Eric Dumazet says:
====================
vxlan: fixes for skb header pulling, cloning, and concurrency in TX path
While working on RTNL-less fill_info for vxlan, Sashiko found annoying
pre-existing issues, adding noise to an already complex work.
This series addresses some of them in VXLAN transmit path,
primarily within route_shortcircuit(), header validation, and neighbour
lookup.
Patch 1 fixes a potential use-after-free in vxlan_xmit() caused by caching
the Ethernet header pointer ('eth') before calling route_shortcircuit(),
which can reallocate skb->head via pskb_may_pull().
Patch 2 calls skb_cow_head() in route_shortcircuit() before modifying the
Ethernet header in-place, preventing packet header corruption when the skb
is cloned (e.g., by packet sockets, tcpdump, or dev_queue_xmit).
Patch 3 replaces direct reads of n->ha in route_shortcircuit() with
neigh_ha_snapshot() to safely snapshot the neighbour hardware address
under seqlock protection, avoiding potential torn reads during
asynchronous updates.
Patch 4 changes route_shortcircuit() to use pskb_network_may_pull() instead
of pskb_may_pull(). Since skb->data points to the MAC header on transmit
(skb_network_offset(skb) == ETH_HLEN), pskb_may_pull() was only checking
6 bytes into the IP header, leaving the remainder un-pulled in non-linear
frags.
Patch 5 applies pskb_network_may_pull() to the remaining transmit-path
header pull checks in arp_reduce(), ND solicitation proxy checks, and
MDB entry lookup, where skb->data similarly points to the Ethernet header.
====================
Link: https://patch.msgid.link/20260723144249.759100-1-edumazet@google.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
In vxlan_xmit(), arp_reduce(), and vxlan_mdb_entry_skb_get(), pskb_may_pull() was
being called to verify the availability of network layer headers (ARP, IPv6/ND,
IP/IPv6 MDB keys).
However, during transmit skb->data points to the MAC header, so skb_network_offset(skb)
is ETH_HLEN (14 bytes). Using pskb_may_pull(skb, len) only checks len bytes from skb->data
rather than skb_network_offset(skb) + len, which can leave part of the network header
in non-linear frags.
Replace these remaining pskb_may_pull() calls with pskb_network_may_pull() to properly
account for the MAC header offset.
Fixes: e4f67addf1 ("add DOVE extensions for VXLAN")
Fixes: f564f45c45 ("vxlan: add ipv6 proxy support")
Fixes: 0f83e69f44 ("vxlan: Add MDB data path support")
Signed-off-by: Eric Dumazet <edumazet@google.com>
Cc: stable@vger.kernel.org
Reviewed-by: Vadim Fedorenko <vadim.fedorenko@linux.dev>
Reviewed-by: Ido Schimmel <idosch@nvidia.com>
Link: https://patch.msgid.link/20260723144249.759100-6-edumazet@google.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
route_shortcircuit() currently calls pskb_may_pull(skb, sizeof(struct iphdr))
(or ipv6hdr), which checks if bytes are available starting from skb->data.
However, in vxlan_xmit(), skb->data points to the MAC header, so
skb_network_offset(skb) is ETH_HLEN (14 bytes). Using pskb_may_pull(skb, 20)
only checks 20 bytes from skb->data (which is 14 bytes MAC header + 6 bytes of
IP header), leaving the rest of the IP header potentially un-pulled in non-linear
frags. Subsequent dereferences of ip_hdr(skb)->daddr can read beyond the pulled
linear buffer length.
Fix this by using pskb_network_may_pull(), which adds skb_network_offset(skb) to
the length check to ensure the full network header is present in the linear buffer.
Fixes: e4f67addf1 ("add DOVE extensions for VXLAN")
Cc: stable@vger.kernel.org
Signed-off-by: Eric Dumazet <edumazet@google.com>
Reviewed-by: Vadim Fedorenko <vadim.fedorenko@linux.dev>
Link: https://patch.msgid.link/20260723144249.759100-5-edumazet@google.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
The neighbour hardware address n->ha can be updated asynchronously by the
neighbour subsystem, protected by n->ha_lock seqlock. Reading n->ha without
holding the seqlock loop can lead to torn reads or reading a partially updated
MAC address.
Use neigh_ha_snapshot() in route_shortcircuit() to safely copy n->ha under
read_seqbegin()/read_seqretry() lock protection before using it.
Note that arp_reduce() and neigh_reduce() seem to have the same issue
left for future patches.
Fixes: e4f67addf1 ("add DOVE extensions for VXLAN")
Cc: stable@vger.kernel.org
Signed-off-by: Eric Dumazet <edumazet@google.com>
Reviewed-by: Vadim Fedorenko <vadim.fedorenko@linux.dev>
Link: https://patch.msgid.link/20260723144249.759100-4-edumazet@google.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
When route_shortcircuit() performs L3 short-circuit routing, it modifies
the Ethernet header of the skb in-place:
memcpy(eth_hdr(skb)->h_source, eth_hdr(skb)->h_dest, dev->addr_len);
memcpy(eth_hdr(skb)->h_dest, n->ha, dev->addr_len);
If the incoming skb is cloned (for example by packet sockets, tcpdump, or
dev_queue_xmit), modifying the Ethernet header without uncloning can corrupt
the packet header for other readers holding a reference to the cloned skb.
Ensure the skb header is writable and unshared by calling skb_cow_head(skb, 0)
prior to updating the Ethernet header. If skb_cow_head() fails, abort short-circuiting
and return false to allow standard packet processing fallback.
Fixes: e4f67addf1 ("add DOVE extensions for VXLAN")
Cc: stable@vger.kernel.org
Signed-off-by: Eric Dumazet <edumazet@google.com>
Link: https://patch.msgid.link/20260723144249.759100-3-edumazet@google.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Before route_shortcircuit(), the eth header pointer is cached from eth_hdr(skb).
Inside route_shortcircuit(), pskb_may_pull() can be called, which may
reallocate skb->head.
In this case, returning to vxlan_xmit() leaves the cached eth pointer pointing to
freed memory, leading to a use-after-free when dereferencing eth->h_dest.
Fix this by updating eth = eth_hdr(skb) after calling route_shortcircuit().
Fixes: ae88408256 ("VXLAN: Allow L2 redirection with L3 switching")
Cc: stable@vger.kernel.org
Signed-off-by: Eric Dumazet <edumazet@google.com>
Reviewed-by: Vadim Fedorenko <vadim.fedorenko@linux.dev>
Link: https://patch.msgid.link/20260723144249.759100-2-edumazet@google.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
When inet_getpeer_v4() or inet_getpeer_v6() fails to allocate a peer entry
under memory pressure or tree size caps, redirect handlers previously fell
back to sending un-rate-limited ICMP/NDISC Redirect messages.
In IPv4, ip_rt_send_redirect() called icmp_send() directly when peer == NULL.
In IPv6, ip6_forward() and ndisc_send_redirect() passed a NULL peer into
inet_peer_xrlim_allow(), which returned true when peer == NULL.
Because ICMP/NDISC Redirects are not part of the default global rate limit
mask (sysctl_icmp_ratemask), sending redirects when peer == NULL creates
an un-rate-limited ICMP packet storm.
Fix this by failing closed in ip_rt_send_redirect(), ip6_forward(), and
ndisc_send_redirect() when peer is NULL.
Fixes: 92d8682926 ("inetpeer: Move ICMP rate limiting state into inet_peer entries.")
Signed-off-by: Eric Dumazet <edumazet@google.com>
Reviewed-by: Ido Schimmel <idosch@nvidia.com>
Link: https://patch.msgid.link/20260724072901.1633601-1-edumazet@google.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Pablo Neira Ayuso says:
====================
Netfilter/IPVS fixes for net
The following batch contains Netfilter/IPVS fixes for net. This batch
includes a mix of IPVS follow ups related to Sashiko reports, as well as
crash fixes for connection tracking expectation, helpers, ipset and
nf_tables mostly for old bugs. This also includes a fix for the
flowtable tunnel selftest.
1) Use s32 instead of s16 to calculate the remaining payload containing
SIP messages, otherwise underflow is possible allowing out-of-bound
memory access beyond the skb->data area. From Xiang Mei.
2) Fix the counter check in the flowtable selftest for tunnels, from
Lorenzo Bianconi.
3) Add and use nf_ct_expect_related_pair() to add the RTP and RTCP
expectations under the expectation lock, this is required by the SIP
and H.323 NAT helpers. This fixes a possible reinsertion of an
expectation with the DEAD flag set on while looping to find
consecutive ports.
