Yonglong Liu [Wed, 3 Jul 2019 11:12:30 +0000 (19:12 +0800)]
net: hns: add support for vlan TSO
The hip07 chip support vlan TSO, this patch adds NETIF_F_TSO
and NETIF_F_TSO6 flags to vlan_features to improve the
performance after adding vlan to the net ports.
Xin Long [Wed, 3 Jul 2019 10:20:20 +0000 (18:20 +0800)]
sctp: count data bundling sack chunk for outctrlchunks
Now all ctrl chunks are counted for asoc stats.octrlchunks and net
SCTP_MIB_OUTCTRLCHUNKS either after queuing up or bundling, other
than the chunk maked and bundled in sctp_packet_bundle_sack, which
caused 'outctrlchunks' not consistent with 'inctrlchunks' in peer.
This issue exists since very beginning, here to fix it by increasing
both net SCTP_MIB_OUTCTRLCHUNKS and asoc stats.octrlchunks when sack
chunk is maked and bundled in sctp_packet_bundle_sack.
Hayes Wang [Wed, 3 Jul 2019 07:11:56 +0000 (15:11 +0800)]
r8152: move calling r8153b_rx_agg_chg_indicate()
r8153b_rx_agg_chg_indicate() needs to be called after enabling TX/RX and
before calling rxdy_gated_en(tp, false). Otherwise, the change of the
settings of RX aggregation wouldn't work.
Besides, adjust rtl8152_set_coalesce() for the same reason. If
rx_coalesce_usecs is changed, restart TX/RX to let the setting work.
net: don't warn in inet diag when IPV6 is disabled
If IPV6 was disabled, then ss command would cause a kernel warning
because the command was attempting to dump IPV6 socket information.
The fix is to just remove the warning.
Bugzilla: https://bugzilla.kernel.org/show_bug.cgi?id=202249 Fixes: 432490f9d455 ("net: ip, diag -- Add diag interface for raw sockets") Signed-off-by: Stephen Hemminger <[email protected]> Signed-off-by: David S. Miller <[email protected]>
Ilya Maximets [Fri, 28 Jun 2019 08:04:07 +0000 (11:04 +0300)]
xdp: fix hang while unregistering device bound to xdp socket
Device that bound to XDP socket will not have zero refcount until the
userspace application will not close it. This leads to hang inside
'netdev_wait_allrefs()' if device unregistering requested:
# ip link del p1
< hang on recvmsg on netlink socket >
# ps -x | grep ip
5126 pts/0 D+ 0:00 ip link del p1
# journalctl -b
Jun 05 07:19:16 kernel:
unregister_netdevice: waiting for p1 to become free. Usage count = 1
Jun 05 07:19:27 kernel:
unregister_netdevice: waiting for p1 to become free. Usage count = 1
...
Fix that by implementing NETDEV_UNREGISTER event notification handler
to properly clean up all the resources and unref device.
This should also allow socket killing via ss(8) utility.
Ilya Maximets [Fri, 28 Jun 2019 08:04:06 +0000 (11:04 +0300)]
xdp: hold device for umem regardless of zero-copy mode
Device pointer stored in umem regardless of zero-copy mode,
so we heed to hold the device in all cases.
Fixes: c9b47cc1fabc ("xsk: fix bug when trying to use both copy and zero-copy on one queue id") Signed-off-by: Ilya Maximets <[email protected]> Acked-by: Jonathan Lemon <[email protected]> Signed-off-by: Daniel Borkmann <[email protected]>
Selftests are reporting this failure in test_lwt_seg6local.sh:
+ ip netns exec ns2 ip -6 route add fb00::6 encap bpf in obj test_lwt_seg6local.o sec encap_srh dev veth2
Error fetching program/map!
Failed to parse eBPF program: Operation not permitted
The problem is __attribute__((always_inline)) alone is not enough to prevent
clang from inserting those functions in .text. In that case, .text is not
marked as relocateable.
See the output of objdump -h test_lwt_seg6local.o:
This causes the iproute bpf loader to fail in bpf_fetch_prog_sec:
bpf_has_call_data returns true but bpf_fetch_prog_relo fails as there's no
relocateable .text section in the file.
To fix this, convert to 'static __always_inline'.
v2: Use 'static __always_inline' instead of 'static inline
__attribute__((always_inline))'
Fixes: c99a84eac026 ("selftests/bpf: test for seg6local End.BPF action") Signed-off-by: Jiri Benc <[email protected]> Acked-by: Yonghong Song <[email protected]> Signed-off-by: Daniel Borkmann <[email protected]>
Luke Nelson [Sat, 29 Jun 2019 05:57:51 +0000 (22:57 -0700)]
selftests: bpf: add tests for shifts by zero
There are currently no tests for ALU64 shift operations when the shift
amount is 0. This adds 6 new tests to make sure they are equivalent
to a no-op. The x32 JIT had such bugs that could have been caught by
these tests.
Luke Nelson [Sat, 29 Jun 2019 05:57:50 +0000 (22:57 -0700)]
bpf, x32: Fix bug with ALU64 {LSH, RSH, ARSH} BPF_K shift by 0
The current x32 BPF JIT does not correctly compile shift operations when
the immediate shift amount is 0. The expected behavior is for this to
be a no-op.
The following program demonstrates the bug. The expexceted result is 1,
but the current JITed code returns 2.
r0 = 1
r1 = 1
r1 <<= 0
if r1 == 1 goto end
r0 = 2
end:
exit
Luke Nelson [Sat, 29 Jun 2019 05:57:49 +0000 (22:57 -0700)]
bpf, x32: Fix bug with ALU64 {LSH, RSH, ARSH} BPF_X shift by 0
The current x32 BPF JIT for shift operations is not correct when the
shift amount in a register is 0. The expected behavior is a no-op, whereas
the current implementation changes bits in the destination register.
The following example demonstrates the bug. The expected result of this
program is 1, but the current JITed code returns 2.
The bug is caused by an incorrect assumption by the JIT that a shift by
32 clear the register. On x32 however, shifts use the lower 5 bits of
the source, making a shift by 32 equivalent to a shift by 0.
This patch fixes the bug using double-precision shifts, which also
simplifies the code.
