openwrt/staging/blogic.git
6 years agoneighbour: support for NTF_EXT_LEARNED flag
Roopa Prabhu [Tue, 24 Apr 2018 20:49:34 +0000 (13:49 -0700)]
neighbour: support for NTF_EXT_LEARNED flag

This patch extends NTF_EXT_LEARNED support to the neighbour system.
Example use-case: An Ethernet VPN implementation (eg in FRR routing suite)
can use this flag to add dynamic reachable external neigh entires
learned via control plane. The use of neigh NTF_EXT_LEARNED in this
patch is consistent with its use with bridge and vxlan fdb entries.

Signed-off-by: Roopa Prabhu <roopa@cumulusnetworks.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agoipv6: addrconf: don't evaluate keep_addr_on_down twice
Ivan Vecera [Tue, 24 Apr 2018 17:51:29 +0000 (19:51 +0200)]
ipv6: addrconf: don't evaluate keep_addr_on_down twice

The addrconf_ifdown() evaluates keep_addr_on_down state twice. There
is no need to do it.

Cc: David Ahern <dsahern@gmail.com>
Signed-off-by: Ivan Vecera <cera@cera.cz>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agoipv6: sr: Compute flowlabel for outer IPv6 header of seg6 encap mode
Ahmed Abdelsalam [Tue, 24 Apr 2018 18:23:16 +0000 (20:23 +0200)]
ipv6: sr: Compute flowlabel for outer IPv6 header of seg6 encap mode

ECMP (equal-cost multipath) hashes are typically computed on the packets'
5-tuple(src IP, dst IP, src port, dst port, L4 proto).

For encapsulated packets, the L4 data is not readily available and ECMP
hashing will often revert to (src IP, dst IP). This will lead to traffic
polarization on a single ECMP path, causing congestion and waste of network
capacity.

In IPv6, the 20-bit flow label field is also used as part of the ECMP hash.
In the lack of L4 data, the hashing will be on (src IP, dst IP, flow
label). Having a non-zero flow label is thus important for proper traffic
load balancing when L4 data is unavailable (i.e., when packets are
encapsulated).

Currently, the seg6_do_srh_encap() function extracts the original packet's
flow label and set it as the outer IPv6 flow label. There are two issues
with this behaviour:

a) There is no guarantee that the inner flow label is set by the source.
b) If the original packet is not IPv6, the flow label will be set to
zero (e.g., IPv4 or L2 encap).

This patch adds a function, named seg6_make_flowlabel(), that computes a
flow label from a given skb. It supports IPv6, IPv4 and L2 payloads, and
leverages the per namespace 'seg6_flowlabel" sysctl value.

The currently support behaviours are as follows:
-1 set flowlabel to zero.
0 copy flowlabel from Inner paceket in case of Inner IPv6
(Set flowlabel to 0 in case IPv4/L2)
1 Compute the flowlabel using seg6_make_flowlabel()

This patch has been tested for IPv6, IPv4, and L2 traffic.

Signed-off-by: Ahmed Abdelsalam <amsalam20@gmail.com>
Acked-by: David Lebrun <dlebrun@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agoMerge git://git.kernel.org/pub/scm/linux/kernel/git/davem/net
David S. Miller [Wed, 25 Apr 2018 03:59:11 +0000 (23:59 -0400)]
Merge git://git./linux/kernel/git/davem/net

6 years agoMerge branch 'userns-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/ebiederm...
Linus Torvalds [Wed, 25 Apr 2018 00:58:51 +0000 (17:58 -0700)]
Merge branch 'userns-linus' of git://git./linux/kernel/git/ebiederm/user-namespace

Pull userns bug fix from Eric Biederman:
 "Just a small fix to properly set the return code on error"

* 'userns-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/ebiederm/user-namespace:
  commoncap: Handle memory allocation failure.

6 years agoMerge git://git.kernel.org/pub/scm/linux/kernel/git/davem/net
Linus Torvalds [Tue, 24 Apr 2018 21:16:40 +0000 (14:16 -0700)]
Merge git://git./linux/kernel/git/davem/net

Pull networking fixes from David Miller:

 1) Fix rtnl deadlock in ipvs, from Julian Anastasov.

 2) s390 qeth fixes from Julian Wiedmann (control IO completion stalls,
    bad MAC address update sequence, request side races on command IO
    timeouts).

 3) Handle seq_file overflow properly in l2tp, from Guillaume Nault.

 4) Fix VLAN priority mappings in cpsw driver, from Ivan Khoronzhuk.

 5) Packet scheduler ife action fixes (malformed TLV lengths, etc.) from
    Alexander Aring.

 6) Fix out of bounds access in tcp md5 option parser, from Jann Horn.

 7) Missing netlink attribute policies in rtm_ipv6_policy table, from
    Eric Dumazet.

 8) Missing socket address length checks in l2tp and pppoe connect, from
    Guillaume Nault.

 9) Fix netconsole over team and bonding, from Xin Long.

10) Fix race with AF_PACKET socket state bitfields, from Willem de
    Bruijn.

* git://git.kernel.org/pub/scm/linux/kernel/git/davem/net: (51 commits)
  ice: Fix insufficient memory issue in ice_aq_manage_mac_read
  sfc: ARFS filter IDs
  net: ethtool: Add missing kernel doc for FEC parameters
  packet: fix bitfield update race
  ice: Do not check INTEVENT bit for OICR interrupts
  ice: Fix incorrect comment for action type
  ice: Fix initialization for num_nodes_added
  igb: Fix the transmission mode of queue 0 for Qav mode
  ixgbevf: ensure xdp_ring resources are free'd on error exit
  team: fix netconsole setup over team
  amd-xgbe: Only use the SFP supported transceiver signals
  amd-xgbe: Improve KR auto-negotiation and training
  amd-xgbe: Add pre/post auto-negotiation phy hooks
  pppoe: check sockaddr length in pppoe_connect()
  l2tp: check sockaddr length in pppol2tp_connect()
  net: phy: marvell: clear wol event before setting it
  ipv6: add RTA_TABLE and RTA_PREFSRC to rtm_ipv6_policy
  bonding: do not set slave_dev npinfo before slave_enable_netpoll in bond_enslave
  tcp: don't read out-of-bounds opsize
  ibmvnic: Clean actual number of RX or TX pools
  ...

6 years agoliquidio: Swap VF representor Tx and Rx statistics
Srinivas Jampala [Tue, 24 Apr 2018 17:23:27 +0000 (10:23 -0700)]
liquidio: Swap VF representor Tx and Rx statistics

Swap VF representor tx and rx interface statistics since it is a
virtual switchdev port and tx for VM should be rx for VF representor
and vice-versa.

Signed-off-by: Srinivas Jampala <srinivasa.jampala@cavium.com>
Acked-by: Derek Chickles <derek.chickles@cavium.com>
Signed-off-by: Felix Manlunas <felix.manlunas@cavium.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet/ipv6: fix LOCKDEP issue in rt6_remove_exception_rt()
Eric Dumazet [Tue, 24 Apr 2018 16:22:49 +0000 (09:22 -0700)]
net/ipv6: fix LOCKDEP issue in rt6_remove_exception_rt()

rt6_remove_exception_rt() is called under rcu_read_lock() only.

We lock rt6_exception_lock a bit later, so we do not hold
rt6_exception_lock yet.

Fixes: 8a14e46f1402 ("net/ipv6: Fix missing rcu dereferences on from")
Signed-off-by: Eric Dumazet <edumazet@google.com>
Reported-by: syzbot <syzkaller@googlegroups.com>
Cc: David Ahern <dsahern@gmail.com>
Acked-by: David Ahern <dsahern@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agoMerge branch '1GbE' of git://git.kernel.org/pub/scm/linux/kernel/git/jkirsher/net...
David S. Miller [Tue, 24 Apr 2018 20:17:59 +0000 (16:17 -0400)]
Merge branch '1GbE' of git://git./linux/kernel/git/jkirsher/net-queue

Jeff Kirsher says:

====================
Intel Wired LAN Driver Updates 2018-04-24

This series contains fixes to ixgbevf, igb and ice drivers.

Colin Ian King fixes the return value on error for the new XDP support
that went into ixgbevf for 4.17.

Vinicius provides a fix for queue 0 for igb, which was not receiving all
the credits it needed when QAV mode was enabled.

Anirudh provides several fixes for the new ice driver, starting with
properly initializing num_nodes_added to zero.  Fixed up a code comment
to better reflect what is really going on in the code.  Fixed how to
detect if an OICR interrupt has occurred to a more reliable method.

Md Fahad fixes the ice driver to allocate the right amount of memory
when reading and storing the devices MAC addresses.  The device can have
up to 2 MAC addresses (LAN and WoL), while WoL is currently not
supported, we need to ensure it can be properly handled when support is
added.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet/tls: remove redundant second null check on sgout
Colin Ian King [Tue, 24 Apr 2018 12:36:58 +0000 (13:36 +0100)]
net/tls: remove redundant second null check on sgout

A duplicated null check on sgout is redundant as it is known to be
already true because of the identical earlier check. Remove it.
Detected by cppcheck:

net/tls/tls_sw.c:696: (warning) Identical inner 'if' condition is always
true.

Signed-off-by: Colin Ian King <colin.king@canonical.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agofsl/fman_port: remove redundant check on port->rev_info.major
Colin Ian King [Tue, 24 Apr 2018 11:39:45 +0000 (12:39 +0100)]
fsl/fman_port: remove redundant check on port->rev_info.major

The check port->rev_info.major >= 6 is being performed twice, thus
the inner second check is always true and is redundant, hence it
can be removed. Detected by cppcheck.

drivers/net/ethernet/freescale/fman/fman_port.c:1394]: (warning)
Identical inner 'if' condition is always true.

Signed-off-by: Colin Ian King <colin.king@canonical.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agoice: Fix insufficient memory issue in ice_aq_manage_mac_read
Md Fahad Iqbal Polash [Mon, 16 Apr 2018 17:07:03 +0000 (10:07 -0700)]
ice: Fix insufficient memory issue in ice_aq_manage_mac_read

For the MAC read operation, the device can return up to two (LAN and WoL)
MAC addresses. Without access to adequate memory, the device will return
an error. Fixed this by allocating the right amount of memory. Also, logic
to detect and copy the LAN MAC address into the port_info structure has
been added. Note that the WoL MAC address is ignored currently as the WoL
feature isn't supported yet.

Fixes: dc49c7723676 ("ice: Get MAC/PHY/link info and scheduler topology")
Signed-off-by: Md Fahad Iqbal Polash <md.fahad.iqbal.polash@intel.com>
Signed-off-by: Anirudh Venkataramanan <anirudh.venkataramanan@intel.com>
Tested-by: Tony Brelinski <tonyx.brelinski@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
6 years agoqed: Fix copying 2 strings
Denis Bolotin [Tue, 24 Apr 2018 12:32:53 +0000 (15:32 +0300)]
qed: Fix copying 2 strings

The strscpy() was a recent fix (net: qed: use correct strncpy() size) to
prevent passing the length of the source buffer to strncpy() and guarantee
null termination.
It misses the goal of overwriting only the first 3 characters in
"???_BIG_RAM" and "???_RAM" while keeping the rest of the string.
Use strncpy() with the length of 3, without null termination.