4) Fix ipset UaF during table resize by blocking comment updates on
kernel-side adds. From David Lee.
5) Do not propagate the IP_VS_CONN_F_ONE_PACKET flag when using IPVS
state synchronization, otherwise reaching stale freed from
ip_vs_conn struct is possible, Zhiling Zou.
6) Adjust the hn1 hash node when the forwarding method changes between
MASQ and non-MASQ for an already hashed connection. This can leave
stale hash nodes pointing to a freed struct ip_vs_conn and trigger
UaF while reading /proc/net/ip_vs_conn. From Julian Anastasov.
7) nft_object rhltable needs to be per table, just like chain rhltable,
otherwise UaF from object lookup path while netns is being released.
There is also the nlevent path that can reach stale objects. Placing
this rhltable under the table hierarchy fixes this issue.
8) Reject invalid combined usage of hashlimit tables with and without
XT_HASHLIMIT_RATE_MATCH flag mode, otherwise access to uninitialized
.burst field of dsthash_ent is possible.
9) Fix checksum validations in IPVS performed from LOCAL_IN,
from Julian Anastasov.
10) Fix incorrect packet offset to layer 4 protocol in IPVS, uncovered
by Sashiko, from Julian Anastasov.
11) Skip the mangling of ICMP replies for non-first fragments, also
reported by Sashiko. Also from Julian.
12) Clear ip_vs_conn flags under the spinlock to fix a possible data
race. From Julian Anastasov.
13) Fix incorrect calculation of the payload bitmask in the nf_tables
hardware offload support, leading to UBSAN splat. From Xiang Mei.
* tag 'nf-26-07-23' of git://git.kernel.org/pub/scm/linux/kernel/git/netfilter/nf:
netfilter: nft_payload: fix mask build for partial field offload
ipvs: clear the nfct flag under lock
ipvs: do not mangle ICMP replies for non-first fragments
ipvs: fix places with wrong packet offsets
ipvs: fix the checksum validations
netfilter: xt_hashlimit: validate hashtable supports XT_HASHLIMIT_RATE_MATCH
netfilter: nf_tables: make nft_object rhltable per table
ipvs: adjust double hashing when fwd method changes
ipvs: do not propagate one-packet flag to synced conns
netfilter: ipset: do not update comments from kernel-side hash adds
netfilter: nf_conntrack_expect: add and use nf_ct_expect_related_pair()
selftests: netfilter: nft_flowtable.sh: fix offload counter verification for tunnel tests
netfilter: nf_conntrack_sip: widen NAT rewrite delta to s32 in sip_help_tcp()
====================
Link: https://patch.msgid.link/20260723163910.274695-1-pablo@netfilter.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
A frag_list skb can reach veth with data_len set but nr_frags zero.
veth_convert_skb_to_xdp_buff() only converts skbs that are shared,
locked, have frags[], or do not have enough headroom. It later uses
skb_is_nonlinear() to decide whether to set XDP_FLAGS_HAS_FRAGS and
xdp_frags_size.
That exposes frag_list data to XDP as if it were stored in frags[], but
frags[] is empty. AF_XDP copy mode can then trust the bogus XDP fragment
metadata, walk an empty fragment entry, and crash in memcpy() from
__xsk_rcv().
Route non-linear skbs through skb_pp_cow_data() before exposing them to
XDP, and only advertise XDP frags when the resulting skb has frags[].
skb_copy_bits() already handles frag_list input, and skb_pp_cow_data()
builds frags[] output with skb_add_rx_frag(), which is the
representation XDP multi-buffer expects.
Fixes: 718a18a0c8 ("veth: Rework veth_xdp_rcv_skb in order to accept non-linear skb")
Cc: stable@vger.kernel.org
Signed-off-by: Matt Fleming <mfleming@cloudflare.com>
Reviewed-by: Toke Høiland-Jørgensen <toke@toke.dk>
Acked-by: Lorenzo Bianconi <lorenzo@kernel.org>
Link: https://patch.msgid.link/20260722191925.2192070-1-matt@readmodwrite.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
pktgen_change_name() replaces pkt_dev->entry while holding t->if_lock.
pktgen_remove_device() removes the same entry before
_rem_dev_from_if_list() takes that lock.
This allows the following interleaving:
CPU 0 (NETDEV_CHANGENAME) CPU 1 (kpktgend)
if_lock(t)
proc_remove(pkt_dev->entry)
proc_remove(pkt_dev->entry)
pkt_dev->entry = proc_create_data(...)
if_unlock(t)
The kthread can pass the stale proc_dir_entry to proc_remove() after the
rename path has freed it. A reproducer with a widened race window reports:
BUG: KASAN: slab-use-after-free in proc_remove+0x78/0x80
Read of size 8 at addr ffff8881478fea70 by task kpktgend_0/67
Call Trace:
proc_remove+0x78/0x80
pktgen_remove_device.isra.0+0x11c/0x4c0
pktgen_thread_worker+0x1214/0x6bc0
kthread+0x2c6/0x3b0
Allocated by task 95:
__proc_create+0x204/0x790
proc_create_data+0x72/0xe0
pktgen_thread_write+0xd61/0x1510
Freed by task 28:
kmem_cache_free+0xcb/0x3d0
proc_free_inode+0x5b/0x80
rcu_core+0x50a/0x1850
The buggy address belongs to the object at ffff8881478fea00
which belongs to the cache proc_dir_entry of size 192
Move proc_remove() into the if_lock-protected list removal helper. Keep it
before list_del_rcu() to preserve the ordering required by add_device().
The rename path must then finish replacing the entry before removal, or
it observes that the device is no longer on the list.
Fixes: 39df232f1a ("[PKTGEN]: fix device name handling")
Cc: stable@vger.kernel.org
Signed-off-by: Chengfeng Ye <nicoyip.dev@gmail.com>
Reviewed-by: Simon Horman <horms@kernel.org>
Link: https://patch.msgid.link/20260719145740.2888967-1-nicoyip.dev@gmail.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
When cake_configure_rates() is called from the dequeue path with
rate_adjust=true, it only needs to update the rate parameters. The
loop that clears the unused tins is both unnecessary and harmful in
this path:
- cake_clear_tin() overwrites q->cur_tin and q->cur_flow, which are
actively used by cake_dequeue(), corrupting the dequeue state.
- iterating over the unused tins and their internal queues to purge
packets adds needless overhead to the hot path.
Skip the entire loop when rate_adjust is set, as neither
cake_clear_tin() nor the mtu_time update are needed when only the
rate changes.
The clearing loop runs on every rate adjustment from the dequeue path,
clearing (max_tins - cur_tins) tins each time, so the cost grows the
fewer tins the configured mode actually uses. Testing cake_mq over veth
(8 rx/tx queues, 2 Gbit limit) with flent's [1] rrul and tcp_nup tests and
32 TCP upstreams shows a large drop in loaded latency and a throughput
gain, restoring behaviour to pre-15c2715a5264 levels:
+------------+------+------+-------+-------+---------+
| kernel | mode | test | base | load | tput |
| | | | (ms) | (ms) | (Mbit) |
+------------+------+------+-------+-------+---------+
| net-next | be | rrul | 0.810 | 11.78 | 1469.67 |
| net-next | be | nup | 0.637 | 85.71 | 1243.15 |
| net-next | ds3 | rrul | 0.397 | 15.28 | 1770.06 |
| net-next | ds3 | nup | 0.351 | 15.98 | 1799.39 |
+------------+------+------+-------+-------+---------+
| patched | be | rrul | 0.092 | 0.56 | 1873.40 |
| patched | be | nup | 0.109 | 1.82 | 1869.12 |
| patched | ds3 | rrul | 0.097 | 0.98 | 1866.10 |
| patched | ds3 | nup | 0.101 | 0.51 | 1861.79 |
+------------+------+------+-------+-------+---------+
The same trend holds on real hardware (IPQ8074A, 4 rx/tx queues,
OpenWrt): in besteffort mode the tcp_nup loaded latency drops from
~470 ms to ~4 ms.
[1] https://flent.org
Fixes: 15c2715a52 ("net/sched: sch_cake: fixup cake_mq rate adjustment for diffserv config")
Signed-off-by: Jonas Köppeler <j.koeppeler@tu-berlin.de>
Tested-by: Mike Pham <mikepham4321@gmail.com>
Acked-by: Toke Høiland-Jørgensen <toke@toke.dk>
Link: https://patch.msgid.link/20260720-sch_cake-skip-clearing-tins-v2-1-e6a8b0275c73@tu-berlin.de
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Commit c191b319f2 ("dpll: allow registering FW-identified pin with a
different DPLL") relaxed dpll_pin_register() to let fwnode-identified pins
register with DPLLs from a different driver. This allows, for example, the
ICE driver to register a zl3073x-created pin with its TXC DPLL using
ice_dpll_txclk_ops, which lack frequency_get and phase_adjust_get
callbacks.
