Age | Commit message (Collapse) | Author |
|
This commit changes virtio/vsock to use sk_buff instead of
virtio_vsock_pkt. Beyond better conforming to other net code, using
sk_buff allows vsock to use sk_buff-dependent features in the future
(such as sockmap) and improves throughput.
This patch introduces the following performance changes:
Tool: Uperf
Env: Phys Host + L1 Guest
Payload: 64k
Threads: 16
Test Runs: 10
Type: SOCK_STREAM
Before: commit b7bfaa761d760 ("Linux 6.2-rc3")
Before
------
g2h: 16.77Gb/s
h2g: 10.56Gb/s
After
-----
g2h: 21.04Gb/s
h2g: 10.76Gb/s
Signed-off-by: Bobby Eshleman <bobby.eshleman@bytedance.com>
Reviewed-by: Stefano Garzarella <sgarzare@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
After switching to TCP_ESTABLISHED or TCP_LISTEN sk_state, alive SOCK_STREAM
and SOCK_SEQPACKET sockets can't change it anymore (since commit 3ff8bff704f4
"unix: Fix race in SOCK_SEQPACKET's unix_dgram_sendmsg()").
Thus, we do not need to take lock here.
Signed-off-by: Kirill Tkhai <tkhai@ya.ru>
Reviewed-by: Kuniyuki Iwashima <kuniyu@amazon.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
Revert a wrong fix that was done during the review process. The
intention was to substitute "if(ret < 0)" with "if(ret)".
Unfortunately, the intended fix did not meet the code.
Besides, after additional review, it was decided that "if(ret < 0)"
was actually the right thing to do.
Fixes: 8580e16c28f3 ("net/ethtool: add netlink interface for the PLCA RS")
Signed-off-by: Piergiorgio Beruto <piergiorgio.beruto@gmail.com>
Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Reviewed-by: Eric Dumazet <edumazet@google.com>
Link: https://lore.kernel.org/r/f2277af8951a51cfee2fb905af8d7a812b7beaf4.1673616357.git.piergiorgio.beruto@gmail.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
|
|
In ip6_dst_gc() replace:
if (entries > gc_thresh)
With:
if (entries > ops->gc_thresh)
Sending Ipv6 packets in a loop via a raw socket triggers an issue where a
route is cloned by ip6_rt_cache_alloc() for each packet sent. This quickly
consumes the Ipv6 max_size threshold which defaults to 4096 resulting in
these warnings:
[1] 99.187805] dst_alloc: 7728 callbacks suppressed
[2] Route cache is full: consider increasing sysctl net.ipv6.route.max_size.
.
.
[300] Route cache is full: consider increasing sysctl net.ipv6.route.max_size.
When this happens the packet is dropped and sendto() gets a network is
unreachable error:
remaining pkt 200557 errno 101
remaining pkt 196462 errno 101
.
.
remaining pkt 126821 errno 101
Implement David Aherns suggestion to remove max_size check seeing that Ipv6
has a GC to manage memory usage. Ipv4 already does not check max_size.
Here are some memory comparisons for Ipv4 vs Ipv6 with the patch:
Test by running 5 instances of a program that sends UDP packets to a raw
socket 5000000 times. Compare Ipv4 and Ipv6 performance with a similar
program.
Ipv4:
Before test:
MemFree: 29427108 kB
Slab: 237612 kB
ip6_dst_cache 1912 2528 256 32 2 : tunables 0 0 0
xfrm_dst_cache 0 0 320 25 2 : tunables 0 0 0
ip_dst_cache 2881 3990 192 42 2 : tunables 0 0 0
During test:
MemFree: 29417608 kB
Slab: 247712 kB
ip6_dst_cache 1912 2528 256 32 2 : tunables 0 0 0
xfrm_dst_cache 0 0 320 25 2 : tunables 0 0 0
ip_dst_cache 44394 44394 192 42 2 : tunables 0 0 0
After test:
MemFree: 29422308 kB
Slab: 238104 kB
ip6_dst_cache 1912 2528 256 32 2 : tunables 0 0 0
xfrm_dst_cache 0 0 320 25 2 : tunables 0 0 0
ip_dst_cache 3048 4116 192 42 2 : tunables 0 0 0
Ipv6 with patch:
Errno 101 errors are not observed anymore with the patch.
Before test:
MemFree: 29422308 kB
Slab: 238104 kB
ip6_dst_cache 1912 2528 256 32 2 : tunables 0 0 0
xfrm_dst_cache 0 0 320 25 2 : tunables 0 0 0
ip_dst_cache 3048 4116 192 42 2 : tunables 0 0 0
During Test:
MemFree: 29431516 kB
Slab: 240940 kB
ip6_dst_cache 11980 12064 256 32 2 : tunables 0 0 0
xfrm_dst_cache 0 0 320 25 2 : tunables 0 0 0
ip_dst_cache 3048 4116 192 42 2 : tunables 0 0 0
After Test:
MemFree: 29441816 kB
Slab: 238132 kB
ip6_dst_cache 1902 2432 256 32 2 : tunables 0 0 0
xfrm_dst_cache 0 0 320 25 2 : tunables 0 0 0
ip_dst_cache 3048 4116 192 42 2 : tunables 0 0 0
Tested-by: Andrea Mayer <andrea.mayer@uniroma2.it>
Signed-off-by: Jon Maxwell <jmaxwell37@gmail.com>
Reviewed-by: David Ahern <dsahern@kernel.org>
Link: https://lore.kernel.org/r/20230112012532.311021-1-jmaxwell37@gmail.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
|
|
The details of the iov_iter types are appropriately abstracted, so
there's no need to check for specific type fields. Just let the
abstractions handle it.
