Blamed commit added these helpers for sake of detecting RAW
sockets specific ioctl.
syzbot complained about it [1].
Issue here is that RAW sockets could pretend there was no need
to call ipmr_sk_ioctl()
Regardless of inet_sk(sk)->inet_num, we must be prepared
for ipmr_ioctl() being called later. This must happen
from ipmr_sk_ioctl() context only.
We could add a safety check in ipmr_ioctl() at the risk of breaking
applications.
Instead, remove sk_is_ipmr() and sk_is_icmpv6() because their
name would be misleading, once we change their implementation.
[1]
BUG: KASAN: stack-out-of-bounds in ipmr_ioctl+0xb12/0xbd0 net/ipv4/ipmr.c:1654
Read of size 4 at addr ffffc90003aefae4 by task syz-executor105/5004
CPU: 0 PID: 5004 Comm: syz-executor105 Not tainted 6.4.0-rc6-syzkaller-01304-gc08afcdcf952 #0
Hardware name: Google Google Compute Engine/Google Compute Engine, BIOS Google 05/27/2023
Call Trace:
<TASK>
__dump_stack lib/dump_stack.c:88 [inline]
dump_stack_lvl+0xd9/0x150 lib/dump_stack.c:106
print_address_description.constprop.0+0x2c/0x3c0 mm/kasan/report.c:351
print_report mm/kasan/report.c:462 [inline]
kasan_report+0x11c/0x130 mm/kasan/report.c:572
ipmr_ioctl+0xb12/0xbd0 net/ipv4/ipmr.c:1654
raw_ioctl+0x4e/0x1e0 net/ipv4/raw.c:881
sock_ioctl_out net/core/sock.c:4186 [inline]
sk_ioctl+0x151/0x440 net/core/sock.c:4214
inet_ioctl+0x18c/0x380 net/ipv4/af_inet.c:1001
sock_do_ioctl+0xcc/0x230 net/socket.c:1189
sock_ioctl+0x1f8/0x680 net/socket.c:1306
vfs_ioctl fs/ioctl.c:51 [inline]
__do_sys_ioctl fs/ioctl.c:870 [inline]
__se_sys_ioctl fs/ioctl.c:856 [inline]
__x64_sys_ioctl+0x197/0x210 fs/ioctl.c:856
do_syscall_x64 arch/x86/entry/common.c:50 [inline]
do_syscall_64+0x39/0xb0 arch/x86/entry/common.c:80
entry_SYSCALL_64_after_hwframe+0x63/0xcd
RIP: 0033:0x7f2944bf6ad9
Code: 28 c3 e8 2a 14 00 00 66 2e 0f 1f 84 00 00 00 00 00 48 89 f8 48 89 f7 48 89 d6 48 89 ca 4d 89 c2 4d 89 c8 4c 8b 4c 24 08 0f 05 <48> 3d 01 f0 ff ff 73 01 c3 48 c7 c1 c0 ff ff ff f7 d8 64 89 01 48
RSP: 002b:00007ffd8897a028 EFLAGS: 00000246 ORIG_RAX: 0000000000000010
RAX: ffffffffffffffda RBX: 0000000000000000 RCX: 00007f2944bf6ad9
RDX: 0000000000000000 RSI: 00000000000089e1 RDI: 0000000000000003
RBP: 00007f2944bbac80 R08: 0000000000000000 R09: 0000000000000000
R10: 0000000000000000 R11: 0000000000000246 R12: 00007f2944bbad10
R13: 0000000000000000 R14: 0000000000000000 R15: 0000000000000000
</TASK>
The buggy address belongs to stack of task syz-executor105/5004
and is located at offset 36 in frame:
sk_ioctl+0x0/0x440 net/core/sock.c:4172
This frame has 2 objects:
[32, 36) 'karg'
[48, 88) 'buffer'
Fixes: e1d001fa5b ("net: ioctl: Use kernel memory on protocol ioctl callbacks")
Reported-by: syzbot <syzkaller@googlegroups.com>
Signed-off-by: Eric Dumazet <edumazet@google.com>
Cc: Breno Leitao <leitao@debian.org>
Cc: Kuniyuki Iwashima <kuniyu@amazon.com>
Reviewed-by: Jiri Pirko <jiri@nvidia.com>
Reviewed-by: David Ahern <dsahern@kernel.org>
Reviewed-by: Willem de Bruijn <willemb@google.com>
Link: https://lore.kernel.org/r/20230619124336.651528-1-edumazet@google.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Enables advertisement of the maximum offset supported by the phase control
functionality of PHCs. The callback is used to return an error if an offset
not supported by the PHC is used in ADJ_OFFSET. The ioctls
PTP_CLOCK_GETCAPS and PTP_CLOCK_GETCAPS2 now advertise the maximum offset a
PHC's phase control functionality is capable of supporting. Introduce new
sysfs node, max_phase_adjustment.
Cc: Jakub Kicinski <kuba@kernel.org>
Cc: Shuah Khan <shuah@kernel.org>
Cc: Richard Cochran <richardcochran@gmail.com>
Cc: Maciek Machnikowski <maciek@machnikowski.net>
Signed-off-by: Rahul Rameshbabu <rrameshbabu@nvidia.com>
Acked-by: Richard Cochran <richardcochran@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
.adjphase expects a PHC to use an internal servo algorithm to correct the
provided phase offset target in the callback. Implementation of the
internal servo algorithm are defined by the individual devices.