After such cross-driver registration, the pin's dpll_refs xarray contains
refs from both drivers. dpll_cmd_pin_get_one() calls
dpll_xa_ref_dpll_first() which returns the ref with the lowest DPLL id.
When the foreign DPLL (e.g. ICE TXC) has a lower id than the owner DPLL
(e.g. zl3073x), the foreign ops are used for reporting. Since those ops
lack callbacks like frequency_get, pin-level attributes are silently
omitted from the netlink response.
For example, a zl3073x output pin that should report frequency and
phase-adjust shows neither:
Before:
# dpll pin show id 45
pin id 45:
module-name: zl3073x
clock-id: 3427468959636104019
board-label: 156M25_NAC0_CLKREF_SYNC
package-label: OUT3
type: synce-eth-port
capabilities: 0x0
phase-adjust-min: -2147483648
phase-adjust-max: 2147483647
phase-adjust-gran: 800
parent-device:
...
After:
# dpll pin show id 19
pin id 19:
module-name: zl3073x
clock-id: 15964355450360090479
board-label: 156M25_NAC0_CLKREF_SYNC
package-label: OUT3
type: synce-eth-port
frequency: 156250000 Hz
frequency-supported:
156250000 Hz
capabilities: 0x0
phase-adjust-min: -2147483648
phase-adjust-max: 2147483647
phase-adjust-gran: 800
phase-adjust: 0
parent-device:
...
Fix this by:
1. Adding dpll_pin_own_dpll_ref_first() helper that returns the first ref
whose DPLL matches the pin's (module, clock_id) tuple -- i.e. the DPLL
from the driver that created the pin and has the complete set of ops.
Return NULL if no owner ref is found.
2. Using dpll_pin_own_dpll_ref_first() in dpll_cmd_pin_get_one() with a
fallback to dpll_xa_ref_dpll_first() for pin-on-pin child pins whose
dpll_refs all point to a different driver's DPLLs.
3. Using dpll_pin_own_dpll_ref_first() in SET operations
(dpll_pin_freq_set, dpll_pin_esync_set, dpll_pin_ref_sync_state_set,
dpll_pin_phase_adj_set) returning -ENODEV if no owner ref exists.
Replacing the validation loops that rejected the entire operation when
any ref's ops lacked the required callback -- instead validate only the
owner refs so that foreign DPLLs with incomplete ops no longer block
SET operations.
4. Guarding all SET and rollback xa_for_each loops against NULL set
callbacks so that foreign refs without the operation are safely skipped
instead of causing a NULL pointer dereference.
Fixes: c191b319f2 ("dpll: allow registering FW-identified pin with a different DPLL")
Signed-off-by: Ivan Vecera <ivecera@redhat.com>
Acked-by: Vadim Fedorenko <vadim.fedorenko@linux.dev>
Link: https://patch.msgid.link/20260714125945.1823269-1-ivecera@redhat.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Maciej Fijalkowski says:
====================
xsk: fix AF_XDP multi-buffer Tx descriptor reclaim
This series fixes several AF_XDP multi-buffer Tx paths where descriptors
consumed from the Tx ring are not consistently returned to userspace
through the completion ring when the packet is later dropped as invalid.
The affected cases are invalid or oversized multi-buffer Tx packets in
both the generic and zero-copy paths. In these cases, the kernel can
consume one or more Tx descriptors while building or validating a
multi-buffer packet, then drop the packet before it reaches the device.
Userspace still owns the UMEM buffers only after the corresponding
addresses are returned through the CQ. Missing completions therefore
make userspace lose track of those buffers.
The generic path fixes cover following related cases:
* partially built multi-buffer skbs dropped by xsk_drop_skb();
continuation descriptors left in the Tx ring after xsk_build_skb()
reports overflow;
* invalid descriptors encountered in the middle of a multi-buffer
packet, including the offending invalid descriptor itself.
The zero-copy path is handled separately. The batched Tx parser now
distinguishes descriptors that can be passed to the driver from
descriptors that are consumed only because they belong to an invalid
multi-buffer packet. Reclaim-only descriptors are written to the CQ
address area and published in completion order, after any earlier
driver-visible Tx descriptors.
The last two patches update xskxceiver so the tests account invalid
multi-buffer Tx packets as descriptors that must be reclaimed, while
still not expecting those invalid packets on the Rx side.
This is a follow-up to Jason's changes [0] which were addressing generic
xmit only and this set allows me to pass full xskxceiver test suite run
against ice driver.
====================
Link: https://patch.msgid.link/20260719135609.147823-1-maciej.fijalkowski@intel.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Invalid Tx descriptors are now returned through the completion ring,
regardless of whether they form a standalone packet or belong to an
invalid multi-buffer packet.
The selftests previously counted only descriptors belonging to valid
packets, with a special exception for some invalid multi-buffer packets
in verbatim streams. This undercounts completion entries when a
standalone invalid descriptor or another invalid packet is reclaimed by
the kernel.
Keep valid_pkts as the number of packets expected on the Rx side, but
count every descriptor submitted to the Tx ring in valid_frags, as every
such descriptor is now expected to be returned through the completion
ring.
Make fragment counting in verbatim mode follow the packet boundary
instead of stopping at the first invalid fragment. Update custom stream
generation so an invalid middle fragment terminates the generated Rx
packet while Tx completion accounting still covers the complete invalid
packet.
Also add explicit end fragments after invalid middle descriptors. This
exercises the kernel drain logic and verifies that subsequent valid
packets are not interpreted as continuations of the invalid packet.
Reviewed-by: Jason Xing <kernelxing@tencent.com>
Signed-off-by: Maciej Fijalkowski <maciej.fijalkowski@intel.com>
Acked-by: Stanislav Fomichev <sdf@fomichev.me>
Link: https://patch.msgid.link/20260719135609.147823-7-maciej.fijalkowski@intel.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
The too-many-frags test describes a packet that is valid from the Tx
ring ownership point of view, but invalid for transmission because it
exceeds the supported number of fragments.
Keep the generated Tx descriptors valid so that __send_pkts() accounts
them as outstanding descriptors that must be reclaimed through the CQ.
Then mark the corresponding Rx packet invalid so the test still does
not expect the oversized packet to appear on the receive side.
Add a valid synchronization packet after the oversized packet so the
test can verify that the Tx path drains the bad packet and resumes at
the next packet boundary.
Reviewed-by: Jason Xing <kernelxing@tencent.com>
Signed-off-by: Maciej Fijalkowski <maciej.fijalkowski@intel.com>
Acked-by: Stanislav Fomichev <sdf@fomichev.me>
Link: https://patch.msgid.link/20260719135609.147823-6-maciej.fijalkowski@intel.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
The zero-copy Tx batch parser stops when it encounters an invalid
descriptor. If this happens after one or more continuation descriptors,
the Tx consumer can be advanced past fragments that are neither submitted
to the driver nor returned to userspace through the completion ring.
A similar problem occurs when a packet exceeds xdp_zc_max_segs. The
descriptors consumed up to the limit are released without completion, and
the remaining continuation descriptors can subsequently be interpreted
as the beginning of another packet.
Parse Tx batches in packet units and distinguish descriptors belonging to
complete valid packets from descriptors consumed while draining an
invalid or oversized packet. Return the former to the driver and append
the latter to the CQ address area so userspace can reclaim their UMEM
frames.
Treat a standalone invalid descriptor as a one-descriptor reclaim-only
packet. Advancing the Tx-ring consumer releases the ring slot, but does
not by itself return ownership of the referenced UMEM frame to userspace.
Once draining starts, continue until the packet's end-of-packet
descriptor is consumed. Preserve the drain state on the socket when EOP
has not yet been supplied, so draining can continue during a later call.
Leave incomplete but otherwise valid packets on the Tx ring.
Shared-UMEM pools using multi-buffer Tx also need packet-framed parsing.
Walk their Tx sockets one packet at a time, preserving the existing
per-socket fairness scheme, instead of using the legacy one-descriptor
fallback. Keep that fallback for shared pools that do not use
multi-buffer Tx. Since the drain state is maintained per socket and both
the singular and shared paths can resume an interrupted drain, changing
the socket list from singular to shared requires no special bind-time
transition.