Nicolas Dichtel [Tue, 2 Jul 2019 15:51:39 +0000 (17:51 +0200)]
xfrm interface: fix memory leak on creation
The following commands produce a backtrace and return an error but the xfrm
interface is created (in the wrong netns):
$ ip netns add foo
$ ip netns add bar
$ ip -n foo netns set bar 0
$ ip -n foo link add xfrmi0 link-netnsid 0 type xfrm dev lo if_id 23
RTNETLINK answers: Invalid argument
$ ip -n bar link ls xfrmi0
2: xfrmi0@lo: <NOARP,M-DOWN> mtu 1500 qdisc noop state DOWN mode DEFAULT group default qlen 1000
link/none 00:00:00:00:00:00 brd 00:00:00:00:00:00
In fact, each time a xfrm interface was created, a netdev was allocated
by __rtnl_newlink()/rtnl_create_link() and then another one by
xfrmi_newlink()/xfrmi_create(). Only the second one was registered, it's
why the previous commands produce a backtrace: dev_change_net_namespace()
was called on a netdev with reg_state set to NETREG_UNINITIALIZED (the
first one).
xfrm: policy: fix bydst hlist corruption on hash rebuild
syzbot reported following spat:
BUG: KASAN: use-after-free in __write_once_size include/linux/compiler.h:221
BUG: KASAN: use-after-free in hlist_del_rcu include/linux/rculist.h:455
BUG: KASAN: use-after-free in xfrm_hash_rebuild+0xa0d/0x1000 net/xfrm/xfrm_policy.c:1318
Write of size 8 at addr ffff888095e79c00 by task kworker/1:3/8066
Workqueue: events xfrm_hash_rebuild
Call Trace:
__write_once_size include/linux/compiler.h:221 [inline]
hlist_del_rcu include/linux/rculist.h:455 [inline]
xfrm_hash_rebuild+0xa0d/0x1000 net/xfrm/xfrm_policy.c:1318
process_one_work+0x814/0x1130 kernel/workqueue.c:2269
Allocated by task 8064:
__kmalloc+0x23c/0x310 mm/slab.c:3669
kzalloc include/linux/slab.h:742 [inline]
xfrm_hash_alloc+0x38/0xe0 net/xfrm/xfrm_hash.c:21
xfrm_policy_init net/xfrm/xfrm_policy.c:4036 [inline]
xfrm_net_init+0x269/0xd60 net/xfrm/xfrm_policy.c:4120
ops_init+0x336/0x420 net/core/net_namespace.c:130
setup_net+0x212/0x690 net/core/net_namespace.c:316
The faulting address is the address of the old chain head,
free'd by xfrm_hash_resize().
In xfrm_hash_rehash(), chain heads get re-initialized without
any hlist_del_rcu:
for (i = hmask; i >= 0; i--)
INIT_HLIST_HEAD(odst + i);
Then, hlist_del_rcu() gets called on the about to-be-reinserted policy
when iterating the per-net list of policies.
hlist_del_rcu() will then make chain->first be nonzero again:
static inline void __hlist_del(struct hlist_node *n)
{
struct hlist_node *next = n->next; // address of next element in list
struct hlist_node **pprev = n->pprev;// location of previous elem, this
// can point at chain->first
WRITE_ONCE(*pprev, next); // chain->first points to next elem
if (next)
next->pprev = pprev;
Then, when we walk chainlist to find insertion point, we may find a
non-empty list even though we're supposedly reinserting the first
policy to an empty chain.
To fix this first unlink all exact and inexact policies instead of
zeroing the list heads.
Add the commands equivalent to the syzbot reproducer to xfrm_policy.sh,
without fix KASAN catches the corruption as it happens, SLUB poisoning
detects it a bit later.
Reported-by: [email protected] Fixes: 1548bc4e0512 ("xfrm: policy: delete inexact policies from inexact list on hash rebuild") Signed-off-by: Florian Westphal <[email protected]> Signed-off-by: Steffen Klassert <[email protected]>
Po-Hsu Lin [Mon, 1 Jul 2019 04:40:31 +0000 (12:40 +0800)]
selftests/net: skip psock_tpacket test if KALLSYMS was not enabled
The psock_tpacket test will need to access /proc/kallsyms, this would
require the kernel config CONFIG_KALLSYMS to be enabled first.
Apart from adding CONFIG_KALLSYMS to the net/config file here, check the
file existence to determine if we can run this test will be helpful to
avoid a false-positive test result when testing it directly with the
following commad against a kernel that have CONFIG_KALLSYMS disabled:
make -C tools/testing/selftests TARGETS=net run_tests
David Howells [Tue, 2 Jul 2019 15:04:19 +0000 (16:04 +0100)]
rxrpc: Fix oops in tracepoint
If the rxrpc_eproto tracepoint is enabled, an oops will be cause by the
trace line that rxrpc_extract_header() tries to emit when a protocol error
occurs (typically because the packet is short) because the call argument is
NULL.
Fix this by using ?: to assume 0 as the debug_id if call is NULL.
Fixes: a25e21f0bcd2 ("rxrpc, afs: Use debug_ids rather than pointers in traces") Signed-off-by: David Howells <[email protected]> Reviewed-by: Marc Dionne <[email protected]> Signed-off-by: David S. Miller <[email protected]>
====================
macsec: fix some bugs in the receive path
This series fixes some bugs in the receive path of macsec. The first
is a use after free when processing macsec frames with a SecTAG that
has the TCI E bit set but the C bit clear. In the 2nd bug, the driver
leaves an invalid checksumming state after decrypting the packet.
This is a combined effort of Sabrina Dubroca <[email protected]> and me.
====================
David Howells [Tue, 2 Jul 2019 14:59:12 +0000 (15:59 +0100)]
rxrpc: Fix send on a connected, but unbound socket
If sendmsg() or sendmmsg() is called on a connected socket that hasn't had
bind() called on it, then an oops will occur when the kernel tries to
connect the call because no local endpoint has been allocated.
Fix this by implicitly binding the socket if it is in the
RXRPC_CLIENT_UNBOUND state, just like it does for the RXRPC_UNBOUND state.
Further, the state should be transitioned to RXRPC_CLIENT_BOUND after this
to prevent further attempts to bind it.
David S. Miller [Tue, 2 Jul 2019 18:53:03 +0000 (11:53 -0700)]
Merge branch 'bridge-stale-ptrs'
Nikolay Aleksandrov says:
====================
net: bridge: fix possible stale skb pointers
In the bridge driver we have a couple of places which call pskb_may_pull
but we've cached skb pointers before that and use them after which can
lead to out-of-bounds/stale pointer use. I've had these in my "to fix"
list for some time and now we got a report (patch 01) so here they are.
Patches 02-04 are fixes based on code inspection. Also patch 01 was
tested by Martin Weinelt, Martin if you don't mind please add your
tested-by tag to it by replying with Tested-by: name <email>.
I've also briefly tested the set by trying to exercise those code paths.
====================
net: bridge: stp: don't cache eth dest pointer before skb pull
Don't cache eth dest pointer before calling pskb_may_pull.
Fixes: cf0f02d04a83 ("[BRIDGE]: use llc for receiving STP packets") Signed-off-by: Nikolay Aleksandrov <[email protected]> Signed-off-by: David S. Miller <[email protected]>
net: bridge: don't cache ether dest pointer on input
We would cache ether dst pointer on input in br_handle_frame_finish but
after the neigh suppress code that could lead to a stale pointer since
both ipv4 and ipv6 suppress code do pskb_may_pull. This means we have to
always reload it after the suppress code so there's no point in having
it cached just retrieve it directly.
Fixes: 057658cb33fbf ("bridge: suppress arp pkts on BR_NEIGH_SUPPRESS ports") Fixes: ed842faeb2bd ("bridge: suppress nd pkts on BR_NEIGH_SUPPRESS ports") Signed-off-by: Nikolay Aleksandrov <[email protected]> Signed-off-by: David S. Miller <[email protected]>
net: bridge: mcast: fix stale ipv6 hdr pointer when handling v6 query
We get a pointer to the ipv6 hdr in br_ip6_multicast_query but we may
call pskb_may_pull afterwards and end up using a stale pointer.
So use the header directly, it's just 1 place where it's needed.