Cc: Jakub Kicinski <kuba@kernel.org>
Cc: Richard Cochran <richardcochran@gmail.com>
Signed-off-by: Rahul Rameshbabu <rrameshbabu@nvidia.com>
Acked-by: Richard Cochran <richardcochran@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
mlx5-updates-2023-06-16
1) Added a new event handler to firmware sync reset, which is used to
support firmware sync reset flow on smart NIC. Adding this new stage to
the flow enables the firmware to ensure host PFs unload before ECPFs
unload, to avoid race of PFs recovery.
2) Debugfs for mlx5 eswitch bridge offloads
3) Added two new counters for vport stats
4) Minor Fixups and cleanups for net-next branch
Signed-off-by: David S. Miller <davem@davemloft.net>
Per-VMA locking allows us to lock a struct vm_area_struct without
taking the process-wide mmap lock in read mode.
Consider a process workload where the mmap lock is taken constantly in
write mode. In this scenario, all zerocopy receives are periodically
blocked during that period of time - though in principle, the memory
ranges being used by TCP are not touched by the operations that need
the mmap write lock. This results in performance degradation.
Now consider another workload where the mmap lock is never taken in
write mode, but there are many TCP connections using receive zerocopy
that are concurrently receiving. These connections all take the mmap
lock in read mode, but this does induce a lot of contention and atomic
ops for this process-wide lock. This results in additional CPU
overhead caused by contending on the cache line for this lock.
However, with per-vma locking, both of these problems can be avoided.
As a test, I ran an RPC-style request/response workload with 4KB
payloads and receive zerocopy enabled, with 100 simultaneous TCP
connections. I measured perf cycles within the
find_tcp_vma/mmap_read_lock/mmap_read_unlock codepath, with and
without per-vma locking enabled.
When using process-wide mmap semaphore read locking, about 1% of
measured perf cycles were within this path. With per-VMA locking, this
value dropped to about 0.45%.
Signed-off-by: Arjun Roy <arjunroy@google.com>
Reviewed-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Under certain circumstances, the tcp receive buffer memory limit
set by autotuning (sk_rcvbuf) is increased due to incoming data
packets as a result of the window not closing when it should be.
This can result in the receive buffer growing all the way up to
tcp_rmem[2], even for tcp sessions with a low BDP.
To reproduce: Connect a TCP session with the receiver doing
nothing and the sender sending small packets (an infinite loop
of socket send() with 4 bytes of payload with a sleep of 1 ms
in between each send()). This will cause the tcp receive buffer
to grow all the way up to tcp_rmem[2].
As a result, a host can have individual tcp sessions with receive
buffers of size tcp_rmem[2], and the host itself can reach tcp_mem
limits, causing the host to go into tcp memory pressure mode.
The fundamental issue is the relationship between the granularity
of the window scaling factor and the number of byte ACKed back
to the sender. This problem has previously been identified in
RFC 7323, appendix F [1].
The Linux kernel currently adheres to never shrinking the window.
In addition to the overallocation of memory mentioned above, the
current behavior is functionally incorrect, because once tcp_rmem[2]
is reached when no remediations remain (i.e. tcp collapse fails to
free up any more memory and there are no packets to prune from the
out-of-order queue), the receiver will drop in-window packets
resulting in retransmissions and an eventual timeout of the tcp
session. A receive buffer full condition should instead result
in a zero window and an indefinite wait.
In practice, this problem is largely hidden for most flows. It
is not applicable to mice flows. Elephant flows can send data
fast enough to "overrun" the sk_rcvbuf limit (in a single ACK),
triggering a zero window.
But this problem does show up for other types of flows. Examples
are websockets and other type of flows that send small amounts of
data spaced apart slightly in time. In these cases, we directly
encounter the problem described in [1].
RFC 7323, section 2.4 [2], says there are instances when a retracted
window can be offered, and that TCP implementations MUST ensure
that they handle a shrinking window, as specified in RFC 1122,
section 4.2.2.16 [3]. All prior RFCs on the topic of tcp window
management have made clear that sender must accept a shrunk window
from the receiver, including RFC 793 [4] and RFC 1323 [5].
This patch implements the functionality to shrink the tcp window
when necessary to keep the right edge within the memory limit by
autotuning (sk_rcvbuf). This new functionality is enabled with
the new sysctl: net.ipv4.tcp_shrink_window
Additional information can be found at:
https://blog.cloudflare.com/unbounded-memory-usage-by-tcp-for-receive-buffers-and-how-we-fixed-it/
[1] https://www.rfc-editor.org/rfc/rfc7323#appendix-F
[2] https://www.rfc-editor.org/rfc/rfc7323#section-2.4
[3] https://www.rfc-editor.org/rfc/rfc1122#page-91
[4] https://www.rfc-editor.org/rfc/rfc793
[5] https://www.rfc-editor.org/rfc/rfc1323
Signed-off-by: Mike Freemon <mfreemon@cloudflare.com>
Reviewed-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Added a new event handler to firmware sync reset, which is used to
support firmware sync reset flow on smart NIC. Adding this new stage to
the flow enables the firmware to ensure host PFs unload before ECPFs
unload, to avoid race of PFs recovery.
If firmware sends sync_reset_unload event to driver the driver should
unload and close all HW resources of the function. Once the driver
finishes unloading part, it can't get any more events from firmware as
event queues are closed, so it polls the reset state field to know when
to continue to next stage of the sync reset flow.
Added capability bit for supporting sync_reset_unload event.
Signed-off-by: Moshe Shemesh <moshe@nvidia.com>
Reviewed-by: Shay Drory <shayd@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
Expose new timoueout in Default Timeouts Register to be used on sync
reset flow running on smart NIC. In this flow the driver should know how
much time to wait from getting unload request till firmware will ask the
PF to continue to next stage of the flow.
