__nh is just a copy of nh with a different type.
Signed-off-by: Benjamin Poirier <bpoirier@nvidia.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
The writer acquires dev_base_lock with disabled bottom halves.
The reader can acquire dev_base_lock without disabling bottom halves
because there is no writer in softirq context.
On PREEMPT_RT the softirqs are preemptible and local_bh_disable() acts
as a lock to ensure that resources, that are protected by disabling
bottom halves, remain protected.
This leads to a circular locking dependency if the lock acquired with
disabled bottom halves (as in write_lock_bh()) and somewhere else with
enabled bottom halves (as by read_lock() in netstat_show()) followed by
disabling bottom halves (cxgb_get_stats() -> t4_wr_mbox_meat_timeout()
-> spin_lock_bh()). This is the reverse locking order.
All read_lock() invocation are from sysfs callback which are not invoked
from softirq context. Therefore there is no need to disable bottom
halves while acquiring a write lock.
Acquire the write lock of dev_base_lock without disabling bottom halves.
Reported-by: Pei Zhang <pezhang@redhat.com>
Reported-by: Luis Claudio R. Goncalves <lgoncalv@redhat.com>
Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Signed-off-by: David S. Miller <davem@davemloft.net>
Previously commit e02d494d2c ("l2tp: Convert rwlock to RCU") converted
most, but not all, rwlock instances in the l2tp subsystem to RCU.
The remaining rwlock protects the per-tunnel hashlist of sessions which
is used for session lookups in the UDP-encap data path.
Convert the remaining rwlock to rcu to improve performance of UDP-encap
tunnels.
Note that the tunnel and session, which both live on RCU-protected
lists, use slightly different approaches to incrementing their refcounts
in the various getter functions.
The tunnel has to use refcount_inc_not_zero because the tunnel shutdown
process involves dropping the refcount to zero prior to synchronizing
RCU readers (via. kfree_rcu).
By contrast, the session shutdown removes the session from the list(s)
it is on, synchronizes with readers, and then decrements the session
refcount. Since the getter functions increment the session refcount
with the RCU read lock held we prevent getters seeing a zero session
refcount, and therefore don't need to use refcount_inc_not_zero.
Signed-off-by: Tom Parkin <tparkin@katalix.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
When we bind an AF_UNIX socket without a name specified, the kernel selects
an available one from 0x00000 to 0xFFFFF. unix_autobind() starts searching
from a number in the 'static' variable and increments it after acquiring
two locks.
If multiple processes try autobind, they obtain the same lock and check if
a socket in the hash list has the same name. If not, one process uses it,
and all except one end up retrying the _next_ number (actually not, it may
be incremented by the other processes). The more we autobind sockets in
parallel, the longer the latency gets. We can avoid such a race by
searching for a name from a random number.
These show latency in unix_autobind() while 64 CPUs are simultaneously
autobind-ing 1024 sockets for each.
Without this patch:
usec : count distribution
0 : 1176 |*** |
2 : 3655 |*********** |
4 : 4094 |************* |
6 : 3831 |************ |
8 : 3829 |************ |
10 : 3844 |************ |
12 : 3638 |*********** |
14 : 2992 |********* |
16 : 2485 |******* |
18 : 2230 |******* |
20 : 2095 |****** |
22 : 1853 |***** |
24 : 1827 |***** |
26 : 1677 |***** |
28 : 1473 |**** |
30 : 1573 |***** |
32 : 1417 |**** |
34 : 1385 |**** |
36 : 1345 |**** |
38 : 1344 |**** |
40 : 1200 |*** |
With this patch:
usec : count distribution
0 : 1855 |****** |
2 : 6464 |********************* |
4 : 9936 |******************************** |
6 : 12107 |****************************************|
8 : 10441 |********************************** |
10 : 7264 |*********************** |
12 : 4254 |************** |
14 : 2538 |******** |
16 : 1596 |***** |
18 : 1088 |*** |
20 : 800 |** |
22 : 670 |** |
24 : 601 |* |
26 : 562 |* |
28 : 525 |* |
30 : 446 |* |
32 : 378 |* |
34 : 337 |* |
36 : 317 |* |
38 : 314 |* |
40 : 298 | |
Signed-off-by: Kuniyuki Iwashima <kuniyu@amazon.co.jp>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
To replace unix_table_lock with per-hash locks in the next patch, we need
to save a hash in each socket because /proc/net/unix or BPF prog iterate
sockets while holding a hash table lock and release it later in a different
function.
Currently, we store a real/pseudo hash in struct unix_address. However, we
do not allocate it to unbound sockets, nor should we do just for that. For
this purpose, we can use sk_hash. Then, we no longer use the hash field in
struct unix_address and can remove it.