Fixes: 08b202b67264 ("bridge br_multicast: IPv6 MLD support.") Signed-off-by: Nikolay Aleksandrov <[email protected]> Tested-by: Martin Weinelt <[email protected]> Signed-off-by: David S. Miller <[email protected]>
net: bridge: mcast: fix stale nsrcs pointer in igmp3/mld2 report handling
We take a pointer to grec prior to calling pskb_may_pull and use it
afterwards to get nsrcs so record nsrcs before the pull when handling
igmp3 and we get a pointer to nsrcs and call pskb_may_pull when handling
mld2 which again could lead to reading 2 bytes out-of-bounds.
==================================================================
BUG: KASAN: use-after-free in br_multicast_rcv+0x480c/0x4ad0 [bridge]
Read of size 2 at addr ffff8880421302b4 by task ksoftirqd/1/16
Fixes: bc8c20acaea1 ("bridge: multicast: treat igmpv3 report with INCLUDE and no sources as a leave") Reported-by: Martin Weinelt <[email protected]> Signed-off-by: Nikolay Aleksandrov <[email protected]> Tested-by: Martin Weinelt <[email protected]> Signed-off-by: David S. Miller <[email protected]>
Hayes Wang [Mon, 1 Jul 2019 07:53:19 +0000 (15:53 +0800)]
r8152: fix the setting of detecting the linking change for runtime suspend
1. Rename r8153b_queue_wake() to r8153_queue_wake().
2. Correct the setting. The enable bit should be 0xd38c bit 0. Besides,
the 0xd38a bit 0 and 0xd398 bit 8 have to be cleared for both enabled
and disabled.
Jakub Kicinski [Fri, 28 Jun 2019 23:11:39 +0000 (16:11 -0700)]
net/tls: make sure offload also gets the keys wiped
Commit 86029d10af18 ("tls: zero the crypto information from tls_context
before freeing") added memzero_explicit() calls to clear the key material
before freeing struct tls_context, but it missed tls_device.c has its
own way of freeing this structure. Replace the missing free.
Fixes: 86029d10af18 ("tls: zero the crypto information from tls_context before freeing") Signed-off-by: Jakub Kicinski <[email protected]> Reviewed-by: Dirk van der Merwe <[email protected]> Signed-off-by: David S. Miller <[email protected]>
Jakub Kicinski [Fri, 28 Jun 2019 23:07:59 +0000 (16:07 -0700)]
net/tls: reject offload of TLS 1.3
Neither drivers nor the tls offload code currently supports TLS
version 1.3. Check the TLS version when installing connection
state. TLS 1.3 will just fallback to the kernel crypto for now.
Fixes: 130b392c6cd6 ("net: tls: Add tls 1.3 support") Signed-off-by: Jakub Kicinski <[email protected]> Reviewed-by: Dirk van der Merwe <[email protected]> Signed-off-by: David S. Miller <[email protected]>
====================
idr: fix overflow cases on 32-bit CPU
idr_get_next_ul() is problematic by design, it can't handle
the following overflow case well on 32-bit CPU:
u32 id = UINT_MAX;
idr_alloc_u32(&id);
while (idr_get_next_ul(&id) != NULL)
id++;
when 'id' overflows and becomes 0 after UINT_MAX, the loop
goes infinite.
Fix this by eliminating external users of idr_get_next_ul()
and migrating them to idr_for_each_entry_continue_ul(). And
add an additional parameter for these iteration macros to detect
overflow properly.
Please merge this through networking tree, as all the users
are in networking subsystem.
====================
Cong Wang [Fri, 28 Jun 2019 18:03:42 +0000 (11:03 -0700)]
idr: introduce idr_for_each_entry_continue_ul()
Similarly, other callers of idr_get_next_ul() suffer the same
overflow bug as they don't handle it properly either.
Introduce idr_for_each_entry_continue_ul() to help these callers
iterate from a given ID.
cls_flower needs more care here because it still has overflow when
does arg->cookie++, we have to fold its nested loops into one
and remove the arg->cookie++.
Cong Wang [Fri, 28 Jun 2019 18:03:41 +0000 (11:03 -0700)]
idr: fix overflow case for idr_for_each_entry_ul()
idr_for_each_entry_ul() is buggy as it can't handle overflow
case correctly. When we have an ID == UINT_MAX, it becomes an
infinite loop. This happens when running on 32-bit CPU where
unsigned long has the same size with unsigned int.
There is no better way to fix this than casting it to a larger
integer, but we can't just 64 bit integer on 32 bit CPU. Instead
we could just use an additional integer to help us to detect this
overflow case, that is, adding a new parameter to this macro.
Fortunately tc action is its only user right now.
David S. Miller [Tue, 2 Jul 2019 02:09:07 +0000 (19:09 -0700)]
Merge branch 'vsock-virtio-fixes'
Stefano Garzarella says:
====================
vsock/virtio: several fixes in the .probe() and .remove()
During the review of "[PATCH] vsock/virtio: Initialize core virtio vsock
before registering the driver", Stefan pointed out some possible issues
in the .probe() and .remove() callbacks of the virtio-vsock driver.
This series tries to solve these issues:
- Patch 1 adds RCU critical sections to avoid use-after-free of
'the_virtio_vsock' pointer.
- Patch 2 stops workers before to call vdev->config->reset(vdev) to
be sure that no one is accessing the device.
- Patch 3 moves the works flush at the end of the .remove() to avoid
use-after-free of 'vsock' object.
v2:
- Patch 1: use RCU to protect 'the_virtio_vsock' pointer
- Patch 2: no changes
- Patch 3: flush works only at the end of .remove()
- Removed patch 4 because virtqueue_detach_unused_buf() returns all the buffers
allocated.
vsock/virtio: fix flush of works during the .remove()
This patch moves the flush of works after vdev->config->del_vqs(vdev),
because we need to be sure that no workers run before to free the
'vsock' object.
Since we stopped the workers using the [tx|rx|event]_run flags,
we are sure no one is accessing the device while we are calling
vdev->config->reset(vdev), so we can safely move the workers' flush.
Before the vdev->config->del_vqs(vdev), workers can be scheduled
by VQ callbacks, so we must flush them after del_vqs(), to avoid
use-after-free of 'vsock' object.
Before to call vdev->config->reset(vdev) we need to be sure that
no one is accessing the device, for this reason, we add new variables
in the struct virtio_vsock to stop the workers during the .remove().
This patch also add few comments before vdev->config->reset(vdev)
and vdev->config->del_vqs(vdev).
vsock/virtio: use RCU to avoid use-after-free on the_virtio_vsock
Some callbacks used by the upper layers can run while we are in the
.remove(). A potential use-after-free can happen, because we free
the_virtio_vsock without knowing if the callbacks are over or not.
To solve this issue we move the assignment of the_virtio_vsock at the
end of .probe(), when we finished all the initialization, and at the
beginning of .remove(), before to release resources.
For the same reason, we do the same also for the vdev->priv.
We use RCU to be sure that all callbacks that use the_virtio_vsock
ended before freeing it. This is not required for callbacks that
use vdev->priv, because after the vdev->config->del_vqs() we are sure
that they are ended and will no longer be invoked.
We also take the mutex during the .remove() to avoid that .probe() can
run while we are resetting the device.
