ip_vs_in_icmp_v6() is missing checksum validation for ICMPv6
packets from clients. In fact, as for TCP/UDP we should
validate the checksum for ICMP packets only when we
mangle the packets on MASQ or on reply for tunnel.
Also, Sashiko points out that handle_response_icmp() being
common for IPv4 and IPv6 is missing the pseudo-header
calculation while validating ICMPv6 messages from real
servers which is a problem if checksum is not validated
by the hardware.
Fix the problems by creating ip_vs_checksum_common_check()
helper and use it for TCP/UDP/ICMP both for IPv4 and IPv6.
Rely on the nf_checksum() for validating the ICMP messages
but use it also for TCP and UDP.
Use correct IP offset for IP_VS_DBG_RL_PKT for TCP/UDP/SCTP.
IPVS packets (TCP/UDP/SCTP/ICMP) do not need checksum
validation on LOCAL_OUT (local clients or local real
servers) and on FORWARD (traffic from servers on LAN).
Do it only on LOCAL_IN, in case nf_checksum() is not
called on PRE_ROUTING.
Also, ip_vs_checksum_complete() can be marked static.
The XT_HASHLIMIT_RATE_MATCH flag mode changes the semantics of the
dsthash_ent structure which represents an entry in the hashtable. There
is a union area which uses a different layout to express the rate match
mode.
Update .checkentry path to validate the XT_HASHLIMIT_RATE_MATCH mode
flag is requested by two or more different rules that refer to the same
hashtable. Otherwise, uninitialized access to the burst field in the
union is possible.
Reject the use of the XT_HASHLIMIT_RATE_MATCH mode flag if set on by
revision less than 3 too.
====================
Intel Wired LAN Driver Updates 2026-07-17 (ice, idpf) [part]
For ice:
Vincent Chen fixes issue preventing VF creation when switchdev is not
enabled in the configuration.
Marcin corrects iteration value for profile association that was
truncating profiles.
Karol bypasses, unnecessary, waiting on sideband queue PTP writes which
can cause failures with phc_ctl program.
Sergey adds READ_ONCE() to access of PHC time to prevent torn read on
32-bit systems.
Paul adds a check for uninitialized PTP state before attempting to
rebuild it and restricts check of TxTime to be for PF VSI only.
Alex adds bounds check on PTYPE to prevent possible out-of-bounds write.
For idpf:
Emil defers setting of adapter max_vports value to prevent inadvertent
use if interim allocation errors are encountered.
====================
Emil Tantilov [Fri, 17 Jul 2026 18:53:34 +0000 (11:53 -0700)]
idpf: fix max_vport related crash on allocation error during init
Set adapter->max_vports only after successful allocation of vports, netdevs
and vport_config buffers. This fixes possible crashes on reset or rmmod,
following failed allocation on init
ice: reject out-of-range ptype in ice_parser_profile_init
set_bit(rslt->ptype, prof->ptypes) operates on a DECLARE_BITMAP of
ICE_FLOW_PTYPE_MAX (1024) bits. Nothing prevents a malicious VF from
providing ptype >= 1024 through VIRTCHNL, resulting in a write past
the end of the bitmap and a kernel page fault.
Reproduced with a custom kernel module injecting a crafted
VIRTCHNL_OP_ADD_RSS_CFG on E810-C QSFP (8086:1592),
FW 4.91 0x800214af 1.3909.0, ICE COMMS DDP 1.3.53.0,
kernel 7.1.0-rc1.
Paul Greenwalt [Fri, 17 Jul 2026 18:53:32 +0000 (11:53 -0700)]
ice: prevent tstamp ring allocation for non-PF VSI types
The pf->txtime_txqs bitmap tracks which Tx queues have ETF (Earliest
TxTime First) offload enabled. This bitmap is indexed by queue number
and is set by ice_offload_txtime(), which only operates on PF VSI
queues.
However, ice_is_txtime_ena() does not check the VSI type before
consulting the bitmap. When ETF offload is enabled on PF Tx queue 0,
bit 0 is set in pf->txtime_txqs. During a subsequent PCI reset
rebuild, the CTRL VSI's Tx queue 0 is reconfigured and
ice_is_txtime_ena() is called for that ring. Since it only checks
pf->txtime_txqs by queue index without distinguishing VSI type, it
finds bit 0 set and returns true, matching the PF VSI's ETF queue,
not the CTRL VSI's. This causes ice_vsi_cfg_txq() to spuriously
allocate a tstamp_ring for the CTRL VSI ring.
Since CTRL VSI rings have no associated netdev, ice_clean_tx_ring()
takes an early return at the !netdev check before reaching
ice_free_tx_tstamp_ring(), leaking the allocation. Each PCI reset
leaks one 64-byte tstamp_ring.
Fix this by restricting ice_is_txtime_ena() to return true only for
PF VSI rings, since txtime_txqs is only meaningful for PF VSI queues.
Fixes: ccde82e90946 ("ice: add E830 Earliest TxTime First Offload support") Signed-off-by: Paul Greenwalt <paul.greenwalt@intel.com> Reviewed-by: Przemek Kitszel <przemyslaw.kitszel@intel.com> Reviewed-by: Aleksandr Loktionov <aleksandr.loktionov@intel.com> Tested-by: Rinitha S <sx.rinitha@intel.com> (A Contingent worker at Intel) Signed-off-by: Tony Nguyen <anthony.l.nguyen@intel.com> Link: https://patch.msgid.link/20260717185340.3595286-11-anthony.l.nguyen@intel.com Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Paul Greenwalt [Fri, 17 Jul 2026 18:53:31 +0000 (11:53 -0700)]
ice: fix PTP Call Trace during PTP release
If a PF reset occurs when the PTP state is ICE_PTP_UNINIT, then
ice_ptp_rebuild() will update the state to ICE_PTP_ERROR. This will
result in the following PTP release call trace during driver unload:
kernel BUG at lib/list_debug.c:52!
ice_ptp_release+0x332/0x3c0 [ice]
ice_deinit_features.part.0+0x10e/0x120 [ice]
ice_remove+0x100/0x220 [ice]
This was observed when passing PF1 through to a VM. ice_ptp_init()
fails because ctrl_pf is NULL and sets the state to ICE_PTP_UNINIT.
Fix by detecting the ICE_PTP_UNINIT state in ice_ptp_rebuild() and
returning without error, preventing the invalid state transition to
ICE_PTP_ERROR. The only valid path to ICE_PTP_ERROR is from
ICE_PTP_RESETTING after a failed rebuild.
Fixes: 8293e4cb2ff5 ("ice: introduce PTP state machine") Cc: stable@vger.kernel.org Signed-off-by: Paul Greenwalt <paul.greenwalt@intel.com> Signed-off-by: Aleksandr Loktionov <aleksandr.loktionov@intel.com> Reviewed-by: Simon Horman <horms@kernel.org> Tested-by: Rinitha S <sx.rinitha@intel.com> (A Contingent worker at Intel) Signed-off-by: Tony Nguyen <anthony.l.nguyen@intel.com> Link: https://patch.msgid.link/20260717185340.3595286-10-anthony.l.nguyen@intel.com Signed-off-by: Jakub Kicinski <kuba@kernel.org>
ptp.cached_phc_time is a 64-bit value updated by a periodic work item
on one CPU and read locklessly on another. On 32-bit or non-atomic
architectures this can result in a torn read. Use READ_ONCE() to
enforce a single atomic load.
Fixes: 77a781155a65 ("ice: enable receive hardware timestamping") Cc: stable@vger.kernel.org Signed-off-by: Sergey Temerkhanov <sergey.temerkhanov@intel.com> Signed-off-by: Aleksandr Loktionov <aleksandr.loktionov@intel.com> Reviewed-by: Simon Horman <horms@kernel.org> Tested-by: Rinitha S <sx.rinitha@intel.com> (A Contingent worker at Intel) Signed-off-by: Tony Nguyen <anthony.l.nguyen@intel.com> Link: https://patch.msgid.link/20260717185340.3595286-9-anthony.l.nguyen@intel.com Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Marcin Szycik [Fri, 17 Jul 2026 18:53:28 +0000 (11:53 -0700)]
ice: fix LAG recipe to profile association
ice_init_lag() associates recipes to profiles, assuming that Link
Aggregation-related profiles will always have profile ID lower than 70
(ICE_PROFID_IPV6_GTPU_IPV6_TCP_INNER). This value seems arbitrary and
might not always be valid for some versions of DDP package, i.e. LAG
profiles may have profile ID greater than 70. This would lead to
misconfigured switch and LAG not working properly.
Fix it by checking up to maximum profile ID.
Fixes: 1e0f9881ef79 ("ice: Flesh out implementation of support for SRIOV on bonded interface") Signed-off-by: Marcin Szycik <marcin.szycik@linux.intel.com> Reviewed-by: Michal Swiatkowski <michal.swiatkowski@linux.intel.com> Reviewed-by: Aleksandr Loktionov <aleksandr.loktionov@intel.com> Reviewed-by: Dave Ertman <david.m.ertman@intel.com> Reviewed-by: Simon Horman <horms@kernel.org> Tested-by: Rinitha S <sx.rinitha@intel.com> (A Contingent worker at Intel) Signed-off-by: Tony Nguyen <anthony.l.nguyen@intel.com> Link: https://patch.msgid.link/20260717185340.3595286-7-anthony.l.nguyen@intel.com Signed-off-by: Jakub Kicinski <kuba@kernel.org>
skb_checksum_help() can fail. Pass its return value back to the caller.
Commonize this software path in goto.
Instead of just returning error try calculating software checksum first.
There is a check for TSO in checksum_sw_fb.
Reviewed-by: Aleksandr Loktionov <aleksandr.loktionov@intel.com> Signed-off-by: Michal Swiatkowski <michal.swiatkowski@linux.intel.com> Tested-by: Rinitha S <sx.rinitha@intel.com> (A Contingent worker at Intel) Signed-off-by: Tony Nguyen <anthony.l.nguyen@intel.com> Link: https://patch.msgid.link/20260717185340.3595286-4-anthony.l.nguyen@intel.com Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Vincent Chen [Fri, 17 Jul 2026 18:53:23 +0000 (11:53 -0700)]
ice: allow creating VFs when !CONFIG_ICE_SWITCHDEV
Currently ice_eswitch_attach_vf() is called unconditionally in
ice_start_vfs(), which causes VF creation to fail when CONFIG_ICE_SWITCHDEV
is not defined.
Fix this by adding switchdev mode checks at the call sites before
calling ice_eswitch_attach_vf(), consistent with how
ice_eswitch_attach_sf() is already handled in ice_devlink_port_new().
This is similar to commit aacca7a83b97 ("ice: allow creating VFs for
!CONFIG_NET_SWITCHDEV") which fixed the same issue for the previous
ice_eswitch_configure() API.
Fixes: 415db8399d06 ("ice: make representor code generic") Signed-off-by: Vincent Chen <vincent.chen@sifive.com> Reviewed-by: Aleksandr Loktionov <aleksandr.loktionov@intel.com> Tested-by: Rafal Romanowski <rafal.romanowski@intel.com> Signed-off-by: Tony Nguyen <anthony.l.nguyen@intel.com> Link: https://patch.msgid.link/20260717185340.3595286-2-anthony.l.nguyen@intel.com Signed-off-by: Jakub Kicinski <kuba@kernel.org>
netfilter: nf_tables: make nft_object rhltable per table
The nft_object rhltable is global, this allows for accessing objects
that are being dismangled from lookup path by other existing netns.
Given the nft_obj_destroy() releases the object inmediately, this might
lead to use-after-free of these objects that are being released.
Make the existing rhltable per table to address this issue to deal with
with the nft_rcv_nl_event() path too.
Update nft_obj_lookup() to take the table as non-const, otherwise,
compiler complains when passing the objname_ht to rhltable_lookup().
Fixes: 4d44175aa5bb ("netfilter: nf_tables: handle nft_object lookups via rhltable") Suggested-by: Florian Westphal <fw@strlen.de> Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
ipvs: adjust double hashing when fwd method changes
Synced conns can be created with one forwarding method
and later updated with different one after the dest
server is configured. This needs adjusting the hashing
for node hn1 because only MASQ supports double hashing.
Modify conn_tab_lock() to support seeking for hash node
hn0 together with adding for hn1. By this way we can
safely modify the forwarding method and hn1.hash_key
under bucket lock for the first node hn0. The forwarding
method is also protected by cp->lock as it is part of
cp->flags.
Fix the usage of stale idx/idx2 values in conn_tab_lock
after jumping to the retry label. Instead, use idx/idx2
values just to order the locking for the old/new tables.
Zhiling Zou [Mon, 13 Jul 2026 11:52:32 +0000 (19:52 +0800)]
ipvs: do not propagate one-packet flag to synced conns
Synced connections can be created before their destination exists. When
the destination is later added, ip_vs_bind_dest() copies connection flags
from the destination into cp->flags.
IP_VS_CONN_F_ONE_PACKET connections are not synced. If a synced
connection inherits IP_VS_CONN_F_ONE_PACKET while it is already hashed,
expiry can treat it as a one-packet connection and skip unlinking the
existing conn_tab node, leaving stale hash nodes pointing at a freed
struct ip_vs_conn.
Drop IP_VS_CONN_F_ONE_PACKET from destination flags when binding synced
connections.
Fixes: 26ec037f9841 ("IPVS: one-packet scheduling") Cc: stable@vger.kernel.org Reported-by: Yuan Tan <yuantan098@gmail.com> Reported-by: Yifan Wu <yifanwucs@gmail.com> Reported-by: Juefei Pu <tomapufckgml@gmail.com> Reported-by: Xin Liu <bird@lzu.edu.cn> Suggested-by: Julian Anastasov <ja@ssi.bg> Signed-off-by: Zhiling Zou <roxy520tt@gmail.com> Signed-off-by: Ren Wei <n05ec@lzu.edu.cn> Acked-by: Julian Anastasov <ja@ssi.bg> Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
David Lee [Mon, 13 Jul 2026 09:59:15 +0000 (09:59 +0000)]
netfilter: ipset: do not update comments from kernel-side hash adds
mtype_resize() copies comment pointers with memcpy(), not the comment
objects themselves. During the window after an entry has been copied but
before the table swap and backlog replay, the old table is still
published for packet-side updates while the replacement-table entry
already holds the same ip_set_comment_rcu pointer.