Also, this patch does
- rename unix_insert_socket() to unix_insert_unbound_socket()
- remove the redundant list argument from __unix_insert_socket() and
unix_insert_unbound_socket()
- use 'unsigned int' instead of 'unsigned' in __unix_set_addr_hash()
- remove 'inline' from unix_remove_socket() and
unix_insert_unbound_socket().
Signed-off-by: Kuniyuki Iwashima <kuniyu@amazon.co.jp>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
This patch adds three helper functions that calculate hashes for unbound
sockets and bound sockets with BSD/abstract addresses.
Signed-off-by: Kuniyuki Iwashima <kuniyu@amazon.co.jp>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
In BSD and abstract address cases, we store sockets in the hash table with
keys between 0 and UNIX_HASH_SIZE - 1. However, the hash saved in a socket
varies depending on its address type; sockets with BSD addresses always
have UNIX_HASH_SIZE in their unix_sk(sk)->addr->hash.
This is just for the UNIX_ABSTRACT() macro used to check the address type.
The difference of the saved hashes comes from the first byte of the address
in the first place. So, we can test it directly.
Then we can keep a real hash in each socket and replace unix_table_lock
with per-hash locks in the later patch.
Signed-off-by: Kuniyuki Iwashima <kuniyu@amazon.co.jp>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
To terminate address with '\0' in unix_bind_bsd(), we add
unix_create_addr() and call it in unix_bind_bsd() and unix_bind_abstract().
Also, unix_bind_abstract() does not return -EEXIST. Only
kern_path_create() and vfs_mknod() in unix_bind_bsd() can return it,
so we move the last error check in unix_bind() to unix_bind_bsd().
Signed-off-by: Kuniyuki Iwashima <kuniyu@amazon.co.jp>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
This patch removes unix_mkname() and postpones calculating a hash to
unix_bind_abstract(). Some BSD stuffs still remain in unix_bind()
though, the next patch packs them into unix_bind_bsd().
Signed-off-by: Kuniyuki Iwashima <kuniyu@amazon.co.jp>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
We should not call unix_mkname() before unix_find_other() and instead do
the same thing where necessary based on the address type:
- terminating the address with '\0' in unix_find_bsd()
- calculating the hash in unix_find_abstract().
Signed-off-by: Kuniyuki Iwashima <kuniyu@amazon.co.jp>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
unix_mkname() tests socket address length and family and does some
processing based on the address type. It is called in the early stage,
and therefore some instructions are redundant and can end up in vain.
The address length/family tests are done twice in unix_bind(). Also, the
address type is rechecked later in unix_bind() and unix_find_other(), where
we can do the same processing. Moreover, in the BSD address case, the hash
is set to 0 but never used and confusing.
This patch moves the address tests out of unix_mkname(), and the following
patches move the other part into appropriate places and remove
unix_mkname() finally.
Signed-off-by: Kuniyuki Iwashima <kuniyu@amazon.co.jp>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
We can return an error as a pointer and need not pass an additional
argument to unix_find_other().
Signed-off-by: Kuniyuki Iwashima <kuniyu@amazon.co.jp>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
As done in the commit fa42d910a3 ("unix_bind(): take BSD and abstract
address cases into new helpers"), this patch moves BSD and abstract address
cases from unix_find_other() into unix_find_bsd() and unix_find_abstract().
Signed-off-by: Kuniyuki Iwashima <kuniyu@amazon.co.jp>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
We do not use struct socket in unix_autobind() and pass struct sock to
unix_bind_bsd() and unix_bind_abstract(). Let's pass it to unix_autobind()
as well.
Also, this patch fixes these errors by checkpatch.pl.
ERROR: do not use assignment in if condition
#1795: FILE: net/unix/af_unix.c:1795:
+ if (test_bit(SOCK_PASSCRED, &sock->flags) && !u->addr
CHECK: Logical continuations should be on the previous line
#1796: FILE: net/unix/af_unix.c:1796:
+ if (test_bit(SOCK_PASSCRED, &sock->flags) && !u->addr
+ && (err = unix_autobind(sock)) != 0)
Signed-off-by: Kuniyuki Iwashima <kuniyu@amazon.co.jp>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
The length of the AF_UNIX socket address contains an offset to the member
sun_path of struct sockaddr_un.
Currently, the preceding member is just sun_family, and its type is
sa_family_t and resolved to short. Therefore, the offset is represented by
sizeof(short). However, it is not clear and fragile to changes in struct
sockaddr_storage or sockaddr_un.
