platform/kernel/linux-rpi.git
5 years agobpfilter: remove extra header search paths for bpfilter_umh
Masahiro Yamada [Thu, 31 Jan 2019 03:15:35 +0000 (12:15 +0900)]
bpfilter: remove extra header search paths for bpfilter_umh

Currently, the header search paths -Itools/include and
-Itools/include/uapi are not used. Let's drop the unused code.

We can remove -I. too by fixing up one C file.

Signed-off-by: Masahiro Yamada <yamada.masahiro@socionext.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoMerge branch 'phy-aquantia-improvements'
David S. Miller [Sun, 3 Feb 2019 22:31:04 +0000 (14:31 -0800)]
Merge branch 'phy-aquantia-improvements'

Heiner Kallweit says:

====================
net: phy: aquantia: number of improvements

This patch series is based on work from Andrew. I adjusted and added
certain parts. The series improves few aspects of driver, no functional
change intended.

v2:
- add my SoB to patch 1
- leave kernel.h in in patch 2
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: phy: aquantia: replace magic numbers with constants
Heiner Kallweit [Sun, 3 Feb 2019 20:19:06 +0000 (21:19 +0100)]
net: phy: aquantia: replace magic numbers with constants

Replace magic numbers with proper constants. The original patch is
from Andrew, I extended / adjusted certain parts:
- Use decimal bit numbers. The datasheet uses hex bit numbers 0 .. F.
- Order defines from highest to lowest bit numbers
- correct some typos
- add constant MDIO_AN_TX_VEND_INT_MASK2_LINK
- Remove few functional improvements from the patch, they will come as
  a separate patch.

Signed-off-by: Andrew Lunn <andrew@lunn.ch>
Signed-off-by: Heiner Kallweit <hkallweit1@gmail.com>
Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: phy: aquantia: use macro PHY_ID_MATCH_MODEL
Heiner Kallweit [Sun, 3 Feb 2019 20:18:03 +0000 (21:18 +0100)]
net: phy: aquantia: use macro PHY_ID_MATCH_MODEL

Make use of macro PHY_ID_MATCH_MODEL to simplify the code.

Signed-off-by: Heiner Kallweit <hkallweit1@gmail.com>
Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: phy: aquantia: remove unneeded includes
Heiner Kallweit [Sun, 3 Feb 2019 20:17:14 +0000 (21:17 +0100)]
net: phy: aquantia: remove unneeded includes

Remove unneeded header includes.

v2:
- leave kernel.h in

Signed-off-by: Heiner Kallweit <hkallweit1@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: phy: aquantia: Shorten name space prefix to aqr_
Andrew Lunn [Sun, 3 Feb 2019 20:16:18 +0000 (21:16 +0100)]
net: phy: aquantia: Shorten name space prefix to aqr_

aquantia_ as a name space prefix is rather long, resulting in lots of
lines needing wrapping, reducing readability. Use the prefix aqr_
instead, which fits with the vendor naming there devices aqr107, for
example.

v2:
- add SoB from Heiner

Signed-off-by: Andrew Lunn <andrew@lunn.ch>
Signed-off-by: Heiner Kallweit <hkallweit1@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: Fix ip_mc_{dec,inc}_group allocation context
Florian Fainelli [Sat, 2 Feb 2019 04:20:52 +0000 (20:20 -0800)]
net: Fix ip_mc_{dec,inc}_group allocation context

After 4effd28c1245 ("bridge: join all-snoopers multicast address"), I
started seeing the following sleep in atomic warnings:

[   26.763893] BUG: sleeping function called from invalid context at mm/slab.h:421
[   26.771425] in_atomic(): 1, irqs_disabled(): 0, pid: 1658, name: sh
[   26.777855] INFO: lockdep is turned off.
[   26.781916] CPU: 0 PID: 1658 Comm: sh Not tainted 5.0.0-rc4 #20
[   26.787943] Hardware name: BCM97278SV (DT)
[   26.792118] Call trace:
[   26.794645]  dump_backtrace+0x0/0x170
[   26.798391]  show_stack+0x24/0x30
[   26.801787]  dump_stack+0xa4/0xe4
[   26.805182]  ___might_sleep+0x208/0x218
[   26.809102]  __might_sleep+0x78/0x88
[   26.812762]  kmem_cache_alloc_trace+0x64/0x28c
[   26.817301]  igmp_group_dropped+0x150/0x230
[   26.821573]  ip_mc_dec_group+0x1b0/0x1f8
[   26.825585]  br_ip4_multicast_leave_snoopers.isra.11+0x174/0x190
[   26.831704]  br_multicast_toggle+0x78/0xcc
[   26.835887]  store_bridge_parm+0xc4/0xfc
[   26.839894]  multicast_snooping_store+0x3c/0x4c
[   26.844517]  dev_attr_store+0x44/0x5c
[   26.848262]  sysfs_kf_write+0x50/0x68
[   26.852006]  kernfs_fop_write+0x14c/0x1b4
[   26.856102]  __vfs_write+0x60/0x190
[   26.859668]  vfs_write+0xc8/0x168
[   26.863059]  ksys_write+0x70/0xc8
[   26.866449]  __arm64_sys_write+0x24/0x30
[   26.870458]  el0_svc_common+0xa0/0x11c
[   26.874291]  el0_svc_handler+0x38/0x70
[   26.878120]  el0_svc+0x8/0xc

while toggling the bridge's multicast_snooping attribute dynamically.

Pass a gfp_t down to igmpv3_add_delrec(), introduce
__igmp_group_dropped() and introduce __ip_mc_dec_group() to take a gfp_t
argument.

Similarly introduce ____ip_mc_inc_group() and __ip_mc_inc_group() to
allow caller to specify gfp_t.

IPv6 part of the patch appears fine.

Fixes: 4effd28c1245 ("bridge: join all-snoopers multicast address")
Signed-off-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: devlink: report cell size of shared buffers
Jakub Kicinski [Sat, 2 Feb 2019 01:56:28 +0000 (17:56 -0800)]
net: devlink: report cell size of shared buffers

Shared buffer allocation is usually done in cell increments.
Drivers will either round up the allocation or refuse the
configuration if it's not an exact multiple of cell size.
Drivers know exactly the cell size of shared buffer, so help
out users by providing this information in dumps.

Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Reviewed-by: Dirk van der Merwe <dirk.vandermerwe@netronome.com>
Reviewed-by: Ido Schimmel <idosch@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoMerge branch 'net-y2038-safe-socket-timestamps'
David S. Miller [Sun, 3 Feb 2019 19:17:31 +0000 (11:17 -0800)]
Merge branch 'net-y2038-safe-socket-timestamps'

Deepa Dinamani says:

====================
net: y2038-safe socket timestamps

The series introduces new socket timestamps that are
y2038 safe.

The time data types used for the existing socket timestamp
options: SO_TIMESTAMP, SO_TIMESTAMPNS and SO_TIMESTAMPING
are not y2038 safe. The series introduces SO_TIMESTAMP_NEW,
SO_TIMESTAMPNS_NEW and SO_TIMESTAMPING_NEW to replace these.
These new timestamps can be used on all architectures.

The alternative considered was to extend the sys_setsockopt()
by using the flags. We did not receive any strong opinions about
either of the approaches. Hence, this was chosen, as glibc folks
preferred this.

The series does not deal with updating the internal kernel socket
calls like rxrpc to make them y2038 safe. This will be dealt
with separately.

Note that the timestamps behavior already does not match the
man page specific behavior:
SIOCGSTAMP
    This ioctl should only be used if the socket option SO_TIMESTAMP
is not set on the socket. Otherwise, it returns the timestamp of
the last packet that was received while SO_TIMESTAMP was not set,
or it fails if no such packet has been received,
(i.e., ioctl(2) returns -1 with errno set to ENOENT).

The recommendation is to update the man page to remove the above statement.

The overview of the socket timestamp series is as below:
1. Delete asm specific socket.h when possible.
2. Support SO/SCM_TIMESTAMP* options only in userspace.
3. Rename current SO/SCM_TIMESTAMP* to SO/SCM_TIMESTAMP*_OLD.
3. Alter socket options so that SOCK_RCVTSTAMPNS does
   not rely on SOCK_RCVTSTAMP.
4. Introduce y2038 safe types for socket timestamp.
5. Introduce new y2038 safe socket options SO/SCM_TIMESTAMP*_NEW.
6. Intorduce new y2038 safe socket timeout options.

Changes since v4:
* Fixed the typo in calling sock_get_timeout()

Changes since v3:
* Rebased onto net-next and fixups as per review comments
* Merged the socket timeout series
* Integrated Arnd's patch to simplify compat handling of timeout syscalls

Changes since v2:
* Removed extra functions to reduce diff churn as per code review

Changes since v1:
* Dropped the change to disentangle sock flags
* Renamed sock_timeval to __kernel_sock_timeval
* Updated a few comments
* Added documentation changes
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agosock: Add SO_RCVTIMEO_NEW and SO_SNDTIMEO_NEW
Deepa Dinamani [Sat, 2 Feb 2019 15:34:54 +0000 (07:34 -0800)]
sock: Add SO_RCVTIMEO_NEW and SO_SNDTIMEO_NEW

Add new socket timeout options that are y2038 safe.