Signed-off-by: Moshe Shemesh <moshe@nvidia.com>
Reviewed-by: Shay Drory <shayd@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
Most of the ioctls to net protocols operates directly on userspace
argument (arg). Usually doing get_user()/put_user() directly in the
ioctl callback. This is not flexible, because it is hard to reuse these
functions without passing userspace buffers.
Change the "struct proto" ioctls to avoid touching userspace memory and
operate on kernel buffers, i.e., all protocol's ioctl callbacks is
adapted to operate on a kernel memory other than on userspace (so, no
more {put,get}_user() and friends being called in the ioctl callback).
This changes the "struct proto" ioctl format in the following way:
int (*ioctl)(struct sock *sk, int cmd,
- unsigned long arg);
+ int *karg);
(Important to say that this patch does not touch the "struct proto_ops"
protocols)
So, the "karg" argument, which is passed to the ioctl callback, is a
pointer allocated to kernel space memory (inside a function wrapper).
This buffer (karg) may contain input argument (copied from userspace in
a prep function) and it might return a value/buffer, which is copied
back to userspace if necessary. There is not one-size-fits-all format
(that is I am using 'may' above), but basically, there are three type of
ioctls:
1) Do not read from userspace, returns a result to userspace
2) Read an input parameter from userspace, and does not return anything
to userspace
3) Read an input from userspace, and return a buffer to userspace.
The default case (1) (where no input parameter is given, and an "int" is
returned to userspace) encompasses more than 90% of the cases, but there
are two other exceptions. Here is a list of exceptions:
* Protocol RAW:
* cmd = SIOCGETVIFCNT:
* input and output = struct sioc_vif_req
* cmd = SIOCGETSGCNT
* input and output = struct sioc_sg_req
* Explanation: for the SIOCGETVIFCNT case, userspace passes the input
argument, which is struct sioc_vif_req. Then the callback populates
the struct, which is copied back to userspace.
* Protocol RAW6:
* cmd = SIOCGETMIFCNT_IN6
* input and output = struct sioc_mif_req6
* cmd = SIOCGETSGCNT_IN6
* input and output = struct sioc_sg_req6
* Protocol PHONET:
* cmd == SIOCPNADDRESOURCE | SIOCPNDELRESOURCE
* input int (4 bytes)
* Nothing is copied back to userspace.
For the exception cases, functions sock_sk_ioctl_inout() will
copy the userspace input, and copy it back to kernel space.
The wrapper that prepare the buffer and put the buffer back to user is
sk_ioctl(), so, instead of calling sk->sk_prot->ioctl(), the callee now
calls sk_ioctl(), which will handle all cases.
Signed-off-by: Breno Leitao <leitao@debian.org>
Reviewed-by: Willem de Bruijn <willemb@google.com>
Reviewed-by: David Ahern <dsahern@kernel.org>
Reviewed-by: Kuniyuki Iwashima <kuniyu@amazon.com>
Link: https://lore.kernel.org/r/20230609152800.830401-1-leitao@debian.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Pull networking fixes from Jakub Kicinski:
"Including fixes from wireless, and netfilter.
Selftests excluded - we have 58 patches and diff of +442/-199, which
isn't really small but perhaps with the exception of the WiFi locking
change it's old(ish) bugs.
We have no known problems with v6.4.
The selftest changes are rather large as MPTCP folks try to apply
Greg's guidance that selftest from torvalds/linux should be able to
run against stable kernels.
Last thing I should call out is the DCCP/UDP-lite deprecation notices.
We are fairly sure those are dead, but if we're wrong reverting them
back in won't be fun.
Current release - regressions:
- wifi:
- cfg80211: fix double lock bug in reg_wdev_chan_valid()
- iwlwifi: mvm: spin_lock_bh() to fix lockdep regression
Current release - new code bugs:
- handshake: remove fput() that causes use-after-free
Previous releases - regressions:
- sched: cls_u32: fix reference counter leak leading to overflow
- sched: cls_api: fix lockup on flushing explicitly created chain
Previous releases - always broken:
- nf_tables: integrate pipapo into commit protocol
- nf_tables: incorrect error path handling with NFT_MSG_NEWRULE, fix
dangling pointer on failure
- ping6: fix send to link-local addresses with VRF
- sched: act_pedit: parse L3 header for L4 offset, the skb may not
have the offset saved
- sched: act_ct: fix promotion of offloaded unreplied tuple
- sched: refuse to destroy an ingress and clsact Qdiscs if there are
lockless change operations in flight
- wifi: mac80211: fix handful of bugs in multi-link operation
- ipvlan: fix bound dev checking for IPv6 l3s mode
- eth: enetc: correct the indexes of highest and 2nd highest TCs
- eth: ice: fix XDP memory leak when NIC is brought up and down
Misc:
- add deprecation notices for UDP-lite and DCCP
- selftests: mptcp: skip tests not supported by old kernels
- sctp: handle invalid error codes without calling BUG()"
* tag 'net-6.4-rc7' of git://git.kernel.org/pub/scm/linux/kernel/git/netdev/net: (91 commits)
dccp: Print deprecation notice.
udplite: Print deprecation notice.
octeon_ep: Add missing check for ioremap
selftests/ptp: Fix timestamp printf format for PTP_SYS_OFFSET
net: ethernet: stmicro: stmmac: fix possible memory leak in __stmmac_open
net: tipc: resize nlattr array to correct size
sfc: fix XDP queues mode with legacy IRQ
net: macsec: fix double free of percpu stats
net: lapbether: only support ethernet devices
MAINTAINERS: add reviewers for SMC Sockets
s390/ism: Fix trying to free already-freed IRQ by repeated ism_dev_exit()
net: dsa: felix: fix taprio guard band overflow at 10Mbps with jumbo frames
net/sched: cls_api: Fix lockup on flushing explicitly created chain
ice: Fix ice module unload
net/handshake: remove fput() that causes use-after-free
selftests: forwarding: hw_stats_l3: Set addrgenmode in a separate step
net/sched: qdisc_destroy() old ingress and clsact Qdiscs before grafting
net/sched: Refactor qdisc_graft() for ingress and clsact Qdiscs
net/sched: act_ct: Fix promotion of offloaded unreplied tuple
wifi: iwlwifi: mvm: spin_lock_bh() to fix lockdep regression
...