If xt_SET --add-set ... --exist hits that old entry in this window,
mtype_add() calls ip_set_init_comment() even though packet-side adds
carry no comment payload. That call frees the shared comment through the
old entry, so the replacement-table entry now holds a stale pointer.
When the queued add is replayed on the new table, mtype_add() calls
ip_set_init_comment() again and strlen() dereferences the stale pointer.
Fix this in mtype_add() by skipping ip_set_init_comment() when
ext->target marks a packet-side add. Userspace adds still update
comments, while packet-side adds can no longer free comment storage
shared with a resize copy.
Fixes: f66ee0410b1c ("netfilter: ipset: Fix "INFO: rcu detected stall in hash_xxx" reports") Cc: stable@vger.kernel.org Signed-off-by: David Lee <david.lee@trailofbits.com> Assisted-by: Codex:gpt-5.5 Acked-by: Jozsef Kadlecsik <kadlec@netfilter.org> Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
Li RongQing [Fri, 17 Jul 2026 14:32:30 +0000 (22:32 +0800)]
net: ipv6: fix dif and sdif mismatch in raw6_icmp_error
In raw6_icmp_error(), raw_v6_match() is called with inet6_iif(skb) passed
to both the 'dif' and 'sdif' arguments. This is a copy-paste or typo error,
as the last argument should represent the secondary interface index (sdif).
This mismatch breaks ICMPv6 error handling for IPv6 raw sockets in VRF
(Virtual Routing and Forwarding) environments. When a raw socket is bound
to a VRF master device, raw_v6_match() fails to find a match because it is
not given the correct sdif value, causing the socket to miss relevant
ICMPv6 error notifications.
Fix this by properly passing inet6_sdif(skb) as the last argument to
raw_v6_match().
Fixes: 5108ab4bf446fa ("net: ipv6: add second dif to raw socket lookups") Signed-off-by: Li RongQing <lirongqing@baidu.com> Reviewed-by: Joe Damato <joe@dama.to> Link: https://patch.msgid.link/20260717143230.1836-1-lirongqing@baidu.com Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Keep div_exp = 0 and derive exp and mantissa from half of the requested
rate. Rates below 2 Mbps are floored to the smallest encodable step
(exp = 0, mantissa = 0).
====================
tcp: validate RST sequence in SYN-RECEIVED
The SYN-RECEIVED request-socket path accepts any in-window RST and
removes the request, even when SEG.SEQ does not exactly match RCV.NXT.
RFC 9293 requires a challenge ACK for a non-exact in-window RST.
Patch 1 applies the RFC 5961 sequence check to request sockets and shares
the per-netns challenge ACK quota with the established-socket path.
Patch 2 adds a compact packetdrill regression test for exact, non-exact,
RST|ACK, and out-of-window cases.
The implementation was tested with a separate raw-socket A/B harness on
IPv4 and IPv6: the unpatched kernel passed 4/12 cases and the patched
kernel passed 12/12. The packetdrill test fails on the unpatched kernel
and passes on the patched kernel for IPv4, IPv6, and IPv4-mapped IPv6
under QEMU/TCG.
====================
Yuxiang Yang [Fri, 17 Jul 2026 08:14:43 +0000 (08:14 +0000)]
selftests/net: packetdrill: cover RST validation in SYN-RECEIVED
Add packetdrill coverage for the RFC 9293 reset checks on request
sockets in SYN-RECEIVED. Verify that an exact RST removes the request,
a non-exact in-window RST sends a challenge ACK without removing it,
and an out-of-window RST is silently discarded.
Also cover an RST|ACK with an unacceptable ACK number to ensure RST
sequence validation runs before ACK-field validation.
Yuxiang Yang [Fri, 17 Jul 2026 08:14:42 +0000 (08:14 +0000)]
tcp: challenge ACK for non-exact RST in SYN-RECEIVED
The SYN-RECEIVED request-socket path in tcp_check_req() accepts an
in-window RST without requiring SEG.SEQ to exactly match RCV.NXT. A
non-exact RST therefore removes the request instead of eliciting a
challenge ACK.
RFC 9293 section 3.10.7.4 applies the RFC 5961 reset check in
SYN-RECEIVED: an exact RST resets the connection, while a non-exact
in-window RST must trigger a challenge ACK and be dropped.
Apply that check before the ACK-field validation, following the RFC
sequence-number, RST, then ACK processing order. Factor the per-netns
challenge ACK quota out of tcp_send_challenge_ack() so request sockets
can share it. Use the request socket's send_ack() callback and its own
out-of-window ACK timestamp to send and rate-limit the response.
Reported-by: Yuxiang Yang <yangyx22@mails.tsinghua.edu.cn> Reported-by: Yizhou Zhao <zhaoyz24@mails.tsinghua.edu.cn> Reported-by: Ao Wang <wangao@seu.edu.cn> Reported-by: Xuewei Feng <fengxw06@126.com> Reported-by: Qi Li <qli01@tsinghua.edu.cn> Reported-by: Ke Xu <xuke@tsinghua.edu.cn> Fixes: 282f23c6ee34 ("tcp: implement RFC 5961 3.2") Cc: stable@vger.kernel.org Signed-off-by: Yuxiang Yang <yangyx22@mails.tsinghua.edu.cn> Reviewed-by: Eric Dumazet <edumazet@google.com> Link: https://patch.msgid.link/20260717081443.809393-2-yangyx22@mails.tsinghua.edu.cn Signed-off-by: Jakub Kicinski <kuba@kernel.org>
usb: atm: ueagle-atm: reject descriptors that confuse probe and disconnect
uea_probe() distinguishes a pre-firmware device from a post-firmware one
using the USB id (UEA_IS_PREFIRM()), and stores a different object as the
interface data in each case: a 'struct completion' for a pre-firmware
device (to be waited on in .disconnect()), or a 'struct usbatm_data' for a
post-firmware one.
uea_disconnect() instead tells the two apart by the number of interfaces
of the active configuration (a pre-firmware device exposes a single
interface, ADI930 has 2 and eagle has 3), and casts the interface data
accordingly.
Because the two handlers use different criteria, a crafted device that
advertises a pre-firmware id together with a multi-interface descriptor
(or a post-firmware id with a single interface) makes them disagree: the
small 'struct completion' stored by uea_probe() is then passed to
usbatm_usb_disconnect(), which casts it to 'struct usbatm_data' and takes
instance->serialize, reading past the end of the allocation:
BUG: KASAN: slab-out-of-bounds in __mutex_lock+0x152a/0x1b80
Read of size 8 at addr ffff8880470e2c60 by task kworker/1:2/982
...
__mutex_lock+0x152a/0x1b80
usbatm_usb_disconnect+0x70/0x820
uea_disconnect+0x133/0x2c0
usb_unbind_interface+0x1dd/0x9e0
...
which belongs to the cache kmalloc-96 of size 96
The buggy address is located 0 bytes to the right of
allocated 96-byte region [ffff8880470e2c00, ffff8880470e2c60)
Reject such inconsistent descriptors in uea_probe() so that both handlers
always make the same pre/post-firmware decision.
Reported-by: syzbot+e62a973f8322b3bbe3ac@syzkaller.appspotmail.com Closes: https://syzkaller.appspot.com/bug?extid=e62a973f8322b3bbe3ac Fixes: e2674dfbed8a ("usb: atm: ueagle-atm: wait for pre-firmware load in .disconnect()") Signed-off-by: Diego Fernando Mancera Gomez <diegomancera.dev@gmail.com> Acked-by: Stanislaw Gruszka <stf_xl@wp.pl> Link: https://patch.msgid.link/20260717080704.1264-1-diegomancera.dev@gmail.com Signed-off-by: Jakub Kicinski <kuba@kernel.org>
====================
net/mlx5e: Report zero bandwidth for non-ETS traffic
The IEEE 802.1Qaz standard restricts bandwidth allocation percentages
to Enhanced Transmission Selection (ETS) traffic classes; STRICT,
VENDOR, and CB Shaper TSA types carry no bandwidth semantics. Two
problems exist in the mlx5e DCBNL ETS implementation: the get path
reports 100% bandwidth for all TCs regardless of TSA type due to a
hardware limitation, introduced by commit 820c2c5e773d ("net/mlx5e:
Read ETS settings directly from firmware"), and the set path does
not reject the unsupported CB Shaper TSA, introduced by commit 08fb1dacdd76 ("net/mlx5e: Support DCBNL IEEE ETS").
This series by Alexei Lazar fixes the get path to report zero
bandwidth for non-ETS traffic classes, and rejects CB Shaper TSA
configurations that the driver does not support.
====================
net/mlx5e: Report zero bandwidth for non-ETS traffic classes
The IEEE 802.1Qaz standard defines that bandwidth allocation percentages
only apply to Enhanced Transmission Selection (ETS) traffic classes.
For STRICT and VENDOR transmission selection algorithms, bandwidth
percentage values are not applicable.
Currently for non-ETS 100 bandwidth is being reported for all traffic
classes in the get operation due to hardware limitation, regardless of
their TSA type.
Fix this by reporting 0 for non-ETS traffic classes.
assoc_array: trim the final shortcut word using the current chunk end
assoc_array_walk() masks off the bits past shortcut->skip_to_level in the
word that contains skip_to_level, gated on
round_up(sc_level, ASSOC_ARRAY_KEY_CHUNK_SIZE) > skip_to_level.
That guard is wrong in two opposite ways:
- When sc_level is word-aligned (every word after the first) round_up()
is a no-op, so the guard is sc_level > skip_to_level and never fires for
the word that holds skip_to_level. A shortcut that spans more than one
word and ends in the middle of its last word leaves that word untrimmed,
and its stale high bits leak into the dissimilarity word and can steer
the walk down the wrong descendant.
- When sc_level is unaligned (the first word) and skip_to_level sits on
the next chunk boundary, sc_level + CHUNK would exceed skip_to_level and
fire the trim with shift = skip_to_level & CHUNK_MASK == 0, which clears
the whole dissimilarity word and makes a differing shortcut compare
equal.
Use the end of the chunk that contains sc_level instead:
For an aligned sc_level whose word holds skip_to_level this now fires (the
first bug); for an unaligned sc_level with skip_to_level on the following
boundary it does not, so shift is never 0 when the branch runs and the trim
never clears the whole word.
Fixes: 3cb989501c26 ("Add a generic associative array implementation.") Assisted-by: Claude:claude-opus-4-8 Signed-off-by: Michael Bommarito <michael.bommarito@gmail.com> Reviewed-by: Jarkko Sakkinen <jarkko@kernel.org> Tested-by: Jarkko Sakkinen <jarkko@kernel.org> Link: https://lore.kernel.org/r/20260719161505.2423935-4-michael.bommarito@gmail.com Signed-off-by: Jarkko Sakkinen <jarkko@kernel.org>
keys: make keyring key-chunk byte order agree with keyring_diff_objects()
keyring_get_key_chunk() loads description bytes into the index chunk low
address first, while keyring_diff_objects() numbers the first differing
bit from the low end and folds the absolute byte index into the level
without removing the inline-prefix offset the level already carries.
The two disagree on byte order and bit position, so the array can be
told two keys first differ at a bit that does not differ in the chunk
the walker uses, letting crafted descriptions collide into one node.
Load the chunk in the order keyring_diff_objects() assumes and drop the
inline-prefix length when folding the byte index into the level. This
only changes the in-memory ordering used to place keys within a keyring;
add, search and read of non-colliding keys are unaffected.
Fixes: f771fde82051 ("keys: Simplify key description management") Assisted-by: Claude:claude-opus-4-8 Signed-off-by: Michael Bommarito <michael.bommarito@gmail.com> Reviewed-by: Jarkko Sakkinen <jarkko@kernel.org> Tested-by: Jarkko Sakkinen <jarkko@kernel.org> Link: https://lore.kernel.org/r/20260719161505.2423935-3-michael.bommarito@gmail.com Signed-off-by: Jarkko Sakkinen <jarkko@kernel.org>
keys: fix out-of-bounds read in keyring_get_key_chunk()
For description-level chunks keyring_get_key_chunk() advances the read
pointer by level * sizeof(long) past the inline prefix but only
bounds-checks the prefix, so a long enough key description is read past
its kmemdup(desc, desc_len + 1) allocation. Compute the full byte
offset and bounds-check the description against it before reading.
The walk only reaches a description-level chunk when two keys collide
through the hash, x, type and domain_tag chunks, so this is reached from
an unprivileged add_key(2) with a crafted pair of same-type keys whose
index hashes collide; KASAN reports a slab-out-of-bounds read.
Fixes: f771fde82051 ("keys: Simplify key description management") Assisted-by: Claude:claude-opus-4-8 Signed-off-by: Michael Bommarito <michael.bommarito@gmail.com> Reviewed-by: Jarkko Sakkinen <jarkko@kernel.org> Tested-by: Jarkko Sakkinen <jarkko@kernel.org> Link: https://lore.kernel.org/r/20260719161505.2423935-2-michael.bommarito@gmail.com Signed-off-by: Jarkko Sakkinen <jarkko@kernel.org>
KEYS: trusted: dcp: fix key_len validation and calc_blob_len() return type
Two correctness and type-hygiene issues exist in the DCP trusted keys
implementation.
First, trusted_dcp_unseal() reads p->key_len from a user-supplied blob
without checking if it exceeds MAX_KEY_SIZE. If a crafted blob provides a
payload_len larger than 128, the subsequent do_aead_crypto() call writes
past the end of the p->key array into the adjacent p->blob buffer within
the same struct trusted_key_payload -- the caller's own input, not
unrelated kernel memory. While not exploitable, this violates strict array
bounds and triggers static analyzers. Fix this by adding a validation
check against MIN_KEY_SIZE and MAX_KEY_SIZE immediately after reading the
length, matching the checks already done in trusted_core.c.
Second, calc_blob_len() calculates a sum in size_t that truncates to
unsigned int on 64-bit platforms. Because the DCP hardware is only present
on 32-bit i.MX SoC platforms, size_t and unsigned int are functionally
equivalent in production, making this truncation harmless in practice.