CQ entries are positional, and drivers may complete only part of the Tx
work returned by xsk_tx_peek_release_desc_batch(). Therefore, reclaim-only
entries cannot be published immediately when earlier driver-visible
descriptors are still outstanding.
Track the number of driver-visible CQ entries preceding the reclaim
entries. Let xsk_tx_completed() publish partial hardware Tx completions,
and publish the reclaim entries only after every earlier Tx descriptor
has completed. Complete a reclaim-only batch immediately when there is no
driver-visible work in front of it, and prevent another Tx batch from
being appended while reclaim entries remain pending.
Also cap batch processing by the size of the pool's temporary descriptor
array, as Tx rings belonging to sockets sharing a UMEM may have different
sizes.
This ensures that every invalid Tx descriptor consumed by the ZC batch
path is either submitted to the driver as part of a valid packet or
returned to userspace without violating CQ completion ordering.
Reviewed-by: Jason Xing <kernelxing@tencent.com>
Signed-off-by: Maciej Fijalkowski <maciej.fijalkowski@intel.com>
Acked-by: Stanislav Fomichev <sdf@fomichev.me>
Fixes: cf24f5a5fe ("xsk: add support for AF_XDP multi-buffer on Tx path")
Link: https://patch.msgid.link/20260719135609.147823-5-maciej.fijalkowski@intel.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
The temporary Tx descriptor array in an XSK buffer pool is currently
sized from the Tx ring of the socket that creates the pool.
This is insufficient for shared-UMEM Tx. A later socket may have a
larger Tx ring and submit a valid multi-buffer packet containing more
descriptors than the first socket's ring, while still remaining within
the device's xdp_zc_max_segs limit.
A packet-framed batch parser bounded by the temporary array cannot reach
the end-of-packet descriptor in that case. It leaves the packet on the
Tx ring and encounters the same packet on every subsequent attempt,
stalling Tx processing for that socket.
Size the temporary descriptor array to the larger of the first Tx ring
and the device's xdp_zc_max_segs capability. This keeps the array large
enough to inspect one maximum-sized valid packet. Larger shared Tx rings
do not require further resizing, as they can be processed over multiple
batches.
Following commit will actually address the data path side.
Fixes: d558196604 ("xsk: support ZC Tx multi-buffer in batch API")
Reviewed-by: Jason Xing <kernelxing@tencent.com>
Signed-off-by: Maciej Fijalkowski <maciej.fijalkowski@intel.com>
Acked-by: Stanislav Fomichev <sdf@fomichev.me>
Link: https://patch.msgid.link/20260719135609.147823-4-maciej.fijalkowski@intel.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
This patch is inspired by the check[1] from sashiko. It says when
overflow happens, the address of cq to be published is invalid.
Actually the severer thing is the whole process of publishing the
address of cq in this particular case is not right: it should truely
publish the address and advance the cached_prod in cq as long as it
reads descriptors from txq.
The following is the full analysis.
xsk_drop_skb() is called in three places, which all discard a partially
built multi-buffer skb:
1) xsk_build_skb() -EOVERFLOW error path: packet exceeds MAX_SKB_FRAGS
2) __xsk_generic_xmit() post-loop cleanup: an invalid descriptor in
the TX ring prevents the partial packet from completing
3) xsk_release(): socket close while xs->skb holds an incomplete packet
In all three cases, the TX descriptors for the already-processed frags
have been consumed from the TX ring (xskq_cons_release), and CQ slots
have been reserved. However, xsk_drop_skb() calls xsk_consume_skb()
which cancels the CQ reservations via xsk_cq_cancel_locked(). Since
the buffer addresses never appear in the completion queue, userspace
permanently loses track of these buffers.
Fix this by letting consume_skb() trigger the existing xsk_destruct_skb
destructor, which already submits buffer addresses to the CQ via
xsk_cq_submit_addr_locked().
Note that cancelling the descriptors back to the TX ring (via
xskq_cons_cancel_n) is not a appropriate option because an oversized
packet that always exceeds MAX_SKB_FRAGS would be retried indefinitely,
which is an obviously deadlock bug in the TX path.
Also move the desc->addr assignment in xsk_build_skb() above the
overflow check so that the current descriptor's address is recorded
before a potential -EOVERFLOW jump to free_err, consistent with the
zerocopy path in xsk_build_skb_zerocopy().
[1]: https://lore.kernel.org/all/20260425041726.85FB3C2BCB2@smtp.kernel.org/
Fixes: cf24f5a5fe ("xsk: add support for AF_XDP multi-buffer on Tx path")
Acked-by: Maciej Fijalkowski <maciej.fijalkowski@intel.com>
Signed-off-by: Jason Xing <kernelxing@tencent.com>
Acked-by: Stanislav Fomichev <sdf@fomichev.me>
Link: https://patch.msgid.link/20260719135609.147823-2-maciej.fijalkowski@intel.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Add a regression test for the unix_listen() state check. The key case is
listen() on a bound socket that has already been connected: it is no
longer in TCP_CLOSE or TCP_LISTEN, so it must fail with EINVAL. A
prepare_peercred() call slipped in ahead of that check once left err at 0
and made listen() silently succeed there instead; this guards against a
repeat.
The neighbouring outcomes are covered too so they cannot regress the same
way: a bound socket in TCP_CLOSE listens fine, calling listen() again on a
socket already in TCP_LISTEN is allowed, and an unbound socket fails with
EINVAL.
Each case runs for both listenable socket types (SOCK_STREAM and
SOCK_SEQPACKET) and both pathname and abstract addresses.
Fixes: fd0a109a0f ("net, pidfs: prepare for handing out pidfds for reaped sk->sk_peer_pid")
Signed-off-by: John Ericson <mail@johnericson.me>
Link: https://patch.msgid.link/20260718182903.2295560-2-John.Ericson@Obsidian.Systems
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Commit fd0a109a0f ("net, pidfs: prepare for handing out pidfds for
reaped sk->sk_peer_pid") inserted a prepare_peercred() call between err
= -EINVAL and the socket-state check in unix_listen(). Since
prepare_peercred() leaves err at 0 on success, listen() on an AF_UNIX
socket that is not in TCP_CLOSE or TCP_LISTEN state (e.g. one that is
already connected) now silently returns success without doing anything,
instead of failing with EINVAL as it did before.
Fixes: fd0a109a0f ("net, pidfs: prepare for handing out pidfds for reaped sk->sk_peer_pid")
Signed-off-by: John Ericson <mail@johnericson.me>
Link: https://patch.msgid.link/20260718182903.2295560-1-John.Ericson@Obsidian.Systems
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
nh_rt_cache_flush() walks nh->f6i_list during an RTNL-serialized nexthop
replace without holding nh->lock, racing the unlocked IPv6 route
add/delete that mutate the list under nh->lock and free fib6_info
entries (nh_rt_cache_flush() is inlined into rtm_new_nexthop()):
BUG: KASAN: slab-use-after-free in nh_rt_cache_flush (net/ipv4/nexthop.c:2243)
Read of size 8 at addr ffff888012953e18 by task exploit/146
nh_rt_cache_flush (net/ipv4/nexthop.c:2243)
replace_nexthop (net/ipv4/nexthop.c:2610)
rtm_new_nexthop (net/ipv4/nexthop.c:3323)
rtnetlink_rcv_msg (net/core/rtnetlink.c:7076)
Unlike the other f6i_list walks, this one bumps each route's sernum via
fib6_update_sernum_upto_root(), which needs tb6_lock; taking nh->lock
around it would invert the established tb6_lock -> nh->lock order and
deadlock. As the only purpose is to invalidate cached dsts, bump the
IPv6 sernum for the whole netns with rt_genid_bump_ipv6() instead,
mirroring the rt_cache_flush() already done for IPv4 just above.
Fixes: 081efd1832 ("ipv6: Protect nh->f6i_list with spinlock and flag.")
