bnge_fix_rings_count() returns 0 on success or a negative errno on failure
However, bnge_adjust_rings() stores its return value in a u16 variable,
causing negative error codes such as -ENOMEM to be converted to a large
positive value.
Use an int for the return code variable so that error values are
preserved and propagated correctly.
Fixes: 627c67f038 ("bng_en: Add resource management support")
Signed-off-by: Alok Tiwari <alok.a.tiwari@oracle.com>
Reviewed-by: Bhargava Marreddy <bhargava.marreddy@broadcom.com>
Link: https://patch.msgid.link/20260801100923.1498570-1-alok.a.tiwari@oracle.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Yangyu Chen says:
====================
net: atlantic: fix two ring teardown leaks
These are the two fixes from the page_pool conversion series [1],
resent against net as requested in the review of that series. The
page_pool conversion itself stays in net-next and is not part of this
posting; it depends on these fixes, but they stand on their own.
Both patches are unchanged from [1] apart from the collected
Reviewed-by tags, and each carries a Fixes tag and a Cc: stable with
the affected range (patch 1: v4.11+, patch 2: v5.2+). They apply and
were build- and runtime-tested independently of each other and of the
conversion.
Patch 1: aq_vec_deinit() drains the TX rings with a single
aq_ring_tx_clean() call, which is capped at AQ_CFG_TX_CLEAN_BUDGET
descriptors and stops at hw_head, frozen once the hardware and NAPI
have been stopped. Everything beyond that keeps its skb or xdp_frame
when the interface goes down and is lost when the buffer ring is
freed.
Patch 2: aq_ring_rx_deinit() only walks [sw_head, sw_tail). Since the
page reuse strategy was added, a cleaned RX buffer keeps its page for
reuse and refill is batched, so consumed but not yet reposted slots
accumulate in the [sw_tail, sw_head) gap and their pages and DMA
mappings are never released.
Reproduction logs for both leaks (as page_pool stalled shutdowns,
which is how they become visible) are in the notes of the respective
patches.
[1] https://lore.kernel.org/lkml/tencent_1F173E0FC1606D2AC704DC9C98AF10984607@qq.com/
====================
Link: https://patch.msgid.link/tencent_29B860317921D68DE77C718242DA418EB608@qq.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
aq_ring_rx_deinit() only walks [sw_head, sw_tail), the region posted to
hardware. Since the page reuse strategy was added, a cleaned RX buffer
keeps its page (and its DMA mapping) in the ring for reuse, and refill
is batched: aq_ring_rx_fill() returns early until AQ_CFG_RX_REFILL_THRES
slots are free. Slots that were consumed but not yet reposted therefore
sit in the complementary [sw_tail, sw_head) gap with a live page, and
the deinit walk never visits them: up to a refill batch worth of pages
and DMA mappings leak on every interface down.
Walk the whole ring instead and release whatever is still there. Also
bail out if the buffer ring is already gone: a partial
aq_ptp_ring_alloc() failure frees the ring but leaves aq_nic set, so
aq_ptp_ring_deinit() still gets here on the unwind path.
Cc: stable@vger.kernel.org # v5.2+
Fixes: 46f4c29d9d ("net: aquantia: optimize rx performance by page reuse strategy")
Reviewed-by: Sukhdeep Singh <sukhdeeps@marvell.com>
Signed-off-by: Yangyu Chen <cyy@cyyself.name>
Acked-by: Mina Almasry <almasrymina@google.com>
Link: https://patch.msgid.link/tencent_607CBA8237DA438E36B844318B21538DE008@qq.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
aq_vec_deinit() drains the TX rings with a single aq_ring_tx_clean()
call, which frees at most AQ_CFG_TX_CLEAN_BUDGET (256) descriptors and
stops at hw_head, which no longer moves once aq_vec_stop() has stopped
the hardware and NAPI. Completed descriptors beyond the budget and
everything still posted in [hw_head, sw_tail) keep their skb or
xdp_frame when the interface goes down: aq_vec_ring_free() then frees
the buffer ring and the references are lost for good.
Today this is a silent memory leak on every interface down under
TX/XDP_TX load. With the conversion of the RX path to page_pool posted
for net-next it becomes much more visible: XDP_TX frames carry fragment
references on the RX ring's page_pool, so a single stranded frame keeps
the pool's inflight count above zero forever. page_pool_destroy() then
never completes, the pool is leaked together with its pages, and
"page_pool_release_retry() stalled pool shutdown" is warned every 60
seconds from that point on, on every ifdown, XDP detach or ring resize
under XDP_TX load.
Bring back aq_ring_tx_deinit() as it was before the removal and use it
for teardown again, with one extension: TX rings can hold xdp_frames
nowadays, so release those too. They are returned with
xdp_return_frame() since this runs in process context.
Fixes: eb36bedf28 ("net: aquantia: remove function aq_ring_tx_deinit")
Cc: stable@vger.kernel.org # v4.11+
Reviewed-by: Sukhdeep Singh <sukhdeeps@marvell.com>
Signed-off-by: Yangyu Chen <cyy@cyyself.name>
Acked-by: Mina Almasry <almasrymina@google.com>
Link: https://patch.msgid.link/tencent_EEDC35FAF2750A3A6A0B39BAE0E2C484860A@qq.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Since the referenced commit, changing the MTU on a running interface no
longer disconnects and reconnects the PHY; __stmmac_release() merely
stops phylink, which also suspends the PHY (BMCR power-down) when WoL
is not enabled. __stmmac_open() then performs the DMA software reset in
stmmac_hw_setup() before phylink_start() resumes the PHY again.
IEEE 802.3 22.2.4.1.5 allows a PHY to stop its receive clock while
powered down, and stmmac requires a running receive clock for the DMA
software reset to complete (the phylink config sets mac_requires_rxc).
On such setups, e.g. the RK3566-based Home Assistant Green with an
RTL8211F-VD PHY in RGMII mode, any runtime MTU change now times out and
leaves the interface dead:
rk_gmac-dwmac fe010000.ethernet end0: Failed to reset the dma
rk_gmac-dwmac fe010000.ethernet end0: stmmac_hw_setup: DMA engine initialization failed
rk_gmac-dwmac fe010000.ethernet end0: __stmmac_open: Hw setup failed
rk_gmac-dwmac fe010000.ethernet end0: failed reopening the interface after MTU change
In the field this is triggered by NetworkManager applying an MTU while
activating the connection, breaking networking entirely. The same
regression has also been reported on i.MX8MP and reproduced on SoCFPGA
based systems.
Resume the PHY in __stmmac_open() before the hardware setup, making it
the counterpart of the phylink_stop() in __stmmac_release(), like
stmmac_resume() already does for the same reason. phylink_start() also
resumes the PHY, but only after stmmac_hw_setup(), and it cannot be
moved before the hardware setup since it may bring the link up
immediately from a workqueue, racing with the initialization (see the
comment in stmmac_resume()). For the regular ndo_open path the PHY has
just been attached and is not suspended, in which case
phylink_prepare_resume() does nothing.
