovery flow is
using functionality that is loaded only by mlx5_init_one(), and there
is no point in running recovery without mlx5_init_one() finished
successfully.
Fix it by waiting for probe flow to finish and checking whether the
device is probed before trying to perform recovery.
Fixes: 51d138c261 ("net/mlx5: Fix health error state handling")
Signed-off-by: Shay Drory <shayd@nvidia.com>
Reviewed-by: Moshe Shemesh <moshe@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
io_eq_size and event_eq_size params are of param type
DEVLINK_PARAM_TYPE_U32. But, the validation callback is addressing them
as DEVLINK_PARAM_TYPE_U16.
This cause mismatch in validation in big-endian systems, in which
values in range were rejected while 268500991 was accepted.
Fix it by checking the U32 value in the validation callback.
Fixes: 0844fa5f7b ("net/mlx5: Let user configure io_eq_size param")
Signed-off-by: Shay Drory <shayd@nvidia.com>
Reviewed-by: Moshe Shemesh <moshe@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