This is preparing for io_uring/net's io_send to utilize the more
efficient ITER_UBUF.
Signed-off-by: Keith Busch <kbusch@kernel.org>
Reviewed-by: Jens Axboe <axboe@kernel.dk>
Link: https://lore.kernel.org/r/20230111184245.3784393-1-kbusch@meta.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
|
|
Add 2 tracepoints to monitor the tcp/udp traffic
of per process and per cgroup.
Regarding monitoring the tcp/udp traffic of each process, there are two
existing solutions, the first one is https://www.atoptool.nl/netatop.php.
The second is via kprobe/kretprobe.
Netatop solution is implemented by registering the hook function at the
hook point provided by the netfilter framework.
These hook functions may be in the soft interrupt context and cannot
directly obtain the pid. Some data structures are added to bind packets
and processes. For example, struct taskinfobucket, struct taskinfo ...
Every time the process sends and receives packets it needs multiple
hashmaps,resulting in low performance and it has the problem fo inaccurate
tcp/udp traffic statistics(for example: multiple threads share sockets).
We can obtain the information with kretprobe, but as we know, kprobe gets
the result by trappig in an exception, which loses performance compared
to tracepoint.
We compared the performance of tracepoints with the above two methods, and
the results are as follows:
ab -n 1000000 -c 1000 -r http://127.0.0.1/index.html
without trace:
Time per request: 39.660 [ms] (mean)
Time per request: 0.040 [ms] (mean, across all concurrent requests)
netatop:
Time per request: 50.717 [ms] (mean)
Time per request: 0.051 [ms] (mean, across all concurrent requests)
kr:
Time per request: 43.168 [ms] (mean)
Time per request: 0.043 [ms] (mean, across all concurrent requests)
tracepoint:
Time per request: 41.004 [ms] (mean)
Time per request: 0.041 [ms] (mean, across all concurrent requests
It can be seen that tracepoint has better performance.
Signed-off-by: Yunhui Cui <cuiyunhui@bytedance.com>
Signed-off-by: Xiongchun Duan <duanxiongchun@bytedance.com>
Reviewed-by: Steven Rostedt (Google) <rostedt@goodmis.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
Add the following ethtool tx aggregation parameters:
ETHTOOL_A_COALESCE_TX_AGGR_MAX_BYTES
Maximum size in bytes of a tx aggregated block of frames.
ETHTOOL_A_COALESCE_TX_AGGR_MAX_FRAMES
Maximum number of frames that can be aggregated into a block.
ETHTOOL_A_COALESCE_TX_AGGR_TIME_USECS
Time in usecs after the first packet arrival in an aggregated
block for the block to be sent.
Signed-off-by: Daniele Palmas <dnlplm@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
For PDelay_Resp messages we will likely have a negative value in the
correction field. The switch hardware cannot correctly update such
values (produces an off by one error in the UDP checksum), so it must be
moved to the time stamp field in the tail tag. Format of the correction
field is 48 bit ns + 16 bit fractional ns. After updating the
correction field, clone is no longer required hence it is freed.
Signed-off-by: Christian Eggers <ceggers@arri.de>
Co-developed-by: Arun Ramadoss <arun.ramadoss@microchip.com>
Signed-off-by: Arun Ramadoss <arun.ramadoss@microchip.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
This patch adds the routines for transmission of ptp packets. When the
ptp pdelay_req packet to be transmitted, it uses the deferred xmit
worker to schedule the packets.
During irq_setup, interrupt for Sync, Pdelay_req and Pdelay_rsp are
enabled. So interrupt is triggered for all three packets. But for
p2p1step, we require only time stamp of Pdelay_req packet. Hence to
avoid posting of the completion from ISR routine for Sync and
Pdelay_resp packets, ts_en flag is introduced. This controls which
packets need to processed for timestamp.
After the packet is transmitted, ISR is triggered. The time at which
packet transmitted is recorded to separate register.
This value is reconstructed to absolute time and posted to the user
application through socket error queue.
Signed-off-by: Christian Eggers <ceggers@arri.de>
Co-developed-by: Arun Ramadoss <arun.ramadoss@microchip.com>
Signed-off-by: Arun Ramadoss <arun.ramadoss@microchip.com>
Reviewed-by: Vladimir Oltean <olteanv@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
Rx Timestamping is done through 4 additional bytes in tail tag.
Whenever the ptp packet is received, the 4 byte hardware time stamped
value is added before 1 byte tail tag. Also, bit 7 in tail tag indicates
it as PTP frame. This 4 byte value is extracted from the tail tag and
reconstructed to absolute time and assigned to skb hwtstamp.
If the packet received in PDelay_Resp, then partial ingress timestamp
is subtracted from the correction field. Since user space tools expects
to be done in hardware.
Signed-off-by: Christian Eggers <ceggers@arri.de>
Co-developed-by: Arun Ramadoss <arun.ramadoss@microchip.com>
Signed-off-by: Arun Ramadoss <arun.ramadoss@microchip.com>
Reviewed-by: Vladimir Oltean <olteanv@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
When the PTP is enabled in hardware bit 6 of PTP_MSG_CONF1 register, the
transmit frame needs additional 4 bytes before the tail tag. It is
needed for all the transmission packets irrespective of PTP packets or
not.
The 4-byte timestamp field is 0 for frames other than Pdelay_Resp. For
the one-step Pdelay_Resp, the switch needs the receive timestamp of the
Pdelay_Req message so that it can put the turnaround time in the
correction field.