Signed-off-by: Deepa Dinamani <deepa.kernel@gmail.com>
Acked-by: Willem de Bruijn <willemb@google.com>
Cc: ccaulfie@redhat.com
Cc: davem@davemloft.net
Cc: deller@gmx.de
Cc: paulus@samba.org
Cc: ralf@linux-mips.org
Cc: rth@twiddle.net
Cc: cluster-devel@redhat.com
Cc: linuxppc-dev@lists.ozlabs.org
Cc: linux-alpha@vger.kernel.org
Cc: linux-arch@vger.kernel.org
Cc: linux-mips@vger.kernel.org
Cc: linux-parisc@vger.kernel.org
Cc: sparclinux@vger.kernel.org
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agosocket: Rename SO_RCVTIMEO/ SO_SNDTIMEO with _OLD suffixes
Deepa Dinamani [Sat, 2 Feb 2019 15:34:53 +0000 (07:34 -0800)]
socket: Rename SO_RCVTIMEO/ SO_SNDTIMEO with _OLD suffixes

SO_RCVTIMEO and SO_SNDTIMEO socket options use struct timeval
as the time format. struct timeval is not y2038 safe.
The subsequent patches in the series add support for new socket
timeout options with _NEW suffix that will use y2038 safe
data structures. Although the existing struct timeval layout
is sufficiently wide to represent timeouts, because of the way
libc will interpret time_t based on user defined flag, these
new flags provide a way of having a structure that is the same
for all architectures consistently.
Rename the existing options with _OLD suffix forms so that the
right option is enabled for userspace applications according
to the architecture and time_t definition of libc.

Signed-off-by: Deepa Dinamani <deepa.kernel@gmail.com>
Acked-by: Willem de Bruijn <willemb@google.com>
Cc: ccaulfie@redhat.com
Cc: deller@gmx.de
Cc: paulus@samba.org
Cc: ralf@linux-mips.org
Cc: rth@twiddle.net
Cc: cluster-devel@redhat.com
Cc: linuxppc-dev@lists.ozlabs.org
Cc: linux-alpha@vger.kernel.org
Cc: linux-arch@vger.kernel.org
Cc: linux-mips@vger.kernel.org
Cc: linux-parisc@vger.kernel.org
Cc: sparclinux@vger.kernel.org
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agosocket: Update timestamping Documentation
Deepa Dinamani [Sat, 2 Feb 2019 15:34:52 +0000 (07:34 -0800)]
socket: Update timestamping Documentation

With the new y2038 safe timestamping options added, update the
documentation to reflect the changes.

Signed-off-by: Deepa Dinamani <deepa.kernel@gmail.com>
Acked-by: Willem de Bruijn <willemb@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agosocket: Add SO_TIMESTAMPING_NEW
Deepa Dinamani [Sat, 2 Feb 2019 15:34:51 +0000 (07:34 -0800)]
socket: Add SO_TIMESTAMPING_NEW

Add SO_TIMESTAMPING_NEW variant of socket timestamp options.
This is the y2038 safe versions of the SO_TIMESTAMPING_OLD
for all architectures.

Signed-off-by: Deepa Dinamani <deepa.kernel@gmail.com>
Acked-by: Willem de Bruijn <willemb@google.com>
Cc: chris@zankel.net
Cc: fenghua.yu@intel.com
Cc: rth@twiddle.net
Cc: tglx@linutronix.de
Cc: ubraun@linux.ibm.com
Cc: linux-alpha@vger.kernel.org
Cc: linux-arch@vger.kernel.org
Cc: linux-ia64@vger.kernel.org
Cc: linux-mips@linux-mips.org
Cc: linux-s390@vger.kernel.org
Cc: linux-xtensa@linux-xtensa.org
Cc: sparclinux@vger.kernel.org
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agosocket: Add SO_TIMESTAMP[NS]_NEW
Deepa Dinamani [Sat, 2 Feb 2019 15:34:50 +0000 (07:34 -0800)]
socket: Add SO_TIMESTAMP[NS]_NEW

Add SO_TIMESTAMP_NEW and SO_TIMESTAMPNS_NEW variants of
socket timestamp options.
These are the y2038 safe versions of the SO_TIMESTAMP_OLD
and SO_TIMESTAMPNS_OLD for all architectures.

Note that the format of scm_timestamping.ts[0] is not changed
in this patch.

Signed-off-by: Deepa Dinamani <deepa.kernel@gmail.com>
Acked-by: Willem de Bruijn <willemb@google.com>
Cc: jejb@parisc-linux.org
Cc: ralf@linux-mips.org
Cc: rth@twiddle.net
Cc: linux-alpha@vger.kernel.org
Cc: linux-mips@linux-mips.org
Cc: linux-parisc@vger.kernel.org
Cc: linux-rdma@vger.kernel.org
Cc: netdev@vger.kernel.org
Cc: sparclinux@vger.kernel.org
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agosocket: Add struct __kernel_sock_timeval
Deepa Dinamani [Sat, 2 Feb 2019 15:34:49 +0000 (07:34 -0800)]
socket: Add struct __kernel_sock_timeval

The new type is meant to be used as a y2038 safe structure
to be used as part of cmsg data.
Presently the SO_TIMESTAMP socket option uses struct timeval
for timestamps. This is not y2038 safe.
Subsequent patches in the series add new y2038 safe socket
option to be used in the place of SO_TIMESTAMP_OLD.
struct __kernel_sock_timeval will be used as the timestamp
format at that time.

struct __kernel_sock_timeval also maintains the same layout
across 32 bit and 64 bit ABIs.

Signed-off-by: Deepa Dinamani <deepa.kernel@gmail.com>
Acked-by: Willem de Bruijn <willemb@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agosocket: Use old_timeval types for socket timestamps
Deepa Dinamani [Sat, 2 Feb 2019 15:34:48 +0000 (07:34 -0800)]
socket: Use old_timeval types for socket timestamps

As part of y2038 solution, all internal uses of
struct timeval are replaced by struct __kernel_old_timeval
and struct compat_timeval by struct old_timeval32.
Make socket timestamps use these new types.

This is mainly to be able to verify that the kernel build
is y2038 safe when such non y2038 safe types are not
supported anymore.

Signed-off-by: Deepa Dinamani <deepa.kernel@gmail.com>
Acked-by: Willem de Bruijn <willemb@google.com>
Cc: isdn@linux-pingi.de
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoarch: sparc: Override struct __kernel_old_timeval
Deepa Dinamani [Sat, 2 Feb 2019 15:34:47 +0000 (07:34 -0800)]
arch: sparc: Override struct __kernel_old_timeval

struct __kernel_old_timeval is supposed to have the same
layout as struct timeval. But, it was inadvarently missed
that __kernel_suseconds has a different definition for
sparc64.
Provide an asm-specific override that fixes it.

Reported-by: Arnd Bergmann <arnd@arndb.de>
Suggested-by: Arnd Bergmann <arnd@arndb.de>
Signed-off-by: Deepa Dinamani <deepa.kernel@gmail.com>
Acked-by: Willem de Bruijn <willemb@google.com>
Cc: sparclinux@vger.kernel.org
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agosockopt: Rename SO_TIMESTAMP* to SO_TIMESTAMP*_OLD
Deepa Dinamani [Sat, 2 Feb 2019 15:34:46 +0000 (07:34 -0800)]
sockopt: Rename SO_TIMESTAMP* to SO_TIMESTAMP*_OLD

SO_TIMESTAMP, SO_TIMESTAMPNS and SO_TIMESTAMPING options, the
way they are currently defined, are not y2038 safe.
Subsequent patches in the series add new y2038 safe versions
of these options which provide 64 bit timestamps on all
architectures uniformly.
Hence, rename existing options with OLD tag suffixes.

Also note that kernel will not use the untagged SO_TIMESTAMP*
and SCM_TIMESTAMP* options internally anymore.

Signed-off-by: Deepa Dinamani <deepa.kernel@gmail.com>
Acked-by: Willem de Bruijn <willemb@google.com>
Cc: deller@gmx.de
Cc: dhowells@redhat.com
Cc: jejb@parisc-linux.org
Cc: ralf@linux-mips.org
Cc: rth@twiddle.net
Cc: linux-afs@lists.infradead.org
Cc: linux-alpha@vger.kernel.org
Cc: linux-arch@vger.kernel.org
Cc: linux-mips@linux-mips.org
Cc: linux-parisc@vger.kernel.org
Cc: linux-rdma@vger.kernel.org
Cc: sparclinux@vger.kernel.org
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoarch: Use asm-generic/socket.h when possible
Deepa Dinamani [Sat, 2 Feb 2019 15:34:45 +0000 (07:34 -0800)]
arch: Use asm-generic/socket.h when possible

Many architectures maintain an arch specific copy of the
file even though there are no differences with the asm-generic
one. Allow these architectures to use the generic one instead.

