Pablo Neira Ayuso says:
====================
Netfilter for net
This is v3, I kept back an ipset fix and another to tigthen the xtables
interface to reject invalid combinations with the NFPROTO_ARP family.
They need a bit more discussion. I fixed the issues reported by AI on
patch 9 (add #ifdef to access ct zone, update nf_conntrack_broadcast
and patch 10 (use better Fixes: tag). Thanks!
The following patchset contains Netfilter fixes for *net*.
Note that most bugs fixed here stem from 2.6 days, the large PR is not
due to an increase in regressions.
1) Fix incorrect reject of set updates with nf_tables pipapo set
avx2 backend. This comes with a regression test in patch 2.
From Florian Westphal.
2) nfnetlink_log needs to zero padding to prevent infoleak to userspace,
from Weiming Shi.
3) xtables ip6t_rt module never validated that addrnr length is within the
allowed array boundary. Reject bogus values. From Ren Wei.
4) Fix high memory usage in rbtree set backend that was unwanted side-effect
of the recently added binary search blob. From Pablo Neira Ayuso.
5) Patches 5 to 10, also from Pablo, address long-standing RCU safety bugs
in conntracks handling of expectations: We can never safely defer
a conntrack extension area without holding a reference. Yet expectation
handling does so in multiple places. Fix this by avoiding the need to
look into the master conntrack to begin with and by extending locked
sections in a few places.
11) Fix use of uninitialized rtp_addr in the sip conntrack helper,
also from Weiming Shi.
12) Add stricter netlink policy checks in ctnetlink, from David Carlier.
This avoids undefined behaviour when userspace provides huge wscale
value.
netfilter pull request 26-03-26
* tag 'nf-26-03-26' of git://git.kernel.org/pub/scm/linux/kernel/git/netfilter/nf:
netfilter: ctnetlink: use netlink policy range checks
netfilter: nf_conntrack_sip: fix use of uninitialized rtp_addr in process_sdp
netfilter: nf_conntrack_expect: skip expectations in other netns via proc
netfilter: nf_conntrack_expect: store netns and zone in expectation
netfilter: ctnetlink: ensure safe access to master conntrack
netfilter: nf_conntrack_expect: use expect->helper
netfilter: nf_conntrack_expect: honor expectation helper field
netfilter: nft_set_rbtree: revisit array resize logic
netfilter: ip6t_rt: reject oversized addrnr in rt_mt6_check()
netfilter: nfnetlink_log: fix uninitialized padding leak in NFULA_PAYLOAD
selftests: netfilter: nft_concat_range.sh: add check for flush+reload bug
netfilter: nft_set_pipapo_avx2: don't return non-matching entry on expiry
====================
Link: https://patch.msgid.link/20260326125153.685915-1-pablo@netfilter.org
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Tony Nguyen says:
====================
For ice:
Michal corrects call to alloc_etherdev_mqs() to provide maximum number
of queues supported rather than currently allocated number of queues.
Petr Oros fixes issues related to some ethtool operations in switchdev
mode.
For iavf:
Kohei Enju corrects number of reported queues for ethtool statistics to
absolute max as using current number could race and cause out-of-bounds
issues.
For idpf:
Josh NULLs cdev_info pointer after freeing to prevent possible subsequent
improper access. He also defers setting of refillqs value until after
allocation to prevent possible NULL pointer dereference.
* '100GbE' of git://git.kernel.org/pub/scm/linux/kernel/git/tnguy/net-queue:
idpf: only assign num refillqs if allocation was successful
idpf: clear stale cdev_info ptr
iavf: fix out-of-bounds writes in iavf_get_ethtool_stats()
ice: use ice_update_eth_stats() for representor stats
ice: fix inverted ready check for VF representors
ice: set max queues in alloc_etherdev_mqs()
====================
Link: https://patch.msgid.link/20260323205843.624704-1-anthony.l.nguyen@intel.com
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
There's a potential mismatch between the memory reserved for statistics
and the amount of memory written.
gem_get_sset_count() correctly computes the number of stats based on the
active queues, whereas gem_get_ethtool_stats() indiscriminately copies
data using the maximum number of queues, and in the case the number of
active queues is less than MACB_MAX_QUEUES, this results in a OOB write
as observed in the KASAN splat.
==================================================================
BUG: KASAN: vmalloc-out-of-bounds in gem_get_ethtool_stats+0x54/0x78
[macb]
Write of size 760 at addr ffff80008080b000 by task ethtool/1027
CPU: [...]
Tainted: [E]=UNSIGNED_MODULE
Hardware name: raspberrypi rpi/rpi, BIOS 2025.10 10/01/2025
Call trace:
show_stack+0x20/0x38 (C)
dump_stack_lvl+0x80/0xf8
print_report+0x384/0x5e0
kasan_report+0xa0/0xf0
kasan_check_range+0xe8/0x190
__asan_memcpy+0x54/0x98
gem_get_ethtool_stats+0x54/0x78 [macb
926c13f3af83b0c6fe64badb21ec87d5e93fcf65]
dev_ethtool+0x1220/0x38c0
dev_ioctl+0x4ac/0xca8
sock_do_ioctl+0x170/0x1d8
sock_ioctl+0x484/0x5d8
__arm64_sys_ioctl+0x12c/0x1b8
invoke_syscall+0xd4/0x258
el0_svc_common.constprop.0+0xb4/0x240
do_el0_svc+0x48/0x68
el0_svc+0x40/0xf8
el0t_64_sync_handler+0xa0/0xe8
el0t_64_sync+0x1b0/0x1b8
The buggy address belongs to a 1-page vmalloc region starting at
0xffff80008080b000 allocated at dev_ethtool+0x11f0/0x38c0
The buggy address belongs to the physical page:
page: refcount:1 mapcount:0 mapping:0000000000000000
index:0xffff00000a333000 pfn:0xa333
flags: 0x7fffc000000000(node=0|zone=0|lastcpupid=0x1ffff)
raw: 007fffc000000000 0000000000000000 dead000000000122 0000000000000000
raw: ffff00000a333000 0000000000000000 00000001ffffffff 0000000000000000
page dumped because: kasan: bad access detected
Memory state around the buggy address:
ffff80008080b080: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00
ffff80008080b100: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00
>ffff80008080b180: 00 00 00 00 00 f8 f8 f8 f8 f8 f8 f8 f8 f8 f8 f8
^
ffff80008080b200: f8 f8 f8 f8 f8 f8 f8 f8 f8 f8 f8 f8 f8 f8 f8 f8
ffff80008080b280: f8 f8 f8 f8 f8 f8 f8 f8 f8 f8 f8 f8 f8 f8 f8 f8
==================================================================
Fix it by making sure the copied size only considers the active number of
queues.