Since PTP has to be enabled for both Transmission and reception
timestamping, driver needs to track of the tx and rx setting of the all
the user ports in the switch.
Two flags hw_tx_en and hw_rx_en are added in ksz_port to track the
timestampping setting of each port. When any one of ports has tx or rx
timestampping enabled, bit 6 of PTP_MSG_CONF1 is set and it is indicated
to tag_ksz.c through tagger bytes. This flag adds 4 additional bytes to
the tail tag. When tx and rx timestamping of all the ports are disabled,
then 4 bytes are not added.
Tested using hwstamp -i <interface>
Signed-off-by: Arun Ramadoss <arun.ramadoss@microchip.com>
Reviewed-by: Vladimir Oltean <olteanv@gmail.com> # mostly api
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
drivers/net/usb/r8152.c
be53771c87f4 ("r8152: add vendor/device ID pair for Microsoft Devkit")
ec51fbd1b8a2 ("r8152: add USB device driver for config selection")
https://lore.kernel.org/all/20230113113339.658c4723@canb.auug.org.au/
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
|
|
git://git.kernel.org/pub/scm/linux/kernel/git/netdev/net
Pull networking fixes from Paolo Abeni:
"Including fixes from rxrpc.
The rxrpc changes are noticeable large: to address a recent regression
has been necessary completing the threaded refactor.
Current release - regressions:
- rxrpc:
- only disconnect calls in the I/O thread
- move client call connection to the I/O thread
- fix incoming call setup race
- eth: mlx5:
- restore pkt rate policing support
- fix memory leak on updating vport counters
Previous releases - regressions:
- gro: take care of DODGY packets
- ipv6: deduct extension header length in rawv6_push_pending_frames
- tipc: fix unexpected link reset due to discovery messages
Previous releases - always broken:
- sched: disallow noqueue for qdisc classes
- eth: ice: fix potential memory leak in ice_gnss_tty_write()
- eth: ixgbe: fix pci device refcount leak
- eth: mlx5:
- fix command stats access after free
- fix macsec possible null dereference when updating MAC security
entity (SecY)"
* tag 'net-6.2-rc4' of git://git.kernel.org/pub/scm/linux/kernel/git/netdev/net: (64 commits)
r8152: add vendor/device ID pair for Microsoft Devkit
net: stmmac: add aux timestamps fifo clearance wait
bnxt: make sure we return pages to the pool
net: hns3: fix wrong use of rss size during VF rss config
ipv6: raw: Deduct extension header length in rawv6_push_pending_frames
net: lan966x: check for ptp to be enabled in lan966x_ptp_deinit()
net: sched: disallow noqueue for qdisc classes
iavf/iavf_main: actually log ->src mask when talking about it
igc: Fix PPS delta between two synchronized end-points
ixgbe: fix pci device refcount leak
octeontx2-pf: Fix resource leakage in VF driver unbind
selftests/net: l2_tos_ttl_inherit.sh: Ensure environment cleanup on failure.
selftests/net: l2_tos_ttl_inherit.sh: Run tests in their own netns.
selftests/net: l2_tos_ttl_inherit.sh: Set IPv6 addresses with "nodad".
net/mlx5e: Fix macsec possible null dereference when updating MAC security entity (SecY)
net/mlx5e: Fix macsec ssci attribute handling in offload path
net/mlx5: E-switch, Coverity: overlapping copy
net/mlx5e: Don't support encap rules with gbp option
net/mlx5: Fix ptp max frequency adjustment range
net/mlx5e: Fix memory leak on updating vport counters
...
|
|
git://git.kernel.org/pub/scm/linux/kernel/git/xen/tip
Pull xen fixes from Juergen Gross:
- two cleanup patches
- a fix of a memory leak in the Xen pvfront driver
- a fix of a locking issue in the Xen hypervisor console driver
* tag 'for-linus-6.2-rc4-tag' of git://git.kernel.org/pub/scm/linux/kernel/git/xen/tip:
xen/pvcalls: free active map buffer on pvcalls_front_free_map
hvc/xen: lock console list traversal
x86/xen: Remove the unused function p2m_index()
xen: make remove callback of xen driver void returned
|
|
This removes behaviour, where error code returned from any transport
was always switched to ENOMEM. For example when user tries to send too
big message via SEQPACKET socket, transport layers return EMSGSIZE, but
this error code was always replaced with ENOMEM and returned to user.
Signed-off-by: Bobby Eshleman <bobby.eshleman@bytedance.com>
Signed-off-by: Arseniy Krasnov <AVKrasnov@sberdevices.ru>
Reviewed-by: Stefano Garzarella <sgarzare@redhat.com>
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
|
|
The reference needs to keep the instance memory around, but also
the instance lock must remain valid. Users will take the lock,
check registration status and release the lock. mutex_destroy()
etc. belong in the same place as the freeing of the memory.
Unfortunately lockdep_unregister_key() sleeps so we need
to switch the an rcu_work.
Note that the problem is a bit hard to repro, because
devlink_pernet_pre_exit() iterates over registered instances.
AFAIU the instances must get devlink_free()d concurrently with
the namespace getting deleted for the problem to occur.
Reported-by: syzbot+d94d214ea473e218fc89@syzkaller.appspotmail.com
Reported-by: syzbot+9f0dd863b87113935acf@syzkaller.appspotmail.com
Fixes: 9053637e0da7 ("devlink: remove the registration guarantee of references")
Reviewed-by: Jiri Pirko <jiri@nvidia.com>
Reviewed-by: Jacob Keller <jacob.e.keller@intel.com>
Link: https://lore.kernel.org/r/20230111042908.988199-1-kuba@kernel.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
|
|
The total cork length created by ip6_append_data includes extension
headers, so we must exclude them when comparing them against the
IPV6_CHECKSUM offset which does not include extension headers.