This commit makes it clear and robust by rewriting sizeof() with
offsetof().
Signed-off-by: Kuniyuki Iwashima <kuniyu@amazon.co.jp>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
The netdev (e.g. ifb, bareudp), which not support ethtool ops
(e.g. .get_drvinfo), we can use the rtnl kind as a default name.
ifb netdev may be created by others prefix, not ifbX.
Cc: Arnd Bergmann <arnd@arndb.de>
Cc: Hao Chen <chenhao288@hisilicon.com>
Cc: Heiner Kallweit <hkallweit1@gmail.com>
Cc: "Gustavo A. R. Silva" <gustavoars@kernel.org>
Cc: Danielle Ratson <danieller@nvidia.com>
Cc: Andrew Lunn <andrew@lunn.ch>
Signed-off-by: Tonghao Zhang <xiangxia.m.yue@gmail.com>
Reviewed-by: Leon Romanovsky <leonro@nvidia.com>
Link: https://lore.kernel.org/r/20211125163049.84970-1-xiangxia.m.yue@gmail.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
When applications call shutdown() with SHUT_RDWR in userspace,
smc_close_active() calls kernel_sock_shutdown(), and it is called
twice in smc_shutdown().
This fixes this by checking sk_state before do clcsock shutdown, and
avoids missing the application's call of smc_shutdown().
Link: https://lore.kernel.org/linux-s390/1f67548e-cbf6-0dce-82b5-10288a4583bd@linux.ibm.com/
Fixes: 606a63c978 ("net/smc: Ensure the active closing peer first closes clcsock")
Signed-off-by: Tony Lu <tonylu@linux.alibaba.com>
Reviewed-by: Wen Gu <guwen@linux.alibaba.com>
Acked-by: Karsten Graul <kgraul@linux.ibm.com>
Link: https://lore.kernel.org/r/20211126024134.45693-1-tonylu@linux.alibaba.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Inject error before dev_hold(real_dev) in register_vlan_dev(),
and execute the following testcase:
ip link add dev dummy1 type dummy
ip link add name dummy1.100 link dummy1 type vlan id 100
ip link del dev dummy1
When the dummy netdevice is removed, we will get a WARNING as following:
=======================================================================
refcount_t: decrement hit 0; leaking memory.
WARNING: CPU: 2 PID: 0 at lib/refcount.c:31 refcount_warn_saturate+0xbf/0x1e0
and an endless loop of:
=======================================================================
unregister_netdevice: waiting for dummy1 to become free. Usage count = -1073741824
That is because dev_put(real_dev) in vlan_dev_free() be called without
dev_hold(real_dev) in register_vlan_dev(). It makes the refcnt of real_dev
underflow.
Move the dev_hold(real_dev) to vlan_dev_init() which is the call-back of
ndo_init(). That makes dev_hold() and dev_put() for vlan's real_dev
symmetrical.
Fixes: 563bcbae3b ("net: vlan: fix a UAF in vlan_dev_real_dev()")
Reported-by: Petr Machata <petrm@nvidia.com>
Suggested-by: Jakub Kicinski <kuba@kernel.org>
Signed-off-by: Ziyang Xuan <william.xuanziyang@huawei.com>
Link: https://lore.kernel.org/r/20211126015942.2918542-1-william.xuanziyang@huawei.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
ethtool_set_coalesce() now uses both the .get_coalesce() and
.set_coalesce() callbacks. But the check for their availability is
buggy, so changing the coalesce settings on a device where the driver
provides only _one_ of the callbacks results in a NULL pointer
dereference instead of an -EOPNOTSUPP.
Fix the condition so that the availability of both callbacks is
ensured. This also matches the netlink code.
Note that reproducing this requires some effort - it only affects the
legacy ioctl path, and needs a specific combination of driver options:
- have .get_coalesce() and .coalesce_supported but no
.set_coalesce(), or
- have .set_coalesce() but no .get_coalesce(). Here eg. ethtool doesn't
cause the crash as it first attempts to call ethtool_get_coalesce()
and bails out on error.
Fixes: f3ccfda193 ("ethtool: extend coalesce setting uAPI with CQE mode")
Cc: Yufeng Mo <moyufeng@huawei.com>
Cc: Huazhong Tan <tanhuazhong@huawei.com>
Cc: Andrew Lunn <andrew@lunn.ch>
Cc: Heiner Kallweit <hkallweit1@gmail.com>
Signed-off-by: Julian Wiedmann <jwi@linux.ibm.com>
Link: https://lore.kernel.org/r/20211126175543.28000-1-jwi@linux.ibm.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Currently, the probe timer is reused as the raise timer when PLPMTUD is in
the Search Complete state. raise_count was introduced to count how many
times the probe timer has timed out. When raise_count reaches to 30, the
raise timer handler will be triggered.