Signed-off-by: Denis Bolotin <denis.bolotin@cavium.com>
Signed-off-by: Ariel Elior <ariel.elior@cavium.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agosfc: ARFS filter IDs
Edward Cree [Tue, 24 Apr 2018 16:09:30 +0000 (17:09 +0100)]
sfc: ARFS filter IDs

Associate an arbitrary ID with each ARFS filter, allowing to properly query
 for expiry.  The association is maintained in a hash table, which is
 protected by a spinlock.

v3: fix build warnings when CONFIG_RFS_ACCEL is disabled (thanks lkp-robot).
v2: fixed uninitialised variable (thanks davem and lkp-robot).

Fixes: 3af0f34290f6 ("sfc: replace asynchronous filter operations")
Signed-off-by: Edward Cree <ecree@solarflare.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agoMerge branch 'ipconfig-NTP-server-support-bug-fixes-documentation-improvements'
David S. Miller [Tue, 24 Apr 2018 17:40:42 +0000 (13:40 -0400)]
Merge branch 'ipconfig-NTP-server-support-bug-fixes-documentation-improvements'

Chris Novakovic says:

====================
ipconfig: NTP server support, bug fixes, documentation improvements

This series (against net-next) makes various improvements to ipconfig:

 - Patch #1 correctly documents the behaviour of parameter 4 in the
   "ip=" and "nfsaddrs=" command line parameter.
 - Patch #2 tidies up the printk()s for reporting configured name
   servers.
 - Patch #3 fixes a bug in autoconfiguration via BOOTP whereby the IP
   addresses of IEN-116 name servers are requested from the BOOTP
   server, rather than those of DNS name servers.
 - Patch #4 requests the number of DNS servers specified by
   CONF_NAMESERVERS_MAX when autoconfiguring via BOOTP, rather than
   hardcoding it to 2.
 - Patch #5 fully documents the contents and format of /proc/net/pnp in
   Documentation/filesystems/nfs/nfsroot.txt.
 - Patch #6 fixes a bug whereby bogus information is written to
   /proc/net/pnp when ipconfig is not used.
 - Patch #7 creates a new procfs directory for ipconfig-related
   configuration reports at /proc/net/ipconfig.
 - Patch #8 allows for NTP servers to be configured (manually on the
   kernel command line or automatically via DHCP), enabling systems with
   an NFS root filesystem to synchronise their clock before mounting
   their root filesystem. NTP server IP addresses are written to
   /proc/net/ipconfig/ntp_servers.

Changes from v1:

 - David requested that a new directory /proc/net/ipconfig be created to
   contain ipconfig-related configuration reports, which is implemented
   in the new patch #7. NTP server IPs are now written to this directory
   instead of /proc/net/ntp in the new patch #8.
 - Cong and David both requested that the modification to CREDITS be
   dropped. This patch has been removed from the series.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agoipconfig: Write NTP server IPs to /proc/net/ipconfig/ntp_servers
Chris Novakovic [Tue, 24 Apr 2018 02:56:39 +0000 (03:56 +0100)]
ipconfig: Write NTP server IPs to /proc/net/ipconfig/ntp_servers

Distributed filesystems are most effective when the server and client
clocks are synchronised. Embedded devices often use NFS for their
root filesystem but typically do not contain an RTC, so the clocks of
the NFS server and the embedded device will be out-of-sync when the root
filesystem is mounted (and may not be synchronised until late in the
boot process).

Extend ipconfig with the ability to export IP addresses of NTP servers
it discovers to /proc/net/ipconfig/ntp_servers. They can be supplied as
follows:

 - If ipconfig is configured manually via the "ip=" or "nfsaddrs="
   kernel command line parameters, one NTP server can be specified in
   the new "<ntp0-ip>" parameter.
 - If ipconfig is autoconfigured via DHCP, request DHCP option 42 in
   the DHCPDISCOVER message, and record the IP addresses of up to three
   NTP servers sent by the responding DHCP server in the subsequent
   DHCPOFFER message.

ipconfig will only write the NTP server IP addresses it discovers to
/proc/net/ipconfig/ntp_servers, one per line (in the order received from
the DHCP server, if DHCP autoconfiguration is used); making use of these
NTP servers is the responsibility of a user space process (e.g. an
initrd/initram script that invokes an NTP client before mounting an NFS
root filesystem).

Signed-off-by: Chris Novakovic <chris@chrisn.me.uk>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agoipconfig: Create /proc/net/ipconfig directory
Chris Novakovic [Tue, 24 Apr 2018 02:56:38 +0000 (03:56 +0100)]
ipconfig: Create /proc/net/ipconfig directory

To allow ipconfig to report IP configuration details to user space
processes without cluttering /proc/net, create a new subdirectory
/proc/net/ipconfig. All files containing IP configuration details should
be written to this directory.

Signed-off-by: Chris Novakovic <chris@chrisn.me.uk>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agoipconfig: Correctly initialise ic_nameservers
Chris Novakovic [Tue, 24 Apr 2018 02:56:37 +0000 (03:56 +0100)]
ipconfig: Correctly initialise ic_nameservers

ic_nameservers, which stores the list of name servers discovered by
ipconfig, is initialised (i.e. has all of its elements set to NONE, or
0xffffffff) by ic_nameservers_predef() in the following scenarios:

 - before the "ip=" and "nfsaddrs=" kernel command line parameters are
   parsed (in ip_auto_config_setup());
 - before autoconfiguring via DHCP or BOOTP (in ic_bootp_init()), in
   order to clear any values that may have been set after parsing "ip="
   or "nfsaddrs=" and are no longer needed.

This means that ic_nameservers_predef() is not called when neither "ip="
nor "nfsaddrs=" is specified on the kernel command line. In this
scenario, every element in ic_nameservers remains set to 0x00000000,
which is indistinguishable from ANY and causes pnp_seq_show() to write
the following (bogus) information to /proc/net/pnp:

  #MANUAL
  nameserver 0.0.0.0
  nameserver 0.0.0.0
  nameserver 0.0.0.0

This is potentially problematic for systems that blindly link
/etc/resolv.conf to /proc/net/pnp.

Ensure that ic_nameservers is also initialised when neither "ip=" nor
"nfsaddrs=" are specified by calling ic_nameservers_predef() in
ip_auto_config(), but only when ip_auto_config_setup() was not called
earlier. This causes the following to be written to /proc/net/pnp, and
is consistent with what gets written when ipconfig is configured
manually but no name servers are specified on the kernel command line:

  #MANUAL

Signed-off-by: Chris Novakovic <chris@chrisn.me.uk>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agoipconfig: Document /proc/net/pnp
Chris Novakovic [Tue, 24 Apr 2018 02:56:36 +0000 (03:56 +0100)]
ipconfig: Document /proc/net/pnp

Fully document the format used by the /proc/net/pnp file written by
ipconfig, explain where its values originate from, and clarify that the
tertiary name server IP and DNS domain name are only written to the file
when autoconfiguration is used.

Signed-off-by: Chris Novakovic <chris@chrisn.me.uk>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agoipconfig: BOOTP: Request CONF_NAMESERVERS_MAX name servers
Chris Novakovic [Tue, 24 Apr 2018 02:56:35 +0000 (03:56 +0100)]
ipconfig: BOOTP: Request CONF_NAMESERVERS_MAX name servers

When ipconfig is autoconfigured via BOOTP, the request packet
initialised by ic_bootp_init_ext() always allocates 8 bytes for the name
server option, limiting the BOOTP server to responding with at most 2
name servers even though ipconfig in fact supports an arbitrary number
of name servers (as defined by CONF_NAMESERVERS_MAX, which is currently
3).

Only request name servers in the request packet if CONF_NAMESERVERS_MAX
is positive (to comply with [1, §3.8]), and allocate enough space in the
packet for CONF_NAMESERVERS_MAX name servers to indicate the maximum
number we can accept in response.

[1] RFC 2132, "DHCP Options and BOOTP Vendor Extensions":
    https://tools.ietf.org/rfc/rfc2132.txt

Signed-off-by: Chris Novakovic <chris@chrisn.me.uk>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agoipconfig: BOOTP: Don't request IEN-116 name servers
Chris Novakovic [Tue, 24 Apr 2018 02:56:34 +0000 (03:56 +0100)]
ipconfig: BOOTP: Don't request IEN-116 name servers

When ipconfig is autoconfigured via BOOTP, the request packet
initialised by ic_bootp_init_ext() allocates 8 bytes for tag 5 ("Name
Server" [1, §3.7]), but tag 5 in the response isn't processed by
ic_do_bootp_ext(). Instead, allocate the 8 bytes to tag 6 ("Domain Name
Server" [1, §3.8]), which is processed by ic_do_bootp_ext(), and appears
to have been the intended tag to request.

This won't cause any breakage for existing users, as tag 5 responses
provided by BOOTP servers weren't being processed anyway.

[1] RFC 2132, "DHCP Options and BOOTP Vendor Extensions":
    https://tools.ietf.org/rfc/rfc2132.txt

Signed-off-by: Chris Novakovic <chris@chrisn.me.uk>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agoipconfig: Tidy up reporting of name servers
Chris Novakovic [Tue, 24 Apr 2018 02:56:33 +0000 (03:56 +0100)]
ipconfig: Tidy up reporting of name servers

Commit 5e953778a2aab04929a5e7b69f53dc26e39b079e ("ipconfig: add
nameserver IPs to kernel-parameter ip=") adds the IP addresses of
discovered name servers to the summary printed by ipconfig when
configuration is complete. It appears the intention in ip_auto_config()
was to print the name servers on a new line (especially given the
spacing and lack of comma before "nameserver0="), but they're actually
printed on the same line as the NFS root filesystem configuration
summary:

  [    0.686186] IP-Config: Complete:
  [    0.686226]      device=eth0, hwaddr=xx:xx:xx:xx:xx:xx, ipaddr=10.0.0.2, mask=255.255.255.0, gw=10.0.0.1
  [    0.686328]      host=test, domain=example.com, nis-domain=(none)
  [    0.686386]      bootserver=10.0.0.1, rootserver=10.0.0.1, rootpath=     nameserver0=10.0.0.1

This makes it harder to read and parse ipconfig's output. Instead, print
the name servers on a separate line:

  [    0.791250] IP-Config: Complete:
  [    0.791289]      device=eth0, hwaddr=xx:xx:xx:xx:xx:xx, ipaddr=10.0.0.2, mask=255.255.255.0, gw=10.0.0.1
  [    0.791407]      host=test, domain=example.com, nis-domain=(none)
  [    0.791475]      bootserver=10.0.0.1, rootserver=10.0.0.1, rootpath=
  [    0.791476]      nameserver0=10.0.0.1

Signed-off-by: Chris Novakovic <chris@chrisn.me.uk>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agoipconfig: Document setting of NIS domain name
Chris Novakovic [Tue, 24 Apr 2018 02:56:32 +0000 (03:56 +0100)]
ipconfig: Document setting of NIS domain name

ic_do_bootp_ext() is responsible for parsing the "ip=" and "nfsaddrs="
kernel parameters. If a "." character is found in parameter 4 (the
client's hostname), everything before the first "." is used as the
hostname, and everything after it is used as the NIS domain name (but
not necessarily the DNS domain name).

Document this behaviour in Documentation/filesystems/nfs/nfsroot.txt,
as it is not made explicit.

Signed-off-by: Chris Novakovic <chris@chrisn.me.uk>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet: ethtool: Add missing kernel doc for FEC parameters
Florian Fainelli [Mon, 23 Apr 2018 22:51:38 +0000 (15:51 -0700)]
net: ethtool: Add missing kernel doc for FEC parameters

While adding support for ethtool::get_fecparam and set_fecparam, kernel
doc for these functions was missed, add those.