Reported-by: Kyle Zeng <zengyhkyle@gmail.com>
Fixes: 357b40a18b04 ("[IPV6]: IPV6_CHECKSUM socket option can corrupt kernel memory")
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
This patch adds the link modes for the IEEE 802.3cg Clause 147 10BASE-T1S
Ethernet PHY. According to the specifications, the 10BASE-T1S supports
Point-To-Point Full-Duplex, Point-To-Point Half-Duplex and/or
Point-To-Multipoint (AKA Multi-Drop) Half-Duplex operations.
Signed-off-by: Piergiorgio Beruto <piergiorgio.beruto@gmail.com>
Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
Add support for configuring the PLCA Reconciliation Sublayer on
multi-drop PHYs that support IEEE802.3cg-2019 Clause 148 (e.g.,
10BASE-T1S). This patch adds the appropriate netlink interface
to ethtool.
Signed-off-by: Piergiorgio Beruto <piergiorgio.beruto@gmail.com>
Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
While experimenting with applying noqueue to a classful queue discipline,
we discovered a NULL pointer dereference in the __dev_queue_xmit()
path that generates a kernel OOPS:
# dev=enp0s5
# tc qdisc replace dev $dev root handle 1: htb default 1
# tc class add dev $dev parent 1: classid 1:1 htb rate 10mbit
# tc qdisc add dev $dev parent 1:1 handle 10: noqueue
# ping -I $dev -w 1 -c 1 1.1.1.1
[ 2.172856] BUG: kernel NULL pointer dereference, address: 0000000000000000
[ 2.173217] #PF: supervisor instruction fetch in kernel mode
...
[ 2.178451] Call Trace:
[ 2.178577] <TASK>
[ 2.178686] htb_enqueue+0x1c8/0x370
[ 2.178880] dev_qdisc_enqueue+0x15/0x90
[ 2.179093] __dev_queue_xmit+0x798/0xd00
[ 2.179305] ? _raw_write_lock_bh+0xe/0x30
[ 2.179522] ? __local_bh_enable_ip+0x32/0x70
[ 2.179759] ? ___neigh_create+0x610/0x840
[ 2.179968] ? eth_header+0x21/0xc0
[ 2.180144] ip_finish_output2+0x15e/0x4f0
[ 2.180348] ? dst_output+0x30/0x30
[ 2.180525] ip_push_pending_frames+0x9d/0xb0
[ 2.180739] raw_sendmsg+0x601/0xcb0
[ 2.180916] ? _raw_spin_trylock+0xe/0x50
[ 2.181112] ? _raw_spin_unlock_irqrestore+0x16/0x30
[ 2.181354] ? get_page_from_freelist+0xcd6/0xdf0
[ 2.181594] ? sock_sendmsg+0x56/0x60
[ 2.181781] sock_sendmsg+0x56/0x60
[ 2.181958] __sys_sendto+0xf7/0x160
[ 2.182139] ? handle_mm_fault+0x6e/0x1d0
[ 2.182366] ? do_user_addr_fault+0x1e1/0x660
[ 2.182627] __x64_sys_sendto+0x1b/0x30
[ 2.182881] do_syscall_64+0x38/0x90
[ 2.183085] entry_SYSCALL_64_after_hwframe+0x63/0xcd
...
[ 2.187402] </TASK>
Previously in commit d66d6c3152e8 ("net: sched: register noqueue
qdisc"), NULL was set for the noqueue discipline on noqueue init
so that __dev_queue_xmit() falls through for the noqueue case. This
also sets a bypass of the enqueue NULL check in the
register_qdisc() function for the struct noqueue_disc_ops.
Classful queue disciplines make it past the NULL check in
__dev_queue_xmit() because the discipline is set to htb (in this case),
and then in the call to __dev_xmit_skb(), it calls into htb_enqueue()
which grabs a leaf node for a class and then calls qdisc_enqueue() by
passing in a queue discipline which assumes ->enqueue() is not set to NULL.
Fix this by not allowing classes to be assigned to the noqueue
discipline. Linux TC Notes states that classes cannot be set to
the noqueue discipline. [1] Let's enforce that here.
Links:
1. https://linux-tc-notes.sourceforge.net/tc/doc/sch_noqueue.txt
Fixes: d66d6c3152e8 ("net: sched: register noqueue qdisc")
Cc: stable@vger.kernel.org
Signed-off-by: Frederick Lawler <fred@cloudflare.com>
Reviewed-by: Jakub Sitnicki <jakub@cloudflare.com>
Link: https://lore.kernel.org/r/20230109163906.706000-1-fred@cloudflare.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
|
|
git://git.kernel.org/pub/scm/linux/kernel/git/cel/linux
Pull nfsd fixes from Chuck Lever:
- Fix a race when creating NFSv4 files
- Revert the use of relaxed bitops
* tag 'nfsd-6.2-3' of git://git.kernel.org/pub/scm/linux/kernel/git/cel/linux:
NFSD: Use set_bit(RQ_DROPME)
Revert "SUNRPC: Use RMW bitops in single-threaded hot paths"
nfsd: fix handling of cached open files in nfsd4_open codepath
|
|
The 'TCA_MPLS_LABEL' attribute is of 'NLA_U32' type, but has a
validation type of 'NLA_VALIDATE_FUNCTION'. This is an invalid
combination according to the comment above 'struct nla_policy':
"
Meaning of `validate' field, use via NLA_POLICY_VALIDATE_FN:
NLA_BINARY Validation function called for the attribute.
