Force of 56G is not supported by hardware in Ethernet devices. This
configuration fails with a bad parameter error from firmware.

Add check of this case. Instead of trying to set 56G with autoneg off,
return a meaningful error.

Fixes: 56ade8fe ("mlxsw: spectrum: Add initial support for Spectrum ASIC")
Signed-off-by: Amit Cohen <amitc@mellanox.com>
Acked-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: Ido Schimmel <idosch@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

When identical rules are inserted, the latter one goes to C-TCAM. For
that, a second eRP with the same mask is created. These 2 eRPs by the
nature cannot be merged and also one cannot be parent of another.
Teach mlxsw_sp_acl_erp_delta_fill() about this possibility and handle it
gracefully.
Reported-by: Alex Kushnarov <alexanderk@mellanox.com>
Fixes: c22291f7 ("mlxsw: spectrum: acl: Implement delta for ERP")
Signed-off-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: Ido Schimmel <idosch@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Currently, the upper half of a 4-byte STATS_TYPE_PORT statistic ends
up in bits 47:32 of the return value, instead of bits 31:16 as they
should.

Fixes: 6e46e2d8 ("net: dsa: mv88e6xxx: Fix u64 statistics")
Signed-off-by: Rasmus Villemoes <rasmus.villemoes@prevas.dk>
Reviewed-by: Vivien Didelot <vivien.didelot@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

(At least) RTL8168e forgets its MAC address in PCI D3. To fix this set
the MAC address when resuming. For resuming from runtime-suspend we
had this in place already, for resuming from S3/S5 it was missing.

The commit referenced as being fixed isn't wrong, it's just the first
one where the patch applies cleanly.

Fixes: 0f07bd85 ("r8169: use dev_get_drvdata where possible")
Signed-off-by: Heiner Kallweit <hkallweit1@gmail.com>
Reported-by: Albert Astals Cid <aacid@kde.org>
Tested-by: Albert Astals Cid <aacid@kde.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

Saeed Mahameed says:

====================
Mellanox, mlx5 fixes 2019-05-28

This series introduces some fixes to mlx5 driver.

Please pull and let me know if there is any problem.

For -stable v4.13:
('net/mlx5: Allocate root ns memory using kzalloc to match kfree')

For -stable v4.16:
('net/mlx5: Avoid double free in fs init error unwinding path')

For -stable v4.18:
('net/mlx5e: Disable rxhash when CQE compress is enabled')
====================
Signed-off-by: David S. Miller <davem@davemloft.net>

Stephen Hemminger says:

====================
XDP generic fixes

This set of patches came about while investigating XDP
generic on Azure. The split brain nature of the accelerated
networking exposed issues with the stack device model.
====================
Signed-off-by: David S. Miller <davem@davemloft.net>

When a device is stacked like (team, bonding, failsafe or netvsc) the
XDP generic program for the parent device was not called.

Move the call to XDP generic inside __netif_receive_skb_core where
it can be done multiple times for stacked case.

Fixes: d4455169 ("net: xdp: support xdp generic on virtual devices")
Signed-off-by: Stephen Hemminger <sthemmin@microsoft.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

The netvsc VF skb handler should make sure that skb is not
shared. Similar logic already exists in bonding and team device
drivers.

This is not an issue in practice because the VF devicex
does not send up shared skb's. But the netvsc driver
should do the right thing if it did.

Fixes: 0c195567 ("netvsc: transparent VF management")
Signed-off-by: Stephen Hemminger <sthemmin@microsoft.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Currently, when resegmenting an unexpected UDP GRO packet, the full UDP
checksum will be calculated for every new SKB created by skb_segment()
because the netdev features passed in by udp_rcv_segment() lack any
information about checksum offload capabilities.

Usually, we have no need to perform this calculation again, as
  1) The GRO implementation guarantees that any packets making it to the
     udp_rcv_segment() function had correct checksums, and, more
     importantly,
  2) Upon the successful return of udp_rcv_segment(), we immediately pull
     the UDP header off and either queue the segment to the socket or
     hand it off to a new protocol handler.

Unless userspace has set the IP_CHECKSUM sockopt to indicate that they
want the final checksum values, we can pass the needed netdev feature
flags to __skb_gso_segment() to avoid checksumming each segment in
skb_segment().