Signed-off-by: Deepa Dinamani <deepa.kernel@gmail.com>
Acked-by: Max Filippov <jcmvbkbc@gmail.com>
Acked-by: Heiko Carstens <heiko.carstens@de.ibm.com>
Acked-by: Willem de Bruijn <willemb@google.com>
Cc: chris@zankel.net
Cc: fenghua.yu@intel.com
Cc: tglx@linutronix.de
Cc: schwidefsky@de.ibm.com
Cc: linux-ia64@vger.kernel.org
Cc: linux-xtensa@linux-xtensa.org
Cc: linux-s390@vger.kernel.org
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agosocket: move compat timeout handling into sock.c
Arnd Bergmann [Sat, 2 Feb 2019 15:34:44 +0000 (07:34 -0800)]
socket: move compat timeout handling into sock.c

This is a cleanup to prepare for the addition of 64-bit time_t
in O_SNDTIMEO/O_RCVTIMEO. The existing compat handler seems
unnecessarily complex and error-prone, moving it all into the
main setsockopt()/getsockopt() implementation requires half
as much code and is easier to extend.

32-bit user space can now use old_timeval32 on both 32-bit
and 64-bit machines, while 64-bit code can use
__old_kernel_timeval.

Signed-off-by: Arnd Bergmann <arnd@arndb.de>
Signed-off-by: Deepa Dinamani <deepa.kernel@gmail.com>
Acked-by: Willem de Bruijn <willemb@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoselftests: add missing include unistd
Deepa Dinamani [Sat, 2 Feb 2019 15:34:43 +0000 (07:34 -0800)]
selftests: add missing include unistd

Compiling rxtimestamp.c generates error messages due to
non-existing declaration for write() library call.

Add missing unistd.h include to provide the declaration and
silence the error.

Signed-off-by: Deepa Dinamani <deepa.kernel@gmail.com>
Acked-by: Willem de Bruijn <willemb@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agocxgb4/cxgb4vf: Program hash region for {t4/t4vf}_change_mac()
Arjun Vynipadath [Fri, 1 Feb 2019 11:37:04 +0000 (17:07 +0530)]
cxgb4/cxgb4vf: Program hash region for {t4/t4vf}_change_mac()

{t4/t4_vf}_change_mac() API's were only doing additions to MPS_TCAM.
This will fail, when the number of tcam entries is limited particularly
in vf's.
This fix programs hash region with the mac address, when TCAM
addtion fails for {t4/t4vf}_change_mac(). Since the locally maintained
driver list for hash entries is shared across mac_{sync/unsync}(),
added an extra parameter if_mac to track the address added thorugh
{t4/t4vf}_change_mac()

Signed-off-by: Arjun Vynipadath <arjun@chelsio.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoipv4/igmp: Don't drop IGMP pkt with zeros src addr
Edward Chron [Thu, 31 Jan 2019 23:00:40 +0000 (15:00 -0800)]
ipv4/igmp: Don't drop IGMP pkt with zeros src addr

Don't drop IGMP packets with a source address of all zeros which are
IGMP proxy reports. This is documented in Section 2.1.1 IGMP
Forwarding Rules of RFC 4541 IGMP and MLD Snooping Switches
Considerations.

Signed-off-by: Edward Chron <echron@arista.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: phy: realtek: add generic Realtek PHY driver
Heiner Kallweit [Sun, 3 Feb 2019 15:07:33 +0000 (16:07 +0100)]
net: phy: realtek: add generic Realtek PHY driver

The integrated PHY's of later RTL8168 network chips report the generic
PHYID 0x001cc800 (Realtek OUI, model and revision number both set to
zero) and therefore currently the genphy driver is used.

To be able to use the paged version of e.g. phy_write() we need a
PHY driver with the read_page and write_page callbacks implemented.
So basically make a copy of the genphy driver, just with the
read_page and write_page callbacks being set.

Signed-off-by: Heiner Kallweit <hkallweit1@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoatheros: atl2: fix an indentaion issue on a return statement
Colin Ian King [Sat, 2 Feb 2019 11:45:26 +0000 (11:45 +0000)]
atheros: atl2: fix an indentaion issue on a return statement

A return statment is not indented correctly, fix this by adding an
extra tab.

Signed-off-by: Colin Ian King <colin.king@canonical.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoatl1c: fix indentation issue on an if statement
Colin Ian King [Sat, 2 Feb 2019 11:41:29 +0000 (11:41 +0000)]
atl1c: fix indentation issue on an if statement

An if statement is indented one level too deep, fix this by removing
the extra tabs. Also add some spaces to the dev_warn arguments to clean
up checkpatch warnings.

Signed-off-by: Colin Ian King <colin.king@canonical.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agobna: fix indentation issue on call to bfa_ioc_pf_failed
Colin Ian King [Sat, 2 Feb 2019 11:37:07 +0000 (11:37 +0000)]
bna: fix indentation issue on call to bfa_ioc_pf_failed

The call to bfa_ioc_pf_failed is indented too far, fix this by
removing a tab.

Signed-off-by: Colin Ian King <colin.king@canonical.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agochelsio: clean up indentation issue
Colin Ian King [Sat, 2 Feb 2019 11:33:51 +0000 (11:33 +0000)]
chelsio: clean up indentation issue

The assignment to size is indented too far, fix this and join
two lines into one.

Signed-off-by: Colin Ian King <colin.king@canonical.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: nixge: Update device-tree bindings with v3.00
Alex Williams [Thu, 31 Jan 2019 21:33:28 +0000 (13:33 -0800)]
net: nixge: Update device-tree bindings with v3.00

Now the DMA engine is free to float elsewhere in the system map.

Signed-off-by: Alex Williams <alex.williams@ni.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: nixge: Separate ctrl and dma resources
Alex Williams [Thu, 31 Jan 2019 21:33:27 +0000 (13:33 -0800)]
net: nixge: Separate ctrl and dma resources

The DMA engine is a separate entity altogether, and this allows the DMA
controller's address to float elsewhere in the FPGA's map.

Signed-off-by: Alex Williams <alex.williams@ni.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agor8169: remove rtl_wol_pll_power_down
Heiner Kallweit [Thu, 31 Jan 2019 21:03:48 +0000 (22:03 +0100)]
r8169: remove rtl_wol_pll_power_down

rtl_wol_pll_power_down() is used in only one place and removing it
makes the code simpler and better readable.

Signed-off-by: Heiner Kallweit <hkallweit1@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoMerge branch 'hns3-next'
David S. Miller [Sat, 2 Feb 2019 16:24:09 +0000 (08:24 -0800)]
Merge branch 'hns3-next'

Huazhong Tan says:

====================
code optimizations & bugfixes for HNS3 driver

This patchset includes bugfixes and code optimizations for the HNS3
ethernet controller driver
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: hns3: MAC table entry count function increases operation 0 value protection...
liuzhongzhu [Sat, 2 Feb 2019 14:39:37 +0000 (22:39 +0800)]
net: hns3: MAC table entry count function increases operation 0 value protection measures

When updating the available MAC VLAN table counts,
MAC VLAN table entry count function adds
operation 0 value protection measures.

Signed-off-by: liuzhongzhu <liuzhongzhu@huawei.com>
Signed-off-by: Peng Li <lipeng321@huawei.com>
Signed-off-by: Huazhong Tan <tanhuazhong@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: hns3: modify the upper limit judgment condition
liuzhongzhu [Sat, 2 Feb 2019 14:39:36 +0000 (22:39 +0800)]
net: hns3: modify the upper limit judgment condition

In order to prevent the variable anomaly from being larger than desc_num,
the upper limit judgment condition becomes >=.

Signed-off-by: liuzhongzhu <liuzhongzhu@huawei.com>
Signed-off-by: Peng Li <lipeng321@huawei.com>
Signed-off-by: Huazhong Tan <tanhuazhong@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: hns3: don't allow user to change vlan filter state
Jian Shen [Sat, 2 Feb 2019 14:39:35 +0000 (22:39 +0800)]
net: hns3: don't allow user to change vlan filter state

When user disables vlan filter, and adds vlan device, it won't
notify the driver the update the vlan filter. In this case, when
user enables vlan filter again, the packets with new vlan tag
will be filtered by vlan filter.

Signed-off-by: Jian Shen <shenjian15@huawei.com>
Signed-off-by: Peng Li <lipeng321@huawei.com>
Signed-off-by: Huazhong Tan <tanhuazhong@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: hns3: optimize the maximum TC macro
liuzhongzhu [Sat, 2 Feb 2019 14:39:34 +0000 (22:39 +0800)]
net: hns3: optimize the maximum TC macro

Multiple macros with the largest number of TCs in the system,
optimized to HCLGE_MAX_TC_NUM.