All other Unused - but note that it's a union
"
This can trigger the warning [1] in nla_get_range_unsigned() when
validation of the attribute fails. Despite being of 'NLA_U32' type, the
associated 'min'/'max' fields in the policy are negative as they are
aliased by the 'validate' field.
Fix by changing the attribute type to 'NLA_BINARY' which is consistent
with the above comment and all other users of NLA_POLICY_VALIDATE_FN().
As a result, move the length validation to the validation function.
No regressions in MPLS tests:
# ./tdc.py -f tc-tests/actions/mpls.json
[...]
# echo $?
0
[1]
WARNING: CPU: 0 PID: 17743 at lib/nlattr.c:118
nla_get_range_unsigned+0x1d8/0x1e0 lib/nlattr.c:117
Modules linked in:
CPU: 0 PID: 17743 Comm: syz-executor.0 Not tainted 6.1.0-rc8 #3
Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS
rel-1.13.0-48-gd9c812dda519-prebuilt.qemu.org 04/01/2014
RIP: 0010:nla_get_range_unsigned+0x1d8/0x1e0 lib/nlattr.c:117
[...]
Call Trace:
<TASK>
__netlink_policy_dump_write_attr+0x23d/0x990 net/netlink/policy.c:310
netlink_policy_dump_write_attr+0x22/0x30 net/netlink/policy.c:411
netlink_ack_tlv_fill net/netlink/af_netlink.c:2454 [inline]
netlink_ack+0x546/0x760 net/netlink/af_netlink.c:2506
netlink_rcv_skb+0x1b7/0x240 net/netlink/af_netlink.c:2546
rtnetlink_rcv+0x18/0x20 net/core/rtnetlink.c:6109
netlink_unicast_kernel net/netlink/af_netlink.c:1319 [inline]
netlink_unicast+0x5e9/0x6b0 net/netlink/af_netlink.c:1345
netlink_sendmsg+0x739/0x860 net/netlink/af_netlink.c:1921
sock_sendmsg_nosec net/socket.c:714 [inline]
sock_sendmsg net/socket.c:734 [inline]
____sys_sendmsg+0x38f/0x500 net/socket.c:2482
___sys_sendmsg net/socket.c:2536 [inline]
__sys_sendmsg+0x197/0x230 net/socket.c:2565
__do_sys_sendmsg net/socket.c:2574 [inline]
__se_sys_sendmsg net/socket.c:2572 [inline]
__x64_sys_sendmsg+0x42/0x50 net/socket.c:2572
do_syscall_x64 arch/x86/entry/common.c:50 [inline]
do_syscall_64+0x2b/0x70 arch/x86/entry/common.c:80
entry_SYSCALL_64_after_hwframe+0x63/0xcd
Link: https://lore.kernel.org/netdev/CAO4mrfdmjvRUNbDyP0R03_DrD_eFCLCguz6OxZ2TYRSv0K9gxA@mail.gmail.com/
Fixes: 2a2ea50870ba ("net: sched: add mpls manipulation actions to TC")
Reported-by: Wei Chen <harperchen1110@gmail.com>
Tested-by: Wei Chen <harperchen1110@gmail.com>
Signed-off-by: Ido Schimmel <idosch@nvidia.com>
Reviewed-by: Alexander Duyck <alexanderduyck@fb.com>
Link: https://lore.kernel.org/r/20230107171004.608436-1-idosch@nvidia.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
|
|
Since commit ce098da1497c ("skbuff: Introduce slab_build_skb()")
drivers trying to build skb around slab-backed buffers should
go via slab_build_skb() rather than passing frag_size = 0 to
the main build_skb().
Remove the copy'n'pasted comments about 0 meaning slab.
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
Jaroslav reported a recent throughput regression with virtio_net
caused by blamed commit.
It is unclear if DODGY GSO packets coming from user space
can be accepted by GRO engine in the future with minimal
changes, and if there is any expected gain from it.
In the meantime, make sure to detect and flush DODGY packets.
Fixes: 5eddb24901ee ("gro: add support of (hw)gro packets to gro stack")
Signed-off-by: Eric Dumazet <edumazet@google.com>
Reported-and-bisected-by: Jaroslav Pulchart <jaroslav.pulchart@gooddata.com>
Cc: Coco Li <lixiaoyan@google.com>
Cc: Paolo Abeni <pabeni@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
Do the statistics of mptcp socket in use with sock_prot_inuse_add().
Therefore, we can get the count of used mptcp socket from
/proc/net/protocols:
& cat /proc/net/protocols
protocol size sockets memory press maxhdr slab module cl co di ac io in de sh ss gs se re sp bi br ha uh gp em
MPTCPv6 2048 0 0 no 0 yes kernel y n y y y y y y y y y y n n n y y y n
MPTCP 1896 1 0 no 0 yes kernel y n y y y y y y y y y y n n n y y y n
Acked-by: Paolo Abeni <pabeni@redhat.com>
Signed-off-by: Menglong Dong <imagedong@tencent.com>
Signed-off-by: Mat Martineau <mathew.j.martineau@linux.intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
'ssk' should be more appropriate to be the name of the first argument
in mptcp_token_new_connect().