Fixes: cf329aa4 ("udp: cope with UDP GRO packet misdirection")
Cc: Paolo Abeni <pabeni@redhat.com>
Cc: Subash Abhinov Kasiviswanathan <subashab@codeaurora.org>
Signed-off-by: Sean Tranchetti <stranche@codeaurora.org>
Acked-by: Paolo Abeni <pabeni@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Max Uvarov says:

====================
net: phy: dp83867: add some fixes

v3: use phy_modify_mmd()
v2: fix minor comments by Heiner Kallweit and Florian Fainelli
====================
Signed-off-by: David S. Miller <davem@davemloft.net>

PHY_INTERFACE_MODE_RGMII_RXID is less then TXID
so code to set tx delay is never called.

Fixes: 2a10154a ("net: phy: dp83867: Add TI dp83867 phy")
Signed-off-by: Max Uvarov <muvarov@gmail.com>
Cc: Florian Fainelli <f.fainelli@gmail.com>
Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Phy state machine calls _config_init just after
reset.
Signed-off-by: Max Uvarov <muvarov@gmail.com>
Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

After reset SGMII Autoneg timer is set to 2us (bits 6 and 5 are 01).
That is not enough to finalize autonegatiation on some devices.
Increase this timer duration to maximum supported 16ms.
Signed-off-by: Max Uvarov <muvarov@gmail.com>
Cc: Heiner Kallweit <hkallweit1@gmail.com>
Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

For supporting 10Mps speed in SGMII mode DP83867_10M_SGMII_RATE_ADAPT bit
of DP83867_10M_SGMII_CFG register has to be cleared by software.
That does not affect speeds 100 and 1000 so can be done on init.
Signed-off-by: Max Uvarov <muvarov@gmail.com>
Cc: Heiner Kallweit <hkallweit1@gmail.com>
Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Some boards do not have the PHY firmware programmed in the 3310's flash,
which leads to the PHY not working as expected.  Warn the user when the
PHY fails to boot the firmware and refuse to initialise.

Fixes: 20b2af32 ("net: phy: add Marvell Alaska X 88X3310 10Gigabit PHY support")
Signed-off-by: Russell King <rmk+kernel@armlinux.org.uk>
Tested-by: Maxime Chevallier <maxime.chevallier@bootlin.com>
Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>

Ensure that we supply the same phy interface mode to mac_link_down() as
we did for the corresponding mac_link_up() call.  This ensures that MAC
drivers that use the phy interface mode in these methods can depend on
mac_link_down() always corresponding to a mac_link_up() call for the
same interface mode.
Signed-off-by: Russell King <rmk+kernel@armlinux.org.uk>
Signed-off-by: David S. Miller <davem@davemloft.net>

The sh_eth_close() resets the MAC and then calls phy_stop()
so that mdio read access result is incorrect without any error
according to kernel trace like below:

ifconfig-216   [003] .n..   109.133124: mdio_access: ee700000.ethernet-ffffffff read  phy:0x01 reg:0x00 val:0xffff

According to the hardware manual, the RMII mode should be set to 1
before operation the Ethernet MAC. However, the previous code was not
set to 1 after the driver issued the soft_reset in sh_eth_dev_exit()
so that the mdio read access result seemed incorrect. To fix the issue,
this patch adds a condition and set the RMII mode register in
sh_eth_dev_exit() for R-Car Gen2 and RZ/A1 SoCs.

Note that when I have tried to move the sh_eth_dev_exit() calling
after phy_stop() on sh_eth_close(), but it gets worse (kernel panic
happened and it seems that a register is accessed while the clock is
off).
Signed-off-by: Yoshihiro Shimoda <yoshihiro.shimoda.uh@renesas.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

When CQE compression is enabled (Multi-host systems), compressed CQEs
might arrive to the driver rx, compressed CQEs don't have a valid hash
offload and the driver already reports a hash value of 0 and invalid hash
type on the skb for compressed CQEs, but this is not good enough.

On a congested PCIe, where CQE compression will kick in aggressively,
gro will deliver lots of out of order packets due to the invalid hash
and this might cause a serious performance drop.

The only valid solution, is to disable rxhash offload at all when CQE
compression is favorable (Multi-host systems).