Nevertheless, updating the return type to size_t (and subsequently updating
'blen' in the seal/unseal paths) resolves type-narrowing warnings and
improves overall code hygiene.
Fixes: 2e8a0f40a39c ("KEYS: trusted: Introduce NXP DCP-backed trusted keys") Signed-off-by: Fabrice Derepas <fabrice.derepas@canonical.com> Reviewed-by: David Gstir <david@sigma-star.at> Reviewed-by: Richard Weinberger <richard@nod.at> Reviewed-by: Jarkko Sakkinen <jarkko@kernel.org> Tested-by: Jarkko Sakkinen <jarkko@kernel.org> Link: https://lore.kernel.org/r/20260719163939.3624767-1-fabrice.derepas@canonical.com Signed-off-by: Jarkko Sakkinen <jarkko@kernel.org>
Commit 2a22b7ae2fa3 ("net: pcs: xpcs: adapt Wangxun NICs for SGMII mode")
added a path in xpcs_get_state_c37_sgmii() that reads speed/duplex from
BMCR after AN completes. However, BMCR does not reflect the negotiated
result on the hardware where this has been tested:
- On RK3568 (MAC side SGMII), BMCR returns a fixed hardware reset value
- Wangxun engineer Jiawen Wu confirmed that on their side, "BMCR looks
like it only wants to be return as 0" [0]
The correct information is available in CL37_ANSGM_STS, which contains
the actual link status and negotiated speed/duplex.
This bug was previously masked by phylink core, which overrides the PCS
link state with the PHY state when a PHY is present:
/* If we have a phy, the "up" state is the union of both the
* PHY and the MAC
*/
if (phy)
link_state.link &= pl->phy_state.link;
Thus, when the link is down, the PHY's link_down state is applied on top
of whatever the PCS reports, hiding the broken PCS state reading path.
Modify xpcs_get_state_c37_sgmii() to:
1. Read link state from CL37_ANSGM_STS
2. If link is up, report speed/duplex from CL37_ANSGM_STS
3. Remove the broken BMCR reading path entirely
Also properly set state->an_complete to reflect the AN completion status,
and clear CL37_ANCMPLT_INTR when link is down to avoid stale state.
net/mlx5: E-Switch, fix zero num_dest in prio_tag egress vlan rule
esw_egress_acl_vlan_create() hardcodes num_dest=0 in its
mlx5_add_flow_rules() call. When invoked from the non-bond path
fwd_dest is NULL and num_dest=0 is correct. When invoked from
esw_acl_egress_ofld_rules_create() during a bond event, fwd_dest is
non-NULL and flow_act.action carries MLX5_FLOW_CONTEXT_ACTION_FWD_DEST,
but _mlx5_add_flow_rules() rejects a non-NULL dest pointer paired with
dest_num<=0 and returns -EINVAL. The error propagates as
"configure slave vport egress fwd, err(-22)". The passive vport's egress
ACL table ends up with its flow groups allocated but no FTEs, so
prio-tagged packets are not popped and bond failover is broken on
prio_tag_required devices.
Fix by passing fwd_dest ? 1 : 0 as num_dest to match the actual number
of destinations supplied.
Gal Pressman [Fri, 17 Jul 2026 07:23:38 +0000 (10:23 +0300)]
net/mlx5: Fix MCIA register buffer overflow on 32 dword reads
The MCIA register can return up to 32 dwords (128 bytes) when the device
advertises the mcia_32dwords capability, but struct
mlx5_ifc_mcia_reg_bits only defines dword_0..11, leaving room for just
12 dwords (48 bytes) of data.
mlx5_query_mcia() clamps the read size to mlx5_mcia_max_bytes() and then
memcpy()s that many bytes out of the register, potentially reading past
the end of the 'out' buffer. On kernels built with FORTIFY_SOURCE this
is caught as a buffer overflow while reading the module EEPROM via
ethtool:
Fixes: 271907ee2f29 ("net/mlx5: Query the maximum MCIA register read size from firmware") Signed-off-by: Gal Pressman <gal@nvidia.com> Reviewed-by: Alex Lazar <alazar@nvidia.com> Signed-off-by: Tariq Toukan <tariqt@nvidia.com> Link: https://patch.msgid.link/20260717072338.1240582-1-tariqt@nvidia.com Signed-off-by: Jakub Kicinski <kuba@kernel.org>
====================
vxlan, geneve: require CAP_NET_ADMIN in the device netns for changelink
The recent series "require CAP_NET_ADMIN in the device netns for
changelink" (8165f7ff57d9..27ccb68e7ccc) added rtnl_dev_link_net_capable()
and gated the eight IP tunnel drivers (ip_gre, ipip, ip_vti, ip6_tunnel,
ip6_gre, ip6_vti, sit, xfrm_interface). VXLAN and GENEVE share the exact
same shape but were not covered: both store the underlay netns sticky at
newlink (vxlan->net / geneve->net) and their changelink() operates on that
netns, while the generic RTM_NEWLINK path only checks CAP_NET_ADMIN against
dev_net(dev). Once such a device is created in or moved to another netns,
a caller privileged in dev_net(dev) but not in the underlay netns can
reconfigure the tunnel'"'"'s underlay.
This completes that series for the two UDP tunnel drivers that were left
out. Same helper, same placement (top of changelink, before any attribute
is parsed).
Verified on next-20260714 in QEMU with CONFIG_VXLAN=y + CONFIG_USER_NS=y:
an unprivileged user namespace holding CAP_NET_ADMIN only in a child netns
issues an IFLA_INFO_DATA changelink on a vxlan device whose underlay lives
in init_net. Before: returns 0 (reconfigures the init_net underlay).
After: returns -EPERM.
====================
geneve: require CAP_NET_ADMIN in the device netns for changelink
A tunnel changelink() operates on at most two netns, dev_net(dev) and
the sticky underlay netns geneve->net. They differ once the device is
created in or moved to a netns other than the one the request runs in.
The rtnl changelink path checks CAP_NET_ADMIN only against dev_net(dev),
so a caller privileged there but not in geneve->net can rewrite a geneve
device whose underlay lives in geneve->net.
geneve_changelink() applies the new configuration against geneve->net:
geneve_link_config() and the geneve_quiesce()/geneve_unquiesce() pair
reopen the underlay sockets in that netns (geneve_sock_add() uses
geneve->net), so the same reasoning as the tunnel changelink series
applies here.
Gate geneve_changelink() with rtnl_dev_link_net_capable(), at the top of
the op before any attribute is parsed, matching ipgre_changelink() and
the rest of the "require CAP_NET_ADMIN in the device netns for
changelink" series.
Found by 0sec automated security-research tooling (https://0sec.ai).
Fixes: 5b861f6baa3a ("geneve: add rtnl changelink support") Cc: stable@vger.kernel.org Signed-off-by: Doruk Tan Ozturk <doruk@0sec.ai> Reviewed-by: Fernando Fernandez Mancera <fmancera@suse.de> Link: https://patch.msgid.link/20260716203500.70573-3-doruk@0sec.ai Signed-off-by: Jakub Kicinski <kuba@kernel.org>
vxlan: require CAP_NET_ADMIN in the device netns for changelink
A tunnel changelink() operates on at most two netns, dev_net(dev) and
the sticky underlay netns vxlan->net. They differ once the device is
created in or moved to a netns other than the one the request runs in.
The rtnl changelink path checks CAP_NET_ADMIN only against dev_net(dev),
so a caller privileged there but not in vxlan->net can rewrite a vxlan
device whose underlay lives in vxlan->net.
vxlan_changelink() validates and applies the new configuration against
vxlan->net (vxlan_config_validate(vxlan->net, ...)) and can reopen the
underlay socket in that netns, so the same reasoning as the tunnel
changelink series applies here.
Gate vxlan_changelink() with rtnl_dev_link_net_capable(), at the top of
the op before any attribute is parsed, matching ipgre_changelink() and
the rest of the "require CAP_NET_ADMIN in the device netns for
changelink" series.
Found by 0sec automated security-research tooling (https://0sec.ai).
Fixes: 8bcdc4f3a20b ("vxlan: add changelink support") Cc: stable@vger.kernel.org Signed-off-by: Doruk Tan Ozturk <doruk@0sec.ai> Reviewed-by: Fernando Fernandez Mancera <fmancera@suse.de> Link: https://patch.msgid.link/20260716203500.70573-2-doruk@0sec.ai Signed-off-by: Jakub Kicinski <kuba@kernel.org>
mac802154: llsec: reject frames shorter than the authentication tag
llsec_do_decrypt_auth() computes the associated-data length for the
AEAD request as
assoclen += datalen - authlen;
where datalen is the number of bytes after the MAC header and authlen
(4, 8 or 16) is the length of the authentication tag. Nothing verifies
that the frame actually carries at least authlen payload bytes. A
secured frame whose payload is shorter than the tag makes
datalen - authlen negative; assoclen is then passed to
aead_request_set_ad() as an unsigned value close to 4 GiB, so
crypto_aead_decrypt() walks far off the end of the scatterlist that
only spans the real frame.
The frame is fully attacker-controlled and reaches this path from any
IEEE 802.15.4 peer in radio range. Reject frames whose payload is
shorter than the authentication tag before the subtraction.
Dynamically reproduced on a KASAN kernel as a general-protection-fault
in the AEAD scatterwalk, and the fix confirmed.
Fixes: 4c14a2fb5d14 ("mac802154: add llsec decryption method") Cc: stable@vger.kernel.org Reviewed-by: Simon Horman <horms@kernel.org> Signed-off-by: Doruk Tan Ozturk <doruk@0sec.ai> Link: https://patch.msgid.link/20260716193423.32498-1-doruk@0sec.ai Signed-off-by: Jakub Kicinski <kuba@kernel.org>
power: supply: bq25890: fix the -10 C NTC lookup entry
The TSPCT lookup table is monotonically decreasing except for ADC code
121, where the sequence reads -9.0 C, -1.0 C, -12.0 C. This makes the
reported battery temperature jump upward by eight degrees for one code
and then downward by eleven degrees for the next code.
The entry is a missing zero: use -10.0 C so the sequence remains
monotonic between -9.0 C and -12.0 C.
raw: annotate lockless match fields in raw_v4_match()
raw_v4_match() is a lockless match helper under sk_for_each_rcu(). It
still reads inet->inet_daddr, inet->inet_rcv_saddr and
sk->sk_bound_dev_if with plain loads while bind, connect and
bind-to-device paths can update the same match fields concurrently.
Annotate only those mutable match fields in raw_v4_match(), and do so
at the point of use instead of hoisting the bound-device read before
the earlier short-circuit tests.
Also annotate the raw bind writer and the shared IPv4 datagram connect
writer used by raw sockets, so the address fields updated on bind and
connect match explicit WRITE_ONCE() updates.
This version intentionally leaves the shared disconnect-side IPv4
writers to follow-up cleanup and limits the writer changes here to the
raw bind path and the datagram connect path directly exercised by raw
sockets.
net: qrtr: restrict socket creation to the initial network namespace
QRTR keeps its entire port and node state in module-global variables
that are not partitioned per network namespace: qrtr_local_nid is a
single global node id (always 1) and qrtr_ports is a single global
xarray. qrtr_port_lookup() and qrtr_local_enqueue() operate on that
global state with no network-namespace check, and qrtr_create() places
no restriction on the namespace a socket is created in.
As a result an unprivileged process that creates an AF_QIPCRTR socket
in a separate network namespace, e.g. via
unshare(CLONE_NEWUSER | CLONE_NEWNET), can send QRTR datagrams -
including control-plane messages such as QRTR_TYPE_NEW_SERVER - to QRTR
sockets owned by another namespace, and vice versa. The receiving
socket sees such a message as coming from node id 1, indistinguishable
from a legitimate local client, breaking the isolation that network
namespaces are expected to provide.
QRTR is a transport to global hardware endpoints (the modem and other
remote processors) and has no per-namespace semantics; its in-kernel
name service already creates its socket in init_net only. Confine the
socket family to the initial network namespace, as other
non-namespace-aware socket families do (see llc_ui_create() and the
ieee802154 socket code).
Nicholas Dudar [Thu, 23 Jul 2026 14:27:35 +0000 (22:27 +0800)]
LoongArch: BPF: Zero-extend signed ALU32 div/mod results
ALU32 operations write a 32-bit result and leave the upper 32 bits of
the BPF register zero. The LoongArch JIT sign-extends the result of
signed ALU32 BPF_DIV and BPF_MOD (off=1), so a negative 32-bit quotient
or remainder leaves bits 63:32 set in JITted code while the verifier
and interpreter model those bits as zero.
Keep sign-extension on the operands, which signed divide needs, and
zero-extend the ALU32 result after the divide or modulo instruction,
matching the unsigned ALU32 div/mod paths and every other ALU32
operation in this JIT.
Fixes: 2425c9e002d2 ("LoongArch: BPF: Support signed div instructions") Fixes: 7b6b13d32965 ("LoongArch: BPF: Support signed mod instructions") Assisted-by: Claude:claude-opus-4-8 Acked-by: Tiezhu Yang <yangtiezhu@loongson.cn> Tested-by: Tiezhu Yang <yangtiezhu@loongson.cn> Signed-off-by: Nicholas Dudar <main.kalliope@gmail.com> Signed-off-by: Huacai Chen <chenhuacai@loongson.cn>
When entering KDB via a breakpoint and then performing single-step
debugging, an oops is triggered. Now during single-step debugging,
kdb_local() expects the reason to be KDB_REASON_SSTEP, but it is
actually KDB_REASON_OOPS. In kdb_stub(), when determining the reason,
the ex_vector for single-step should be 0, as already implemented on
other architectures such as arm64 and riscv.