During the whole processing above, the timer keeps timing out every probe_
interval. It is a waste for the Search Complete state, as the raise timer
only needs to time out after 30 * probe_interval.
Since the raise timer and probe timer are never used at the same time, it
is no need to keep probe timer 'alive' in the Search Complete state. This
patch to introduce sctp_transport_reset_raise_timer() to start the timer
as the raise timer when entering the Search Complete state. When entering
the other states, sctp_transport_reset_probe_timer() will still be called
to reset the timer to the probe timer.
raise_count can be removed from sctp_transport as no need to count probe
timer timeout for raise timer timeout. last_rtx_chunks can be removed as
sctp_transport_reset_probe_timer() can be called in the place where asoc
rtx_data_chunks is changed.
Signed-off-by: Xin Long <lucien.xin@gmail.com>
Acked-by: Marcelo Ricardo Leitner <marcelo.leitner@gmail.com>
Link: https://lore.kernel.org/r/edb0e48988ea85997488478b705b11ddc1ba724a.1637781974.git.lucien.xin@gmail.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
When a skb comes to tipc_aead_encrypt(), it's always linear. The
unlikely check 'skb_cloned(skb) && tailen <= skb_tailroom(skb)'
can completely be taken care of in skb_cow_data() by the code
in branch "if (!skb_has_frag_list())".
Also, remove the 'TODO:' annotation, as the pages in skbs are not
writable, see more on commit 3cf4375a09 ("tipc: do not write
skb_shinfo frags when doing decrytion").
Signed-off-by: Xin Long <lucien.xin@gmail.com>
Acked-by: Jon Maloy <jmaloy@redhat.com>
Link: https://lore.kernel.org/r/47a478da0b6095b76e3cbe7a75cbd25d9da1df9a.1637773872.git.lucien.xin@gmail.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
We replace proto_ops whenever TLS is configured for RX. But our
replacement also overrides sendpage_locked, which will crash
unless TX is also configured. Similarly we plug both of those
in for TLS_HW (NIC crypto offload) even tho TLS_HW has a completely
different implementation for TX.
Last but not least we always plug in something based on inet_stream_ops
even though a few of the callbacks differ for IPv6 (getname, release,
bind).
Use a callback building method similar to what we do for struct proto.
Fixes: c46234ebb4 ("tls: RX path for ktls")
Fixes: d4ffb02dee ("net/tls: enable sk_msg redirect to tls socket egress")
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
recvmsg() will put peek()ed and partially read records onto the rx_list.
splice_read() needs to consult that list otherwise it may miss data.
Align with recvmsg() and also put partially-read records onto rx_list.
tls_sw_advance_skb() is pretty pointless now and will be removed in
net-next.
Fixes: 692d7b5d1f ("tls: Fix recvmsg() to be able to peek across multiple records")
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
We don't support splicing control records. TLS 1.3 changes moved
the record type check into the decrypt if(). The skb may already
be decrypted and still be an alert.
Note that decrypt_skb_update() is idempotent and updates ctx->decrypted
so the if() is pointless.
Reorder the check for decryption errors with the content type check
while touching them. This part is not really a bug, because if
decryption failed in TLS 1.3 content type will be DATA, and for
TLS 1.2 it will be correct. Nevertheless its strange to touch output
before checking if the function has failed.
Fixes: fedf201e12 ("net: tls: Refactor control message handling on recv")
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
The kernel_listen function in smc_listen will fail when all the available
ports are occupied. At this point smc->clcsock->sk->sk_data_ready has
been changed to smc_clcsock_data_ready. When we call smc_listen again,
now both smc->clcsock->sk->sk_data_ready and smc->clcsk_data_ready point
to the smc_clcsock_data_ready function.
The smc_clcsock_data_ready() function calls lsmc->clcsk_data_ready which
now points to itself resulting in an infinite loop.
This patch restores smc->clcsock->sk->sk_data_ready with the old value.
Fixes: a60a2b1e0a ("net/smc: reduce active tcp_listen workers")
Signed-off-by: Guo DaXing <guodaxing@huawei.com>
Acked-by: Tony Lu <tonylu@linux.alibaba.com>
Signed-off-by: Karsten Graul <kgraul@linux.ibm.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Coverity reports a possible NULL dereferencing problem:
in smc_vlan_by_tcpsk():
6. returned_null: netdev_lower_get_next returns NULL (checked 29 out of 30 times).