Reported-by: AutonomousCodeSecurity@microsoft.com
Signed-off-by: Xiang Mei (Microsoft) <xmei5@asu.edu>
Reviewed-by: Ido Schimmel <idosch@nvidia.com>
Link: https://patch.msgid.link/20260722002951.2614721-2-xmei5@asu.edu
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
fib6_check_nh_list() and __nexthop_replace_notify() walk nh->f6i_list
during an RTNL-serialized nexthop replace without holding nh->lock. IPv6
RTM_NEWROUTE/RTM_DELROUTE run without RTNL and mutate that list under
nh->lock (fib6_add_rt2node_nh(), fib6_purge_rt()), so both walks race a
concurrent route delete that unlinks and frees a fib6_info:
BUG: KASAN: slab-use-after-free in rt6_fill_node.isra.0 (net/ipv6/route.c:5799)
Read of size 4 at addr ffff888014607e64 by task exploit/143
rt6_fill_node.isra.0 (net/ipv6/route.c:5799)
fib6_rt_update (net/ipv6/route.c:6412)
__nexthop_replace_notify (net/ipv4/nexthop.c:2542)
rtm_new_nexthop (net/ipv4/nexthop.c:2554)
rtnetlink_rcv_msg (net/core/rtnetlink.c:7076)
BUG: KASAN: slab-use-after-free in fib6_check_nh_list (net/ipv4/nexthop.c:1605)
Read of size 8 at addr ffff888014a7d068 by task exploit/142
fib6_check_nh_list (net/ipv4/nexthop.c:1605)
rtm_new_nexthop (net/ipv4/nexthop.c:2575)
rtnetlink_rcv_msg (net/core/rtnetlink.c:7076)
Both walks only read the entries and take no tb6_lock, so protect them
with nh->lock; fib6_rt_update() uses gfp_any(), which returns GFP_ATOMIC
under the lock.
Fixes: 081efd1832 ("ipv6: Protect nh->f6i_list with spinlock and flag.")
Reported-by: AutonomousCodeSecurity@microsoft.com
Signed-off-by: Xiang Mei (Microsoft) <xmei5@asu.edu>
Reviewed-by: Ido Schimmel <idosch@nvidia.com>
Link: https://patch.msgid.link/20260722002951.2614721-1-xmei5@asu.edu
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
rds_tcp_laddr_check() looks up a scoped IPv6 interface with
dev_get_by_index_rcu(), drops the RCU read-side lock, and only then
passes the bare struct net_device * into ipv6_chk_addr().
dev_get_by_index_rcu() only keeps the device alive within the same RCU
read-side section. After rcu_read_unlock(), a concurrent RTM_DELLINK can
free the net_device; ipv6_chk_addr() then dereferences the stale pointer
in __ipv6_chk_addr_and_flags() (e.g. l3mdev_master_dev_rcu(dev)), reading
freed memory.
Keep the RCU read-side lock held across the ipv6_chk_addr() call instead
of dropping it right after the lookup, so the device cannot be freed
while it is in use.
BUG: KASAN: slab-use-after-free in __ipv6_chk_addr_and_flags (... net/ipv6/addrconf.c:1998)
Read of size 8 at addr ffff8880106ec000 by task exploit/153
Call Trace:
...
kasan_report (mm/kasan/report.c:595)
__ipv6_chk_addr_and_flags (... net/ipv6/addrconf.c:1998)
ipv6_chk_addr (net/ipv6/addrconf.c:2031 net/ipv6/addrconf.c:1972)
rds_tcp_laddr_check (net/rds/tcp.c:370)
rds_bind (net/rds/bind.c:248)
__sys_bind (net/socket.c:1920)
__x64_sys_bind (net/socket.c:1956)
do_syscall_64 (arch/x86/entry/syscall_64.c:63)
entry_SYSCALL_64_after_hwframe (arch/x86/entry/entry_64.S:121)
Fixes: eee2fa6ab3 ("rds: Changing IP address internal representation to struct in6_addr")
Reported-by: Weiming Shi <bestswngs@gmail.com>
Signed-off-by: Xiang Mei <xmei5@asu.edu>
Reviewed-by: Allison Henderson <achender@kernel.org>
Link: https://patch.msgid.link/20260722210203.565803-1-xmei5@asu.edu
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Pull networking fixes from Jakub Kicinski:
"Lots of fixes, double the count even for the 'new normal'. Largely due
to my time off followed by a networking conference which distracted
most maintainers (less so the AI generators).
Including fixes from Bluetooth and WiFi.
Current release - regressions:
- wifi: mt76: fix MAC address for non OF pcie cards
Current release - new code bugs:
- mptcp: fix BUILD_BUG_ON on legacy ARM config
- wifi: cfg80211: guard optional PMSR nominal time
Previous releases - regressions:
- qrtr: ns: raise node count limit to 512, we arbitrarily picked
256 as a limit, turns out it was too low for real world deployments
- vhost-net: fix TX stall when vhost owns virtio-net header
- eth: amd-xgbe: fix MAC_AUTO_SW handling in CL37 AN
- wifi: ath12k: fix low MLO RX throughput on WCN7850
Previous releases - always broken:
- number of random AI fixes for SCTP, RDS and TIPC protocols
- more AI-looking fixes for WiFi drivers
- number of fixes for missing pointer reloading after skb pull
- reject BPF redirect use from qdisc qevent block
- tcp: initialize standalone TCP-AO response padding
- vsock/virtio: collapse receive queue under memory pressure to avoid
client OOMing the host with tiny messages
- ipv4: icmp: fill flow parameters in icmp_route_lookup decoy lookup,
make sure the ICMP response routing follows the routing policy
- gro: fix double aggregation of flush-marked skbs
- ovpn: fix various refcount bugs
- tls: device: push pending open record on splice EOF
- eth: mlx5:
- use sender devcom for MPV master-up
- fix MCIA register buffer overflow on 32 dword reads"
* tag 'net-7.2-rc5' of git://git.kernel.org/pub/scm/linux/kernel/git/netdev/net: (234 commits)
drop_monitor: perform u64_stats updates under IRQ-disabled section
drop_monitor: fix size calculations for 64-bit attributes
net: drop_monitor: fix info leak in NET_DM_ATTR_PAYLOAD
mptcp: fix BUILD_BUG_ON on legacy ARM config
selftests: mptcp: userspace_pm: fix undefined variable port
mptcp: fix stale skb->sk reference on subflow close
mptcp: pm: userspace: fix use-after-free in get_local_id
mptcp: decrement subflows counter on failed passive join
mac802154: hold an interface reference across the scan worker
sctp: don't free the ASCONF's own transport in DEL-IP processing
phonet: check register_netdevice_notifier() error in phonet_device_init()
phonet: pep: fix use-after-free in pep_get_sb()
bnge/bng_re: fix ring ID widths
tipc: fix integer overflow in tipc_recvmsg() and tipc_recvstream()
net: airoha: fix ETS channel derivation in airoha_tc_setup_qdisc_ets()
mctp: check register_netdevice_notifier() error in mctp_device_init()
ptp: netc: explicitly clear TMR_OFF during initialization
rds: tcp: unregister sysctl before tearing down listen socket
ipv6: Change allocation flags to match rcu_read_lock section requirements
net: slip: serialize receive against buffer reallocation
...
Eric Dumazet says:
====================
drop_monitor: take care of 32bit kernels
This series fixes two drop_monitor issues on 32-bit architectures:
- Patch 1 uses nla_total_size_64bit() for PC and TIMESTAMP attributes to
account for alignment padding added by nla_put_u64_64bit(), avoiding
potential skb_over_panic() crashes.
- Patch 2 moves u64_stats updates before spin_unlock_irqrestore(), ensuring
local interrupts are disabled to prevent seqcount corruption from nested
interrupts in probe context.
====================
Link: https://patch.msgid.link/20260722141743.3266924-1-edumazet@google.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
In net_dm_packet_trace_kfree_skb_hit() and net_dm_hw_trap_packet_probe(),
u64_stats_update_begin() / u64_stats_inc() / u64_stats_update_end() were
called after spin_unlock_irqrestore(&...drop_queue.lock, flags), when local
IRQs had already been re-enabled.
Tracepoint probes can execute in IRQ or softirq context. On 32-bit
architectures, u64_stats_update_begin() disables preemption but not interrupts,
relying on seqcount writes. If a nested interrupt occurs on the same CPU during
the 64-bit stats update, the reentrant seqcount update can corrupt the
seqcount state or stats value.
Fix this by performing the 64-bit per-CPU stats update before releasing
drop_queue.lock via spin_unlock_irqrestore(), ensuring local interrupts remain
disabled during the u64_stats update.
Fixes: e9feb58020 ("drop_monitor: Expose tail drop counter")
Fixes: 5e58109b1e ("drop_monitor: Add support for packet alert mode for hardware drops")
Signed-off-by: Eric Dumazet <edumazet@google.com>
Link: https://patch.msgid.link/20260722141743.3266924-3-edumazet@google.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
net_dm_packet_report_fill() and net_dm_hw_packet_report_fill() use
nla_put_u64_64bit() to append 64-bit attributes (NET_DM_ATTR_PC and
NET_DM_ATTR_TIMESTAMP).