Fixes: 7219ab34 ("net/mlx5e: CQE compression")
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>

When register indr block for vlan device, it should check the real_dev
of vlan device is same as uplink device. Or it will set offload rule
to mlx5e which will never hit.

Fixes: 35a605db ("net/mlx5e: Offload TC e-switch rules with ingress VLAN device")
Signed-off-by: wenxu <wenxu@ucloud.cn>
Reviewed-by: Roi Dayan <roid@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>

root ns is yet another fs core node which is freed using kfree() by
tree_put_node().
Rest of the other fs core objects are also allocated using kmalloc
variants.

However, root ns memory is allocated using kvzalloc().
Hence allocate root ns memory using kzalloc().

Fixes: 25302363 ("net/mlx5_core: Flow steering tree initialization")
Signed-off-by: Parav Pandit <parav@mellanox.com>
Reviewed-by: Daniel Jurgens <danielj@mellanox.com>
Reviewed-by: Mark Bloch <markb@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>

In below code flow, for ingress acl table root ns memory leads
to double free.

mlx5_init_fs
  init_ingress_acls_root_ns()
    init_ingress_acl_root_ns
       kfree(steering->esw_ingress_root_ns);
       /* steering->esw_ingress_root_ns is not marked NULL */
  mlx5_cleanup_fs
    cleanup_ingress_acls_root_ns
       steering->esw_ingress_root_ns non NULL check passes.
       kfree(steering->esw_ingress_root_ns);
       /* double free */

Similar issue exist for other tables.

Hence zero out the pointers to not process the table again.

Fixes: 9b93ab98 ("net/mlx5: Separate ingress/egress namespaces for each vport")
Fixes: 40c3eebb49e51 ("net/mlx5: Add support in RDMA RX steering")
Signed-off-by: Parav Pandit <parav@mellanox.com>
Reviewed-by: Mark Bloch <markb@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>

When root ns setup for rdma, sniffer tx and sniffer rx fails,
such root ns cleanup is done by the error unwinding path of
mlx5_cleanup_fs().
Below call graph shows an example for sniffer_rx_root_ns.

mlx5_init_fs()
  init_sniffer_rx_root_ns()
    cleanup_root_ns(steering->sniffer_rx_root_ns);
mlx5_cleanup_fs()
  cleanup_root_ns(steering->sniffer_rx_root_ns);
  /* double free of sniffer_rx_root_ns */

Hence, use the existing cleanup_fs to cleanup.

Fixes: d83eb50e ("net/mlx5: Add support in RDMA RX steering")
Fixes: 87d22483 ("net/mlx5: Add sniffer namespaces")
Signed-off-by: Parav Pandit <parav@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>

In case mlx5_core_set_hca_defaults fails, it should jump to
mlx5_cleanup_fs, fix that.

Fixes: c85023e1 ("IB/mlx5: Add raw ethernet local loopback support")
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
Reviewed-by: Huy Nguyen <huyn@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>

Both sysfs-bus-mdio and sysfs-class-net-phydev contain the same
duplication information. There is not currently any MDIO bus specific
attribute, but there are PHY device (struct phy_device) specific
attributes. Use the more precise description from sysfs-bus-mdio and
carry that over to sysfs-class-net-phydev.

Fixes: 86f22d04 ("net: sysfs: Document PHY device sysfs attributes")
Signed-off-by: Florian Fainelli <f.fainelli@gmail.com>
Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>

If llc_mac_hdr_init() returns an error, we must drop the skb
since no llc_build_and_send_ui_pkt() caller will take care of this.