Signed-off-by: liuzhongzhu <liuzhongzhu@huawei.com>
Signed-off-by: Peng Li <lipeng321@huawei.com>
Signed-off-by: Huazhong Tan <tanhuazhong@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: hns3: fix the problem that the supported port is empty
liuzhongzhu [Sat, 2 Feb 2019 14:39:33 +0000 (22:39 +0800)]
net: hns3: fix the problem that the supported port is empty

Run ethtool ethx when displaying device information in VF,
the supported port and link mode items will be empty.

This patch fixes it.

Fixes: e2cb1dec9779 ("net: hns3: Add HNS3 VF HCL(Hardware Compatibility Layer) Support")
Signed-off-by: liuzhongzhu <liuzhongzhu@huawei.com>
Signed-off-by: Peng Li <lipeng321@huawei.com>
Signed-off-by: Huazhong Tan <tanhuazhong@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: hns3: fix a wrong checking in the hclge_tx_buffer_calc()
Huazhong Tan [Sat, 2 Feb 2019 14:39:32 +0000 (22:39 +0800)]
net: hns3: fix a wrong checking in the hclge_tx_buffer_calc()

Only the TC is enabled, we need to check whether the buffer is enough,
otherwise it may lead to a wrong -ENOMEM case.

Fixes: 9ffe79a9c2ee ("net: hns3: Support for dynamically assigning tx buffer to TC")
Signed-off-by: Huazhong Tan <tanhuazhong@huawei.com>
Signed-off-by: Peng Li <lipeng321@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: hns3: move some set_bit statement into hclge_prepare_mac_addr
Weihang Li [Sat, 2 Feb 2019 14:39:31 +0000 (22:39 +0800)]
net: hns3: move some set_bit statement into hclge_prepare_mac_addr

This patch does not change the code logic. There are some same
set_bit statements called by add/rm_uc/mc_addr_common, and move
this statements into hclge_prepare_mac_addr to reduce duplicate
code.

Signed-off-by: Weihang Li <liweihang@hisilicon.com>
Signed-off-by: Peng Li <lipeng321@huawei.com>
Signed-off-by: Huazhong Tan <tanhuazhong@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: hns3: add hclge_cmd_check_retval() to parse comman's return value
Weihang Li [Sat, 2 Feb 2019 14:39:30 +0000 (22:39 +0800)]
net: hns3: add hclge_cmd_check_retval() to parse comman's return value

For simplifying the code, this patch adds hclge_cmd_check_retval() to
check the return value of the command.

Also, according the IMP's description, when there are several descriptors
in a command, then the IMP will save the return value on the last
description, so hclge_cmd_check_retval() just check the last one for this
case.

Signed-off-by: Weihang Li <liweihang@hisilicon.com>
Signed-off-by: Peng Li <lipeng321@huawei.com>
Signed-off-by: Huazhong Tan <tanhuazhong@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: hns3: code optimization for hclge_rx_buffer_calc
Yunsheng Lin [Sat, 2 Feb 2019 14:39:29 +0000 (22:39 +0800)]
net: hns3: code optimization for hclge_rx_buffer_calc

There are four steps to calcuate the rx private buffer, each step
can be done in a function to avoid code duplication and aid code
readability.

This patch adds three separate functions do the job. Also, the
function name more or less make the comment redundant, so remove
some obvious comment.

Signed-off-by: Yunsheng Lin <linyunsheng@huawei.com>
Signed-off-by: Peng Li <lipeng321@huawei.com>
Signed-off-by: Huazhong Tan <tanhuazhong@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: hns3: Modify parameter type from int to bool in set_gro_en
Yonglong Liu [Sat, 2 Feb 2019 14:39:28 +0000 (22:39 +0800)]
net: hns3: Modify parameter type from int to bool in set_gro_en

The second parameter to the hook function set_gro_en is always passed in
true/false, so modify it's type from int to bool.

Signed-off-by: Yonglong Liu <liuyonglong@huawei.com>
Signed-off-by: Peng Li <lipeng321@huawei.com>
Signed-off-by: Huazhong Tan <tanhuazhong@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: hns3: fix an issue for hns3_update_new_int_gl
Peng Li [Sat, 2 Feb 2019 14:39:27 +0000 (22:39 +0800)]
net: hns3: fix an issue for hns3_update_new_int_gl

HNS3 supports setting rx-usecs|tx-usecs as 0, but it will not
update dynamically when adaptive-tx or adaptive-rx is enable.
This patch removes the Redundant check.

Fixes: a95e1f8666e9 ("net: hns3: change the time interval of int_gl calculating")
Signed-off-by: Peng Li <lipeng321@huawei.com>
Signed-off-by: Huazhong Tan <tanhuazhong@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: hns3: fix a code style issue for hns3_update_new_int_gl()
Peng Li [Sat, 2 Feb 2019 14:39:26 +0000 (22:39 +0800)]
net: hns3: fix a code style issue for hns3_update_new_int_gl()

Use the same code style for rx_group and tx_group in the
hns3_update_new_int_gl().

Signed-off-by: Peng Li <lipeng321@huawei.com>
Signed-off-by: Huazhong Tan <tanhuazhong@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoMerge git://git.kernel.org/pub/scm/linux/kernel/git/bpf/bpf-next
David S. Miller [Sat, 2 Feb 2019 04:12:18 +0000 (20:12 -0800)]
Merge git://git./linux/kernel/git/bpf/bpf-next

Alexei Starovoitov says:

====================
pull-request: bpf-next 2019-02-01

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

The main changes are:

1) introduce bpf_spin_lock, from Alexei.

2) convert xdp samples to libbpf, from Maciej.

3) skip verifier tests for unsupported program/map types, from Stanislav.

4) powerpc64 JIT support for BTF line info, from Sandipan.

5) assorted fixed, from Valdis, Jesper, Jiong.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoMerge branch 'shifts-cleanup'
Alexei Starovoitov [Sat, 2 Feb 2019 02:03:50 +0000 (18:03 -0800)]
Merge branch 'shifts-cleanup'

Jiong Wang says:

====================
NFP JIT back-end is missing several ALU32 logic shifts support.

Also, shifts with shift amount be zero are not handled properly.

This set cleans up these issues.
====================

Signed-off-by: Alexei Starovoitov <ast@kernel.org>
5 years agonfp: bpf: complete ALU32 logic shift supports
Jiong Wang [Fri, 1 Feb 2019 22:39:29 +0000 (22:39 +0000)]
nfp: bpf: complete ALU32 logic shift supports

The following ALU32 logic shift supports are missing:

  BPF_ALU | BPF_LSH | BPF_X
  BPF_ALU | BPF_RSH | BPF_X
  BPF_ALU | BPF_RSH | BPF_K

For BPF_RSH | BPF_K, it could be implemented using NFP direct shift
instruction. For the other BPF_X shifts, NFP indirect shifts sequences need
to be used.

Separate code-gen hook is assigned to each instruction to make the
implementation clear.

Reviewed-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Signed-off-by: Jiong Wang <jiong.wang@netronome.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
5 years agonfp: bpf: correct the behavior for shifts by zero
Jiong Wang [Fri, 1 Feb 2019 22:39:28 +0000 (22:39 +0000)]
nfp: bpf: correct the behavior for shifts by zero

Shifts by zero do nothing, and should be treated as nops.

Even though compiler is not supposed to generate such instructions and
manual written assembly is unlikely to have them, but they are legal
instructions and have defined behavior.

This patch correct existing shifts code-gen to make sure they do nothing
when shift amount is zero except when the instruction is ALU32 for which
high bits need to be cleared.

For shift amount bigger than type size, already, NFP JIT back-end errors
out for immediate shift and only low 5 bits will be taken into account for
indirect shift which is the same as x86.

Reviewed-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Signed-off-by: Jiong Wang <jiong.wang@netronome.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
5 years agoselftests/bpf: remove generated verifier/tests.h on 'make clean'
Stanislav Fomichev [Fri, 1 Feb 2019 23:46:38 +0000 (15:46 -0800)]
selftests/bpf: remove generated verifier/tests.h on 'make clean'

'make clean' is supposed to remove generated files.

Signed-off-by: Stanislav Fomichev <sdf@google.com>
Acked-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
5 years agoMerge branch 'devlink-add-device-driver-information-API'
David S. Miller [Fri, 1 Feb 2019 23:30:31 +0000 (15:30 -0800)]
Merge branch 'devlink-add-device-driver-information-API'

Jakub Kicinski says:

====================
devlink: add device (driver) information API

fw_version field in ethtool -i does not suit modern needs with 31
characters being quite limiting on more complex systems.  There is
also no distinction between the running and flashed versions of
the firmware.

Since the driver information pertains to the entire device, rather
than a particular netdev, it seems wise to move it do devlink, at
the same time fixing the aforementioned issues.

The new API allows exposing the device serial number and versions
of the components of the card - both hardware, firmware (running
and flashed).  Driver authors can choose descriptive identifiers
for the version fields.  A few version identifiers which seemed
relevant for most devices have been added to the global devlink
header.