Pull rdma fixes from Jason Gunthorpe:
"This is an unusually large bunch of bug fixes for the later rc cycle,
rxe and mlx5 both dumped a lot of things at once. rxe continues to fix
itself, and mlx5 is fixing a bunch of "queue counters" related bugs.
There is one highly notable bug fix regarding the qkey. This small
security check was missed in the original 2005 implementation and it
allows some significant issues.
Summary:
- Two rtrs bug fixes for error unwind bugs
- Several rxe bug fixes:
* Incorrect Rx packet validation
* Using memory without a refcount
* Syzkaller found use before initialization
* Regression fix for missing locking with the tasklet conversion
from this merge window
- Have bnxt report the correct link properties to userspace, this was
a regression in v6.3
- Several mlx5 bug fixes:
* Kernel crash triggerable by userspace for the RAW ethernet
profile
* Defend against steering refcounting issues created by userspace
* Incorrect change of QP port affinity parameters in some LAG
configurations
- Fix mlx5 Q counters:
* Do not over allocate Q counters to allow userspace to use the
full port capacity
* Kernel crash triggered by eswitch due to mis-use of Q counters
* Incorrect mlx5_device for Q counters in some LAG configurations
- Properly implement the IBA spec restricting privileged qkeys to
root
- Always an error when reading from a disassociated device's event
queue
- isert bug fixes:
* Avoid a deadlock with the CM handler and CM ID destruction
* Correct list corruption due to incorrect locking
* Fix a use after free around connection tear down"
* tag 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/rdma/rdma:
RDMA/rxe: Fix rxe_cq_post
IB/isert: Fix incorrect release of isert connection
IB/isert: Fix possible list corruption in CMA handler
IB/isert: Fix dead lock in ib_isert
RDMA/mlx5: Fix affinity assignment
IB/uverbs: Fix to consider event queue closing also upon non-blocking mode
RDMA/uverbs: Restrict usage of privileged QKEYs
RDMA/cma: Always set static rate to 0 for RoCE
RDMA/mlx5: Fix Q-counters query in LAG mode
RDMA/mlx5: Remove vport Q-counters dependency on normal Q-counters
RDMA/mlx5: Fix Q-counters per vport allocation
RDMA/mlx5: Create an indirect flow table for steering anchor
RDMA/mlx5: Initiate dropless RQ for RAW Ethernet functions
RDMA/rxe: Fix the use-before-initialization error of resp_pkts
RDMA/bnxt_re: Fix reporting active_{speed,width} attributes
RDMA/rxe: Fix ref count error in check_rkey()
RDMA/rxe: Fix packet length checks
RDMA/rtrs: Fix rxe_dealloc_pd warning
RDMA/rtrs: Fix the last iu->buf leak in err path
Pull media fixes from Mauro Carvalho Chehab:
"A fix for dvb-core to avoid a race condition during DVB board
registration"
* tag 'media/v6.4-6' of git://git.kernel.org/pub/scm/linux/kernel/git/mchehab/linux-media:
Revert "media: dvb-core: Fix use-after-free on race condition at dvb_frontend"
All callers of tls_is_sk_tx_device_offloaded() currently do
an equivalent of:
if (skb->sk && tls_is_skb_tx_device_offloaded(skb->sk))
Have the helper accept skb and do the skb->sk check locally.
Two drivers have local static inlines with similar wrappers
already.
While at it change the ifdef condition to TLS_DEVICE.
Only TLS_DEVICE selects SOCK_VALIDATE_XMIT, so the two are
equivalent. This makes removing the duplicated IS_ENABLED()
check in funeth more obviously correct.
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Acked-by: Maxim Mikityanskiy <maxtram95@gmail.com>
Reviewed-by: Simon Horman <simon.horman@corigine.com>
Acked-by: Tariq Toukan <tariqt@nvidia.com>
Acked-by: Dimitris Michailidis <dmichail@fungible.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
New users of dev_get_by_index() and dev_get_by_name() keep
getting added and it would be nice to steer them towards
the APIs with reference tracking.
Add variants of those calls which allocate the reference
tracker and use them in a couple of places.
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Reviewed-by: Eric Dumazet <edumazet@google.com>
Reviewed-by: David Ahern <dsahern@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
mini_Qdisc_pair::p_miniq is a double pointer to mini_Qdisc, initialized
in ingress_init() to point to net_device::miniq_ingress. ingress Qdiscs
access this per-net_device pointer in mini_qdisc_pair_swap(). Similar
for clsact Qdiscs and miniq_egress.
Unfortunately, after introducing RTNL-unlocked RTM_{NEW,DEL,GET}TFILTER
requests (thanks Hillf Danton for the hint), when replacing ingress or
clsact Qdiscs, for example, the old Qdisc ("@old") could access the same
miniq_{in,e}gress pointer(s) concurrently with the new Qdisc ("@new"),
causing race conditions [1] including a use-after-free bug in
mini_qdisc_pair_swap() reported by syzbot:
BUG: KASAN: slab-use-after-free in mini_qdisc_pair_swap+0x1c2/0x1f0 net/sched/sch_generic.c:1573
Write of size 8 at addr ffff888045b31308 by task syz-executor690/14901
...