Before the patch:
[112]kdb> ss
Entering kdb (current=0x900020009f520000, pid 10661) on
processor 112 Oops: (null)
due to oops @ 0x90000000005b57a4
George Guo [Thu, 23 Jul 2026 14:27:30 +0000 (22:27 +0800)]
LoongArch: Fix address space mismatch in kexec command line lookup
When searching the loaded segments for the "kexec" command line marker,
the kexec_load(2) path (file_mode == 0) passes the user-space segment
buffer straight to strncmp() through a bogus (char __user *) cast. This
dereferences a user pointer in kernel context, which is wrong and is
flagged by sparse:
Here copy the marker-sized prefix of each segment into a small on-stack
buffer with copy_from_user() before comparing, and skip segments that
fault. The subsequent copy_from_user() that stages the full command line
into the safe area is left unchanged.
Cc: stable@vger.kernel.org Fixes: 4a03b2ac06a5 ("LoongArch: Add kexec support") Reported-by: kernel test robot <lkp@intel.com> Closes: https://lore.kernel.org/oe-kbuild-all/202605051639.aEPioXdD-lkp@intel.com/ Co-developed-by: Kexin Liu <liukexin@kylinos.cn> Signed-off-by: Kexin Liu <liukexin@kylinos.cn> Signed-off-by: George Guo <guodongtai@kylinos.cn> Signed-off-by: Huacai Chen <chenhuacai@loongson.cn>
Rong Bao [Thu, 23 Jul 2026 14:27:29 +0000 (22:27 +0800)]
LoongArch: Retrieve CPU package ID from PPTT when available
Currently, the LoongArch CPU topology initialization code calculates
each core's package ID by dividing its physical ID by loongson_sysconf.
cores_per_package. This relies on the assumption that cores_per_package
counts in the same domain as physical IDs.
On Loongson-3B6000 (XB612B0V_1.2), cores_per_package matches the visible
core count -- 24 in this case. However, the physical IDs range from 0 to
31 in a noncontinuous fashion:
Retrieve the exact package ID from ACPI PPTT when available, in the same
style as retrieving the core ID and thread ID in parse_acpi_topology().
Use this information in loongson_init_secondary() when the PPTT readout
is successful. The original division logic is kept as a fallback.
Meanwhile, since some existing code paths like loongson3_cpufreq expect
a continuous integer sequence of package IDs in [0, MAX_PACKAGES) when
retrieving from cpu_data[], here we also canonicalize the package ID to
be filled in parse_acpi_topology() to meet such an expectation.
Cc: stable@vger.kernel.org Tested-by: Mingcong Bai <jeffbai@aosc.io> Co-developed-by: Xi Ruoyao <xry111@xry111.site> Signed-off-by: Xi Ruoyao <xry111@xry111.site> Signed-off-by: Rong Bao <rong.bao@csmantle.top> Signed-off-by: Huacai Chen <chenhuacai@loongson.cn>
Kanglong Wang [Thu, 23 Jul 2026 14:27:29 +0000 (22:27 +0800)]
LoongArch: Move jump_label_init() before parse_early_param()
When enabling both CONFIG_MEM_ALLOC_PROFILING=y and
CONFIG_MEM_ALLOC_PROFILING_ENABLED_BY_DEFAULT=y, then diabling memory
profiling by adding the boot parameter 'sysctl.vm.mem_profiling=0' will
cause the kernel failed to boot.
After analysis, this is because jump_label_init() must be called before
parse_early_param(), the early param handlers may modify static keys by
static_branch_enable/disable().
Fix this by moving jump_label_init() to before parse_early_param(). The
solution is similar to other architectures.
Cc: <stable@vger.kernel.org> Signed-off-by: Kanglong Wang <wangkanglong@loongson.cn> Signed-off-by: Huacai Chen <chenhuacai@loongson.cn>
LoongArch: Increase TASK_STRUCT_OFFSET up to 2040 for 32BIT
THREAD_INFO_IN_TASK increase the size of task_struct, which casuses a
build error for the 32BIT kernel if RANDSTRUCT is enabled. So increase
TASK_STRUCT_OFFSET as big as possible (2040), but can still be aligned
and be fit in the addi.w instruction.
hinic: remove unused ethtool RSS user configuration buffers
rss_indir_user and rss_hkey_user are allocated and filled in
__set_rss_rxfh() when the user configures RSS via ethtool, but
nothing ever reads them. hinic_get_rxfh() fetches the state from
the device, and the hardware is programmed from the original
indir/key arguments. These buffers only leaked on driver unload.
Drop the unused allocations, memcpys, and struct fields.
Fixes: 4fdc51bb4e92 ("hinic: add support for rss parameters with ethtool") Signed-off-by: Chenguang Zhao <zhaochenguang@kylinos.cn> Reviewed-by: Joe Damato <joe@dama.to> Link: https://patch.msgid.link/20260722025353.328179-1-chenguang.zhao@linux.dev Signed-off-by: Jakub Kicinski <kuba@kernel.org>
====================
Add missing facility check to ptp_s390 driver
This patchset adds a missing facility check and a check that the 'query
physical clock' (PTFF QPT) function is actually available. If it's not
present, no qpt ptp device will be registered. In order to use ptff_query()
in a module, the first patch adds a EXPORT_SYMBOL() to export
ptff_function_mask.
====================
MAINTAINERS: Add myself for stmmac ethernet driver maintainance
The stmmac driver based on Synopsys' dwmac IP is used in a very wide
variety of SoCs and is currently very actively used and contributed to.
It has been orphaned in January 2025 after the previous maintainers
became inactive, but Russell King was providing very valuable reviews
and fixes for the driver at that point.
Now we're seeing more and more activity on the driver, but are lacking
people to test and review contributions to both glue drivers as well as
core stmmac code.
I have access to some variety of stmmac-based platforms such as socfpga
CycloneV, imx8mp, some Allwinner SoCs and stm32mp1xx boards that I can
run regression tests on, and I'm offering to step-up as a maintainer for
driver, for the time being at least.
Let's hope other people will eventually join this effort.
pppoe: reload header pointer after dev_hard_header()
pppoe_sendmsg() saves a pointer to the PPPoE header before calling
dev_hard_header(). Device header callbacks are allowed to reallocate the
skb head, invalidating pointers into it.
This can happen when a send is blocked in copy_from_user() while the first
non-Ethernet port is added to an empty team device. The team's delegated
GRE header callback then expands the skb head. PPPoE subsequently writes
six bytes through the stale pointer into the freed head.
Reload the PPPoE header through the skb's network-header offset after
device header creation. pskb_expand_head() updates that offset when it
relocates the head.
Eric Dumazet [Wed, 22 Jul 2026 10:16:05 +0000 (10:16 +0000)]
ppp: annotate data races in ppp_generic
Several fields in struct ppp can be read or updated concurrently
from multiple CPUs without synchronization, causing data races:
1. ppp->mru is read concurrently in ppp_receive_nonmp_frame() while
being updated via PPPIOCSMRU ioctl. Protect ppp->mru updates in
PPPIOCSMRU with ppp_recv_lock(ppp).
2. PPPIOCGFLAGS reads ppp->flags, ppp->xstate, and ppp->rstate
unlocked. Wrap the read in ppp_lock(ppp) to get a consistent
snapshot.
3. ppp->debug is updated via PPPIOCSDEBUG and read concurrently on
fast paths. Annotate reads with READ_ONCE() and writes with
WRITE_ONCE().
4. ppp->last_xmit and ppp->last_recv are updated on TX/RX data paths
and read via PPPIOCGIDLE32 / PPPIOCGIDLE64 ioctls. Annotate with
WRITE_ONCE() / READ_ONCE() and use max() to handle jiffies
subtraction.
5. ppp->npmode[] is updated via PPPIOCSNPMODE and read on TX/RX
paths. Annotate with WRITE_ONCE() / READ_ONCE().
Eric Dumazet [Wed, 22 Jul 2026 10:42:36 +0000 (10:42 +0000)]
ipv4: icmp: fill flow parameters in icmp_route_lookup decoy lookup
When Linux forwards a packet and needs to generate an ICMP error,
icmp_route_lookup() performs a reverse-path relookup. For non-local
destinations, it performs a decoy lookup to find the expected egress
interface (rt2->dst.dev) before validating the path with ip_route_input().
Currently, the decoy flow structure (fl4_2) only sets .daddr = fl4_dec.saddr,
leaving .saddr, .flowi4_dscp, .flowi4_proto, .flowi4_mark, .flowi4_oif,
.fl4_sport, .fl4_dport, and .flowi4_uid zeroed out.
When policy routing rules (such as ip rule add from $SRC lookup 100, or
dscp/fwmark/ipproto/port rules, or VRF bindings) are configured:
1. The decoy lookup fails to match the policy rule because saddr and other
key flow selectors are missing in fl4_2.
2. It resolves a route using the default table instead, returning an incorrect
egress netdev.
3. Passing the wrong netdev to ip_route_input() causes strict reverse-path
filtering (rp_filter=1) to fail, logging false-positive "martian source"
warnings and causing the relookup to fail.
Fix this by initializing fl4_2 from fl4_dec and:
- Swapping source/destination IP addresses.
- Swapping L4 ports for transport protocols with ports (TCP, UDP, SCTP, DCCP)
so port-based policy routing matches correctly. Non-port protocols (such as
ICMP or GRE) leave the flowi_uli union fields intact to prevent corruption.
- Setting .flowi4_oif = l3mdev_master_ifindex(route_lookup_dev) to ensure
VRF routing tables are respected.
- Setting .flowi4_flags |= FLOWI_FLAG_ANYSRC to allow output route lookups
for non-local source IP addresses.
- Using __ip_route_output_key() instead of ip_route_output_key() for fl4_2
so that raw FIB routing is used without triggering spurious XFRM policy
lookups on the decoy flow (the actual XFRM lookup is performed later using
fl4_dec).
Fixes: 415b3334a21a ("icmp: Fix regression in nexthop resolution during replies.") Reported-by: Muhammad Ziad <muhzi100@gmail.com> Closes: https://lore.kernel.org/netdev/CAOAwikA60AYKdFr_UDLyja3oU4hqyAE7uFZWqum5uRdaQsgRYg@mail.gmail.com/ Signed-off-by: Eric Dumazet <edumazet@google.com> Reviewed-by: David Ahern <dsahern@kernel.org> Link: https://patch.msgid.link/20260722104236.2938082-1-edumazet@google.com Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Shihuang Liu [Wed, 22 Jul 2026 11:39:19 +0000 (19:39 +0800)]
amt: fix use-after-free in AMT delayed works
When an AMT device is removed, pending delayed works can still access
the freed amt_dev structure, which may result in kernel crashes or
memory corruption.
amt_dev_stop() cancels req_wq and discovery_wq with
cancel_delayed_work_sync(), but these works can be scheduled again
from event_wq after the cancellation. This allows delayed works to
access the freed amt_dev structure after the netdev has been released.
Use disable_delayed_work_sync() in amt_dev_stop() to prevent req_wq and
discovery_wq from being queued again and wait for running work items
to complete.
The delayed works are disabled after initialization in
amt_newlink() and enabled only when the device is successfully opened.
This keeps the delayed work lifecycle synchronized with the lifetime
of the AMT device.
Fixes: cbc21dc1cfe9 ("amt: add data plane of amt interface") Cc: stable@vger.kernel.org Signed-off-by: Shihuang Liu <shlomojune6@gmail.com> Reviewed-by: Simon Horman <horms@kernel.org> Reviewed-by: Taehee Yoo <ap420073@gmail.com> Link: https://patch.msgid.link/20260722113919.7723-1-shlomojune6@gmail.com Signed-off-by: Jakub Kicinski <kuba@kernel.org>
A length of zero passes this check, so rxlen is set to 0 and the state
machine advances to STATE_DATA. In mctp_serial_push() STATE_DATA, the
incoming byte is stored and rxpos incremented before the terminator is
tested:
With rxlen == 0 the "rxpos == rxlen" terminator can never fire (rxpos is
already 1 on the first data byte), so subsequent bytes are written past
the end of the fixed 74-byte rxbuf, which is the last member of the
netdev private area. Every following data byte is an attacker-controlled
1-byte out-of-bounds heap write, and the overflow continues until a
frame (0x7e) or escape byte resets the parser -- effectively unbounded.
Reaching this requires CAP_NET_ADMIN to attach the N_MCTP line
discipline and bring the resulting mctpserialN netdev up, after which
the bytes arrive via the tty receive path.
Route a zero-length frame straight to STATE_TRAILER instead of
STATE_DATA. The trailer/framing bytes are still consumed, and the frame
resolves to a zero-length skb that the MCTP core rejects; the parser
never enters STATE_DATA with rxlen == 0, so the out-of-bounds write can
no longer occur.
KASAN, on a frame of 0x7e 0x01 0x00 followed by data bytes (before this
change):
UBSAN: array-index-out-of-bounds in drivers/net/mctp/mctp-serial.c:370
index 74 is out of range for type 'u8 [74]'
BUG: KASAN: slab-out-of-bounds in mctp_serial_tty_receive_buf
Write of size 1 at addr ... by task kworker/u16:0
mctp_serial_tty_receive_buf
tty_ldisc_receive_buf
flush_to_ldisc
Allocated by task 152:
alloc_netdev_mqs
mctp_serial_open
v2: route zero-length frames to STATE_TRAILER instead of STATE_ERR so
the trailer/framing bytes are still consumed (Jeremy Kerr).
Found by 0sec automated security-research tooling (https://0sec.ai). Fixes: a0c2ccd9b5ad ("mctp: Add MCTP-over-serial transport binding") Cc: stable@vger.kernel.org Suggested-by: Jeremy Kerr <jk@codeconstruct.com.au> Assisted-by: 0sec:multi-model Signed-off-by: Doruk Tan Ozturk <doruk@0sec.ai> Reviewed-by: Simon Horman <horms@kernel.org> Link: https://patch.msgid.link/20260715082021.46315-1-doruk@0sec.ai Signed-off-by: Paolo Abeni <pabeni@redhat.com>
octeontx2-vf: set TC flower flag on MCAM entry allocation
When MCAM entries are allocated for a VF netdev via the devlink
mcam_count parameter, only OTX2_FLAG_NTUPLE_SUPPORT was set. That
enabled ethtool ntuple filters but not tc flower offload. Also set
OTX2_FLAG_TC_FLOWER_SUPPORT when entries are successfully allocated.