7. var_assigned: Assigning: ndev = NULL return value from netdev_lower_get_next.
1623 ndev = (struct net_device *)netdev_lower_get_next(ndev, &lower);
CID 1468509 (#1 of 1): Dereference null return value (NULL_RETURNS)
8. dereference: Dereferencing a pointer that might be NULL ndev when calling is_vlan_dev.
1624 if (is_vlan_dev(ndev)) {
Remove the manual implementation and use netdev_walk_all_lower_dev() to
iterate over the lower devices. While on it remove an obsolete function
parameter comment.
Fixes: cb9d43f677 ("net/smc: determine vlan_id of stacked net_device")
Suggested-by: Julian Wiedmann <jwi@linux.ibm.com>
Signed-off-by: Karsten Graul <kgraul@linux.ibm.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
This is to match ipv4 behaviour, see __ip_sock_set_tos()
implementation.
Technically for ipv6 this might not be required because normally we
do not allow tclass to influence routing, yet the cli tooling does
support it:
lpk11:~# ip -6 rule add pref 5 tos 45 lookup 5
lpk11:~# ip -6 rule
5: from all tos 0x45 lookup 5
and in general dscp/tclass based routing does make sense.
We already have cases where dscp can affect vlan priority and/or
transmit queue (especially on wifi).
So let's just make things match. Easier to reason about and no harm.
Cc: Eric Dumazet <edumazet@google.com>
Cc: Neal Cardwell <ncardwell@google.com>
Signed-off-by: Maciej Żenczykowski <maze@google.com>
Link: https://lore.kernel.org/r/20211123223208.1117871-1-zenczykowski@gmail.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
This is to match ipv4 behaviour, see __ip_sock_set_tos()
implementation at ipv4/ip_sockglue.c:579
void __ip_sock_set_tos(struct sock *sk, int val)
{
if (sk->sk_type == SOCK_STREAM) {
val &= ~INET_ECN_MASK;
val |= inet_sk(sk)->tos & INET_ECN_MASK;
}
if (inet_sk(sk)->tos != val) {
inet_sk(sk)->tos = val;
sk->sk_priority = rt_tos2priority(val);
sk_dst_reset(sk);
}
}
Cc: Neal Cardwell <ncardwell@google.com>
Signed-off-by: Maciej Żenczykowski <maze@google.com>
Reviewed-by: Eric Dumazet <edumazet@google.com>
Link: https://lore.kernel.org/r/20211123223154.1117794-1-zenczykowski@gmail.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
A CAP_NET_RAW capable process can already spoof (on transmit) anything
it desires via raw packet sockets... There is no good reason to not
allow it to also be able to play routing tricks on packets from its
own normal sockets.
There is a desire to be able to use SO_MARK for routing table selection
(via ip rule fwmark) from within a user process without having to run
it as root. Granting it CAP_NET_RAW is much less dangerous than
CAP_NET_ADMIN (CAP_NET_RAW doesn't permit persistent state change,
while CAP_NET_ADMIN does - by for example allowing the reconfiguration
of the routing tables and/or bringing up/down devices).
Let's keep CAP_NET_ADMIN for persistent state changes,
while using CAP_NET_RAW for non-configuration related stuff.
Signed-off-by: Maciej Żenczykowski <maze@google.com>
Link: https://lore.kernel.org/r/20211123203715.193413-1-zenczykowski@gmail.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
CAP_NET_ADMIN is and should continue to be about configuring the
system as a whole, not about configuring per-socket or per-packet
parameters.
Sending and receiving raw packets is what CAP_NET_RAW is all about.
It can already send packets with any VLAN tag, and any IPv4 TOS
mark, and any IPv6 TCLASS mark, simply by virtue of building
such a raw packet. Not to mention using any protocol and source/
/destination ip address/port tuple.
These are the fields that networking gear uses to prioritize packets.
Hence, a CAP_NET_RAW process is already capable of affecting traffic
prioritization after it hits the wire. This change makes it capable
of affecting traffic prioritization even in the host at the nic and
before that in the queueing disciplines (provided skb->priority is
actually being used for prioritization, and not the TOS/TCLASS field)
Hence it makes sense to allow a CAP_NET_RAW process to set the
priority of sockets and thus packets it sends.
Signed-off-by: Maciej Żenczykowski <maze@google.com>
Link: https://lore.kernel.org/r/20211123203702.193221-1-zenczykowski@gmail.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
While testing BIG TCP patch series, I was expecting that TCP_RR workloads
with 80KB requests/answers would send one 80KB TSO packet,
then being received as a single GRO packet.