Fixes: 1a5f3da20bd9 ("net: ethtool: add support for forward error correction modes")
Signed-off-by: Florian Fainelli <f.fainelli@gmail.com>
Acked-by: Roopa Prabhu <roopa@cumulusnetworks.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agoMerge branch 'rhash-cleanups'
David S. Miller [Tue, 24 Apr 2018 17:21:46 +0000 (13:21 -0400)]
Merge branch 'rhash-cleanups'

NeilBrown says:

====================
A few rhashtables cleanups

2 patches fixes documentation
1 fixes a bit in rhashtable_walk_start()
1 improves rhashtable_walk stability.

All reviewed and Acked.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agorhashtable: improve rhashtable_walk stability when stop/start used.
NeilBrown [Mon, 23 Apr 2018 22:29:13 +0000 (08:29 +1000)]
rhashtable: improve rhashtable_walk stability when stop/start used.

When a walk of an rhashtable is interrupted with rhastable_walk_stop()
and then rhashtable_walk_start(), the location to restart from is based
on a 'skip' count in the current hash chain, and this can be incorrect
if insertions or deletions have happened.  This does not happen when
the walk is not stopped and started as iter->p is a placeholder which
is safe to use while holding the RCU read lock.

In rhashtable_walk_start() we can revalidate that 'p' is still in the
same hash chain.  If it isn't then the current method is still used.

With this patch, if a rhashtable walker ensures that the current
object remains in the table over a stop/start period (possibly by
elevating the reference count if that is sufficient), it can be sure
that a walk will not miss objects that were in the hashtable for the
whole time of the walk.

rhashtable_walk_start() may not find the object even though it is
still in the hashtable if a rehash has moved it to a new table.  In
this case it will (eventually) get -EAGAIN and will need to proceed
through the whole table again to be sure to see everything at least
once.

Acked-by: Herbert Xu <herbert@gondor.apana.org.au>
Signed-off-by: NeilBrown <neilb@suse.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agorhashtable: reset iter when rhashtable_walk_start sees new table
NeilBrown [Mon, 23 Apr 2018 22:29:13 +0000 (08:29 +1000)]
rhashtable: reset iter when rhashtable_walk_start sees new table

The documentation claims that when rhashtable_walk_start_check()
detects a resize event, it will rewind back to the beginning
of the table.  This is not true.  We need to set ->slot and
->skip to be zero for it to be true.

Acked-by: Herbert Xu <herbert@gondor.apana.org.au>
Signed-off-by: NeilBrown <neilb@suse.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agorhashtable: Revise incorrect comment on r{hl, hash}table_walk_enter()
NeilBrown [Mon, 23 Apr 2018 22:29:13 +0000 (08:29 +1000)]
rhashtable: Revise incorrect comment on r{hl, hash}table_walk_enter()

Neither rhashtable_walk_enter() or rhltable_walk_enter() sleep, though
they do take a spinlock without irq protection.
So revise the comments to accurately state the contexts in which
these functions can be called.

Acked-by: Herbert Xu <herbert@gondor.apana.org.au>
Signed-off-by: NeilBrown <neilb@suse.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agorhashtable: remove outdated comments about grow_decision etc
NeilBrown [Mon, 23 Apr 2018 22:29:13 +0000 (08:29 +1000)]
rhashtable: remove outdated comments about grow_decision etc

grow_decision and shink_decision no longer exist, so remove
the remaining references to them.

Acked-by: Herbert Xu <herbert@gondor.apana.org.au>
Signed-off-by: NeilBrown <neilb@suse.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agotcp: md5: only call tp->af_specific->md5_lookup() for md5 sockets
Eric Dumazet [Mon, 23 Apr 2018 21:46:25 +0000 (14:46 -0700)]
tcp: md5: only call tp->af_specific->md5_lookup() for md5 sockets

RETPOLINE made calls to tp->af_specific->md5_lookup() quite expensive,
given they have no result.
We can omit the calls for sockets that have no md5 keys.

Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agopacket: fix bitfield update race
Willem de Bruijn [Mon, 23 Apr 2018 21:37:03 +0000 (17:37 -0400)]
packet: fix bitfield update race

Updates to the bitfields in struct packet_sock are not atomic.
Serialize these read-modify-write cycles.

Move po->running into a separate variable. Its writes are protected by
po->bind_lock (except for one startup case at packet_create). Also
replace a textual precondition warning with lockdep annotation.

All others are set only in packet_setsockopt. Serialize these
updates by holding the socket lock. Analogous to other field updates,
also hold the lock when testing whether a ring is active (pg_vec).

Fixes: 8dc419447415 ("[PACKET]: Add optional checksum computation for recvmsg")
Reported-by: DaeRyong Jeong <threeearcat@gmail.com>
Reported-by: Byoungyoung Lee <byoungyoung@purdue.edu>
Signed-off-by: Willem de Bruijn <willemb@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agoice: Do not check INTEVENT bit for OICR interrupts
Ben Shelton [Wed, 11 Apr 2018 19:21:33 +0000 (12:21 -0700)]
ice: Do not check INTEVENT bit for OICR interrupts

According to the hardware spec, checking the INTEVENT bit isn't a
reliable way to detect if an OICR interrupt has occurred. This is
because this bit can be cleared by the hardware/firmware before the
interrupt service routine has run. So instead, just check for OICR
events every time.

Fixes: 940b61af02f4 ("ice: Initialize PF and setup miscellaneous interrupt")
Signed-off-by: Ben Shelton <benjamin.h.shelton@intel.com>
Signed-off-by: Anirudh Venkataramanan <anirudh.venkataramanan@intel.com>
Tested-by: Tony Brelinski <tonyx.brelinski@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
6 years agoice: Fix incorrect comment for action type
Anirudh Venkataramanan [Wed, 11 Apr 2018 17:41:47 +0000 (10:41 -0700)]
ice: Fix incorrect comment for action type

Action type 5 defines large action generic values. Fix comment to
reflect that better.

Signed-off-by: Anirudh Venkataramanan <anirudh.venkataramanan@intel.com>
Tested-by: Tony Brelinski <tonyx.brelinski@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
6 years agoice: Fix initialization for num_nodes_added
Anirudh Venkataramanan [Tue, 10 Apr 2018 17:49:49 +0000 (10:49 -0700)]
ice: Fix initialization for num_nodes_added

ice_sched_add_nodes_to_layer is used recursively, and so we start
with num_nodes_added being 0. This way, in case of an error or if
num_nodes is NULL, the function just returns 0 to indicate that no
nodes were added.

Fixes: 5513b920a4f7 ("ice: Update Tx scheduler tree for VSI multi-Tx queue support")
Signed-off-by: Anirudh Venkataramanan <anirudh.venkataramanan@intel.com>
Tested-by: Tony Brelinski <tonyx.brelinski@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
6 years agoigb: Fix the transmission mode of queue 0 for Qav mode
Vinicius Costa Gomes [Sat, 31 Mar 2018 00:06:52 +0000 (17:06 -0700)]
igb: Fix the transmission mode of queue 0 for Qav mode

When Qav mode is enabled, queue 0 should be kept on Stream Reservation
mode. From the i210 datasheet, section 8.12.19:

"Note: Queue0 QueueMode must be set to 1b when TransmitMode is set to
Qav." ("QueueMode 1b" represents the Stream Reservation mode)

The solution is to give queue 0 the all the credits it might need, so
it has priority over queue 1.

A situation where this can happen is when cbs is "installed" only on
queue 1, leaving queue 0 alone. For example:

$ tc qdisc replace dev enp2s0 handle 100: parent root mqprio num_tc 3 \
         map 2 2 1 0 2 2 2 2 2 2 2 2 2 2 2 2 queues 1@0 1@1 2@2 hw 0

$ tc qdisc replace dev enp2s0 parent 100:2 cbs locredit -1470 \
         hicredit 30 sendslope -980000 idleslope 20000 offload 1

Signed-off-by: Vinicius Costa Gomes <vinicius.gomes@intel.com>
Tested-by: Aaron Brown <aaron.f.brown@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
6 years agoixgbevf: ensure xdp_ring resources are free'd on error exit
Colin Ian King [Tue, 27 Mar 2018 14:21:48 +0000 (15:21 +0100)]
ixgbevf: ensure xdp_ring resources are free'd on error exit

The current error handling for failed resource setup for xdp_ring
data is a break out of the loop and returning 0 indicated everything
was OK, when in fact it is not.  Fix this by exiting via the
error exit label err_setup_tx that will clean up the resources
correctly and return and error status.

Detected by CoverityScan, CID#1466879 ("Logically dead code")

Fixes: 21092e9ce8b1 ("ixgbevf: Add support for XDP_TX action")
Signed-off-by: Colin Ian King <colin.king@canonical.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
6 years agoRevert "net: init sk_cookie for inet socket"
Yafang Shao [Tue, 24 Apr 2018 15:07:45 +0000 (23:07 +0800)]
Revert "net: init sk_cookie for inet socket"

This reverts commit <c6849a3ac17e> ("net: init sk_cookie for inet socket")

Per discussion with Eric, when update sock_net(sk)->cookie_gen, the
whole cache cache line will be invalidated, as this cache line is shared
with all cpus, that may cause great performace hit.

Bellow is the data form Eric.
"Performance is reduced from ~5 Mpps to ~3.8 Mpps with 16 RX queues on
my host" when running synflood test.

Have to revert it to prevent from cache line false sharing.

Signed-off-by: Yafang Shao <laoar.shao@gmail.com>
Reviewed-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agoMerge branch 'net-DIM-tx'
David S. Miller [Tue, 24 Apr 2018 14:15:08 +0000 (10:15 -0400)]
Merge branch 'net-DIM-tx'

Tal Gilboa says:

====================
Introduce adaptive TX interrupt moderation to net DIM

Net DIM is a library designed for dynamic interrupt moderation. It was
implemented and optimized with receive side interrupts in mind, since these
are usually the CPU expensive ones. This patch-set introduces adaptive transmit
interrupt moderation to net DIM, complete with a usage in the mlx5e driver.
Using adaptive TX behavior would reduce interrupt rate for multiple scenarios.
Furthermore, it is essential for increasing bandwidth on cases where payload
aggregation is required.

v3: Remove "inline" from functions in .c files (requested by DaveM). Revert
adding "enabled" field from struct net_dim and applied mlx5e structural
suggestions (suggested by SaeedM).

v2: Rebase over proper tree.

v1: Fix compilation issues due to missed function renaming.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet/mlx5e: Enable adaptive-TX moderation
Tal Gilboa [Tue, 24 Apr 2018 10:36:03 +0000 (13:36 +0300)]
net/mlx5e: Enable adaptive-TX moderation

Add support for adaptive TX moderation. This greatly reduces TX interrupt
rate and increases bandwidth, mostly for TCP bandwidth over ARM
architecture (below). There is a slight single stream TCP with very large
message sizes degradation (x86). In this case if there's any moderation on
transmitted packets the bandwidth would reduce due to hitting TCP output limit.
Since this is a synthetic case, this is still worth doing.

Performance improvement (ConnectX-4Lx 40GbE, ARM)
TCP 64B bandwidth with 1-50 streams increased 6-35%.
TCP 64B bandwidth with 100-500 streams increased 20-70%.

Performance improvement (ConnectX-5 100GbE, x86)
Bandwidth: increased up to 40% (1024B with 10s of streams).
Interrupt rate: reduced up to 50% (1024B with 1000s of streams).

Performance degradation (ConnectX-5 100GbE, x86)
Bandwidth: up to 10% decrease single stream TCP (1MB message size from
51Gb/s to 47Gb/s).