Fixes: 512286bbd4 ("net: macb: Added some queue statistics")
Signed-off-by: Paolo Valerio <pvalerio@redhat.com>
Reviewed-by: Nicolai Buchwitz <nb@tipi-net.de>
Link: https://patch.msgid.link/20260323191634.2185840-1-pvalerio@redhat.com
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Luiz Augusto von Dentz says:
====================
bluetooth pull request for net:
- L2CAP: Fix deadlock in l2cap_conn_del()
- L2CAP: Fix ERTM re-init and zero pdu_len infinite loop
- L2CAP: Fix send LE flow credits in ACL link
- btintel: serialize btintel_hw_error() with hci_req_sync_lock
- btusb: clamp SCO altsetting table indices
* tag 'for-net-2026-03-25' of git://git.kernel.org/pub/scm/linux/kernel/git/bluetooth/bluetooth:
Bluetooth: btusb: clamp SCO altsetting table indices
Bluetooth: L2CAP: Fix ERTM re-init and zero pdu_len infinite loop
Bluetooth: L2CAP: Fix deadlock in l2cap_conn_del()
Bluetooth: btintel: serialize btintel_hw_error() with hci_req_sync_lock
Bluetooth: L2CAP: Fix send LE flow credits in ACL link
====================
Link: https://patch.msgid.link/20260325194358.618892-1-luiz.dentz@gmail.com
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Replace manual range and mask validations with netlink policy
annotations in ctnetlink code paths, so that the netlink core rejects
invalid values early and can generate extack errors.
- CTA_PROTOINFO_TCP_STATE: reject values > TCP_CONNTRACK_SYN_SENT2 at
policy level, removing the manual >= TCP_CONNTRACK_MAX check.
- CTA_PROTOINFO_TCP_WSCALE_ORIGINAL/REPLY: reject values > TCP_MAX_WSCALE
(14). The normal TCP option parsing path already clamps to this value,
but the ctnetlink path accepted 0-255, causing undefined behavior when
used as a u32 shift count.
- CTA_FILTER_ORIG_FLAGS/REPLY_FLAGS: use NLA_POLICY_MASK with
CTA_FILTER_F_ALL, removing the manual mask checks.
- CTA_EXPECT_FLAGS: use NLA_POLICY_MASK with NF_CT_EXPECT_MASK, adding
a new mask define grouping all valid expect flags.
Extracted from a broader nf-next patch by Florian Westphal, scoped to
ctnetlink for the fixes tree.
Fixes: c8e2078cfe ("[NETFILTER]: ctnetlink: add support for internal tcp connection tracking flags handling")
Signed-off-by: David Carlier <devnexen@gmail.com>
Co-developed-by: Florian Westphal <fw@strlen.de>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
process_sdp() declares union nf_inet_addr rtp_addr on the stack and
passes it to the nf_nat_sip sdp_session hook after walking the SDP
media descriptions. However rtp_addr is only initialized inside the
media loop when a recognized media type with a non-zero port is found.
If the SDP body contains no m= lines, only inactive media sections
(m=audio 0 ...) or only unrecognized media types, rtp_addr is never
assigned. Despite that, the function still calls hooks->sdp_session()
with &rtp_addr, causing nf_nat_sdp_session() to format the stale stack
value as an IP address and rewrite the SDP session owner and connection
lines with it.
With CONFIG_INIT_STACK_ALL_ZERO (default on most distributions) this
results in the session-level o= and c= addresses being rewritten to
0.0.0.0 for inactive SDP sessions. Without stack auto-init the
rewritten address is whatever happened to be on the stack.
Fix this by pre-initializing rtp_addr from the session-level connection
address (caddr) when available, and tracking via a have_rtp_addr flag
whether any valid address was established. Skip the sdp_session hook
entirely when no valid address exists.
Fixes: 4ab9e64e5e ("[NETFILTER]: nf_nat_sip: split up SDP mangling")
Reported-by: Xiang Mei <xmei5@asu.edu>
Signed-off-by: Weiming Shi <bestswngs@gmail.com>
Signed-off-by: Florian Westphal <fw@strlen.de>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
Skip expectations that do not reside in this netns.
Similar to e77e6ff502 ("netfilter: conntrack: do not dump other netns's
conntrack entries via proc").
Fixes: 9b03f38d04 ("netfilter: netns nf_conntrack: per-netns expectations")
Signed-off-by: Florian Westphal <fw@strlen.de>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
__nf_ct_expect_find() and nf_ct_expect_find_get() are called under
rcu_read_lock() but they dereference the master conntrack via
exp->master.
Since the expectation does not hold a reference on the master conntrack,
this could be dying conntrack or different recycled conntrack than the
real master due to SLAB_TYPESAFE_RCU.
Store the netns, the master_tuple and the zone in struct
nf_conntrack_expect as a safety measure.
This patch is required by the follow up fix not to dump expectations
that do not belong to this netns.
Signed-off-by: Florian Westphal <fw@strlen.de>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
Holding reference on the expectation is not sufficient, the master
conntrack object can just go away, making exp->master invalid.
To access exp->master safely:
- Grab the nf_conntrack_expect_lock, this gets serialized with
clean_from_lists() which also holds this lock when the master
conntrack goes away.
- Hold reference on master conntrack via nf_conntrack_find_get().
Not so easy since the master tuple to look up for the master conntrack
is not available in the existing problematic paths.
This patch goes for extending the nf_conntrack_expect_lock section
to address this issue for simplicity, in the cases that are described
below this is just slightly extending the lock section.
The add expectation command already holds a reference to the master
conntrack from ctnetlink_create_expect().
However, the delete expectation command needs to grab the spinlock
before looking up for the expectation. Expand the existing spinlock
section to address this to cover the expectation lookup. Note that,
the nf_ct_expect_iterate_net() calls already grabs the spinlock while
iterating over the expectation table, which is correct.
The get expectation command needs to grab the spinlock to ensure master
conntrack does not go away. This also expands the existing spinlock
section to cover the expectation lookup too. I needed to move the
netlink skb allocation out of the spinlock to keep it GFP_KERNEL.
For the expectation events, the IPEXP_DESTROY event is already delivered
under the spinlock, just move the delivery of IPEXP_NEW under the
spinlock too because the master conntrack event cache is reached through
exp->master.
While at it, add lockdep notations to help identify what codepaths need
to grab the spinlock.
Signed-off-by: Florian Westphal <fw@strlen.de>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
Use expect->helper in ctnetlink and /proc to dump the helper name.
Using nfct_help() without holding a reference to the master conntrack
is unsafe.
Use exp->master->helper in ctnetlink path if userspace does not provide
an explicit helper when creating an expectation to retain the existing
behaviour. The ctnetlink expectation path holds the reference on the
master conntrack and nf_conntrack_expect lock and the nfnetlink glue
path refers to the master ct that is attached to the skb.
Reported-by: Hyunwoo Kim <imv4bel@gmail.com>
Signed-off-by: Florian Westphal <fw@strlen.de>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
The expectation helper field is mostly unused. As a result, the
netfilter codebase relies on accessing the helper through exp->master.
Always set on the expectation helper field so it can be used to reach
the helper.
nf_ct_expect_init() is called from packet path where the skb owns
the ct object, therefore accessing exp->master for the newly created
expectation is safe. This saves a lot of updates in all callsites
to pass the ct object as parameter to nf_ct_expect_init().
This is a preparation patches for follow up fixes.
Signed-off-by: Florian Westphal <fw@strlen.de>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
Chris Arges reports high memory consumption with thousands of
containers, this patch revisits the array allocation logic.
For anonymous sets, start by 16 slots (which takes 256 bytes on x86_64).
Expand it by x2 until threshold of 512 slots is reached, over that
threshold, expand it by x1.5.
For non-anonymous set, start by 1024 slots in the array (which takes 16
Kbytes initially on x86_64). Expand it by x1.5.