Example:
$ devlink dev info pci/0000:05:00.0
pci/0000:05:00.0:
  driver nfp
  serial_number 16240145
  versions:
    fixed:
      board.id AMDA0099-0001
      board.rev 07
      board.vendor SMA
      board.model carbon
    running:
      fw.mgmt: 010156.010156.010156
      fw.cpld: 0x44
      fw.app: sriov-2.1.16
    stored:
      fw.mgmt: 010158.010158.010158
      fw.cpld: 0x44
      fw.app: sriov-2.1.20

Last patch also includes a compat code for ethtool.  If driver
reports no fw_version via the traditional ethtool API, ethtool
can call into devlink and try to cram as many versions as possible
into the 31 characters.

v4:
 - use IS_REACHABLE instead of IS_ENABLED in last patch.

v3 (Jiri):
 - rename various functions and attributes;
 - break out the version helpers per-type;
 - make the compat code parse a dump instead of special casing
   in each helper;
 - move generic version defines to a separate patch.

v2:
 - rebase.

this non-RFC, v3 some would say:
 - add three more versions in the NFP patches;
 - add last patch (ethool compat) - Andrew & Michal.

RFCv2:
 - use one driver op;
 - allow longer serial number;
 - wrap the skb into an opaque request struct;
 - add some common identifier into the devlink header.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoethtool: add compat for devlink info
Jakub Kicinski [Thu, 31 Jan 2019 18:50:47 +0000 (10:50 -0800)]
ethtool: add compat for devlink info

If driver did not fill the fw_version field, try to call into
the new devlink get_info op and collect the versions that way.
We assume ethtool was always reporting running versions.

v4:
 - use IS_REACHABLE() to avoid problems with DEVLINK=m (kbuildbot).
v3 (Jiri):
 - do a dump and then parse it instead of special handling;
 - concatenate all versions (well, all that fit :)).

Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Acked-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonfp: devlink: report the running and flashed versions
Jakub Kicinski [Thu, 31 Jan 2019 18:50:46 +0000 (10:50 -0800)]
nfp: devlink: report the running and flashed versions

Report versions of firmware components using the new NSP command.

Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Acked-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonfp: nsp: add support for versions command
Jakub Kicinski [Thu, 31 Jan 2019 18:50:45 +0000 (10:50 -0800)]
nfp: nsp: add support for versions command

Retrieve the FW versions with the new command.

Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonfp: devlink: report fixed versions
Jakub Kicinski [Thu, 31 Jan 2019 18:50:44 +0000 (10:50 -0800)]
nfp: devlink: report fixed versions

Report information about the hardware.

RFCv2:
 - add defines for board IDs which are likely to be reusable for
   other drivers (Jiri).

Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Acked-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonfp: devlink: report driver name and serial number
Jakub Kicinski [Thu, 31 Jan 2019 18:50:43 +0000 (10:50 -0800)]
nfp: devlink: report driver name and serial number

Report the basic info through new devlink info API.

RFCv2:
 - add driver name;
 - align serial to core changes.

Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Acked-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agodevlink: add generic info version names
Jakub Kicinski [Thu, 31 Jan 2019 18:50:42 +0000 (10:50 -0800)]
devlink: add generic info version names

Add defines and docs for generic info versions.

v3:
 - add docs;
 - separate patch (Jiri).

Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Acked-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agodevlink: add version reporting to devlink info API
Jakub Kicinski [Thu, 31 Jan 2019 18:50:41 +0000 (10:50 -0800)]
devlink: add version reporting to devlink info API

ethtool -i has a few fixed-size fields which can be used to report
firmware version and expansion ROM version. Unfortunately, modern
hardware has more firmware components. There is usually some
datapath microcode, management controller, PXE drivers, and a
CPLD load. Running ethtool -i on modern controllers reveals the
fact that vendors cram multiple values into firmware version field.

Here are some examples from systems I could lay my hands on quickly:

tg3:  "FFV20.2.17 bc 5720-v1.39"
i40e: "6.01 0x800034a4 1.1747.0"
nfp:  "0.0.3.5 0.25 sriov-2.1.16 nic"

Add a new devlink API to allow retrieving multiple versions, and
provide user-readable name for those versions.

While at it break down the versions into three categories:
 - fixed - this is the board/fixed component version, usually vendors
           report information like the board version in the PCI VPD,
           but it will benefit from naming and common API as well;
 - running - this is the running firmware version;
 - stored - this is firmware in the flash, after firmware update
            this value will reflect the flashed version, while the
            running version may only be updated after reboot.

v3:
 - add per-type helpers instead of using the special argument (Jiri).
RFCv2:
 - remove the nesting in attr DEVLINK_ATTR_INFO_VERSIONS (now
   versions are mixed with other info attrs)l
 - have the driver report versions from the same callback as
   other info.

Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Acked-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agodevlink: add device information API
Jakub Kicinski [Thu, 31 Jan 2019 18:50:40 +0000 (10:50 -0800)]
devlink: add device information API

ethtool -i has served us well for a long time, but its showing
its limitations more and more. The device information should
also be reported per device not per-netdev.

Lay foundation for a simple devlink-based way of reading device
info. Add driver name and device serial number as initial pieces
of information exposed via this new API.

v3:
 - rename helpers (Jiri);
 - rename driver name attr (Jiri);
 - remove double spacing in commit message (Jiri).
RFC v2:
 - wrap the skb into an opaque structure (Jiri);
 - allow the serial number of be any length (Jiri & Andrew);
 - add driver name (Jonathan).

Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Acked-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoMerge branch 'selftests-Various-fixes'
David S. Miller [Fri, 1 Feb 2019 23:26:37 +0000 (15:26 -0800)]
Merge branch 'selftests-Various-fixes'

Petr Machata says:

====================
selftests: Various fixes

This patch set contains various fixes whose common denominator is
improving quality of forwarding and mlxsw selftests.

Most of the fixes are improvements in determinism (such that timing and
latency don't impact the test performance). These were prompted by
regular runs of the test suite on a hardware emulator, the performance
of which is necessarily lower than that of the real device.

Patches #1 (from Ido), #2 and #3 make changes to ping limits.

Patches #4 and #5 add more sleep in places where things need more time
to finish.

Patches #6 and #7 fix two tests in the suite of mirror-to-gretap tests
where underlay involves a VLAN device over an 802.1q bridge.

Patches #8, #9 and #10 fix bugs in mirror-to-gretap test where underlay
involves a LAG device.

Patch #11 fixes a missed RET initialization in mirror-to-gretap flower
test.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoselftests: forwarding: mirror_gre_flower: Fix test result handling
Petr Machata [Thu, 31 Jan 2019 22:35:27 +0000 (22:35 +0000)]
selftests: forwarding: mirror_gre_flower: Fix test result handling

The global variable RET needs to be initialized before each call to
log_test. This test case sets it once before running the tests, but then
calls log_tests for every individual test. Thus a failure in one of the
tests causes spurious failures in follow-up tests as well.

Fix by moving the initialization of RET from test_all() to
full_test_span_gre_dir_acl(), a function that implements the test.

Signed-off-by: Petr Machata <petrm@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoselftests: forwarding: mirror_gre_bridge_1q_lag: Ignore ARP
Petr Machata [Thu, 31 Jan 2019 22:35:25 +0000 (22:35 +0000)]
selftests: forwarding: mirror_gre_bridge_1q_lag: Ignore ARP

This test sets up mirroring such that it mirrors all overlay traffic.
That includes ARP, which causes occasional miscounts and spurious
failures. Ignore ARP explicitly to avoid these problems.

Signed-off-by: Petr Machata <petrm@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoselftests: forwarding: mirror_gre_bridge_1q_lag: Enable forwarding
Petr Machata [Thu, 31 Jan 2019 22:35:24 +0000 (22:35 +0000)]
selftests: forwarding: mirror_gre_bridge_1q_lag: Enable forwarding

This test relies on routing in the primary traffic path, but neglects to
enable forwarding. Do so.

Signed-off-by: Petr Machata <petrm@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoselftests: forwarding: mirror_gre_bridge_1q_lag: Flush neighbors
Petr Machata [Thu, 31 Jan 2019 22:35:22 +0000 (22:35 +0000)]
selftests: forwarding: mirror_gre_bridge_1q_lag: Flush neighbors

After one LAG slave is downed and another upped, it takes a while for
the neighbor on a bridge to time out and get renegotiated. The test does
prompt update of FDB entries by arpinging. But because the neighbor
still references another address, offloading is not possible, and some
packets may end up not being mirrored.

To force the neighbor renegotiation, simply flush the neighbor table at
the bridge.

Signed-off-by: Petr Machata <petrm@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoselftests: forwarding: mirror_gre_vlan_bridge_1q: Fix roaming test
Petr Machata [Thu, 31 Jan 2019 22:35:21 +0000 (22:35 +0000)]
selftests: forwarding: mirror_gre_vlan_bridge_1q: Fix roaming test

ARP or ND traffic can cause spurious migration of FDB back to $swp3.
Mirroring is then updated in accordance with the change, and mirrored
packets are seen at h3, causing a failure.