Signed-off-by: Tal Gilboa <talgi@mellanox.com>
Reviewed-by: Tariq Toukan <tariqt@mellanox.com>
Acked-by: Saeed Mahameed <saeedm@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet/dim: Support adaptive TX moderation
Tal Gilboa [Tue, 24 Apr 2018 10:36:02 +0000 (13:36 +0300)]
net/dim: Support adaptive TX moderation

Interrupt moderation for TX traffic requires different profiles than RX
interrupt moderation. The main goal here is to reduce interrupt rate and
allow better payload aggregation by keeping SKBs in the TX queue a bit
longer. Ping-pong behavior would get a profile with a short timer, so
latency wouldn't increase for these scenarios. There might be a slight
degradation in bandwidth for single stream with large message sizes, since
net.ipv4.tcp_limit_output_bytes is limiting the allowed TX traffic, but
with many streams performance is always improved.

Signed-off-by: Tal Gilboa <talgi@mellanox.com>
Reviewed-by: Tariq Toukan <tariqt@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet/dim: Rename *_get_profile() functions to *_get_rx_moderation()
Tal Gilboa [Tue, 24 Apr 2018 10:36:01 +0000 (13:36 +0300)]
net/dim: Rename *_get_profile() functions to *_get_rx_moderation()

Preparation for introducing adaptive TX to net DIM.

Signed-off-by: Tal Gilboa <talgi@mellanox.com>
Reviewed-by: Tariq Toukan <tariqt@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agovhost_net: use packet weight for rx handler, too
Paolo Abeni [Tue, 24 Apr 2018 08:34:36 +0000 (10:34 +0200)]
vhost_net: use packet weight for rx handler, too

Similar to commit a2ac99905f1e ("vhost-net: set packet weight of
tx polling to 2 * vq size"), we need a packet-based limit for
handler_rx, too - elsewhere, under rx flood with small packets,
tx can be delayed for a very long time, even without busypolling.

The pkt limit applied to handle_rx must be the same applied by
handle_tx, or we will get unfair scheduling between rx and tx.
Tying such limit to the queue length makes it less effective for
large queue length values and can introduce large process
scheduler latencies, so a constant valued is used - likewise
the existing bytes limit.

The selected limit has been validated with PVP[1] performance
test with different queue sizes:

queue size 256 512 1024

baseline 366 354 362
weight 128 715 723 670
weight 256 740 745 733
weight 512 600 460 583
weight 1024 423 427 418

A packet weight of 256 gives peek performances in under all the
tested scenarios.

No measurable regression in unidirectional performance tests has
been detected.

[1] https://developers.redhat.com/blog/2017/06/05/measuring-and-comparing-open-vswitch-performance/

Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Acked-by: Jason Wang <jasowang@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agoteam: fix netconsole setup over team
Xin Long [Tue, 24 Apr 2018 06:33:37 +0000 (14:33 +0800)]
team: fix netconsole setup over team

The same fix in Commit dbe173079ab5 ("bridge: fix netconsole
setup over bridge") is also needed for team driver.

While at it, remove the unnecessary parameter *team from
team_port_enable_netpoll().

v1->v2:
  - fix it in a better way, as does bridge.

Fixes: 0fb52a27a04a ("team: cleanup netpoll clode")
Reported-by: João Avelino Bellomo Filho <jbellomo@redhat.com>
Signed-off-by: Xin Long <lucien.xin@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet: fib_rules: fix l3mdev netlink attr processing
Roopa Prabhu [Tue, 24 Apr 2018 03:08:41 +0000 (20:08 -0700)]
net: fib_rules: fix l3mdev netlink attr processing

Fixes: b16fb418b1bf ("net: fib_rules: add extack support")
Signed-off-by: Roopa Prabhu <roopa@cumulusnetworks.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agoMerge branch 'amd-xgbe-fixes'
David S. Miller [Tue, 24 Apr 2018 01:24:23 +0000 (21:24 -0400)]
Merge branch 'amd-xgbe-fixes'

aTom Lendacky says:

====================
amd-xgbe: AMD XGBE driver fixes 2018-04-23

This patch series addresses some issues in the AMD XGBE driver.

The following fixes are included in this driver update series:

- Improve KR auto-negotiation and training (2 patches)
  - Add pre and post auto-negotiation hooks
  - Use the pre and post auto-negotiation hooks to disable CDR tracking
    during auto-negotiation page exchange in KR mode
- Check for SFP tranceiver signal support and only use the signal if the
  SFP indicates that it is supported

This patch series is based on net.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agoamd-xgbe: Only use the SFP supported transceiver signals
Tom Lendacky [Mon, 23 Apr 2018 16:43:34 +0000 (11:43 -0500)]
amd-xgbe: Only use the SFP supported transceiver signals

The SFP eeprom indicates the transceiver signals (Rx LOS, Tx Fault, etc.)
that it supports.  Update the driver to include checking the eeprom data
when deciding whether to use a transceiver signal.

Signed-off-by: Tom Lendacky <thomas.lendacky@amd.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agoamd-xgbe: Improve KR auto-negotiation and training
Tom Lendacky [Mon, 23 Apr 2018 16:43:17 +0000 (11:43 -0500)]
amd-xgbe: Improve KR auto-negotiation and training

Update xgbe-phy-v2.c to make use of the auto-negotiation (AN) phy hooks
to improve the ability to successfully complete Clause 73 AN when running
at 10gbps.  Hardware can sometimes have issues with CDR lock when the
AN DME page exchange is being performed.

The AN and KR training hooks are used as follows:
- The pre AN hook is used to disable CDR tracking in the PHY so that the
  DME page exchange can be successfully and consistently completed.
- The post KR training hook is used to re-enable the CDR tracking so that
  KR training can successfully complete.
- The post AN hook is used to check for an unsuccessful AN which will
  increase a CDR tracking enablement delay (up to a maximum value).

Add two debugfs entries to allow control over use of the CDR tracking
workaround.  The debugfs entries allow the CDR tracking workaround to
be disabled and determine whether to re-enable CDR tracking before or
after link training has been initiated.

Also, with these changes the receiver reset cycle that is performed during
the link status check can be performed less often.

Signed-off-by: Tom Lendacky <thomas.lendacky@amd.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agoamd-xgbe: Add pre/post auto-negotiation phy hooks
Tom Lendacky [Mon, 23 Apr 2018 16:43:08 +0000 (11:43 -0500)]
amd-xgbe: Add pre/post auto-negotiation phy hooks

Add hooks to the driver auto-negotiation (AN) flow to allow the different
phy implementations to perform any steps necessary to improve AN.

Signed-off-by: Tom Lendacky <thomas.lendacky@amd.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agopppoe: check sockaddr length in pppoe_connect()
Guillaume Nault [Mon, 23 Apr 2018 14:38:27 +0000 (16:38 +0200)]
pppoe: check sockaddr length in pppoe_connect()

We must validate sockaddr_len, otherwise userspace can pass fewer data
than we expect and we end up accessing invalid data.

Fixes: 224cf5ad14c0 ("ppp: Move the PPP drivers")
Reported-by: syzbot+4f03bdf92fdf9ef5ddab@syzkaller.appspotmail.com
Signed-off-by: Guillaume Nault <g.nault@alphalink.fr>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agol2tp: check sockaddr length in pppol2tp_connect()
Guillaume Nault [Mon, 23 Apr 2018 14:15:14 +0000 (16:15 +0200)]
l2tp: check sockaddr length in pppol2tp_connect()

Check sockaddr_len before dereferencing sp->sa_protocol, to ensure that
it actually points to valid data.

Fixes: fd558d186df2 ("l2tp: Split pppol2tp patch into separate l2tp and ppp parts")
Reported-by: syzbot+a70ac890b23b1bf29f5c@syzkaller.appspotmail.com
Signed-off-by: Guillaume Nault <g.nault@alphalink.fr>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agoselftests: net: update .gitignore with missing test
Anders Roxell [Mon, 23 Apr 2018 14:00:50 +0000 (16:00 +0200)]
selftests: net: update .gitignore with missing test

Fixes: 192dc405f308 ("selftests: net: add tcp_mmap program")
Signed-off-by: Anders Roxell <anders.roxell@linaro.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet: phy: marvell: clear wol event before setting it
Jingju Hou [Mon, 23 Apr 2018 07:22:49 +0000 (15:22 +0800)]
net: phy: marvell: clear wol event before setting it

If WOL event happened once, the LED[2] interrupt pin will not be
cleared unless we read the CSISR register. If interrupts are in use,
the normal interrupt handling will clear the WOL event. Let's clear the
WOL event before enabling it if !phy_interrupt_is_valid().

Signed-off-by: Jingju Hou <Jingju.Hou@synaptics.com>
Signed-off-by: Jisheng Zhang <Jisheng.Zhang@synaptics.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agodca: make function dca_common_get_tag static
Colin Ian King [Mon, 23 Apr 2018 12:49:38 +0000 (13:49 +0100)]
dca: make function dca_common_get_tag static

Function dca_common_get_tag is local to the source and does not need to be
in global scope, so make it static.

Cleans up sparse warning:
drivers/dca/dca-core.c:273:4: warning: symbol 'dca_common_get_tag' was
not declared. Should it be static?

Signed-off-by: Colin Ian King <colin.king@canonical.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agoMerge git://git.kernel.org/pub/scm/linux/kernel/git/pablo/nf
David S. Miller [Mon, 23 Apr 2018 20:22:24 +0000 (16:22 -0400)]
Merge git://git./pub/scm/linux/kernel/git/pablo/nf

Pablo Neira Ayuso says:

====================
Netfilter/IPVS fixes for net

The following patchset contains Netfilter/IPVS fixes for your net tree,
they are:

1) Fix SIP conntrack with phones sending session descriptions for different
   media types but same port numbers, from Florian Westphal.

2) Fix incorrect rtnl_lock mutex logic from IPVS sync thread, from Julian
   Anastasov.

3) Skip compat array allocation in ebtables if there is no entries, also
   from Florian.

4) Do not lose left/right bits when shifting marks from xt_connmark, from
   Jack Ma.

5) Silence false positive memleak in conntrack extensions, from Cong Wang.

6) Fix CONFIG_NF_REJECT_IPV6=m link problems, from Arnd Bergmann.

7) Cannot kfree rule that is already in list in nf_tables, switch order
   so this error handling is not required, from Florian Westphal.

8) Release set name in error path, from Florian.

9) include kmemleak.h in nf_conntrack_extend.c, from Stepheh Rothwell.

10) NAT chain and extensions depend on NF_TABLES.

11) Out of bound access when renaming chains, from Taehee Yoo.

12) Incorrect casting in xt_connmark leads to wrong bitshifting.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agoMerge branch 'ipv6-couple-of-fixes-for-rcu-change-to-from'
David S. Miller [Mon, 23 Apr 2018 20:12:55 +0000 (16:12 -0400)]
Merge branch 'ipv6-couple-of-fixes-for-rcu-change-to-from'

David Ahern says:

====================
net/ipv6: couple of fixes for rcu change to from

So many details... I am thankful for all the robots running the
permutations and tools.

Two bug fixes from the rcu change to rt->from:
1. missing rcu lock in ip6_negative_advice
2. rcu dereferences in 2 sites
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet/ipv6: Fix missing rcu dereferences on from
David Ahern [Mon, 23 Apr 2018 18:32:07 +0000 (11:32 -0700)]
net/ipv6: Fix missing rcu dereferences on from

kbuild test robot reported 2 uses of rt->from not properly accessed
using rcu_dereference:
1. add rcu_dereference_protected to rt6_remove_exception_rt and make
   sure it is always called with rcu lock held.