Fixes: db299a0c09 ("net: stmmac: move PHY handling out of __stmmac_open()/release()")
Link: https://github.com/home-assistant/operating-system/issues/4858
Tested-by: Alexander Stein <alexander.stein@ew.tq-group.com>
Signed-off-by: Stefan Agner <stefan@agner.ch>
Tested-by: Maxime Chevallier <maxime.chevallier@bootlin.com>
Reviewed-by: Maxime Chevallier <maxime.chevallier@bootlin.com>
Link: https://patch.msgid.link/20260803095156.132827-1-stefan@agner.ch
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Antonio Quartulli says:
====================
Included fixes:
* use rcu_dereference_bh() instead of rcu_access_pointer() where the
pointer is actually dereferenced
* ensure TCP global variables are initialized before they can be
accessed via netlink (e.g. when attaching a TCP socket)
* actually disable IPv4 redirects on multipeer interfaces (the
previous attempt was a no-op and did not survive netns moves)
* hash a floated peer by its transport identity only, consistently
with the add and lookup paths
* zero the sockaddr padding before learning a floated endpoint so it
does not leak into the by_transp_addr hash key
* ensure the socket is owned by ovpn before dereferencing
sk_user_data
* rehash a peer in the by_transp_addr table when its remote endpoint
is updated via CMD_PEER_SET
* avoid re-adding to the hashtables a peer that was concurrently
removed (use-after-free)
* limit keepalive values to one day to avoid overflowing the
delayed-work delay on 32-bit systems
* add the missing rtnl_link_ops->get_size callback so link messages
account for the nested mode attribute
* tag 'ovpn-net-20260730' of https://github.com/OpenVPN/ovpn-net-next:
ovpn: fix incorrect use of rcu_access_pointer()
ovpn: ensure TCP vars are initialized first
ovpn: disable IPv4 redirects on MP interfaces
ovpn: hash floated peer by transport identity only
ovpn: zero-initialize sockaddr before learning a floated endpoint
ovpn: ensure socket is owned by ovpn before deref sk_user_data
ovpn: rehash peer in by_transp_addr table on CMD_PEER_SET
ovpn: skip rehash for peers already removed from by_id
ovpn: limit keepalive values to one day
ovpn: add missing rtnl_link_ops->get_size callback
====================
Link: https://patch.msgid.link/20260730094624.4102963-1-antonio@openvpn.net
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
in6_dev_get() reads dev->ip6_ptr under RCU and then unconditionally
increments its refcount. Device teardown can clear the pointer and drop
the last reference between these operations. The increment then
resurrects an object whose RCU free has already been queued, so callers
can use it after it is freed.
Use refcount_inc_not_zero() and return NULL when the object has already
reached zero. RCU keeps the memory accessible through the attempted
reference acquisition, and a successful increment pins the object for
the caller.
An independent run on the exact unpatched 6f5156d7a3 (v7.2-rc3)
kernel reproduced the invalid reference acquisition as UID 1000:
refcount_t: addition on 0; use-after-free.
ip6_mc_source+0xef4/0x17e0
It was followed by the corresponding reference underflow in
ip6_mc_source(). The supplied trace from the same unpatched revision
additionally shows the access after the RCU read-side section ends:
BUG: KASAN: slab-use-after-free in mutex_lock+0x76/0xe0
Write of size 8 at addr ffff888015b50240 by task poc/1219
Bug found and triaged by OpenAI Security Research and
validated by Trail of Bits.
Fixes: 8814c4b533 ("[IPV6] ADDRCONF: Convert addrconf_lock to RCU.")
Cc: stable@vger.kernel.org
Signed-off-by: Kyle Zeng <kylebot@openai.com>
Co-developed-by: David Lee <david.lee@trailofbits.com>
Signed-off-by: David Lee <david.lee@trailofbits.com>
Reviewed-by: Ido Schimmel <idosch@nvidia.com>
Link: https://patch.msgid.link/20260803122758.666112-1-david.lee@trailofbits.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Another challenge with unlocked filters.
There is a short window in tc_new_tfilter where a tcf_proto can be found
and briefly referenced by a totally unrelated, unlocked classifier's request
and cause a race.
Feng created a poc which created this race with two threads, one creating a
u32 filter and other a flower filter in the same chain/prio:
1. Both threads enter tc_new_tfilter, both find the chain empty, both
drop filter_chain_lock
2. u32 finishes tcf_proto_create("u32") first, calls
tcf_chain_tp_insert_unique() -> inserts u32_tp into the chain
3. flower finishes tcf_proto_create("flower") later, calls
tcf_chain_tp_insert_unique() -> tcf_chain_tp_find() now sees u32_tp
already there, takes a reference on it, destroys flower's own tp_new
and returns u32_tp to the caller.
Flower then hits the kind mismatch check (because it requested for kind
"flower" but tp->ops->kind is "u32") and goes through the errout path
which calls tcf_proto_put() on u32_tp. If the u32 thread has already
gone through its own errout (its change() call failed on the PoC's empty
options) and dropped its create and insert refs, flower's put is the
last one and drops u32_tp's refcnt to zero.
At this point tp->ops->destroy() runs in a context that never took
rtnl_lock. When that happens, it might cause a UAF like the following
(illustrated by the PoC):
[ +0.000710] BUG: KASAN: slab-use-after-free in u32_init (net/sched/cls_u32.c:393)
[ +0.000281] Read of size 8 at addr ffff888120022f00 by task poc_feng_xue/524
Call Trace:
u32_init (net/sched/cls_u32.c:393)
tc_new_tfilter (net/sched/cls_api.c:2378)
Allocated by task 526:
u32_init (net/sched/cls_u32.c:378)
tc_new_tfilter (net/sched/cls_api.c:2378)
Freed by task 522:
kfree
u32_destroy (net/sched/cls_u32.c:662)
tcf_proto_destroy (net/sched/cls_api.c:446)
tcf_proto_put (net/sched/cls_api.c:459)
tc_new_tfilter (net/sched/cls_api.c:2459)
Fix this by having tcf_proto_destroy() take rtnl_lock around
tp->ops->destroy() for locked classifiers whenever rtnl is not held.
To explain why I used a temp variable "not_lockless" I'd like to point to a
semi-related note on rtnl_held vs TCF_PROTO_OPS_DOIT_UNLOCKED (adding here
for future cleanup if deemed necessary):
The rtnl_held parameter and the TCF_PROTO_OPS_DOIT_UNLOCKED flag are
redundant sources of truth for whether rtnl_lock is held. Among the nine
classifier destroy(..rtnl_held..) callbacks, only flower consults the
rtnl_held parameter which it propagates to tc_setup_cb_destroy()
and tc_setup_cb_call(). The other eight (u32, flow, bpf, cgroup, route, basic,
fw, mall) ignore it entirely;-> those that call tc_setup_cb_destroy()
(u32, bpf, mall) hardcode true always instead of forwarding the parameter.
A future cleanup should remove the rtnl_held parameter from the destroy callback
signature entirely and have callers rely solely on their knowledge whether
they are running in an unlocked context.
Fixes: 12db03b65c ("net: sched: extend proto ops to support unlocked classifiers")
Reported-by: Feng Xue <feng.xue@outlook.com>
Tested-by: Victor Nogueira <victor@mojatatu.com>
Signed-off-by: Jamal Hadi Salim <jhs@mojatatu.com>
Link: https://patch.msgid.link/20260801125632.360365-1-jhs@mojatatu.com
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
When a packet arrives on an ARPHRD_NONE device (e.g. TUN),
ovs_flow_key_extract() trusts the user-provided skb->protocol field: if
it is ETH_P_TEB, the packet is classified as MAC_PROTO_ETHERNET and
key_extract() is called without ensuring the skb has ETH_HLEN (14) bytes
of linear data. key_extract() unconditionally pulls 2 * ETH_ALEN bytes
for MAC addresses and parse_ethertype() pulls 2 more, either of which
triggers a kernel BUG in __skb_pull() when the linear area is too small.
kernel BUG at include/linux/skbuff.h:2848!
RIP: 0010:key_extract+0xa7e/0xd90 net/openvswitch/flow.c:933
ovs_flow_key_extract+0x419/0xa70
ovs_vport_receive+0x222/0x390
netdev_frame_hook+0x3e0/0x630
tun_get_user+0x2d0c/0x38e0
Fixed by calling check_header() in key_extract() before accessing the
Ethernet header.
Fixes: 217ac77a3c ("openvswitch: allow L3 netdev ports")
Reported-by: AutonomousCodeSecurity@microsoft.com
Reviewed-by: Eelco Chaudron <echaudro@redhat.com>
Signed-off-by: Cen Zhang (Microsoft) <blbllhy@gmail.com>
Reviewed-by: Ilya Maximets <i.maximets@ovn.org>
Link: https://patch.msgid.link/20260730222006.118652-1-blbllhy@gmail.com
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
packet_set_ring() updates the RX ring state under sk_receive_queue.lock,
but used to publish the tpacket receive mode through po->prot_hook.func
after releasing that lock. packet_poll() and packet_recvmsg() can then
run the pressure clearing path after the ring has been cleared while
still seeing tpacket_rcv, causing __packet_rcv_has_room() to dereference
stale or NULL ring storage.
Move the existing receive hook assignment into the same
sk_receive_queue.lock section as the ring state update. Keep the
assignment otherwise unchanged, including on TX ring reconfiguration, to
avoid adding behavior changes that are not required for the fix.