It turns out this was not happening, and the root cause was that
cubic Hystart ACK train was triggering after a few (2 or 3) rounds of RPC.
Hystart was wrongly setting CWND/SSTHRESH to 30, while my RPC
needed a budget of ~20 segments.
Ideally these TCP_RR flows should not exit slow start.
Cubic Hystart should reset itself at each round, instead of assuming
every TCP flow is a bulk one.
Note that even after this patch, Hystart can still trigger, depending
on scheduling artifacts, but at a higher CWND/SSTHRESH threshold,
keeping optimal TSO packet sizes.
Tested:
ip link set dev eth0 gro_ipv6_max_size 131072 gso_ipv6_max_size 131072
nstat -n; netperf -H ... -t TCP_RR -l 5 -- -r 80000,80000 -K cubic; nstat|egrep "Ip6InReceives|Hystart|Ip6OutRequests"
Before:
8605
Ip6InReceives 87541 0.0
Ip6OutRequests 129496 0.0
TcpExtTCPHystartTrainDetect 1 0.0
TcpExtTCPHystartTrainCwnd 30 0.0
After:
8760
Ip6InReceives 88514 0.0
Ip6OutRequests 87975 0.0
Fixes: ae27e98a51 ("[TCP] CUBIC v2.3")
Co-developed-by: Neal Cardwell <ncardwell@google.com>
Signed-off-by: Neal Cardwell <ncardwell@google.com>
Signed-off-by: Eric Dumazet <edumazet@google.com>
Cc: Stephen Hemminger <stephen@networkplumber.org>
Cc: Yuchung Cheng <ycheng@google.com>
Cc: Soheil Hassas Yeganeh <soheil@google.com>
Link: https://lore.kernel.org/r/20211123202535.1843771-1-eric.dumazet@gmail.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Cited commit converted simple_strtoul() to kstrtoul() as suggested by
the former's documentation. However, it also forced all the inputs to be
decimal resulting in user space breakage.
Fix by setting the base to '0' so that the base is automatically
detected.
Before:
# ip link add name br0 type bridge vlan_filtering 1
# echo "0x88a8" > /sys/class/net/br0/bridge/vlan_protocol
bash: echo: write error: Invalid argument
After:
# ip link add name br0 type bridge vlan_filtering 1
# echo "0x88a8" > /sys/class/net/br0/bridge/vlan_protocol
# echo $?
0
Fixes: 520fbdf7fb ("net/bridge: replace simple_strtoul to kstrtol")
Signed-off-by: Ido Schimmel <idosch@nvidia.com>
Acked-by: Nikolay Aleksandrov <nikolay@nvidia.com>
Link: https://lore.kernel.org/r/20211124101122.3321496-1-idosch@idosch.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
All gro_complete() handlers are called from napi_gro_complete()
while rcu_read_lock() has been called.
There is no point stacking more rcu_read_lock()
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
All gro_receive() handlers are called from dev_gro_receive()
while rcu_read_lock() has been called.
There is no point stacking more rcu_read_lock()
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Update NC-SI command handler (both standard and OEM) to take into
account of payload paddings in allocating skb (in case of payload
size is not 32-bit aligned).
The checksum field follows payload field, without taking payload
padding into account can cause checksum being truncated, leading to
dropped packets.
Fixes: fb4ee67529 ("net/ncsi: Add NCSI OEM command support")
Signed-off-by: Kumar Thangavel <thangavel.k@hcl.com>
Acked-by: Samuel Mendoza-Jonas <sam@mendozajonas.com>
Reviewed-by: Paul Menzel <pmenzel@molgen.mpg.de>
Signed-off-by: David S. Miller <davem@davemloft.net>
This patch inlines dccp_listen_start() and removes a stale comment in
inet_dccp_listen() so that it looks like inet_listen().
Signed-off-by: Kuniyuki Iwashima <kuniyu@amazon.co.jp>
Reviewed-by: Richard Sailer <richard_siegfried@systemli.org>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
The commit 1295e2cf30 ("inet: minor optimization for backlog setting in
listen(2)") added change so that sk_max_ack_backlog is initialised earlier
in inet_dccp_listen() and inet_listen(). Since then, we no longer use
backlog in inet_csk_listen_start(), so let's remove it.
Signed-off-by: Kuniyuki Iwashima <kuniyu@amazon.co.jp>
Acked-by: Yafang Shao <laoar.shao@gmail.com>
Reviewed-by: Richard Sailer <richard_siegfried@systemli.org>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
.ndo_change_proto_down was added seemingly to enable out-of-tree
implementations. Over 2.5yrs later we still have no real users
upstream. Hardwire the generic implementation for now, we can
revert once real users materialize. (rocker is a test vehicle,
not a user.)