Acked-by: Paolo Abeni <pabeni@redhat.com>
Signed-off-by: Menglong Dong <imagedong@tencent.com>
Signed-off-by: Mat Martineau <mathew.j.martineau@linux.intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
The 'sk_prot' field in token KUNIT self-tests will be dereferenced in
mptcp_token_new_connect(). Therefore, init it with tcp_prot.
Acked-by: Paolo Abeni <pabeni@redhat.com>
Signed-off-by: Menglong Dong <imagedong@tencent.com>
Signed-off-by: Mat Martineau <mathew.j.martineau@linux.intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
'sock->sk' is used frequently in mptcp_listen(). Therefore, we can
introduce the 'sk' and replace 'sock->sk' with it.
Acked-by: Paolo Abeni <pabeni@redhat.com>
Signed-off-by: Menglong Dong <imagedong@tencent.com>
Signed-off-by: Mat Martineau <mathew.j.martineau@linux.intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
The local variable 'ssk' has been defined at the beginning of the function
mptcp_write_options(), use it instead of getting 'ssk' again.
Reviewed-by: Mat Martineau <mathew.j.martineau@linux.intel.com>
Signed-off-by: Geliang Tang <geliang.tang@suse.com>
Signed-off-by: Mat Martineau <mathew.j.martineau@linux.intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
Use the local variable 'net' instead of sock_net() in the functions where
the variable 'struct net *net' has been defined.
Reviewed-by: Mat Martineau <mathew.j.martineau@linux.intel.com>
Signed-off-by: Geliang Tang <geliang.tang@suse.com>
Signed-off-by: Mat Martineau <mathew.j.martineau@linux.intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
The helper msk_owned_by_me() is defined in protocol.h, so use it instead
of sock_owned_by_me().
Reviewed-by: Mat Martineau <mathew.j.martineau@linux.intel.com>
Signed-off-by: Geliang Tang <geliang.tang@suse.com>
Signed-off-by: Mat Martineau <mathew.j.martineau@linux.intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
git://git.kernel.org/pub/scm/linux/kernel/git/dhowells/linux-fs
David Howells says:
====================
rxrpc: Fix race between call connection, data transmit and call disconnect
Here are patches to fix an oops[1] caused by a race between call
connection, initial packet transmission and call disconnection which
results in something like:
kernel BUG at net/rxrpc/peer_object.c:413!
when the syzbot test is run. The problem is that the connection procedure
is effectively split across two threads and can get expanded by taking an
interrupt, thereby adding the call to the peer error distribution list
*after* it has been disconnected (say by the rxrpc socket shutting down).
The easiest solution is to look at the fourth set of I/O thread
conversion/SACK table expansion patches that didn't get applied[2] and take
from it those patches that move call connection and disconnection into the
I/O thread. Moving these things into the I/O thread means that the
sequencing is managed by all being done in the same thread - and the race
can no longer happen.
This is preferable to introducing an extra lock as adding an extra lock
would make the I/O thread have to wait for the app thread in yet another
place.
The changes can be considered as a number of logical parts:
(1) Move all of the call state changes into the I/O thread.
(2) Make client connection ID space per-local endpoint so that the I/O
thread doesn't need locks to access it.
(3) Move actual abort generation into the I/O thread and clean it up. If
sendmsg or recvmsg want to cause an abort, they have to delegate it.
(4) Offload the setting up of the security context on a connection to the
thread of one of the apps that's starting a call. We don't want to be
doing any sort of crypto in the I/O thread.
(5) Connect calls (ie. assign them to channel slots on connections) in the
I/O thread. Calls are set up by sendmsg/kafs and passed to the I/O
thread to connect. Connections are allocated in the I/O thread after
this.
(6) Disconnect calls in the I/O thread.
I've also added a patch for an unrelated bug that cropped up during
testing, whereby a race can occur between an incoming call and socket
shutdown.
Note that whilst this fixes the original syzbot bug, another bug may get
triggered if this one is fixed:
INFO: rcu detected stall in corrupted
rcu: INFO: rcu_preempt detected expedited stalls on CPUs/tasks: { P5792 } 2657 jiffies s: 2825 root: 0x0/T
rcu: blocking rcu_node structures (internal RCU debug):
It doesn't look this should be anything to do with rxrpc, though, as I've
tested an additional patch[3] that removes practically all the RCU usage
from rxrpc and it still occurs. It seems likely that it is being caused by
something in the tunnelling setup that the syzbot test does, but there's
not enough info to go on. It also seems unlikely to be anything to do with
the afs driver as the test doesn't use that.
====================
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
Pull NFS client fixes from Trond Myklebust:
- Fix a race in the RPCSEC_GSS upcall code that causes hung RPC calls
- Fix a broken coalescing test in the pNFS file layout driver
- Ensure that the access cache rcu path also applies the login test
- Fix up for a sparse warning
* tag 'nfs-for-6.2-2' of git://git.linux-nfs.org/projects/trondmy/linux-nfs:
NFS: Fix up a sparse warning
NFS: Judge the file access cache's timestamp in rcu path
pNFS/filelayout: Fix coalescing test for single DS
SUNRPC: ensure the matching upcall is in-flight upon downcall
|
|
An incoming call can race with rxrpc socket destruction, leading to a
leaked call. This may result in an oops when the call timer eventually
expires:
BUG: kernel NULL pointer dereference, address: 0000000000000874
RIP: 0010:_raw_spin_lock_irqsave+0x2a/0x50
Call Trace:
<IRQ>
try_to_wake_up+0x59/0x550
? __local_bh_enable_ip+0x37/0x80
? rxrpc_poke_call+0x52/0x110 [rxrpc]
? rxrpc_poke_call+0x110/0x110 [rxrpc]
? rxrpc_poke_call+0x110/0x110 [rxrpc]
call_timer_fn+0x24/0x120
with a warning in the kernel log looking something like:
rxrpc: Call 00000000ba5e571a still in use (1,SvAwtACK,1061d,0)!
incurred during rmmod of rxrpc. The 1061d is the call flags:
RECVMSG_READ_ALL, RX_HEARD, BEGAN_RX_TIMER, RX_LAST, EXPOSED,
IS_SERVICE, RELEASED
but no DISCONNECTED flag (0x800), so it's an incoming (service) call and
it's still connected.