On 32-bit architectures without CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS,
nla_put_u64_64bit() may append a 4-byte NET_DM_ATTR_PAD attribute for
64-bit alignment.
However, net_dm_packet_report_size() and net_dm_hw_packet_report_size()
used nla_total_size(sizeof(u64)) instead of nla_total_size_64bit(sizeof(u64)),
budgeting 12 bytes instead of up to 16 bytes.
This under-estimation of SKB size can lead to an skb_over_panic() when
__nla_reserve() or skb_put() is subsequently called.
Fix this by using nla_total_size_64bit(sizeof(u64)) in both size calculations.
Fixes: ca30707dee ("drop_monitor: Add packet alert mode")
Fixes: 5e58109b1e ("drop_monitor: Add support for packet alert mode for hardware drops")
Signed-off-by: Eric Dumazet <edumazet@google.com>
Link: https://patch.msgid.link/20260722141743.3266924-2-edumazet@google.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
net_dm_packet_report_fill() and net_dm_hw_packet_report_fill() open code
the NET_DM_ATTR_PAYLOAD attribute to avoid zeroing the packet payload
before overwriting it with skb_copy_bits().
skb_put() reserves nla_total_size(payload_len), i.e. the header plus the
NLA_ALIGN() padding, but only payload_len bytes are copied in. When
payload_len is not a multiple of 4 the 1-3 padding bytes are never
initialized and are leaked to user space inside the netlink message.
KMSAN confirms the leak for the software path when the packet payload
length is not 4-byte aligned:
BUG: KMSAN: kernel-infoleak in _copy_to_iter
_copy_to_iter
__skb_datagram_iter
skb_copy_datagram_iter
netlink_recvmsg
sock_recvmsg
__sys_recvfrom
Uninit was created at:
kmem_cache_alloc_node_noprof
__alloc_skb
net_dm_packet_work
Bytes 173-175 of 176 are uninitialized
Use __nla_reserve(), which sets up the attribute header and zeroes the
padding, instead of open coding the attribute construction.
Fixes: ca30707dee ("drop_monitor: Add packet alert mode")
Fixes: 5e58109b1e ("drop_monitor: Add support for packet alert mode for hardware drops")
Suggested-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: Yehyeong Lee <yhlee@isslab.korea.ac.kr>
Link: https://patch.msgid.link/20260722122817.5548-1-yhlee@isslab.korea.ac.kr
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Matthieu Baerts says:
====================
mptcp: misc fixes for v7.2-rc5
Here are various unrelated fixes:
- Patch 1: decrement extra subflows counter in case of errors with
passive MP_JOIN. A fix for v5.7.
- Patch 2: fix use-after-free in userspace_pm_get_local_id. A fix for
v5.19.
- Patch 3: fix stale skb->sk reference on subflow close, in case of
concurrent read operation. A fix for v6.19.
- Patch 4: wait on the correct port in the userspace_pm.sh selftest. A
fix for v6.19.
- Patch 5: fix a BUILD_BUG_ON on legacy ARM config. A fix for v7.1.
====================
Link: https://patch.msgid.link/20260722-net-mptcp-misc-fixes-7-2-rc5-v1-0-6fb595bc86ef@kernel.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
In make_connection(), the variable "port" is used but never defined.
This leads to an empty argument being passed to wait_local_port_listen(),
causing "printf: : invalid number" errors:
# INFO: Init
# 01 Created network namespaces ns1, ns2 [ OK ]
# INFO: Make connections
# ./../lib.sh: line 651: printf: : invalid number
# 02 Established IPv4 MPTCP Connection ns2 => ns1 [ OK ]
# INFO: Connection info: 10.0.1.2:59516 -> 10.0.1.1:50002
# ./../lib.sh: line 651: printf: : invalid number
# 03 Established IPv6 MPTCP Connection ns2 => ns1 [ OK ]
Fix it by using the correctly defined variable "app_port", which holds the
appropriate port number for the connection.
Fixes: 39348f5f2f ("selftests: mptcp: wait for port instead of sleep")
Cc: stable@vger.kernel.org
Signed-off-by: Geliang Tang <tanggeliang@kylinos.cn>
Reviewed-by: Matthieu Baerts (NGI0) <matttbe@kernel.org>
Signed-off-by: Matthieu Baerts (NGI0) <matttbe@kernel.org>
Link: https://patch.msgid.link/20260722-net-mptcp-misc-fixes-7-2-rc5-v1-4-6fb595bc86ef@kernel.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
The backlog list is updated by mptcp_data_ready() under
mptcp_data_lock(). The cleanup of backlog references to a closing
subflow, however, was performed in mptcp_close_ssk(), before
__mptcp_close_ssk() acquires the ssk lock, and while holding neither
the ssk lock nor mptcp_data_lock().
Because that traversal ran without mptcp_data_lock(), concurrent softirq
RX processing on another CPU (subflow_data_ready() -> mptcp_data_ready()
-> __mptcp_add_backlog(), under mptcp_data_lock()) could add a backlog
entry referencing the ssk while the cleanup loop was in progress. Such
an entry could be missed by the cleanup, or the concurrent list update
could corrupt the traversal, leaving skb->sk pointing at the ssk after
it is freed.
A later mptcp_backlog_purge() then dereferences the stale pointer,
triggering a warning in inet_sock_destruct() (ssk->sk_rmem_alloc != 0)
followed by a use-after-free in mptcp_backlog_purge().
Fix this by moving the backlog cleanup into __mptcp_close_ssk(), after
subflow->closing is set to 1 and while the ssk lock is still held,
serialized under mptcp_data_lock(). The cleanup runs only on the push
path (MPTCP_CF_PUSH), where backlog references accumulate; on other
teardown paths the caller already handles cleanup.
With subflow->closing set and mptcp_data_lock() held across the purge,
any concurrent mptcp_data_ready() either completes its enqueue before
the purge runs and is caught, or observes closing=1 and bails out. Once
mptcp_data_unlock() is reached, no new skb referencing the ssk can be
enqueued, so the cleanup is exhaustive.
Remove the unprotected traversal from mptcp_close_ssk() entirely.
Fixes: ee458a3f31 ("mptcp: introduce mptcp-level backlog")
Cc: stable@vger.kernel.org
Suggested-by: Paolo Abeni <pabeni@redhat.com>
Reported-by: Matthieu Baerts (NGI0) <matttbe@kernel.org>
Closes: https://github.com/multipath-tcp/mptcp_net-next/issues/621
Signed-off-by: Kalpan Jani <kalpan.jani@mpiricsoftware.com>
Acked-by: Paolo Abeni <pabeni@redhat.com>
Signed-off-by: Matthieu Baerts (NGI0) <matttbe@kernel.org>
Link: https://patch.msgid.link/20260722-net-mptcp-misc-fixes-7-2-rc5-v1-3-6fb595bc86ef@kernel.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
In mptcp_pm_userspace_get_local_id(), the address entry is looked up under
spinlock, but its id is read after dropping the lock. A concurrent deletion
can free the entry between the unlock and the read, leading to UAF.
The race window is narrow. It was reproduced only with a locally
constructed stress test that repeatedly overlaps an MP_JOIN SYN with a
MPTCP_PM_CMD_SUBFLOW_DESTROY request.
However, the KASAN report below confirms that the race is reachable:
[ 666.319376] BUG: KASAN: slab-use-after-free in mptcp_userspace_pm_get_local_id+0x1dc/0x1f0
[ 666.319386] Read of size 1 at addr ffff888124845610 by task swapper/0/0
...
[ 666.319401] Call Trace:
[ 666.319405] <IRQ>
[ 666.319408] dump_stack_lvl+0x53/0x70
[ 666.319412] print_address_description.constprop.0+0x2c/0x3b0
[ 666.319418] print_report+0xbe/0x2b0
[ 666.319421] ? mptcp_userspace_pm_get_local_id+0x1dc/0x1f0
[ 666.319423] kasan_report+0xce/0x100
[ 666.319426] ? mptcp_userspace_pm_get_local_id+0x1dc/0x1f0
[ 666.319429] mptcp_userspace_pm_get_local_id+0x1dc/0x1f0
[ 666.319433] mptcp_pm_get_local_id+0x371/0x440
...