Call Trace:
<TASK>
__dump_stack lib/dump_stack.c:88 [inline]
dump_stack_lvl+0xd9/0x150 lib/dump_stack.c:106
print_address_description.constprop.0+0x2c/0x3c0 mm/kasan/report.c:319
print_report mm/kasan/report.c:430 [inline]
kasan_report+0x11c/0x130 mm/kasan/report.c:536
mini_qdisc_pair_swap+0x1c2/0x1f0 net/sched/sch_generic.c:1573
tcf_chain_head_change_item net/sched/cls_api.c:495 [inline]
tcf_chain0_head_change.isra.0+0xb9/0x120 net/sched/cls_api.c:509
tcf_chain_tp_insert net/sched/cls_api.c:1826 [inline]
tcf_chain_tp_insert_unique net/sched/cls_api.c:1875 [inline]
tc_new_tfilter+0x1de6/0x2290 net/sched/cls_api.c:2266
...
@old and @new should not affect each other. In other words, @old should
never modify miniq_{in,e}gress after @new, and @new should not update
@old's RCU state.
Fixing without changing sch_api.c turned out to be difficult (please
refer to Closes: for discussions). Instead, make sure @new's first call
always happen after @old's last call (in {ingress,clsact}_destroy()) has
finished:
In qdisc_graft(), return -EBUSY if @old has any ongoing filter requests,
and call qdisc_destroy() for @old before grafting @new.
Introduce qdisc_refcount_dec_if_one() as the counterpart of
qdisc_refcount_inc_nz() used for filter requests. Introduce a
non-static version of qdisc_destroy() that does a TCQ_F_BUILTIN check,
just like qdisc_put() etc.
Depends on patch "net/sched: Refactor qdisc_graft() for ingress and
clsact Qdiscs".
[1] To illustrate, the syzkaller reproducer adds ingress Qdiscs under
TC_H_ROOT (no longer possible after commit c7cfbd1150 ("net/sched:
sch_ingress: Only create under TC_H_INGRESS")) on eth0 that has 8
transmission queues:
Thread 1 creates ingress Qdisc A (containing mini Qdisc a1 and a2),
then adds a flower filter X to A.
Thread 2 creates another ingress Qdisc B (containing mini Qdisc b1 and
b2) to replace A, then adds a flower filter Y to B.
Thread 1 A's refcnt Thread 2
RTM_NEWQDISC (A, RTNL-locked)
qdisc_create(A) 1
qdisc_graft(A) 9
RTM_NEWTFILTER (X, RTNL-unlocked)
__tcf_qdisc_find(A) 10
tcf_chain0_head_change(A)
mini_qdisc_pair_swap(A) (1st)
|
| RTM_NEWQDISC (B, RTNL-locked)
RCU sync 2 qdisc_graft(B)
| 1 notify_and_destroy(A)
|
tcf_block_release(A) 0 RTM_NEWTFILTER (Y, RTNL-unlocked)
qdisc_destroy(A) tcf_chain0_head_change(B)
tcf_chain0_head_change_cb_del(A) mini_qdisc_pair_swap(B) (2nd)
mini_qdisc_pair_swap(A) (3rd) |
... ...
Here, B calls mini_qdisc_pair_swap(), pointing eth0->miniq_ingress to
its mini Qdisc, b1. Then, A calls mini_qdisc_pair_swap() again during
ingress_destroy(), setting eth0->miniq_ingress to NULL, so ingress
packets on eth0 will not find filter Y in sch_handle_ingress().
This is just one of the possible consequences of concurrently accessing
miniq_{in,e}gress pointers.
Fixes: 7a096d579e ("net: sched: ingress: set 'unlocked' flag for Qdisc ops")
Fixes: 87f373921c ("net: sched: ingress: set 'unlocked' flag for clsact Qdisc ops")
Reported-by: syzbot+b53a9c0d1ea4ad62da8b@syzkaller.appspotmail.com
Closes: https://lore.kernel.org/r/0000000000006cf87705f79acf1a@google.com/
Cc: Hillf Danton <hdanton@sina.com>
Cc: Vlad Buslov <vladbu@mellanox.com>
Signed-off-by: Peilin Ye <peilin.ye@bytedance.com>
Acked-by: Jamal Hadi Salim <jhs@mojatatu.com>
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Currently UNREPLIED and UNASSURED connections are added to the nf flow
table. This causes the following connection packets to be processed
by the flow table which then skips conntrack_in(), and thus such the
connections will remain UNREPLIED and UNASSURED even if reply traffic
is then seen. Even still, the unoffloaded reply packets are the ones
triggering hardware update from new to established state, and if
there aren't any to triger an update and/or previous update was
missed, hardware can get out of sync with sw and still mark
packets as new.
Fix the above by:
1) Not skipping conntrack_in() for UNASSURED packets, but still
refresh for hardware, as before the cited patch.
2) Try and force a refresh by reply-direction packets that update
the hardware rules from new to established state.
3) Remove any bidirectional flows that didn't failed to update in
hardware for re-insertion as bidrectional once any new packet
arrives.