The race appears to be that:
(1) rxrpc_new_incoming_call() consults the service struct, checks sk_state
and allocates a call - then pauses, possibly for an interrupt.
(2) rxrpc_release_sock() sets RXRPC_CLOSE, nulls the service pointer,
discards the prealloc and releases all calls attached to the socket.
(3) rxrpc_new_incoming_call() resumes, launching the new call, including
its timer and attaching it to the socket.
Fix this by read-locking local->services_lock to access the AF_RXRPC socket
providing the service rather than RCU in rxrpc_new_incoming_call().
There's no real need to use RCU here as local->services_lock is only
write-locked by the socket side in two places: when binding and when
shutting down.
Fixes: 5e6ef4f1017c ("rxrpc: Make the I/O thread take over the call and local processor work")
Reported-by: Marc Dionne <marc.dionne@auristor.com>
Signed-off-by: David Howells <dhowells@redhat.com>
cc: linux-afs@lists.infradead.org
|
|
Zero-length arrays are deprecated[1]. Replace struct ipv6_rpl_sr_hdr's
"segments" union of 0-length arrays with flexible arrays. Detected with
GCC 13, using -fstrict-flex-arrays=3:
In function 'rpl_validate_srh',
inlined from 'rpl_build_state' at ../net/ipv6/rpl_iptunnel.c:96:7:
../net/ipv6/rpl_iptunnel.c:60:28: warning: array subscript <unknown> is outside array bounds of 'struct in6_addr[0]' [-Warray-bounds=]
60 | if (ipv6_addr_type(&srh->rpl_segaddr[srh->segments_left - 1]) &
| ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
In file included from ../include/net/rpl.h:12,
from ../net/ipv6/rpl_iptunnel.c:13:
../include/uapi/linux/rpl.h: In function 'rpl_build_state':
../include/uapi/linux/rpl.h:40:33: note: while referencing 'addr'
40 | struct in6_addr addr[0];
| ^~~~
[1] https://www.kernel.org/doc/html/latest/process/deprecated.html#zero-length-and-one-element-arrays
Cc: Hideaki YOSHIFUJI <yoshfuji@linux-ipv6.org>
Signed-off-by: Kees Cook <keescook@chromium.org>
Reviewed-by: Gustavo A. R. Silva <gustavoars@kernel.org>
Reviewed-by: David Ahern <dsahern@kernel.org>
Link: https://lore.kernel.org/r/20230105221533.never.711-kees@kernel.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
|
|
The premise that "Once an svc thread is scheduled and executing an
RPC, no other processes will touch svc_rqst::rq_flags" is false.
svc_xprt_enqueue() examines the RQ_BUSY flag in scheduled nfsd
threads when determining which thread to wake up next.
Found via KCSAN.
Fixes: 28df0988815f ("SUNRPC: Use RMW bitops in single-threaded hot paths")
Reviewed-by: Jeff Layton <jlayton@kernel.org>
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
|
|
It's most natural to register the instance first and then its
subobjects. Now that we can use the instance lock to protect
the atomicity of all init - it should also be safe.
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
Requiring devlink_set_features() to be run before devlink is
registered is overzealous. devlink_set_features() itself is
a leftover from old workarounds which were trying to prevent
initiating reload before probe was complete.
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Reviewed-by: Jiri Pirko <jiri@nvidia.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
The objective of exposing the devlink instance locks to
drivers was to let them use these locks to prevent user space
from accessing the device before it's fully initialized.
This is difficult because devlink_unregister() waits for all
references to be released, meaning that devlink_unregister()
can't itself be called under the instance lock.
To avoid this issue devlink_register() was moved after subobject
registration a while ago. Unfortunately the netdev paths get
a hold of the devlink instances _before_ they are registered.
Ideally netdev should wait for devlink init to finish (synchronizing
on the instance lock). This can't work because we don't know if the
instance will _ever_ be registered (in case of failures it may not).
The other option of returning an error until devlink_register()
is called is unappealing (user space would get a notification
netdev exist but would have to wait arbitrary amount of time
before accessing some of its attributes).
Weaken the guarantees of the devlink references.
Holding a reference will now only guarantee that the memory
of the object is around. Another way of looking at it is that
the reference now protects the object not its "registered" status.
Use devlink instance lock to synchronize unregistration.
This implies that releasing of the "main" reference of the devlink
instance moves from devlink_unregister() to devlink_free().
Reviewed-by: Jacob Keller <jacob.e.keller@intel.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Reviewed-by: Jiri Pirko <jiri@nvidia.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
Always check under the instance lock whether the devlink instance
is still / already registered.
This is a no-op for the most part, as the unregistration path currently
waits for all references. On the init path, however, we may temporarily
open up a race with netdev code, if netdevs are registered before the
devlink instance. This is temporary, the next change fixes it, and this
commit has been split out for the ease of review.
Note that in case of iterating over sub-objects which have their
own lock (regions and line cards) we assume an implicit dependency
between those objects existing and devlink unregistration.