[ 666.319821] Allocated by task 45539:
[ 666.319844] kasan_save_stack+0x33/0x60
[ 666.319855] kasan_save_track+0x14/0x30
[ 666.319858] __kasan_kmalloc+0x8f/0xa0
[ 666.319863] __kmalloc_noprof+0x1e7/0x520
[ 666.319867] sock_kmalloc+0xdf/0x130
[ 666.319885] sock_kmemdup+0x1b/0x40
[ 666.319888] mptcp_userspace_pm_append_new_local_addr+0x261/0x500
[ 666.319910] mptcp_pm_nl_announce_doit+0x16a/0x610
...
[ 666.319967] Freed by task 45560:
[ 666.319988] kasan_save_stack+0x33/0x60
[ 666.319991] kasan_save_track+0x14/0x30
[ 666.319994] kasan_save_free_info+0x3b/0x60
[ 666.319998] __kasan_slab_free+0x43/0x70
[ 666.320000] kfree+0x166/0x440
[ 666.320003] sock_kfree_s+0x1d/0x50
[ 666.320007] mptcp_userspace_pm_delete_local_addr.isra.0+0x157/0x200
[ 666.320011] mptcp_pm_nl_subflow_destroy_doit+0x51d/0xea0
Fix by copying the id into a local variable while still holding the lock,
and use -1 as a "not found" sentinel.
Fixes: f012d796a6 ("mptcp: check addrs list in userspace_pm_get_local_id")
Cc: stable@vger.kernel.org
Signed-off-by: Geliang Tang <tanggeliang@kylinos.cn>
Tested-by: Xuanqiang Luo <luoxuanqiang@kylinos.cn>
Reviewed-by: Matthieu Baerts (NGI0) <matttbe@kernel.org>
Signed-off-by: Matthieu Baerts (NGI0) <matttbe@kernel.org>
Link: https://patch.msgid.link/20260722-net-mptcp-misc-fixes-7-2-rc5-v1-2-6fb595bc86ef@kernel.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
mac802154_scan_worker() captures the scanning sub-interface under RCU
and then keeps dereferencing sdata->dev after rcu_read_unlock() and
outside the rtnl -- in the failure traces, in
mac802154_transmit_beacon_req() (skb->dev = sdata->dev), and in the
end_scan cleanup. Nothing keeps that netdev alive across the worker
iteration.
A concurrent DEL_INTERFACE or PHY removal can unregister the interface
once the worker drops the rtnl between its two drv_set_channel()
sections. unregister_netdevice() frees the netdev asynchronously from
netdev_run_todo() with the rtnl already dropped, so neither holding the
rtnl nor the per-PHY IEEE802154_IS_SCANNING flag prevents a stale worker
iteration from dereferencing the freed netdev -- a KASAN
slab-use-after-free, reachable by racing TRIGGER_SCAN against
DEL_INTERFACE (both CAP_NET_ADMIN).
Pin the netdev with netdev_hold() while the RCU read lock is still held,
and release it at every worker exit.
Fixes: 57588c7117 ("mac802154: Handle passive scanning")
Cc: stable@vger.kernel.org
Signed-off-by: Ibrahim Hashimov <security@auditcode.ai>
Link: https://patch.msgid.link/20260721211228.34578-1-security@auditcode.ai
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
sctp_process_asconf() caches the transport the ASCONF chunk is processed
against in asconf->transport (== chunk->transport, set once in sctp_rcv()).
For an ASCONF located through its Address Parameter by
__sctp_rcv_asconf_lookup(), that cached transport corresponds to the
Address Parameter, which need not be the packet's source address.
sctp_process_asconf_param() rejects a DEL-IP for the packet source address
(ADDIP D8, SCTP_ERROR_DEL_SRC_IP), but nothing protects asconf->transport.
A single ASCONF can therefore carry, in order:
[Address Parameter L] [DEL-IP L] [DEL-IP 0.0.0.0]
where L differs from the source. The DEL-IP for L passes the D8 check and
calls sctp_assoc_rm_peer() on the transport that asconf->transport still
points at, freeing it (RCU-deferred). The following wildcard DEL-IP then
reuses the now-dangling asconf->transport in sctp_assoc_set_primary() and
sctp_assoc_del_nonprimary_peers(): set_primary() dereferences the freed
transport (->ipaddr, ->state) and plants the dangling pointer into
asoc->peer.primary_path / active_path, and del_nonprimary_peers(), keeping
only the pointer that is no longer on the list, removes every real
transport, leaving the association with a transport_count of 0 and
primary_path/active_path pointing at freed memory.
Reject a DEL-IP that targets the transport the ASCONF is being processed
against, mirroring the existing source-address guard, so the wildcard
branch can never reuse a freed transport.
Fixes: 42e30bf346 ("[SCTP]: Handle the wildcard ADD-IP Address parameter")
Cc: stable@kernel.org
Signed-off-by: Jun Yang <junvyyang@tencent.com>
Acked-by: Xin Long <lucien.xin@gmail.com>
Link: https://patch.msgid.link/tencent_73762ED1DF08CC9D5F5F61954B01350CFE0A@qq.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
phonet_device_init() registers a netdevice notifier before calling
phonet_netlink_register(), but does not check whether notifier
registration succeeded. On failure, netlink setup still proceeds and
init may return success without the notifier in place.
Also, the existing phonet_netlink_register() failure path called
phonet_device_exit(), which runs rtnl_unregister_all() even though
rtnl_register_many() already unwound any partial registration. Calling
the full exit helper on a partial init is not correct.
Check each registration error, including proc_create_net(), and unwind
only the steps that have succeeded so far, in reverse order.
Signed-off-by: Minhong He <heminhong@kylinos.cn>
Link: https://patch.msgid.link/20260721093956.162617-1-heminhong@kylinos.cn
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
pep_get_sb() doesn't consider that pskb_may_pull() might have relocated
the skb data, and continue to access the older pointer, causing UAF.
Reproduced under KASAN:
BUG: KASAN: slab-use-after-free in pep_get_sb+0x234/0x3b0
Read of size 1 at addr ff11000105510f50 by task repro/157
pep_get_sb+0x234/0x3b0
pipe_handler_do_rcv+0x5f7/0xa10
pep_do_rcv+0x203/0x410
__sk_receive_skb+0x471/0x4a0
phonet_rcv+0x5b3/0x6c0
__netif_receive_skb+0xcc/0x1d0
Refetch the header with skb_header_pointer() after pskb_may_pull(), so
the possibly stale pointer is no longer dereferenced. There are better
ways to solve this, but, this is the less instrusive one.
Fixes: 9641458d3e ("Phonet: Pipe End Point for Phonet Pipes protocol")
Cc: stable@vger.kernel.org
Signed-off-by: Breno Leitao <leitao@debian.org>
Link: https://patch.msgid.link/20260721-phonet_get_sb_uaf-v1-1-95fd7881cc4e@debian.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Firmware requires more than 16 bits to address TX ring IDs for its
internal QP management. Widen the associated HSI ring ID fields to
32 bits. The values firmware assigns remain within 24 bits, bounded
by the hardware doorbell XID field.
The fw_ring_id field belongs to bnge_ring_struct, a common struct
shared by all ring types, so widening it to u32 applies uniformly
across TX, RX, CP, and NQ rings but firmware assigns values within
16-bit range for all ring types except TX, which requires the wider
field.
Note that, Thor Ultra hardware has not yet been deployed and no
firmware has been released to field, so backward compatibility
is not a concern.
Fixes: 42d1c54d62 ("bnge/bng_re: Add a new HSI")
Signed-off-by: Vikas Gupta <vikas.gupta@broadcom.com>
Reviewed-by: Siva Reddy Kallam <siva.kallam@broadcom.com>
Reviewed-by: Dharmender Garg <dharmender.garg@broadcom.com>
Reviewed-by: Yendapally Reddy Dhananjaya Reddy <yendapally.reddy@broadcom.com>
Link: https://patch.msgid.link/20260721063731.2622500-1-vikas.gupta@broadcom.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
In tipc_recvmsg(), the copy length is computed as:
copy = min_t(int, dlen - offset, buflen);
buflen is size_t but min_t(int, ...) casts it to int. When buflen
exceeds INT_MAX (e.g. 0xFFFFFFFF via io_uring provided buffers), it
wraps negative, wins the comparison, and the negative copy length
propagates to simple_copy_to_iter() where int-to-size_t promotion
makes it SIZE_MAX, triggering a WARN_ON. tipc_recvstream() has the
same pattern.
Kernel panic - not syncing: kernel: panic_on_warn set ...