We need to drop the optimization on the sysfs side, because
unlike ndos priv_flags will be changed at runtime, so we'd
need READ_ONCE/WRITE_ONCE everywhere..
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
The side that actively closed socket, it's clcsock doesn't enter
TIME_WAIT state, but the passive side does it. It should show the same
behavior as TCP sockets.
Consider this, when client actively closes the socket, the clcsock in
server enters TIME_WAIT state, which means the address is occupied and
won't be reused before TIME_WAIT dismissing. If we restarted server, the
service would be unavailable for a long time.
To solve this issue, shutdown the clcsock in [A], perform the TCP active
close progress first, before the passive closed side closing it. So that
the actively closed side enters TIME_WAIT, not the passive one.
Client | Server
close() // client actively close |
smc_release() |
smc_close_active() // PEERCLOSEWAIT1 |
smc_close_final() // abort or closed = 1|
smc_cdc_get_slot_and_msg_send() |
[A] |
|smc_cdc_msg_recv_action() // ACTIVE
| queue_work(smc_close_wq, &conn->close_work)
| smc_close_passive_work() // PROCESSABORT or APPCLOSEWAIT1
| smc_close_passive_abort_received() // only in abort
|
|close() // server recv zero, close
| smc_release() // PROCESSABORT or APPCLOSEWAIT1
| smc_close_active()
| smc_close_abort() or smc_close_final() // CLOSED
| smc_cdc_get_slot_and_msg_send() // abort or closed = 1
smc_cdc_msg_recv_action() | smc_clcsock_release()
queue_work(smc_close_wq, &conn->close_work) | sock_release(tcp) // actively close clc, enter TIME_WAIT
smc_close_passive_work() // PEERCLOSEWAIT1 | smc_conn_free()
smc_close_passive_abort_received() // CLOSED|
smc_conn_free() |
smc_clcsock_release() |
sock_release(tcp) // passive close clc |
Link: https://www.spinics.net/lists/netdev/msg780407.html
Fixes: b38d732477 ("smc: socket closing and linkgroup cleanup")
Signed-off-by: Tony Lu <tonylu@linux.alibaba.com>
Reviewed-by: Wen Gu <guwen@linux.alibaba.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
There remains some variables to replace with local struct sock. So clean
them up all.
Fixes: 3163c5071f ("net/smc: use local struct sock variables consistently")
Signed-off-by: Tony Lu <tonylu@linux.alibaba.com>
Reviewed-by: Wen Gu <guwen@linux.alibaba.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Add a new device generic parameter to enable and disable
iWARP functionality on a multi-protocol RDMA device.
Signed-off-by: Shiraz Saleem <shiraz.saleem@intel.com>
Tested-by: Leszek Kaliszczuk <leszek.kaliszczuk@intel.com>
Signed-off-by: Tony Nguyen <anthony.l.nguyen@intel.com>
When replacing a nexthop group, we must release the IPv6 per-cpu dsts of
the removed nexthop entries after an RCU grace period because they
contain references to the nexthop's net device and to the fib6 info.
With specific series of events[1] we can reach net device refcount
imbalance which is unrecoverable. IPv4 is not affected because dsts
don't take a refcount on the route.
[1]
$ ip nexthop list
id 200 via 2002:db8::2 dev bridge.10 scope link onlink
id 201 via 2002:db8::3 dev bridge scope link onlink
id 203 group 201/200
$ ip -6 route
2001:db8::10 nhid 203 metric 1024 pref medium
nexthop via 2002:db8::3 dev bridge weight 1 onlink
nexthop via 2002:db8::2 dev bridge.10 weight 1 onlink
Create rt6_info through one of the multipath legs, e.g.:
$ taskset -a -c 1 ./pkt_inj 24 bridge.10 2001:db8::10
(pkt_inj is just a custom packet generator, nothing special)
Then remove that leg from the group by replace (let's assume it is id
200 in this case):
$ ip nexthop replace id 203 group 201
Now remove the IPv6 route:
$ ip -6 route del 2001:db8::10/128
The route won't be really deleted due to the stale rt6_info holding 1
refcnt in nexthop id 200.
At this point we have the following reference count dependency:
(deleted) IPv6 route holds 1 reference over nhid 203
nh 203 holds 1 ref over id 201
nh 200 holds 1 ref over the net device and the route due to the stale
rt6_info
Now to create circular dependency between nh 200 and the IPv6 route, and
also to get a reference over nh 200, restore nhid 200 in the group:
$ ip nexthop replace id 203 group 201/200
And now we have a permanent circular dependncy because nhid 203 holds a
reference over nh 200 and 201, but the route holds a ref over nh 203 and
is deleted.