Use set->ndeact to subtract deactivated elements when calculating the
number of the slots in the array, otherwise the array size array gets
increased artifically. Add special case shrink logic to deal with flush
set too.
The shrink logic is skipped by anonymous sets.
Use check_add_overflow() to calculate the new array size.
Add a WARN_ON_ONCE check to make sure elements fit into the new array
size.
Reported-by: Chris Arges <carges@cloudflare.com>
Fixes: 7e43e0a114 ("netfilter: nft_set_rbtree: translate rbtree to array for binary search")
Signed-off-by: Florian Westphal <fw@strlen.de>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
Reject rt match rules whose addrnr exceeds IP6T_RT_HOPS.
rt_mt6() expects addrnr to stay within the bounds of rtinfo->addrs[].
Validate addrnr during rule installation so malformed rules are rejected
before the match logic can use an out-of-range value.
Fixes: 1da177e4c3 ("Linux-2.6.12-rc2")
Reported-by: Yifan Wu <yifanwucs@gmail.com>
Reported-by: Juefei Pu <tomapufckgml@gmail.com>
Co-developed-by: Yuan Tan <yuantan098@gmail.com>
Signed-off-by: Yuan Tan <yuantan098@gmail.com>
Suggested-by: Xin Liu <bird@lzu.edu.cn>
Tested-by: Yuhang Zheng <z1652074432@gmail.com>
Signed-off-by: Ren Wei <n05ec@lzu.edu.cn>
Signed-off-by: Florian Westphal <fw@strlen.de>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
__build_packet_message() manually constructs the NFULA_PAYLOAD netlink
attribute using skb_put() and skb_copy_bits(), bypassing the standard
nla_reserve()/nla_put() helpers. While nla_total_size(data_len) bytes
are allocated (including NLA alignment padding), only data_len bytes
of actual packet data are copied. The trailing nla_padlen(data_len)
bytes (1-3 when data_len is not 4-byte aligned) are never initialized,
leaking stale heap contents to userspace via the NFLOG netlink socket.
Replace the manual attribute construction with nla_reserve(), which
handles the tailroom check, header setup, and padding zeroing via
__nla_reserve(). The subsequent skb_copy_bits() fills in the payload
data on top of the properly initialized attribute.
Fixes: df6fb868d6 ("[NETFILTER]: nfnetlink: convert to generic netlink attribute functions")
Reported-by: Xiang Mei <xmei5@asu.edu>
Signed-off-by: Weiming Shi <bestswngs@gmail.com>
Signed-off-by: Florian Westphal <fw@strlen.de>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
The async_hold queue pins encrypted input skbs while
the AEAD engine references their scatterlist data. Once
tls_decrypt_async_wait() returns, every AEAD operation
has completed and the engine no longer references those
skbs, so they can be freed unconditionally.
A subsequent patch adds batch async decryption to
tls_sw_read_sock(), introducing a new call site that
must drain pending AEAD operations and release held
skbs. Move __skb_queue_purge(&ctx->async_hold) into
tls_decrypt_async_wait() so the purge is centralized
and every caller -- recvmsg's drain path, the -EBUSY
fallback in tls_do_decryption(), and the new read_sock
batch path -- releases held skbs on synchronization
without each site managing the purge independently.
This fixes a leak when tls_strp_msg_hold() fails part-way through,
after having added some cloned skbs to the async_hold
queue. tls_decrypt_sg() will then call tls_decrypt_async_wait() to
process all pending decrypts, and drop back to synchronous mode, but
tls_sw_recvmsg() only flushes the async_hold queue when one record has
been processed in "fully-async" mode, which may not be the case here.
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Reported-by: Yiming Qian <yimingqian591@gmail.com>
Fixes: b8a6ff84ab ("tls: wait for pending async decryptions if tls_strp_msg_hold fails")
Link: https://patch.msgid.link/20260324-tls-read-sock-v5-1-5408befe5774@oracle.com
[pabeni@redhat.com: added leak comment]
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
This test will fail without
the preceding commit ("netfilter: nft_set_pipapo_avx2: fix match retart if found element is expired"):
reject overlapping range on add 0s [ OK ]
reload with flush /dev/stdin:59:32-52: Error: Could not process rule: File exists
add element inet filter test { 10.0.0.29 . 10.0.2.29 }
Reviewed-by: Stefano Brivio <sbrivio@redhat.com>
Signed-off-by: Florian Westphal <fw@strlen.de>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
New test case fails unexpectedly when avx2 matching functions are used.
The test first loads a ranomly generated pipapo set
with 'ipv4 . port' key, i.e. nft -f foo.
This works. Then, it reloads the set after a flush:
(echo flush set t s; cat foo) | nft -f -
This is expected to work, because its the same set after all and it was
already loaded once.
But with avx2, this fails: nft reports a clashing element.
The reported clash is of following form:
We successfully re-inserted
a . b
c . d
Then we try to insert a . d
avx2 finds the already existing a . d, which (due to 'flush set') is marked
as invalid in the new generation. It skips the element and moves to next.
Due to incorrect masking, the skip-step finds the next matching
element *only considering the first field*,
i.e. we return the already reinserted "a . b", even though the
last field is different and the entry should not have been matched.
No such error is reported for the generic c implementation (no avx2) or when
the last field has to use the 'nft_pipapo_avx2_lookup_slow' fallback.
Bisection points to
7711f4bb4b ("netfilter: nft_set_pipapo: fix range overlap detection")
but that fix merely uncovers this bug.
Before this commit, the wrong element is returned, but erronously
reported as a full, identical duplicate.
The root-cause is too early return in the avx2 match functions.
When we process the last field, we should continue to process data
until the entire input size has been consumed to make sure no stale
bits remain in the map.
Link: https://lore.kernel.org/netfilter-devel/20260321152506.037f68c0@elisabeth/
Signed-off-by: Florian Westphal <fw@strlen.de>
Reviewed-by: Stefano Brivio <sbrivio@redhat.com>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
btusb_work() maps the number of active SCO links to USB alternate
settings through a three-entry lookup table when CVSD traffic uses
transparent voice settings. The lookup currently indexes alts[] with
data->sco_num - 1 without first constraining sco_num to the number of
available table entries.
While the table only defines alternate settings for up to three SCO
links, data->sco_num comes from hci_conn_num() and is used directly.
Cap the lookup to the last table entry before indexing it so the
driver keeps selecting the highest supported alternate setting without
reading past alts[].
Fixes: baac6276c0 ("Bluetooth: btusb: handle mSBC audio over USB Endpoints")
Signed-off-by: Pengpeng Hou <pengpeng@iscas.ac.cn>
Signed-off-by: Luiz Augusto von Dentz <luiz.von.dentz@intel.com>
l2cap_config_req() processes CONFIG_REQ for channels in BT_CONNECTED
state to support L2CAP reconfiguration (e.g. MTU changes). However,
since both CONF_INPUT_DONE and CONF_OUTPUT_DONE are already set from
the initial configuration, the reconfiguration path falls through to
l2cap_ertm_init(), which re-initializes tx_q, srej_q, srej_list, and
retrans_list without freeing the previous allocations and sets
chan->sdu to NULL without freeing the existing skb. This leaks all
previously allocated ERTM resources.