Fixes: 6a9bad0069 ("net/sched: act_ct: offload UDP NEW connections")
Co-developed-by: Vlad Buslov <vladbu@nvidia.com>
Signed-off-by: Vlad Buslov <vladbu@nvidia.com>
Signed-off-by: Paul Blakey <paulb@nvidia.com>
Reviewed-by: Florian Westphal <fw@strlen.de>
Link: https://lore.kernel.org/r/1686313379-117663-1-git-send-email-paulb@nvidia.com
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Rewrite the AF_KCM transmission loop to send all the fragments in a single
skb or frag_list-skb in one sendmsg() with MSG_SPLICE_PAGES set. The list
of fragments in each skb is conveniently a bio_vec[] that can just be
attached to a BVEC iter.
Note: I'm working out the size of each fragment-skb by adding up bv_len for
all the bio_vecs in skb->frags[] - but surely this information is recorded
somewhere? For the skbs in head->frag_list, this is equal to
skb->data_len, but not for the head. head->data_len includes all the tail
frags too.
Signed-off-by: David Howells <dhowells@redhat.com>
cc: Tom Herbert <tom@herbertland.com>
cc: Tom Herbert <tom@quantonium.net>
cc: Jens Axboe <axboe@kernel.dk>
cc: Matthew Wilcox <willy@infradead.org>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Add support to the tc flower classifier to match based on fields in CFM
information elements like level and opcode.
tc filter add dev ens6 ingress protocol 802.1q \
flower vlan_id 698 vlan_ethtype 0x8902 cfm mdl 5 op 46 \
action drop
Signed-off-by: Zahari Doychev <zdoychev@maxlinear.com>
Reviewed-by: Simon Horman <simon.horman@corigine.com>
Reviewed-by: Ido Schimmel <idosch@nvidia.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Add support for dissecting cfm packets. The cfm packet header
fields maintenance domain level and opcode can be dissected.
Signed-off-by: Zahari Doychev <zdoychev@maxlinear.com>
Reviewed-by: Simon Horman <simon.horman@corigine.com>
Reviewed-by: Ido Schimmel <idosch@nvidia.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
mlx5-updates-2023-06-09
1) Embedded CPU Virtual Functions
2) Lightweight local SFs
Daniel Jurgens says:
====================
Embedded CPU Virtual Functions
This series enables the creation of virtual functions on Bluefield (the
embedded CPU platform). Embedded CPU virtual functions (EC VFs). EC VF
creation, deletion and management interfaces are the same as those for
virtual functions in a server with a Connect-X NIC.
When using EC VFs on the ARM the creation of virtual functions on the
host system is still supported. Host VFs eswitch vports occupy a range
of 1..max_vfs, the EC VF vport range is max_vfs+1..max_ec_vfs.
Every function (PF, ECPF, VF, EC VF, and subfunction) has a function ID
associated with it. Prior to this series the function ID and the eswitch
vport were the same. That is no longer the case, the EC VF function ID
range is 1..max_ec_vfs. When querying or setting the capabilities of an
EC VF function an new bit must be set in the query/set HCA cap
structure.
This is a high level overview of the changes made:
- Allocate vports for EC VFs if they are enabled.
- Create representors and devlink ports for the EC VF vports.
- When querying/setting HCA caps by vport break the assumption
that function ID is the same a vport number and adjust
accordingly.
- Create a new type of page, so that when SRIOV on the ARM is
disabled, but remains enabled on the host, the driver can
wait for the correct pages.
- Update SRIOV code to support EC VF creation/deletion.
===================
Lightweight local SFs:
Last 3 patches form Shay Drory:
SFs are heavy weight and by default they come with the full package of
ConnectX features. Usually users want specialized SFs for one specific
purpose and using devlink users will almost always override the set of
advertises features of an SF and reload it.
Shay Drory says:
================
In order to avoid the wasted time and resources on the reload, local SFs
will probe without any auxiliary sub-device, so that the SFs can be
configured prior to its full probe.
The defaults of the enable_* devlink params of these SFs are set to
false.
Usage example:
Create SF:
$ devlink port add pci/0000:08:00.0 flavour pcisf pfnum 0 sfnum 11
$ devlink port function set pci/0000:08:00.0/32768 \
hw_addr 00:00:00:00:00:11 state active
Enable ETH auxiliary device:
$ devlink dev param set auxiliary/mlx5_core.sf.1 \
name enable_eth value true cmode driverinit
Now, in order to fully probe the SF, use devlink reload:
$ devlink dev reload auxiliary/mlx5_core.sf.1
At this point the user have SF devlink instance with auxiliary device
for the Ethernet functionality only.
================
Now all tcp_stream_alloc_skb() callers pass @size == 0, we can
remove this parameter.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
tcp_send_syn_data() is the last component in TCP transmit
path to put payload in skb->head.
Switch it to use page frags, so that we can remove dead
code later.
This allows to put more payload than previous implementation.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Commit 4a19edb60d ("netlink: Pass extack to dump handlers")
added extack support to netlink dumps. It was focused on rtnl
and since rtnl does not use ->start(), ->done() callbacks
it ignored those. Genetlink on the other hand uses ->start()
extensively, for parsing and input validation.
Pass the extact in via struct netlink_dump_control and link
it to cb for the time of ->start(). Both struct netlink_dump_control
and extack itself live on the stack so we can't keep the same
extack for the duration of the dump. This means that the extack
visible in ->start() and each ->dump() callbacks will be different.
Corner cases like reporting a warning message in DONE across dump
calls are still not supported.
We could put the extack (for dumps) in the socket struct,
but layering makes it slightly awkward (extack pointer is decided
before the DO / DUMP split).