Fixes: 2da489432747 ("octeontx2-pf: devlink params support to set mcam entry count") Signed-off-by: Suman Ghosh <sumang@marvell.com> Signed-off-by: Ratheesh Kannoth <rkannoth@marvell.com> Reviewed-by: Simon Horman <horms@kernel.org> Link: https://patch.msgid.link/20260715052007.2099851-1-rkannoth@marvell.com Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Jorn Baayen [Tue, 21 Jul 2026 15:19:23 +0000 (17:19 +0200)]
ASoC: amd: yc: Add DMI quirk for Acer Aspire AG14-22P
The Acer Aspire AG14-22P has its internal microphone connected to the
ACP as a digital microphone, but the BIOS does not advertise it via the
AcpDmicConnected ACPI property, so the internal microphone does not work
out of the box. Add a DMI quirk to enable it.
Tested on an Aspire AG14-22P (board Dove2_MDU, BIOS V1.03): the acp6x
DMIC card is created and the internal microphone captures audio.
mpls: Set rt->rt_nhn just before returning from mpls_nh_build_multi().
Commit f0914b8436c5 ("mpls: Hold dev refcnt for mpls_nh.") added
change_nexthops() loop to call netdev_put() for the nexthop devices
before freeing mpls_route.
Then, mpls_nh_build_multi() was also changed to avoid iterating
uninitialised nexthops in mpls_rt_free_rcu().
However, setting rt->rt_nhn to 0 at the entry of mpls_nh_build_multi()
makes the following change_nexthops() no-op.
Let's set rt->rt_nhn just before returning from mpls_nh_build_multi().
Fixes: f0914b8436c5 ("mpls: Hold dev refcnt for mpls_nh.") Reported-by: Anthony Doeraene <anthony.doeraene@uclouvain.be> Closes: https://lore.kernel.org/netdev/036a0c95-f5d4-46ab-88e7-1eab567d7a84@uclouvain.be/ Signed-off-by: Kuniyuki Iwashima <kuniyu@google.com> Link: https://patch.msgid.link/20260716170609.804629-1-kuniyu@google.com Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Merge tag 'amd-pstate-v7.2-2026-07-22' of ssh://gitolite.kernel.org/pub/scm/linux/kernel/git/superm1/linux
Merge amd-pstate fixes for 7.2 (7/22/26) from Mario Limonciello:
"* Fix a case blocking amd-pstate from binding
when lowest nonlinear freq == minimum freq
* Stop trying to bind in guests"
* tag 'amd-pstate-v7.2-2026-07-22' of ssh://gitolite.kernel.org/pub/scm/linux/kernel/git/superm1/linux:
cpufreq/amd-pstate: Prevent the driver from loading on unsupported hardware
cpufreq/amd-pstate: Loosen requirement on lowest nonlinear frequency != min freq
net: gre: fix lltx regression for GRE tunnels with SEQ/CSUM
Before commit 00d066a4d4ed ("netdev_features: convert NETIF_F_LLTX to
dev->lltx"), NETIF_F_LLTX was set unconditionally in both
__gre_tunnel_init() and ip6gre_tnl_init_features() alongside
GRE_FEATURES:
dev->features |= GRE_FEATURES | NETIF_F_LLTX;
When that commit converted NETIF_F_LLTX to the dev->lltx flag, it
placed 'dev->lltx = true' after the SEQ/CSUM early returns instead
of before them. This causes GRE/GRETAP/ip6gre tunnels with SEQ or
CSUM+encap to lose lockless TX, reintroducing _xmit_lock acquisition
around their ndo_start_xmit. Since GRE xmit re-enters the stack via
ip_tunnel_xmit(), holding _xmit_lock risks ABBA deadlock with the
underlay device.
Daehyeon Ko [Tue, 14 Jul 2026 13:19:39 +0000 (22:19 +0900)]
tipc: clear sock->sk on the failed-insert path in tipc_sk_create()
When tipc_sk_create() fails to insert the new socket (tipc_sk_insert()
returns non-zero), its error path frees the sk with sk_free() but leaves
sock->sk pointing at the freed object:
if (tipc_sk_insert(tsk)) {
sk_free(sk);
pr_warn("Socket create failed; port number exhausted\n");
return -EINVAL;
}
This is harmless for plain socket(): the syscall layer clears sock->ops
before releasing, so tipc_release() is never called. It is not harmless
on the accept() path. tipc_accept() creates the pre-allocated child
socket with tipc_sk_create(net, new_sock, 0, kern); on failure it leaves
new_sock->sk dangling and new_sock->ops non-NULL, and do_accept() then
fput()s the new file, so __sock_release() -> tipc_release() runs
lock_sock(new_sock->sk) on the freed sk -- a use-after-free write of the
sk_lock spinlock.
tipc_release() already guards this exact "failed accept() releases a
pre-allocated child" case with "if (sk == NULL) return 0;", but the
guard is bypassed because tipc_sk_create() left sock->sk non-NULL
(dangling) rather than NULL.
Clear sock->sk on the failed-insert path so the existing tipc_release()
NULL check fires and the use-after-free is avoided.
The tipc_sk_insert() failure is reached when the per-netns socket
rhashtable hits its max_size (tsk_rht_params.max_size = 1048576, ~2M
elements) -- i.e. once a netns holds ~2M TIPC sockets every insert
returns -E2BIG.
BUG: KASAN: slab-use-after-free in lock_sock_nested (net/core/sock.c:3839)
Write of size 8 at addr ffff8880047cdc38 by task init/1
lock_sock_nested (net/core/sock.c:3839)
tipc_release (net/tipc/socket.c:638)
__sock_release (net/socket.c:710)
sock_close (net/socket.c:1501)
__fput (fs/file_table.c:512)
Allocated by task 1:
sk_alloc (net/core/sock.c:2308)
tipc_sk_create (net/tipc/socket.c:487)
tipc_accept (net/tipc/socket.c:2744)
do_accept (net/socket.c:2034)
Freed by task 1:
__sk_destruct (net/core/sock.c:2391)
tipc_sk_create (net/tipc/socket.c:504)
tipc_accept (net/tipc/socket.c:2744)
do_accept (net/socket.c:2034)
Fixes: 00aff3590fc0 ("net: tipc: fix possible refcount leak in tipc_sk_create()") Cc: stable@vger.kernel.org Reviewed-by: Tung Nguyen <tung.quang.nguyen@est.tech> Reviewed-by: Breno Leitao <leitao@debian.org> Signed-off-by: Daehyeon Ko <4ncienth@gmail.com> Reviewed-by: Simon Horman <horms@kernel.org> Link: https://patch.msgid.link/20260714131939.1255974-1-4ncienth@gmail.com Signed-off-by: Paolo Abeni <pabeni@redhat.com>
net: stmmac: enable the MAC on link up for all supported speeds
stmmac_mac_link_down() clears the MAC's transmit and receive enable bits.
stmmac_mac_link_up() is expected to set them again through
stmmac_mac_set(..., true), but it first switches on the negotiated speed
and returns early for a speed the switch does not list. The MAC is then
left gated off.
The speed selection is split into three switches, keyed on the interface.
The generic branch -- taken for everything that is neither USXGMII nor
XLGMII, so including PHY_INTERFACE_MODE_10GBASER -- lists only SPEED_2500,
SPEED_1000, SPEED_100 and SPEED_10.
MGBE on Tegra234 runs 10GBASE-R into an Aquantia AQR113C. That PHY does
rate matching, so phylink_link_up() replaces the media speed with the
MAC-side interface speed before calling into the MAC:
case RATE_MATCH_PAUSE:
speed = phylink_interface_max_speed(link_state.interface);
duplex = DUPLEX_FULL;
which falls through to "default: return;". The interface stops passing
traffic after the first link flap.
The failure is easy to misread. The link still comes up, because the PHY
is polled over MDIO and needs no MAC, so the interface reports carrier 1
at the media speed. The DMA is untouched, so its start bits stay set and
descriptors are still consumed. Only the MAC itself is gated off: the
receiver counts nothing (mmc_rx_framecount_gb stops advancing, RE is 0)
and nothing reaches the wire (TE is 0). The interface survives boot only
because stmmac_hw_setup(), called from ndo_open, enables the MAC
unconditionally -- so the problem appears only once the cable has been
unplugged and plugged back in, and "ip link set dev <ethX> down && ip
link set dev <ethX> up" appears to fix it.
The interface is not what the speed bits depend on: with the single
exception of 2.5G, which is selected through the XGMII block on USXGMII
and through the regular speed bits otherwise, each speed maps to one
field of struct mac_link. The per-interface switches are speed
validation, and phylink already validates the speed against
priv->hw->link.caps. So collapse the three switches into one keyed on the
speed alone, keeping the interface test only for the 2.5G case. This
covers 10G on 10GBASE-R, and equally 5G, and 1G/100/10 on USXGMII, all of
which hit "default: return;" today.
A core that does not support a speed leaves the corresponding mac_link
field at 0, and phylink will not offer it that speed in the first place.
For dwxgmac2 at 10G, link.xgmii.speed10000 is XGMAC_CONFIG_SS_10000,
which is 0 and is the correct speed selection for a 10GBASE-R MAC: ctrl
then equals old_ctrl, the register write is skipped, and execution
reaches stmmac_mac_set(..., true).
Log an error in the default case, since a speed with no entry here leaves
the MAC disabled and the symptom does not point at the cause.
Fixes: d8ca113724e7 ("net: stmmac: tegra: Add MGBE support") Suggested-by: Maxime Chevallier <maxime.chevallier@bootlin.com> Signed-off-by: vadik likholetov <vadikas@gmail.com> Reviewed-by: Jacob Keller <jacob.e.keller@intel.com> Reviewed-by: Maxime Chevallier <maxime.chevallier@bootlin.com> Link: https://patch.msgid.link/20260713074911.30090-1-vadikas@gmail.com Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Guidong Han [Sat, 18 Jul 2026 10:44:06 +0000 (18:44 +0800)]
eventpoll: pin files while checking reverse paths
Commit 319c15174757 ("epoll: take epitem list out of struct file")
intentionally removed temporary file references from the reverse path
check list. At the time, both epitems and their files were freed after
an RCU grace period, so unlist_file() could obtain file->f_lock through
an epitem while clear_tfile_check_list() held rcu_read_lock().
Commit 0ede61d8589c ("file: convert to SLAB_TYPESAFE_BY_RCU") made
struct file SLAB_TYPESAFE_BY_RCU and removed its RCU-delayed freeing.
RCU still protects the epitem, but no longer keeps the referenced file
from being freed and reused. A concurrent close can therefore make
unlist_file() lock or unlock f_lock in a recycled file object.
This violates the documented SLAB_TYPESAFE_BY_RCU rule requiring a
reference before acquiring an object's lock. The race was reproduced,
causing a wild unlock of f_lock in a recycled file and breaking its
mutual exclusion.
Add ->file to epitems_head to remember the pinned file independently of
->epitems. A concurrent EPOLL_CTL_DEL can empty ->epitems before the head
is unlisted, leaving no epi->ffd.file from which to drop the reference.
In list_file(), acquire the reference before adding the head to the
check list. The caller either owns a reference or holds the ep->mtx for
the epitem leading to the file. In the latter case, file_ref_get() can
fail after the last reference is dropped, but eventpoll_release_file()
must acquire the same mutex before the file can be freed. The dying leaf
can be skipped because removing links cannot increase the reverse path
count.
In unlist_file(), epnested_mutex excludes another list_file() or
unlist_file(), while head->next prevents a concurrent EPOLL_CTL_DEL from
freeing the head. Save head->file locally, clear it with head->next
under f_lock, and drop the reference after the RCU-protected operation.
Christian Brauner <brauner@kernel.org> quotes:
> SLAB_TYPESAFE_BY_RCU allows a slab slot to be reused while an RCU reader
> still holds its old address. Once that address contains a new live
> struct file, KASAN sees valid, unpoisoned memory and cannot distinguish
> the stale object identity. CONFIG_DEBUG_SPINLOCK exposes the failure
> instead.
>
> The failing interleaving is:
>
> CPU0: nested EPOLL_CTL_ADD CPU1: close/open churn
> ------------------------------------ ---------------------------------
> p = hlist_first_rcu(&head->epitems)
> epi = container_of(p, ...)
> close(victim)
> __fput()
> eventpoll_release_file()
> file_free(victim)
> // the slot is free; f_lock remains
> spin_lock(&epi->ffd.file->f_lock)
> open() reuses the slot as new_file
> spin_lock_init(&new_file->f_lock)
> spin_unlock(&epi->ffd.file->f_lock) // wild unlock of new_file's lock
>
> CONFIG_DEBUG_SPINLOCK reports:
>
> BUG: spinlock already unlocked on CPU#0, poc_unlist/150
> lock: 0xffff8880067fb200, .magic: dead4ead, .owner: <none>/-1, .owner_cpu: -1
> CPU: 0 UID: 1000 PID: 150 Comm: poc_unlist Not tainted 7.2.0-rc3-dirty #22 PREEMPTLAZY
> Hardware name: QEMU Ubuntu 24.04 PC v2 (i440FX + PIIX, arch_caps fix, 1996), BIOS 1.16.3-debian-1.16.3-2 04/01/2014
> Call Trace:
> <TASK>
> dump_stack_lvl+0x64/0x80
> do_raw_spin_unlock+0x75/0xb0
> _raw_spin_unlock+0xe/0x30
> clear_tfile_check_list+0x88/0xe0
> do_epoll_ctl_file+0x519/0xcf0
> ? __pfx_ep_ptable_queue_proc+0x10/0x10
> do_epoll_ctl+0x8f/0x100
> __x64_sys_epoll_ctl+0x6f/0xa0
> do_syscall_64+0xdc/0x520
> ? srso_alias_return_thunk+0x5/0xfbef5
> entry_SYSCALL_64_after_hwframe+0x76/0x7e
> RIP: 0033:0x42034e
> Code: 48 c7 c1 b8 ff ff ff f7 d8 64 89 01 48 83 c8 ff c3 66 2e 0f 1f 84 00 00 00 00 00 90 f3 0f 1e fa 49 89 ca b8 e9 00 00 00 0f 05 <48> 3d 01 f0 ff ff 73 01 c3 48 c7 c1 b8 ff ff ff f7 d8 64 89 01 48
> RSP: 002b:00007a657ff3c198 EFLAGS: 00000202 ORIG_RAX: 00000000000000e9
> RAX: ffffffffffffffda RBX: 00007a657ff3ccdc RCX: 000000000042034e
> RDX: 0000000000000003 RSI: 0000000000000001 RDI: 0000000000000004
> RBP: 00007a657ff3c2f0 R08: 0000000000000000 R09: 00007a657ff3c6c0
> R10: 00007a657ff3c1a4 R11: 0000000000000202 R12: 00007a657ff3c6c0
> R13: ffffffffffffffb8 R14: 000000000000000d R15: 00007fffb7de0210
> </TASK>
> ------------[ cut here ]------------
>
> unlist_file() does not appear as a separate frame because it was inlined
> into clear_tfile_check_list(). This report was obtained with mdelay()
> instrumentation immediately before spin_lock() and spin_unlock() in
> unlist_file() to widen the two race windows.