Taehee Yoo [Fri, 28 Jun 2019 05:07:25 +0000 (14:07 +0900)]
vxlan: do not destroy fdb if register_netdevice() is failed
__vxlan_dev_create() destroys FDB using specific pointer which indicates
a fdb when error occurs.
But that pointer should not be used when register_netdevice() fails because
register_netdevice() internally destroys fdb when error occurs.
This patch makes vxlan_fdb_create() to do not link fdb entry to vxlan dev
internally.
Instead, a new function vxlan_fdb_insert() is added to link fdb to vxlan
dev.
vxlan_fdb_insert() is called after calling register_netdevice().
This routine can avoid situation that ->ndo_uninit() destroys fdb entry
in error path of register_netdevice().
Hence, error path of __vxlan_dev_create() routine can have an opportunity
to destroy default fdb entry by hand.
Test command
ip link add bonding_masters type vxlan id 0 group 239.1.1.1 \
dev enp0s9 dstport 4789
sctp: fix error handling on stream scheduler initialization
It allocates the extended area for outbound streams only on sendmsg
calls, if they are not yet allocated. When using the priority
stream scheduler, this initialization may imply into a subsequent
allocation, which may fail. In this case, it was aborting the stream
scheduler initialization but leaving the ->ext pointer (allocated) in
there, thus in a partially initialized state. On a subsequent call to
sendmsg, it would notice the ->ext pointer in there, and trip on
uninitialized stuff when trying to schedule the data chunk.
The fix is undo the ->ext initialization if the stream scheduler
initialization fails and avoid the partially initialized state.
Although syzkaller bisected this to commit 4ff40b86262b ("sctp: set
chunk transport correctly when it's a new asoc"), this bug was actually
introduced on the commit I marked below.
Cong Wang [Thu, 27 Jun 2019 21:30:58 +0000 (14:30 -0700)]
netrom: fix a memory leak in nr_rx_frame()
When the skb is associated with a new sock, just assigning
it to skb->sk is not sufficient, we have to set its destructor
to free the sock properly too.
Avoid the situation where an IPV6 only flag is applied to an IPv4 address:
# ip addr add 192.0.2.1/24 dev dummy0 nodad home mngtmpaddr noprefixroute
# ip -4 addr show dev dummy0
2: dummy0: <BROADCAST,NOARP,UP,LOWER_UP> mtu 1500 qdisc noqueue state UNKNOWN group default qlen 1000
inet 192.0.2.1/24 scope global noprefixroute dummy0
valid_lft forever preferred_lft forever
Or worse, by sending a malicious netlink command:
# ip -4 addr show dev dummy0
2: dummy0: <BROADCAST,NOARP,UP,LOWER_UP> mtu 1500 qdisc noqueue state UNKNOWN group default qlen 1000
inet 192.0.2.1/24 scope global nodad optimistic dadfailed home tentative mngtmpaddr noprefixroute stable-privacy dummy0
valid_lft forever preferred_lft forever
John Hurley [Thu, 27 Jun 2019 13:37:30 +0000 (14:37 +0100)]
net: openvswitch: fix csum updates for MPLS actions
Skbs may have their checksum value populated by HW. If this is a checksum
calculated over the entire packet then the CHECKSUM_COMPLETE field is
marked. Changes to the data pointer on the skb throughout the network
stack still try to maintain this complete csum value if it is required
through functions such as skb_postpush_rcsum.
The MPLS actions in Open vSwitch modify a CHECKSUM_COMPLETE value when
changes are made to packet data without a push or a pull. This occurs when
the ethertype of the MAC header is changed or when MPLS lse fields are
modified.
The modification is carried out using the csum_partial function to get the
csum of a buffer and add it into the larger checksum. The buffer is an
inversion of the data to be removed followed by the new data. Because the
csum is calculated over 16 bits and these values align with 16 bits, the
effect is the removal of the old value from the CHECKSUM_COMPLETE and
addition of the new value.
However, the csum fed into the function and the outcome of the
calculation are also inverted. This would only make sense if it was the
new value rather than the old that was inverted in the input buffer.
Fix the issue by removing the bit inverts in the csum_partial calculation.
The bug was verified and the fix tested by comparing the folded value of
the updated CHECKSUM_COMPLETE value with the folded value of a full
software checksum calculation (reset skb->csum to 0 and run
skb_checksum_complete(skb)). Prior to the fix the outcomes differed but
after they produce the same result.
Fixes: 25cd9ba0abc0 ("openvswitch: Add basic MPLS support to kernel") Fixes: bc7cc5999fd3 ("openvswitch: update checksum in {push,pop}_mpls") Signed-off-by: John Hurley <[email protected]> Reviewed-by: Jakub Kicinski <[email protected]> Reviewed-by: Simon Horman <[email protected]> Acked-by: Pravin B Shelar <[email protected]> Signed-off-by: David S. Miller <[email protected]>
David S. Miller [Sun, 30 Jun 2019 23:00:26 +0000 (16:00 -0700)]
Merge branch 'bnxt_en-Bug-fixes'
Michael Chan says:
====================
bnxt_en: Bug fixes.
Miscellaneous bug fix patches, including two resource handling fixes for
the RDMA driver, a PCI shutdown patch to add pci_disable_device(), a patch
to fix ethtool selftest crash, and the last one suppresses an unnecessry
error message.
Please also queue patches 1, 2, and 3 for -stable. Thanks.
====================
Michael Chan [Sat, 29 Jun 2019 15:16:48 +0000 (11:16 -0400)]
bnxt_en: Suppress error messages when querying DSCP DCB capabilities.
Some firmware versions do not support this so use the silent variant
to send the message to firmware to suppress the harmless error. This
error message is unnecessarily alarming the user.
Fixes: afdc8a84844a ("bnxt_en: Add DCBNL DSCP application protocol support.") Signed-off-by: Michael Chan <[email protected]> Signed-off-by: David S. Miller <[email protected]>
Michael Chan [Sat, 29 Jun 2019 15:16:47 +0000 (11:16 -0400)]
bnxt_en: Cap the returned MSIX vectors to the RDMA driver.
In an earlier commit to improve NQ reservations on 57500 chips, we
set the resv_irqs on the 57500 VFs to the fixed value assigned by
the PF regardless of how many are actually used. The current
code assumes that resv_irqs minus the ones used by the network driver
must be the ones for the RDMA driver. This is no longer true and
we may return more MSIX vectors than requested, causing inconsistency.
Fix it by capping the value.
Fixes: 01989c6b69d9 ("bnxt_en: Improve NQ reservations.") Signed-off-by: Michael Chan <[email protected]> Signed-off-by: David S. Miller <[email protected]>
Michael Chan [Sat, 29 Jun 2019 15:16:46 +0000 (11:16 -0400)]
bnxt_en: Fix statistics context reservation logic for RDMA driver.
The current logic assumes that the RDMA driver uses one statistics
context adjacent to the ones used by the network driver. This
assumption is not true and the statistics context used by the
RDMA driver is tied to its MSIX base vector. This wrong assumption
can cause RDMA driver failure after changing ethtool rings on the
network side. Fix the statistics reservation logic accordingly.