2. change rt6_do_redirect to take a reference on 'from' when accessed
   the first time so it can be used the sceond time outside of the lock

Fixes: a68886a69180 ("net/ipv6: Make from in rt6_info rcu protected")
Reported-by: kbuild test robot <lkp@intel.com>
Signed-off-by: David Ahern <dsahern@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet/ipv6: add rcu locking to ip6_negative_advice
David Ahern [Mon, 23 Apr 2018 18:32:06 +0000 (11:32 -0700)]
net/ipv6: add rcu locking to ip6_negative_advice

syzbot reported a suspicious rcu_dereference_check:
  __dump_stack lib/dump_stack.c:77 [inline]
  dump_stack+0x1b9/0x294 lib/dump_stack.c:113
  lockdep_rcu_suspicious+0x14a/0x153 kernel/locking/lockdep.c:4592
  rt6_check_expired+0x38b/0x3e0 net/ipv6/route.c:410
  ip6_negative_advice+0x67/0xc0 net/ipv6/route.c:2204
  dst_negative_advice include/net/sock.h:1786 [inline]
  sock_setsockopt+0x138f/0x1fe0 net/core/sock.c:1051
  __sys_setsockopt+0x2df/0x390 net/socket.c:1899
  SYSC_setsockopt net/socket.c:1914 [inline]
  SyS_setsockopt+0x34/0x50 net/socket.c:1911

Add rcu locking around call to rt6_check_expired in
ip6_negative_advice.

Fixes: a68886a69180 ("net/ipv6: Make from in rt6_info rcu protected")
Reported-by: syzbot+2422c9e35796659d2273@syzkaller.appspotmail.com
Signed-off-by: David Ahern <dsahern@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agoMerge branch 'qed-debug-data'
David S. Miller [Mon, 23 Apr 2018 16:05:58 +0000 (12:05 -0400)]
Merge branch 'qed-debug-data'

Denis Bolotin says:

====================
Add configuration information to register dump and debug data

The purpose of this patchset is to add configuration information to the
debug data collection, which already contains register dump.
The first patch (removing the ptt) is essential because it prevents the
unnecessary ptt acquirement when calling mcp APIs.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agoqed: Add configuration information to register dump and debug data
Denis Bolotin [Mon, 23 Apr 2018 11:56:05 +0000 (14:56 +0300)]
qed: Add configuration information to register dump and debug data

Configuration information is added to the debug data collection, in
addition to register dump.
Added qed_dbg_nvm_image() that receives an image type, allocates a
buffer and reads the image. The images are saved in the buffers and the
dump size is updated.

Signed-off-by: Denis Bolotin <denis.bolotin@cavium.com>
Signed-off-by: Ariel Elior <ariel.elior@cavium.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agoqed: Delete unused parameter p_ptt from mcp APIs
Denis Bolotin [Mon, 23 Apr 2018 11:56:04 +0000 (14:56 +0300)]
qed: Delete unused parameter p_ptt from mcp APIs

Since nvm images attributes are cached during driver load, acquiring ptt
is not needed when calling qed_mcp_get_nvm_image().

Signed-off-by: Denis Bolotin <denis.bolotin@cavium.com>
Signed-off-by: Ariel Elior <ariel.elior@cavium.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet: stmmac: Implement logic to automatically select HW Interface
Jose Abreu [Mon, 23 Apr 2018 08:05:15 +0000 (09:05 +0100)]
net: stmmac: Implement logic to automatically select HW Interface

Move all the core version detection to a common place ("hwif.c") and
implement a table which can be used to lookup the correct callbacks for
each IP version.

This simplifies the initialization flow of each IP version and eases
future implementation of new IP versions.

Signed-off-by: Jose Abreu <joabreu@synopsys.com>
Cc: David S. Miller <davem@davemloft.net>
Cc: Joao Pinto <jpinto@synopsys.com>
Cc: Vitor Soares <soares@synopsys.com>
Cc: Giuseppe Cavallaro <peppe.cavallaro@st.com>
Cc: Alexandre Torgue <alexandre.torgue@st.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agoipv6: add RTA_TABLE and RTA_PREFSRC to rtm_ipv6_policy
Eric Dumazet [Mon, 23 Apr 2018 01:29:23 +0000 (18:29 -0700)]
ipv6: add RTA_TABLE and RTA_PREFSRC to rtm_ipv6_policy

KMSAN reported use of uninit-value that I tracked to lack
of proper size check on RTA_TABLE attribute.

I also believe RTA_PREFSRC lacks a similar check.

Fixes: 86872cb57925 ("[IPv6] route: FIB6 configuration using struct fib6_config")
Fixes: c3968a857a6b ("ipv6: RTA_PREFSRC support for ipv6 route source address selection")
Signed-off-by: Eric Dumazet <edumazet@google.com>
Reported-by: syzbot <syzkaller@googlegroups.com>
Acked-by: David Ahern <dsahern@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agor8169: don't use netif_info et al before net_device has been registered
Heiner Kallweit [Sun, 22 Apr 2018 15:15:15 +0000 (17:15 +0200)]
r8169: don't use netif_info et al before net_device has been registered

There's no benefit in using netif_info et al before the net_device has
been registered. We get messages like
r8169 0000:03:00.0 (unnamed net_device) (uninitialized): [message]
Therefore use dev_info/dev_err instead.

As a side effect we don't need parameter dev for function
rtl8169_get_mac_version() any longer.

Signed-off-by: Heiner Kallweit <hkallweit1@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet: init sk_cookie for inet socket
Yafang Shao [Sun, 22 Apr 2018 13:50:04 +0000 (21:50 +0800)]
net: init sk_cookie for inet socket

With sk_cookie we can identify a socket, that is very helpful for
traceing and statistic, i.e. tcp tracepiont and ebpf.
So we'd better init it by default for inet socket.
When using it, we just need call atomic64_read(&sk->sk_cookie).

Signed-off-by: Yafang Shao <laoar.shao@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agobonding: do not set slave_dev npinfo before slave_enable_netpoll in bond_enslave
Xin Long [Sun, 22 Apr 2018 11:11:50 +0000 (19:11 +0800)]
bonding: do not set slave_dev npinfo before slave_enable_netpoll in bond_enslave

After Commit 8a8efa22f51b ("bonding: sync netpoll code with bridge"), it
would set slave_dev npinfo in slave_enable_netpoll when enslaving a dev
if bond->dev->npinfo was set.

However now slave_dev npinfo is set with bond->dev->npinfo before calling
slave_enable_netpoll. With slave_dev npinfo set, __netpoll_setup called
in slave_enable_netpoll will not call slave dev's .ndo_netpoll_setup().
It causes that the lower dev of this slave dev can't set its npinfo.

One way to reproduce it:

  # modprobe bonding
  # brctl addbr br0
  # brctl addif br0 eth1
  # ifconfig bond0 192.168.122.1/24 up
  # ifenslave bond0 eth2
  # systemctl restart netconsole
  # ifenslave bond0 br0
  # ifconfig eth2 down
  # systemctl restart netconsole

The netpoll won't really work.

This patch is to remove that slave_dev npinfo setting in bond_enslave().

Fixes: 8a8efa22f51b ("bonding: sync netpoll code with bridge")
Signed-off-by: Xin Long <lucien.xin@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agoMerge branch 'fib-rules-extack-support'
David S. Miller [Mon, 23 Apr 2018 14:21:25 +0000 (10:21 -0400)]
Merge branch 'fib-rules-extack-support'

Roopa Prabhu says:

====================
fib rules extack support

First patch refactors code to move fib rule netlink handling
into a common function. This became obvious when adding
duplicate extack msgs in add and del paths. Second patch
adds extack msgs.

v2 - Dropped the ip route get support and selftests from
     the series to look at the input path some more (as pointed
     out by ido). Will come back to that next week when i have
     some time. resending just the extack part for now.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet: fib_rules: add extack support
Roopa Prabhu [Sat, 21 Apr 2018 16:41:31 +0000 (09:41 -0700)]
net: fib_rules: add extack support

Signed-off-by: Roopa Prabhu <roopa@cumulusnetworks.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agofib_rules: move common handling of newrule delrule msgs into fib_nl2rule
Roopa Prabhu [Sat, 21 Apr 2018 16:41:30 +0000 (09:41 -0700)]
fib_rules: move common handling of newrule delrule msgs into fib_nl2rule

This reduces code duplication in the fib rule add and del paths.
Get rid of validate_rulemsg. This became obvious when adding duplicate
extack support in fib newrule/delrule error paths.

Signed-off-by: Roopa Prabhu <roopa@cumulusnetworks.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agotc-testing: updated ife test cases
Roman Mashak [Sat, 21 Apr 2018 03:56:51 +0000 (23:56 -0400)]
tc-testing: updated ife test cases

Signed-off-by: Roman Mashak <mrv@mojatatu.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet: introduce a new tracepoint for tcp_rcv_space_adjust
Yafang Shao [Fri, 20 Apr 2018 15:18:26 +0000 (23:18 +0800)]
net: introduce a new tracepoint for tcp_rcv_space_adjust

tcp_rcv_space_adjust is called every time data is copied to user space,
introducing a tcp tracepoint for which could show us when the packet is
copied to user.

When a tcp packet arrives, tcp_rcv_established() will be called and with
the existed tracepoint tcp_probe we could get the time when this packet
arrives.
Then this packet will be copied to user, and tcp_rcv_space_adjust will
be called and with this new introduced tracepoint we could get the time
when this packet is copied to user.
With these two tracepoints, we could figure out whether the user program
processes this packet immediately or there's latency.

Hence in the printk message, sk_cookie is printed as a key to relate
tcp_rcv_space_adjust with tcp_probe.

Maybe we could export sockfd in this new tracepoint as well, then we
could relate this new tracepoint with epoll/read/recv* tracepoints, and
finally that could show us the whole lifespan of this packet. But we
could also implement that with pid as these functions are executed in
process context.

Signed-off-by: Yafang Shao <laoar.shao@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agotcp: don't read out-of-bounds opsize
Jann Horn [Fri, 20 Apr 2018 13:57:30 +0000 (15:57 +0200)]
tcp: don't read out-of-bounds opsize

The old code reads the "opsize" variable from out-of-bounds memory (first
byte behind the segment) if a broken TCP segment ends directly after an
opcode that is neither EOL nor NOP.

The result of the read isn't used for anything, so the worst thing that
could theoretically happen is a pagefault; and since the physmap is usually
mostly contiguous, even that seems pretty unlikely.

The following C reproducer triggers the uninitialized read - however, you
can't actually see anything happen unless you put something like a
pr_warn() in tcp_parse_md5sig_option() to print the opsize.