Additionally, l2cap_parse_conf_req() does not validate the minimum
value of remote_mps derived from the RFC max_pdu_size option. A zero
value propagates to l2cap_segment_sdu() where pdu_len becomes zero,
causing the while loop to never terminate since len is never
decremented, exhausting all available memory.
Fix the double-init by skipping l2cap_ertm_init() and
l2cap_chan_ready() when the channel is already in BT_CONNECTED state,
while still allowing the reconfiguration parameters to be updated
through l2cap_parse_conf_req(). Also add a pdu_len zero check in
l2cap_segment_sdu() as a safeguard.
Fixes: 96298f6401 ("Bluetooth: L2CAP: handle l2cap config request during open state")
Signed-off-by: Hyunwoo Kim <imv4bel@gmail.com>
Signed-off-by: Luiz Augusto von Dentz <luiz.von.dentz@intel.com>
l2cap_conn_del() calls cancel_delayed_work_sync() for both info_timer
and id_addr_timer while holding conn->lock. However, the work functions
l2cap_info_timeout() and l2cap_conn_update_id_addr() both acquire
conn->lock, creating a potential AB-BA deadlock if the work is already
executing when l2cap_conn_del() takes the lock.
Move the work cancellations before acquiring conn->lock and use
disable_delayed_work_sync() to additionally prevent the works from
being rearmed after cancellation, consistent with the pattern used in
hci_conn_del().
Fixes: ab4eedb790 ("Bluetooth: L2CAP: Fix corrupted list in hci_chan_del")
Signed-off-by: Hyunwoo Kim <imv4bel@gmail.com>
Signed-off-by: Luiz Augusto von Dentz <luiz.von.dentz@intel.com>
btintel_hw_error() issues two __hci_cmd_sync() calls (HCI_OP_RESET
and Intel exception-info retrieval) without holding
hci_req_sync_lock(). This lets it race against
hci_dev_do_close() -> btintel_shutdown_combined(), which also runs
__hci_cmd_sync() under the same lock. When both paths manipulate
hdev->req_status/req_rsp concurrently, the close path may free the
response skb first, and the still-running hw_error path hits a
slab-use-after-free in kfree_skb().
Wrap the whole recovery sequence in hci_req_sync_lock/unlock so it
is serialized with every other synchronous HCI command issuer.
Below is the data race report and the kasan report:
BUG: data-race in __hci_cmd_sync_sk / btintel_shutdown_combined
read of hdev->req_rsp at net/bluetooth/hci_sync.c:199
by task kworker/u17:1/83:
__hci_cmd_sync_sk+0x12f2/0x1c30 net/bluetooth/hci_sync.c:200
__hci_cmd_sync+0x55/0x80 net/bluetooth/hci_sync.c:223
btintel_hw_error+0x114/0x670 drivers/bluetooth/btintel.c:254
hci_error_reset+0x348/0xa30 net/bluetooth/hci_core.c:1030
write/free by task ioctl/22580:
btintel_shutdown_combined+0xd0/0x360
drivers/bluetooth/btintel.c:3648
hci_dev_close_sync+0x9ae/0x2c10 net/bluetooth/hci_sync.c:5246
hci_dev_do_close+0x232/0x460 net/bluetooth/hci_core.c:526
BUG: KASAN: slab-use-after-free in
sk_skb_reason_drop+0x43/0x380 net/core/skbuff.c:1202
Read of size 4 at addr ffff888144a738dc
by task kworker/u17:1/83:
__hci_cmd_sync_sk+0x12f2/0x1c30 net/bluetooth/hci_sync.c:200
__hci_cmd_sync+0x55/0x80 net/bluetooth/hci_sync.c:223
btintel_hw_error+0x186/0x670 drivers/bluetooth/btintel.c:260
Fixes: 973bb97e5a ("Bluetooth: btintel: Add generic function for handling hardware errors")
Signed-off-by: Cen Zhang <zzzccc427@gmail.com>
Signed-off-by: Luiz Augusto von Dentz <luiz.von.dentz@intel.com>
When the L2CAP channel mode is L2CAP_MODE_ERTM/L2CAP_MODE_STREAMING,
l2cap_publish_rx_avail will be called and le flow credits will be sent in
l2cap_chan_rx_avail, even though the link type is ACL.
The logs in question as follows:
> ACL Data RX: Handle 129 flags 0x02 dlen 12
L2CAP: Unknown (0x16) ident 4 len 4
40 00 ed 05
< ACL Data TX: Handle 129 flags 0x00 dlen 10
L2CAP: Command Reject (0x01) ident 4 len 2
Reason: Command not understood (0x0000)
Bluetooth: Unknown BR/EDR signaling command 0x16
Bluetooth: Wrong link type (-22)
Fixes: ce60b9231b ("Bluetooth: compute LE flow credits based on recvbuf space")
Signed-off-by: Zhang Chen <zhangchen01@kylinos.cn>
Signed-off-by: Luiz Augusto von Dentz <luiz.von.dentz@intel.com>
If auxiliary_device_add() fails, add_adev() jumps to add_fail and calls
auxiliary_device_uninit(adev).
The auxiliary device has its release callback set to adev_release(),
which frees the containing struct mana_adev. Since adev is embedded in
struct mana_adev, the subsequent fall-through to init_fail and access
to adev->id may result in a use-after-free.
Fix this by saving the allocated auxiliary device id in a local
variable before calling auxiliary_device_add(), and use that saved id
in the cleanup path after auxiliary_device_uninit().
Fixes: a69839d432 ("net: mana: Add support for auxiliary device")
Cc: stable@vger.kernel.org
Reviewed-by: Long Li <longli@microsoft.com>
Signed-off-by: Guangshuo Li <lgs201920130244@gmail.com>
Link: https://patch.msgid.link/20260323165730.945365-1-lgs201920130244@gmail.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
When codel_dequeue() finds an empty queue, it resets vars->dropping
but does not reset vars->first_above_time. The reference CoDel
algorithm (Nichols & Jacobson, ACM Queue 2012) resets both:
dodeque_result codel_queue_t::dodeque(time_t now) {
...
if (r.p == NULL) {
first_above_time = 0; // <-- Linux omits this
}
...
}
Note that codel_should_drop() does reset first_above_time when called
with a NULL skb, but codel_dequeue() returns early before ever calling
codel_should_drop() in the empty-queue case. The post-drop code paths
do reach codel_should_drop(NULL) and correctly reset the timer, so a
dropped packet breaks the cycle -- but the next delivered packet
re-arms first_above_time and the cycle repeats.
For sparse flows such as ICMP ping (one packet every 200ms-1s), the
first packet arms first_above_time, the flow goes empty, and the
second packet arrives after the interval has elapsed and gets dropped.
The pattern repeats, producing sustained loss on flows that are not
actually congested.
Test: veth pair, fq_codel, BQL disabled, 30000 iptables rules in the
consumer namespace (NAPI-64 cycle ~14ms, well above fq_codel's 5ms
target), ping at 5 pps under UDP flood:
Before fix: 26% ping packet loss
After fix: 0% ping packet loss
Fix by resetting first_above_time to zero in the empty-queue path
of codel_dequeue(), matching the reference algorithm.