The genetlink dump error extacks are now surfaced:
$ cli.py --spec netlink/specs/ethtool.yaml --dump channels-get
lib.ynl.NlError: Netlink error: Invalid argument
nl_len = 64 (48) nl_flags = 0x300 nl_type = 2
error: -22 extack: {'msg': 'request header missing'}
Previously extack was missing:
$ cli.py --spec netlink/specs/ethtool.yaml --dump channels-get
lib.ynl.NlError: Netlink error: Invalid argument
nl_len = 36 (20) nl_flags = 0x100 nl_type = 2
error: -22
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
Since its introduction, the ovs module execute_hash action allowed
hash algorithms other than the skb->l4_hash to be used. However,
additional hash algorithms were not implemented. This means flows
requiring different hash distributions weren't able to use the
kernel datapath.
Now, introduce support for symmetric hashing algorithm as an
alternative hash supported by the ovs module using the flow
dissector.
Output of flow using l4_sym hash:
recirc_id(0),in_port(3),eth(),eth_type(0x0800),
ipv4(dst=64.0.0.0/192.0.0.0,proto=6,frag=no), packets:30473425,
bytes:45902883702, used:0.000s, flags:SP.,
actions:hash(sym_l4(0)),recirc(0xd)
Some performance testing with no GRO/GSO, two veths, single flow:
hash(l4(0)): 4.35 GBits/s
hash(l4_sym(0)): 4.24 GBits/s
Signed-off-by: Aaron Conole <aconole@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
The taprio Qdisc creates child classes per netdev TX queue, but
taprio_dump_class_stats() currently reports offload statistics per
traffic class. Traffic classes are groups of TXQs sharing the same
dequeue priority, so this is incorrect and we shouldn't be bundling up
the TXQ stats when reporting them, as we currently do in enetc.
Modify the API from taprio to drivers such that they report TXQ offload
stats and not TC offload stats.
There is no change in the UAPI or in the global Qdisc stats.
Fixes: 6c1adb650c ("net/sched: taprio: add netlink reporting for offload statistics counters")
Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
When compiling YNL generated code compiler complains about
array-initializer-out-of-bounds. Turns out the MAX value
for STATS_GRP uses the value for STATS.
This may lead to random corruptions in user space (kernel
itself doesn't use this value as it never parses stats).
Fixes: f09ea6fb12 ("ethtool: add a new command for reading standard stats")
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Reviewed-by: David Ahern <dsahern@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
The cited commit aimed to ensure that Virtual Functions (VFs) assign a
queue affinity to a Queue Pair (QP) to distribute traffic when
the LAG master creates a hardware LAG. If the affinity was set while
the hardware was not in LAG, the firmware would ignore the affinity value.
However, this commit unintentionally assigned an affinity to QPs on the LAG
master's VPORT even if the RDMA device was not marked as LAG-enabled.
In most cases, this was not an issue because when the hardware entered
hardware LAG configuration, the RDMA device of the LAG master would be
destroyed and a new one would be created, marked as LAG-enabled.
The problem arises when a user configures Equal-Cost Multipath (ECMP).
In ECMP mode, traffic can be directed to different physical ports based on
the queue affinity, which is intended for use by VPORTS other than the
E-Switch manager. ECMP mode is supported only if both E-Switch managers are
in switchdev mode and the appropriate route is configured via IP. In this
configuration, the RDMA device is not destroyed, and we retain the RDMA
device that is not marked as LAG-enabled.
To ensure correct behavior, Send Queues (SQs) opened by the E-Switch
manager through verbs should be assigned strict affinity. This means they
will only be able to communicate through the native physical port
associated with the E-Switch manager. This will prevent the firmware from
assigning affinity and will not allow the SQs to be remapped in case of
failover.
Fixes: 802dcc7fc5 ("RDMA/mlx5: Support TX port affinity for VF drivers in LAG mode")
Reviewed-by: Maor Gottlieb <maorg@nvidia.com>
Signed-off-by: Mark Bloch <mbloch@nvidia.com>
Link: https://lore.kernel.org/r/425b05f4da840bc684b0f7e8ebf61aeb5cef09b0.1685960567.git.leon@kernel.org
Signed-off-by: Leon Romanovsky <leon@kernel.org>
Pull ARM SoC fixes from Arnd Bergmann:
"Most of the changes this time are for the Qualcomm Snapdragon
platforms.
There are bug fixes for error handling in Qualcomm icc-bwmon,
rpmh-rsc, ramp_controller and rmtfs driver as well as the AMD tee
firmware driver and a missing initialization in the Arm ff-a firmware
driver. The Qualcomm RPMh and EDAC drivers need some rework to work
correctly on all supported chips.
The DT fixes include:
- i.MX8 fixes for gpio, pinmux and clock settings
- ADS touchscreen gpio polarity settings in several machines
- Address dtb warnings for caches, panel and input-enable properties
on Qualcomm platforms
- Incorrect data on qualcomm platforms fir SA8155P power domains,
SM8550 LLCC, SC7180-lite SDRAM frequencies and SM8550 soundwire
- Remoteproc firmware paths are corrected for Sony Xperia 10 IV"
* tag 'arm-fixes-6.4-2' of git://git.kernel.org/pub/scm/linux/kernel/git/soc/soc: (36 commits)
firmware: arm_ffa: Set handle field to zero in memory descriptor
ARM: dts: Fix erroneous ADS touchscreen polarities
arm64: dts: imx8mn-beacon: Fix SPI CS pinmux
arm64: dts: imx8-ss-dma: assign default clock rate for lpuarts
arm64: dts: imx8qm-mek: correct GPIOs for USDHC2 CD and WP signals
EDAC/qcom: Get rid of hardcoded register offsets
EDAC/qcom: Remove superfluous return variable assignment in qcom_llcc_core_setup()
arm64: dts: qcom: sm8550: Use the correct LLCC register scheme
dt-bindings: cache: qcom,llcc: Fix SM8550 description
arm64: dts: qcom: sc7180-lite: Fix SDRAM freq for misidentified sc7180-lite boards
arm64: dts: qcom: sm8550: use uint16 for Soundwire interval
soc: qcom: rpmhpd: Add SA8155P power domains
arm64: dts: qcom: Split out SA8155P and use correct RPMh power domains
dt-bindings: power: qcom,rpmpd: Add SA8155P
soc: qcom: Rename ice to qcom_ice to avoid module name conflict
soc: qcom: rmtfs: Fix error code in probe()
soc: qcom: ramp_controller: Fix an error handling path in qcom_ramp_controller_probe()
ARM: dts: at91: sama7g5ek: fix debounce delay property for shdwc
ARM: at91: pm: fix imbalanced reference counter for ethernet devices
arm64: dts: qcom: sm6375-pdx225: Fix remoteproc firmware paths
...