====================================
#define _GNU_SOURCE
#include <arpa/inet.h>
#include <stdlib.h>
#include <errno.h>
#include <stdarg.h>
#include <net/if.h>
#include <linux/if.h>
#include <linux/ip.h>
#include <linux/tcp.h>
#include <linux/in.h>
#include <linux/if_tun.h>
#include <err.h>
#include <sys/types.h>
#include <sys/stat.h>
#include <fcntl.h>
#include <string.h>
#include <stdio.h>
#include <unistd.h>
#include <sys/ioctl.h>
#include <assert.h>

void systemf(const char *command, ...) {
  char *full_command;
  va_list ap;
  va_start(ap, command);
  if (vasprintf(&full_command, command, ap) == -1)
    err(1, "vasprintf");
  va_end(ap);
  printf("systemf: <<<%s>>>\n", full_command);
  system(full_command);
}

char *devname;

int tun_alloc(char *name) {
  int fd = open("/dev/net/tun", O_RDWR);
  if (fd == -1)
    err(1, "open tun dev");
  static struct ifreq req = { .ifr_flags = IFF_TUN|IFF_NO_PI };
  strcpy(req.ifr_name, name);
  if (ioctl(fd, TUNSETIFF, &req))
    err(1, "TUNSETIFF");
  devname = req.ifr_name;
  printf("device name: %s\n", devname);
  return fd;
}

#define IPADDR(a,b,c,d) (((a)<<0)+((b)<<8)+((c)<<16)+((d)<<24))

void sum_accumulate(unsigned int *sum, void *data, int len) {
  assert((len&2)==0);
  for (int i=0; i<len/2; i++) {
    *sum += ntohs(((unsigned short *)data)[i]);
  }
}

unsigned short sum_final(unsigned int sum) {
  sum = (sum >> 16) + (sum & 0xffff);
  sum = (sum >> 16) + (sum & 0xffff);
  return htons(~sum);
}

void fix_ip_sum(struct iphdr *ip) {
  unsigned int sum = 0;
  sum_accumulate(&sum, ip, sizeof(*ip));
  ip->check = sum_final(sum);
}

void fix_tcp_sum(struct iphdr *ip, struct tcphdr *tcp) {
  unsigned int sum = 0;
  struct {
    unsigned int saddr;
    unsigned int daddr;
    unsigned char pad;
    unsigned char proto_num;
    unsigned short tcp_len;
  } fakehdr = {
    .saddr = ip->saddr,
    .daddr = ip->daddr,
    .proto_num = ip->protocol,
    .tcp_len = htons(ntohs(ip->tot_len) - ip->ihl*4)
  };
  sum_accumulate(&sum, &fakehdr, sizeof(fakehdr));
  sum_accumulate(&sum, tcp, tcp->doff*4);
  tcp->check = sum_final(sum);
}

int main(void) {
  int tun_fd = tun_alloc("inject_dev%d");
  systemf("ip link set %s up", devname);
  systemf("ip addr add 192.168.42.1/24 dev %s", devname);

  struct {
    struct iphdr ip;
    struct tcphdr tcp;
    unsigned char tcp_opts[20];
  } __attribute__((packed)) syn_packet = {
    .ip = {
      .ihl = sizeof(struct iphdr)/4,
      .version = 4,
      .tot_len = htons(sizeof(syn_packet)),
      .ttl = 30,
      .protocol = IPPROTO_TCP,
      /* FIXUP check */
      .saddr = IPADDR(192,168,42,2),
      .daddr = IPADDR(192,168,42,1)
    },
    .tcp = {
      .source = htons(1),
      .dest = htons(1337),
      .seq = 0x12345678,
      .doff = (sizeof(syn_packet.tcp)+sizeof(syn_packet.tcp_opts))/4,
      .syn = 1,
      .window = htons(64),
      .check = 0 /*FIXUP*/
    },
    .tcp_opts = {
      /* INVALID: trailing MD5SIG opcode after NOPs */
      1, 1, 1, 1, 1,
      1, 1, 1, 1, 1,
      1, 1, 1, 1, 1,
      1, 1, 1, 1, 19
    }
  };
  fix_ip_sum(&syn_packet.ip);
  fix_tcp_sum(&syn_packet.ip, &syn_packet.tcp);
  while (1) {
    int write_res = write(tun_fd, &syn_packet, sizeof(syn_packet));
    if (write_res != sizeof(syn_packet))
      err(1, "packet write failed");
  }
}
====================================

Fixes: cfb6eeb4c860 ("[TCP]: MD5 Signature Option (RFC2385) support.")
Signed-off-by: Jann Horn <jannh@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agoLinux 4.17-rc2
Linus Torvalds [Mon, 23 Apr 2018 02:20:09 +0000 (19:20 -0700)]
Linux 4.17-rc2

6 years agoMerge git://git.kernel.org/pub/scm/linux/kernel/git/bpf/bpf
David S. Miller [Mon, 23 Apr 2018 01:15:59 +0000 (21:15 -0400)]
Merge git://git./pub/scm/linux/kernel/git/bpf/bpf

Daniel Borkmann says:

====================
pull-request: bpf 2018-04-21

The following pull-request contains BPF updates for your *net* tree.

The main changes are:

1) Fix a deadlock between mm->mmap_sem and bpf_event_mutex when
   one task is detaching a BPF prog via perf_event_detach_bpf_prog()
   and another one dumping through bpf_prog_array_copy_info(). For
   the latter we move the copy_to_user() out of the bpf_event_mutex
   lock to fix it, from Yonghong.

2) Fix test_sock and test_sock_addr.sh failures. The former was
   hitting rlimit issues and the latter required ping to specify
   the address family, from Yonghong.

3) Remove a dead check in sockmap's sock_map_alloc(), from Jann.

4) Add generated files to BPF kselftests gitignore that were previously
   missed, from Anders.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agoibmvnic: Clean actual number of RX or TX pools
Thomas Falcon [Fri, 20 Apr 2018 19:25:32 +0000 (14:25 -0500)]
ibmvnic: Clean actual number of RX or TX pools

Avoid using value stored in the login response buffer when
cleaning TX and RX buffer pools since these could be inconsistent
depending on the device state. Instead use the field in the driver's
private data that tracks the number of active pools.

Signed-off-by: Thomas Falcon <tlfalcon@linux.vnet.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agoMerge branch 'net-sched-ife-malformed-ife-packet-fixes'
David S. Miller [Mon, 23 Apr 2018 01:12:00 +0000 (21:12 -0400)]
Merge branch 'net-sched-ife-malformed-ife-packet-fixes'

Alexander Aring says:

====================
net: sched: ife: malformed ife packet fixes

As promised at netdev 2.2 tc workshop I am working on adding scapy support for
tdc testing. It is still work in progress. I will submit the patches to tdc
later (they are not in good shape yet). The good news is I have been able to
find bugs which normal packet testing would not be able to find.
With fuzzy testing I was able to craft certain malformed packets that IFE
action was not able to deal with. This patch set fixes those bugs.

changes since v4:
 - use pskb_may_pull before pointer assign

changes since v3:
 - use pskb_may_pull

changes since v2:
 - remove inline from __ife_tlv_meta_valid
 - add const to cast to meta_tlvhdr
 - add acked and reviewed tags
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet: sched: ife: check on metadata length
Alexander Aring [Fri, 20 Apr 2018 19:15:05 +0000 (15:15 -0400)]
net: sched: ife: check on metadata length

This patch checks if sk buffer is available to dererence ife header. If
not then NULL will returned to signal an malformed ife packet. This
avoids to crashing the kernel from outside.

Signed-off-by: Alexander Aring <aring@mojatatu.com>
Reviewed-by: Yotam Gigi <yotam.gi@gmail.com>
Acked-by: Jamal Hadi Salim <jhs@mojatatu.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet: sched: ife: handle malformed tlv length
Alexander Aring [Fri, 20 Apr 2018 19:15:04 +0000 (15:15 -0400)]
net: sched: ife: handle malformed tlv length

There is currently no handling to check on a invalid tlv length. This
patch adds such handling to avoid killing the kernel with a malformed
ife packet.

Signed-off-by: Alexander Aring <aring@mojatatu.com>
Reviewed-by: Yotam Gigi <yotam.gi@gmail.com>
Acked-by: Jamal Hadi Salim <jhs@mojatatu.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet: sched: ife: signal not finding metaid
Alexander Aring [Fri, 20 Apr 2018 19:15:03 +0000 (15:15 -0400)]
net: sched: ife: signal not finding metaid

We need to record stats for received metadata that we dont know how
to process. Have find_decode_metaid() return -ENOENT to capture this.

Signed-off-by: Alexander Aring <aring@mojatatu.com>
Reviewed-by: Yotam Gigi <yotam.gi@gmail.com>
Acked-by: Jamal Hadi Salim <jhs@mojatatu.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agostrparser: Do not call mod_delayed_work with a timeout of LONG_MAX
Doron Roberts-Kedes [Fri, 20 Apr 2018 19:11:11 +0000 (12:11 -0700)]
strparser: Do not call mod_delayed_work with a timeout of LONG_MAX

struct sock's sk_rcvtimeo is initialized to
LONG_MAX/MAX_SCHEDULE_TIMEOUT in sock_init_data. Calling
mod_delayed_work with a timeout of LONG_MAX causes spurious execution of
the work function. timer->expires is set equal to jiffies + LONG_MAX.
When timer_base->clk falls behind the current value of jiffies,
the delta between timer_base->clk and jiffies + LONG_MAX causes the
expiration to be in the past. Returning early from strp_start_timer if
timeo == LONG_MAX solves this problem.

Found while testing net/tls_sw recv path.

Fixes: 43a0c6751a322847 ("strparser: Stream parser for messages")
Reviewed-by: Tejun Heo <tj@kernel.org>
Signed-off-by: Doron Roberts-Kedes <doronrk@fb.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agohv_netvsc: select needed ucs2_string routine
Stephen Hemminger [Fri, 20 Apr 2018 15:48:47 +0000 (08:48 -0700)]
hv_netvsc: select needed ucs2_string routine

The conversion of rndis friendly name to utf8 uses a standard
kernel routine which is optional in config. Therefore build
would fail for some configurations. Resolve by selecting needed
library.

Fixes: 0fe554a46a0f ("hv_netvsc: propogate Hyper-V friendly name into interface alias")
Signed-off-by: Stephen Hemminger <sthemmin@microsoft.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agoipv6: sr: fix NULL pointer dereference in seg6_do_srh_encap()- v4 pkts
Ahmed Abdelsalam [Fri, 20 Apr 2018 13:58:05 +0000 (15:58 +0200)]
ipv6: sr: fix NULL pointer dereference in seg6_do_srh_encap()- v4 pkts

In case of seg6 in encap mode, seg6_do_srh_encap() calls set_tun_src()
in order to set the src addr of outer IPv6 header.

The net_device is required for set_tun_src(). However calling ip6_dst_idev()
on dst_entry in case of IPv4 traffic results on the following bug.

Using just dst->dev should fix this BUG.