BUG: memory leak
unreferenced object 0xffff8881202b6800 (size 2048):
  comm "syz-executor907", pid 7074, jiffies 4294943781 (age 8.590s)
  hex dump (first 32 bytes):
    00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00  ................
    1a 00 07 40 00 00 00 00 00 00 00 00 00 00 00 00  ...@............
  backtrace:
    [<00000000e25b5abe>] kmemleak_alloc_recursive include/linux/kmemleak.h:55 [inline]
    [<00000000e25b5abe>] slab_post_alloc_hook mm/slab.h:439 [inline]
    [<00000000e25b5abe>] slab_alloc mm/slab.c:3326 [inline]
    [<00000000e25b5abe>] __do_kmalloc mm/slab.c:3658 [inline]
    [<00000000e25b5abe>] __kmalloc+0x161/0x2c0 mm/slab.c:3669
    [<00000000a1ae188a>] kmalloc include/linux/slab.h:552 [inline]
    [<00000000a1ae188a>] sk_prot_alloc+0xd6/0x170 net/core/sock.c:1608
    [<00000000ded25bbe>] sk_alloc+0x35/0x2f0 net/core/sock.c:1662
    [<000000002ecae075>] llc_sk_alloc+0x35/0x170 net/llc/llc_conn.c:950
    [<00000000551f7c47>] llc_ui_create+0x7b/0x140 net/llc/af_llc.c:173
    [<0000000029027f0e>] __sock_create+0x164/0x250 net/socket.c:1430
    [<000000008bdec225>] sock_create net/socket.c:1481 [inline]
    [<000000008bdec225>] __sys_socket+0x69/0x110 net/socket.c:1523
    [<00000000b6439228>] __do_sys_socket net/socket.c:1532 [inline]
    [<00000000b6439228>] __se_sys_socket net/socket.c:1530 [inline]
    [<00000000b6439228>] __x64_sys_socket+0x1e/0x30 net/socket.c:1530
    [<00000000cec820c1>] do_syscall_64+0x76/0x1a0 arch/x86/entry/common.c:301
    [<000000000c32554f>] entry_SYSCALL_64_after_hwframe+0x44/0xa9

BUG: memory leak
unreferenced object 0xffff88811d750d00 (size 224):
  comm "syz-executor907", pid 7074, jiffies 4294943781 (age 8.600s)
  hex dump (first 32 bytes):
    00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00  ................
    00 f0 0c 24 81 88 ff ff 00 68 2b 20 81 88 ff ff  ...$.....h+ ....
  backtrace:
    [<0000000053026172>] kmemleak_alloc_recursive include/linux/kmemleak.h:55 [inline]
    [<0000000053026172>] slab_post_alloc_hook mm/slab.h:439 [inline]
    [<0000000053026172>] slab_alloc_node mm/slab.c:3269 [inline]
    [<0000000053026172>] kmem_cache_alloc_node+0x153/0x2a0 mm/slab.c:3579
    [<00000000fa8f3c30>] __alloc_skb+0x6e/0x210 net/core/skbuff.c:198
    [<00000000d96fdafb>] alloc_skb include/linux/skbuff.h:1058 [inline]
    [<00000000d96fdafb>] alloc_skb_with_frags+0x5f/0x250 net/core/skbuff.c:5327
    [<000000000a34a2e7>] sock_alloc_send_pskb+0x269/0x2a0 net/core/sock.c:2225
    [<00000000ee39999b>] sock_alloc_send_skb+0x32/0x40 net/core/sock.c:2242
    [<00000000e034d810>] llc_ui_sendmsg+0x10a/0x540 net/llc/af_llc.c:933
    [<00000000c0bc8445>] sock_sendmsg_nosec net/socket.c:652 [inline]
    [<00000000c0bc8445>] sock_sendmsg+0x54/0x70 net/socket.c:671
    [<000000003b687167>] __sys_sendto+0x148/0x1f0 net/socket.c:1964
    [<00000000922d78d9>] __do_sys_sendto net/socket.c:1976 [inline]
    [<00000000922d78d9>] __se_sys_sendto net/socket.c:1972 [inline]
    [<00000000922d78d9>] __x64_sys_sendto+0x2a/0x30 net/socket.c:1972
    [<00000000cec820c1>] do_syscall_64+0x76/0x1a0 arch/x86/entry/common.c:301
    [<000000000c32554f>] entry_SYSCALL_64_after_hwframe+0x44/0xa9

Fixes: 1da177e4 ("Linux-2.6.12-rc2")
Signed-off-by: Eric Dumazet <edumazet@google.com>
Reported-by: syzbot <syzkaller@googlegroups.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

In the pmtu_vti6_link_change_mtu test, both local and remote addresses
for the vti6 tunnel are assigned to the same address given to the dummy
interface that we use as encapsulating device with a known MTU.

This works as long as the dummy interface is actually selected, via
rt6_lookup(), as encapsulating device. But if the remote address of the
tunnel is a local address too, the loopback interface could also be
selected, and there's nothing wrong with it.