Fixes: 780baad44f0f ("bnxt_en: Reserve 1 stat_ctx for RDMA driver.") Signed-off-by: Michael Chan <[email protected]> Signed-off-by: David S. Miller <[email protected]>
Michael Chan [Sat, 29 Jun 2019 15:16:45 +0000 (11:16 -0400)]
bnxt_en: Fix ethtool selftest crash under error conditions.
After ethtool loopback packet tests, we re-open the nic for the next
IRQ test. If the open fails, we must not proceed with the IRQ test
or we will crash with NULL pointer dereference. Fix it by checking
the bnxt_open_nic() return code before proceeding.
Reported-by: Somasundaram Krishnasamy <[email protected]> Fixes: 67fea463fd87 ("bnxt_en: Add interrupt test to ethtool -t selftest.") Signed-off-by: Michael Chan <[email protected]> Signed-off-by: David S. Miller <[email protected]>
Michael Chan [Sat, 29 Jun 2019 15:16:44 +0000 (11:16 -0400)]
bnxt_en: Disable bus master during PCI shutdown and driver unload.
Some chips with older firmware can continue to perform DMA read from
context memory even after the memory has been freed. In the PCI shutdown
method, we need to call pci_disable_device() to shutdown DMA to prevent
this DMA before we put the device into D3hot. DMA memory request in
D3hot state will generate PCI fatal error. Similarly, in the driver
remove method, the context memory should only be freed after DMA has
been shutdown for correctness.
Fixes: 98f04cf0f1fc ("bnxt_en: Check context memory requirements from firmware.") Signed-off-by: Michael Chan <[email protected]> Signed-off-by: David S. Miller <[email protected]>
bnx2x: Prevent ptp_task to be rescheduled indefinitely
Currently bnx2x ptp worker tries to read a register with timestamp
information in case of TX packet timestamping and in case it fails,
the routine reschedules itself indefinitely. This was reported as a
kworker always at 100% of CPU usage, which was narrowed down to be
bnx2x ptp_task.
By following the ioctl handler, we could narrow down the problem to
an NTP tool (chrony) requesting HW timestamping from bnx2x NIC with
RX filter zeroed; this isn't reproducible for example with ptp4l
(from linuxptp) since this tool requests a supported RX filter.
It seems NIC FW timestamp mechanism cannot work well with
RX_FILTER_NONE - driver's PTP filter init routine skips a register
write to the adapter if there's not a supported filter request.
This patch addresses the problem of bnx2x ptp thread's everlasting
reschedule by retrying the register read 10 times; between the read
attempts the thread sleeps for an increasing amount of time starting
in 1ms to give FW some time to perform the timestamping. If it still
fails after all retries, we bail out in order to prevent an unbound
resource consumption from bnx2x.
The patch also adds an ethtool statistic for accounting the skipped
TX timestamp packets and it reduces the priority of timestamping
error messages to prevent log flooding. The code was tested using
both linuxptp and chrony.
David S. Miller [Sat, 29 Jun 2019 18:09:18 +0000 (11:09 -0700)]
Merge branch 'Sub-ns-increment-fixes-in-Macb-PTP'
Harini Katakam says:
====================
Sub ns increment fixes in Macb PTP
The subns increment register fields are not captured correctly in the
driver. Fix the same and also increase the subns incr resolution.
Sub ns resolution was increased to 24 bits in r1p06f2 version. To my
knowledge, this PTP driver, with its current BD time stamp
implementation, is only useful to that version or above. So, I have
increased the resolution unconditionally. Please let me know if there
is any IP versions incompatible with this - there is no register to
obtain this information from.
Harini Katakam [Thu, 27 Jun 2019 06:21:00 +0000 (11:51 +0530)]
net: macb: Fix SUBNS increment and increase resolution
The subns increment register has 24 bits as follows:
RegBit[15:0] = Subns[23:8]; RegBit[31:24] = Subns[7:0]
Fix the same in the driver and increase sub ns resolution to the
best capable, 24 bits. This should be the case on all GEM versions
that this PTP driver supports.
Harini Katakam [Thu, 27 Jun 2019 06:20:59 +0000 (11:50 +0530)]
net: macb: Add separate definition for PPM fraction
The scaled ppm parameter passed to _adjfine() contains a 16 bit
fraction. This just happens to be the same as SUBNSINCR_SIZE now.
Hence define this separately.
Florian Westphal [Wed, 26 Jun 2019 18:40:45 +0000 (20:40 +0200)]
net: make skb_dst_force return true when dst is refcounted
netfilter did not expect that skb_dst_force() can cause skb to lose its
dst entry.
I got a bug report with a skb->dst NULL dereference in netfilter
output path. The backtrace contains nf_reinject(), so the dst might have
been cleared when skb got queued to userspace.
Other users were fixed via
if (skb_dst(skb)) {
skb_dst_force(skb);
if (!skb_dst(skb))
goto handle_err;
}
But I think its preferable to make the 'dst might be cleared' part
of the function explicit.
In netfilter case, skb with a null dst is expected when queueing in
prerouting hook, so drop skb for the other hooks.
v2:
v1 of this patch returned true in case skb had no dst entry.
Eric said:
Say if we have two skb_dst_force() calls for some reason
on the same skb, only the first one will return false.
This now returns false even when skb had no dst, as per Erics
suggestion, so callers might need to check skb_dst() first before
skb_dst_force().
Xin Long [Wed, 26 Jun 2019 08:31:39 +0000 (16:31 +0800)]
sctp: not bind the socket in sctp_connect
Now when sctp_connect() is called with a wrong sa_family, it binds
to a port but doesn't set bp->port, then sctp_get_af_specific will
return NULL and sctp_connect() returns -EINVAL.
Then if sctp_bind() is called to bind to another port, the last
port it has bound will leak due to bp->port is NULL by then.
sctp_connect() doesn't need to bind ports, as later __sctp_connect
will do it if bp->port is NULL. So remove it from sctp_connect().
While at it, remove the unnecessary sockaddr.sa_family len check
as it's already done in sctp_inet_connect.
Fixes: 644fbdeacf1d ("sctp: fix the issue that flags are ignored when using kernel_connect") Reported-by: [email protected] Signed-off-by: Xin Long <[email protected]> Acked-by: Marcelo Ricardo Leitner <[email protected]> Signed-off-by: David S. Miller <[email protected]>
Baruch Siach [Fri, 28 Jun 2019 04:08:45 +0000 (07:08 +0300)]
bpf: fix uapi bpf_prog_info fields alignment
Merge commit 1c8c5a9d38f60 ("Merge
git://git.kernel.org/pub/scm/linux/kernel/git/davem/net-next") undid the
fix from commit 36f9814a494 ("bpf: fix uapi hole for 32 bit compat
applications") by taking the gpl_compatible 1-bit field definition from
commit b85fab0e67b162 ("bpf: Add gpl_compatible flag to struct
bpf_prog_info") as is. That breaks architectures with 16-bit alignment
like m68k. Add 31-bit pad after gpl_compatible to restore alignment of
following fields.
Thanks to Dmitry V. Levin his analysis of this bug history.
net: mvpp2: prs: Don't override the sign bit in SRAM parser shift
The Header Parser allows identifying various fields in the packet
headers, used for various kind of filtering and classification
steps.
This is a re-entrant process, where the offset in the packet header
depends on the previous lookup results. This offset is represented in
the SRAM results of the TCAM, as a shift to be operated.