Detect the case of this spurious roaming, and retry the test.

Signed-off-by: Petr Machata <petrm@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoselftests: forwarding: mirror_gre_vlan_bridge_1q: Fix untagged test
Petr Machata [Thu, 31 Jan 2019 22:35:18 +0000 (22:35 +0000)]
selftests: forwarding: mirror_gre_vlan_bridge_1q: Fix untagged test

The untagged egress test sets up mirroring to {,ip6}gretap such that the
underlay goes through a bridge. Then VLAN flags are manipulated to test
that the traffic leaves the bridge 802.1q-tagged or not, as appropriate.

However, when a neighbor expires at the time that the bridge VLAN is
configured as PVID and egress untagged, the following discovery process
can't finish, because the IP address on H3 is still at the VLAN-tagged
netdevice. This manifests by occasional failures where only several of
the 10 required packets get through.

Therefore, when reconfiguring the VLAN flags, move the IP address to the
appropriate device in the H3 VRF.

In addition to that, take this opportunity to embed an ASCII art diagram
to make the topology move obvious.

Signed-off-by: Petr Machata <petrm@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoselftests: forwarding: mirror_lib: Wait for tardy mirrored packets
Petr Machata [Thu, 31 Jan 2019 22:35:17 +0000 (22:35 +0000)]
selftests: forwarding: mirror_lib: Wait for tardy mirrored packets

When running in an environment with poor performance (such as a
simulator), processing mirrored packets can take a while. Evaluating the
condition too soon leads to spurious "seen 9, expected 10" failures as
the last packet doesn't have enough time to get mirrored and the mirror
to arrive and bump the observed counters.

Wait for one ping interval before evaluating the test.

Signed-off-by: Petr Machata <petrm@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoselftests: forwarding: mirror_gre_changes: Fix TTL test
Petr Machata [Thu, 31 Jan 2019 22:35:15 +0000 (22:35 +0000)]
selftests: forwarding: mirror_gre_changes: Fix TTL test

When running in a simulator, the TTL change takes a while to settle and
during this time the performance of the packet processing is lowered.
The resulting instability leads to ping sending more packets as it
assumes some have been dropped. This then leads to regular spurious
failures as more packets than expected are observed.

Sleep a bit to give the system time to stabilize.

Signed-off-by: Petr Machata <petrm@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoselftests: mlxsw: Update ping limits
Petr Machata [Thu, 31 Jan 2019 22:35:14 +0000 (22:35 +0000)]
selftests: mlxsw: Update ping limits

The current ping intervals are too short for running mirroring tests in
simulator. This leads to ping sending a follow-up ping before the reply
arrives, thus sending more than the requested 10 ICMP requests. This
traffic is seen at the counters, and causes spurious failures.

Bump interval and timeout numbers 5x in mirroring tests to address the
spurious failures.

Signed-off-by: Petr Machata <petrm@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoselftests: forwarding: mirror_lib: Update ping limits
Petr Machata [Thu, 31 Jan 2019 22:35:12 +0000 (22:35 +0000)]
selftests: forwarding: mirror_lib: Update ping limits

The current ping intervals are too short for running mirroring tests in
simulator. This leads to ping sending a follow-up ping before the reply
arrives, thus sending more than the requested 10 ICMP requests. Those
are mirrored, and over a certain threshold the test case run is
considered a failure, because too much traffic is observed.

Bump interval and timeout numbers 5x in mirroring tests to address the
spurious failures.

Signed-off-by: Petr Machata <petrm@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoselftests: forwarding: Make ping timeout configurable
Ido Schimmel [Thu, 31 Jan 2019 22:35:10 +0000 (22:35 +0000)]
selftests: forwarding: Make ping timeout configurable

The current timeout (2 seconds) proved to be too low for some (emulated)
systems where we run the tests.

Make the timeout configurable and default to 5 seconds.

Signed-off-by: Ido Schimmel <idosch@mellanox.com>
Signed-off-by: Petr Machata <petrm@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoipconfig: add carrier_timeout kernel parameter
Martin Kepplinger [Thu, 31 Jan 2019 10:14:18 +0000 (11:14 +0100)]
ipconfig: add carrier_timeout kernel parameter

commit 3fb72f1e6e61 ("ipconfig wait for carrier") added a
"wait for carrier" policy, with a fixed worst case maximum wait
of two minutes.

Now make the wait for carrier timeout configurable on the kernel
commandline and use the 120s as the default.

The timeout messages introduced with
commit 5e404cd65860 ("ipconfig: add informative timeout messages while
waiting for carrier") are done in a fixed interval of 20 seconds, just
like they were before (240/12).

Signed-off-by: Martin Kepplinger <martin.kepplinger@ginzinger.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoipv4: fib: use struct_size() in kzalloc()
Gustavo A. R. Silva [Thu, 31 Jan 2019 00:51:48 +0000 (18:51 -0600)]
ipv4: fib: use struct_size() in kzalloc()

One of the more common cases of allocation size calculations is finding
the size of a structure that has a zero-sized array at the end, along
with memory for some number of elements for that array. For example:

struct foo {
    int stuff;
    struct boo entry[];
};

instance = kzalloc(sizeof(struct foo) + count * sizeof(struct boo), GFP_KERNEL);

Instead of leaving these open-coded and prone to type mistakes, we can
now use the new struct_size() helper:

instance = kzalloc(struct_size(instance, entry, count), GFP_KERNEL);

This code was detected with the help of Coccinelle.

Signed-off-by: Gustavo A. R. Silva <gustavo@embeddedor.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonfp: use struct_size() in kzalloc()
Gustavo A. R. Silva [Thu, 31 Jan 2019 00:38:59 +0000 (18:38 -0600)]
nfp: use struct_size() in kzalloc()

One of the more common cases of allocation size calculations is finding
the size of a structure that has a zero-sized array at the end, along
with memory for some number of elements for that array. For example:

struct foo {
    int stuff;
    struct boo entry[];
};

instance = kzalloc(sizeof(struct foo) + count * sizeof(struct boo), GFP_KERNEL);

Instead of leaving these open-coded and prone to type mistakes, we can
now use the new struct_size() helper:

instance = kzalloc(struct_size(instance, entry, count), GFP_KERNEL);

This code was detected with the help of Coccinelle.

Signed-off-by: Gustavo A. R. Silva <gustavo@embeddedor.com>
Acked-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agotulip: eeprom: use struct_size() in kmalloc()
Gustavo A. R. Silva [Thu, 31 Jan 2019 00:33:55 +0000 (18:33 -0600)]
tulip: eeprom: use struct_size() in kmalloc()

One of the more common cases of allocation size calculations is finding
the size of a structure that has a zero-sized array at the end, along
with memory for some number of elements for that array. For example:

struct foo {
    int stuff;
    struct boo entry[];
};

instance = kmalloc(sizeof(struct foo) + count * sizeof(struct boo), GFP_KERNEL);

Instead of leaving these open-coded and prone to type mistakes, we can
now use the new struct_size() helper:

instance = kmalloc(struct_size(instance, entry, count), GFP_KERNEL);

This code was detected with the help of Coccinelle.

Signed-off-by: Gustavo A. R. Silva <gustavo@embeddedor.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agocxgb4: smt: use struct_size() in kvzalloc()
Gustavo A. R. Silva [Thu, 31 Jan 2019 00:27:44 +0000 (18:27 -0600)]
cxgb4: smt: use struct_size() in kvzalloc()

One of the more common cases of allocation size calculations is
finding the size of a structure that has a zero-sized array at
the end, along with memory for some number of elements for that
array. For example:

struct foo {
    int stuff;
    struct boo entry[];
};

instance = kvzalloc(sizeof(struct foo) + count * sizeof(struct boo), GFP_KERNEL);

Instead of leaving these open-coded and prone to type mistakes, we can now
use the new struct_size() helper:

instance = kvzalloc(struct_size(instance, entry, count), GFP_KERNEL);

This code was detected with the help of Coccinelle.

Signed-off-by: Gustavo A. R. Silva <gustavo@embeddedor.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agocxgb4: sched: use struct_size() in kvzalloc()
Gustavo A. R. Silva [Thu, 31 Jan 2019 00:23:14 +0000 (18:23 -0600)]
cxgb4: sched: use struct_size() in kvzalloc()

One of the more common cases of allocation size calculations is
finding the size of a structure that has a zero-sized array at
the end, along with memory for some number of elements for that
array. For example:

struct foo {
    int stuff;
    struct boo entry[];
};

instance = kvzalloc(sizeof(struct foo) + count * sizeof(struct boo), GFP_KERNEL);

Instead of leaving these open-coded and prone to type mistakes, we can now
use the new struct_size() helper:

instance = kvzalloc(struct_size(instance, entry, count), GFP_KERNEL);

This code was detected with the help of Coccinelle.