To trigger the bug just delete the group (nhid 203):
$ ip nexthop del id 203
It won't really be deleted due to the IPv6 route dependency, and now we
have 2 unlinked and deleted objects that reference each other: the group
and the IPv6 route. Since the group drops the reference it holds over its
entries at free time (i.e. its own refcount needs to drop to 0) that will
never happen and we get a permanent ref on them, since one of the entries
holds a reference over the IPv6 route it will also never be released.
At this point the dependencies are:
(deleted, only unlinked) IPv6 route holds reference over group nh 203
(deleted, only unlinked) group nh 203 holds reference over nh 201 and 200
nh 200 holds 1 ref over the net device and the route due to the stale
rt6_info
This is the last point where it can be fixed by running traffic through
nh 200, and specifically through the same CPU so the rt6_info (dst) will
get released due to the IPv6 genid, that in turn will free the IPv6
route, which in turn will free the ref count over the group nh 203.
If nh 200 is deleted at this point, it will never be released due to the
ref from the unlinked group 203, it will only be unlinked:
$ ip nexthop del id 200
$ ip nexthop
$
Now we can never release that stale rt6_info, we have IPv6 route with ref
over group nh 203, group nh 203 with ref over nh 200 and 201, nh 200 with
rt6_info (dst) with ref over the net device and the IPv6 route. All of
these objects are only unlinked, and cannot be released, thus they can't
release their ref counts.
Message from syslogd@dev at Nov 19 14:04:10 ...
kernel:[73501.828730] unregister_netdevice: waiting for bridge.10 to become free. Usage count = 3
Message from syslogd@dev at Nov 19 14:04:20 ...
kernel:[73512.068811] unregister_netdevice: waiting for bridge.10 to become free. Usage count = 3
Fixes: 7bf4796dd0 ("nexthops: add support for replace")
Signed-off-by: Nikolay Aleksandrov <nikolay@nvidia.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
We need a way to release a fib6_nh's per-cpu dsts when replacing
nexthops otherwise we can end up with stale per-cpu dsts which hold net
device references, so add a new IPv6 stub called fib6_nh_release_dsts.
It must be used after an RCU grace period, so no new dsts can be created
through a group's nexthop entry.
Similar to fib6_nh_release it shouldn't be used if fib6_nh_init has failed
so it doesn't need a dummy stub when IPv6 is not enabled.
Fixes: 7bf4796dd0 ("nexthops: add support for replace")
Signed-off-by: Nikolay Aleksandrov <nikolay@nvidia.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
In preparation for FORTIFY_SOURCE performing compile-time and run-time
field bounds checking for memcpy(), memmove(), and memset(), avoid
intentionally writing across neighboring fields.
Replace the existing empty member position markers "headers_start" and
"headers_end" with a struct_group(). This will allow memcpy() and sizeof()
to more easily reason about sizes, and improve readability.
"pahole" shows no size nor member offset changes to struct sk_buff.
"objdump -d" shows no object code changes (outside of WARNs affected by
source line number changes).
Signed-off-by: Kees Cook <keescook@chromium.org>
Reviewed-by: Gustavo A. R. Silva <gustavoars@kernel.org>
Reviewed-by: Jason A. Donenfeld <Jason@zx2c4.com> # drivers/net/wireguard/*
Link: https://lore.kernel.org/lkml/20210728035006.GD35706@embeddedor
Signed-off-by: David S. Miller <davem@davemloft.net>
In preparation for using the struct_group() macro in struct sk_buff,
move the conditional preprocessor directives out of the region of struct
sk_buff that will be enclosed by struct_group(). While GCC and Clang are
happy with conditional preprocessor directives here, sparse is not, even
under -Wno-directive-within-macro[1], as would be seen under a C=1 build:
net/core/filter.c: note: in included file (through include/linux/netlink.h, include/linux/sock_diag.h):
./include/linux/skbuff.h:820:1: warning: directive in macro's argument list
./include/linux/skbuff.h:822:1: warning: directive in macro's argument list
./include/linux/skbuff.h:846:1: warning: directive in macro's argument list
./include/linux/skbuff.h:848:1: warning: directive in macro's argument list
Additionally remove empty macro argument definitions and usage.
"objdump -d" shows no object code differences.
[1] https://www.spinics.net/lists/linux-sparse/msg10857.html
Signed-off-by: Kees Cook <keescook@chromium.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