Fixes: 76e3cc126b ("codel: Controlled Delay AQM")
Fixes: d068ca2ae2 ("codel: split into multiple files")
Co-developed-by: Jesper Dangaard Brouer <hawk@kernel.org>
Signed-off-by: Jesper Dangaard Brouer <hawk@kernel.org>
Signed-off-by: Jonas Köppeler <j.koeppeler@tu-berlin.de>
Reported-by: Chris Arges <carges@cloudflare.com>
Tested-by: Jonas Köppeler <j.koeppeler@tu-berlin.de>
Reviewed-by: Toke Høiland-Jørgensen <toke@redhat.com>
Link: https://lore.kernel.org/all/20260318134826.1281205-7-hawk@kernel.org/
Reviewed-by: Eric Dumazet <edumazet@google.com>
Link: https://patch.msgid.link/20260323174920.253526-1-hawk@kernel.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
The driver does not explicitly configure the MAC duplex mode when
bringing the link up. As a result, the MAC may retain a stale duplex
setting from a previous link state, leading to duplex mismatches with
the link partner and degraded network performance.
Update lan743x_phylink_mac_link_up() to set or clear the MAC_CR_DPX_
bit according to the negotiated duplex mode.
This ensures the MAC configuration is consistent with the phylink
resolved state.
Fixes: a5f199a8d8 ("net: lan743x: Migrate phylib to phylink")
Signed-off-by: Thangaraj Samynathan <thangaraj.s@microchip.com>
Reviewed-by: Russell King (Oracle) <rmk+kernel@armlinux.org.uk>
Link: https://patch.msgid.link/20260323065345.144915-1-thangaraj.s@microchip.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
A UAF issue occurs when the virtio_net driver is configured with napi_tx=N
and the device's IFF_XMIT_DST_RELEASE flag is cleared
(e.g., during the configuration of tc route filter rules).
When IFF_XMIT_DST_RELEASE is removed from the net_device, the network stack
expects the driver to hold the reference to skb->dst until the packet
is fully transmitted and freed. In virtio_net with napi_tx=N,
skbs may remain in the virtio transmit ring for an extended period.
If the network namespace is destroyed while these skbs are still pending,
the corresponding dst_ops structure has freed. When a subsequent packet
is transmitted, free_old_xmit() is triggered to clean up old skbs.
It then calls dst_release() on the skb associated with the stale dst_entry.
Since the dst_ops (referenced by the dst_entry) has already been freed,
a UAF kernel paging request occurs.
fix it by adds skb_dst_drop(skb) in start_xmit to explicitly release
the dst reference before the skb is queued in virtio_net.
Call Trace:
Unable to handle kernel paging request at virtual address ffff80007e150000
CPU: 2 UID: 0 PID: 6236 Comm: ping Kdump: loaded Not tainted 7.0.0-rc1+ #6 PREEMPT
...
percpu_counter_add_batch+0x3c/0x158 lib/percpu_counter.c:98 (P)
dst_release+0xe0/0x110 net/core/dst.c:177
skb_release_head_state+0xe8/0x108 net/core/skbuff.c:1177
sk_skb_reason_drop+0x54/0x2d8 net/core/skbuff.c:1255
dev_kfree_skb_any_reason+0x64/0x78 net/core/dev.c:3469
napi_consume_skb+0x1c4/0x3a0 net/core/skbuff.c:1527
__free_old_xmit+0x164/0x230 drivers/net/virtio_net.c:611 [virtio_net]
free_old_xmit drivers/net/virtio_net.c:1081 [virtio_net]
start_xmit+0x7c/0x530 drivers/net/virtio_net.c:3329 [virtio_net]
...
Reproduction Steps:
NETDEV="enp3s0"
config_qdisc_route_filter() {
tc qdisc del dev $NETDEV root
tc qdisc add dev $NETDEV root handle 1: prio
tc filter add dev $NETDEV parent 1:0 \
protocol ip prio 100 route to 100 flowid 1:1
ip route add 192.168.1.100/32 dev $NETDEV realm 100
}
test_ns() {
ip netns add testns
ip link set $NETDEV netns testns
ip netns exec testns ifconfig $NETDEV 10.0.32.46/24
ip netns exec testns ping -c 1 10.0.32.1
ip netns del testns
}
config_qdisc_route_filter
test_ns
sleep 2
test_ns
Fixes: f2fc6a5458 ("[NETNS][IPV6] route6 - move ip6_dst_ops inside the network namespace")
Cc: stable@vger.kernel.org
Signed-off-by: xietangxin <xietangxin@yeah.net>
Reviewed-by: Xuan Zhuo <xuanzhuo@linux.alibaba.com>
Fixes: 0287587884 ("net: better IFF_XMIT_DST_RELEASE support")
Link: https://patch.msgid.link/20260312025406.15641-1-xietangxin@yeah.net
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Steffen Klassert says:
====================
pull request (net): ipsec 2026-03-23
1) Add missing extack for XFRMA_SA_PCPU in add_acquire and allocspi.
From Sabrina Dubroca.
2) Fix the condition on x->pcpu_num in xfrm_sa_len by using the
proper check. From Sabrina Dubroca.
3) Call xdo_dev_state_delete during state update to properly cleanup
the xdo device state. From Sabrina Dubroca.
4) Fix a potential skb leak in espintcp when async crypto is used.
From Sabrina Dubroca.
5) Validate inner IPv4 header length in IPTFS payload to avoid
parsing malformed packets. From Roshan Kumar.
6) Fix skb_put() panic on non-linear skb during IPTFS reassembly.
From Fernando Fernandez Mancera.
7) Silence various sparse warnings related to RCU, state, and policy
handling. From Sabrina Dubroca.
8) Fix work re-schedule race after cancel in xfrm_nat_keepalive_net_fini().
From Hyunwoo Kim.
9) Prevent policy_hthresh.work from racing with netns teardown by using
a proper cleanup mechanism. From Minwoo Ra.
10) Validate that the family of the source and destination addresses match
in pfkey_send_migrate(). From Eric Dumazet.
11) Only publish mode_data after the clone is setup in the IPTFS receive path.
This prevents leaving x->mode_data pointing at freed memory on error.
From Paul Moses.
Please pull or let me know if there are problems.
ipsec-2026-03-23
* tag 'ipsec-2026-03-23' of git://git.kernel.org/pub/scm/linux/kernel/git/klassert/ipsec:
xfrm: iptfs: only publish mode_data after clone setup
af_key: validate families in pfkey_send_migrate()
xfrm: prevent policy_hthresh.work from racing with netns teardown
xfrm: Fix work re-schedule after cancel in xfrm_nat_keepalive_net_fini()
xfrm: avoid RCU warnings around the per-netns netlink socket
xfrm: add rcu_access_pointer to silence sparse warning for xfrm_input_afinfo
xfrm: policy: silence sparse warning in xfrm_policy_unregister_afinfo
xfrm: policy: fix sparse warnings in xfrm_policy_{init,fini}
xfrm: state: silence sparse warnings during netns exit
xfrm: remove rcu/state_hold from xfrm_state_lookup_spi_proto
xfrm: state: add xfrm_state_deref_prot to state_by* walk under lock
xfrm: state: fix sparse warnings around XFRM_STATE_INSERT
xfrm: state: fix sparse warnings in xfrm_state_init
xfrm: state: fix sparse warnings on xfrm_state_hold_rcu
xfrm: iptfs: fix skb_put() panic on non-linear skb during reassembly
xfrm: iptfs: validate inner IPv4 header length in IPTFS payload
esp: fix skb leak with espintcp and async crypto
xfrm: call xdo_dev_state_delete during state update
xfrm: fix the condition on x->pcpu_num in xfrm_sa_len
xfrm: add missing extack for XFRMA_SA_PCPU in add_acquire and allocspi
====================
Link: https://patch.msgid.link/20260323083440.2741292-1-steffen.klassert@secunet.com
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
NETIF_F_IPV6_CSUM only advertises support for checksum offload of
packets without IPv6 extension headers. Packets with extension
headers must fall back onto software checksumming. Since TSO
depends on checksum offload, those must revert to GSO.