Serialize packet_recvmsg() pressure clearing with the same queue lock
only after PACKET_SOCK_PRESSURE has been observed. If the flag is clear
and the socket has moved away from tpacket_rcv, packet_set_ring() has
already detached the socket and waited for synchronize_net(), so no new
packet input can set the flag again.
packet_poll() already holds sk_receive_queue.lock, so it uses the new
unlocked helper directly.
Fixes: 2ccdbaa6d5 ("packet: rollover lock contention avoidance")
Cc: stable@vger.kernel.org
Reported-by: Vega <vega@nebusec.ai>
Assisted-by: Codex:gpt-5.4
Signed-off-by: Zihan Xi <zihanx@nebusec.ai>
Link: https://patch.msgid.link/f90b5688311fa278d1361ea8c6be0bf25967d591.1785247446.git.zihanx@nebusec.ai
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
The sch_cake ACK filter parses packets to find the TCP header and filter
duplicated ACKs if the flow is backlogged. The parsing code contains a
WARN_ON(1) which can be triggered by a malformed IP header in certain
cases. Depending on the system configuration, this leads either to
either spamming dmesg with warnings, or a panic if panic_on_warn is set.
The code already correctly skips the offending packet in the branch that
triggers the warning, so the WARN_ON itself doesn't really serve any
purpose. So just drop it altogether to avoid the inconvenient side
effects.
Fixes: 8b7138814f ("sch_cake: Add optional ACK filter")
Reported-by: Zhiling Zou <zhilinz@nebusec.ai>
Reported-by: Ren Wei <enjou1224z@gmail.com>
Signed-off-by: Toke Høiland-Jørgensen <toke@toke.dk>
Link: https://patch.msgid.link/20260729191417.45665-1-toke@toke.dk
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
__skb_udp_tunnel_segment() gets the UDP header before ensuring the
tunnel header is in the skb head. If the pull reallocates skb->head,
the saved UDP header pointer is no longer valid.
Get the UDP header after the pull to avoid a potential use-after-free.
Fixes: dbef491ebe ("udp: Use uh->len instead of skb->len to compute checksum in segmentation")
Signed-off-by: Xuanqiang Luo <luoxuanqiang@kylinos.cn>
Reviewed-by: Antoine Tenart <atenart@kernel.org>
Link: https://patch.msgid.link/20260730093554.68127-1-xuanqiang.luo@linux.dev
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
qeth_snmp_command() and qeth_l3_arp_query() allocate a buffer sized by
a user-supplied length (udata_len) without checking a lower bound, then
set udata_offset to a fixed non-zero value and pass both to a reply
callback. The callback bounds-checks the copy with
if ((udata_len - udata_offset) < len)
Both fields are u32, so a udata_len smaller than udata_offset makes the
subtraction wrap and the check pass, and the following memcpy() writes
past the allocation. A udata_len of 0 also yields ZERO_SIZE_PTR from
kzalloc(), which the existing NULL check does not catch.
Reject buffers smaller than udata_offset before allocating, so the
callback subtraction can no longer underflow.
Fixes: 4a71df5004 ("qeth: new qeth device driver")
Cc: stable@vger.kernel.org
Reviewed-by: Alexandra Winter <wintera@linux.ibm.com>
Signed-off-by: Hidayath Khan <hidayath@linux.ibm.com>
Reviewed-by: Joe Damato <joe@dama.to>
Link: https://patch.msgid.link/20260730142216.218309-1-hidayath@linux.ibm.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
fib_nlmsg_size() still estimates nexthop space as if every gateway is
encoded as an IPv4 RTA_GATEWAY attribute. IPv4 routes can also carry an
IPv6 gateway, which fib_nexthop_info() dumps as RTA_VIA.
As a result, route notifications can allocate an skb that is too small.
fib_dump_info() then fails with -EMSGSIZE and rtmsg_fib() hits the
WARN_ON() that marks such failures as a fib_nlmsg_size() bug. With
panic_on_warn set, this becomes a kernel panic.
Mirror the actual nexthop dump layout in fib_nlmsg_size(): account for
IPv6 nexthop gateways dumped as RTA_VIA, for the no-header rtnexthop
layout used inside RTA_MULTIPATH, and for RTA_FLOW only when it is
actually present.
Fixes: d15662682d ("ipv4: Allow ipv6 gateway with ipv4 routes")
Cc: stable@vger.kernel.org
Reported-by: Vega <vega@nebusec.ai>
Signed-off-by: Zihan Xi <zihanx@nebusec.ai>
Reviewed-by: Ido Schimmel <idosch@nvidia.com>
Link: https://patch.msgid.link/6f53fa797fcaeb26966432ed7ae9bb87c4961f37.1785411220.git.zihanx@nebusec.ai
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
smc_rx_splice() passes pages to splice_to_pipe() before taking the
references that cover the lifetime of each splice entry. In the
VM-backed RMB path, splice_to_pipe() may drop unqueued entries through
smc_rx_spd_release(), while queued entries are released later via the
pipe buffer callback.
The old post-splice accounting also derives the number of queued VM pages
from an offset mutated while building the descriptor, and a multi-page
splice pairs one sock_hold() with multiple sock_put() calls.
Take the page and socket references for every candidate entry before
splice_to_pipe(), and drop the matching private state, page reference,
and socket reference from smc_rx_spd_release() for entries that never
get queued. This fixes a refcount imbalance that can underflow page
refcounts and trigger a use-after-free.
Fixes: 9014db202c ("smc: add support for splice()")
Cc: stable@vger.kernel.org
Reported-by: Vega <vega@nebusec.ai>
Co-developed-by: Xiao Liu <lx24@stu.ynu.edu.cn>
Signed-off-by: Xiao Liu <lx24@stu.ynu.edu.cn>
Signed-off-by: Daming Li <d4n.for.sec@gmail.com>
Signed-off-by: Ren Wei <enjou1224z@gmail.com>
Reviewed-by: Dust Li <dust.li@linux.alibaba.com>
Reviewed-by: Sidraya Jayagond <sidraya@linux.ibm.com>
Link: https://patch.msgid.link/20260730145552.360287-2-enjou1224z@gmail.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
If allocation of auxr_dev fails during auxiliary device setup, the error
path calls auxiliary_device_uninit(), which eventually invokes
bnge_aux_dev_release().
The release callback unconditionally dereferences aux_priv->auxr_dev->pdev
to retrieve the parent bnge_dev. Since auxr_dev has not yet been allocated
on this failure path, the dereference results in a NULL pointer exception
Retrieve the parent bnge_dev from the auxiliary device's parent instead of
auxr_dev, and free auxr_dev only when it was successfully allocated. This
allows the release callback to correctly clean up partially initialized
auxiliary devices.
Fixes: 8ac050ec3b ("bng_en: Add RoCE aux device support")
Signed-off-by: Alok Tiwari <alok.a.tiwari@oracle.com>
Reviewed-by: Bhargava Marreddy <bhargava.marreddy@broadcom.com>
Link: https://patch.msgid.link/20260731192301.1427645-1-alok.a.tiwari@oracle.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
When the interface has NETIF_F_SG enabled and skb_linearize() fails in
ax88179_tx_fixup(), the function returns NULL without freeing the skb.
usbnet_start_xmit() treats a NULL return from tx_fixup() as a drop
(info->flags does not set FLAG_MULTI_PACKET for this driver), jumping
to the "drop" label where it does `if (skb) dev_kfree_skb_any(skb)`.
Because tx_fixup() returned NULL, the local skb variable in
usbnet_start_xmit() is NULL, so the original skb is never freed — a
memory leak on every TX frame whose linearization fails (i.e. under
memory pressure).
Free the skb before returning, matching the error handling already used
for the pskb_expand_head() failure path in the same function.
Fixes: 16b1c4e01c ("net: usb: ax88179_178a: add TSO feature")
Cc: stable@vger.kernel.org
Signed-off-by: Yi Cong <yicong@kylinos.cn>
Link: https://patch.msgid.link/20260729030436.3420477-1-cong.yi@linux.dev
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Stanislav Fomichev says:
====================
xsk: harden TX metadata validation against races
Cen Zhang reported a KASAN out-of-bounds read when AF_XDP is configured
with a TX metadata area smaller than struct xsk_tx_metadata. The metadata
is also shared with user space, so reading its flags more than once can
produce inconsistent validation and processing decisions.