Signed-off-by: Gustavo A. R. Silva <gustavo@embeddedor.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: tls: Set async_capable for tls zerocopy only if we see EINPROGRESS
Dave Watson [Wed, 30 Jan 2019 22:08:21 +0000 (22:08 +0000)]
net: tls: Set async_capable for tls zerocopy only if we see EINPROGRESS

Currently we don't zerocopy if the crypto framework async bit is set.
However some crypto algorithms (such as x86 AESNI) support async,
but in the context of sendmsg, will never run asynchronously.  Instead,
check for actual EINPROGRESS return code before assuming algorithm is
async.

Signed-off-by: Dave Watson <davejwatson@fb.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoMerge branch 'tls-1.3-support'
David S. Miller [Fri, 1 Feb 2019 23:00:55 +0000 (15:00 -0800)]
Merge branch 'tls-1.3-support'

Dave Watson says:

====================
net: tls: TLS 1.3 support

This patchset adds 256bit keys and TLS1.3 support to the kernel TLS
socket.

TLS 1.3 is requested by passing TLS_1_3_VERSION in the setsockopt
call, which changes the framing as required for TLS1.3.

256bit keys are requested by passing TLS_CIPHER_AES_GCM_256 in the
sockopt.  This is a fairly straightforward passthrough to the crypto
framework.

256bit keys work with both TLS 1.2 and TLS 1.3

TLS 1.3 requires a different AAD layout, necessitating some minor
refactoring.  It also moves the message type byte to the encrypted
portion of the message, instead of the cleartext header as it was in
TLS1.2.  This requires moving the control message handling to after
decryption, but is otherwise similar.

V1 -> V2

The first two patches were dropped, and sent separately, one as a
bugfix to the net tree.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: tls: Add tests for TLS 1.3
Dave Watson [Wed, 30 Jan 2019 21:58:37 +0000 (21:58 +0000)]
net: tls: Add tests for TLS 1.3

Change most tests to TLS 1.3, while adding tests for previous TLS 1.2
behavior.

Signed-off-by: Dave Watson <davejwatson@fb.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: tls: Add tls 1.3 support
Dave Watson [Wed, 30 Jan 2019 21:58:31 +0000 (21:58 +0000)]
net: tls: Add tls 1.3 support

TLS 1.3 has minor changes from TLS 1.2 at the record layer.

* Header now hardcodes the same version and application content type in
  the header.
* The real content type is appended after the data, before encryption (or
  after decryption).
* The IV is xored with the sequence number, instead of concatinating four
  bytes of IV with the explicit IV.
* Zero-padding:  No exlicit length is given, we search backwards from the
  end of the decrypted data for the first non-zero byte, which is the
  content type.  Currently recv supports reading zero-padding, but there
  is no way for send to add zero padding.

Signed-off-by: Dave Watson <davejwatson@fb.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: tls: Refactor control message handling on recv
Dave Watson [Wed, 30 Jan 2019 21:58:24 +0000 (21:58 +0000)]
net: tls: Refactor control message handling on recv

For TLS 1.3, the control message is encrypted.  Handle control
message checks after decryption.

Signed-off-by: Dave Watson <davejwatson@fb.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: tls: Refactor tls aad space size calculation
Dave Watson [Wed, 30 Jan 2019 21:58:12 +0000 (21:58 +0000)]
net: tls: Refactor tls aad space size calculation

TLS 1.3 has a different AAD size, use a variable in the code to
make TLS 1.3 support easy.

Signed-off-by: Dave Watson <davejwatson@fb.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: tls: Support 256 bit keys
Dave Watson [Wed, 30 Jan 2019 21:58:05 +0000 (21:58 +0000)]
net: tls: Support 256 bit keys

Wire up support for 256 bit keys from the setsockopt to the crypto
framework

Signed-off-by: Dave Watson <davejwatson@fb.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoMerge branch 'bpf-xdp-sample-libbpf'
Daniel Borkmann [Fri, 1 Feb 2019 22:37:52 +0000 (23:37 +0100)]
Merge branch 'bpf-xdp-sample-libbpf'

Maciej Fijalkowski says:

====================
This patchset tries to address the situation where:
* user loads a particular xdp sample application that does stats polling
* user loads another sample application on the same interface
* then, user sends SIGINT/SIGTERM to the app that was attached as a first one
* second application ends up with an unloaded xdp program

1st patch contains a helper libbpf function for getting the map fd by a
given map name.
In patch 2 Jesper removes the read_trace_pipe usage from xdp_redirect_cpu which
was a blocker for converting this sample to libbpf usage.
3rd patch updates a bunch of xdp samples to make the use of libbpf.
Patch 4 adjusts RLIMIT_MEMLOCK for two samples touched in this patchset.
In patch 5 extack messages are added for cases where dev_change_xdp_fd returns
with an error so user has an idea what was the reason for not attaching the
xdp program onto interface.
Patch 6 makes the samples behavior similar to what iproute2 does when loading
xdp prog - the "force" flag is introduced.
Patch 7 introduces the libbpf function that will query the driver from
userspace about the currently attached xdp prog id.

Use it in samples that do polling by checking the prog id in signal handler
and comparing it with previously stored one which is the scope of patch 8.

Thanks!

v1->v2:
* add a libbpf helper for getting a prog via relative index
* include xdp_redirect_cpu into conversion

v2->v3: mostly addressing Daniel's/Jesper's comments
* get rid of the helper from v1->v2
* feed the xdp_redirect_cpu with program name instead of number

v3->v4:
* fix help message in xdp_sample_pkts

v4->v5:
* in get_link_xdp_fd, assign prog_id only when libbpf_nl_get_link returned
  with 0
* add extack messages in dev_change_xdp_fd
* check the return value of bpf_get_link_xdp_id when exiting from sample progs

v5->v6:
* rebase
====================

Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
5 years agosamples/bpf: Check the prog id before exiting
Maciej Fijalkowski [Fri, 1 Feb 2019 21:42:30 +0000 (22:42 +0100)]
samples/bpf: Check the prog id before exiting

Check the program id within the signal handler on polling xdp samples
that were previously converted to libbpf usage. Avoid the situation of
unloading the program that was not attached by sample that is exiting.
Handle also the case where bpf_get_link_xdp_id didn't exit with an error
but the xdp program was not found on an interface.

Reported-by: Michal Papaj <michal.papaj@intel.com>
Reported-by: Jakub Spizewski <jakub.spizewski@intel.com>
Signed-off-by: Maciej Fijalkowski <maciej.fijalkowski@intel.com>
Reviewed-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
5 years agolibbpf: Add a support for getting xdp prog id on ifindex
Maciej Fijalkowski [Fri, 1 Feb 2019 21:42:29 +0000 (22:42 +0100)]
libbpf: Add a support for getting xdp prog id on ifindex

Since we have a dedicated netlink attributes for xdp setup on a
particular interface, it is now possible to retrieve the program id that
is currently attached to the interface. The use case is targeted for
sample xdp programs, which will store the program id just after loading
bpf program onto iface. On shutdown, the sample will make sure that it
can unload the program by querying again the iface and verifying that
both program id's matches.

Signed-off-by: Maciej Fijalkowski <maciej.fijalkowski@intel.com>
Reviewed-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
5 years agosamples/bpf: Add a "force" flag to XDP samples
Maciej Fijalkowski [Fri, 1 Feb 2019 21:42:28 +0000 (22:42 +0100)]
samples/bpf: Add a "force" flag to XDP samples

Make xdp samples consistent with iproute2 behavior and set the
XDP_FLAGS_UPDATE_IF_NOEXIST by default when setting the xdp program on
interface. Provide an option for user to force the program loading,
which as a result will not include the mentioned flag in
bpf_set_link_xdp_fd call.

Signed-off-by: Maciej Fijalkowski <maciej.fijalkowski@intel.com>
Reviewed-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Acked-by: John Fastabend <john.fastabend@gmail.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
5 years agoxdp: Provide extack messages when prog attachment failed
Maciej Fijalkowski [Fri, 1 Feb 2019 21:42:27 +0000 (22:42 +0100)]
xdp: Provide extack messages when prog attachment failed

In order to provide more meaningful messages to user when the process of
loading xdp program onto network interface failed, let's add extack
messages within dev_change_xdp_fd.

Suggested-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Signed-off-by: Maciej Fijalkowski <maciej.fijalkowski@intel.com>
Acked-by: Jesper Dangaard Brouer <brouer@redhat.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
5 years agosamples/bpf: Extend RLIMIT_MEMLOCK for xdp_{sample_pkts, router_ipv4}
Maciej Fijalkowski [Fri, 1 Feb 2019 21:42:26 +0000 (22:42 +0100)]
samples/bpf: Extend RLIMIT_MEMLOCK for xdp_{sample_pkts, router_ipv4}

There is a common problem with xdp samples that happens when user wants
to run a particular sample and some bpf program is already loaded. The
default 64kb RLIMIT_MEMLOCK resource limit will cause a following error
(assuming that xdp sample that is failing was converted to libbpf
usage):

libbpf: Error in bpf_object__probe_name():Operation not permitted(1).
Couldn't load basic 'r0 = 0' BPF program.
libbpf: failed to load object './xdp_sample_pkts_kern.o'

Fix it in xdp_sample_pkts and xdp_router_ipv4 by setting RLIMIT_MEMLOCK
to RLIM_INFINITY.