The below commit introduces that fallback. It always checks
network header length. For tunneled packets, the inner header length
must be checked instead. Extend the check accordingly.
A special case is tunneled packets without inner IP protocol. Such as
RFC 6951 SCTP in UDP. Those are not standard IPv6 followed by
transport header either, so also must revert to the software GSO path.
Cc: stable@vger.kernel.org
Fixes: 864e339697 ("net: gso: Forbid IPv6 TSO with extensions on devices with only IPV6_CSUM")
Reported-by: Tangxin Xie <xietangxin@yeah.net>
Closes: https://lore.kernel.org/netdev/0414e7e2-9a1c-4d7c-a99d-b9039cf68f40@yeah.net/
Suggested-by: Paolo Abeni <pabeni@redhat.com>
Signed-off-by: Willem de Bruijn <willemb@google.com>
Link: https://patch.msgid.link/20260320190148.2409107-1-willemdebruijn.kernel@gmail.com
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Marc Kleine-Budde says:
====================
pull-request: can 2026-03-23
this is a pull request of 5 patches for net/main.
The first patch is by me and adds missing error handling to the CAN
netlink device configuration code.
Wenyuan Li contributes a patch for the mcp251x drier to add missing
error handling for power enabling in th open and resume functions.
Oliver Hartkopp's patch adds missing atomic access in hot path for the
CAN procfs statistics.
A series by Ali Norouzi and Oliver Hartkopp fix a can-Out-of-Bounds
Heap R/W in the can-gw protocol and a UAF in the CAN isotp protocol.
linux-can-fixes-for-7.0-20260323
* tag 'linux-can-fixes-for-7.0-20260323' of git://git.kernel.org/pub/scm/linux/kernel/git/mkl/linux-can:
can: isotp: fix tx.buf use-after-free in isotp_sendmsg()
can: gw: fix OOB heap access in cgw_csum_crc8_rel()
can: statistics: add missing atomic access in hot path
can: mcp251x: add error handling for power enable in open and resume
can: netlink: can_changelink(): add missing error handling to call can_ctrlmode_changelink()
====================
Link: https://patch.msgid.link/20260323103224.218099-1-mkl@pengutronix.de
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
cppi5_hdesc_get_psdata() returns a pointer into the CPPI descriptor.
In both emac_rx_packet() and emac_rx_packet_zc(), the descriptor is
freed via k3_cppi_desc_pool_free() before the psdata pointer is used
by emac_rx_timestamp(), which dereferences psdata[0] and psdata[1].
This constitutes a use-after-free on every received packet that goes
through the timestamp path.
Defer the descriptor free until after all accesses through the psdata
pointer are complete. For emac_rx_packet(), move the free into the
requeue label so both early-exit and success paths free the descriptor
after all accesses are done. For emac_rx_packet_zc(), move the free to
the end of the loop body after emac_dispatch_skb_zc() (which calls
emac_rx_timestamp()) has returned.
Fixes: 46eeb90f03 ("net: ti: icssg-prueth: Use page_pool API for RX buffer allocation")
Signed-off-by: David Carlier <devnexen@gmail.com>
Reviewed-by: Simon Horman <horms@kernel.org>
Link: https://patch.msgid.link/20260320174439.41080-1-devnexen@gmail.com
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Similar to commit 950803f725 ("bonding: fix type confusion in
bond_setup_by_slave()") team has the same class of header_ops type
confusion.
For non-Ethernet ports, team_setup_by_port() copies port_dev->header_ops
directly. When the team device later calls dev_hard_header() or
dev_parse_header(), these callbacks can run with the team net_device
instead of the real lower device, so netdev_priv(dev) is interpreted as
the wrong private type and can crash.
The syzbot report shows a crash in bond_header_create(), but the root
cause is in team: the topology is gre -> bond -> team, and team calls
the inherited header_ops with its own net_device instead of the lower
device, so bond_header_create() receives a team device and interprets
netdev_priv() as bonding private data, causing a type confusion crash.
Fix this by introducing team header_ops wrappers for create/parse,
selecting a team port under RCU, and calling the lower device callbacks
with port->dev, so each callback always sees the correct net_device
context.
Also pass the selected lower device to the lower parse callback, so
recursion is bounded in stacked non-Ethernet topologies and parse
callbacks always run with the correct device context.
Fixes: 1d76efe157 ("team: add support for non-ethernet devices")
Reported-by: syzbot+3d8bc31c45e11450f24c@syzkaller.appspotmail.com
Closes: https://lore.kernel.org/all/69b46af7.050a0220.36eb34.000e.GAE@google.com/T/
Cc: Jiayuan Chen <jiayuan.chen@linux.dev>
Signed-off-by: Jiayuan Chen <jiayuan.chen@shopee.com>
Link: https://patch.msgid.link/20260320072139.134249-2-jiayuan.chen@linux.dev
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Xuan Zhuo says:
====================
virtio-net: fix for VIRTIO_NET_F_GUEST_HDRLEN
The commit be50da3e9d ("net: virtio_net: implement exact header length
guest feature") introduces support for the VIRTIO_NET_F_GUEST_HDRLEN
feature in virtio-net.
This feature requires virtio-net to set hdr_len to the actual header
length of the packet when transmitting, the number of
bytes from the start of the packet to the beginning of the
transport-layer payload.
However, in practice, hdr_len was being set using skb_headlen(skb),
which is clearly incorrect. This path set fixes that issue.
As discussed in [0], this version checks the VIRTIO_NET_F_GUEST_HDRLEN is
negotiated.
[0]: http://lore.kernel.org/all/20251029030913.20423-1-xuanzhuo@linux.alibaba.com
v10: fix http://lore.kernel.org/all/202603122214.8Anoxrmq-lkp@intel.com
====================
Link: https://patch.msgid.link/20260320021818.111741-1-xuanzhuo@linux.alibaba.com
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
The commit a2fb4bc4e2 ("net: implement virtio helpers to handle UDP
GSO tunneling.") introduces support for the UDP GSO tunnel feature in
virtio-net.
The virtio spec says:
If the \field{gso_type} has the VIRTIO_NET_HDR_GSO_UDP_TUNNEL_IPV4 bit or
VIRTIO_NET_HDR_GSO_UDP_TUNNEL_IPV6 bit set, \field{hdr_len} accounts for
all the headers up to and including the inner transport.
The commit did not update the hdr_len to include the inner transport.