Require enough space for the flags and one request field, validate the
launch-time field against the configured metadata length, and use one
snapshot of the flags while processing each request. Carry the validated
decision through completion handling so later user-space changes cannot
enable an unrequested completion timestamp.
Reported-by: AutonomousCodeSecurity@microsoft.com
Reported-by: Cen Zhang (Microsoft) <blbllhy@gmail.com>
Link: https://lore.kernel.org/netdev/20260720155210.34229-1-blbllhy@gmail.com/
====================
Link: https://patch.msgid.link/20260727161959.885642-1-sdf@fomichev.me
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
The zero-copy path validates TX metadata while obtaining the descriptor
context, then reads it again later when preparing the hardware request.
User space can change the metadata between those operations and bypass the
original validation.
Validate the metadata in xsk_tx_metadata_request() and use the resulting
flags snapshot for every feature check. Read request fields once so all
zero-copy drivers process only values observed after successful
validation.
Fixes: ca4419f15a ("xsk: Add launch time hardware offload support to XDP Tx metadata")
Cc: Cen Zhang (Microsoft) <blbllhy@gmail.com>
Signed-off-by: Stanislav Fomichev <sdf@fomichev.me>
Link: https://patch.msgid.link/20260727161959.885642-7-sdf@fomichev.me
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
xsk_tx_metadata_request() must validate metadata with
xsk_buff_valid_tx_metadata(), which is defined in xdp_sock_drv.h. Move the
helper there before adding that dependency. All callers already include
the destination header, so this has no functional effect.
Fixes: ca4419f15a ("xsk: Add launch time hardware offload support to XDP Tx metadata")
Cc: Cen Zhang (Microsoft) <blbllhy@gmail.com>
Signed-off-by: Stanislav Fomichev <sdf@fomichev.me>
Link: https://patch.msgid.link/20260727161959.885642-6-sdf@fomichev.me
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Launch-time metadata extends beyond the first 16 bytes of struct
xsk_tx_metadata. Reject the request when the registered metadata area does
not contain the complete field.
Snapshot the validated flags for the generic transmit path and use that
snapshot for request and completion processing, avoiding inconsistent
decisions if user space changes the flags concurrently.
Note that only xsk_skb_metadata is properly using the flags,
__xsk_buff_get_metadata ignores them. Next commits address that.
Fixes: ca4419f15a ("xsk: Add launch time hardware offload support to XDP Tx metadata")
Cc: Cen Zhang (Microsoft) <blbllhy@gmail.com>
Signed-off-by: Stanislav Fomichev <sdf@fomichev.me>
Link: https://patch.msgid.link/20260727161959.885642-5-sdf@fomichev.me
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
User space can change metadata flags after request processing. Rereading
them during completion can therefore make the kernel write a timestamp
that was not requested when the packet was submitted.
Clear the metadata pointer during request processing unless timestamp
completion is requested. Completion handling can then use the pointer
itself instead of rereading the flags.
On the mlx5 multi-packet WQE path metadata is evaluated per batch:
xsk_tx_metadata_request() runs only for the descriptor that starts a
session, just like the checksum offload that is applied once through the
shared WQE. Only that descriptor's pointer is reset, so completion
handling can record a timestamp for the other descriptors of the session
regardless of their own XDP_TXMD_FLAGS_TIMESTAMP bit. The write stays
inside the metadata area; the single-WQE, other zero-copy, and generic
paths reset the pointer per descriptor and are unaffected.
Fixes: ca4419f15a ("xsk: Add launch time hardware offload support to XDP Tx metadata")
Cc: Cen Zhang (Microsoft) <blbllhy@gmail.com>
Signed-off-by: Stanislav Fomichev <sdf@fomichev.me>
Link: https://patch.msgid.link/20260727161959.885642-4-sdf@fomichev.me
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Completion handling needs to know whether a timestamp was requested when
the metadata was processed. Let xsk_tx_metadata_request() update the
caller's metadata pointer so that decision can be carried forward without
rereading user-controlled flags.
This only changes the interface; behavior remains unchanged.
Fixes: ca4419f15a ("xsk: Add launch time hardware offload support to XDP Tx metadata")
Cc: Cen Zhang (Microsoft) <blbllhy@gmail.com>
Signed-off-by: Stanislav Fomichev <sdf@fomichev.me>
Link: https://patch.msgid.link/20260727161959.885642-3-sdf@fomichev.me
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Weiming Shi says:
====================
vsock/virtio: fix worker access after virtqueue teardown
Virtio-vsock workers can remain queued while freeze deletes the
virtqueues. This series prevents workers delayed across freeze and
restore from retaining pointers to deleted queues, and prevents the RX
worker from refilling its queue after teardown.
====================
Link: https://patch.msgid.link/cover.1785352330.git.bestswngs@gmail.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Commit b917507e5a ("vsock/virtio: stop workers during the .remove()")
made the RX worker jump to its common exit when rx_run is clear. That
exit still refills the RX queue when the buffer count is low, so work
queued across virtio_vsock_vqs_del() can add buffers after the virtqueues
have been deleted.
BUG: KASAN: slab-use-after-free in virtqueue_add_sgs
Read of size 4 by task kworker/0:1
Workqueue: virtio_vsock virtio_transport_rx_work
Call Trace:
virtqueue_add_sgs (drivers/virtio/virtio_ring.c:2796)
virtio_vsock_rx_fill (net/vmw_vsock/virtio_transport.c:332)
virtio_transport_rx_work (net/vmw_vsock/virtio_transport.c:701)
process_one_work (kernel/workqueue.c:3314)
worker_thread (kernel/workqueue.c:3478)
kthread (kernel/kthread.c:436)
ret_from_fork (arch/x86/kernel/process.c:158)
ret_from_fork_asm (arch/x86/entry/entry_64.S:245)
...
Freed by task 141:
kfree (mm/slub.c:6566)
vp_del_vq (drivers/virtio/virtio_pci_common.c:259)
vp_del_vqs (drivers/virtio/virtio_pci_common.c:285)
virtio_vsock_freeze (net/vmw_vsock/virtio_transport.c:912)
virtio_device_freeze (drivers/virtio/virtio.c:658)
virtio_pci_freeze (drivers/virtio/virtio_pci_common.c:601)
pci_pm_freeze (drivers/pci/pci-driver.c:1098)
device_suspend (drivers/base/power/main.c:1968)
Kernel panic - not syncing: KASAN: panic_on_warn set ...
Jump to a no-refill exit when rx_run is clear, leaving the normal exit
to replenish a running queue.
Fixes: b917507e5a ("vsock/virtio: stop workers during the .remove()")
Cc: stable@vger.kernel.org
Reported-by: Xiang Mei <xmei5@asu.edu>
Link: https://lore.kernel.org/r/20260727035804.1860862-1-bestswngs@gmail.com
Suggested-by: Stefano Garzarella <sgarzare@redhat.com>
Signed-off-by: Weiming Shi <bestswngs@gmail.com>
Reviewed-by: Bobby Eshleman <bobbyeshleman@meta.com>
Link: https://patch.msgid.link/f9c8c1d64cad9d262f305d02ffe164c2f900fadf.1785352330.git.bestswngs@gmail.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
There is a small memory leak in bnxt_queue_mem_alloc:
when bnxt_alloc_rx_agg_bmap() succeeds
but bnxt_alloc_one_tpa_info() later fails,
the rx_agg_bmap allocated by bnxt_alloc_rx_agg_bmap()
is not freed in the fallthrough cleanup cases.
Free the rx_agg_bmap in the err_free_rx_agg_ring case
and initialize clone->rx_agg_bmap = NULL earlier in the function
to allow for safe fallthrough.