This is what some older -stable kernels do (3.18.z, at least), and
nothing prevents us from subtly changing FIB implementation to revert
back to that behaviour in the future.

Define an IPv6 prefix instead, and use two separate addresses as local
and remote for vti6, so that the encapsulating device can't be a
loopback interface.
Reported-by: Xiumei Mu <xmu@redhat.com>
Fixes: 1fad59ea ("selftests: pmtu: Add pmtu_vti6_link_change_mtu test")
Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Make sure only the portals for the online CPUs are used.
Without this change, there are issues when someone boots with
maxcpus=n, with n < actual number of cores available as frames
either received or corresponding to the transmit confirmation
path would be offered for dequeue to the offline CPU portals,
getting lost.
Signed-off-by: Madalin Bucur <madalin.bucur@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Fix below issues in err code path of probe:
1. we don't need to unregister_netdev() because the netdev isn't
registered.
2. when register_netdev() fails, we also need to destroy bm pool for
HWBM case.

Fixes: dc35a10f ("net: mvneta: bm: add support for hardware buffer management")
Signed-off-by: Jisheng Zhang <Jisheng.Zhang@synaptics.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

If the subdriver defers probe, do not show an error message. It's
perfectly fine for this error to occur since the driver will get another
chance to probe after some time and will usually succeed after all of
the resources that it requires have been registered.
Signed-off-by: Thierry Reding <treding@nvidia.com>
Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>

Igor Russkikh says:

====================
net: aquantia: various fixes May, 2019

Here is a set of various bug fixes found on recent verification stage.
====================
Signed-off-by: David S. Miller <davem@davemloft.net>

Thats a known quirk in windows tcp stack it can produce 0xffff checksum.
Thats incorrect but it is.

Atlantic HW with LRO enabled handles that incorrectly and changes csum to
0xfffe - but indicates that csum is invalid. This causes driver to pass
packet to linux networking stack with CSUM_NONE, stack eventually drops
the packet.

There is a quirk in atlantic HW to enable correct processing of
0xffff incorrect csum. Enable it.

The visible bug is that windows link partner with software generated csums
caused TCP connection to be unstable since all packets that csum value
are dropped.
Reported-by: Dmitry Bezrukov <dmitry.bezrukov@aquantia.com>
Signed-off-by: Nikita Danilov <ndanilov@aquantia.com>
Signed-off-by: Igor Russkikh <igor.russkikh@aquantia.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Driver stops producing skbs on ring if a packet with FCS error
was coalesced into LRO session. Ring gets hang forever.

Thats a logical error in driver processing descriptors:
When rx_stat indicates MAC Error, next pointer and eop flags
are not filled. This confuses driver so it waits for descriptor 0
to be filled by HW.

Solution is fill next pointer and eop flag even for packets with FCS error.

Fixes: bab6de8f ("net: ethernet: aquantia: Atlantic A0 and B0 specific functions.")
Signed-off-by: Igor Russkikh <igor.russkikh@aquantia.com>
Signed-off-by: Dmitry Bogdanov <dmitry.bogdanov@aquantia.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Atlantic hardware does not aggregate nor breaks LRO sessions
with bad csum packets. This means driver should take care of that.

If in LRO session there is a non-first descriptor with invalid
checksum (L2/L3/L4), the driver must account this information
in csum application logic.

Fixes: 018423e9 ("net: ethernet: aquantia: Add ring support code")
Signed-off-by: Igor Russkikh <igor.russkikh@aquantia.com>
Signed-off-by: Dmitry Bogdanov <dmitry.bogdanov@aquantia.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

In case no other traffic happening on the ring, full tx cleanup
may not be completed. That may cause socket buffer to overflow
and tx traffic to stuck until next activity on the ring happens.

This is due to logic error in budget variable decrementor.
Variable is compared with zero, and then post decremented,
causing it to become MAX_INT. Solution is remove decrementor
from the `for` statement and rewrite it in a clear way.