I observed that the "hdr_len" is 116 for this packet:
17:36:18.241105 52:55:00:d1:27:0a > 2e:2c:df:46:a9:e1, ethertype IPv4 (0x0800), length 2912: (tos 0x0, ttl 64, id 45197, offset 0, flags [none], proto UDP (17), length 2898)
192.168.122.100.50613 > 192.168.122.1.4789: [bad udp cksum 0x8106 -> 0x26a0!] VXLAN, flags [I] (0x08), vni 1
fa:c3:ba:82:05:ee > ce:85:0c:31:77:e5, ethertype IPv4 (0x0800), length 2862: (tos 0x0, ttl 64, id 14678, offset 0, flags [DF], proto TCP (6), length 2848)
192.168.3.1.49880 > 192.168.3.2.9898: Flags [P.], cksum 0x9266 (incorrect -> 0xaa20), seq 515667:518463, ack 1, win 64, options [nop,nop,TS val 2990048824 ecr 2798801412], length 2796
116 = 14(mac) + 20(ip) + 8(udp) + 8(vxlan) + 14(inner mac) + 20(inner ip) + 32(innner tcp)
Fixes: a2fb4bc4e2 ("net: implement virtio helpers to handle UDP GSO tunneling.")
Signed-off-by: Xuan Zhuo <xuanzhuo@linux.alibaba.com>
Link: https://patch.msgid.link/20260320021818.111741-3-xuanzhuo@linux.alibaba.com
Acked-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
The commit be50da3e9d ("net: virtio_net: implement exact header length
guest feature") introduces support for the VIRTIO_NET_F_GUEST_HDRLEN
feature in virtio-net.
This feature requires virtio-net to set hdr_len to the actual header
length of the packet when transmitting, the number of
bytes from the start of the packet to the beginning of the
transport-layer payload.
However, in practice, hdr_len was being set using skb_headlen(skb),
which is clearly incorrect. This commit fixes that issue.
Fixes: be50da3e9d ("net: virtio_net: implement exact header length guest feature")
Signed-off-by: Xuan Zhuo <xuanzhuo@linux.alibaba.com>
Link: https://patch.msgid.link/20260320021818.111741-2-xuanzhuo@linux.alibaba.com
Acked-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Currently, enetc_get_ringparam() only provides rx_pending and tx_pending,
but 'ethtool --show-ring' no longer displays these fields. Because the
ringparam retrieval path has moved to the new netlink interface, where
rings_fill_reply() emits the *x_pending only if the *x_max_pending values
are non-zero. So rx_max_pending and tx_max_pending to are added to
enetc_get_ringparam() to fix the issue.
Note that the maximum tx/rx ring size of hardware is 64K, but we haven't
added set_ringparam() to make the ring size configurable. To avoid users
mistakenly believing that the ring size can be increased, so set
the *x_max_pending to priv->*x_bd_count.
Fixes: e4a1717b67 ("ethtool: provide ring sizes with RINGS_GET request")
Signed-off-by: Wei Fang <wei.fang@nxp.com>
Link: https://patch.msgid.link/20260320094222.706339-1-wei.fang@nxp.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
When binding a udp_sock to a local address and port, UDP uses
two hashes (udptable->hash and udptable->hash2) for collision
detection. The current code switches to "hash2" when
hslot->count > 10.
"hash2" is keyed by local address and local port.
"hash" is keyed by local port only.
The issue can be shown in the following bind sequence (pseudo code):
bind(fd1, "[fd00::1]:8888")
bind(fd2, "[fd00::2]:8888")
bind(fd3, "[fd00::3]:8888")
bind(fd4, "[fd00::4]:8888")
bind(fd5, "[fd00::5]:8888")
bind(fd6, "[fd00::6]:8888")
bind(fd7, "[fd00::7]:8888")
bind(fd8, "[fd00::8]:8888")
bind(fd9, "[fd00::9]:8888")
bind(fd10, "[fd00::10]:8888")
/* Correctly return -EADDRINUSE because "hash" is used
* instead of "hash2". udp_lib_lport_inuse() detects the
* conflict.
*/
bind(fail_fd, "[::]:8888")
/* After one more socket is bound to "[fd00::11]:8888",
* hslot->count exceeds 10 and "hash2" is used instead.
*/
bind(fd11, "[fd00::11]:8888")
bind(fail_fd, "[::]:8888") /* succeeds unexpectedly */
The same issue applies to the IPv4 wildcard address "0.0.0.0"
and the IPv4-mapped wildcard address "::ffff:0.0.0.0". For
example, if there are existing sockets bound to
"192.168.1.[1-11]:8888", then binding "0.0.0.0:8888" or
"[::ffff:0.0.0.0]:8888" can also miss the conflict when
hslot->count > 10.
TCP inet_csk_get_port() already has the correct check in
inet_use_bhash2_on_bind(). Rename it to
inet_use_hash2_on_bind() and move it to inet_hashtables.h
so udp.c can reuse it in this fix.
Fixes: 30fff9231f ("udp: bind() optimisation")
Reported-by: Andrew Onyshchuk <oandrew@meta.com>
Signed-off-by: Martin KaFai Lau <martin.lau@kernel.org>
Reviewed-by: Kuniyuki Iwashima <kuniyu@google.com>
Link: https://patch.msgid.link/20260319181817.1901357-1-martin.lau@linux.dev
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
When CONFIG_FIXED_PHY=m but CONFIG_B44=y, the kernel fails to link:
ld.lld: error: undefined symbol: fixed_phy_unregister
>>> referenced by b44.c
>>> drivers/net/ethernet/broadcom/b44.o:(b44_remove_one) in archive vmlinux.a
ld.lld: error: undefined symbol: fixed_phy_register_100fd
>>> referenced by b44.c
>>> drivers/net/ethernet/broadcom/b44.o:(b44_register_phy_one) in archive vmlinux.a
The fixed phy support is small enough that just always enabling it
for b44 is the simplest solution, and it avoids adding ugly #ifdef
checks.
Fixes: 10d2f15afb ("net: b44: register a fixed phy using fixed_phy_register_100fd if needed")
Signed-off-by: Arnd Bergmann <arnd@arndb.de>
Link: https://patch.msgid.link/20260320154927.674555-1-arnd@kernel.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Since 0417adf367 ("ppp: fix race conditions in ppp_fill_forward_path")
dev_fill_forward_path() should be called with RCU read lock held. This
fix was applied to net, while the Airoha flowtable commit was applied to
net-next, so it hadn't been an issue until net was merged into net-next.
Fixes: a8bdd935d1 ("net: airoha: Add wlan flowtable TX offload")
Signed-off-by: Qingfang Deng <dqfext@gmail.com>
Acked-by: Lorenzo Bianconi <lorenzo@kernel.org>
Link: https://patch.msgid.link/20260320094315.525126-1-dqfext@gmail.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
`packet_release()` has a race window where `NETDEV_UP` can re-register a
socket into a fanout group's `arr[]` array. The re-registration is not
cleaned up by `fanout_release()`, leaving a dangling pointer in the fanout
array.
`packet_release()` does NOT zero `po->num` in its `bind_lock` section.
After releasing `bind_lock`, `po->num` is still non-zero and `po->ifindex`
still matches the bound device. A concurrent `packet_notifier(NETDEV_UP)`
that already found the socket in `sklist` can re-register the hook.
For fanout sockets, this re-registration calls `__fanout_link(sk, po)`
which adds the socket back into `f->arr[]` and increments `f->num_members`,
but does NOT increment `f->sk_ref`.