Fixes: bd649c5cc9 ("bnxt_en: handle tpa_info in queue API implementation")
Signed-off-by: Will Chen <will.chen.tty@gmail.com>
Reviewed-by: Joe Damato <joe@dama.to>
Reviewed-by: Michael Chan <michael.chan@broadcom.com>
Link: https://patch.msgid.link/20260729220132.1256924-1-will.chen.tty@gmail.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
rt6_route_rcv() validates the Route Information option (RFC 4191) length
against the prefix length, but both checks are off by one.
rinfo->length is the ND option length in units of 8 octets and it
*includes* the 8-byte option header, so an option carrying N bytes of
prefix has length == 1 + N/8. RFC 4191 section 2.3 requires length 3
when Prefix Length is greater than 64, and 2 or 3 when it is greater
than 0. The code accepts length >= 2 and length >= 1 respectively.
ipv6_addr_prefix() then copies prefix_len/8 bytes out of rinfo->prefix,
so a Router Advertisement with (prefix_len=128, length=2) or
(prefix_len=64, length=1) makes the kernel read up to 8 bytes past the
end of the option. Those bytes end up in the prefix of the route that
gets installed, so they are visible to userspace:
# RA with a Route Information option (prefix_len=128, length=2)
# followed by a source link-layer address option, 01 01 de ad be ef ca fe
$ ip -6 route show
2001:db8:dead:beef:101:dead:beef:cafe via fe80::1234 dev veth0 proto ra
^^^^^^^^^^^^^^^^^^ the next option, read out of bounds
When the Route Information option is the last one in the packet, those
eight bytes come from the skb tail room instead.
Reject the option lengths RFC 4191 does not allow.
Fixes: 70ceb4f539 ("[IPV6]: ROUTE: Add experimental support for Route Information Option in RA (RFC4191).")
Cc: stable@vger.kernel.org
Signed-off-by: Yuejie Shi <syjcnss@gmail.com>
Reviewed-by: Ido Schimmel <idosch@nvidia.com>
Link: https://patch.msgid.link/20260730035310.74584-1-syjcnss@gmail.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
__sctp_outq_flush_rtx() moves a gap-acked chunk onto another transport's
transmitted list without updating chunk->transport:
if (chunk->tsn_gap_acked) {
list_move_tail(&chunk->transmitted_list,
&transport->transmitted);
continue;
}
The chunk then sits on a live transport's list while chunk->transport still
names a different one. If that transport is removed - sctp_assoc_rm_peer()
from an ASCONF Delete-IP - sctp_transport_free() RCU-frees it and the chunk
is left with a dangling pointer. sctp_assoc_rm_peer() scrubs
peer->transmitted and asoc->outqueue.out_chunk_list, but the chunk is on
neither.
The pointer is not followed while tsn_gap_acked is set. A SACK that
reneges on the TSN clears the flag, and the next SACK reaches
tchunk->transport->flight_size -= sctp_data_size(tchunk);
inside the freed transport. KASAN reports a slab-use-after-free read in
sctp_check_transmitted(), freed from sctp_assoc_rm_peer(). Both the
removal and the SACKs come from the association peer.
Set chunk->transport at the move. The ordinary resend path needs nothing:
it reaches its list_move_tail() only after sctp_packet_append_chunk()
returned SCTP_XMIT_OK, and __sctp_packet_append_chunk() has rebound the
chunk by then.
Discovered by XBOW, triaged by Baul Lee <baul.lee@xbow.com>
Fixes: 1da177e4c3 ("Linux-2.6.12-rc2")
Cc: stable@vger.kernel.org
Signed-off-by: Baul Lee <baul.lee@xbow.com>
Acked-by: Xin Long <lucien.xin@gmail.com>
Link: https://patch.msgid.link/20260729160028.54546-1-baul.lee@xbow.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Commit f5da7c4518 ("tcp: adjust rcvq_space after updating scaling
ratio") replaced the direct window_clamp update in tcp_measure_rcv_mss()
with a call to tcp_set_window_clamp(), a helper that implements the
TCP_WINDOW_CLAMP setsockopt. As a side effect, the helper also shrinks
rcv_ssthresh via __tcp_adjust_rcv_ssthresh().
As a result, each scaling_ratio decrease detected by
tcp_measure_rcv_mss() also cuts rcv_ssthresh. Elsewhere in TCP,
rcv_ssthresh is usually cut under memory pressure and grows via
tcp_grow_window().
Flows whose segment sizes vary keep scaling_ratio oscillating, which
leads to an unstable rcv_ssthresh: a dip of rcv_ssthresh only recovers
via tcp_grow_window(), keeping the advertised window at a relatively
low level even after the ratio itself has recovered, and can even stall
the sender.
Observed on a customer's proxy gateway after upgrading from kernel 6.1
to 6.12: in the worst case, rcv_ssthresh was cut in half by a
scaling_ratio dip. P99 latency jumped from <10ms on 6.1 to ~100ms on
6.12, and almost returned to the 6.1 level with this patch applied.
Restore the plain WRITE_ONCE() update of window_clamp, as introduced
in commit a2cbb16039 ("tcp: Update window clamping condition"), and
keep the rcvq_space.space adjustment. Now rcv_ssthresh is decoupled from
scaling_ratio changes in tcp_measure_rcv_mss().
Fixes: f5da7c4518 ("tcp: adjust rcvq_space after updating scaling ratio")
Signed-off-by: Nathan Gao <zcgao@amazon.com>
Link: https://patch.msgid.link/20260725030806.28135-1-zcgao@amazon.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
When host is powered off, firmware does not send vhca_state event
for every probed host SF on the DPU because it may have deployed
thousands of SFs to the host. Instead it sends a function changed
event. Currently, only VFs handle this event. This commit extends
support to SFs.
When DPU user deactivates[1] SFs, mlx5 expects vhca_state event
and leaves the SF in dangling state[2].
When DPU user deletes[3] SFs, mlx5 also expects vhca_state event
and destroys the SF resources[4].
Fix it by changing SF to the right state and freeing SF resources
when the function changed event is received.
When this event is received, driver checks all SF states.
- If state is in_use, change it to active.
- If state is teardown_request, change it to allocated.
And SF hardware table entry is freed if it is pending for delete.
[1]
# devlink port function set en3f0c1pf0sf0 state inactive
[2]
# devlink port function set en3f0c1pf0sf0 state active
Error: mlx5_core: SF is inactivated but it is still attached.
kernel answers: Device or resource busy
[3]
# devlink port show
pci/0000:03:00.0/229376: type eth netdev en3f0c1pf0sf0 \
flavour pcisf controller 1 pfnum 0 sfnum 0 splittable false
function:
hw_addr 00:00:00:00:00:00 state active opstate attached \
roce enable trust off max_uc_macs 4096 max_io_eqs 8
# devlink port del en3f0c1pf0sf0
[4]
# devlink port add pci/0000:03:00.0 flavour pcisf pfnum 0 sfnum 0 \
controller 1
Error: mlx5_core: SF already exist. Choose different sfnum.
kernel answers: File exists
Fixes: 6a32732174 ("net/mlx5: SF, Port function state change support")
Signed-off-by: Chris Mi <cmi@nvidia.com>
Reviewed-by: Shay Drori <shayd@nvidia.com>
Signed-off-by: Tariq Toukan <tariqt@nvidia.com>
Link: https://patch.msgid.link/20260729071622.2423270-1-tariqt@nvidia.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
devlink_nl_reload_doit() calls devlink_netns_get(), which returns a net
with a held reference. When the requested namespace differs from the
current one and the reload action is not DRIVER_REINIT, the function
returns -EOPNOTSUPP without releasing the reference. Add the missing
put_net() on this error path.
Fixes: 2edd925704 ("devlink: don't allow to change net namespace for FW_ACTIVATE reload action")
Signed-off-by: Or Har-Toov <ohartoov@nvidia.com>
Reviewed-by: Jiri Pirko <jiri@nvidia.com>
Signed-off-by: Tariq Toukan <tariqt@nvidia.com>
Reviewed-by: Antoine Tenart <atenart@kernel.org>
Link: https://patch.msgid.link/20260729080600.2427721-1-tariqt@nvidia.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
hix5hd2_dev_remove() calls netif_napi_del() before unregister_netdev().
This is not needed because free_netdev() deletes all NAPI instances
attached to the net_device.
Remove the redundant call and let the networking core tear down the NAPI
instance during unregister_netdev(). The probe error path still keeps its
explicit netif_napi_del(), because the device has not been registered
there.
Fixes: 57c5bc9ad7 ("net: hisilicon: add hix5hd2 mac driver")
Signed-off-by: Jiawen Liu <1298662399@qq.com>
Link: https://patch.msgid.link/tencent_5FFD37A252B4FEA6A80AD25B17C8E904F005@qq.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
The route4 classifier maintains a 16-slot fastmap cache that stores raw
struct route4_filter pointers indexed by (id, iif). The reader
(route4_classify) populates this cache via route4_set_fastmap() for every
classified packet that hits a filter. The writer (route4_delete,
route4_change) clears the cache via route4_reset_fastmap() before
RCU-deferred kfree of the filter.