This shift can be negative in some cases, such as in IPv6 parsing.
This commit prevents overriding the sign bit when setting the shift
value, which could cause instabilities when parsing IPv6 flows.
Fixes: 3f518509dedc ("ethernet: Add new driver for Marvell Armada 375 network unit") Suggested-by: Alan Winkowski <[email protected]> Signed-off-by: Maxime Chevallier <[email protected]> Signed-off-by: David S. Miller <[email protected]>
Russell King [Tue, 25 Jun 2019 09:44:33 +0000 (10:44 +0100)]
net: phylink: further documentation clarifications
Clarify the validate() behaviour in a few cases which weren't mentioned
in the documentation, but which are necessary for users to get the
correct behaviour.
Marek Vasut [Sun, 23 Jun 2019 12:11:43 +0000 (14:11 +0200)]
net: ethernet: ti: cpsw: Assign OF node to slave devices
Assign OF node to CPSW slave devices, otherwise it is not possible to
bind e.g. DSA switch to them. Without this patch, the DSA code tries
to find the ethernet device by OF match, but fails to do so because
the slave device has NULL OF node.
A b53 device may configured through an external EEPROM like the switch
device on the Lamobo R1 router board. The configuration of a port may
therefore differ from the reset configuration of the switch.
The switch configuration reported by the DSA subsystem is different until
the port is configured by DSA i.e. a port can be active, while the DSA
subsystem reports the port is inactive. Disable all ports and not only
the unused ones to put all ports into a well defined state.
We should have got the following reply from command line, but we did not.
From 127.0.0.1 icmp_seq=1 Destination Port Unreachable
Yi Zhao reported it and narrowed it down to: 7fc38225363d ("netfilter: reject: skip csum verification for protocols that don't support it"),
This is because nf_ip_checksum still expects pseudo-header protocol type 0 for
packets that are of neither TCP or UDP, and thus ICMP packets are mistakenly
treated as TCP/UDP.
This patch corrects the conditions in nf_ip_checksum and all other places that
still call it with protocol 0.
Fixes: 7fc38225363d ("netfilter: reject: skip csum verification for protocols that don't support it") Reported-by: Yi Zhao <[email protected]> Signed-off-by: He Zhe <[email protected]> Signed-off-by: Pablo Neira Ayuso <[email protected]>
Linus Torvalds [Fri, 28 Jun 2019 00:50:09 +0000 (08:50 +0800)]
Merge tag 'clk-fixes-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/clk/linux
Pull clk fixes from Stephen Boyd:
"A handful of clk driver fixes and one core framework fix
- Do a DT/firmware lookup in clk_core_get() even when the DT index is
a nonsensical value
- Fix some clk data typos in the Amlogic DT headers/code
- Avoid returning junk in the TI clk driver when an invalid clk is
looked for
- Fix dividers for the emac clks on Stratix10 SoCs
- Fix default HDA rates on Tegra210 to correct distorted audio"
* tag 'clk-fixes-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/clk/linux:
clk: socfpga: stratix10: fix divider entry for the emac clocks
clk: Do a DT parent lookup even when index < 0
clk: tegra210: Fix default rates for HDA clocks
clk: ti: clkctrl: Fix returning uninitialized data
clk: meson: meson8b: fix a typo in the VPU parent names array variable
clk: meson: fix MPLL 50M binding id typo
Linus Torvalds [Fri, 28 Jun 2019 00:48:21 +0000 (08:48 +0800)]
Merge tag 'for-5.2/dm-fixes-2' of git://git.kernel.org/pub/scm/linux/kernel/git/device-mapper/linux-dm
Pull device mapper fixes from Mike Snitzer:
- Fix incorrect uses of kstrndup and DM logging macros in DM's early
init code.
- Fix DM log-writes target's handling of super block sectors so updates
are made in order through use of completion.
- Fix DM core's argument splitting code to avoid undefined behaviour
reported as a side-effect of UBSAN analysis on ppc64le.
- Fix DM verity target to limit the amount of error messages that can
result from a corrupt block being found.
* tag 'for-5.2/dm-fixes-2' of git://git.kernel.org/pub/scm/linux/kernel/git/device-mapper/linux-dm:
dm verity: use message limit for data block corruption message
dm table: don't copy from a NULL pointer in realloc_argv()
dm log writes: make sure super sector log updates are written in order
dm init: remove trailing newline from calls to DMERR() and DMINFO()
dm init: fix incorrect uses of kstrndup()
Linus Torvalds [Fri, 28 Jun 2019 00:41:18 +0000 (08:41 +0800)]
Merge tag 'for-linus-20190627' of gitolite.kernel.org:pub/scm/linux/kernel/git/brauner/linux
Pull pidfd fixes from Christian Brauner:
"Userspace tools and libraries such as strace or glibc need a cheap and
reliable way to tell whether CLONE_PIDFD is supported. The easiest way
is to pass an invalid fd value in the return argument, perform the
syscall and verify the value in the return argument has been changed
to a valid fd.
However, if CLONE_PIDFD is specified we currently check if pidfd == 0
and return EINVAL if not.
The check for pidfd == 0 was originally added to enable us to abuse
the return argument for passing additional flags along with
CLONE_PIDFD in the future.
However, extending legacy clone this way would be a terrible idea and
with clone3 on the horizon and the ability to reuse CLONE_DETACHED
with CLONE_PIDFD there's no real need for this clutch. So remove the
pidfd == 0 check and help userspace out.
Also, accordig to Al, anon_inode_getfd() should only be used past the
point of no failure and ksys_close() should not be used at all since
it is far too easy to get wrong. Al's motto being "basically, once
it's in descriptor table, it's out of your control". So Al's patch
switches back to what we already had in v1 of the original patchset
and uses a anon_inode_getfile() + put_user() + fd_install() sequence
in the success path and a fput() + put_unused_fd() in the failure
path.
The other two changes should be trivial"
* tag 'for-linus-20190627' of gitolite.kernel.org:pub/scm/linux/kernel/git/brauner/linux:
proc: remove useless d_is_dir() check
copy_process(): don't use ksys_close() on cleanups
samples: make pidfd-metadata fail gracefully on older kernels
fork: don't check parent_tidptr with CLONE_PIDFD
Linus Torvalds [Fri, 28 Jun 2019 00:39:18 +0000 (08:39 +0800)]
Merge branch 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/hid/hid
Pull HID fixes from Jiri Kosina:
- fix for one corner case in HID++ protocol with respect to handling
very long reports, from Hans de Goede
- power management fix in Intel-ISH driver, from Hyungwoo Yang
- use-after-free fix in Intel-ISH driver, from Dan Carpenter
- a couple of new device IDs/quirks from Kai-Heng Feng, Kyle Godbey and
Oleksandr Natalenko
* 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/hid/hid:
HID: intel-ish-hid: fix wrong driver_data usage
HID: multitouch: Add pointstick support for ALPS Touchpad
HID: logitech-dj: Fix forwarding of very long HID++ reports
HID: uclogic: Add support for Huion HS64 tablet
HID: chicony: add another quirk for PixArt mouse
HID: intel-ish-hid: Fix a use after free in load_fw_from_host()
Linus Torvalds [Fri, 28 Jun 2019 00:37:04 +0000 (08:37 +0800)]
Merge tag 'armsoc-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/soc/soc
Pull ARM SoC fixes from Olof Johansson:
"A smaller batch of fixes, nothing that stands out as risky or scary.