[  196.242461] BUG: unable to handle kernel NULL pointer dereference at 0000000000000000
[  196.242975] PGD 800000010f076067 P4D 800000010f076067 PUD 10f060067 PMD 0
[  196.243329] Oops: 0000 [#1] SMP PTI
[  196.243468] Modules linked in: nfsd auth_rpcgss nfs_acl nfs lockd grace fscache sunrpc crct10dif_pclmul crc32_pclmul ghash_clmulni_intel pcbc aesni_intel aes_x86_64 crypto_simd cryptd input_leds glue_helper led_class pcspkr serio_raw mac_hid video autofs4 hid_generic usbhid hid e1000 i2c_piix4 ahci pata_acpi libahci
[  196.244362] CPU: 2 PID: 1089 Comm: ping Not tainted 4.16.0+ #1
[  196.244606] Hardware name: innotek GmbH VirtualBox/VirtualBox, BIOS VirtualBox 12/01/2006
[  196.244968] RIP: 0010:seg6_do_srh_encap+0x1ac/0x300
[  196.245236] RSP: 0018:ffffb2ce00b23a60 EFLAGS: 00010202
[  196.245464] RAX: 0000000000000000 RBX: ffff8c7f53eea300 RCX: 0000000000000000
[  196.245742] RDX: 0000f10000000000 RSI: ffff8c7f52085a6c RDI: ffff8c7f41166850
[  196.246018] RBP: ffffb2ce00b23aa8 R08: 00000000000261e0 R09: ffff8c7f41166800
[  196.246294] R10: ffffdce5040ac780 R11: ffff8c7f41166828 R12: ffff8c7f41166808
[  196.246570] R13: ffff8c7f52085a44 R14: ffffffffb73211c0 R15: ffff8c7e69e44200
[  196.246846] FS:  00007fc448789700(0000) GS:ffff8c7f59d00000(0000) knlGS:0000000000000000
[  196.247286] CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
[  196.247526] CR2: 0000000000000000 CR3: 000000010f05a000 CR4: 00000000000406e0
[  196.247804] Call Trace:
[  196.247972]  seg6_do_srh+0x15b/0x1c0
[  196.248156]  seg6_output+0x3c/0x220
[  196.248341]  ? prandom_u32+0x14/0x20
[  196.248526]  ? ip_idents_reserve+0x6c/0x80
[  196.248723]  ? __ip_select_ident+0x90/0x100
[  196.248923]  ? ip_append_data.part.50+0x6c/0xd0
[  196.249133]  lwtunnel_output+0x44/0x70
[  196.249328]  ip_send_skb+0x15/0x40
[  196.249515]  raw_sendmsg+0x8c3/0xac0
[  196.249701]  ? _copy_from_user+0x2e/0x60
[  196.249897]  ? rw_copy_check_uvector+0x53/0x110
[  196.250106]  ? _copy_from_user+0x2e/0x60
[  196.250299]  ? copy_msghdr_from_user+0xce/0x140
[  196.250508]  sock_sendmsg+0x36/0x40
[  196.250690]  ___sys_sendmsg+0x292/0x2a0
[  196.250881]  ? _cond_resched+0x15/0x30
[  196.251074]  ? copy_termios+0x1e/0x70
[  196.251261]  ? _copy_to_user+0x22/0x30
[  196.251575]  ? tty_mode_ioctl+0x1c3/0x4e0
[  196.251782]  ? _cond_resched+0x15/0x30
[  196.251972]  ? mutex_lock+0xe/0x30
[  196.252152]  ? vvar_fault+0xd2/0x110
[  196.252337]  ? __do_fault+0x1f/0xc0
[  196.252521]  ? __handle_mm_fault+0xc1f/0x12d0
[  196.252727]  ? __sys_sendmsg+0x63/0xa0
[  196.252919]  __sys_sendmsg+0x63/0xa0
[  196.253107]  do_syscall_64+0x72/0x200
[  196.253305]  entry_SYSCALL_64_after_hwframe+0x3d/0xa2
[  196.253530] RIP: 0033:0x7fc4480b0690
[  196.253715] RSP: 002b:00007ffde9f252f8 EFLAGS: 00000246 ORIG_RAX: 000000000000002e
[  196.254053] RAX: ffffffffffffffda RBX: 0000000000000040 RCX: 00007fc4480b0690
[  196.254331] RDX: 0000000000000000 RSI: 000000000060a360 RDI: 0000000000000003
[  196.254608] RBP: 00007ffde9f253f0 R08: 00000000002d1e81 R09: 0000000000000002
[  196.254884] R10: 00007ffde9f250c0 R11: 0000000000000246 R12: 0000000000b22070
[  196.255205] R13: 20c49ba5e353f7cf R14: 431bde82d7b634db R15: 00007ffde9f278fe
[  196.255484] Code: a5 0f b6 45 c0 41 88 41 28 41 0f b6 41 2c 48 c1 e0 04 49 8b 54 01 38 49 8b 44 01 30 49 89 51 20 49 89 41 18 48 8b 83 b0 00 00 00 <48> 8b 30 49 8b 86 08 0b 00 00 48 8b 40 20 48 8b 50 08 48 0b 10
[  196.256190] RIP: seg6_do_srh_encap+0x1ac/0x300 RSP: ffffb2ce00b23a60
[  196.256445] CR2: 0000000000000000
[  196.256676] ---[ end trace 71af7d093603885c ]---

Fixes: 8936ef7604c11 ("ipv6: sr: fix NULL pointer dereference when setting encap source address")
Signed-off-by: Ahmed Abdelsalam <amsalam20@gmail.com>
Acked-by: David Lebrun <dlebrun@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agoMerge tag 'drm-fixes-for-v4.17-rc2' of git://people.freedesktop.org/~airlied/linux
Linus Torvalds [Mon, 23 Apr 2018 00:14:29 +0000 (17:14 -0700)]
Merge tag 'drm-fixes-for-v4.17-rc2' of git://people.freedesktop.org/~airlied/linux

Pull drm fixes from Dave Airlie:
 "Exynos, i915, vc4, amdgpu fixes.

  i915:
   - an oops fix
   - two race fixes
   - some gvt fixes

  amdgpu:
   - dark screen fix
   - clk/voltage fix
   - vega12 smu fix

  vc4:
   - memory leak fix

  exynos just drops some code"

* tag 'drm-fixes-for-v4.17-rc2' of git://people.freedesktop.org/~airlied/linux: (23 commits)
  drm/amd/powerplay: header file interface to SMU update
  drm/amd/pp: Fix bug voltage can't be OD separately on VI
  drm/amd/display: Don't program bypass on linear regamma LUT
  drm/i915: Fix LSPCON TMDS output buffer enabling from low-power state
  drm/i915/audio: Fix audio detection issue on GLK
  drm/i915: Call i915_perf_fini() on init_hw error unwind
  drm/i915/bios: filter out invalid DDC pins from VBT child devices
  drm/i915/pmu: Inspect runtime PM state more carefully while estimating RC6
  drm/i915: Do no use kfree() to free a kmem_cache_alloc() return value
  drm/exynos: exynos_drm_fb -> drm_framebuffer
  drm/exynos: Move dma_addr out of exynos_drm_fb
  drm/exynos: Move GEM BOs to drm_framebuffer
  drm: Fix HDCP downstream dev count read
  drm/vc4: Fix memory leak during BO teardown
  drm/i915/execlists: Clear user-active flag on preemption completion
  drm/i915/gvt: Add drm_format_mod update
  drm/i915/gvt: Disable primary/sprite/cursor plane at virtual display initialization
  drm/i915/gvt: Delete redundant error message in fb_decode.c
  drm/i915/gvt: Cancel dma map when resetting ggtt entries
  drm/i915/gvt: Missed to cancel dma map for ggtt entries
  ...

6 years agoMerge branch 'drm-next-4.17' of git://people.freedesktop.org/~agd5f/linux into drm...
Dave Airlie [Sun, 22 Apr 2018 22:54:06 +0000 (08:54 +1000)]
Merge branch 'drm-next-4.17' of git://people.freedesktop.org/~agd5f/linux into drm-next

- Fix a dark screen issue in DC
- Fix clk/voltage dependency tracking for wattman
- Update SMU interface for vega12

* 'drm-next-4.17' of git://people.freedesktop.org/~agd5f/linux:
  drm/amd/powerplay: header file interface to SMU update
  drm/amd/pp: Fix bug voltage can't be OD separately on VI
  drm/amd/display: Don't program bypass on linear regamma LUT

6 years agoMerge tag 'exynos-drm-fixes-for-v4.17-rc2' of git://git.kernel.org/pub/scm/linux...
Dave Airlie [Sun, 22 Apr 2018 22:53:41 +0000 (08:53 +1000)]
Merge tag 'exynos-drm-fixes-for-v4.17-rc2' of git://git./linux/kernel/git/daeinki/drm-exynos into drm-next

Remove Exynos specific framebuffer structure and
relevant functions.
- it removes exynos_drm_fb structure which is a wrapper of
  drm_framebuffer and unnecessary two exynos specific callback
  functions, exynos_drm_destory() and exynos_drm_fb_create_handle()
  because we can reuse existing drm common callback ones instead.

* tag 'exynos-drm-fixes-for-v4.17-rc2' of git://git.kernel.org/pub/scm/linux/kernel/git/daeinki/drm-exynos:
  drm/exynos: exynos_drm_fb -> drm_framebuffer
  drm/exynos: Move dma_addr out of exynos_drm_fb
  drm/exynos: Move GEM BOs to drm_framebuffer
  drm/amdkfd: Deallocate SDMA queues correctly
  drm/amdkfd: Fix scratch memory with HWS enabled

6 years agoMerge tag 'drm-intel-next-fixes-2018-04-19' of git://anongit.freedesktop.org/drm...
Dave Airlie [Sun, 22 Apr 2018 22:53:27 +0000 (08:53 +1000)]
Merge tag 'drm-intel-next-fixes-2018-04-19' of git://anongit.freedesktop.org/drm/drm-intel into drm-next

- Fix for FDO #105549: Avoid OOPS on bad VBT (Jani)
- Fix rare pre-emption race (Chris)
- Fix RC6 race against PM transitions (Tvrtko)

* tag 'drm-intel-next-fixes-2018-04-19' of git://anongit.freedesktop.org/drm/drm-intel:
  drm/i915/audio: Fix audio detection issue on GLK
  drm/i915: Call i915_perf_fini() on init_hw error unwind
  drm/i915/bios: filter out invalid DDC pins from VBT child devices
  drm/i915/pmu: Inspect runtime PM state more carefully while estimating RC6
  drm/i915: Do no use kfree() to free a kmem_cache_alloc() return value
  drm/i915/execlists: Clear user-active flag on preemption completion
  drm/i915/gvt: Add drm_format_mod update
  drm/i915/gvt: Disable primary/sprite/cursor plane at virtual display initialization
  drm/i915/gvt: Delete redundant error message in fb_decode.c
  drm/i915/gvt: Cancel dma map when resetting ggtt entries
  drm/i915/gvt: Missed to cancel dma map for ggtt entries
  drm/i915/gvt: Make MI_USER_INTERRUPT nop in cmd parser
  drm/i915/gvt: Mark expected switch fall-through in handle_g2v_notification
  drm/i915/gvt: throw error on unhandled vfio ioctls

6 years agoMerge tag 'drm-misc-fixes-2018-04-18-1' of git://anongit.freedesktop.org/drm/drm...
Dave Airlie [Sun, 22 Apr 2018 22:52:54 +0000 (08:52 +1000)]
Merge tag 'drm-misc-fixes-2018-04-18-1' of git://anongit.freedesktop.org/drm/drm-misc into drm-next

drm-misc-fixes:

stable: vc4: Fix memory leak during BO teardown (Daniel)
dp: Add i2c retry for LSPCON adapters (Imre)
hdcp: Fix device count mask (Ramalingam)

Cc: Daniel J Blueman <daniel@quora.org
Cc: Imre Deak <imre.deak@intel.com>
Cc: Ramalingam C <ramalingam.c@intel.com>
* tag 'drm-misc-fixes-2018-04-18-1' of git://anongit.freedesktop.org/drm/drm-misc:
  drm/i915: Fix LSPCON TMDS output buffer enabling from low-power state
  drm: Fix HDCP downstream dev count read
  drm/vc4: Fix memory leak during BO teardown

6 years agoMerge tag '4.17-rc1-SMB3-CIFS' of git://git.samba.org/sfrench/cifs-2.6
Linus Torvalds [Sun, 22 Apr 2018 19:13:04 +0000 (12:13 -0700)]
Merge tag '4.17-rc1-SMB3-CIFS' of git://git.samba.org/sfrench/cifs-2.6

Pull cifs fixes from Steve French:
 "Various SMB3/CIFS fixes.