Fixes: b647d398 ("net: aquantia: Add tx clean budget and valid budget handling logic")
Signed-off-by: Igor Russkikh <igor.russkikh@aquantia.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Building with Clang reports the redundant use of MODULE_DEVICE_TABLE():

drivers/net/ethernet/dec/tulip/de4x5.c:2110:1: error: redefinition of '__mod_eisa__de4x5_eisa_ids_device_table'
MODULE_DEVICE_TABLE(eisa, de4x5_eisa_ids);
^
./include/linux/module.h:229:21: note: expanded from macro 'MODULE_DEVICE_TABLE'
extern typeof(name) __mod_##type##__##name##_device_table               \
                    ^
<scratch space>:90:1: note: expanded from here
__mod_eisa__de4x5_eisa_ids_device_table
^
drivers/net/ethernet/dec/tulip/de4x5.c:2100:1: note: previous definition is here
MODULE_DEVICE_TABLE(eisa, de4x5_eisa_ids);
^
./include/linux/module.h:229:21: note: expanded from macro 'MODULE_DEVICE_TABLE'
extern typeof(name) __mod_##type##__##name##_device_table               \
                    ^
<scratch space>:85:1: note: expanded from here
__mod_eisa__de4x5_eisa_ids_device_table
^

This drops the one further from the table definition to match the common
use of MODULE_DEVICE_TABLE().

Fixes: 07563c71 ("EISA bus MODALIAS attributes support")
Signed-off-by: Kees Cook <keescook@chromium.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

Jakub Kicinski says:

====================
net/tls: two fixes for rx_list pre-handling

tls_sw_recvmsg() had been modified to cater better to async decrypt.
Partially read records now live on the rx_list. Data is copied from
this list before the old do {} while loop, and the not included
correctly in deciding whether to sleep or not and lowat threshold
handling. These modifications, unfortunately, added some bugs.

First patch fixes lowat - we need to calculate the threshold early
and make sure all copied data is compared to the threshold, not just
the freshly decrypted data.

Third patch fixes sleep - if data is picked up from rx_list and
no flags are set, we should not put the process to sleep, but
rather return the partial read.

Patches 2 and 4 add test cases for these bugs, both will cause
a sleep and test timeout before the fix.
====================
Signed-off-by: David S. Miller <davem@davemloft.net>

Add a test which sends 15 bytes of data, and then tries
to read 10 byes twice.  Previously the second read would
sleep indifinitely, since the record was already decrypted
and there is only 5 bytes left, not full 10.
Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Reviewed-by: Dirk van der Merwe <dirk.vandermerwe@netronome.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

When tls_sw_recvmsg() partially copies a record it pops that
record from ctx->recv_pkt and places it on rx_list.

Next iteration of tls_sw_recvmsg() reads from rx_list via
process_rx_list() before it enters the decryption loop.
If there is no more records to be read tls_wait_data()
will put the process on the wait queue and got to sleep.
This is incorrect, because some data was already copied
in process_rx_list().

In case of RPC connections process may never get woken up,
because peer also simply blocks in read().

I think this may also fix a similar issue when BPF is at
play, because after __tcp_bpf_recvmsg() returns some data
we subtract it from len and use continue to restart the
loop, but len could have just reached 0, so again we'd
sleep unnecessarily. That's added by:
commit d3b18ad3 ("tls: add bpf support to sk_msg handling")

Fixes: 692d7b5d ("tls: Fix recvmsg() to be able to peek across multiple records")
Reported-by: David Beckett <david.beckett@netronome.com>
Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Reviewed-by: Dirk van der Merwe <dirk.vandermerwe@netronome.com>
Tested-by: David Beckett <david.beckett@netronome.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Set SO_RCVLOWAT and test it gets respected when gathering
data from multiple records.
Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Reviewed-by: Dirk van der Merwe <dirk.vandermerwe@netronome.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

If some of the data came from the previous record, i.e. from
the rx_list it had already been decrypted, so it's not counted
towards the "decrypted" variable, but the "copied" variable.
Take that into account when checking lowat.

When calculating lowat target we need to pass the original len.
E.g. if lowat is at 80, len is 100 and we had 30 bytes on rx_list
target would currently be incorrectly calculated as 70, even though
we only need 50 more bytes to make up the 80.

Fixes: 692d7b5d ("tls: Fix recvmsg() to be able to peek across multiple records")
Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Reviewed-by: Dirk van der Merwe <dirk.vandermerwe@netronome.com>
Tested-by: David Beckett <david.beckett@netronome.com>
Signed-off-by: David S. Miller <davem@davemloft.net>