>
> More importantly, this is a wild unlock. The stale unlock can target
> f_lock of a different live file and invalidate mutual exclusion for
> state protected by that lock. Turning this into a reliable exploit
> would require precise scheduling and same-slot reuse and is likely
> difficult, but the primitive is potentially exploitable.
Reported-by: Qi Tang <tpluszz77@gmail.com> Reported-by: Junxi Qian <qjx1298677004@gmail.com> Fixes: 0ede61d8589c ("file: convert to SLAB_TYPESAFE_BY_RCU") Cc: stable@vger.kernel.org Signed-off-by: Guidong Han <2045gemini@gmail.com> Link: https://patch.msgid.link/20260718104406.27897-1-2045gemini@gmail.com Signed-off-by: Christian Brauner (Amutable) <brauner@kernel.org>
This series fixes three bugs in the stmmac L3/L4 TC flower filter
implementation for the XGMAC2 core. All three patches target net.
The L3/L4 filter match count statistics patch (originally patch 4/4)
has been split out and will be sent separately against net-next per
Andrew Lunn's review of v1.
Patch 1 fixes a register corruption bug in the L4 filter port configuration.
The XGMAC_L4_ADDR register holds both source and destination port match
values in a single register. The original code overwrites the entire register
when setting either field, silently erasing the other. This is fixed by
using a read-modify-write sequence.
Patch 2 fixes the basic flow match parser to properly reject unsupported
offload requests with -EOPNOTSUPP instead of silently accepting them.
Unsupported cases include partial protocol masks, non-IPv4 network proto,
and non-TCP/UDP transport proto. Extack messages are now included so users
know exactly which part of the match is unsupported. The -EOPNOTSUPP is
also now returned directly instead of using break, which was silently
discarding the error on FLOW_CLS_REPLACE operations.
Patch 3 fixes a stale action bug on filter deletion. When a filter entry
with a drop action is deleted, the action field was not reset, causing
it to persist and potentially affect subsequent filter configurations.
All three patches fix the original L3/L4 filter implementation introduced in 425eabddaf0f ("net: stmmac: Implement L3/L4 Filters using TC Flower").
====================
Nazim Amirul [Tue, 14 Jul 2026 02:37:16 +0000 (19:37 -0700)]
net: stmmac: reset residual action in L3L4 filters on delete
When deleting an L3/L4 flower filter entry, the action field is not
reset. If a filter was previously configured with a drop action, that
action may persist and affect subsequent filter configurations
unintentionally.
Clear the action field when the filter entry is deleted.
Fixes: 425eabddaf0f ("net: stmmac: Implement L3/L4 Filters using TC Flower") Signed-off-by: Rohan G Thomas <rohan.g.thomas@altera.com> Signed-off-by: Nazim Amirul <muhammad.nazim.amirul.nazle.asmade@altera.com> Reviewed-by: Maxime Chevallier <maxime.chevallier@bootlin.com> Link: https://patch.msgid.link/20260714023716.29865-5-muhammad.nazim.amirul.nazle.asmade@altera.com Reviewed-by: Jakub Raczynski <j.raczynski@samsung.com> Signed-off-by: Paolo Abeni <pabeni@redhat.com>
The basic flow parser in tc_add_basic_flow() does not validate match
keys before proceeding. Unsupported offload configurations such as
partial protocol masks, non-IPv4 network proto, or non-TCP/UDP transport
proto are silently accepted instead of returning -EOPNOTSUPP.
Add validation to return -EOPNOTSUPP early for:
- No network or transport proto present in the key
- Partial protocol mask (only full mask supported)
- Network proto is not IPv4
- Transport proto is not TCP or UDP
Each rejection includes an extack message so the user knows which part
of the match is unsupported.
Also propagate -EOPNOTSUPP from tc_add_basic_flow() in tc_add_flow()
by returning it directly rather than using break. The break was silently
discarding the error for FLOW_CLS_REPLACE operations where entry->in_use
is already true, causing tc_add_flow() to return 0 (success) for
unsupported replace requests.
Fixes: 425eabddaf0f ("net: stmmac: Implement L3/L4 Filters using TC Flower") Signed-off-by: Rohan G Thomas <rohan.g.thomas@altera.com> Signed-off-by: Nazim Amirul <muhammad.nazim.amirul.nazle.asmade@altera.com> Reviewed-by: Maxime Chevallier <maxime.chevallier@bootlin.com> Link: https://patch.msgid.link/20260714023716.29865-4-muhammad.nazim.amirul.nazle.asmade@altera.com Reviewed-by: Jakub Raczynski <j.raczynski@samsung.com> Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Nazim Amirul [Tue, 14 Jul 2026 02:37:14 +0000 (19:37 -0700)]
net: stmmac: xgmac: fix l4 filter port overwrite on register update
The XGMAC_L4_ADDR register holds both source and destination port
match values. The current implementation overwrites the entire register
when configuring either port, so setting one silently erases the other.
Fix this by reading the register first, then masking and updating only
the relevant field before writing back.
Fixes: 425eabddaf0f ("net: stmmac: Implement L3/L4 Filters using TC Flower") Signed-off-by: Rohan G Thomas <rohan.g.thomas@altera.com> Signed-off-by: Nazim Amirul <muhammad.nazim.amirul.nazle.asmade@altera.com> Reviewed-by: Maxime Chevallier <maxime.chevallier@bootlin.com> Link: https://patch.msgid.link/20260714023716.29865-3-muhammad.nazim.amirul.nazle.asmade@altera.com Reviewed-by: Jakub Raczynski <j.raczynski@samsung.com> Signed-off-by: Paolo Abeni <pabeni@redhat.com>
bpf: tcp: fix double sock release on batch realloc
bpf_iter_tcp_batch() releases the current batch via
bpf_iter_tcp_put_batch(), which drops the socket refs and rewrites
each slot with the socket cookie, then grows the batch. cur_sk/end_sk
are kept for bpf_iter_tcp_resume(), but on realloc failure the function
returns ERR_PTR() before resume runs, leaving cur_sk < end_sk over
slots that now hold cookies rather than sock pointers.
bpf_iter_tcp_seq_stop() then calls bpf_iter_tcp_put_batch() again and
dereferences a cookie as a struct sock.
Empty the batch on the failure path so stop() does not release it
again. The sockets were already freed by the first
bpf_iter_tcp_put_batch(), so nothing leaks, and a later read() rescans
the bucket from the start instead of skipping it. The sibling
GFP_NOWAIT failure path still holds real socket references and is left
for stop() to release.
BUG: KASAN: null-ptr-deref in __sock_gen_cookie
Read of size 8 at addr 0000000000000059 by task exploit
...
__sock_gen_cookie (net/core/sock_diag.c:28)
bpf_iter_tcp_put_batch (net/ipv4/tcp_ipv4.c:2918)
bpf_iter_tcp_seq_stop (net/ipv4/tcp_ipv4.c:3270)
bpf_seq_read (kernel/bpf/bpf_iter.c:205)
vfs_read (fs/read_write.c:572)
ksys_read (fs/read_write.c:716)
do_syscall_64
entry_SYSCALL_64_after_hwframe
Kernel panic - not syncing: Fatal exception
Fixes: cdec67a489d4 ("bpf: tcp: Make sure iter->batch always contains a full bucket snapshot") Reported-by: AutonomousCodeSecurity@microsoft.com Signed-off-by: Xiang Mei (Microsoft) <xmei5@asu.edu> Reviewed-by: Eric Dumazet <edumazet@google.com> Reviewed-by: Jordan Rife <jordan@jrife.io> Link: https://patch.msgid.link/20260713233230.3553593-1-xmei5@asu.edu Signed-off-by: Paolo Abeni <pabeni@redhat.com>
David Lee [Mon, 13 Jul 2026 10:47:50 +0000 (10:47 +0000)]
net/x25: fix use-after-free in x25_kill_by_neigh()
x25_kill_by_neigh() walks the global X.25 socket list looking for sockets
attached to a terminating neighbour. x25_list_lock protects list membership
while the lookup is in progress, but it does not pin a socket's lifetime
after the lock is dropped.
The function currently drops x25_list_lock before calling lock_sock(s). A
concurrent close can run x25_release(), remove the same socket from
x25_list, and drop the last socket reference in that window. The neighbour
teardown path can then lock or inspect a freed struct sock/struct x25_sock.
Take sock_hold(s) while x25_list_lock still proves that the list entry is
live, then drop the temporary reference after the socket has been locked,
rechecked, and released. Recheck x25_sk(s)->neighbour after lock_sock(),
because another path may have disconnected the socket before this path
acquired the socket lock. Restart the list walk after each disconnect
because the list lock was dropped and the previous iterator state may no
longer be valid.
A QEMU/KASAN run against origin/master reproduced a slab-use-after-free in
x25_kill_by_neigh().
Fixes: 7781607938c8 ("net/x25: Fix null-ptr-deref caused by x25_disconnect") Cc: stable@vger.kernel.org Signed-off-by: David Lee <david.lee@trailofbits.com> Assisted-by: Codex:gpt-5.5 Acked-by: Martin Schiller <ms@dev.tdt.de> Link: https://patch.msgid.link/20260713104752.241175-1-david.lee@trailofbits.com Signed-off-by: Paolo Abeni <pabeni@redhat.com>
The root cause is that KUnit devices are initialized with a 32-bit DMA
mask (DMA_BIT_MASK(32)) in lib/kunit/device.c. On ppc64le and s390x
systems with physical memory above 4GB, page allocations can land at
addresses that exceed this mask. When drm_gem_shmem_get_pages_sgt()
attempts to DMA-map these pages via dma_map_sgtable(), the DMA layer
rejects the mapping because the physical address overflows the 32-bit
mask.
The failure is intermittent because pages may or may not be allocated
above 4GB on any given run depend on memory pressure.
Fix by setting a 64-bit DMA mask on the device before calling
drm_gem_shmem_get_pages_sgt() for all tests, following the same pattern
already used in drm_gem_shmem_test_obj_create_private().
Fixes: 93032ae634d4 ("drm/test: add a test suite for GEM objects backed by shmem") Closes: https://datawarehouse.cki-project.org/issue/5345 Closes: https://datawarehouse.cki-project.org/issue/3184 Assisted-by: Claude:claude-4.6-opus Reviewed-by: Thomas Zimmermann <tzimmermann@suse.de> Signed-off-by: José Expósito <jose.exposito@redhat.com> Link: https://patch.msgid.link/20260703150808.3832-1-jose.exposito89@gmail.com
tipc: fix u16 MTU truncation in media and bearer MTU validation
Both TIPC_NL_MEDIA_SET and TIPC_NL_BEARER_SET accept user-supplied
MTU values but only enforce a minimum bound, not a maximum. When a user
sets the MTU to a value exceeding U16_MAX (65535), it passes validation
but is silently truncated when assigned to u16 fields l->mtu and
l->advertised_mtu in tipc_link_create(). Values like 65536 (0x10000)
truncate to 0, causing a division by zero in tipc_link_set_queue_limits()
which computes TIPC_MAX_PUBL / (l->mtu / ITEM_SIZE). Other overflowing
values (e.g. 65537-131071) produce small incorrect MTU values, resulting
in link malfunction behaviors.
Crash stack (triggered as unprivileged user via user namespace):
Two independent paths lack the upper bound check:
1. tipc_udp_mtu_bad() -- called from __tipc_nl_media_set() (MEDIA_SET)
2. inline check in __tipc_nl_bearer_set() at bearer.c:1160 (BEARER_SET)
Fix both by rejecting MTU values above U16_MAX.
Fixes: 901271e0403a ("tipc: implement configuration of UDP media MTU") Reported-by: AutonomousCodeSecurity@microsoft.com Closes: https://lore.kernel.org/all/CAB8m9WgETt0AjmFwE=F-CKjGXsK6_WDv0=kbYRcC8-noo+amnA@mail.gmail.com Reviewed-by: Vadim Fedorenko <vadim.fedorenko@linux.dev> Signed-off-by: Cen Zhang (Microsoft) <blbllhy@gmail.com> Reviewed-by: Simon Horman <horms@kernel.org> Link: https://patch.msgid.link/20260714041541.307702-1-blbllhy@gmail.com Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Usama Arif [Wed, 15 Jul 2026 10:35:16 +0000 (03:35 -0700)]
fs: push nr_cached_objects memcg gating into individual filesystems
Commit 0baad6f9b997 ("fs/super: skip non-memcg-aware nr_cached_objects
in memcg slab shrink") added a check in fs/super.c that skipped every
->nr_cached_objects() hook whenever the shrinker was invoked for a
non-root memcg, on the assumption that none of them honour sc->memcg.
That assumption is wrong for XFS, whose inode-reclaim hook is
intentionally driven from per-memcg contexts to free memcg-charged
slab. Encoding a blanket "never memcg-aware" policy in fs/super.c
short-circuits that path.
Push the check down into the callbacks whose counters really are
irrelevant to per-memcg reclaim - btrfs_nr_cached_objects() and
shmem_unused_huge_count() - and drop the fs/super.c gate. Each
filesystem can now lift the restriction independently if its counter
later grows memcg awareness, without touching fs/super.c.