  There are three more security related fixes in progress that are not
  included in this set but they are still being tested and reviewed, so
  sending this unrelated set of smaller fixes now"

* tag '4.17-rc1-SMB3-CIFS' of git://git.samba.org/sfrench/cifs-2.6:
  CIFS: fix typo in cifs_dbg
  cifs: do not allow creating sockets except with SMB1 posix exensions
  cifs: smbd: Dump SMB packet when configured
  cifs: smbd: Check for iov length on sending the last iov
  fs: cifs: Adding new return type vm_fault_t
  cifs: smb2ops: Fix NULL check in smb2_query_symlink

6 years agoMerge tag 'for-4.17-rc1-tag' of git://git.kernel.org/pub/scm/linux/kernel/git/kdave...
Linus Torvalds [Sun, 22 Apr 2018 19:09:27 +0000 (12:09 -0700)]
Merge tag 'for-4.17-rc1-tag' of git://git./linux/kernel/git/kdave/linux

Pull btrfs fixes from David Sterba:
 "This contains a few fixups to the qgroup patches that were merged this
  dev cycle, unaligned access fix, blockgroup removal corner case fix
  and a small debugging output tweak"

* tag 'for-4.17-rc1-tag' of git://git.kernel.org/pub/scm/linux/kernel/git/kdave/linux:
  btrfs: print-tree: debugging output enhancement
  btrfs: Fix race condition between delayed refs and blockgroup removal
  btrfs: fix unaligned access in readdir
  btrfs: Fix wrong btrfs_delalloc_release_extents parameter
  btrfs: delayed-inode: Remove wrong qgroup meta reservation calls
  btrfs: qgroup: Use independent and accurate per inode qgroup rsv
  btrfs: qgroup: Commit transaction in advance to reduce early EDQUOT

6 years agollc: fix NULL pointer deref for SOCK_ZAPPED
Cong Wang [Fri, 20 Apr 2018 04:54:34 +0000 (21:54 -0700)]
llc: fix NULL pointer deref for SOCK_ZAPPED

For SOCK_ZAPPED socket, we don't need to care about llc->sap,
so we should just skip these refcount functions in this case.

Fixes: f7e43672683b ("llc: hold llc_sap before release_sock()")
Reported-by: kernel test robot <lkp@intel.com>
Signed-off-by: Cong Wang <xiyou.wangcong@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet: ethernet: ti: cpsw: fix tx vlan priority mapping
Ivan Khoronzhuk [Thu, 19 Apr 2018 19:49:09 +0000 (22:49 +0300)]
net: ethernet: ti: cpsw: fix tx vlan priority mapping

The CPDMA_TX_PRIORITY_MAP in real is vlan pcp field priority mapping
register and basically replaces vlan pcp field for tagged packets.
So, set it to be 1:1 mapping. Otherwise, it will cause unexpected
change of egress vlan tagged packets, like prio 2 -> prio 5.

Fixes: e05107e6b747 ("net: ethernet: ti: cpsw: add multi queue support")
Reviewed-by: Grygorii Strashko <grygorii.strashko@ti.com>
Signed-off-by: Ivan Khoronzhuk <ivan.khoronzhuk@linaro.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agollc: delete timers synchronously in llc_sk_free()
Cong Wang [Thu, 19 Apr 2018 19:25:38 +0000 (12:25 -0700)]
llc: delete timers synchronously in llc_sk_free()

The connection timers of an llc sock could be still flying
after we delete them in llc_sk_free(), and even possibly
after we free the sock. We could just wait synchronously
here in case of troubles.

Note, I leave other call paths as they are, since they may
not have to wait, at least we can change them to synchronously
when needed.

Also, move the code to net/llc/llc_conn.c, which is apparently
a better place.

Reported-by: <syzbot+f922284c18ea23a8e457@syzkaller.appspotmail.com>
Signed-off-by: Cong Wang <xiyou.wangcong@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agol2tp: fix {pppol2tp, l2tp_dfs}_seq_stop() in case of seq_file overflow
Guillaume Nault [Thu, 19 Apr 2018 14:20:48 +0000 (16:20 +0200)]
l2tp: fix {pppol2tp, l2tp_dfs}_seq_stop() in case of seq_file overflow

Commit 0e0c3fee3a59 ("l2tp: hold reference on tunnels printed in pppol2tp proc file")
assumed that if pppol2tp_seq_stop() was called with non-NULL private
data (the 'v' pointer), then pppol2tp_seq_start() would not be called
again. It turns out that this isn't guaranteed, and overflowing the
seq_file's buffer in pppol2tp_seq_show() is a way to get into this
situation.

Therefore, pppol2tp_seq_stop() needs to reset pd->tunnel, so that
pppol2tp_seq_start() won't drop a reference again if it gets called.
We also have to clear pd->session, because the rest of the code expects
a non-NULL tunnel when pd->session is set.

The l2tp_debugfs module has the same issue. Fix it in the same way.

Fixes: 0e0c3fee3a59 ("l2tp: hold reference on tunnels printed in pppol2tp proc file")
Fixes: f726214d9b23 ("l2tp: hold reference on tunnels printed in l2tp/tunnels debugfs file")
Signed-off-by: Guillaume Nault <g.nault@alphalink.fr>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agoMerge branch 's390-qeth-fixes'
David S. Miller [Sun, 22 Apr 2018 18:42:32 +0000 (14:42 -0400)]
Merge branch 's390-qeth-fixes'

Julian Wiedmann says:

====================
s390/qeth: fixes 2018-04-19

Please apply the following qeth fixes for 4.17. The common theme
seems to be error handling improvements in various areas of cmd IO.

Patches 1-3 should also go back to stable.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agos390/qeth: use Read device to query hypervisor for MAC
Julian Wiedmann [Thu, 19 Apr 2018 10:52:11 +0000 (12:52 +0200)]
s390/qeth: use Read device to query hypervisor for MAC

For z/VM NICs, qeth needs to consider which of the three CCW devices in
an MPC group it uses for requesting a managed MAC address.

On the Base device, the hypervisor returns a default MAC which is
pre-assigned when creating the NIC (this MAC is also returned by the
READ MAC primitive). Querying any other device results in the allocation
of an additional MAC address.

For consistency with READ MAC and to avoid using up more addresses than
necessary, it is preferable to use the NIC's default MAC. So switch the
the diag26c over to using a NIC's Read device, which should always be
identical to the Base device.

Fixes: ec61bd2fd2a2 ("s390/qeth: use diag26c to get MAC address on L2")
Signed-off-by: Julian Wiedmann <jwi@linux.vnet.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agos390/qeth: fix request-side race during cmd IO timeout
Julian Wiedmann [Thu, 19 Apr 2018 10:52:10 +0000 (12:52 +0200)]
s390/qeth: fix request-side race during cmd IO timeout

Submitting a cmd IO request (usually on the WRITE device, but for IDX
also on the READ device) is currently done with ccw_device_start()
and a manual timeout in the caller.
On timeout, the caller cleans up the related resources (eg. IO buffer).
But 1) the IO might still be active and utilize those resources, and
    2) when the IO completes, qeth_irq() will attempt to clean up the
       same resources again.

Instead of introducing additional resource locking, switch to
ccw_device_start_timeout() to ensure IO termination after timeout, and
let the IRQ handler alone deal with cleaning up after a request.

This also removes a stray write->irq_pending reset from
clear_ipacmd_list(). The routine doesn't terminate any pending IO on
the WRITE device, so this should be handled properly via IO timeout
in the IRQ handler.

Signed-off-by: Julian Wiedmann <jwi@linux.vnet.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agos390/qeth: fix MAC address update sequence
Julian Wiedmann [Thu, 19 Apr 2018 10:52:09 +0000 (12:52 +0200)]
s390/qeth: fix MAC address update sequence

When changing the MAC address on a L2 qeth device, current code first
unregisters the old address, then registers the new one.
If HW rejects the new address (or the IO fails), the device ends up with
no operable address at all.

Re-order the code flow so that the old address only gets dropped if the
new address was registered successfully. While at it, add logic to catch
some corner-cases.

Signed-off-by: Julian Wiedmann <jwi@linux.vnet.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agos390/qeth: handle failure on workqueue creation
Julian Wiedmann [Thu, 19 Apr 2018 10:52:08 +0000 (12:52 +0200)]
s390/qeth: handle failure on workqueue creation

Creating the global workqueue during driver init may fail, deal with it.
Also, destroy the created workqueue on any subsequent error.

Fixes: 0f54761d167f ("qeth: Support VEPA mode")
Signed-off-by: Julian Wiedmann <jwi@linux.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agos390/qeth: avoid control IO completion stalls
Julian Wiedmann [Thu, 19 Apr 2018 10:52:07 +0000 (12:52 +0200)]
s390/qeth: avoid control IO completion stalls

For control IO, qeth currently tracks the index of the buffer that it
expects to complete the next IO on each qeth_channel. If the channel
presents an IRQ while this buffer has not yet completed, no completion
processing for _any_ completed buffer takes place.
So if the 'next buffer' is skipped for any sort of reason* (eg. when it
is released due to error conditions, before the IO is started), the
buffer obviously won't switch to PROCESSED until it is eventually
allocated for a _different_ IO and completes.
Until this happens, all completion processing on that channel stalls
and pending requests possibly time out.

As a fix, remove the whole 'next buffer' logic and simply process any
IO buffer right when it completes. A channel will never have more than
one IO pending, so there's no risk of processing out-of-sequence.

*Note: currently just one location in the code really handles this problem,
       by advancing the 'next' index manually.

Signed-off-by: Julian Wiedmann <jwi@linux.vnet.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agos390/qeth: fix error handling in adapter command callbacks
Julian Wiedmann [Thu, 19 Apr 2018 10:52:06 +0000 (12:52 +0200)]
s390/qeth: fix error handling in adapter command callbacks

Make sure to check both return code fields before(!) processing the
command response. Otherwise we risk operating on invalid data.

This matches an earlier fix for SETASSPARMS commands, see
commit ad3cbf613329 ("s390/qeth: fix error handling in checksum cmd callback").

Signed-off-by: Julian Wiedmann <jwi@linux.vnet.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agoMerge branch 'x86-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel...
Linus Torvalds [Sun, 22 Apr 2018 18:40:52 +0000 (11:40 -0700)]
Merge branch 'x86-urgent-for-linus' of git://git./linux/kernel/git/tip/tip

Pull x86 fixes from Thomas Gleixner:
 "A small set of fixes for x86:

   - Prevent X2APIC ID 0xFFFFFFFF from being treated as valid, which
     causes the possible CPU count to be wrong.

   - Prevent 32bit truncation in calc_hpet_ref() which causes the TSC
     calibration to fail

   - Fix the page table setup for temporary text mappings in the resume
     code which causes resume failures

   - Make the page table dump code handle HIGHPTE correctly instead of
     oopsing

   - Support for topologies where NUMA nodes share an LLC to prevent a
     invalid topology warning and further malfunction on such systems.

   - Remove the now unused pci-nommu code

   - Remove stale function declarations"

* 'x86-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
  x86/power/64: Fix page-table setup for temporary text mapping
  x86/mm: Prevent kernel Oops in PTDUMP code with HIGHPTE=y
  x86,sched: Allow topologies where NUMA nodes share an LLC
  x86/processor: Remove two unused function declarations
  x86/acpi: Prevent X2APIC id 0xffffffff from being accounted
  x86/tsc: Prevent 32bit truncation in calc_hpet_ref()
  x86: Remove pci-nommu.c