Mostly DTS tweaks for a few issues:
- GPU fixlets for Meson
- CPU idle fix for LS1028A
- PWM interrupt fixes for i.MX6UL
Also, enable a driver (FSL_EDMA) on arm64 defconfig, and a warning and
two MAINTAINER tweaks"
* tag 'armsoc-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/soc/soc:
ARM: dts: imx6ul: fix PWM[1-4] interrupts
ARM: omap2: remove incorrect __init annotation
ARM: dts: gemini Fix up DNS-313 compatible string
ARM: dts: Blank D-Link DIR-685 console
arm64: defconfig: Enable FSL_EDMA driver
arm64: dts: ls1028a: Fix CPU idle fail.
MAINTAINERS: BCM53573: Add internal Broadcom mailing list
MAINTAINERS: BCM2835: Add internal Broadcom mailing list
ARM: dts: meson8b: fix the operating voltage of the Mali GPU
ARM: dts: meson8b: drop undocumented property from the Mali GPU node
ARM: dts: meson8: fix GPU interrupts and drop an undocumented property
Linus Torvalds [Fri, 28 Jun 2019 00:34:12 +0000 (08:34 +0800)]
Merge tag 'afs-fixes-20190620' of git://git.kernel.org/pub/scm/linux/kernel/git/dhowells/linux-fs
Pull AFS fixes from David Howells:
"The in-kernel AFS client has been undergoing testing on opendev.org on
one of their mirror machines. They are using AFS to hold data that is
then served via apache, and Ian Wienand had reported seeing oopses,
spontaneous machine reboots and updates to volumes going missing. This
patch series appears to have fixed the problem, very probably due to
patch (2), but it's not 100% certain.
(1) Fix the printing of the "vnode modified" warning to exclude checks
on files for which we don't have a callback promise from the
server (and so don't expect the server to tell us when it
changes).
Without this, for every file or directory for which we still have
an in-core inode that gets changed on the server, we may get a
message logged when we next look at it. This can happen in bulk
if, for instance, someone does "vos release" to update a R/O
volume from a R/W volume and a whole set of files are all changed
together.
We only really want to log a message if the file changed and the
server didn't tell us about it or we failed to track the state
internally.
(2) Fix accidental corruption of either afs_vlserver struct objects or
the the following memory locations (which could hold anything).
The issue is caused by a union that points to two different
structs in struct afs_call (to save space in the struct). The call
cleanup code assumes that it can simply call the cleanup for one
of those structs if not NULL - when it might be actually pointing
to the other struct.
This means that every Volume Location RPC op is going to corrupt
something.
(3) Fix an uninitialised spinlock. This isn't too bad, it just causes
a one-off warning if lockdep is enabled when "vos release" is
called, but the spinlock still behaves correctly.
(4) Fix the setting of i_block in the inode. This causes du, for
example, to produce incorrect results, but otherwise should not be
dangerous to the kernel"
* tag 'afs-fixes-20190620' of git://git.kernel.org/pub/scm/linux/kernel/git/dhowells/linux-fs:
afs: Fix setting of i_blocks
afs: Fix uninitialised spinlock afs_volume::cb_break_lock
afs: Fix vlserver record corruption
afs: Fix over zealous "vnode modified" warnings
1) Fix ppp_mppe crypto soft dependencies, from Takashi Iawi.
2) Fix TX completion to be finite, from Sergej Benilov.
3) Use register_pernet_device to avoid a dst leak in tipc, from Xin
Long.
4) Double free of TX cleanup in Dirk van der Merwe.
5) Memory leak in packet_set_ring(), from Eric Dumazet.
6) Out of bounds read in qmi_wwan, from Bjørn Mork.
7) Fix iif used in mcast/bcast looped back packets, from Stephen
Suryaputra.
8) Fix neighbour resolution on raw ipv6 sockets, from Nicolas Dichtel.
* git://git.kernel.org/pub/scm/linux/kernel/git/davem/net: (25 commits)
af_packet: Block execution of tasks waiting for transmit to complete in AF_PACKET
sctp: change to hold sk after auth shkey is created successfully
ipv6: fix neighbour resolution with raw socket
ipv6: constify rt6_nexthop()
net: dsa: microchip: Use gpiod_set_value_cansleep()
net: aquantia: fix vlans not working over bridged network
ipv4: reset rt_iif for recirculated mcast/bcast out pkts
team: Always enable vlan tx offload
net/smc: Fix error path in smc_init
net/smc: hold conns_lock before calling smc_lgr_register_conn()
bonding: Always enable vlan tx offload
net/ipv6: Fix misuse of proc_dointvec "skip_notify_on_dev_down"
ipv4: Use return value of inet_iif() for __raw_v4_lookup in the while loop
qmi_wwan: Fix out-of-bounds read
tipc: check msg->req data len in tipc_nl_compat_bearer_disable
net: macb: do not copy the mac address if NULL
net/packet: fix memory leak in packet_set_ring()
net/tls: fix page double free on TX cleanup
net/sched: cbs: Fix error path of cbs_module_init
tipc: change to use register_pernet_device
...
Lorenzo Bianconi [Sat, 15 Jun 2019 14:03:32 +0000 (16:03 +0200)]
mt76: usb: fix rx A-MSDU support
Commit f8f527b16db5 ("mt76: usb: use EP max packet aligned buffer sizes
for rx") breaks A-MSDU support. When A-MSDU is enable the device can
receive frames up to q->buf_size but they will be discarded in
mt76u_process_rx_entry since there is no enough room for
skb_shared_info. Fix the issue reallocating the skb and copying in the
linear area the first 128B of the received frames and in the frag_list
the remaining part
Fixes: f8f527b16db5 ("mt76: usb: use EP max packet aligned buffer sizes for rx") Signed-off-by: Lorenzo Bianconi <[email protected]> Signed-off-by: Kalle Valo <[email protected]>
Al Viro [Thu, 27 Jun 2019 02:22:09 +0000 (22:22 -0400)]
copy_process(): don't use ksys_close() on cleanups
anon_inode_getfd() should be used *ONLY* in situations when we are
guaranteed to be past the last failure point (including copying the
descriptor number to userland, at that). And ksys_close() should
not be used for cleanups at all.
anon_inode_getfile() is there for all nontrivial cases like that.
Just use that...
Neil Horman [Tue, 25 Jun 2019 21:57:49 +0000 (17:57 -0400)]
af_packet: Block execution of tasks waiting for transmit to complete in AF_PACKET
When an application is run that:
a) Sets its scheduler to be SCHED_FIFO
and
b) Opens a memory mapped AF_PACKET socket, and sends frames with the
MSG_DONTWAIT flag cleared, its possible for the application to hang
forever in the kernel. This occurs because when waiting, the code in
tpacket_snd calls schedule, which under normal circumstances allows
other tasks to run, including ksoftirqd, which in some cases is
responsible for freeing the transmitted skb (which in AF_PACKET calls a
destructor that flips the status bit of the transmitted frame back to
available, allowing the transmitting task to complete).