This creates a UAF race:
1. Reader walks the RCU-protected bucket chain, finds filter f
2. Writer unlinks f, calls route4_reset_fastmap(), then tcf_queue_work()
3. Reader calls route4_set_fastmap() and writes f into the cache
*after* the writer's reset, caching a pointer about to be freed
4. After the RCU grace period, kfree(f) executes
5. Next classified packet on the same (id, iif) tuple hits the stale
fastmap entry and reads f->res from freed memory
Reproduced with an mdelay(100) accelerator in route4_set_fastmap() and a
concurrent add/delete stress test (provided by both zdi and Santosh).
Both triggered KASAN slab-use-after-free reports in the route4 fastmap
paths.
Fix:
Introduce a per-filter boolean dying flag to suppress stale fastmap
republishing by in-flight readers.
Fixes: 1109c00547 ("net: sched: RCU cls_route")
Reported-by: zdi-disclosures@trendmicro.com
Reported-by: Santosh Kalluri <santosh.kalluri129@gmail.com>
Suggested-by: Paolo Abeni <pabeni@redhat.com>
Tested-by: Victor Nogueira <victor@mojatatu.com>
Tested-by: Santosh Kalluri <santosh.kalluri129@gmail.com>
Signed-off-by: Jamal Hadi Salim <jhs@mojatatu.com>
Link: https://patch.msgid.link/20260729094411.46257-1-jhs@mojatatu.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
inet_frag_create() arms the fragment queue timer before inserting the
queue into the fqdir rhashtable. If the namespace fragment timeout is
zero or negative, the timer can run before the queue is published.
The timer callback then marks the queue complete, tries to remove a node
that is not in the hash table yet, and drops the anticipated hash
reference. Creation can subsequently publish the completed queue without
restoring that reference, leaving a stale hash node after the caller drops
the remaining reference.
Publish the queue first and arm the timer while holding the queue lock.
This makes timer expiry wait until the queue is visible in the hash table,
so inet_frag_kill() can remove the node and balance the hash reference.
Fixes: 648700f76b ("inet: frags: use rhashtables for reassembly units")
Cc: stable@vger.kernel.org
Reported-by: Vega <vega@nebusec.ai>
Signed-off-by: Zhiling Zou <zhilinz@nebusec.ai>
Signed-off-by: Ren Wei <enjou1224z@gmail.com>
Link: https://patch.msgid.link/bf66785e7c0c139d7a1900e2f01faeeab344b960.1784948849.git.zhilinz@nebusec.ai
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
br_mrp_alloc_test_skb() builds MRP test frames on an skb from
dev_alloc_skb(), which does not clear the linear data area. On the MRA
ring-role branch the sub-option TLV header is appended with
sub_tlv = skb_put(skb, sizeof(*sub_tlv));
sub_tlv->type = BR_MRP_SUB_TLV_HEADER_TEST_AUTO_MGR;
so sub_tlv->length is never written, and the two trailing alignment bytes
are appended with a bare skb_put() that does not clear them either. The
neighbouring oui and sub_opt regions are explicitly zeroed, so three
uninitialised bytes are left in every MRA MRP_Test frame that goes out.
Put the sub-option TLV header and the alignment padding in a single
skb_put_zero(), which clears both. The AUTO_MGR sub-TLV carries no
payload, so the zeroed length field is already the value it should have.
Fixes: f7458934b0 ("net: bridge: mrp: Update the Test frames for MRA")
Suggested-by: Nikolay Aleksandrov <razor@blackwall.org>
Cc: stable@vger.kernel.org
Signed-off-by: Baul Lee <baul.lee@xbow.com>
Acked-by: Nikolay Aleksandrov <razor@blackwall.org>
Link: https://patch.msgid.link/20260729131941.10254-1-baul.lee@xbow.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
The SMC_LLC_CONFIRM_LINK / SMC_LLC_ADD_LINK_CONT branch in
smc_llc_event_handler() stores an incoming qentry into the local LLC flow
without first checking whether a qentry is already pending. If a malicious or
buggy peer sends a second CONFIRM_LINK or ADD_LINK_CONT request while a flow is
active and flow->qentry is already set, smc_llc_flow_qentry_set() overwrites the
pointer without freeing the previous allocation, leaking one kmalloc-96 object
per spurious message.
The sibling SMC_LLC_DELETE_LINK branch already has the correct !flow->qentry
guard. Apply the same guard to the CONFIRM_LINK/ADD_LINK_CONT branch so that a
duplicate message when qentry is already occupied falls through to break and is
freed by the kfree(qentry) at the out: label, rather than silently leaking the
existing allocation.
The response direction (smc_llc_rx_response()) is unaffected: it already guards
with flow->qentry at the equivalent site and drops duplicate responses
correctly.
Fixes: 0fb0b02bd6 ("net/smc: adapt SMC client code to use the LLC flow")
Signed-off-by: Mahanta Jambigi <mjambigi@linux.ibm.com>
Reviewed-by: Hidayath Khan <hidayath@linux.ibm.com>
Reviewed-by: Sidraya Jayagond <sidraya@linux.ibm.com>
Reviewed-by: Dust Li <dust.li@linux.alibaba.com>
Link: https://patch.msgid.link/20260729130153.970800-1-mjambigi@linux.ibm.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
The x25 timers are armed with mod_timer() and cancelled with
timer_delete(), so a pending timer holds no reference on the socket and a
cancel does not wait for a callback already running on another CPU.
x25_heartbeat_expiry() also rearms unconditionally, so it can reinstall
sk->sk_timer after __x25_destroy_socket() has passed its cancel point.
The following __sock_put() frees the socket while the timer is still
queued, and the next expiry uses freed memory. KASAN reports a
slab-use-after-free on the kmalloc-2k object freed by close().
timer_delete_sync() cannot be used here: x25_heartbeat_expiry() and
x25_timer_expiry() both reach the cancels from inside the timer they
would wait on, through __x25_destroy_socket() and x25_disconnect().
Arm the timers with sk_reset_timer() and cancel them with sk_stop_timer()
so that an armed timer owns a reference, and release it in both expiry
handlers. Rearm the heartbeat only while sk_hashed(sk) is still true,
since __x25_destroy_socket() unlinks the socket before dropping it. Arm
the deferred destroy timer the same way and drop its reference in
x25_destroy_timer().
Reproduced on net with KASAN, with the heartbeat period shortened so the
window recurs. With this patch the reproducer no longer triggers a
report and /proc/net/x25 drains.
Discovered by XBOW, triaged by Baul Lee <baul.lee@xbow.com>
Fixes: 1da177e4c3 ("Linux-2.6.12-rc2")
Cc: stable@vger.kernel.org
Signed-off-by: Baul Lee <baul.lee@xbow.com>
Link: https://patch.msgid.link/20260726220342.47245-1-baul.lee@xbow.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
pdsc_check_pci_health() queues pci_reset_work when it sees a broken PCI
connection, and nothing cancels it. When the PCI core starts AER
recovery, pdsc_pci_error_detected() runs pdsc_reset_prepare() and
recovers the device, but a pci_reset_work queued just before is left
pending. If it runs after recovery released the device lock, it resets a
device the driver now considers healthy, bouncing the link for no reason.
Cancel pci_reset_work in pdsc_pci_error_detected() after
pdsc_reset_prepare(), which has already stopped the health thread so it
cannot requeue the work. cancel_work_sync() is safe under the device
lock here because pdsc_pci_reset_thread() uses pci_try_reset_function(),
which returns instead of blocking on the lock. Only PFs initialize
pci_reset_work, so guard the cancel with !is_virtfn.