RIP: 0010:simple_copy_to_iter+0x9e/0xd0 (net/core/datagram.c:521)
Call Trace:
__skb_datagram_iter+0x123/0x8b0 (net/core/datagram.c:402)
skb_copy_datagram_iter+0x77/0x1a0 (net/core/datagram.c:534)
tipc_recvmsg+0x3d7/0xe80 (net/tipc/socket.c:1934)
io_recvmsg+0x47e/0xda0
Fix by changing min_t(int, ...) to min_t(size_t, ...) in both
functions. The result is always <= (dlen - offset), which is bounded
by TIPC maximum message size (0x1ffff bytes), so the implicit
narrowing on assignment to int copy is always safe.
Fixes: e9f8b10101 ("tipc: refactor function tipc_sk_recvmsg()")
Fixes: ec8a09fbbe ("tipc: refactor function tipc_sk_recv_stream()")
Reported-by: AutonomousCodeSecurity@microsoft.com
Signed-off-by: Cen Zhang (Microsoft) <blbllhy@gmail.com>
Reviewed-by: Tung Nguyen <tung.quang.nguyen@est.tech>
Link: https://patch.msgid.link/20260720214103.47732-1-blbllhy@gmail.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Antonio Quartulli says:
====================
Included fixes:
* ensure keepalive timestamps are computed using monotonic source
* avoid UAF in unlock_ovpn() when iterating over release_list
* fix memleak in selftest tool
* ensure reference to peer is acquired before scheduling worker
(which may drop the not-yet-taken ref)
* fix refcount leak in case of concurrent TX and RX TCP error
* fix potential refcount unbalance in case of sock release in
P2P mode
* tag 'ovpn-net-20260720' of https://github.com/OpenVPN/ovpn-net-next:
ovpn: use monotonic clock for peer keepalive timeouts
ovpn: fix use after free in unlock_ovpn()
selftests/net: ovpn: fix getaddrinfo memory leak in ovpn_parse_remote()
ovpn: hold peer before scheduling keepalive work
ovpn: fix peer refcount leak in TCP error paths
ovpn: avoid putting unrelated P2P peer on socket release
====================
Link: https://patch.msgid.link/20260720144131.3657121-1-antonio@openvpn.net
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Derive the hardware QoS channel from opt->parent instead of opt->handle
in airoha_tc_setup_qdisc_ets(). The ETS qdisc handle is either
user-specified or auto-allocated by qdisc_alloc_handle() and bears no
relation to the HTB leaf classid that identifies the hardware channel.
HTB derives the channel from TC_H_MIN(opt->classid), and ETS is always
attached as a child of an HTB leaf, so its opt->parent matches that
classid. Using opt->handle instead can cause two ETS qdiscs on different
HTB leaves to collide on the same hardware channel, corrupting scheduler
configuration and stats.
Fixes: 20bf7d07c9 ("net: airoha: Add sched ETS offload support")
Reviewed-by: Simon Horman <horms@kernel.org>
Signed-off-by: Lorenzo Bianconi <lorenzo@kernel.org>
Link: https://patch.msgid.link/20260720-airoha-ets-handle-fix-v2-1-6f7129ddc06f@kernel.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
nft_payload_offload_mask() builds the offload match mask for a payload
expression that covers only part of a header field. For a partial IPv6
address match (field_len = 16, priv_len = 1) that shift is 1 << 120, which
is undefined on the 32-bit int operand. It also trims only one word, so
the remaining words stay 0xffffffff (and when priv_len is a multiple of 4
the trim is skipped entirely), leaving the mask covering more bytes than
the rule matches.
UBSAN: shift-out-of-bounds in net/netfilter/nft_payload.c:278:20
shift exponent 120 is too large for 32-bit type 'int'
...
The match is byte-granular and struct nft_data is zero-initialised, so the
correct mask is simply the first priv_len bytes set to 0xff. Set those
bytes directly and drop the word/shift trimming; this removes the undefined
shift and no longer over-masks the trailing bytes.
Fixes: a5d45bc0dc ("netfilter: nftables_offload: build mask based from the matching bytes")
Reported-by: AutonomousCodeSecurity@microsoft.com
Signed-off-by: Xiang Mei (Microsoft) <xmei5@asu.edu>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
mctp_device_init() handles errors from rtnl_af_register() and
rtnl_register_many(), but ignores the return value of
register_netdevice_notifier(). If notifier registration fails, init can
still return success while the module is only partially initialized.
Check the notifier registration error and fail module init early.
Fixes: 583be982d9 ("mctp: Add device handling and netlink interface")
Signed-off-by: Minhong He <heminhong@kylinos.cn>
Link: https://patch.msgid.link/20260720072518.112614-1-heminhong@kylinos.cn
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
The NETC timer does not support function level reset, so TMR_OFF_L/H
registers are not cleared by pcie_flr(). If TMR_OFF was set to a
non-zero value in a previous binding, it will persist across driver
rebind and cause inaccurate PTP time.
There is also a hardware issue: after a warm reset or soft reset,
TMR_OFF_L/H registers appear to be cleared to zero, but the timer clock
domain internally retains the stale value. When the timer is re-enabled,
TMR_CUR_TIME continues to track the old offset until TMR_OFF is written
explicitly. This can cause incorrect PTP timestamps and even PTP clock
synchronization failures.
Per the recommendation from the IP team, explicitly write 0 to TMR_OFF
in netc_timer_init() to flush the internally cached value and ensure
TMR_CUR_TIME follows the freshly initialized counter.
Fixes: 87a201d599 ("ptp: netc: add NETC V4 Timer PTP driver support")
Signed-off-by: Clark Wang <xiaoning.wang@nxp.com>
Signed-off-by: Wei Fang <wei.fang@nxp.com>
Reviewed-by: Vadim Fedorenko <vadim.fedorenko@linux.dev>
Reviewed-by: Breno Leitao <leitao@debian.org>
Link: https://patch.msgid.link/20260720012508.23227-1-wei.fang@oss.nxp.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
rds_tcp_exit_net() frees the per-netns RDS TCP listen socket via
rds_tcp_kill_sock() before unregistering the per-netns sysctl table. Since
rds_tcp_skbuf_handler() derives the netns from
rtn->rds_tcp_listen_sock->sk, a concurrent sysctl write can race with
netns teardown and dereference the freed socket/sk.
KASAN reports the race as:
BUG: KASAN: slab-use-after-free in rds_tcp_skbuf_handler+0x2aa/0x2e0
rds_tcp_skbuf_handler net/rds/tcp.c:721
proc_sys_call_handler fs/proc/proc_sysctl.c
vfs_write fs/read_write.c
__x64_sys_pwrite64 fs/read_write.c
Fix this by unregistering the RDS TCP sysctl table before calling
rds_tcp_kill_sock(). unregister_net_sysctl_table() prevents new sysctl
handlers from starting and waits for in-flight handlers to finish, so
the listen socket can then be released safely. The fix was tested
against the linked reproducer.
Fixes: 7f5611cbc4 ("rds: sysctl: rds_tcp_{rcv,snd}buf: avoid using current->nsproxy")
Reported-by: AutonomousCodeSecurity@microsoft.com
Link: https://lore.kernel.org/all/20260719203718.9680-1-blbllhy@gmail.com
Reviewed-by: Allison Henderson <achender@kernel.org>
Signed-off-by: Cen Zhang (Microsoft) <blbllhy@gmail.com>
Link: https://patch.msgid.link/20260719210357.10179-1-blbllhy@gmail.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Since the call to __ip6_del_rt_siblings has been converted under
rcu read lock and it only has one call point
we should no longer block or yield.
Our stack trace from the syzbot reproducer looks as follows:
__ip6_del_rt_siblings
rtnl_notify (Here we pass gfp_any() -> GFP_KERNEL)
nlmsg_notify
nlmsg_multicast
nlmsg_multicast_filtered
netlink_broadcast_filtered (GFP_KERNEL passed from earlier)
netlink_broadcast_filtered can yield if GFP_KERNEL
is passed, which we do not want to happen.
Fix this by changing the allocation flag of rtnl_notify.
Also change the flag passed to nlmsg_new. Even though it
is not related to the syzbot generated bug it still falls
under the same requirements.
Reported-by: syzbot+84d4a405ed798b40c96d@syzkaller.appspotmail.com
Closes: https://syzkaller.appspot.com/bug?extid=84d4a405ed798b40c96d
Fixes: bd11ff421d ("ipv6: Get rid of RTNL for SIOCDELRT and RTM_DELROUTE.")
Signed-off-by: Nikola Z. Ivanov <zlatistiv@gmail.com>
Reviewed-by: Ido Schimmel <idosch@nvidia.com>
Link: https://patch.msgid.link/20260719105759.558050-1-zlatistiv@gmail.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