Signed-off-by: Maciej Fijalkowski <maciej.fijalkowski@intel.com>
Reviewed-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Acked-by: Jesper Dangaard Brouer <brouer@redhat.com>
Acked-by: John Fastabend <john.fastabend@gmail.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
5 years agosamples/bpf: Convert XDP samples to libbpf usage
Maciej Fijalkowski [Fri, 1 Feb 2019 21:42:25 +0000 (22:42 +0100)]
samples/bpf: Convert XDP samples to libbpf usage

Some of XDP samples that are attaching the bpf program to the interface
via libbpf's bpf_set_link_xdp_fd are still using the bpf_load.c for
loading and manipulating the ebpf program and maps. Convert them to do
this through libbpf usage and remove bpf_load from the picture.

While at it remove what looks like debug leftover in
xdp_redirect_map_user.c

In xdp_redirect_cpu, change the way that the program to be loaded onto
interface is chosen - user now needs to pass the program's section name
instead of the relative number. In case of typo print out the section
names to choose from.

Signed-off-by: Maciej Fijalkowski <maciej.fijalkowski@intel.com>
Reviewed-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Acked-by: Jesper Dangaard Brouer <brouer@redhat.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
5 years agosamples/bpf: xdp_redirect_cpu have not need for read_trace_pipe
Jesper Dangaard Brouer [Fri, 1 Feb 2019 21:42:24 +0000 (22:42 +0100)]
samples/bpf: xdp_redirect_cpu have not need for read_trace_pipe

The sample xdp_redirect_cpu is not using helper bpf_trace_printk.
Thus it makes no sense that the --debug option us reading
from /sys/kernel/debug/tracing/trace_pipe via read_trace_pipe.
Simply remove it.

Signed-off-by: Jesper Dangaard Brouer <brouer@redhat.com>
Acked-by: John Fastabend <john.fastabend@gmail.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
5 years agolibbpf: Add a helper for retrieving a map fd for a given name
Maciej Fijalkowski [Fri, 1 Feb 2019 21:42:23 +0000 (22:42 +0100)]
libbpf: Add a helper for retrieving a map fd for a given name

XDP samples are mostly cooperating with eBPF maps through their file
descriptors. In case of a eBPF program that contains multiple maps it
might be tiresome to iterate through them and call bpf_map__fd for each
one. Add a helper mostly based on bpf_object__find_map_by_name, but
instead of returning the struct bpf_map pointer, return map fd.

Suggested-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Signed-off-by: Maciej Fijalkowski <maciej.fijalkowski@intel.com>
Reviewed-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Acked-by: John Fastabend <john.fastabend@gmail.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
5 years agobpf: powerpc64: add JIT support for bpf line info
Sandipan Das [Fri, 1 Feb 2019 10:32:32 +0000 (16:02 +0530)]
bpf: powerpc64: add JIT support for bpf line info

This adds support for generating bpf line info for
JITed programs.

Signed-off-by: Sandipan Das <sandipan@linux.ibm.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
5 years agoMerge branch 'bpf-spinlocks'
Daniel Borkmann [Fri, 1 Feb 2019 19:55:40 +0000 (20:55 +0100)]
Merge branch 'bpf-spinlocks'

Alexei Starovoitov says:

====================
Many algorithms need to read and modify several variables atomically.
Until now it was hard to impossible to implement such algorithms in BPF.
Hence introduce support for bpf_spin_lock.

The api consists of 'struct bpf_spin_lock' that should be placed
inside hash/array/cgroup_local_storage element
and bpf_spin_lock/unlock() helper function.

Example:
struct hash_elem {
    int cnt;
    struct bpf_spin_lock lock;
};
struct hash_elem * val = bpf_map_lookup_elem(&hash_map, &key);
if (val) {
    bpf_spin_lock(&val->lock);
    val->cnt++;
    bpf_spin_unlock(&val->lock);
}

and BPF_F_LOCK flag for lookup/update bpf syscall commands that
allows user space to read/write map elements under lock.

Together these primitives allow race free access to map elements
from bpf programs and from user space.

Key restriction: root only.
Key requirement: maps must be annotated with BTF.

This concept was discussed at Linux Plumbers Conference 2018.
Thank you everyone who participated and helped to iron out details
of api and implementation.

Patch 1: bpf_spin_lock support in the verifier, BTF, hash, array.
Patch 2: bpf_spin_lock in cgroup local storage.
Patches 3,4,5: tests
Patch 6: BPF_F_LOCK flag to lookup/update
Patches 7,8,9: tests

v6->v7:
- fixed this_cpu->__this_cpu per Peter's suggestion and added Ack.
- simplified bpf_spin_lock and load/store overlap check in the verifier
  as suggested by Andrii
- rebase

v5->v6:
- adopted arch_spinlock approach suggested by Peter
- switched to spin_lock_irqsave equivalent as the simplest way
  to avoid deadlocks in rare case of nested networking progs
  (cgroup-bpf prog in preempt_disable vs clsbpf in softirq sharing
  the same map with bpf_spin_lock)
  bpf_spin_lock is only allowed in networking progs that don't
  have arbitrary entry points unlike tracing progs.
- rebase and split test_verifier tests

v4->v5:
- disallow bpf_spin_lock for tracing progs due to insufficient preemption checks
- socket filter progs cannot use bpf_spin_lock due to missing preempt_disable
- fix atomic_set_release. Spotted by Peter.
- fixed hash_of_maps

v3->v4:
- fix BPF_EXIST | BPF_NOEXIST check patch 6. Spotted by Jakub. Thanks!
- rebase

v2->v3:
- fixed build on ia64 and archs where qspinlock is not supported
- fixed missing lock init during lookup w/o BPF_F_LOCK. Spotted by Martin

v1->v2:
- addressed several issues spotted by Daniel and Martin in patch 1
- added test11 to patch 4 as suggested by Daniel
====================

Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
5 years agoselftests/bpf: test for BPF_F_LOCK
Alexei Starovoitov [Thu, 31 Jan 2019 23:40:12 +0000 (15:40 -0800)]
selftests/bpf: test for BPF_F_LOCK

Add C based test that runs 4 bpf programs in parallel
that update the same hash and array maps.
And another 2 threads that read from these two maps
via lookup(key, value, BPF_F_LOCK) api
to make sure the user space sees consistent value in both
hash and array elements while user space races with kernel bpf progs.

Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
5 years agolibbpf: introduce bpf_map_lookup_elem_flags()
Alexei Starovoitov [Thu, 31 Jan 2019 23:40:11 +0000 (15:40 -0800)]
libbpf: introduce bpf_map_lookup_elem_flags()

Introduce
int bpf_map_lookup_elem_flags(int fd, const void *key, void *value, __u64 flags)
helper to lookup array/hash/cgroup_local_storage elements with BPF_F_LOCK flag.

Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
5 years agotools/bpf: sync uapi/bpf.h
Alexei Starovoitov [Thu, 31 Jan 2019 23:40:10 +0000 (15:40 -0800)]
tools/bpf: sync uapi/bpf.h

add BPF_F_LOCK definition to tools/include/uapi/linux/bpf.h

Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
5 years agobpf: introduce BPF_F_LOCK flag
Alexei Starovoitov [Thu, 31 Jan 2019 23:40:09 +0000 (15:40 -0800)]
bpf: introduce BPF_F_LOCK flag

Introduce BPF_F_LOCK flag for map_lookup and map_update syscall commands
and for map_update() helper function.
In all these cases take a lock of existing element (which was provided
in BTF description) before copying (in or out) the rest of map value.

Implementation details that are part of uapi:

Array:
The array map takes the element lock for lookup/update.

Hash:
hash map also takes the lock for lookup/update and tries to avoid the bucket lock.
If old element exists it takes the element lock and updates the element in place.
If element doesn't exist it allocates new one and inserts into hash table
while holding the bucket lock.
In rare case the hashmap has to take both the bucket lock and the element lock
to update old value in place.

Cgroup local storage:
It is similar to array. update in place and lookup are done with lock taken.

Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
5 years agoselftests/bpf: add bpf_spin_lock C test
Alexei Starovoitov [Thu, 31 Jan 2019 23:40:08 +0000 (15:40 -0800)]
selftests/bpf: add bpf_spin_lock C test

add bpf_spin_lock C based test that requires latest llvm with BTF support

Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
5 years agoselftests/bpf: add bpf_spin_lock verifier tests
Alexei Starovoitov [Thu, 31 Jan 2019 23:40:07 +0000 (15:40 -0800)]
selftests/bpf: add bpf_spin_lock verifier tests

add bpf_spin_lock tests to test_verifier.c that don't require
latest llvm with BTF support

Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>