Fixes: 81665adf25 ("pds_core: Fix pdsc_check_pci_health function to use work thread")
Reported-by: sashiko-bot <sashiko-bot@kernel.org>
Closes: https://sashiko.dev/#/patchset/20260714180223.1642792-2-nikhil.rao%40amd.com?part=1
Signed-off-by: Nikhil P. Rao <nikhil.rao@amd.com>
Reviewed-by: Jacob Keller <jacob.e.keller@intel.com>
Link: https://patch.msgid.link/20260727170030.361116-1-nikhil.rao@amd.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Commit d9407ff118 ("pds_core: Prevent health thread from running
during reset/remove") stops the health thread with cancel_work_sync()
before a reset, but a devcmd timeout during pdsc_fw_down() re-queues
health_work, so pdsc_health_thread() runs again mid-reset and double
allocates the core DMA queues via pdsc_fw_up().
Only the reset path is affected: on remove PDSC_S_STOPPING_DRIVER gates
the health thread and the workqueue is destroyed.
Use disable_work_sync() to cancel health_work and block further
queue_work() on it, and enable_work() in pdsc_restart_health_thread() to
re-allow it after the reset.
disable_work_sync() keeps a disable depth, so every disable must be
matched by one enable. pdsc_reset_prepare() stops the health thread and
pdsc_reset_done() restarts it. On the AER path pdsc_pci_error_detected()
calls pdsc_reset_prepare(), then pdsc_pci_error_resume() re-inits via
pci_reset_function_locked() (pds_core has no .slot_reset handler), which
runs the pair again - stopping the thread twice but restarting it once.
Gate the disable and enable on a health_stopped flag so each fires at
most once per stopped/running transition.
Fixes: d9407ff118 ("pds_core: Prevent health thread from running during reset/remove")
Reported-by: sashiko-bot <sashiko-bot@kernel.org>
Closes: https://sashiko.dev/#/patchset/20260629200358.2626129-1-nikhil.rao%40amd.com?part=2
Signed-off-by: Nikhil P. Rao <nikhil.rao@amd.com>
Reviewed-by: Jacob Keller <jacob.e.keller@intel.com>
Link: https://patch.msgid.link/20260727164548.359562-1-nikhil.rao@amd.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
In case __mlx5e_add_fdb_flow() fails in lower levels, the flow is
deleted via mlx5e_tc_del_flow(), and mlx5e_tc_del_flow() is acquiring
ESW devcom lock without condition. In addition, in case of peer_flow,
__mlx5e_add_fdb_flow() is called while holding ESW devcom comp lock.
This results in an AA deadlock.
To fix this, introduce a new PEER flag that is set on flows created as
peer flows (the duplicate flows on peer devices), and check it in
mlx5e_tc_del_flow() before acquiring ESW devcom lock.
Lockdep splat:
============================================
WARNING: possible recursive locking detected
============================================
Possible unsafe locking scenario:
CPU0
----
lock(&comp->lock_key#2);
lock(&comp->lock_key#2);
*** DEADLOCK ***
Call Trace:
<TASK>
dump_stack_lvl+0x69/0xa0
print_deadlock_bug.cold+0xbd/0xca
__lock_acquire+0x1671/0x2ec0
lock_acquire+0x10e/0x2e0
down_read+0x95/0x430
mlx5_devcom_for_each_peer_begin+0x4e/0xe0 [mlx5_core]
mlx5e_tc_del_flow+0x11d/0xa70 [mlx5_core]
mlx5e_flow_put+0x99/0x100 [mlx5_core]
__mlx5e_add_fdb_flow+0x409/0xf00 [mlx5_core]
mlx5e_configure_flower+0x2a86/0x4100 [mlx5_core]
mlx5e_rep_setup_tc_cls_flower+0x12f/0x1b0 [mlx5_core]
mlx5e_rep_setup_tc_cb+0x153/0x750 [mlx5_core]
tc_setup_cb_add+0x1dc/0x470
fl_change+0x2f4d/0x626d [cls_flower]
tc_new_tfilter+0x79b/0x2310
rtnetlink_rcv_msg+0x778/0xad0
do_syscall_64+0x70/0x960
entry_SYSCALL_64_after_hwframe+0x4b/0x53
</TASK>
Fixes: 04de7dda73 ("net/mlx5e: Infrastructure for duplicated offloading of TC flows")
Signed-off-by: Shay Drory <shayd@nvidia.com>
Reviewed-by: Cosmin Ratiu <cratiu@nvidia.com>
Signed-off-by: Tariq Toukan <tariqt@nvidia.com>
Link: https://patch.msgid.link/20260728044338.2271143-1-tariqt@nvidia.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
enic_remove() cancels the reset and change_mtu_work items but does not
cancel tx_hang_reset. A TX timeout that fires while the device is being
removed can schedule enic_tx_hang_reset() so that it runs after
free_netdev(), resulting in a use-after-free.
cancel_work_sync() alone is not sufficient here: the still-live watchdog
and notify paths can re-schedule these work items in the window between
the cancel and unregister_netdev(). Use disable_work_sync(), which
cancels the work and blocks any subsequent schedule_work() from
requeuing it, and apply it to the reset and change_mtu_work items as
well so the same requeue race is closed for all teardown work.
Fixes: 937317c7c1 ("enic: do hang reset only in case of tx timeout")
Signed-off-by: Satish Kharat <satishkh@cisco.com>
Link: https://patch.msgid.link/20260728062730.2394873-1-satishkh@cisco.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
packet_parse_headers() resets the MAC header only for a SOCK_RAW frame
whose socket did not bind a protocol. A protocol-bound SOCK_RAW socket,
any SOCK_DGRAM frame, and the legacy SOCK_PACKET path therefore leave
skb->mac_header unset here.
For frames sent via __dev_queue_xmit() this is harmless: it resets the
MAC header unconditionally. But the packet-socket PACKET_QDISC_BYPASS
path uses dev_direct_xmit(), which does not, so the frame reaches
ndo_start_xmit() with the MAC header unset. A driver that reads
eth_hdr(skb) on transmit then dereferences skb->head + (u16)~0, an
out-of-bounds access ~64 KiB past the head -- the same class fixed for
one consumer in commit f5089008f9 ("macsec: do not read an unset MAC
header in macsec_encrypt()").
packet_parse_headers() runs only on the transmit path, where skb->data
points at the start of the L2 header for every packet-socket type
regardless of its length: SOCK_RAW and SOCK_PACKET carry a user-supplied
header and SOCK_DGRAM has one built by dev_hard_header(). Reset the MAC
header unconditionally, mirroring __dev_queue_xmit(), so the frame is
anchored on the bypass path too.
Found by 0sec (https://0sec.ai) using automated source analysis;
verified against source and matched to the macsec KASAN report in
f5089008f9. Compile-tested.
Fixes: 75c65772c3 ("net/packet: Ask driver for protocol if not provided by user")
Cc: stable@vger.kernel.org
Signed-off-by: Doruk Tan Ozturk <doruk@0sec.ai>
Reviewed-by: Willem de Bruijn <willemb@google.com>
Link: https://patch.msgid.link/20260724144015.63219-1-doruk@0sec.ai
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
bond_alb_monitor() reads primary_is_promisc under RCU, then drops RCU and
takes RTNL via rtnl_trylock() before undoing the promiscuity it set on the
active slave. In that window the active slave can change under RTNL
(RTM_DELLINK -> __bond_release_one() -> bond_alb_handle_active_change()),
which already drops the promiscuity and clears primary_is_promisc. The
monitor still acts on the stale decision: if the slave was removed with no
failover, curr_active_slave is now NULL and the deref faults; if it failed
over, the stale dev_set_promiscuity(-1) underflows the new slave's
promiscuity counter and pins it in IFF_PROMISC.
Oops: general protection fault, probably for non-canonical address ...
KASAN: null-ptr-deref in range [0x0000000000000000-0x0000000000000007]
Workqueue: b42 bond_alb_monitor
RIP: 0010:bond_alb_monitor (drivers/net/bonding/bond_alb.c:1600)
process_one_work (kernel/workqueue.c:3322)
worker_thread (kernel/workqueue.c:3486)
kthread (kernel/kthread.c:436)
ret_from_fork (arch/x86/kernel/process.c:158)
Kernel panic - not syncing: Fatal exception
Re-check primary_is_promisc (and curr_active_slave) after taking RTNL so
the monitor only undoes an increment it still owns. The other bonding
monitors already re-read state under RTNL in their commit phase
(bond_miimon_commit/bond_ab_arp_commit); bond_alb_monitor() was the only
one acting on the pre-trylock decision.