Introduce mem_cgroup_shrink_is_root() in <linux/memcontrol.h> so the
callbacks don't open-code "sc->memcg is NULL or root".
Fixes: 0baad6f9b997 ("fs/super: skip non-memcg-aware nr_cached_objects in memcg slab shrink") Acked-by: Qi Zheng <qi.zheng@linux.dev> Reviewed-by: Jan Kara <jack@suse.cz> Reviewed-by: Shakeel Butt <shakeel.butt@linux.dev> Signed-off-by: Usama Arif <usama.arif@linux.dev> Link: https://patch.msgid.link/20260715103516.2410175-1-usama.arif@linux.dev Acked-by: David Sterba <dsterba@suse.com> Reviewed-by: Baolin Wang <baolin.wang@linux.alibaba.com> Signed-off-by: Christian Brauner (Amutable) <brauner@kernel.org>
David Howells [Mon, 13 Jul 2026 15:22:55 +0000 (16:22 +0100)]
afs: Fix afs_edit_dir_remove() to get, not find, block 0
Fix afs_edit_dir_remove() to use afs_dir_get_block() to get block 0 rather
than afs_dir_find_block() as the latter caches the found block in the
afs_dir_iter and may[*] switch out the page it's on if another
afs_dir_find_block() is done. This parallels what afs_edit_dir_add() does.
[*] There's more than one block per page.
Fixes: a5b5beebcf96 ("afs: Use the contained hashtable to search a directory") Closes: https://sashiko.dev/#/patchset/20260706153408.1231650-1-dhowells%40redhat.com Signed-off-by: David Howells <dhowells@redhat.com> Link: https://patch.msgid.link/2380759.1783956175@warthog.procyon.org.uk
cc: Marc Dionne <marc.dionne@auristor.com>
cc: linux-afs@lists.infradead.org
cc: linux-fsdevel@vger.kernel.org
cc: stable@vger.kernel.org Signed-off-by: Christian Brauner (Amutable) <brauner@kernel.org>
Zhang Yi [Mon, 13 Jul 2026 07:42:06 +0000 (15:42 +0800)]
iomap: prevent ioend merge when io_private differs
Different io_private values indicate distinct completion contexts that
must not be merged together, as this could leak or corrupt the private
data associated with each ioend.
Merge patch series "iomap: trivial fixes for ext4 conversion"
Zhang Yi <yi.zhang@huaweicloud.com> says:
This patch series contains a few trivial iomap-related fixes in
preparation for converting ext4 buffered I/O to use iomap.
The first three patches are taken from my ext4 conversion series [1], as
suggested by Christoph. The fourth patch fixes a bug originally reported
by Sashiko during review of my series; although unrelated to the ext4
conversion, it is worth fixing on its own. Please see the following
patches for detail. The fifth patch add comments for
ifs_clear/set_range_dirty(), and the last patch avoids merging ioends
that have different private data.
* patches from https://patch.msgid.link/20260714082325.325163-1-yi.zhang@huaweicloud.com:
iomap: add comments for ifs_clear/set_range_dirty()
iomap: fix out-of-bounds bitmap_set() with zero-length range
iomap: fix incorrect did_zero setting in iomap_zero_iter()
iomap: support invalidating partial folios
iomap: correct the range of a partial dirty clear
Zhang Yi [Tue, 14 Jul 2026 08:23:25 +0000 (16:23 +0800)]
iomap: add comments for ifs_clear/set_range_dirty()
The range alignment strategy differs between ifs_clear_range_dirty() and
ifs_set_range_dirty(). The former rounds inwards to clear only
fully-covered blocks, while the latter rounds outwards to mark any
partially-touched block as dirty. Add comments to document this
asymmetry in block range calculation.
Suggested-by: "Darrick J. Wong" <djwong@kernel.org> Signed-off-by: Zhang Yi <yi.zhang@huawei.com> Link: https://patch.msgid.link/20260714082325.325163-6-yi.zhang@huaweicloud.com Reviewed-by: Joanne Koong <joannelkoong@gmail.com> Reviewed-by: Christoph Hellwig <hch@lst.de> Signed-off-by: Christian Brauner (Amutable) <brauner@kernel.org>
Zhang Yi [Tue, 14 Jul 2026 08:23:24 +0000 (16:23 +0800)]
iomap: fix out-of-bounds bitmap_set() with zero-length range
ifs_set_range_dirty() and ifs_set_range_uptodate() compute last_blk
as (off + len - 1) >> i_blkbits. When off is 0 and len is 0, the
unsigned subtraction underflows to SIZE_MAX, producing a huge
last_blk and nr_blks value that causes bitmap_set() to write far
beyond the ifs->state allocation.
Regarding ifs_set_range_uptodate(), it is temporarily safe because len
cannot be passed in as 0. However, for ifs_set_range_dirty() this is
reachable from __iomap_write_end(): when copy_folio_from_iter_atomic()
returns 0 (e.g. user buffer fault) and the folio is already uptodate,
the guard at the top of __iomap_write_end() does not trigger because
!folio_test_uptodate() is false, and iomap_set_range_dirty() is called
with copied == 0.
Add a !len guard to both functions before the computation, so that a
zero-length range is a no-op.
Fixes: 4ce02c679722 ("iomap: Add per-block dirty state tracking to improve performance") Cc: stable@vger.kernel.org # v6.6 Signed-off-by: Zhang Yi <yi.zhang@huawei.com> Link: https://patch.msgid.link/20260714082325.325163-5-yi.zhang@huaweicloud.com Reviewed-by: Joanne Koong <joannelkoong@gmail.com> Reviewed-by: "Darrick J. Wong" <djwong@kernel.org> Reviewed-by: Christoph Hellwig <hch@lst.de> Signed-off-by: Christian Brauner (Amutable) <brauner@kernel.org>
Zhang Yi [Tue, 14 Jul 2026 08:23:23 +0000 (16:23 +0800)]
iomap: fix incorrect did_zero setting in iomap_zero_iter()
The did_zero output parameter was unconditionally set after the loop,
which is incorrect. It should only be set when the zeroing operation
actually completes, not when IOMAP_F_STALE is set or when
IOMAP_F_FOLIO_BATCH is set but !folio causes the loop to break early,
or when iomap_iter_advance() returns an error.
This causes did_zero to be incorrectly set when zeroing a clean
unwritten extent because the loop exits early without actually zeroing
any data.
Fix it by using a local variable to track whether any folio was actually
zeroed, and only set did_zero after the loop if zeroing happened.
Fixes: 98eb8d95025b ("iomap: set did_zero to true when zeroing successfully") Signed-off-by: Zhang Yi <yi.zhang@huawei.com> Link: https://patch.msgid.link/20260714082325.325163-4-yi.zhang@huaweicloud.com Reviewed-by: "Darrick J. Wong" <djwong@kernel.org> Reviewed-by: Christoph Hellwig <hch@lst.de> Signed-off-by: Christian Brauner (Amutable) <brauner@kernel.org>
Zhang Yi [Tue, 14 Jul 2026 08:23:22 +0000 (16:23 +0800)]
iomap: support invalidating partial folios
Current iomap_invalidate_folio() can only invalidate an entire folio. If
we truncate a partial folio on a filesystem where the block size is
smaller than the folio size, it will leave behind dirty bits for the
truncated or punched blocks. During the write-back process, it will
attempt to map the invalid hole range. Fortunately, this has not caused
any real problems so far because the ->writeback_range() function
corrects the length.
However, the implementation of FALLOC_FL_ZERO_RANGE in ext4 depends on
the support for invalidating partial folios. When ext4 partially zeroes
out a dirty and unwritten folio, it does not perform a flush first like
XFS. Therefore, if the dirty bits of the corresponding area cannot be
cleared, the zeroed area after writeback remains in the written state
rather than reverting to the unwritten state. Fix this by supporting
invalidation of partial folios.
Signed-off-by: Zhang Yi <yi.zhang@huawei.com> Link: https://patch.msgid.link/20260714082325.325163-3-yi.zhang@huaweicloud.com Reviewed-by: "Darrick J. Wong" <djwong@kernel.org> Reviewed-by: Joanne Koong <joannelkoong@gmail.com> Reviewed-by: Christoph Hellwig <hch@lst.de> Signed-off-by: Christian Brauner (Amutable) <brauner@kernel.org>
Zhang Yi [Tue, 14 Jul 2026 08:23:21 +0000 (16:23 +0800)]
iomap: correct the range of a partial dirty clear
The block range calculation in ifs_clear_range_dirty() is incorrect when
partially clearing a range in a folio. We cannot clear the dirty bit of
the first block or the last block if the start or end offset is not
blocksize-aligned. This has not yet caused any issues since we always
clear a whole folio in iomap_writeback_folio().
Fix this by rounding up the first block to blocksize alignment, and
calculate the last block by rounding down (using truncation). Correct
the nr_blks calculation accordingly.
Fixes: 4ce02c679722 ("iomap: Add per-block dirty state tracking to improve performance") Signed-off-by: Zhang Yi <yi.zhang@huawei.com> Link: https://patch.msgid.link/20260714082325.325163-2-yi.zhang@huaweicloud.com Reviewed-by: Joanne Koong <joannelkoong@gmail.com> Reviewed-by: "Darrick J. Wong" <djwong@kernel.org> Reviewed-by: Christoph Hellwig <hch@lst.de> Signed-off-by: Christian Brauner (Amutable) <brauner@kernel.org>
Alexandru Elisei [Wed, 22 Jul 2026 09:03:54 +0000 (10:03 +0100)]
KVM: arm64: Reject guest_memfd memslots when the VM has MTE
The user cannot use MTE on VMAs created by mapping a guest_memfd file,
as arch_calc_vm_flag_bits() does not set VM_MTE_ALLOWED.
When creating a guest_memfd backed memslot,
kvm_arch_prepare_memory_region() rejects the memslot if MTE is enabled for
the VM and if guest_memfd has been mapped in a VMA that intersects the
memslot.
However, the documentation for KVM_SET_USER_MEMORY_REGION2 explicitly
states that the only condition for userspace_addr is for it to be a legal
userspace address, but the mapping is not required to be valid nor
populated at memslot creation.
If userspace sets userspace_addr to an address that hasn't been mapped, or
if userspace_addr belongs to a VMA that isn't backed by the guest_memfd
file, or if the VMA doesn't intersect the memslot, memslot creation is
successful and KVM ends up with a VM with MTE and guest_memfd-backed
memslots.
The same happens if the order is reversed: when userspace enables MTE, KVM
does not check if memslots backed by guest_memfd are already present.
Fix both issues by rejecting guest_memfd-backed memslots when MTE is
enabled, and by rejecting MTE when guest_memfd-backed memslots are already
present.
Fixes: 32e200bd6e44 ("KVM: arm64: Enable support for guest_memfd backed memory") Tested-by: Fuad Tabba <fuad.tabba@linux.dev> Reviewed-by: Fuad Tabba <fuad.tabba@linux.dev> Signed-off-by: Alexandru Elisei <alexandru.elisei@arm.com> Link: https://patch.msgid.link/20260722090354.94245-1-alexandru.elisei@arm.com Signed-off-by: Marc Zyngier <maz@kernel.org>
KVM: arm64: Fix hyp_trace_desc allocation size in hyp_trace_load()
The footprint calculated for struct hyp_trace_desc sizes only
trace_buffer_desc and do not take into account the other fields. It
worked so far thanks to the follow-up PAGE_ALIGN().
Fix the descriptor size and while at it, enforce an overflow check after
PAGE_ALIGN().
Reported-by: Sashiko <sashiko-bot@kernel.org> Fixes: 3aed038aac8d ("KVM: arm64: Add trace remote for the nVHE/pKVM hyp") Signed-off-by: Vincent Donnefort <vdonnefort@google.com> Reviewed-by: Fuad Tabba <fuad.tabba@linux.dev> Tested-by: Fuad Tabba <fuad.tabba@linux.dev> Link: https://patch.msgid.link/20260710114819.2689386-3-vdonnefort@google.com Signed-off-by: Marc Zyngier <maz@kernel.org>
KVM: arm64: Fix potential leak in hyp_trace_buffer_alloc_bpages_backing
In the very unlikely event of a failure in __map_hyp, the allocated
backing pages are leaked in hyp_trace_buffer_alloc_bpages_backing(). Fix
this by freeing the pages on error.
Fixes: 3aed038aac8d ("KVM: arm64: Add trace remote for the nVHE/pKVM hyp") Reported-by: Sashiko <sashiko-bot@kernel.org> Reviewed-by: Fuad Tabba <fuad.tabba@linux.dev> Tested-by: Fuad Tabba <fuad.tabba@linux.dev> Signed-off-by: Vincent Donnefort <vdonnefort@google.com> Link: https://patch.msgid.link/20260710114819.2689386-2-vdonnefort@google.com Signed-off-by: Marc Zyngier <maz@kernel.org>
Fix the disable path in hyp_trace_clock_enable(), which fell through to
re-initialize and reschedule the clock after cancelling the work. Return
early instead.
While at it, cleanup hyp_trace_clock::lock which is unused and
hyp_trace_clock::running which is redundant: the trace_remote framework
already serializes calls to the callback enable_tracing.
Fixes: b22888917fa4 ("KVM: arm64: Sync boot clock with the nVHE/pKVM hyp") Signed-off-by: Vincent Donnefort <vdonnefort@google.com> Reviewed-by: Fuad Tabba <fuad.tabba@linux.dev> (✓ DKIM/linux.dev) Link: https://patch.msgid.link/20260715105100.3178255-1-vdonnefort@google.com Signed-off-by: Marc Zyngier <maz@kernel.org>
Mitigate a potential failure when inserting a new LPI into the VGIC LPI
xarray.
When vgic_add_lpi() is preparing to register a new LPI, it pre-allocates
an xarray entry using xa_reserve_irq(), so that it can later perform the
insertion under the xarray lock without allocating.
However, since xa_reserve_irq() is called before acquiring such lock,
there is a potential race where xa_reserve_irq() observes a populated
entry, thus not performing the allocation, and another CPU removes that
entry before the xarray lock is grabbed to perform the insertion.