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Reviewed-by: Jiri Pirko <jiri@nvidia.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
devlink->dev is assumed to be always valid as long as any
outstanding reference to the devlink instance exists.
In prep for weakening of the references take the instance lock.
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Reviewed-by: Jiri Pirko <jiri@nvidia.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
devlink_pernet_pre_exit() is the only obvious place which takes
the instance lock without using the devl_ helpers. Update the code
and move the error print after releasing the reference
(having unlock and put together feels slightly idiomatic).
Reviewed-by: Jiri Pirko <jiri@nvidia.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
xa_find_after() is designed to handle multi-index entries correctly.
If a xarray has two entries one which spans indexes 0-3 and one at
index 4 xa_find_after(0) will return the entry at index 4.
Having to juggle the two callbacks, however, is unnecessary in case
of the devlink xarray, as there is 1:1 relationship with indexes.
Always use xa_find() and increment the index manually.
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Reviewed-by: Jiri Pirko <jiri@nvidia.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
This unexpected behavior is observed:
node 1 | node 2
------ | ------
link is established | link is established
reboot | link is reset
up | send discovery message
receive discovery message |
link is established | link is established
send discovery message |
| receive discovery message
| link is reset (unexpected)
| send reset message
link is reset |
It is due to delayed re-discovery as described in function
tipc_node_check_dest(): "this link endpoint has already reset
and re-established contact with the peer, before receiving a
discovery message from that node."
However, commit 598411d70f85 has changed the condition for calling
tipc_node_link_down() which was the acceptance of new media address.
This commit fixes this by restoring the old and correct behavior.
Fixes: 598411d70f85 ("tipc: make resetting of links non-atomic")
Acked-by: Jon Maloy <jmaloy@redhat.com>
Signed-off-by: Tung Nguyen <tung.q.nguyen@dektech.com.au>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
All were not visible to the non-priv users inside netns. However,
with 4ecb90090c84 ("sysctl: allow override of /proc/sys/net with
CAP_NET_ADMIN"), these vars are protected from getting modified.
A proc with capable(CAP_NET_ADMIN) can change the values so
not having them visible inside netns is just causing nuisance to
process that check certain values (e.g. net.core.somaxconn) and
see different behavior in root-netns vs. other-netns
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>
Cc: Soheil Hassas Yeganeh <soheil@google.com>
Signed-off-by: Mahesh Bandewar <maheshb@google.com>
Acked-by: Soheil Hassas Yeganeh <soheil@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
Move the connection setup of client calls to the I/O thread so that a whole
load of locking and barrierage can be eliminated. This necessitates the
app thread waiting for connection to complete before it can begin
encrypting data.
This also completes the fix for a race that exists between call connection
and call disconnection whereby the data transmission code adds the call to
the peer error distribution list after the call has been disconnected (say
by the rxrpc socket getting closed).
The fix is to complete the process of moving call connection, data
transmission and call disconnection into the I/O thread and thus forcibly
serialising them.
Note that the issue may predate the overhaul to an I/O thread model that
were included in the merge window for v6.2, but the timing is very much
changed by the change given below.
Fixes: cf37b5987508 ("rxrpc: Move DATA transmission into call processor work item")
Reported-by: syzbot+c22650d2844392afdcfd@syzkaller.appspotmail.com
Signed-off-by: David Howells <dhowells@redhat.com>
cc: Marc Dionne <marc.dionne@auristor.com>
cc: linux-afs@lists.infradead.org
|
|
Move the management of the client connection cache to the I/O thread rather
than managing it from the namespace as an aggregate across all the local
endpoints within the namespace.
This will allow a load of locking to be got rid of in a future patch as
only the I/O thread will be looking at the this.
The downside is that the total number of cached connections on the system
can get higher because the limit is now per-local rather than per-netns.
We can, however, keep the number of client conns in use across the entire
netfs and use that to reduce the expiration time of idle connection.
Signed-off-by: David Howells <dhowells@redhat.com>
cc: Marc Dionne <marc.dionne@auristor.com>
cc: linux-afs@lists.infradead.org
|
|
All the setters of call->state are now in the I/O thread and thus the state
lock is now unnecessary.
Signed-off-by: David Howells <dhowells@redhat.com>
cc: Marc Dionne <marc.dionne@auristor.com>
cc: linux-afs@lists.infradead.org
|
|
Move the call state changes that are made in rxrpc_recvmsg() to the I/O
thread. This means that, thenceforth, only the I/O thread does this and
the call state lock can be removed.
This requires the Rx phase to be ended when the last packet is received,
not when it is processed.
Since this now changes the rxrpc call state to SUCCEEDED before we've
consumed all the data from it, rxrpc_kernel_check_life() mustn't say the
call is dead until the recvmsg queue is empty (unless the call has failed).
Signed-off-by: David Howells <dhowells@redhat.com>
cc: Marc Dionne <marc.dionne@auristor.com>
cc: linux-afs@lists.infradead.org
|
|
Move all the call state changes that are made in rxrpc_sendmsg() to the I/O
thread. This is a step towards removing the call state lock.
This requires the switch to the RXRPC_CALL_CLIENT_AWAIT_REPLY and
RXRPC_CALL_SERVER_SEND_REPLY states to be done when the last packet is
decanted from ->tx_sendmsg to ->tx_buffer in the I/O thread, not when it is
added to ->tx_sendmsg by sendmsg().
Signed-off-by: David Howells <dhowells@redhat.com>
cc: Marc Dionne <marc.dionne@auristor.com>
cc: linux-afs@lists.infradead.org
|