However, when the calling application is SCHED_FIFO, its priority is
such that the schedule call immediately places the task back on the cpu,
preventing ksoftirqd from freeing the skb, which in turn prevents the
transmitting task from detecting that the transmission is complete.
We can fix this by converting the schedule call to a completion
mechanism. By using a completion queue, we force the calling task, when
it detects there are no more frames to send, to schedule itself off the
cpu until such time as the last transmitted skb is freed, allowing
forward progress to be made.
Tested by myself and the reporter, with good results
Change Notes:
V1->V2:
Enhance the sleep logic to support being interruptible and
allowing for honoring to SK_SNDTIMEO (Willem de Bruijn)
V2->V3:
Rearrage the point at which we wait for the completion queue, to
avoid needing to check for ph/skb being null at the end of the loop.
Also move the complete call to the skb destructor to avoid needing to
modify __packet_set_status. Also gate calling complete on
packet_read_pending returning zero to avoid multiple calls to complete.
(Willem de Bruijn)
Move timeo computation within loop, to re-fetch the socket
timeout since we also use the timeo variable to record the return code
from the wait_for_complete call (Neil Horman)
V3->V4:
Willem has requested that the control flow be restored to the
previous state. Doing so lets us eliminate the need for the
po->wait_on_complete flag variable, and lets us get rid of the
packet_next_frame function, but introduces another complexity.
Specifically, but using the packet pending count, we can, if an
applications calls sendmsg multiple times with MSG_DONTWAIT set, each
set of transmitted frames, when complete, will cause
tpacket_destruct_skb to issue a complete call, for which there will
never be a wait_on_completion call. This imbalance will lead to any
future call to wait_for_completion here to return early, when the frames
they sent may not have completed. To correct this, we need to re-init
the completion queue on every call to tpacket_snd before we enter the
loop so as to ensure we wait properly for the frames we send in this
iteration.
Change the timeout and interrupted gotos to out_put rather than
out_status so that we don't try to free a non-existant skb
Clean up some extra newlines (Willem de Bruijn)
Xin Long [Mon, 24 Jun 2019 16:21:45 +0000 (00:21 +0800)]
sctp: change to hold sk after auth shkey is created successfully
Now in sctp_endpoint_init(), it holds the sk then creates auth
shkey. But when the creation fails, it doesn't release the sk,
which causes a sk defcnf leak,
Here to fix it by only holding the sk when auth shkey is created
successfully.
Nicolas Dichtel [Mon, 24 Jun 2019 14:01:09 +0000 (16:01 +0200)]
ipv6: fix neighbour resolution with raw socket
The scenario is the following: the user uses a raw socket to send an ipv6
packet, destinated to a not-connected network, and specify a connected nh.
Here is the corresponding python script to reproduce this scenario:
fd00:175::/64 is a connected route and fd00:200::fa is not a connected
host.
With this scenario, the kernel starts by sending a NS to resolve
fd00:175::2. When it receives the NA, it flushes its queue and try to send
the initial packet. But instead of sending it, it sends another NS to
resolve fd00:200::fa, which obvioulsy fails, thus the packet is dropped. If
the user sends again the packet, it now uses the right nh (fd00:175::2).
The problem is that ip6_dst_lookup_neigh() uses the rt6i_gateway, which is
:: because the associated route is a connected route, thus it uses the dst
addr of the packet. Let's use rt6_nexthop() to choose the right nh.
Marek Vasut [Sun, 23 Jun 2019 15:12:57 +0000 (17:12 +0200)]
net: dsa: microchip: Use gpiod_set_value_cansleep()
Replace gpiod_set_value() with gpiod_set_value_cansleep(), as the switch
reset GPIO can be connected to e.g. I2C GPIO expander and it is perfectly
fine for the kernel to sleep for a bit in ksz_switch_register().
Dmitry Bogdanov [Sat, 22 Jun 2019 08:46:37 +0000 (08:46 +0000)]
net: aquantia: fix vlans not working over bridged network
In configuration of vlan over bridge over aquantia device
it was found that vlan tagged traffic is dropped on chip.
The reason is that bridge device enables promisc mode,
but in atlantic chip vlan filters will still apply.
So we have to corellate promisc settings with vlan configuration.
The solution is to track in a separate state variable the
need of vlan forced promisc. And also consider generic
promisc configuration when doing vlan filter config.
Fixes: 7975d2aff5af ("net: aquantia: add support of rx-vlan-filter offload") Signed-off-by: Dmitry Bogdanov <[email protected]> Signed-off-by: Igor Russkikh <[email protected]> Signed-off-by: David S. Miller <[email protected]>
ipv4: reset rt_iif for recirculated mcast/bcast out pkts
Multicast or broadcast egress packets have rt_iif set to the oif. These
packets might be recirculated back as input and lookup to the raw
sockets may fail because they are bound to the incoming interface
(skb_iif). If rt_iif is not zero, during the lookup, inet_iif() function
returns rt_iif instead of skb_iif. Hence, the lookup fails.
v2: Make it non vrf specific (David Ahern). Reword the changelog to
reflect it. Signed-off-by: Stephen Suryaputra <[email protected]> Reviewed-by: David Ahern <[email protected]> Signed-off-by: David S. Miller <[email protected]>
Huaping Zhou [Wed, 26 Jun 2019 15:47:49 +0000 (17:47 +0200)]
net/smc: hold conns_lock before calling smc_lgr_register_conn()
After smc_lgr_create(), the newly created link group is added
to smc_lgr_list, thus is accessible from other context.
Although link group creation is serialized by
smc_create_lgr_pending, the new link group may still be accessed
concurrently. For example, if ib_device is no longer active,
smc_ib_port_event_work() will call smc_port_terminate(), which
in turn will call __smc_lgr_terminate() on every link group of
this device. So conns_lock is required here.
YueHaibing [Wed, 26 Jun 2019 08:08:44 +0000 (16:08 +0800)]
bonding: Always enable vlan tx offload
We build vlan on top of bonding interface, which vlan offload
is off, bond mode is 802.3ad (LACP) and xmit_hash_policy is
BOND_XMIT_POLICY_ENCAP34.
Because vlan tx offload is off, vlan tci is cleared and skb push
the vlan header in validate_xmit_vlan() while sending from vlan
devices. Then in bond_xmit_hash, __skb_flow_dissect() fails to
get information from protocol headers encapsulated within vlan,
because 'nhoff' is points to IP header, so bond hashing is based
on layer 2 info, which fails to distribute packets across slaves.
This patch always enable bonding's vlan tx offload, pass the vlan
packets to the slave devices with vlan tci, let them to handle
vlan implementation.
Kalle Valo [Tue, 25 Jun 2019 14:52:25 +0000 (17:52 +0300)]
ath: fix SPDX tags
Commit ec8f24b7faaf ("treewide: Add SPDX license identifier -
Makefile/Kconfig") marked various Makefiles and Kconfig files within ath
directories as GPL-2.0. But these modules and drivers are actually ISC:
and are causing failure of test_verifier test 'arsh32 on imm 2' on BE
arches.
The code is taking address and interpreting memory directly, so is not
endianness neutral. We should instead perform standard C type casting on
the variable. A u64 to s32 conversion will drop the high 32-bit and reserve
the low 32-bit as signed integer, this is all we want.
projects / linux.git / log