This can happen e.g. if the guest issues a DISCARD while the LPI is
still referenced from a vCPU's active-pending list (ap_list), and the
same INTID is re-mapped via MAPTI.
Mitigate this by passing GFP_NOWAIT to __xa_store(), so that the
allocation can happen under the lock in the rare case that this
condition is hit. Add __GFP_ACCOUNT as well to match xa_reserve_irq()'s
flags.
Reported-by: Sashiko <sashiko-bot@kernel.org> Fixes: 1d6f83f60f79 ("KVM: arm64: vgic: Store LPIs in an xarray") Signed-off-by: Carlos López <clopez@suse.de> Link: https://patch.msgid.link/20260715105137.3973823-5-clopez@suse.de Signed-off-by: Marc Zyngier <maz@kernel.org>
Carlos López [Wed, 15 Jul 2026 10:51:37 +0000 (12:51 +0200)]
KVM: arm64: vgic: Fix race between LPI release and re-registration
Fix a potential race between decrementing an LPI's reference count and
evicting that structure from the LPI xarray.
LPI structures are maintained in the VGIC LPI xarray (dist->lpi_xa).
When the reference count of an LPI structure drops to zero,
vgic_release_lpi_locked() removes the structure from the xarray and
frees it under the xarray lock.
However, the release of an LPI can race with a concurrent LPI
re-registration with the same INTID via vgic_add_lpi() on another CPU,
since the reference count drop and the xarray eviction are not performed
in a single atomic step. This can happen e.g. if the guest issues a
DISCARD while the LPI is still referenced from a vCPU's active-pending
list (ap_list), and the same INTID is re-mapped via MAPTI.
Particularly, vgic_release_lpi_locked() is called from two distinct
paths: direct release via vgic_put_irq(), and deferred release via
vgic_release_deleted_lpis(). During direct release, the issue can result
in deleting a newly registered LPI from the xarray:
CPU0 (Releasing LPI) CPU1 (Adding new LPI)
==================== =====================
vgic_put_irq()
__vgic_put_irq()
refcount_dec_and_test()
vgic_add_lpi()
xa_lock_irqsave()
old_irq = xa_load(.., intid)
vgic_try_get_irq_ref(old_irq) == false
new IRQ inserted --> __xa_store(.., intid, ..)
xa_unlock_irqrestore()
xa_lock_irqsave();
vgic_release_lpi_locked()
__xa_erase(.., irq->intid) <-- BUG: new IRQ is erased
kfree_rcu(old_irq)
During the deferred release path, the old IRQ can be leaked:
vgic_release_deleted_lpis()
xa_lock_irqsave()
xa_for_each() { .. } <-- old IRQ with pending_release = true
is gone, so it cannot be released
To fix the direct release path, move the reference count drop inside
the xarray lock, making sure that vgic_add_lpi() never encounters the
to-be-released LPI.
In the deferred release path, the refcount drop must happen under a raw
spinlock, so the xarray lock cannot be grabbed, and the same solution
does not work. Instead, update vgic_add_lpi(), so that if it evicts
an LPI from the xarray, it takes on the responsibility of freeing it.
Consequently, an LPI may now be freed concurrently after a deferred
release drops the refcount, so accessing the pending_release field is no
longer safe from use-after-free. Delete all uses of the flag, and update
vgic_release_deleted_lpis() to identify orphaned LPIs purely based on
their refcount.
Reported-by: Claude:claude-opus-4-6 Fixes: 3a08a6ca7c37 ("KVM: arm64: vgic-v3: Use bare refcount for VGIC LPIs") Fixes: d54594accf73 ("KVM: arm64: vgic-v3: Erase LPIs from xarray outside of raw spinlocks") Signed-off-by: Carlos López <clopez@suse.de> Link: https://patch.msgid.link/20260715105137.3973823-4-clopez@suse.de Signed-off-by: Marc Zyngier <maz@kernel.org>
Andrei Kuchynski [Fri, 17 Jul 2026 10:46:14 +0000 (10:46 +0000)]
usb: typec: ucsi: Correct teardown ordering in ucsi_init() error path
The commit 7aa7d4bf9d3f ("usb: typec: ucsi: Fix race condition and
ordering in port unregistration") consolidated port teardown into the
ucsi_unregister_port() helper. However, it introduced an ordering problem
in the ucsi_init() error path.
Fix this by ensuring ucsi_unregister_port() is called before we unregister
their corresponding lockdep keys.
Cc: stable@vger.kernel.org Fixes: 7aa7d4bf9d3f ("usb: typec: ucsi: Fix race condition and ordering in port unregistration") Reported-by: "Borah, Chaitanya Kumar" <chaitanya.kumar.borah@intel.com> Closes: https://lore.kernel.org/all/22064276-6c56-411a-9f20-6917ceeb865f@intel.com/ Signed-off-by: Andrei Kuchynski <akuchynski@chromium.org> Tested-by: Chaitanya Kumar Borah <chaitanya.kumar.borah@intel.com> Reviewed-by: Heikki Krogerus <heikki.krogerus@linux.intel.com> Link: https://patch.msgid.link/20260717104614.325250-1-akuchynski@chromium.org Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
fs/super: fix emergency thaw double-unlock of s_umount
do_thaw_all() iterates over all superblocks via __iterate_supers()
with SUPER_ITER_EXCL, which acquires s_umount exclusively before
calling the callback and releases it afterwards. However, the
callback do_thaw_all_callback() calls thaw_super_locked() which
unconditionally releases s_umount on every code path. This results
in a second unlock attempt in __iterate_supers() that corrupts the
rwsem state, triggering a DEBUG_RWSEMS warning:
Fix this by switching to SUPER_ITER_UNLOCKED and acquiring s_umount
in the callback via super_lock_excl() before calling
thaw_super_locked(). This matches the locking pattern expected by
thaw_super_locked() and eliminates the double unlock.
While at it, remove the dead 'return;' at the end of
do_thaw_all_callback().
After a recent upstream LLVM change to start generating jump and lookup
tables in switch statements in more instances [1], linking the
compressed x86 boot image when CONFIG_KERNEL_ZSTD is enabled fails with:
Dumping the relocations in misc.o, which is the only file influenced by
CONFIG_KERNEL_ZSTD in the decompressor, shows dynamic relocations to
some string constants, which correspond to the string literals in the
switch statement in handle_zstd_error():
This optimization is problematic for the decompressor environment, as it
is built as -fPIE without any explicit absolute references (as described
at the top of misc.c) while not applying any dynamic relocations, hence
the linker assertion. To opt out of this optimization, which is of
little value in this special early boot code, and to mirror the other
x86 startup code in arch/x86/boot/startup, disable jump tables in the
decompressor.
Shuicheng Lin [Tue, 21 Jul 2026 20:55:14 +0000 (20:55 +0000)]
drm/xe/vm: Fix SVM leak on resv obj alloc failure in xe_vm_create()
Commit 9e9787414882 ("drm/xe/userptr: replace xe_hmm with gpusvm") made
xe_svm_init() unconditional in xe_vm_create() and extended it to also
initialize a "simple" gpusvm state for non-fault-mode VMs. The matching
xe_svm_fini() call in xe_vm_close_and_put() was updated to run
unconditionally, but the error unwind path in xe_vm_create() was not.
On the drm_gpuvm_resv_object_alloc() failure path, xe_svm_init() has
already succeeded but xe_svm_fini() is only called when
XE_VM_FLAG_FAULT_MODE is set. For non-fault-mode VMs this leaves
vm->svm.gpusvm partially initialized and leaks the resources allocated
by drm_gpusvm_init().
For fault-mode VMs, xe_svm_init() additionally acquires the pagemap
owner via drm_pagemap_acquire_owner() and the pagemaps via
xe_svm_get_pagemaps(). Those resources are released by xe_svm_close(),
not xe_svm_fini(). On the same error path, xe_svm_close() is not
called either, so fault-mode VMs leak the pagemap owner and pagemaps.
Fix both leaks:
- Call xe_svm_fini() unconditionally on the err_svm_fini path, matching
the unconditional xe_svm_init() call. Move the vm->size = 0
assignment out of the conditional so the xe_vm_is_closed() assert in
xe_svm_fini() (and xe_svm_close()) holds for both modes.
- Call xe_svm_close() for fault-mode VMs before xe_svm_fini(), matching
the ordering used in xe_vm_close_and_put().
Fixes: 9e9787414882 ("drm/xe/userptr: replace xe_hmm with gpusvm") Cc: Matthew Auld <matthew.auld@intel.com> Assisted-by: Claude:claude-opus-4.7 Reviewed-by: Matthew Brost <matthew.brost@intel.com> Link: https://patch.msgid.link/20260721205516.4058959-2-shuicheng.lin@intel.com Signed-off-by: Shuicheng Lin <shuicheng.lin@intel.com>
(cherry picked from commit ca2a3587d577ba764e0fe628fb676244fc33ddd4) Signed-off-by: Thomas Hellström <thomas.hellstrom@linux.intel.com>
PCI bus, device and function can be same for devices existing across
different domains. Allow per domain unique identifier while registering
platform device to prevent name conflict.
Fixes: f0e53aadd702 ("drm/xe: Support for I2C attached MCUs") Signed-off-by: Raag Jadav <raag.jadav@intel.com> Reviewed-by: Heikki Krogerus <heikki.krogerus@linux.intel.com> Link: https://patch.msgid.link/20260721113438.651100-1-raag.jadav@intel.com Signed-off-by: Matt Roper <matthew.d.roper@intel.com>
(cherry picked from commit a79f6abc8b516b5bd906e2eca8121e3549ee163f) Signed-off-by: Thomas Hellström <thomas.hellstrom@linux.intel.com>
mshv: Fix sleeping under spinlock in mshv_portid_alloc
idr_alloc() is called with GFP_KERNEL inside idr_lock(), which holds a
spinlock. GFP_KERNEL allows the allocator to sleep, triggering a
sleeping-while-atomic bug.
Fix by using idr_preload(GFP_KERNEL) before taking the lock to
pre-allocate memory in a sleepable context, then idr_alloc() with
GFP_NOWAIT inside the spinlock-protected section.
Fixes: 621191d709b1 ("Drivers: hv: Introduce mshv_root module to expose /dev/mshv to VMMs") Signed-off-by: Stanislav Kinsburskii <skinsburskii@linux.microsoft.com> Reviewed-by: Anirudh Rayabharam (Microsoft) <anirudh@anirudhrb.com> Signed-off-by: Wei Liu <wei.liu@kernel.org>
The duplicate routing entry check in mshv_update_routing_table() uses
guest_irq_num != 0 to detect whether a GSI slot is already occupied.
This fails for GSI 0 because its guest_irq_num is 0 both when the slot
is unused (zero-initialized) and when legitimately assigned. As a
result, duplicate entries for GSI 0 are silently accepted, with the
second entry overwriting the first — corrupting the routing table
without any error reported to userspace.
While GSI 0 (legacy timer) is unlikely to appear in MSI-based routing
in practice, the check is semantically wrong — it conflates
"uninitialized" with "GSI number 0." Use girq_entry_valid instead,
which is explicitly set to true when an entry is populated and remains
zero for unused slots regardless of the GSI number.
Fixes: 621191d709b14 ("Drivers: hv: Introduce mshv_root module to expose /dev/mshv to VMMs") Signed-off-by: Stanislav Kinsburskii <skinsburskii@linux.microsoft.com> Reviewed-by: Anirudh Rayabharam (Microsoft) <anirudh@anirudhrb.com> Signed-off-by: Wei Liu <wei.liu@kernel.org>
vmbus_irq_initialized is only true if the registration of the per-CPU
threads succeeded. If it failed, the whole registration aborts and the
vmbus_exit() path is never called.
Remove vmbus_irq_initialized.
Reviewed-by: Michael Kelley <mhklinux@outlook.com> Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de> Signed-off-by: Wei Liu <wei.liu@kernel.org>
Drivers: hv: vmbus: Replace lockdep_hardirq_threaded() with lockdep annotation
lockdep_hardirq_threaded() is supposed to be used within IRQ core code
and not within drivers. It is not obvious from within the driver, that
this is the only interrupt service routing and that it is not shared
handler.
Replace lockdep_hardirq_threaded() with a lockdep annotation limiting
threaded context on PREEMPT_RT to __vmbus_isr().
Fixes: f8e6343b7a89c ("Drivers: hv: vmbus: Use kthread for vmbus interrupts on PREEMPT_RT") Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de> Reviewed-by: Michael Kelley <mhklinux@outlook.com> Signed-off-by: Wei Liu <wei.liu@kernel.org>
Michael Kelley [Mon, 8 Jun 2026 02:06:16 +0000 (19:06 -0700)]
Drivers: hv: vmbus: Set DMA coherent mask for VMBus devices
In current code, the coherent_dma_mask for VMBus devices is not set, so
it has the default value of 0, which essentially means "invalid". Because
drivers for VMBus devices do not use dma_alloc_*() functions, the usual
use of the coherent mask does not occur, and no errors result.
However, a valid coherent_dma_mask may be needed even though the drivers
don't use dma_alloc_*() functions. In a CoCo VM, the VMBus storvsc and
netvsc drivers must bounce buffer DMA operations through the swiotlb
because the Hyper-V host can't DMA into encrypted guest memory. If the
kernel is built with CONFIG_SWIOTLB_DYNAMIC and the initial swiotlb size
is small, swiotlb code may need to grow the swiotlb in response to a DMA
mapping request. That growth first allocates a transient pool while the
swiotlb is expanded in the background. The transient pool memory is
allocated from the DMA atomic pools, and the allocation code checks for
a valid coherent_dma_mask. With current code, this check fails, then the
DMA mapping request from the storvsc or netvsc driver fails, and finally
an I/O error occurs.
Fix this problem by setting coherent_dma_mask for VMBus devices at the
same time that dma_mask is set. Being a synthetic bus, VMBus does not
have any restrictions on coherent DMA, so the coherent mask is set to
the full 64 bits for all VMBus devices, just like with dma_mask.
Signed-off-by: Michael Kelley <mhklinux@outlook.com> Signed-off-by: Wei Liu <wei.liu@kernel.org>
projects / thirdparty / kernel / linux.git / log