Fixes: d0e81b7e22 ("bonding: Acquire correct locks in alb for promisc change")
Reported-by: AutonomousCodeSecurity@microsoft.com
Signed-off-by: Xiang Mei (Microsoft) <xmei5@asu.edu>
Reviewed-by: Nikolay Aleksandrov <razor@blackwall.org>
Acked-by: Jay Vosburgh <jv@jvosburgh.net>
Link: https://patch.msgid.link/20260725233930.2957317-1-xmei5@asu.edu
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
This reverts commit a8065af334.
Per the USB4 spec, a Transmit Descriptor Ring with E2E flow control
disabled does not require any credits to be available before the Host
Interface Adapter Layer transmits a tunneled packet from it. Once E2E is
enabled on that ring the controller must first obtain end-to-end
credits.
The ASMedia ASM4242 USB4 host router (PCI 1b21:2425) never delivers
those credits. The controller does accept the configuration: reading the
ring OPTIONS register back right after tb_ring_start() returns exactly
what was written, including RING_FLAG_E2E_FLOW_CONTROL (bit 28) and the
E2E HopID field. No credit ever arrives though, so the Tx ring's
hardware consumer index never advances and the link carries no traffic
at all.
Measured on two hosts connected point to point, onboard ASM4242 on MSI
X870E and X870, v6.17, stock drivers/net/thunderbolt/main.c with only
this revert applied on top:
before: 100% packet loss to the peer; thunderbolt0 is up and the
XDomain handshake completes ("new host found"), but iperf3
fails with "No route to host" once the neighbour entry
expires
after: 0% packet loss, 0.28 ms RTT; iperf3 4.21 Gb/s one way and
5.17 Gb/s the other (5 runs each, stddev <= 0.02), 1
retransmit in 10 s
An instrumented build additionally showed a frozen-Tx-consumer watchdog
firing ~30k times in a 10 s window before this change.
Rx-side E2E is not touched by this revert, so peers that do return
credits keep receive-side flow control.
ASMedia does not look like an isolated case. The out-of-tree
thunderbolt-ibverbs project disables native E2E on AMD NHI by default,
noting that "Strix Halo has reproduced TX completion wedges with
multiple native E2E rings active" -- the same failure mode, on a
different vendor. Since the driver has no way to tell in advance which
host router returns the credits, going back to the previous behaviour
looks safer than adding a quirk per affected part; Tx-side E2E can be
reintroduced as an opt-in for controllers that are known to implement
the credit return.
Note that the reverted commit was not fixing a reported problem, it was
derived from the spec wording alone, so this revert is not expected to
regress a known workload. Cc'ing the original author in case there was
one.
Fixes: a8065af334 ("net: thunderbolt: Enable end-to-end flow control also in transmit")
Cc: zhangjianrong <zhangjianrong5@huawei.com>
Signed-off-by: Fan Ye <fy15309206903@gmail.com>
Acked-by: Mika Westerberg <mika.westerberg@linux.intel.com>
Link: https://patch.msgid.link/20260727123002.25225-1-fy15309206903@gmail.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
After setting a 100G optical port to 40G via ethtool and reloading
the driver, the port remains at 40G instead of reverting to the
firmware default speed of 100G.
The commit referenced in Fixes: added two overwrites in
hclge_init_ae_dev() for non-copper media, so that optical ports
connected to forced-mode remotes inherit the firmware-preset
autoneg and speed instead of the hardcoded defaults:
req_autoneg = mac.autoneg
req_speed = mac.speed (when autoneg disabled)
The autoneg overwrite keeps existing behavior:
hclge_set_autoneg_speed_dup() already uses mac.autoneg (not
req_autoneg) since it was introduced, so autoneg inheritance from
firmware was already in place. This part is kept.
The speed overwrite, however, introduces the residue: mac.speed
reflects whatever was last programmed into the MAC, and after unload
firmware does not restore the MAC speed to the flash default. So if
the user changed speed via ethtool in a prior load, mac.speed still
carries that value on reload and req_speed inherits it.
Fix by dropping the req_speed overwrite only. req_speed keeps the
firmware default value set in hclge_configure() (cfg.default_speed),
so a reload reverts the speed to default, matching the expectation
that a driver reload resets link configuration.
Trade-off: on optical ports whose firmware default speed does not
match a forced-mode remote, reload now drops the link and the user
must re-apply ethtool configuration. This is acceptable: a driver
reload is expected to reset link configuration, not to inherit
runtime state from before unload. The autoneg inheritance is left
in place as established behavior; changing it is out of scope for
this patch and would itself be a user-perceivable behavior change.
Fixes: d9d349c4e8 ("net: hns3: differentiate autoneg default values between copper and fiber")
Signed-off-by: Jijie Shao <shaojijie@huawei.com>
Reviewed-by: Simon Horman <horms@kernel.org>
Link: https://patch.msgid.link/20260724093036.426631-1-shaojijie@huawei.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Pull networking fixes from Paolo Abeni:
"This is again larger than usual: the backlog accumulated in the past weeks
is not done yet. I'm not aware of any known pending regression.
Including fixes from netfilter, Bluetooth, WiFi and CAN.
Current release - regressions:
- bluetooth: remove unnecessary hci_conn_get in create_conn_sync
- can: isotp: fix timer drain order, wakeup handling and tx_gen
ordering
- eth:
- tun/vhost: revert avoid ptr_ring tail-drop when a qdisc is
present
Previous releases - regressions:
- core: do not send ICMP/NDISC Redirects when peer allocation fails
- ipv6: take nexthop lock for f6i_list walks in replace check and
notify
- wifi: fix an ath12k MLO regression impacting WCN7850/QCC2072.
- netfilter: nf_tables: make nft_object rhltable per table
- af_unix: fix listen() succeeding on sockets in the wrong state
- openvswitch: fix potential UAF on meter attach failure
- bluetooth:
- fix advertising data UAFs
- avoid deadlocks in iso_sock_timeout
- smc: fix socket use-after-free during link group termination
- dpll: use pin owner's dpll ref for pin-level attribute reporting
- eth:
- veth: convert frag_list skbs before running XDP
- ice: wait for reset completion in ice_resume()
- igc: remove napi_synchronize() in igc_down()
- vxlan: use pskb_network_may_pull() for transmit path header pulls
Previous releases - always broken:
- xsk: fix AF_XDP multi-buffer Tx descriptor reclaim
- psp: fix NULL genl_sock deref race with concurrent netns teardown
- netfilter: widen NAT rewrite delta to s32 in sip_help_tcp()
- can: peak_usb: fix double free of transfer buffer on URB submit error
- dibs: fix use-after-free of dmb_node in loopback attach/detach/unregister
- sctp: prevent peer transport count overflow
- dsa: mt7530: error out on failed reads in MT7531 PHY polling
- eth:
- idpf: bound interrupt-vector register fill to the allocated array"
* tag 'net-7.2-rc6' of git://git.kernel.org/pub/scm/linux/kernel/git/netdev/net: (156 commits)
qede: sync udp_tunnel ports outside qede_lock in the recovery path
net: openvswitch: fix potential UAF on meter attach failure
octeontx2-pf: Set correct sequence for carrier off and tx queue stop
net: libwx: fix FDIR ATR queue mismatch for software VLAN packets
net: dsa: realtek: use devm_mutex_init for l2_lock
net: dsa: realtek: use devm_mutex_init for vlan_lock
net: dsa: realtek: use devm_mutex_init for regmap lock
net: dsa: realtek: rtl8365mb: use devm_mutex_init for mib_lock
ptp: netc: fix potential interrupt storm caused by incorrect unbind order
net: mana: Return error code from mana_create_rxq()
net: openvswitch: fix skb leak on flow key update failure during ct
net: openvswitch: fix skb leak on flow key update failure during recirculation
net: stmmac: Fix E2E delay mechanism
net: dsa: mt7530: error out on failed reads in MT7531 PHY polling
net: dsa: mt7530: error out on failed reads in ATC/VTCR command polling
net: dsa: mt7530: check bus->read() errors in the MDIO regmap backend
Revert "tun/tap: add ptr_ring consume helper with netdev queue wakeup"
Revert "vhost-net: wake queue of tun/tap after ptr_ring consume"
Revert "ptr_ring: move free-space check into separate helper"
Revert "tun/tap & vhost-net: avoid ptr_ring tail-drop when a qdisc is present"
...