netfilter pull request 23-06-08
Pablo Neira Ayuso says:
====================
The following patchset contains Netfilter fixes for net:
1) Add commit and abort set operation to pipapo set abort path.
2) Bail out immediately in case of ENOMEM in nfnetlink batch.
3) Incorrect error path handling when creating a new rule leads to
dangling pointer in set transaction list.
====================
Signed-off-by: David S. Miller <davem@davemloft.net>
Kalle Valo says:
====================
wireless-next patches for v6.5
The second pull request for v6.5. We have support for three new
Realtek chipsets, all from different generations. Shows how active
Realtek development is right now, even older generations are being
worked on.
Note: We merged wireless into wireless-next to avoid complex conflicts
between the trees.
Major changes:
rtl8xxxu
- RTL8192FU support
rtw89
- RTL8851BE support
rtw88
- RTL8723DS support
ath11k
- Multiple Basic Service Set Identifier (MBSSID) and Enhanced MBSSID
Advertisement (EMA) support in AP mode
iwlwifi
- support for segmented PNVM images and power tables
- new vendor entries for PPAG (platform antenna gain) feature
cfg80211/mac80211
- more Multi-Link Operation (MLO) support such as hardware restart
- fixes for a potential work/mutex deadlock and with it beginnings of
the previously discussed locking simplifications
* tag 'wireless-next-2023-06-09' of git://git.kernel.org/pub/scm/linux/kernel/git/wireless/wireless-next: (162 commits)
wifi: rtlwifi: remove misused flag from HAL data
wifi: rtlwifi: remove unused dualmac control leftovers
wifi: rtlwifi: remove unused timer and related code
wifi: rsi: Do not set MMC_PM_KEEP_POWER in shutdown
wifi: rsi: Do not configure WoWlan in shutdown hook if not enabled
wifi: brcmfmac: Detect corner error case earlier with log
wifi: rtw89: 8852c: update RF radio A/B parameters to R63
wifi: rtw89: 8852c: update TX power tables to R63 with 6 GHz power type (3 of 3)
wifi: rtw89: 8852c: update TX power tables to R63 with 6 GHz power type (2 of 3)
wifi: rtw89: 8852c: update TX power tables to R63 with 6 GHz power type (1 of 3)
wifi: rtw89: process regulatory for 6 GHz power type
wifi: rtw89: regd: update regulatory map to R64-R40
wifi: rtw89: regd: judge 6 GHz according to chip and BIOS
wifi: rtw89: refine clearing supported bands to check 2/5 GHz first
wifi: rtw89: 8851b: configure CRASH_TRIGGER feature for 8851B
wifi: rtw89: set TX power without precondition during setting channel
wifi: rtw89: debug: txpwr table access only valid page according to chip
wifi: rtw89: 8851b: enable hw_scan support
wifi: cfg80211: move scan done work to wiphy work
wifi: cfg80211: move sched scan stop to wiphy work
...
====================
Link: https://lore.kernel.org/r/87bkhohkbg.fsf@kernel.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
When the embedded cpu supports SRIOV it can be enabled and disabled
independently from the host SRIOV. Track the pages separately so we can
properly wait for returned VF pages.
Signed-off-by: Daniel Jurgens <danielj@nvidia.com>
Reviewed-by: William Tu <witu@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
These functions are for query/set by vport, there was an underlying
assumption that vport was equal to function ID. That's not the case for
EC VF functions. Set the ec_vf_function bit accordingly.
Signed-off-by: Daniel Jurgens <danielj@nvidia.com>
Reviewed-by: William Tu <witu@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
Add ec_vf_vport_base to HCA Capabilities 2. This indicates the base vport
of embedded CPU virtual functions that are connected to the eswitch.
Add ec_vf_function to query/set_hca_caps. If set this indicates
accessing a virtual function on the embedded CPU by function ID. This
should only be used with other_function set to 1.
Signed-off-by: Daniel Jurgens <danielj@nvidia.com>
Reviewed-by: Bodong Wang <bodong@nvidia.com>
Reviewed-by: William Tu <witu@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
Broadcom PHYs have two LEDs selector registers which allow us to control
the LED assignment, including how to turn them on/off.
Signed-off-by: Florian Fainelli <florian.fainelli@broadcom.com>
Reviewed-by: Simon Horman <simon.horman@corigine.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
These registers are common to most PHYs and are not specific to the
BCM5482, renamed the constants accordingly, no functional change.
Signed-off-by: Florian Fainelli <florian.fainelli@broadcom.com>
Reviewed-by: Simon Horman <simon.horman@corigine.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