The fix sets `po->num` to zero in `packet_release` while `bind_lock` is
held to prevent NETDEV_UP from linking, preventing the race window.
This bug was found following an additional audit with Claude Code based
on CVE-2025-38617.
Fixes: ce06b03e60 ("packet: Add helpers to register/unregister ->prot_hook")
Link: https://blog.calif.io/p/a-race-within-a-race-exploiting-cve
Signed-off-by: Yochai Eisenrich <echelonh@gmail.com>
Reviewed-by: Willem de Bruijn <willemb@google.com>
Link: https://patch.msgid.link/20260319200610.25101-1-echelonh@gmail.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Kuniyuki Iwashima says:
====================
ipv6: Fix two GC issues with permanent routes.
Patch 1 fixes the unbounded growth of tb6_gc_hlist due to
permanent routes whose exception routes have all expired.
Patch 2 fixes an issue where exception routes tied to
permanent routes are not properly aged.
Patch 3 is a selftest for the issue fixed by Patch 2.
====================
Link: https://patch.msgid.link/20260320072317.2561779-1-kuniyu@google.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Without the prior commit, IPv6 GC cannot track exceptions tied
to permanent routes if they were originally added as temporary
routes.
Let's add a test case for the issue.
1. Add temporary routes
2. Create exceptions for the temporary routes
3. Promote the routes to permanent routes
4. Check if GC can find and purge the exceptions
A few notes:
+ At step 4, unlike other test cases, we cannot wait for
$GC_WAIT_TIME. While the exceptions are always iterable via
netlink (since it traverses the entire fib tree instead of
tb6_gc_hlist), rt6_nh_dump_exceptions() skips expired entries.
If we waited for the expiration time, we would be unable to
distinguish whether the exceptions were truly purged by GC or
just hidden due to being expired.
+ For the same reason, at step 2, we use ICMPv6 redirect message
instead of Packet Too Big message. This is because MTU exceptions
always have RTF_EXPIRES, and rt6_age_examine_exception() does not
respect the period specified by net.ipv6.route.flush=1.
+ We add a neighbour entry for the redirect target with NTF_ROUTER.
Without this, the exceptions would be removed at step 3 when the
fib6_may_remove_gc_list() is called.
Without the fix, the exceptions remain even after GC is triggered
by sysctl -wq net.ipv6.route.flush=1.
FAIL: Expected 0 routes, got 5
TEST: ipv6 route garbage collection (promote to permanent routes) [FAIL]
With the fix, GC purges the exceptions properly.
TEST: ipv6 route garbage collection (promote to permanent routes) [ OK ]
Signed-off-by: Kuniyuki Iwashima <kuniyu@google.com>
Reviewed-by: David Ahern <dsahern@kernel.org>
Link: https://patch.msgid.link/20260320072317.2561779-4-kuniyu@google.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
The cited commit mechanically put fib6_remove_gc_list()
just after every fib6_clean_expires() call.
When a temporary route is promoted to a permanent route,
there may already be exception routes tied to it.
If fib6_remove_gc_list() removes the route from tb6_gc_hlist,
such exception routes will no longer be aged.
Let's replace fib6_remove_gc_list() with a new helper
fib6_may_remove_gc_list() and use fib6_age_exceptions() there.
Note that net->ipv6 is only compiled when CONFIG_IPV6 is
enabled, so fib6_{add,remove,may_remove}_gc_list() are guarded.
Fixes: 5eb902b8e7 ("net/ipv6: Remove expired routes with a separated list of routes.")
Signed-off-by: Kuniyuki Iwashima <kuniyu@google.com>
Reviewed-by: David Ahern <dsahern@kernel.org>
Link: https://patch.msgid.link/20260320072317.2561779-3-kuniyu@google.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Commit 5eb902b8e7 ("net/ipv6: Remove expired routes with a
separated list of routes.") introduced a per-table GC list and
changed GC to iterate over that list instead of traversing
the entire route table.
However, it forgot to add permanent routes to tb6_gc_hlist
when exception routes are added.
Commit cfe82469a0 ("ipv6: add exception routes to GC list
in rt6_insert_exception") fixed that issue but introduced
another one.
Even after all exception routes expire, the permanent routes
remain in tb6_gc_hlist, potentially negating the performance
benefits intended by the initial change.
Let's count gc_args->more before and after rt6_age_exceptions()
and remove the permanent route when the delta is 0.
Note that the next patch will reuse fib6_age_exceptions().
Fixes: cfe82469a0 ("ipv6: add exception routes to GC list in rt6_insert_exception")
Signed-off-by: Kuniyuki Iwashima <kuniyu@google.com>
Reviewed-by: Xin Long <lucien.xin@gmail.com>
Reviewed-by: David Ahern <dsahern@kernel.org>
Link: https://patch.msgid.link/20260320072317.2561779-2-kuniyu@google.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Deinit calls idpf_idc_deinit_core_aux_device to free the cdev_info
memory, but leaves the adapter->cdev_info field with a stale pointer
value. This will bypass subsequent "if (!cdev_info)" checks if cdev_info
is not reallocated. For example, if idc_init fails after a reset,
cdev_info will already have been freed during the reset handling, but it
will not have been reallocated. The next reset or rmmod will result in a
crash.
[ +0.000008] BUG: kernel NULL pointer dereference, address: 00000000000000d0
[ +0.000033] #PF: supervisor read access in kernel mode
[ +0.000020] #PF: error_code(0x0000) - not-present page
[ +0.000017] PGD 2097dfa067 P4D 0
[ +0.000017] Oops: Oops: 0000 [#1] SMP NOPTI
...
[ +0.000018] RIP: 0010:device_del+0x3e/0x3d0
[ +0.000010] Call Trace:
[ +0.000010] <TASK>
[ +0.000012] idpf_idc_deinit_core_aux_device+0x36/0x70 [idpf]
[ +0.000034] idpf_vc_core_deinit+0x3e/0x180 [idpf]
[ +0.000035] idpf_remove+0x40/0x1d0 [idpf]
[ +0.000035] pci_device_remove+0x42/0xb0
[ +0.000020] device_release_driver_internal+0x19c/0x200
[ +0.000024] driver_detach+0x48/0x90
[ +0.000018] bus_remove_driver+0x6d/0x100
[ +0.000023] pci_unregister_driver+0x2e/0xb0
[ +0.000022] __do_sys_delete_module.isra.0+0x18c/0x2b0
[ +0.000025] ? kmem_cache_free+0x2c2/0x390
[ +0.000023] do_syscall_64+0x107/0x7d0
[ +0.000023] entry_SYSCALL_64_after_hwframe+0x76/0x7e
Pass the adapter struct into idpf_idc_deinit_core_aux_device instead and
clear the cdev_info ptr.
Fixes: f4312e6bfa ("idpf: implement core RDMA auxiliary dev create, init, and destroy")
Signed-off-by: Joshua Hay <joshua.a.hay@intel.com>
Reviewed-by: Przemek Kitszel <przemyslaw.kitszel@intel.com>
Reviewed-by: Aleksandr Loktionov <aleksandr.loktionov@intel.com>
Reviewed-by: Simon Horman <horms@kernel.org>
Tested-by: Samuel Salin <Samuel.salin@intel.com>
Signed-off-by: Tony Nguyen <anthony.l.nguyen@intel.com>
